#include <linux/udp.h>
#include <linux/version.h>
#include <linux/ethtool.h>
- #include <linux/random.h>
#include <linux/wait.h>
#include <asm/system.h>
#include <asm/div64.h>
#include <linux/inetdevice.h>
#include <linux/list.h>
#include <linux/rculist.h>
- #include <linux/workqueue.h>
#include <linux/dmi.h>
#include <net/inet_ecn.h>
#include <linux/compat.h>
-#include "openvswitch/datapath-protocol.h"
+#include "openvswitch/xflow.h"
#include "datapath.h"
#include "actions.h"
#include "flow.h"
-#include "odp-compat.h"
+#include "xflow-compat.h"
#include "table.h"
#include "vport-internal_dev.h"
* It is safe to access the datapath and dp_port structures with just
* dp_mutex.
*/
-static struct datapath *dps[ODP_MAX];
+static struct datapath *dps[XFLOW_MAX];
static DEFINE_MUTEX(dp_mutex);
/* We limit the number of times that we pass into dp_process_received_packet()
static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters);
-static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
+static int new_dp_port(struct datapath *, struct xflow_port *, int port_no);
/* Must be called with rcu_read_lock or dp_mutex. */
struct datapath *get_dp(int dp_idx)
{
- if (dp_idx < 0 || dp_idx >= ODP_MAX)
+ if (dp_idx < 0 || dp_idx >= XFLOW_MAX)
return NULL;
return rcu_dereference(dps[dp_idx]);
}
/* Must be called with rcu_read_lock or RTNL lock. */
const char *dp_name(const struct datapath *dp)
{
- return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
+ return vport_get_name(dp->ports[XFLOWP_LOCAL]->vport);
}
static inline size_t br_nlmsg_size(void)
hdr->ifi_change = 0;
NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
- NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
+ NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[XFLOWP_LOCAL]->vport));
NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
static int create_dp(int dp_idx, const char __user *devnamep)
{
- struct odp_port internal_dev_port;
+ struct xflow_port internal_dev_port;
char devname[IFNAMSIZ];
struct datapath *dp;
int err;
/* Set up our datapath device. */
BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
strcpy(internal_dev_port.devname, devname);
- internal_dev_port.flags = ODP_PORT_INTERNAL;
- err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
+ internal_dev_port.flags = XFLOW_PORT_INTERNAL;
+ err = new_dp_port(dp, &internal_dev_port, XFLOWP_LOCAL);
if (err) {
if (err == -EBUSY)
err = -EEXIST;
return 0;
err_destroy_local_port:
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[XFLOWP_LOCAL], 1);
err_destroy_table:
tbl_destroy(dp->table, NULL);
err_free_dp:
int i;
list_for_each_entry_safe (p, n, &dp->port_list, node)
- if (p->port_no != ODPP_LOCAL)
+ if (p->port_no != XFLOWP_LOCAL)
dp_detach_port(p, 1);
dp_sysfs_del_dp(dp);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[XFLOWP_LOCAL], 1);
tbl_destroy(dp->table, flow_free_tbl);
};
/* Called with RTNL lock and dp_mutex. */
-static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
+static int new_dp_port(struct datapath *dp, struct xflow_port *xflow_port, int port_no)
{
struct vport *vport;
struct dp_port *p;
int err;
- vport = vport_locate(odp_port->devname);
+ vport = vport_locate(xflow_port->devname);
if (!vport) {
vport_lock();
- if (odp_port->flags & ODP_PORT_INTERNAL)
- vport = vport_add(odp_port->devname, "internal", NULL);
+ if (xflow_port->flags & XFLOW_PORT_INTERNAL)
+ vport = vport_add(xflow_port->devname, "internal", NULL);
else
- vport = vport_add(odp_port->devname, "netdev", NULL);
+ vport = vport_add(xflow_port->devname, "netdev", NULL);
vport_unlock();
return 0;
}
-static int attach_port(int dp_idx, struct odp_port __user *portp)
+static int attach_port(int dp_idx, struct xflow_port __user *portp)
{
struct datapath *dp;
- struct odp_port port;
+ struct xflow_port port;
int port_no;
int err;
ASSERT_RTNL();
- if (p->port_no != ODPP_LOCAL)
+ if (p->port_no != XFLOWP_LOCAL)
dp_sysfs_del_if(p);
dp_ifinfo_notify(RTM_DELLINK, p);
int err;
err = -EINVAL;
- if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
+ if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == XFLOWP_LOCAL)
goto out;
rtnl_lock();
struct datapath *dp = p->dp;
struct dp_stats_percpu *stats;
int stats_counter_off;
- struct xflow_key key;
- struct tbl_node *flow_node;
- struct sw_flow *flow;
struct sw_flow_actions *acts;
struct loop_counter *loop;
int error;
OVS_CB(skb)->dp_port = p;
- /* Extract flow from 'skb' into 'key'. */
- error = flow_extract(skb, p ? p->port_no : XFLOWP_NONE, &key);
- if (unlikely(error)) {
- kfree_skb(skb);
- return;
- }
+ if (!OVS_CB(skb)->flow) {
- struct odp_flow_key key;
++ struct xflow_key key;
+ struct tbl_node *flow_node;
+ bool is_frag;
- if (OVS_CB(skb)->is_frag && dp->drop_frags) {
- kfree_skb(skb);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
- goto out;
- }
+ /* Extract flow from 'skb' into 'key'. */
- error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
++ error = flow_extract(skb, p ? p->port_no : XFLOWP_NONE, &key, &is_frag);
+ if (unlikely(error)) {
+ kfree_skb(skb);
+ return;
+ }
- /* Look up flow. */
- flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp);
- if (unlikely(!flow_node)) {
- dp_output_control(dp, skb, _XFLOWL_MISS_NR, OVS_CB(skb)->tun_id);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
- goto out;
+ if (is_frag && dp->drop_frags) {
+ kfree_skb(skb);
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
+ goto out;
+ }
+
+ /* Look up flow. */
+ flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
+ flow_hash(&key), flow_cmp);
+ if (unlikely(!flow_node)) {
- dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
++ dp_output_control(dp, skb, _XFLOWL_MISS_NR, OVS_CB(skb)->tun_id);
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
+ goto out;
+ }
+
+ OVS_CB(skb)->flow = flow_cast(flow_node);
}
- flow = flow_cast(flow_node);
- flow_used(flow, skb);
+ flow_used(OVS_CB(skb)->flow, skb);
- acts = rcu_dereference(flow->sf_acts);
+ acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
/* Check whether we've looped too much. */
loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()];
}
/* Execute actions. */
- execute_actions(dp, skb, &key, acts->actions, acts->n_actions, GFP_ATOMIC);
+ execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
+ acts->n_actions, GFP_ATOMIC);
stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
/* Check whether sub-actions looped too much. */
if (OVS_CB(skb)->dp_port)
port_no = OVS_CB(skb)->dp_port->port_no;
else
- port_no = ODPP_LOCAL;
+ port_no = XFLOWP_LOCAL;
do {
- struct odp_msg *header;
+ struct xflow_msg *header;
nskb = skb->next;
skb->next = NULL;
if (err)
goto err_kfree_skbs;
- header = (struct odp_msg*)__skb_push(skb, sizeof *header);
+ header = (struct xflow_msg*)__skb_push(skb, sizeof *header);
header->type = queue_no;
header->length = skb->len;
header->port = port_no;
int err;
WARN_ON_ONCE(skb_shared(skb));
- BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
+ BUG_ON(queue_no != _XFLOWL_MISS_NR && queue_no != _XFLOWL_ACTION_NR && queue_no != _XFLOWL_SFLOW_NR);
queue = &dp->queues[queue_no];
err = -ENOBUFS;
if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
unsigned int i;
for (i = 0; i < actions->n_actions; i++) {
- const union odp_action *a = &actions->actions[i];
+ const union xflow_action *a = &actions->actions[i];
+ __be16 mask;
+
switch (a->type) {
- case ODPAT_OUTPUT:
+ case XFLOWAT_OUTPUT:
if (a->output.port >= DP_MAX_PORTS)
return -EINVAL;
break;
- case ODPAT_OUTPUT_GROUP:
+ case XFLOWAT_OUTPUT_GROUP:
if (a->output_group.group >= DP_MAX_GROUPS)
return -EINVAL;
break;
- case ODPAT_SET_VLAN_VID:
- if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
+ case XFLOWAT_SET_DL_TCI:
+ mask = a->dl_tci.mask;
+ if (mask != htons(VLAN_VID_MASK) &&
+ mask != htons(VLAN_PCP_MASK) &&
+ mask != htons(VLAN_VID_MASK | VLAN_PCP_MASK))
return -EINVAL;
- break;
-
- case ODPAT_SET_VLAN_PCP:
- if (a->vlan_pcp.vlan_pcp
- & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
+ if (a->dl_tci.tci & ~mask)
return -EINVAL;
break;
- case ODPAT_SET_NW_TOS:
+ case XFLOWAT_SET_NW_TOS:
if (a->nw_tos.nw_tos & INET_ECN_MASK)
return -EINVAL;
break;
default:
- if (a->type >= ODPAT_N_ACTIONS)
+ if (a->type >= XFLOWAT_N_ACTIONS)
return -EOPNOTSUPP;
break;
}
return 0;
}
-static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
+static struct sw_flow_actions *get_actions(const struct xflow_flow *flow)
{
struct sw_flow_actions *actions;
int error;
error = -EFAULT;
if (copy_from_user(actions->actions, flow->actions,
- flow->n_actions * sizeof(union odp_action)))
+ flow->n_actions * sizeof(union xflow_action)))
goto error_free_actions;
error = validate_actions(actions);
if (error)
return timespec_sub(now_mono, now_jiffies);
}
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats,
+static void get_stats(struct sw_flow *flow, struct xflow_flow_stats *stats,
struct timespec time_offset)
{
if (flow->used) {
return 0;
}
-static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
- struct odp_flow_stats *stats)
+static int do_put_flow(struct datapath *dp, struct xflow_flow_put *uf,
+ struct xflow_flow_stats *stats)
{
struct tbl_node *flow_node;
struct sw_flow *flow;
struct tbl *table;
int error;
- memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved);
-
table = rcu_dereference(dp->table);
flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
if (!flow_node) {
struct sw_flow_actions *acts;
error = -ENOENT;
- if (!(uf->flags & ODPPF_CREATE))
+ if (!(uf->flags & XFLOWPF_CREATE))
goto error;
/* Expand table, if necessary, to make room. */
}
/* Allocate flow. */
- error = -ENOMEM;
- flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
- if (flow == NULL)
+ flow = flow_alloc();
+ if (IS_ERR(flow)) {
+ error = PTR_ERR(flow);
goto error;
+ }
flow->key = uf->flow.key;
- spin_lock_init(&flow->lock);
clear_stats(flow);
/* Obtain actions. */
if (error)
goto error_free_flow_acts;
- memset(stats, 0, sizeof(struct odp_flow_stats));
+ memset(stats, 0, sizeof(struct xflow_flow_stats));
} else {
/* We found a matching flow. */
struct sw_flow_actions *old_acts, *new_acts;
/* Bail out if we're not allowed to modify an existing flow. */
error = -EEXIST;
- if (!(uf->flags & ODPPF_MODIFY))
+ if (!(uf->flags & XFLOWPF_MODIFY))
goto error;
/* Swap actions. */
old_acts = rcu_dereference(flow->sf_acts);
if (old_acts->n_actions != new_acts->n_actions ||
memcmp(old_acts->actions, new_acts->actions,
- sizeof(union odp_action) * old_acts->n_actions)) {
+ sizeof(union xflow_action) * old_acts->n_actions)) {
rcu_assign_pointer(flow->sf_acts, new_acts);
flow_deferred_free_acts(old_acts);
} else {
/* Fetch stats, then clear them if necessary. */
spin_lock_bh(&flow->lock);
get_stats(flow, stats, get_time_offset());
- if (uf->flags & ODPPF_ZERO_STATS)
+ if (uf->flags & XFLOWPF_ZERO_STATS)
clear_stats(flow);
spin_unlock_bh(&flow->lock);
}
error_free_flow_acts:
kfree(flow->sf_acts);
error_free_flow:
- kmem_cache_free(flow_cache, flow);
+ flow->sf_acts = NULL;
+ flow_put(flow);
error:
return error;
}
-static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
+static int put_flow(struct datapath *dp, struct xflow_flow_put __user *ufp)
{
- struct odp_flow_stats stats;
- struct odp_flow_put uf;
+ struct xflow_flow_stats stats;
+ struct xflow_flow_put uf;
int error;
- if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
+ if (copy_from_user(&uf, ufp, sizeof(struct xflow_flow_put)))
return -EFAULT;
error = do_put_flow(dp, &uf, &stats);
return error;
if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
+ sizeof(struct xflow_flow_stats)))
return -EFAULT;
return 0;
static int do_answer_query(struct sw_flow *flow, u32 query_flags,
struct timespec time_offset,
- struct odp_flow_stats __user *ustats,
- union odp_action __user *actions,
+ struct xflow_flow_stats __user *ustats,
+ union xflow_action __user *actions,
u32 __user *n_actionsp)
{
struct sw_flow_actions *sf_acts;
- struct odp_flow_stats stats;
+ struct xflow_flow_stats stats;
u32 n_actions;
spin_lock_bh(&flow->lock);
get_stats(flow, &stats, time_offset);
- if (query_flags & ODPFF_ZERO_TCP_FLAGS)
+ if (query_flags & XFLOWFF_ZERO_TCP_FLAGS)
flow->tcp_flags = 0;
spin_unlock_bh(&flow->lock);
- if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
+ if (copy_to_user(ustats, &stats, sizeof(struct xflow_flow_stats)) ||
get_user(n_actions, n_actionsp))
return -EFAULT;
sf_acts = rcu_dereference(flow->sf_acts);
if (put_user(sf_acts->n_actions, n_actionsp) ||
(actions && copy_to_user(actions, sf_acts->actions,
- sizeof(union odp_action) *
+ sizeof(union xflow_action) *
min(sf_acts->n_actions, n_actions))))
return -EFAULT;
static int answer_query(struct sw_flow *flow, u32 query_flags,
struct timespec time_offset,
- struct odp_flow __user *ufp)
+ struct xflow_flow __user *ufp)
{
- union odp_action *actions;
+ union xflow_action *actions;
if (get_user(actions, &ufp->actions))
return -EFAULT;
&ufp->stats, actions, &ufp->n_actions);
}
-static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
+static struct sw_flow *do_del_flow(struct datapath *dp, struct xflow_key *key)
{
struct tbl *table = rcu_dereference(dp->table);
struct tbl_node *flow_node;
int error;
- memset(key->reserved, 0, sizeof key->reserved);
flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
if (!flow_node)
return ERR_PTR(-ENOENT);
return flow_cast(flow_node);
}
-static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
+static int del_flow(struct datapath *dp, struct xflow_flow __user *ufp)
{
struct sw_flow *flow;
- struct odp_flow uf;
+ struct xflow_flow uf;
int error;
if (copy_from_user(&uf, ufp, sizeof uf))
return error;
}
-static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
+static int do_query_flows(struct datapath *dp, const struct xflow_flowvec *flowvec)
{
struct tbl *table = rcu_dereference(dp->table);
struct timespec time_offset;
time_offset = get_time_offset();
for (i = 0; i < flowvec->n_flows; i++) {
- struct odp_flow __user *ufp = &flowvec->flows[i];
- struct odp_flow uf;
+ struct xflow_flow __user *ufp = &flowvec->flows[i];
+ struct xflow_flow uf;
struct tbl_node *flow_node;
int error;
if (copy_from_user(&uf, ufp, sizeof uf))
return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
if (!flow_node)
}
struct list_flows_cbdata {
- struct odp_flow __user *uflows;
+ struct xflow_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
struct timespec time_offset;
{
struct sw_flow *flow = flow_cast(node);
struct list_flows_cbdata *cbdata = cbdata_;
- struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
+ struct xflow_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
int error;
if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
return 0;
}
-static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
+static int do_list_flows(struct datapath *dp, const struct xflow_flowvec *flowvec)
{
struct list_flows_cbdata cbdata;
int error;
static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
int (*function)(struct datapath *,
- const struct odp_flowvec *))
+ const struct xflow_flowvec *))
{
- struct odp_flowvec __user *uflowvec;
- struct odp_flowvec flowvec;
+ struct xflow_flowvec __user *uflowvec;
+ struct xflow_flowvec flowvec;
int retval;
- uflowvec = (struct odp_flowvec __user *)argp;
+ uflowvec = (struct xflow_flowvec __user *)argp;
if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
return -EFAULT;
- if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
+ if (flowvec.n_flows > INT_MAX / sizeof(struct xflow_flow))
return -EINVAL;
retval = function(dp, &flowvec);
: put_user(retval, &uflowvec->n_flows));
}
-static int do_execute(struct datapath *dp, const struct odp_execute *execute)
+static int do_execute(struct datapath *dp, const struct xflow_execute *execute)
{
- struct odp_flow_key key;
+ struct xflow_key key;
struct sk_buff *skb;
struct sw_flow_actions *actions;
struct ethhdr *eth;
+ bool is_frag;
int err;
err = -EINVAL;
if (execute->length < ETH_HLEN || execute->length > 65535)
goto error;
- err = -ENOMEM;
actions = flow_actions_alloc(execute->n_actions);
- if (!actions)
+ if (IS_ERR(actions)) {
+ err = PTR_ERR(actions);
goto error;
+ }
err = -EFAULT;
if (copy_from_user(actions->actions, execute->actions,
else
skb->protocol = htons(ETH_P_802_2);
- err = flow_extract(skb, execute->in_port, &key);
+ err = flow_extract(skb, execute->in_port, &key, &is_frag);
if (err)
goto error_free_skb;
return err;
}
-static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
+static int execute_packet(struct datapath *dp, const struct xflow_execute __user *executep)
{
- struct odp_execute execute;
+ struct xflow_execute execute;
if (copy_from_user(&execute, executep, sizeof execute))
return -EFAULT;
return do_execute(dp, &execute);
}
-static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
+static int get_dp_stats(struct datapath *dp, struct xflow_stats __user *statsp)
{
struct tbl *table = rcu_dereference(dp->table);
- struct odp_stats stats;
+ struct xflow_stats stats;
int i;
stats.n_flows = tbl_count(table);
}
}
-static int put_port(const struct dp_port *p, struct odp_port __user *uop)
+static int put_port(const struct dp_port *p, struct xflow_port __user *uop)
{
- struct odp_port op;
+ struct xflow_port op;
memset(&op, 0, sizeof op);
rcu_read_unlock();
op.port = p->port_no;
- op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
+ op.flags = is_internal_vport(p->vport) ? XFLOW_PORT_INTERNAL : 0;
return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
}
-static int query_port(struct datapath *dp, struct odp_port __user *uport)
+static int query_port(struct datapath *dp, struct xflow_port __user *uport)
{
- struct odp_port port;
+ struct xflow_port port;
if (copy_from_user(&port, uport, sizeof port))
return -EFAULT;
return put_port(dp->ports[port.port], uport);
}
-static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
+static int do_list_ports(struct datapath *dp, struct xflow_port __user *uports,
int n_ports)
{
int idx = 0;
return idx;
}
-static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
+static int list_ports(struct datapath *dp, struct xflow_portvec __user *upv)
{
- struct odp_portvec pv;
+ struct xflow_portvec pv;
int retval;
if (copy_from_user(&pv, upv, sizeof pv))
}
static int set_port_group(struct datapath *dp,
- const struct odp_port_group __user *upg)
+ const struct xflow_port_group __user *upg)
{
- struct odp_port_group pg;
+ struct xflow_port_group pg;
if (copy_from_user(&pg, upg, sizeof pg))
return -EFAULT;
return 0;
}
-static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg)
+static int get_port_group(struct datapath *dp, struct xflow_port_group __user *upg)
{
- struct odp_port_group pg;
+ struct xflow_port_group pg;
if (copy_from_user(&pg, upg, sizeof pg))
return -EFAULT;
/* Handle commands with special locking requirements up front. */
switch (cmd) {
- case ODP_DP_CREATE:
+ case XFLOW_DP_CREATE:
err = create_dp(dp_idx, (char __user *)argp);
goto exit;
- case ODP_DP_DESTROY:
+ case XFLOW_DP_DESTROY:
err = destroy_dp(dp_idx);
goto exit;
- case ODP_PORT_ATTACH:
- err = attach_port(dp_idx, (struct odp_port __user *)argp);
+ case XFLOW_PORT_ATTACH:
+ err = attach_port(dp_idx, (struct xflow_port __user *)argp);
goto exit;
- case ODP_PORT_DETACH:
+ case XFLOW_PORT_DETACH:
err = get_user(port_no, (int __user *)argp);
if (!err)
err = detach_port(dp_idx, port_no);
goto exit;
- case ODP_VPORT_ADD:
- err = vport_user_add((struct odp_vport_add __user *)argp);
+ case XFLOW_VPORT_ADD:
+ err = vport_user_add((struct xflow_vport_add __user *)argp);
goto exit;
- case ODP_VPORT_MOD:
- err = vport_user_mod((struct odp_vport_mod __user *)argp);
+ case XFLOW_VPORT_MOD:
+ err = vport_user_mod((struct xflow_vport_mod __user *)argp);
goto exit;
- case ODP_VPORT_DEL:
+ case XFLOW_VPORT_DEL:
err = vport_user_del((char __user *)argp);
goto exit;
- case ODP_VPORT_STATS_GET:
- err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
+ case XFLOW_VPORT_STATS_GET:
+ err = vport_user_stats_get((struct xflow_vport_stats_req __user *)argp);
goto exit;
- case ODP_VPORT_STATS_SET:
- err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
+ case XFLOW_VPORT_STATS_SET:
+ err = vport_user_stats_set((struct xflow_vport_stats_req __user *)argp);
goto exit;
- case ODP_VPORT_ETHER_GET:
- err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
+ case XFLOW_VPORT_ETHER_GET:
+ err = vport_user_ether_get((struct xflow_vport_ether __user *)argp);
goto exit;
- case ODP_VPORT_ETHER_SET:
- err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
+ case XFLOW_VPORT_ETHER_SET:
+ err = vport_user_ether_set((struct xflow_vport_ether __user *)argp);
goto exit;
- case ODP_VPORT_MTU_GET:
- err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
+ case XFLOW_VPORT_MTU_GET:
+ err = vport_user_mtu_get((struct xflow_vport_mtu __user *)argp);
goto exit;
- case ODP_VPORT_MTU_SET:
- err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
+ case XFLOW_VPORT_MTU_SET:
+ err = vport_user_mtu_set((struct xflow_vport_mtu __user *)argp);
goto exit;
}
goto exit;
switch (cmd) {
- case ODP_DP_STATS:
- err = get_dp_stats(dp, (struct odp_stats __user *)argp);
+ case XFLOW_DP_STATS:
+ err = get_dp_stats(dp, (struct xflow_stats __user *)argp);
break;
- case ODP_GET_DROP_FRAGS:
+ case XFLOW_GET_DROP_FRAGS:
err = put_user(dp->drop_frags, (int __user *)argp);
break;
- case ODP_SET_DROP_FRAGS:
+ case XFLOW_SET_DROP_FRAGS:
err = get_user(drop_frags, (int __user *)argp);
if (err)
break;
err = 0;
break;
- case ODP_GET_LISTEN_MASK:
+ case XFLOW_GET_LISTEN_MASK:
err = put_user(get_listen_mask(f), (int __user *)argp);
break;
- case ODP_SET_LISTEN_MASK:
+ case XFLOW_SET_LISTEN_MASK:
err = get_user(listeners, (int __user *)argp);
if (err)
break;
err = -EINVAL;
- if (listeners & ~ODPL_ALL)
+ if (listeners & ~XFLOWL_ALL)
break;
err = 0;
set_listen_mask(f, listeners);
break;
- case ODP_GET_SFLOW_PROBABILITY:
+ case XFLOW_GET_SFLOW_PROBABILITY:
err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
break;
- case ODP_SET_SFLOW_PROBABILITY:
+ case XFLOW_SET_SFLOW_PROBABILITY:
err = get_user(sflow_probability, (unsigned int __user *)argp);
if (!err)
dp->sflow_probability = sflow_probability;
break;
- case ODP_PORT_QUERY:
- err = query_port(dp, (struct odp_port __user *)argp);
+ case XFLOW_PORT_QUERY:
+ err = query_port(dp, (struct xflow_port __user *)argp);
break;
- case ODP_PORT_LIST:
- err = list_ports(dp, (struct odp_portvec __user *)argp);
+ case XFLOW_PORT_LIST:
+ err = list_ports(dp, (struct xflow_portvec __user *)argp);
break;
- case ODP_PORT_GROUP_SET:
- err = set_port_group(dp, (struct odp_port_group __user *)argp);
+ case XFLOW_PORT_GROUP_SET:
+ err = set_port_group(dp, (struct xflow_port_group __user *)argp);
break;
- case ODP_PORT_GROUP_GET:
- err = get_port_group(dp, (struct odp_port_group __user *)argp);
+ case XFLOW_PORT_GROUP_GET:
+ err = get_port_group(dp, (struct xflow_port_group __user *)argp);
break;
- case ODP_FLOW_FLUSH:
+ case XFLOW_FLOW_FLUSH:
err = flush_flows(dp);
break;
- case ODP_FLOW_PUT:
- err = put_flow(dp, (struct odp_flow_put __user *)argp);
+ case XFLOW_FLOW_PUT:
+ err = put_flow(dp, (struct xflow_flow_put __user *)argp);
break;
- case ODP_FLOW_DEL:
- err = del_flow(dp, (struct odp_flow __user *)argp);
+ case XFLOW_FLOW_DEL:
+ err = del_flow(dp, (struct xflow_flow __user *)argp);
break;
- case ODP_FLOW_GET:
+ case XFLOW_FLOW_GET:
err = do_flowvec_ioctl(dp, argp, do_query_flows);
break;
- case ODP_FLOW_LIST:
+ case XFLOW_FLOW_LIST:
err = do_flowvec_ioctl(dp, argp, do_list_flows);
break;
- case ODP_EXECUTE:
- err = execute_packet(dp, (struct odp_execute __user *)argp);
+ case XFLOW_EXECUTE:
+ err = execute_packet(dp, (struct xflow_execute __user *)argp);
break;
default:
}
#ifdef CONFIG_COMPAT
-static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
+static int compat_list_ports(struct datapath *dp, struct compat_xflow_portvec __user *upv)
{
- struct compat_odp_portvec pv;
+ struct compat_xflow_portvec pv;
int retval;
if (copy_from_user(&pv, upv, sizeof pv))
return put_user(retval, &upv->n_ports);
}
-static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg)
+static int compat_set_port_group(struct datapath *dp, const struct compat_xflow_port_group __user *upg)
{
- struct compat_odp_port_group pg;
+ struct compat_xflow_port_group pg;
if (copy_from_user(&pg, upg, sizeof pg))
return -EFAULT;
return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group);
}
-static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg)
+static int compat_get_port_group(struct datapath *dp, struct compat_xflow_port_group __user *upg)
{
- struct compat_odp_port_group pg;
+ struct compat_xflow_port_group pg;
if (copy_from_user(&pg, upg, sizeof pg))
return -EFAULT;
pg.group, &upg->n_ports);
}
-static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
+static int compat_get_flow(struct xflow_flow *flow, const struct compat_xflow_flow __user *compat)
{
compat_uptr_t actions;
- if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
- __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
- __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
+ if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_xflow_flow)) ||
+ __copy_from_user(&flow->stats, &compat->stats, sizeof(struct xflow_flow_stats)) ||
+ __copy_from_user(&flow->key, &compat->key, sizeof(struct xflow_key)) ||
__get_user(actions, &compat->actions) ||
__get_user(flow->n_actions, &compat->n_actions) ||
__get_user(flow->flags, &compat->flags))
return 0;
}
-static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
+static int compat_put_flow(struct datapath *dp, struct compat_xflow_flow_put __user *ufp)
{
- struct odp_flow_stats stats;
- struct odp_flow_put fp;
+ struct xflow_flow_stats stats;
+ struct xflow_flow_put fp;
int error;
if (compat_get_flow(&fp.flow, &ufp->flow) ||
return error;
if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
+ sizeof(struct xflow_flow_stats)))
return -EFAULT;
return 0;
static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
struct timespec time_offset,
- struct compat_odp_flow __user *ufp)
+ struct compat_xflow_flow __user *ufp)
{
compat_uptr_t actions;
compat_ptr(actions), &ufp->n_actions);
}
-static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
+static int compat_del_flow(struct datapath *dp, struct compat_xflow_flow __user *ufp)
{
struct sw_flow *flow;
- struct odp_flow uf;
+ struct xflow_flow uf;
int error;
if (compat_get_flow(&uf, ufp))
return error;
}
-static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int compat_query_flows(struct datapath *dp, struct compat_xflow_flow *flows, u32 n_flows)
{
struct tbl *table = rcu_dereference(dp->table);
struct timespec time_offset;
time_offset = get_time_offset();
for (i = 0; i < n_flows; i++) {
- struct compat_odp_flow __user *ufp = &flows[i];
- struct odp_flow uf;
+ struct compat_xflow_flow __user *ufp = &flows[i];
+ struct xflow_flow uf;
struct tbl_node *flow_node;
int error;
if (compat_get_flow(&uf, ufp))
return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
if (!flow_node)
}
struct compat_list_flows_cbdata {
- struct compat_odp_flow __user *uflows;
+ struct compat_xflow_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
struct timespec time_offset;
{
struct sw_flow *flow = flow_cast(node);
struct compat_list_flows_cbdata *cbdata = cbdata_;
- struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
+ struct compat_xflow_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
int error;
if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
return 0;
}
-static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int compat_list_flows(struct datapath *dp, struct compat_xflow_flow *flows, u32 n_flows)
{
struct compat_list_flows_cbdata cbdata;
int error;
static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
int (*function)(struct datapath *,
- struct compat_odp_flow *,
+ struct compat_xflow_flow *,
u32 n_flows))
{
- struct compat_odp_flowvec __user *uflowvec;
- struct compat_odp_flow __user *flows;
- struct compat_odp_flowvec flowvec;
+ struct compat_xflow_flowvec __user *uflowvec;
+ struct compat_xflow_flow __user *flows;
+ struct compat_xflow_flowvec flowvec;
int retval;
uflowvec = compat_ptr(argp);
copy_from_user(&flowvec, uflowvec, sizeof flowvec))
return -EFAULT;
- if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
+ if (flowvec.n_flows > INT_MAX / sizeof(struct compat_xflow_flow))
return -EINVAL;
flows = compat_ptr(flowvec.flows);
if (!access_ok(VERIFY_WRITE, flows,
- flowvec.n_flows * sizeof(struct compat_odp_flow)))
+ flowvec.n_flows * sizeof(struct compat_xflow_flow)))
return -EFAULT;
retval = function(dp, flows, flowvec.n_flows);
: put_user(retval, &uflowvec->n_flows));
}
-static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
+static int compat_execute(struct datapath *dp, const struct compat_xflow_execute __user *uexecute)
{
- struct odp_execute execute;
+ struct xflow_execute execute;
compat_uptr_t actions;
compat_uptr_t data;
- if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
+ if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_xflow_execute)) ||
__get_user(execute.in_port, &uexecute->in_port) ||
__get_user(actions, &uexecute->actions) ||
__get_user(execute.n_actions, &uexecute->n_actions) ||
int err;
switch (cmd) {
- case ODP_DP_DESTROY:
- case ODP_FLOW_FLUSH:
+ case XFLOW_DP_DESTROY:
+ case XFLOW_FLOW_FLUSH:
/* Ioctls that don't need any translation at all. */
return openvswitch_ioctl(f, cmd, argp);
- case ODP_DP_CREATE:
- case ODP_PORT_ATTACH:
- case ODP_PORT_DETACH:
- case ODP_VPORT_DEL:
- case ODP_VPORT_MTU_SET:
- case ODP_VPORT_MTU_GET:
- case ODP_VPORT_ETHER_SET:
- case ODP_VPORT_ETHER_GET:
- case ODP_VPORT_STATS_SET:
- case ODP_VPORT_STATS_GET:
- case ODP_DP_STATS:
- case ODP_GET_DROP_FRAGS:
- case ODP_SET_DROP_FRAGS:
- case ODP_SET_LISTEN_MASK:
- case ODP_GET_LISTEN_MASK:
- case ODP_SET_SFLOW_PROBABILITY:
- case ODP_GET_SFLOW_PROBABILITY:
- case ODP_PORT_QUERY:
+ case XFLOW_DP_CREATE:
+ case XFLOW_PORT_ATTACH:
+ case XFLOW_PORT_DETACH:
+ case XFLOW_VPORT_DEL:
+ case XFLOW_VPORT_MTU_SET:
+ case XFLOW_VPORT_MTU_GET:
+ case XFLOW_VPORT_ETHER_SET:
+ case XFLOW_VPORT_ETHER_GET:
+ case XFLOW_VPORT_STATS_SET:
+ case XFLOW_VPORT_STATS_GET:
+ case XFLOW_DP_STATS:
+ case XFLOW_GET_DROP_FRAGS:
+ case XFLOW_SET_DROP_FRAGS:
+ case XFLOW_SET_LISTEN_MASK:
+ case XFLOW_GET_LISTEN_MASK:
+ case XFLOW_SET_SFLOW_PROBABILITY:
+ case XFLOW_GET_SFLOW_PROBABILITY:
+ case XFLOW_PORT_QUERY:
/* Ioctls that just need their pointer argument extended. */
return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
- case ODP_VPORT_ADD32:
+ case XFLOW_VPORT_ADD32:
return compat_vport_user_add(compat_ptr(argp));
- case ODP_VPORT_MOD32:
+ case XFLOW_VPORT_MOD32:
return compat_vport_user_mod(compat_ptr(argp));
}
goto exit;
switch (cmd) {
- case ODP_PORT_LIST32:
+ case XFLOW_PORT_LIST32:
err = compat_list_ports(dp, compat_ptr(argp));
break;
- case ODP_PORT_GROUP_SET32:
+ case XFLOW_PORT_GROUP_SET32:
err = compat_set_port_group(dp, compat_ptr(argp));
break;
- case ODP_PORT_GROUP_GET32:
+ case XFLOW_PORT_GROUP_GET32:
err = compat_get_port_group(dp, compat_ptr(argp));
break;
- case ODP_FLOW_PUT32:
+ case XFLOW_FLOW_PUT32:
err = compat_put_flow(dp, compat_ptr(argp));
break;
- case ODP_FLOW_DEL32:
+ case XFLOW_FLOW_DEL32:
err = compat_del_flow(dp, compat_ptr(argp));
break;
- case ODP_FLOW_GET32:
+ case XFLOW_FLOW_GET32:
err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
break;
- case ODP_FLOW_LIST32:
+ case XFLOW_FLOW_LIST32:
err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
break;
- case ODP_EXECUTE32:
+ case XFLOW_EXECUTE32:
err = compat_execute(dp, compat_ptr(argp));
break;
* @waitqueue: Waitqueue, for waiting for new packets in @queues.
* @n_flows: Number of flows currently in flow table.
* @table: Current flow table (RCU protected).
- * @groups: Port groups, used by ODPAT_OUTPUT_GROUP action (RCU protected).
+ * @groups: Port groups, used by XFLOWAT_OUTPUT_GROUP action (RCU protected).
* @n_ports: Number of ports currently in @ports.
- * @ports: Map from port number to &struct dp_port. %ODPP_LOCAL port
+ * @ports: Map from port number to &struct dp_port. %XFLOWP_LOCAL port
* always exists, other ports may be %NULL.
* @port_list: List of all ports in @ports in arbitrary order.
* @stats_percpu: Per-CPU datapath statistics.
* @sflow_probability: Number of packets out of UINT_MAX to sample to the
- * %ODPL_SFLOW queue, e.g. (@sflow_probability/UINT_MAX) is the probability of
+ * %XFLOWL_SFLOW queue, e.g. (@sflow_probability/UINT_MAX) is the probability of
* sampling a given packet.
*/
struct datapath {
/**
* struct ovs_skb_cb - OVS data in skb CB
* @dp_port: The datapath port on which the skb entered the switch.
+ * @flow: The flow associated with this packet. May be %NULL if no flow.
* @ip_summed: Consistently stores L4 checksumming status across different
* kernel versions.
* @tun_id: ID (in network byte order) of the tunnel that encapsulated this
* packet. It is 0 if the packet was not received on a tunnel.
- * @is_frag: %true if this packet is an IPv4 fragment, %false otherwise.
*/
struct ovs_skb_cb {
struct dp_port *dp_port;
+ struct sw_flow *flow;
enum csum_type ip_summed;
__be32 tun_id;
- bool is_frag;
};
#define OVS_CB(skb) ((struct ovs_skb_cb *)(skb)->cb)
{
struct sw_flow_actions *sfa;
- if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union xflow_action))
+ /* At least DP_MAX_PORTS actions are required to be able to flood a
+ * packet to every port. Factor of 2 allows for setting VLAN tags,
+ * etc. */
+ if (n_actions > 2 * DP_MAX_PORTS)
return ERR_PTR(-EINVAL);
- sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
+ sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union xflow_action),
GFP_KERNEL);
if (!sfa)
return ERR_PTR(-ENOMEM);
return sfa;
}
-
- /* Frees 'flow' immediately. */
- static void flow_free(struct sw_flow *flow)
+ struct sw_flow *flow_alloc(void)
{
- if (unlikely(!flow))
- return;
- kfree(flow->sf_acts);
- kmem_cache_free(flow_cache, flow);
+ struct sw_flow *flow;
+
+ flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
+ if (!flow)
+ return ERR_PTR(-ENOMEM);
+
+ spin_lock_init(&flow->lock);
+ atomic_set(&flow->refcnt, 1);
+ flow->dead = false;
+
+ return flow;
}
void flow_free_tbl(struct tbl_node *node)
{
struct sw_flow *flow = flow_cast(node);
- flow_free(flow);
+
+ flow->dead = true;
+ flow_put(flow);
}
/* RCU callback used by flow_deferred_free. */
static void rcu_free_flow_callback(struct rcu_head *rcu)
{
struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
- flow_free(flow);
+
+ flow->dead = true;
+ flow_put(flow);
}
/* Schedules 'flow' to be freed after the next RCU grace period.
call_rcu(&flow->rcu, rcu_free_flow_callback);
}
+ void flow_hold(struct sw_flow *flow)
+ {
+ atomic_inc(&flow->refcnt);
+ }
+
+ void flow_put(struct sw_flow *flow)
+ {
+ if (unlikely(!flow))
+ return;
+
+ if (atomic_dec_and_test(&flow->refcnt)) {
+ kfree(flow->sf_acts);
+ kmem_cache_free(flow_cache, flow);
+ }
+ }
+
/* RCU callback used by flow_deferred_free_acts. */
static void rcu_free_acts_callback(struct rcu_head *rcu)
{
call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
}
-static void parse_vlan(struct sk_buff *skb, struct odp_flow_key *key)
+static void parse_vlan(struct sk_buff *skb, struct xflow_key *key)
{
struct qtag_prefix {
__be16 eth_type; /* ETH_P_8021Q */
return;
qp = (struct qtag_prefix *) skb->data;
- key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
- key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
+ key->dl_tci = qp->tci | htons(XFLOW_TCI_PRESENT);
__skb_pull(skb, sizeof(struct qtag_prefix));
}
proto = *(__be16 *) skb->data;
__skb_pull(skb, sizeof(__be16));
- if (ntohs(proto) >= ODP_DL_TYPE_ETH2_CUTOFF)
+ if (ntohs(proto) >= XFLOW_DL_TYPE_ETH2_CUTOFF)
return proto;
if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
- return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
+ return htons(XFLOW_DL_TYPE_NOT_ETH_TYPE);
llc = (struct llc_snap_hdr *) skb->data;
if (llc->dsap != LLC_SAP_SNAP ||
llc->ssap != LLC_SAP_SNAP ||
(llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
- return htons(ODP_DL_TYPE_NOT_ETH_TYPE);
+ return htons(XFLOW_DL_TYPE_NOT_ETH_TYPE);
__skb_pull(skb, sizeof(struct llc_snap_hdr));
return llc->ethertype;
* Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
* %false.
*/
- int flow_extract(struct sk_buff *skb, u16 in_port, struct xflow_key *key)
-int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key,
++int flow_extract(struct sk_buff *skb, u16 in_port, struct xflow_key *key,
+ bool *is_frag)
{
struct ethhdr *eth;
memset(key, 0, sizeof *key);
key->tun_id = OVS_CB(skb)->tun_id;
key->in_port = in_port;
- OVS_CB(skb)->is_frag = false;
- key->dl_vlan = htons(ODP_VLAN_NONE);
+ *is_frag = false;
/*
* We would really like to pull as many bytes as we could possibly
memcpy(key->dl_src, eth->h_source, ETH_ALEN);
memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
- /* dl_type, dl_vlan, dl_vlan_pcp. */
+ /* dl_type, dl_tci. */
__skb_pull(skb, 2 * ETH_ALEN);
if (eth->h_proto == htons(ETH_P_8021Q))
parse_vlan(skb, key);
key->tp_dst = htons(icmp->code);
}
}
- } else {
- OVS_CB(skb)->is_frag = true;
- }
+ } else
+ *is_frag = true;
+
} else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
struct arp_eth_header *arp;
&& arp->ar_pln == 4) {
/* We only match on the lower 8 bits of the opcode. */
- if (ntohs(arp->ar_op) <= 0xff) {
+ if (ntohs(arp->ar_op) <= 0xff)
key->nw_proto = ntohs(arp->ar_op);
- }
if (key->nw_proto == ARPOP_REQUEST
|| key->nw_proto == ARPOP_REPLY) {
return 0;
}
-u32 flow_hash(const struct odp_flow_key *key)
+u32 flow_hash(const struct xflow_key *key)
{
return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
}
int flow_cmp(const struct tbl_node *node, void *key2_)
{
- const struct odp_flow_key *key1 = &flow_cast(node)->key;
- const struct odp_flow_key *key2 = key2_;
+ const struct xflow_key *key1 = &flow_cast(node)->key;
+ const struct xflow_key *key2 = key2_;
- return !memcmp(key1, key2, sizeof(struct odp_flow_key));
+ return !memcmp(key1, key2, sizeof(struct xflow_key));
}
/* Initializes the flow module.
#include <linux/jiffies.h>
#include <linux/time.h>
-#include "openvswitch/datapath-protocol.h"
+#include "openvswitch/xflow.h"
#include "table.h"
struct sk_buff;
struct sw_flow_actions {
struct rcu_head rcu;
unsigned int n_actions;
- union odp_action actions[];
+ union xflow_action actions[];
};
struct sw_flow {
struct rcu_head rcu;
struct tbl_node tbl_node;
- struct odp_flow_key key;
+ struct xflow_key key;
struct sw_flow_actions *sf_acts;
+ atomic_t refcnt;
+ bool dead;
+
spinlock_t lock; /* Lock for values below. */
unsigned long used; /* Last used time (in jiffies). */
u64 packet_count; /* Number of packets matched. */
unsigned char ar_tip[4]; /* target IP address */
} __attribute__((packed));
- extern struct kmem_cache *flow_cache;
+ int flow_init(void);
+ void flow_exit(void);
- struct sw_flow_actions *flow_actions_alloc(size_t n_actions);
+ struct sw_flow *flow_alloc(void);
void flow_deferred_free(struct sw_flow *);
+ void flow_free_tbl(struct tbl_node *);
+
+ struct sw_flow_actions *flow_actions_alloc(size_t n_actions);
void flow_deferred_free_acts(struct sw_flow_actions *);
- int flow_extract(struct sk_buff *, u16 in_port, struct xflow_key *);
+
+ void flow_hold(struct sw_flow *);
+ void flow_put(struct sw_flow *);
+
-int flow_extract(struct sk_buff *, u16 in_port, struct odp_flow_key *, bool *is_frag);
++int flow_extract(struct sk_buff *, u16 in_port, struct xflow_key *, bool *is_frag);
void flow_used(struct sw_flow *, struct sk_buff *);
-u32 flow_hash(const struct odp_flow_key *key);
+u32 flow_hash(const struct xflow_key *key);
int flow_cmp(const struct tbl_node *, void *target);
- void flow_free_tbl(struct tbl_node *);
-
- int flow_init(void);
- void flow_exit(void);
static inline struct sw_flow *flow_cast(const struct tbl_node *node)
{
#include <linux/jhash.h>
#include <linux/kernel.h>
#include <linux/version.h>
+ #include <linux/workqueue.h>
#include <net/dsfield.h>
#include <net/dst.h>
#include "tunnel.h"
#include "vport.h"
#include "vport-generic.h"
+ #include "vport-internal_dev.h"
+
+ #ifdef NEED_CACHE_TIMEOUT
+ /*
+ * On kernels where we can't quickly detect changes in the rest of the system
+ * we use an expiration time to invalidate the cache. A shorter expiration
+ * reduces the length of time that we may potentially blackhole packets while
+ * a longer time increases performance by reducing the frequency that the
+ * cache needs to be rebuilt. A variety of factors may cause the cache to be
+ * invalidated before the expiration time but this is the maximum. The time
+ * is expressed in jiffies.
+ */
+ #define MAX_CACHE_EXP HZ
+ #endif
+
+ /*
+ * Interval to check for and remove caches that are no longer valid. Caches
+ * are checked for validity before they are used for packet encapsulation and
+ * old caches are removed at that time. However, if no packets are sent through
+ * the tunnel then the cache will never be destroyed. Since it holds
+ * references to a number of system objects, the cache will continue to use
+ * system resources by not allowing those objects to be destroyed. The cache
+ * cleaner is periodically run to free invalid caches. It does not
+ * significantly affect system performance. A lower interval will release
+ * resources faster but will itself consume resources by requiring more frequent
+ * checks. A longer interval may result in messages being printed to the kernel
+ * message buffer about unreleased resources. The interval is expressed in
+ * jiffies.
+ */
+ #define CACHE_CLEANER_INTERVAL (5 * HZ)
+
+ #define CACHE_DATA_ALIGN 16
/* Protected by RCU. */
static struct tbl *port_table;
+ static void cache_cleaner(struct work_struct *work);
+ DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
+
/*
* These are just used as an optimization: they don't require any kind of
* synchronization because we could have just as easily read the value before
return container_of(node, struct tnl_vport, tbl_node);
}
- /* RCU callback. */
- static void free_config(struct rcu_head *rcu)
+ static inline void schedule_cache_cleaner(void)
+ {
+ schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
+ }
+
+ static void free_cache(struct tnl_cache *cache)
+ {
+ if (!cache)
+ return;
+
+ flow_put(cache->flow);
+ ip_rt_put(cache->rt);
+ kfree(cache);
+ }
+
+ static void free_config_rcu(struct rcu_head *rcu)
{
struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
kfree(c);
}
+ static void free_cache_rcu(struct rcu_head *rcu)
+ {
+ struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
+ free_cache(c);
+ }
+
static void assign_config_rcu(struct vport *vport,
struct tnl_mutable_config *new_config)
{
struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
struct tnl_mutable_config *old_config;
- old_config = rcu_dereference(tnl_vport->mutable);
+ old_config = tnl_vport->mutable;
rcu_assign_pointer(tnl_vport->mutable, new_config);
- call_rcu(&old_config->rcu, free_config);
+ call_rcu(&old_config->rcu, free_config_rcu);
+ }
+
+ static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
+ {
+ struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ struct tnl_cache *old_cache;
+
+ old_cache = tnl_vport->cache;
+ rcu_assign_pointer(tnl_vport->cache, new_cache);
+
+ if (old_cache)
+ call_rcu(&old_cache->rcu, free_cache_rcu);
}
static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
return jhash2(lookup->vals, ARRAY_SIZE(lookup->vals), 0);
}
+ static u32 mutable_hash(const struct tnl_mutable_config *mutable)
+ {
+ struct port_lookup_key lookup;
+
+ lookup.vals[LOOKUP_SADDR] = mutable->port_config.saddr;
+ lookup.vals[LOOKUP_DADDR] = mutable->port_config.daddr;
+ lookup.vals[LOOKUP_KEY] = mutable->port_config.in_key;
+ lookup.vals[LOOKUP_TUNNEL_TYPE] = mutable->tunnel_type;
+
+ return port_hash(&lookup);
+ }
+
+ static void check_table_empty(void)
+ {
+ if (tbl_count(port_table) == 0) {
+ struct tbl *old_table = port_table;
+
+ cancel_delayed_work_sync(&cache_cleaner_wq);
+ rcu_assign_pointer(port_table, NULL);
+ tbl_deferred_destroy(old_table, NULL);
+ }
+ }
+
static int add_port(struct vport *vport)
{
struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
- struct port_lookup_key lookup;
int err;
if (!port_table) {
return -ENOMEM;
rcu_assign_pointer(port_table, new_table);
+ schedule_cache_cleaner();
} else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
struct tbl *old_table = port_table;
tbl_deferred_destroy(old_table, NULL);
}
- lookup.vals[LOOKUP_SADDR] = tnl_vport->mutable->port_config.saddr;
- lookup.vals[LOOKUP_DADDR] = tnl_vport->mutable->port_config.daddr;
- lookup.vals[LOOKUP_KEY] = tnl_vport->mutable->port_config.in_key;
- lookup.vals[LOOKUP_TUNNEL_TYPE] = tnl_vport->mutable->tunnel_type;
+ err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable));
+ if (err) {
+ check_table_empty();
+ return err;
+ }
+
+ (*find_port_pool(tnl_vport->mutable))++;
+
+ return 0;
+ }
+
+ static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
+ {
+ int err;
+ struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ u32 hash;
+
+ hash = mutable_hash(new_mutable);
+ if (hash == tnl_vport->tbl_node.hash)
+ goto table_updated;
- err = tbl_insert(port_table, &tnl_vport->tbl_node, port_hash(&lookup));
+ /*
+ * Ideally we should make this move atomic to avoid having gaps in
+ * finding tunnels or the possibility of failure. However, if we do
+ * find a tunnel it will always be consistent.
+ */
+ err = tbl_remove(port_table, &tnl_vport->tbl_node);
if (err)
return err;
- (*find_port_pool(tnl_vport->mutable))++;
+ err = tbl_insert(port_table, &tnl_vport->tbl_node, hash);
+ if (err) {
+ check_table_empty();
+ return err;
+ }
+
+ table_updated:
+ assign_config_rcu(vport, new_mutable);
return 0;
}
if (err)
return err;
+ check_table_empty();
(*find_port_pool(tnl_vport->mutable))--;
return 0;
struct tbl *table = rcu_dereference(port_table);
struct tbl_node *tbl_node;
- if (!table)
+ if (unlikely(!table))
return NULL;
lookup.vals[LOOKUP_SADDR] = saddr;
return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
}
+ static inline void ecn_decapsulate(struct sk_buff *skb)
+ {
+ u8 tos = ip_hdr(skb)->tos;
+
+ if (INET_ECN_is_ce(tos)) {
+ __be16 protocol = skb->protocol;
+ unsigned int nw_header = skb_network_offset(skb);
+
+ if (skb->protocol == htons(ETH_P_8021Q)) {
+ if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
+ return;
+
+ protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
+ nw_header += VLAN_HLEN;
+ }
+
+ if (protocol == htons(ETH_P_IP)) {
+ if (unlikely(!pskb_may_pull(skb, nw_header
+ + sizeof(struct iphdr))))
+ return;
+
+ IP_ECN_set_ce((struct iphdr *)(skb->data + nw_header));
+ }
+ #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+ else if (protocol == htons(ETH_P_IPV6)) {
+ if (unlikely(!pskb_may_pull(skb, nw_header
+ + sizeof(struct ipv6hdr))))
+ return;
+
+ IP6_ECN_set_ce((struct ipv6hdr *)(skb->data + nw_header));
+ }
+ #endif
+ }
+ }
+
+ /* Called with rcu_read_lock. */
+ void tnl_rcv(struct vport *vport, struct sk_buff *skb)
+ {
+ skb->pkt_type = PACKET_HOST;
+ skb->protocol = eth_type_trans(skb, skb->dev);
+
+ skb_dst_drop(skb);
+ nf_reset(skb);
+ secpath_reset(skb);
+ skb_reset_network_header(skb);
+
+ ecn_decapsulate(skb);
+
+ skb_push(skb, ETH_HLEN);
+ compute_ip_summed(skb, false);
+
+ vport_receive(vport, skb);
+ }
+
static bool check_ipv4_address(__be32 addr)
{
if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
return true;
}
- static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
+ static bool check_mtu(struct sk_buff *skb,
+ struct vport *vport,
+ const struct tnl_mutable_config *mutable,
+ const struct rtable *rt, __be16 *frag_offp)
{
- if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
- struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
- if (unlikely(!nskb)) {
- kfree_skb(skb);
- return ERR_PTR(-ENOMEM);
+ int mtu;
+ __be16 frag_off;
+
+ frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
+ if (frag_off)
+ mtu = dst_mtu(&rt_dst(rt))
+ - ETH_HLEN
+ - mutable->tunnel_hlen
+ - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
+ else
+ mtu = mutable->mtu;
+
+ if (skb->protocol == htons(ETH_P_IP)) {
+ struct iphdr *old_iph = ip_hdr(skb);
+
+ frag_off |= old_iph->frag_off & htons(IP_DF);
+ mtu = max(mtu, IP_MIN_MTU);
+
+ if ((old_iph->frag_off & htons(IP_DF)) &&
+ mtu < ntohs(old_iph->tot_len)) {
+ if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
+ goto drop;
}
+ }
+ #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+ else if (skb->protocol == htons(ETH_P_IPV6)) {
+ unsigned int packet_length = skb->len - ETH_HLEN
+ - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
- set_skb_csum_bits(skb, nskb);
+ mtu = max(mtu, IPV6_MIN_MTU);
- if (skb->sk)
- skb_set_owner_w(nskb, skb->sk);
+ /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
+ if (packet_length > IPV6_MIN_MTU)
+ frag_off = htons(IP_DF);
- dev_kfree_skb(skb);
- return nskb;
+ if (mtu < packet_length) {
+ if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
+ goto drop;
+ }
}
+ #endif
- return skb;
+ *frag_offp = frag_off;
+ return true;
+
+ drop:
+ *frag_offp = 0;
+ return false;
+ }
+
+ static void create_tunnel_header(const struct vport *vport,
+ const struct tnl_mutable_config *mutable,
+ const struct rtable *rt, void *header)
+ {
+ struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ struct iphdr *iph = header;
+
+ iph->version = 4;
+ iph->ihl = sizeof(struct iphdr) >> 2;
+ iph->frag_off = htons(IP_DF);
+ iph->protocol = tnl_vport->tnl_ops->ipproto;
+ iph->tos = mutable->port_config.tos;
+ iph->daddr = rt->rt_dst;
+ iph->saddr = rt->rt_src;
+ iph->ttl = mutable->port_config.ttl;
+ if (!iph->ttl)
+ iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
+
+ tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
}
- static inline u8 ecn_encapsulate(u8 tos, struct sk_buff *skb)
+ static inline void *get_cached_header(const struct tnl_cache *cache)
{
- u8 inner;
+ return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
+ }
- if (skb->protocol == htons(ETH_P_IP))
- inner = ((struct iphdr *)skb_network_header(skb))->tos;
- #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (skb->protocol == htons(ETH_P_IPV6))
- inner = ipv6_get_dsfield((struct ipv6hdr *)skb_network_header(skb));
+ static inline bool check_cache_valid(const struct tnl_cache *cache,
+ const struct tnl_mutable_config *mutable)
+ {
+ return cache &&
+ #ifdef NEED_CACHE_TIMEOUT
+ time_before(jiffies, cache->expiration) &&
#endif
- else
- inner = 0;
-
- return INET_ECN_encapsulate(tos, inner);
+ #ifdef HAVE_RT_GENID
+ atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
+ #endif
+ #ifdef HAVE_HH_SEQ
+ rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
+ #endif
+ mutable->seq == cache->mutable_seq &&
+ (!is_internal_dev(rt_dst(cache->rt).dev) ||
+ (cache->flow && !cache->flow->dead));
}
- static inline void ecn_decapsulate(struct sk_buff *skb)
+ static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
{
- u8 tos = ip_hdr(skb)->tos;
+ struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
+ const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
+ const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
- if (INET_ECN_is_ce(tos)) {
- __be16 protocol = skb->protocol;
- unsigned int nw_header = skb_network_header(skb) - skb->data;
+ if (cache && !check_cache_valid(cache, mutable) &&
+ spin_trylock_bh(&tnl_vport->cache_lock)) {
+ assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
+ spin_unlock_bh(&tnl_vport->cache_lock);
+ }
- if (skb->protocol == htons(ETH_P_8021Q)) {
- if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
- return;
+ return 0;
+ }
- protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
- nw_header += VLAN_HLEN;
- }
+ static void cache_cleaner(struct work_struct *work)
+ {
+ schedule_cache_cleaner();
- if (protocol == htons(ETH_P_IP)) {
- if (unlikely(!pskb_may_pull(skb, nw_header
- + sizeof(struct iphdr))))
- return;
+ rcu_read_lock();
+ tbl_foreach(port_table, cache_cleaner_cb, NULL);
+ rcu_read_unlock();
+ }
- IP_ECN_set_ce((struct iphdr *)(nw_header + skb->data));
- }
- #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (protocol == htons(ETH_P_IPV6)) {
- if (unlikely(!pskb_may_pull(skb, nw_header
- + sizeof(struct ipv6hdr))))
- return;
+ static inline void create_eth_hdr(struct tnl_cache *cache,
+ const struct rtable *rt)
+ {
+ void *cache_data = get_cached_header(cache);
+ int hh_len = rt_dst(rt).hh->hh_len;
+ int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
- IP6_ECN_set_ce((struct ipv6hdr *)(nw_header
- + skb->data));
- }
+ #ifdef HAVE_HH_SEQ
+ unsigned hh_seq;
+
+ do {
+ hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
+ memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
+ } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
+
+ cache->hh_seq = hh_seq;
+ #else
+ read_lock_bh(&rt_dst(rt).hh->hh_lock);
+ memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
+ read_unlock_bh(&rt_dst(rt).hh->hh_lock);
#endif
- }
}
- static struct sk_buff *handle_gso(struct sk_buff *skb)
+ static struct tnl_cache *build_cache(struct vport *vport,
+ const struct tnl_mutable_config *mutable,
+ struct rtable *rt)
{
- if (skb_is_gso(skb)) {
- struct sk_buff *nskb = skb_gso_segment(skb, 0);
+ struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ struct tnl_cache *cache;
+ void *cache_data;
+ int cache_len;
- dev_kfree_skb(skb);
- return nskb;
+ if (!(mutable->port_config.flags & TNL_F_HDR_CACHE))
+ return NULL;
+
+ /*
+ * If there is no entry in the ARP cache or if this device does not
+ * support hard header caching just fall back to the IP stack.
+ */
+ if (!rt_dst(rt).hh)
+ return NULL;
+
+ /*
+ * If lock is contended fall back to directly building the header.
+ * We're not going to help performance by sitting here spinning.
+ */
+ if (!spin_trylock_bh(&tnl_vport->cache_lock))
+ return NULL;
+
+ cache = tnl_vport->cache;
+ if (check_cache_valid(cache, mutable))
+ goto unlock;
+ else
+ cache = NULL;
+
+ cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
+
+ cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
+ cache_len, GFP_ATOMIC);
+ if (!cache)
+ goto unlock;
+
+ cache->len = cache_len;
+
+ create_eth_hdr(cache, rt);
+ cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
+
+ create_tunnel_header(vport, mutable, rt, cache_data);
+
+ cache->mutable_seq = mutable->seq;
+ cache->rt = rt;
+ #ifdef NEED_CACHE_TIMEOUT
+ cache->expiration = jiffies + tnl_vport->cache_exp_interval;
+ #endif
+
+ if (is_internal_dev(rt_dst(rt).dev)) {
+ int err;
+ struct vport *vport;
+ struct dp_port *dp_port;
+ struct sk_buff *skb;
+ bool is_frag;
- struct odp_flow_key flow_key;
++ struct xflow_key flow_key;
+ struct tbl_node *flow_node;
+
+ vport = internal_dev_get_vport(rt_dst(rt).dev);
+ if (!vport)
+ goto done;
+
+ dp_port = vport_get_dp_port(vport);
+ if (!dp_port)
+ goto done;
+
+ skb = alloc_skb(cache->len, GFP_ATOMIC);
+ if (!skb)
+ goto done;
+
+ __skb_put(skb, cache->len);
+ memcpy(skb->data, get_cached_header(cache), cache->len);
+
+ err = flow_extract(skb, dp_port->port_no, &flow_key, &is_frag);
+
+ kfree_skb(skb);
+ if (err || is_frag)
+ goto done;
+
+ flow_node = tbl_lookup(rcu_dereference(dp_port->dp->table),
+ &flow_key, flow_hash(&flow_key),
+ flow_cmp);
+ if (flow_node) {
+ struct sw_flow *flow = flow_cast(flow_node);
+
+ cache->flow = flow;
+ flow_hold(flow);
+ }
}
- return skb;
+ done:
+ assign_cache_rcu(vport, cache);
+
+ unlock:
+ spin_unlock_bh(&tnl_vport->cache_lock);
+
+ return cache;
}
- static int handle_csum_offload(struct sk_buff *skb)
+ static struct rtable *find_route(struct vport *vport,
+ const struct tnl_mutable_config *mutable,
+ u8 tos, struct tnl_cache **cache)
{
- if (skb->ip_summed == CHECKSUM_PARTIAL)
- return skb_checksum_help(skb);
- else {
- skb->ip_summed = CHECKSUM_NONE;
- return 0;
+ struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
+
+ *cache = NULL;
+ tos = RT_TOS(tos);
+
+ if (likely(tos == mutable->port_config.tos &&
+ check_cache_valid(cur_cache, mutable))) {
+ *cache = cur_cache;
+ return cur_cache->rt;
+ } else {
+ struct rtable *rt;
+ struct flowi fl = { .nl_u = { .ip4_u =
+ { .daddr = mutable->port_config.daddr,
+ .saddr = mutable->port_config.saddr,
+ .tos = tos } },
+ .proto = tnl_vport->tnl_ops->ipproto };
+
+ if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
+ return NULL;
+
+ if (likely(tos == mutable->port_config.tos))
+ *cache = build_cache(vport, mutable, rt);
+
+ return rt;
}
}
- /* Called with rcu_read_lock. */
- void tnl_rcv(struct vport *vport, struct sk_buff *skb)
+ static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
{
- skb->pkt_type = PACKET_HOST;
- skb->protocol = eth_type_trans(skb, skb->dev);
+ if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
+ struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
+ if (unlikely(!nskb)) {
+ kfree_skb(skb);
+ return ERR_PTR(-ENOMEM);
+ }
- skb_dst_drop(skb);
- nf_reset(skb);
- secpath_reset(skb);
- skb_reset_network_header(skb);
+ set_skb_csum_bits(skb, nskb);
- ecn_decapsulate(skb);
+ if (skb->sk)
+ skb_set_owner_w(nskb, skb->sk);
- skb_push(skb, ETH_HLEN);
- compute_ip_summed(skb, false);
+ kfree_skb(skb);
+ return nskb;
+ }
- vport_receive(vport, skb);
+ return skb;
}
- static int build_packet(struct vport *vport, const struct tnl_mutable_config *mutable,
- struct iphdr *iph, struct rtable *rt, int max_headroom,
- int mtu, struct sk_buff *skb)
+ static inline bool need_linearize(const struct sk_buff *skb)
{
- struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
+ int i;
+
+ if (unlikely(skb_shinfo(skb)->frag_list))
+ return true;
+
+ /*
+ * Generally speaking we should linearize if there are paged frags.
+ * However, if all of the refcounts are 1 we know nobody else can
+ * change them from underneath us and we can skip the linearization.
+ */
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
+ if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1))
+ return true;
+
+ return false;
+ }
+
+ static struct sk_buff *handle_offloads(struct sk_buff *skb,
+ const struct tnl_mutable_config *mutable,
+ const struct rtable *rt)
+ {
+ int min_headroom;
int err;
- struct iphdr *new_iph;
- int orig_len = skb->len;
- __be16 frag_off = iph->frag_off;
- skb = check_headroom(skb, max_headroom);
- if (unlikely(IS_ERR(skb)))
- goto error;
+ forward_ip_summed(skb);
- err = handle_csum_offload(skb);
+ err = vswitch_skb_checksum_setup(skb);
if (unlikely(err))
goto error_free;
- if (skb->protocol == htons(ETH_P_IP)) {
- struct iphdr *old_iph = ip_hdr(skb);
+ min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
+ + mutable->tunnel_hlen;
- if ((old_iph->frag_off & htons(IP_DF)) &&
- mtu < ntohs(old_iph->tot_len)) {
- if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
- goto error_free;
+ if (skb_is_gso(skb)) {
+ struct sk_buff *nskb;
+
+ /*
+ * If we are doing GSO on a pskb it is better to make sure that
+ * the headroom is correct now. We will only have to copy the
+ * portion in the linear data area and GSO will preserve
+ * headroom when it creates the segments. This is particularly
+ * beneficial on Xen where we get a lot of GSO pskbs.
+ * Conversely, we avoid copying if it is just to get our own
+ * writable clone because GSO will do the copy for us.
+ */
+ if (skb_headroom(skb) < min_headroom) {
+ skb = check_headroom(skb, min_headroom);
+ if (unlikely(IS_ERR(skb))) {
+ err = PTR_ERR(skb);
+ goto error;
+ }
}
- }
- #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (skb->protocol == htons(ETH_P_IPV6)) {
- unsigned int packet_length = skb->len - ETH_HLEN
- - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
+ nskb = skb_gso_segment(skb, 0);
+ kfree_skb(skb);
+ if (unlikely(IS_ERR(nskb))) {
+ err = PTR_ERR(nskb);
+ goto error;
+ }
- /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
- if (packet_length > IPV6_MIN_MTU)
- frag_off = htons(IP_DF);
+ skb = nskb;
+ } else {
+ skb = check_headroom(skb, min_headroom);
+ if (unlikely(IS_ERR(skb))) {
+ err = PTR_ERR(skb);
+ goto error;
+ }
- if (mtu < packet_length) {
- if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
+ if (skb->ip_summed == CHECKSUM_PARTIAL) {
+ /*
+ * Pages aren't locked and could change at any time.
+ * If this happens after we compute the checksum, the
+ * checksum will be wrong. We linearize now to avoid
+ * this problem.
+ */
+ if (unlikely(need_linearize(skb))) {
+ err = __skb_linearize(skb);
+ if (unlikely(err))
+ goto error_free;
+ }
+
+ err = skb_checksum_help(skb);
+ if (unlikely(err))
goto error_free;
- }
+ } else if (skb->ip_summed == CHECKSUM_COMPLETE)
+ skb->ip_summed = CHECKSUM_NONE;
}
- #endif
- new_iph = (struct iphdr *)skb_push(skb, mutable->tunnel_hlen);
- skb_reset_network_header(skb);
- skb_set_transport_header(skb, sizeof(struct iphdr));
-
- memcpy(new_iph, iph, sizeof(struct iphdr));
- new_iph->frag_off = frag_off;
- ip_select_ident(new_iph, &rt_dst(rt), NULL);
+ return skb;
- memset(&IPCB(skb)->opt, 0, sizeof(IPCB(skb)->opt));
- IPCB(skb)->flags = 0;
+ error_free:
+ kfree_skb(skb);
+ error:
+ return ERR_PTR(err);
+ }
- skb = tnl_vport->tnl_ops->build_header(skb, vport, mutable, &rt_dst(rt));
- if (unlikely(!skb))
- goto error;
+ static int send_frags(struct sk_buff *skb,
+ const struct tnl_mutable_config *mutable)
+ {
+ int sent_len;
+ int err;
+ sent_len = 0;
while (skb) {
struct sk_buff *next = skb->next;
int frag_len = skb->len - mutable->tunnel_hlen;
skb->next = NULL;
err = ip_local_out(skb);
- if (unlikely(net_xmit_eval(err) != 0)) {
- orig_len -= frag_len;
+ if (likely(net_xmit_eval(err) == 0))
+ sent_len += frag_len;
+ else {
skb = next;
goto free_frags;
}
skb = next;
- };
+ }
- return orig_len;
+ return sent_len;
- error_free:
- kfree_skb(skb);
- error:
- return 0;
free_frags:
/*
* There's no point in continuing to send fragments once one has been
* dropped so just free the rest. This may help improve the congestion
* that caused the first packet to be dropped.
*/
- while (skb) {
- struct sk_buff *next = skb->next;
- orig_len -= skb->len - mutable->tunnel_hlen;
- kfree_skb(skb);
- skb = next;
- };
- return orig_len;
+ tnl_free_linked_skbs(skb);
+ return sent_len;
}
int tnl_send(struct vport *vport, struct sk_buff *skb)
struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
- struct iphdr *old_iph;
- int orig_len;
- struct iphdr iph;
+ enum vport_err_type err = VPORT_E_TX_ERROR;
struct rtable *rt;
- int max_headroom;
- int mtu;
+ struct dst_entry *unattached_dst = NULL;
+ struct tnl_cache *cache;
+ int sent_len = 0;
+ __be16 frag_off;
+ u8 ttl;
+ u8 inner_tos;
+ u8 tos;
/* Validate the protocol headers before we try to use them. */
if (skb->protocol == htons(ETH_P_8021Q)) {
}
if (skb->protocol == htons(ETH_P_IP)) {
- if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
- + sizeof(struct iphdr) - skb->data)))
+ if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
+ + sizeof(struct iphdr))))
skb->protocol = 0;
}
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
else if (skb->protocol == htons(ETH_P_IPV6)) {
- if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
- + sizeof(struct ipv6hdr) - skb->data)))
+ if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
+ + sizeof(struct ipv6hdr))))
skb->protocol = 0;
}
#endif
- old_iph = ip_hdr(skb);
-
- iph.tos = mutable->port_config.tos;
- if (mutable->port_config.flags & TNL_F_TOS_INHERIT) {
- if (skb->protocol == htons(ETH_P_IP))
- iph.tos = old_iph->tos;
- #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (skb->protocol == htons(ETH_P_IPV6))
- iph.tos = ipv6_get_dsfield(ipv6_hdr(skb));
- #endif
- }
- iph.tos = ecn_encapsulate(iph.tos, skb);
-
- {
- struct flowi fl = { .nl_u = { .ip4_u =
- { .daddr = mutable->port_config.daddr,
- .saddr = mutable->port_config.saddr,
- .tos = RT_TOS(iph.tos) } },
- .proto = tnl_vport->tnl_ops->ipproto };
-
- if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
- goto error_free;
- }
- iph.ttl = mutable->port_config.ttl;
- if (mutable->port_config.flags & TNL_F_TTL_INHERIT) {
- if (skb->protocol == htons(ETH_P_IP))
- iph.ttl = old_iph->ttl;
+ /* ToS */
+ if (skb->protocol == htons(ETH_P_IP))
+ inner_tos = ip_hdr(skb)->tos;
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (skb->protocol == htons(ETH_P_IPV6))
- iph.ttl = ipv6_hdr(skb)->hop_limit;
+ else if (skb->protocol == htons(ETH_P_IPV6))
+ inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
#endif
- }
- if (!iph.ttl)
- iph.ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
+ else
+ inner_tos = 0;
- iph.frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
- if (iph.frag_off)
- mtu = dst_mtu(&rt_dst(rt))
- - ETH_HLEN
- - mutable->tunnel_hlen
- - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
+ if (mutable->port_config.flags & TNL_F_TOS_INHERIT)
+ tos = inner_tos;
else
- mtu = mutable->mtu;
+ tos = mutable->port_config.tos;
- if (skb->protocol == htons(ETH_P_IP)) {
- iph.frag_off |= old_iph->frag_off & htons(IP_DF);
- mtu = max(mtu, IP_MIN_MTU);
- }
- #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- else if (skb->protocol == htons(ETH_P_IPV6))
- mtu = max(mtu, IPV6_MIN_MTU);
- #endif
+ tos = INET_ECN_encapsulate(tos, inner_tos);
- iph.version = 4;
- iph.ihl = sizeof(struct iphdr) >> 2;
- iph.protocol = tnl_vport->tnl_ops->ipproto;
- iph.daddr = rt->rt_dst;
- iph.saddr = rt->rt_src;
+ /* Route lookup */
+ rt = find_route(vport, mutable, tos, &cache);
+ if (unlikely(!rt))
+ goto error_free;
+ if (unlikely(!cache))
+ unattached_dst = &rt_dst(rt);
+ /* Reset SKB */
nf_reset(skb);
secpath_reset(skb);
skb_dst_drop(skb);
- skb_dst_set(skb, &rt_dst(rt));
- /*
- * If we are doing GSO on a pskb it is better to make sure that the
- * headroom is correct now. We will only have to copy the portion in
- * the linear data area and GSO will preserve headroom when it creates
- * the segments. This is particularly beneficial on Xen where we get
- * lots of GSO pskbs. Conversely, we delay copying if it is just to
- * get our own writable clone because GSO may do the copy for us.
- */
- max_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
- + mutable->tunnel_hlen;
+ /* Offloading */
+ skb = handle_offloads(skb, mutable, rt);
+ if (unlikely(IS_ERR(skb)))
+ goto error;
- if (skb_headroom(skb) < max_headroom) {
- skb = check_headroom(skb, max_headroom);
- if (unlikely(IS_ERR(skb))) {
- vport_record_error(vport, VPORT_E_TX_DROPPED);
- goto error;
- }
+ /* MTU */
+ if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
+ err = VPORT_E_TX_DROPPED;
+ goto error_free;
}
- forward_ip_summed(skb);
+ /*
+ * If we are over the MTU, allow the IP stack to handle fragmentation.
+ * Fragmentation is a slow path anyways.
+ */
+ if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
+ cache)) {
+ unattached_dst = &rt_dst(rt);
+ dst_hold(unattached_dst);
+ cache = NULL;
+ }
- if (unlikely(vswitch_skb_checksum_setup(skb)))
- goto error_free;
+ /* TTL */
+ ttl = mutable->port_config.ttl;
+ if (!ttl)
+ ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
- skb = handle_gso(skb);
- if (unlikely(IS_ERR(skb))) {
- vport_record_error(vport, VPORT_E_TX_DROPPED);
- goto error;
+ if (mutable->port_config.flags & TNL_F_TTL_INHERIT) {
+ if (skb->protocol == htons(ETH_P_IP))
+ ttl = ip_hdr(skb)->ttl;
+ #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+ else if (skb->protocol == htons(ETH_P_IPV6))
+ ttl = ipv6_hdr(skb)->hop_limit;
+ #endif
}
- /*
- * Process GSO segments. Try to do any work for the entire packet that
- * doesn't involve actually writing to it before this point.
- */
- orig_len = 0;
- do {
+ while (skb) {
+ struct iphdr *iph;
struct sk_buff *next_skb = skb->next;
skb->next = NULL;
- orig_len += build_packet(vport, mutable, &iph, rt, max_headroom, mtu, skb);
+ if (likely(cache)) {
+ skb_push(skb, cache->len);
+ memcpy(skb->data, get_cached_header(cache), cache->len);
+ skb_reset_mac_header(skb);
+ skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
+
+ } else {
+ skb_push(skb, mutable->tunnel_hlen);
+ create_tunnel_header(vport, mutable, rt, skb->data);
+ skb_reset_network_header(skb);
+
+ if (next_skb)
+ skb_dst_set(skb, dst_clone(unattached_dst));
+ else {
+ skb_dst_set(skb, unattached_dst);
+ unattached_dst = NULL;
+ }
+
+ memset(&IPCB(skb)->opt, 0, sizeof(IPCB(skb)->opt));
+ IPCB(skb)->flags = 0;
+ }
+ skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
+
+ iph = ip_hdr(skb);
+ iph->tos = tos;
+ iph->ttl = ttl;
+ iph->frag_off = frag_off;
+ ip_select_ident(iph, &rt_dst(rt), NULL);
+
+ skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
+ if (unlikely(!skb))
+ goto next;
+
+ if (likely(cache)) {
+ int orig_len = skb->len - cache->len;
+
+ skb->protocol = htons(ETH_P_IP);
+ iph->tot_len = htons(skb->len - skb_network_offset(skb));
+ ip_send_check(iph);
+
+ if (is_internal_dev(rt_dst(rt).dev)) {
+ OVS_CB(skb)->flow = cache->flow;
+ compute_ip_summed(skb, true);
+ vport_receive(internal_dev_get_vport(rt_dst(rt).dev), skb);
+ sent_len += orig_len;
+ } else {
+ int err;
+
+ skb->dev = rt_dst(rt).dev;
+ err = dev_queue_xmit(skb);
+
+ if (likely(net_xmit_eval(err) == 0))
+ sent_len += orig_len;
+ }
+ } else
+ sent_len += send_frags(skb, mutable);
+
+ next:
skb = next_skb;
- } while (skb);
+ }
- if (unlikely(orig_len == 0))
+ if (unlikely(sent_len == 0))
vport_record_error(vport, VPORT_E_TX_DROPPED);
- return orig_len;
+ goto out;
error_free:
- kfree_skb(skb);
- vport_record_error(vport, VPORT_E_TX_ERROR);
+ tnl_free_linked_skbs(skb);
error:
- return 0;
- }
-
- int tnl_init(void)
- {
- return 0;
- }
-
- void tnl_exit(void)
- {
- tbl_destroy(port_table, NULL);
- port_table = NULL;
+ dst_release(unattached_dst);
+ vport_record_error(vport, err);
+ out:
+ return sent_len;
}
static int set_config(const void __user *uconfig, const struct tnl_ops *tnl_ops,
if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct tnl_port_config)))
return -EFAULT;
+ if (mutable->port_config.daddr == 0)
+ return -EINVAL;
+
+ if (mutable->port_config.tos != RT_TOS(mutable->port_config.tos))
+ return -EINVAL;
+
mutable->tunnel_hlen = tnl_ops->hdr_len(&mutable->port_config);
if (mutable->tunnel_hlen < 0)
return mutable->tunnel_hlen;
mutable->tunnel_hlen += sizeof(struct iphdr);
- if (mutable->port_config.daddr == 0)
- return -EINVAL;
-
mutable->tunnel_type = tnl_ops->tunnel_type;
if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
mutable->tunnel_type |= TNL_T_KEY_MATCH;
strcpy(tnl_vport->name, name);
tnl_vport->tnl_ops = tnl_ops;
- tnl_vport->mutable = kmalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
+ tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
if (!tnl_vport->mutable) {
err = -ENOMEM;
goto error_free_vport;
if (err)
goto error_free_mutable;
+ spin_lock_init(&tnl_vport->cache_lock);
+
+ #ifdef NEED_CACHE_TIMEOUT
+ tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
+ (net_random() % (MAX_CACHE_EXP / 2));
+ #endif
+
err = add_port(vport);
if (err)
goto error_free_mutable;
struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
struct tnl_mutable_config *mutable;
int err;
- bool update_hash = false;
mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
if (!mutable) {
if (err)
goto error_free;
- /*
- * Only remove the port from the hash table if something that would
- * affect the lookup has changed.
- */
- if (tnl_vport->mutable->port_config.saddr != mutable->port_config.saddr ||
- tnl_vport->mutable->port_config.daddr != mutable->port_config.daddr ||
- tnl_vport->mutable->port_config.in_key != mutable->port_config.in_key ||
- (tnl_vport->mutable->port_config.flags & TNL_F_IN_KEY_MATCH) !=
- (mutable->port_config.flags & TNL_F_IN_KEY_MATCH))
- update_hash = true;
-
-
- /*
- * This update is not atomic but the lookup uses the config, which
- * serves as an inherent double check.
- */
- if (update_hash) {
- err = del_port(vport);
- if (err)
- goto error_free;
- }
-
- assign_config_rcu(vport, mutable);
+ mutable->seq++;
- if (update_hash) {
- err = add_port(vport);
- if (err)
- goto error_free;
- }
+ err = move_port(vport, mutable);
+ if (err)
+ goto error_free;
return 0;
return err;
}
- static void free_port(struct rcu_head *rcu)
+ static void free_port_rcu(struct rcu_head *rcu)
{
struct tnl_vport *tnl_vport = container_of(rcu, struct tnl_vport, rcu);
+ spin_lock_bh(&tnl_vport->cache_lock);
+ free_cache(tnl_vport->cache);
+ spin_unlock_bh(&tnl_vport->cache_lock);
+
kfree(tnl_vport->mutable);
vport_free(tnl_vport_to_vport(tnl_vport));
}
&old_mutable))
del_port(vport);
- call_rcu(&tnl_vport->rcu, free_port);
+ call_rcu(&tnl_vport->rcu, free_port_rcu);
return 0;
}
return 0;
}
-
const char *tnl_get_name(const struct vport *vport)
{
const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
return rcu_dereference(tnl_vport->mutable)->mtu;
}
+
+ void tnl_free_linked_skbs(struct sk_buff *skb)
+ {
+ if (unlikely(!skb))
+ return;
+
+ while (skb) {
+ struct sk_buff *next = skb->next;
+ kfree_skb(skb);
+ skb = next;
+ }
+ }
struct net_device_stats *stats = &internal_dev_priv(netdev)->stats;
if (vport) {
- struct odp_vport_stats vport_stats;
+ struct xflow_vport_stats vport_stats;
vport_get_stats(vport, &vport_stats);
skb_reset_mac_header(skb);
compute_ip_summed(skb, true);
+ OVS_CB(skb)->flow = NULL;
vport_receive(vport, skb);
struct vport_ops internal_vport_ops = {
.type = "internal",
- .flags = VPORT_F_REQUIRED | VPORT_F_GEN_STATS,
+ .flags = VPORT_F_REQUIRED | VPORT_F_GEN_STATS | VPORT_F_FLOW,
.create = internal_dev_create,
.destroy = internal_dev_destroy,
.attach = internal_dev_attach,
kfree(dev_table);
}
-static int do_vport_add(struct odp_vport_add *vport_config)
+static int do_vport_add(struct xflow_vport_add *vport_config)
{
struct vport *vport;
int err = 0;
* on device type). This function is for userspace callers and assumes no
* locks are held.
*/
-int vport_user_add(const struct odp_vport_add __user *uvport_config)
+int vport_user_add(const struct xflow_vport_add __user *uvport_config)
{
- struct odp_vport_add vport_config;
+ struct xflow_vport_add vport_config;
- if (copy_from_user(&vport_config, uvport_config, sizeof(struct odp_vport_add)))
+ if (copy_from_user(&vport_config, uvport_config, sizeof(struct xflow_vport_add)))
return -EFAULT;
return do_vport_add(&vport_config);
}
#ifdef CONFIG_COMPAT
-int compat_vport_user_add(struct compat_odp_vport_add *ucompat)
+int compat_vport_user_add(struct compat_xflow_vport_add *ucompat)
{
- struct compat_odp_vport_add compat;
- struct odp_vport_add vport_config;
+ struct compat_xflow_vport_add compat;
+ struct xflow_vport_add vport_config;
- if (copy_from_user(&compat, ucompat, sizeof(struct compat_odp_vport_add)))
+ if (copy_from_user(&compat, ucompat, sizeof(struct compat_xflow_vport_add)))
return -EFAULT;
memcpy(vport_config.port_type, compat.port_type, VPORT_TYPE_SIZE);
}
#endif
-static int do_vport_mod(struct odp_vport_mod *vport_config)
+static int do_vport_mod(struct xflow_vport_mod *vport_config)
{
struct vport *vport;
int err;
* dependent on device type). This function is for userspace callers and
* assumes no locks are held.
*/
-int vport_user_mod(const struct odp_vport_mod __user *uvport_config)
+int vport_user_mod(const struct xflow_vport_mod __user *uvport_config)
{
- struct odp_vport_mod vport_config;
+ struct xflow_vport_mod vport_config;
- if (copy_from_user(&vport_config, uvport_config, sizeof(struct odp_vport_mod)))
+ if (copy_from_user(&vport_config, uvport_config, sizeof(struct xflow_vport_mod)))
return -EFAULT;
return do_vport_mod(&vport_config);
}
#ifdef CONFIG_COMPAT
-int compat_vport_user_mod(struct compat_odp_vport_mod *ucompat)
+int compat_vport_user_mod(struct compat_xflow_vport_mod *ucompat)
{
- struct compat_odp_vport_mod compat;
- struct odp_vport_mod vport_config;
+ struct compat_xflow_vport_mod compat;
+ struct xflow_vport_mod vport_config;
- if (copy_from_user(&compat, ucompat, sizeof(struct compat_odp_vport_mod)))
+ if (copy_from_user(&compat, ucompat, sizeof(struct compat_xflow_vport_mod)))
return -EFAULT;
memcpy(vport_config.devname, compat.devname, IFNAMSIZ);
* Retrieves transmit, receive, and error stats for the given device. This
* function is for userspace callers and assumes no locks are held.
*/
-int vport_user_stats_get(struct odp_vport_stats_req __user *ustats_req)
+int vport_user_stats_get(struct xflow_vport_stats_req __user *ustats_req)
{
- struct odp_vport_stats_req stats_req;
+ struct xflow_vport_stats_req stats_req;
struct vport *vport;
int err;
- if (copy_from_user(&stats_req, ustats_req, sizeof(struct odp_vport_stats_req)))
+ if (copy_from_user(&stats_req, ustats_req, sizeof(struct xflow_vport_stats_req)))
return -EFAULT;
stats_req.devname[IFNAMSIZ - 1] = '\0';
vport_unlock();
if (!err)
- if (copy_to_user(ustats_req, &stats_req, sizeof(struct odp_vport_stats_req)))
+ if (copy_to_user(ustats_req, &stats_req, sizeof(struct xflow_vport_stats_req)))
err = -EFAULT;
return err;
* -EOPNOTSUPP. This function is for userspace callers and assumes no locks
* are held.
*/
-int vport_user_stats_set(struct odp_vport_stats_req __user *ustats_req)
+int vport_user_stats_set(struct xflow_vport_stats_req __user *ustats_req)
{
- struct odp_vport_stats_req stats_req;
+ struct xflow_vport_stats_req stats_req;
struct vport *vport;
int err;
- if (copy_from_user(&stats_req, ustats_req, sizeof(struct odp_vport_stats_req)))
+ if (copy_from_user(&stats_req, ustats_req, sizeof(struct xflow_vport_stats_req)))
return -EFAULT;
stats_req.devname[IFNAMSIZ - 1] = '\0';
* Retrieves the Ethernet address of the given device. This function is for
* userspace callers and assumes no locks are held.
*/
-int vport_user_ether_get(struct odp_vport_ether __user *uvport_ether)
+int vport_user_ether_get(struct xflow_vport_ether __user *uvport_ether)
{
- struct odp_vport_ether vport_ether;
+ struct xflow_vport_ether vport_ether;
struct vport *vport;
int err = 0;
- if (copy_from_user(&vport_ether, uvport_ether, sizeof(struct odp_vport_ether)))
+ if (copy_from_user(&vport_ether, uvport_ether, sizeof(struct xflow_vport_ether)))
return -EFAULT;
vport_ether.devname[IFNAMSIZ - 1] = '\0';
vport_unlock();
if (!err)
- if (copy_to_user(uvport_ether, &vport_ether, sizeof(struct odp_vport_ether)))
+ if (copy_to_user(uvport_ether, &vport_ether, sizeof(struct xflow_vport_ether)))
err = -EFAULT;
return err;
* -EOPNOTSUPP. This function is for userspace callers and assumes no locks
* are held.
*/
-int vport_user_ether_set(struct odp_vport_ether __user *uvport_ether)
+int vport_user_ether_set(struct xflow_vport_ether __user *uvport_ether)
{
- struct odp_vport_ether vport_ether;
+ struct xflow_vport_ether vport_ether;
struct vport *vport;
int err;
- if (copy_from_user(&vport_ether, uvport_ether, sizeof(struct odp_vport_ether)))
+ if (copy_from_user(&vport_ether, uvport_ether, sizeof(struct xflow_vport_ether)))
return -EFAULT;
vport_ether.devname[IFNAMSIZ - 1] = '\0';
* Retrieves the MTU of the given device. This function is for userspace
* callers and assumes no locks are held.
*/
-int vport_user_mtu_get(struct odp_vport_mtu __user *uvport_mtu)
+int vport_user_mtu_get(struct xflow_vport_mtu __user *uvport_mtu)
{
- struct odp_vport_mtu vport_mtu;
+ struct xflow_vport_mtu vport_mtu;
struct vport *vport;
int err = 0;
- if (copy_from_user(&vport_mtu, uvport_mtu, sizeof(struct odp_vport_mtu)))
+ if (copy_from_user(&vport_mtu, uvport_mtu, sizeof(struct xflow_vport_mtu)))
return -EFAULT;
vport_mtu.devname[IFNAMSIZ - 1] = '\0';
vport_unlock();
if (!err)
- if (copy_to_user(uvport_mtu, &vport_mtu, sizeof(struct odp_vport_mtu)))
+ if (copy_to_user(uvport_mtu, &vport_mtu, sizeof(struct xflow_vport_mtu)))
err = -EFAULT;
return err;
* MTU, in which case the result will always be -EOPNOTSUPP. This function is
* for userspace callers and assumes no locks are held.
*/
-int vport_user_mtu_set(struct odp_vport_mtu __user *uvport_mtu)
+int vport_user_mtu_set(struct xflow_vport_mtu __user *uvport_mtu)
{
- struct odp_vport_mtu vport_mtu;
+ struct xflow_vport_mtu vport_mtu;
struct vport *vport;
int err;
- if (copy_from_user(&vport_mtu, uvport_mtu, sizeof(struct odp_vport_mtu)))
+ if (copy_from_user(&vport_mtu, uvport_mtu, sizeof(struct xflow_vport_mtu)))
return -EFAULT;
vport_mtu.devname[IFNAMSIZ - 1] = '\0';
* support setting the stats, in which case the result will always be
* -EOPNOTSUPP. RTNL lock must be held.
*/
-int vport_set_stats(struct vport *vport, struct odp_vport_stats *stats)
+int vport_set_stats(struct vport *vport, struct xflow_vport_stats *stats)
{
ASSERT_RTNL();
if (vport->ops->flags & VPORT_F_GEN_STATS) {
spin_lock_bh(&vport->stats_lock);
- memcpy(&vport->offset_stats, stats, sizeof(struct odp_vport_stats));
+ memcpy(&vport->offset_stats, stats, sizeof(struct xflow_vport_stats));
spin_unlock_bh(&vport->stats_lock);
return 0;
*
* Retrieves transmit, receive, and error stats for the given device.
*/
-int vport_get_stats(struct vport *vport, struct odp_vport_stats *stats)
+int vport_get_stats(struct vport *vport, struct xflow_vport_stats *stats)
{
- struct odp_vport_stats dev_stats;
- struct odp_vport_stats *dev_statsp = NULL;
+ struct xflow_vport_stats dev_stats;
+ struct xflow_vport_stats *dev_statsp = NULL;
int err;
if (vport->ops->get_stats) {
spin_lock_bh(&vport->stats_lock);
- memcpy(stats, &vport->offset_stats, sizeof(struct odp_vport_stats));
+ memcpy(stats, &vport->offset_stats, sizeof(struct xflow_vport_stats));
stats->rx_errors += vport->err_stats.rx_errors
+ vport->err_stats.rx_frame_err
if (!dp_port)
return -EAGAIN;
- return vport_get_ifindex(dp_port->dp->ports[ODPP_LOCAL]->vport);
+ return vport_get_ifindex(dp_port->dp->ports[XFLOWP_LOCAL]->vport);
}
/**
local_bh_enable();
}
+ if (!(vport->ops->flags & VPORT_F_FLOW))
+ OVS_CB(skb)->flow = NULL;
+
if (!(vport->ops->flags & VPORT_F_TUN_ID))
OVS_CB(skb)->tun_id = 0;
#include <linux/spinlock.h>
#include "datapath.h"
-#include "openvswitch/datapath-protocol.h"
-#include "odp-compat.h"
+#include "openvswitch/xflow.h"
+#include "xflow-compat.h"
struct vport;
struct dp_port;
/* The following definitions are for users of the vport subsytem: */
-int vport_user_add(const struct odp_vport_add __user *);
-int vport_user_mod(const struct odp_vport_mod __user *);
+int vport_user_add(const struct xflow_vport_add __user *);
+int vport_user_mod(const struct xflow_vport_mod __user *);
int vport_user_del(const char __user *udevname);
#ifdef CONFIG_COMPAT
-int compat_vport_user_add(struct compat_odp_vport_add __user *);
-int compat_vport_user_mod(struct compat_odp_vport_mod __user *);
+int compat_vport_user_add(struct compat_xflow_vport_add __user *);
+int compat_vport_user_mod(struct compat_xflow_vport_mod __user *);
#endif
-int vport_user_stats_get(struct odp_vport_stats_req __user *);
-int vport_user_stats_set(struct odp_vport_stats_req __user *);
-int vport_user_ether_get(struct odp_vport_ether __user *);
-int vport_user_ether_set(struct odp_vport_ether __user *);
-int vport_user_mtu_get(struct odp_vport_mtu __user *);
-int vport_user_mtu_set(struct odp_vport_mtu __user *);
+int vport_user_stats_get(struct xflow_vport_stats_req __user *);
+int vport_user_stats_set(struct xflow_vport_stats_req __user *);
+int vport_user_ether_get(struct xflow_vport_ether __user *);
+int vport_user_ether_set(struct xflow_vport_ether __user *);
+int vport_user_mtu_get(struct xflow_vport_mtu __user *);
+int vport_user_mtu_set(struct xflow_vport_mtu __user *);
void vport_lock(void);
void vport_unlock(void);
int vport_set_mtu(struct vport *, int mtu);
int vport_set_addr(struct vport *, const unsigned char *);
-int vport_set_stats(struct vport *, struct odp_vport_stats *);
+int vport_set_stats(struct vport *, struct xflow_vport_stats *);
const char *vport_get_name(const struct vport *);
const char *vport_get_type(const struct vport *);
struct dp_port *vport_get_dp_port(const struct vport *);
struct kobject *vport_get_kobj(const struct vport *);
-int vport_get_stats(struct vport *, struct odp_vport_stats *);
+int vport_get_stats(struct vport *, struct xflow_vport_stats *);
unsigned vport_get_flags(const struct vport *);
int vport_is_running(const struct vport *);
spinlock_t stats_lock;
struct vport_err_stats err_stats;
- struct odp_vport_stats offset_stats;
+ struct xflow_vport_stats offset_stats;
};
#define VPORT_F_REQUIRED (1 << 0) /* If init fails, module loading fails. */
#define VPORT_F_GEN_STATS (1 << 1) /* Track stats at the generic layer. */
- #define VPORT_F_TUN_ID (1 << 2) /* Sets OVS_CB(skb)->tun_id. */
+ #define VPORT_F_FLOW (1 << 2) /* Sets OVS_CB(skb)->flow. */
+ #define VPORT_F_TUN_ID (1 << 3) /* Sets OVS_CB(skb)->tun_id. */
/**
* struct vport_ops - definition of a type of virtual port
int (*set_mtu)(struct vport *, int mtu);
int (*set_addr)(struct vport *, const unsigned char *);
- int (*set_stats)(const struct vport *, struct odp_vport_stats *);
+ int (*set_stats)(const struct vport *, struct xflow_vport_stats *);
/* Called with rcu_read_lock or RTNL lock. */
const char *(*get_name)(const struct vport *);
const unsigned char *(*get_addr)(const struct vport *);
struct kobject *(*get_kobj)(const struct vport *);
- int (*get_stats)(const struct vport *, struct odp_vport_stats *);
+ int (*get_stats)(const struct vport *, struct xflow_vport_stats *);
unsigned (*get_dev_flags)(const struct vport *);
int (*is_running)(const struct vport *);
#include "openflow/openflow.h"
-#define NICIRA_OUI_STR "002320"
-
/* The following vendor extensions, proposed by Nicira Networks, are not yet
- * ready for standardization (and may never be), so they are not included in
- * openflow.h. */
+ * standardized, so they are not included in openflow.h. Some of them may be
+ * suitable for standardization; others we never expect to standardize. */
#define NX_VENDOR_ID 0x00002320
+\f
+/* Nicira vendor-specific error messages extension.
+ *
+ * OpenFlow 1.0 has a set of predefined error types (OFPET_*) and codes (which
+ * are specific to each type). It does not have any provision for
+ * vendor-specific error codes, and it does not even provide "generic" error
+ * codes that can apply to problems not anticipated by the OpenFlow
+ * specification authors.
+ *
+ * This extension attempts to address the problem by adding a generic "error
+ * vendor extension". The extension works as follows: use NXET_VENDOR as type
+ * and NXVC_VENDOR_CODE as code, followed by struct nx_vendor_error with
+ * vendor-specific details, followed by at least 64 bytes of the failed
+ * request.
+ *
+ * It would be better to have type-specific vendor extension, e.g. so that
+ * OFPET_BAD_ACTION could be used with vendor-specific code values. But
+ * OFPET_BAD_ACTION and most other standardized types already specify that
+ * their 'data' values are (the start of) the OpenFlow message being replied
+ * to, so there is no room to insert a vendor ID.
+ *
+ * Currently this extension is only implemented by Open vSwitch, but it seems
+ * like a reasonable candidate for future standardization.
+ */
+
+/* This is a random number to avoid accidental collision with any other
+ * vendor's extension. */
+#define NXET_VENDOR 0xb0c2
+
+/* ofp_error msg 'code' values for NXET_VENDOR. */
+enum nx_vendor_code {
+ NXVC_VENDOR_ERROR /* 'data' contains struct nx_vendor_error. */
+};
+
+/* 'data' for 'type' == NXET_VENDOR, 'code' == NXVC_VENDOR_ERROR. */
+struct nx_vendor_error {
+ uint32_t vendor; /* Vendor ID as in struct ofp_vendor_header. */
+ uint16_t type; /* Vendor-defined type. */
+ uint16_t code; /* Vendor-defined subtype. */
+ /* Followed by at least the first 64 bytes of the failed request. */
+};
+\f
+/* Specific Nicira extension error numbers.
+ *
+ * These are the "code" values used in nx_vendor_error. So far, the "type"
+ * values in nx_vendor_error are the same as those in ofp_error_msg. That is,
+ * at Nicira so far we've only needed additional vendor-specific 'code' values,
+ * so we're using the existing 'type' values to avoid having to invent new ones
+ * that duplicate the current ones' meanings. */
+
+/* Additional "code" values for OFPET_FLOW_MOD_FAILED. */
+enum {
+ /* Generic hardware error. */
+ NXFMFC_HARDWARE = 0x100,
+
+ /* A nonexistent table ID was specified in the "command" field of struct
+ * ofp_flow_mod, when the nxt_flow_mod_table_id extension is enabled. */
+ NXFMFC_BAD_TABLE_ID
+};
+\f
+/* Nicira vendor requests and replies. */
enum nicira_type {
/* Switch status request. The request body is an ASCII string that
/* Controller role support. The request body is struct nx_role_request.
* The reply echos the request. */
NXT_ROLE_REQUEST,
- NXT_ROLE_REPLY
+ NXT_ROLE_REPLY,
+
+ /* Use the upper 8 bits of the 'command' member in struct ofp_flow_mod to
+ * designate the table to which a flow is to be added? See the big comment
+ * on struct nxt_flow_mod_table_id for more information. */
+ NXT_FLOW_MOD_TABLE_ID
};
struct nicira_header {
};
OFP_ASSERT(sizeof(struct nxt_tun_id_cookie) == 24);
+/* This command enables or disables an Open vSwitch extension that allows a
+ * controller to specify the OpenFlow table to which a flow should be added,
+ * instead of having the switch decide which table is most appropriate as
+ * required by OpenFlow 1.0. By default, the extension is disabled.
+ *
+ * When this feature is enabled, Open vSwitch treats struct ofp_flow_mod's
+ * 16-bit 'command' member as two separate fields. The upper 8 bits are used
+ * as the table ID, the lower 8 bits specify the command as usual. A table ID
+ * of 0xff is treated like a wildcarded table ID.
+ *
+ * The specific treatment of the table ID depends on the type of flow mod:
+ *
+ * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that
+ * table. If an identical flow already exists in that table only, then it
+ * is replaced. If the flow cannot be placed in the specified table,
+ * either because the table is full or because the table cannot support
+ * flows of the given type, the switch replies with an
+ * OFPFMFC_ALL_TABLES_FULL error. (A controller can distinguish these
+ * cases by comparing the current and maximum number of entries reported
+ * in ofp_table_stats.)
+ *
+ * If the table ID is wildcarded, the switch picks an appropriate table
+ * itself. If an identical flow or flows already exist in some flow
+ * table, then one of them is replaced. The choice of table might depend
+ * on the flows that are already in the switch; for example, if one table
+ * fills up then the switch might fall back to another one.
+ *
+ * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows
+ * within that table are matched and modified or deleted. If the table ID
+ * is wildcarded, flows within any table may be matched and modified or
+ * deleted.
+ *
+ * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID,
+ * only a flow within that table may be matched and modified or deleted.
+ * If the table ID is wildcarded and exactly one flow within any table
+ * matches, then it is modified or deleted; if flows in more than one
+ * table match, then none is modified or deleted.
+ */
+struct nxt_flow_mod_table_id {
+ struct ofp_header header;
+ uint32_t vendor; /* NX_VENDOR_ID. */
+ uint32_t subtype; /* NXT_FLOW_MOD_TABLE_ID. */
+ uint8_t set; /* Nonzero to enable, zero to disable. */
+ uint8_t pad[7];
+};
+OFP_ASSERT(sizeof(struct nxt_flow_mod_table_id) == 24);
+
/* Configures the "role" of the sending controller. The default role is:
*
* - Other (NX_ROLE_OTHER), which allows the controller access to all
NX_ROLE_MASTER, /* Full access, at most one. */
NX_ROLE_SLAVE /* Read-only access. */
};
+\f
+/* Nicira vendor flow actions. */
enum nx_action_subtype {
NXAST_SNAT__OBSOLETE, /* No longer used. */
*
* This is useful because OpenFlow does not provide a way to match on the
* Ethernet addresses inside ARP packets, so there is no other way to drop
- * spoofed ARPs other than sending every packet up to the controller. */
- NXAST_DROP_SPOOFED_ARP
+ * spoofed ARPs other than sending every ARP packet to a controller. */
+ NXAST_DROP_SPOOFED_ARP,
+
+ /* Set the queue that should be used when packets are output. This
+ * is similar to the OpenFlow OFPAT_ENQUEUE action, but does not
+ * take the output port as an argument. This allows the queue
+ * to be defined before the port is known. */
+ NXAST_SET_QUEUE,
+
+ /* Restore the queue to the value it was before any NXAST_SET_QUEUE
+ * actions were used. */
+ NXAST_POP_QUEUE
};
/* Action structure for NXAST_RESUBMIT. */
};
OFP_ASSERT(sizeof(struct nx_action_set_tunnel) == 16);
+ /* Action structure for NXAST_SET_QUEUE. */
+ struct nx_action_set_queue {
+ uint16_t type; /* OFPAT_VENDOR. */
+ uint16_t len; /* Length is 16. */
+ uint32_t vendor; /* NX_VENDOR_ID. */
+ uint16_t subtype; /* NXAST_SET_QUEUE. */
+ uint8_t pad[2];
+ uint32_t queue_id; /* Where to enqueue packets. */
+ };
+ OFP_ASSERT(sizeof(struct nx_action_set_queue) == 16);
+
/* Header for Nicira-defined actions. */
struct nx_action_header {
uint16_t type; /* OFPAT_VENDOR. */
static bool rules_match_2wild(const struct cls_rule *wild1,
const struct cls_rule *wild2, int field_idx);
-/* Converts the flow in 'flow' into a cls_rule in 'rule', with the given
- * 'wildcards' and 'priority'.*/
+/* Converts the flow in 'flow' into a cls_rule in 'rule'. */
void
-cls_rule_from_flow(const flow_t *flow, uint32_t wildcards,
- unsigned int priority, struct cls_rule *rule)
+cls_rule_from_flow(const flow_t *flow, struct cls_rule *rule)
{
- assert(!flow->reserved[0] && !flow->reserved[1] && !flow->reserved[2]);
rule->flow = *flow;
- flow_wildcards_init(&rule->wc, wildcards);
- rule->priority = priority;
- rule->table_idx = table_idx_from_wildcards(rule->wc.wildcards);
+ if (!rule->flow.wildcards && rule->flow.priority < UINT16_MAX) {
+ rule->flow.priority = UINT16_MAX;
+ }
+ flow_wildcards_init(&rule->wc, flow->wildcards);
+ rule->table_idx = table_idx_from_wildcards(flow->wildcards);
}
/* Converts the ofp_match in 'match' into a cls_rule in 'rule', with the given
bool tun_id_from_cookie, uint64_t cookie,
struct cls_rule *rule)
{
- uint32_t wildcards;
- flow_from_match(match, tun_id_from_cookie, cookie, &rule->flow, &wildcards);
- flow_wildcards_init(&rule->wc, wildcards);
- rule->priority = rule->wc.wildcards ? priority : UINT16_MAX;
- rule->table_idx = table_idx_from_wildcards(rule->wc.wildcards);
+ flow_from_match(match, rule->flow.wildcards ? priority : UINT16_MAX,
+ tun_id_from_cookie, cookie, &rule->flow);
+ flow_wildcards_init(&rule->wc, rule->flow.wildcards);
+ rule->table_idx = table_idx_from_wildcards(rule->flow.wildcards);
}
/* Converts 'rule' to a string and returns the string. The caller must free
{
struct ds s = DS_EMPTY_INITIALIZER;
ds_put_format(&s, "wildcards=%x priority=%u ",
- rule->wc.wildcards, rule->priority);
+ rule->flow.wildcards, rule->flow.priority);
flow_format(&s, &rule->flow);
return ds_cstr(&s);
}
void
cls_rule_print(const struct cls_rule *rule)
{
- printf("wildcards=%x priority=%u ", rule->wc.wildcards, rule->priority);
+ printf("wildcards=%x priority=%u ",
+ rule->flow.wildcards, rule->flow.priority);
flow_print(stdout, &rule->flow);
putc('\n', stdout);
}
struct cls_rule *new)
{
if (old != new) {
- if (new->wc.wildcards) {
+ if (new->flow.wildcards) {
list_moved(&new->node.list);
} else {
hmap_node_moved(&cls->exact_table,
struct cls_rule *new)
{
assert(old != new);
- assert(old->wc.wildcards == new->wc.wildcards);
- assert(old->priority == new->priority);
+ assert(old->flow.wildcards == new->flow.wildcards);
+ assert(old->flow.priority == new->flow.priority);
- if (new->wc.wildcards) {
+ if (new->flow.wildcards) {
list_replace(&new->node.list, &old->node.list);
} else {
hmap_replace(&cls->exact_table, &old->node.hmap, &new->node.hmap);
struct hmap *tbl;
for (tbl = &cls->tables[0]; tbl < &cls->tables[CLS_N_FIELDS]; tbl++) {
- HMAP_FOR_EACH_SAFE (bucket, next_bucket,
- struct cls_bucket, hmap_node, tbl) {
+ HMAP_FOR_EACH_SAFE (bucket, next_bucket, hmap_node, tbl) {
free(bucket);
}
hmap_destroy(tbl);
return hmap_count(&cls->exact_table);
}
+/* Returns the number of rules in 'classifier' that have at least one
+ * wildcard. */
+int
+classifier_count_wild(const struct classifier *cls)
+{
+ return classifier_count(cls) - classifier_count_exact(cls);
+}
+
/* Inserts 'rule' into 'cls'. Transfers ownership of 'rule' to 'cls'.
*
* If 'cls' already contains an identical rule (including wildcards, values of
classifier_insert(struct classifier *cls, struct cls_rule *rule)
{
struct cls_rule *old;
- assert((rule->wc.wildcards == 0) == (rule->table_idx == CLS_F_IDX_EXACT));
- old = (rule->wc.wildcards
+ assert((rule->flow.wildcards == 0) == (rule->table_idx == CLS_F_IDX_EXACT));
+ old = (rule->flow.wildcards
? table_insert(&cls->tables[rule->table_idx], rule)
: insert_exact_rule(cls, rule));
if (!old) {
/* Inserts 'rule' into 'cls'. Transfers ownership of 'rule' to 'cls'.
*
- * 'rule' must be an exact-match rule (rule->wc.wildcards must be 0) and 'cls'
+ * 'rule' must be an exact-match rule (rule->flow.wildcards must be 0) and 'cls'
* must not contain any rule with an identical key. */
void
classifier_insert_exact(struct classifier *cls, struct cls_rule *rule)
{
hmap_insert(&cls->exact_table, &rule->node.hmap,
- flow_hash(&rule->flow, 0));
+ flow_hash_headers(&rule->flow, 0));
cls->n_rules++;
}
void
classifier_remove(struct classifier *cls, struct cls_rule *rule)
{
- if (rule->wc.wildcards) {
+ if (rule->flow.wildcards) {
/* Remove 'rule' from bucket. If that empties the bucket, remove the
* bucket from its table. */
struct hmap *table = &cls->tables[rule->table_idx];
classifier_lookup_exact(const struct classifier *cls, const flow_t *flow)
{
return (!hmap_is_empty(&cls->exact_table)
- ? search_exact_table(cls, flow_hash(flow, 0), flow)
+ ? search_exact_table(cls, flow_hash_headers(flow, 0), flow)
: NULL);
}
struct cls_rule target;
int i;
- cls_rule_from_flow(flow, 0, 0, &target);
+ cls_rule_from_flow(flow, &target);
for (i = 0; i < CLS_N_FIELDS; i++) {
struct cls_rule *rule = search_table(&cls->tables[i], i, &target);
- if (rule && (!best || rule->priority > best->priority)) {
+ if (rule && (!best || rule->flow.priority > best->flow.priority)) {
best = rule;
}
}
struct cls_rule *
classifier_find_rule_exactly(const struct classifier *cls,
- const flow_t *target, uint32_t wildcards,
- unsigned int priority)
+ const flow_t *target)
{
struct cls_bucket *bucket;
int table_idx;
uint32_t hash;
- if (!wildcards) {
+ if (!target->wildcards) {
/* Ignores 'priority'. */
- return search_exact_table(cls, flow_hash(target, 0), target);
+ return search_exact_table(cls, flow_hash_headers(target, 0), target);
}
- assert(wildcards == (wildcards & OVSFW_ALL));
- table_idx = table_idx_from_wildcards(wildcards);
+ assert(target->wildcards == (target->wildcards & OVSFW_ALL));
+ table_idx = table_idx_from_wildcards(target->wildcards);
hash = hash_fields(target, table_idx);
- HMAP_FOR_EACH_WITH_HASH (bucket, struct cls_bucket, hmap_node, hash,
+ HMAP_FOR_EACH_WITH_HASH (bucket, hmap_node, hash,
&cls->tables[table_idx]) {
if (equal_fields(&bucket->fixed, target, table_idx)) {
struct cls_rule *pos;
- LIST_FOR_EACH (pos, struct cls_rule, node.list, &bucket->rules) {
+ LIST_FOR_EACH (pos, node.list, &bucket->rules) {
- if (pos->priority < priority) {
+ if (pos->flow.priority < target->priority) {
return NULL;
- } else if (pos->priority == priority &&
- pos->wc.wildcards == wildcards &&
- flow_equal(target, &pos->flow)) {
+ } else if (pos->flow.priority == target->priority &&
+ pos->flow.wildcards == target->wildcards &&
+ flow_equal_headers(target, &pos->flow)) {
return pos;
}
}
return NULL;
}
-/* Checks if the flow defined by 'target' with 'wildcards' at 'priority'
- * overlaps with any other rule at the same priority in the classifier.
- * Two rules are considered overlapping if a packet could match both. */
+/* Checks if the flow defined by 'target' overlaps with any other rule at the
+ * same priority in the classifier. Two rules are considered overlapping if a
+ * packet could match both. */
bool
-classifier_rule_overlaps(const struct classifier *cls,
- const flow_t *target, uint32_t wildcards,
- unsigned int priority)
+classifier_rule_overlaps(const struct classifier *cls, const flow_t *target)
{
struct cls_rule target_rule;
const struct hmap *tbl;
- if (!wildcards) {
- return search_exact_table(cls, flow_hash(target, 0), target) ?
+ if (!target->wildcards) {
+ return search_exact_table(cls, flow_hash_headers(target, 0), target) ?
true : false;
}
- cls_rule_from_flow(target, wildcards, priority, &target_rule);
+ cls_rule_from_flow(target, &target_rule);
for (tbl = &cls->tables[0]; tbl < &cls->tables[CLS_N_FIELDS]; tbl++) {
struct cls_bucket *bucket;
- HMAP_FOR_EACH (bucket, struct cls_bucket, hmap_node, tbl) {
+ HMAP_FOR_EACH (bucket, hmap_node, tbl) {
struct cls_rule *rule;
- LIST_FOR_EACH (rule, struct cls_rule, node.list,
- &bucket->rules) {
+ LIST_FOR_EACH (rule, node.list, &bucket->rules) {
- if (rule->priority == priority
+ if (rule->flow.priority == target->priority
- && rules_match_2wild(rule, &target_rule, 0)) {
+ && rules_match_2wild(rule, &target_rule, 0)) {
return true;
}
}
return false;
}
-/* Ignores target->priority.
+/* Ignores target->flow.priority.
*
* 'callback' is allowed to delete the rule that is passed as its argument, but
* it must not delete (or move) any other rules in 'cls' that are in the same
* table as the argument rule. Two rules are in the same table if their
* cls_rule structs have the same table_idx; as a special case, a rule with
- * wildcards and an exact-match rule will never be in the same table. */
-void
+ * wildcards and an exact-match rule will never be in the same table.
+ *
+ * If 'callback' returns nonzero then the iteration stops immediately and
+ * classifier_for_each_match() passes up the return value. Otherwise,
+ * classifier_for_each_match() returns 0 after completing the iteration. */
+int
classifier_for_each_match(const struct classifier *cls,
- const struct cls_rule *target,
+ const flow_t *target_flow,
int include, cls_cb_func *callback, void *aux)
{
+ struct cls_rule target;
+
+ cls_rule_from_flow(target_flow, &target);
if (include & CLS_INC_WILD) {
const struct hmap *table;
table++) {
struct cls_bucket *bucket, *next_bucket;
- HMAP_FOR_EACH_SAFE (bucket, next_bucket,
- struct cls_bucket, hmap_node, table) {
+ HMAP_FOR_EACH_SAFE (bucket, next_bucket, hmap_node, table) {
/* XXX there is a bit of room for optimization here based on
* rejecting entire buckets on their fixed fields, but it will
* only be worthwhile for big buckets (which we hope we won't
* bucket itself will be destroyed. The bucket contains the
* list head so that's a use-after-free error. */
prev_rule = NULL;
- LIST_FOR_EACH (rule, struct cls_rule, node.list,
- &bucket->rules) {
+ LIST_FOR_EACH (rule, node.list, &bucket->rules) {
- if (rules_match_1wild(rule, target, 0)) {
+ if (rules_match_1wild(rule, &target, 0)) {
if (prev_rule) {
- callback(prev_rule, aux);
+ int retval = callback(prev_rule, aux);
+ if (retval) {
+ return retval;
+ }
}
prev_rule = rule;
}
}
if (prev_rule) {
- callback(prev_rule, aux);
+ int retval = callback(prev_rule, aux);
+ if (retval) {
+ return retval;
+ }
}
}
}
}
if (include & CLS_INC_EXACT) {
- if (target->wc.wildcards) {
+ if (target.flow.wildcards) {
struct cls_rule *rule, *next_rule;
- HMAP_FOR_EACH_SAFE (rule, next_rule, struct cls_rule, node.hmap,
+ HMAP_FOR_EACH_SAFE (rule, next_rule, node.hmap,
&cls->exact_table) {
- if (rules_match_1wild(rule, target, 0)) {
- callback(rule, aux);
+ if (rules_match_1wild(rule, &target, 0)) {
+ int retval = callback(rule, aux);
+ if (retval) {
+ return retval;
+ }
}
}
} else {
/* Optimization: there can be at most one match in the exact
* table. */
- size_t hash = flow_hash(&target->flow, 0);
+ size_t hash = flow_hash_headers(&target.flow, 0);
struct cls_rule *rule = search_exact_table(cls, hash,
- &target->flow);
+ &target.flow);
if (rule) {
- callback(rule, aux);
+ int retval = callback(rule, aux);
+ if (retval) {
+ return retval;
+ }
}
}
}
+
+ return 0;
}
/* 'callback' is allowed to delete the rule that is passed as its argument, but
* it must not delete (or move) any other rules in 'cls' that are in the same
* table as the argument rule. Two rules are in the same table if their
* cls_rule structs have the same table_idx; as a special case, a rule with
- * wildcards and an exact-match rule will never be in the same table. */
-void
+ * wildcards and an exact-match rule will never be in the same table.
+ *
+ * If 'callback' returns nonzero then the iteration stops immediately and
+ * classifier_for_each() passes up the return value. Otherwise,
+ * classifier_for_each() returns 0 after completing the iteration. */
+int
classifier_for_each(const struct classifier *cls, int include,
- void (*callback)(struct cls_rule *, void *aux),
+ int (*callback)(struct cls_rule *, void *aux),
void *aux)
{
if (include & CLS_INC_WILD) {
for (tbl = &cls->tables[0]; tbl < &cls->tables[CLS_N_FIELDS]; tbl++) {
struct cls_bucket *bucket, *next_bucket;
- HMAP_FOR_EACH_SAFE (bucket, next_bucket,
- struct cls_bucket, hmap_node, tbl) {
+ HMAP_FOR_EACH_SAFE (bucket, next_bucket, hmap_node, tbl) {
struct cls_rule *prev_rule, *rule;
/* We can't just use LIST_FOR_EACH_SAFE here because, if the
* bucket itself will be destroyed. The bucket contains the
* list head so that's a use-after-free error. */
prev_rule = NULL;
- LIST_FOR_EACH (rule, struct cls_rule, node.list,
- &bucket->rules) {
+ LIST_FOR_EACH (rule, node.list, &bucket->rules) {
if (prev_rule) {
- callback(prev_rule, aux);
+ int retval = callback(prev_rule, aux);
+ if (retval) {
+ return retval;
+ }
}
prev_rule = rule;
}
if (prev_rule) {
- callback(prev_rule, aux);
+ int retval = callback(prev_rule, aux);
+ if (retval) {
+ return retval;
+ }
}
}
}
if (include & CLS_INC_EXACT) {
struct cls_rule *rule, *next_rule;
- HMAP_FOR_EACH_SAFE (rule, next_rule,
- struct cls_rule, node.hmap, &cls->exact_table) {
+ HMAP_FOR_EACH_SAFE (rule, next_rule, node.hmap, &cls->exact_table) {
- callback(rule, aux);
+ int retval = callback(rule, aux);
+ if (retval) {
+ return retval;
+ }
}
}
+
+ return 0;
}
\f
static struct cls_bucket *create_bucket(struct hmap *, size_t hash,
bucket_insert(struct cls_bucket *bucket, struct cls_rule *rule)
{
struct cls_rule *pos;
- LIST_FOR_EACH (pos, struct cls_rule, node.list, &bucket->rules) {
+ LIST_FOR_EACH (pos, node.list, &bucket->rules) {
- if (pos->priority == rule->priority) {
- if (pos->wc.wildcards == rule->wc.wildcards
+ if (pos->flow.priority == rule->flow.priority) {
+ if (pos->flow.wildcards == rule->flow.wildcards
&& rules_match_1wild(pos, rule, rule->table_idx))
{
list_replace(&rule->node.list, &pos->node.list);
return pos;
}
- } else if (pos->priority < rule->priority) {
+ } else if (pos->flow.priority < rule->flow.priority) {
break;
}
}
struct cls_rule *old_rule;
size_t hash;
- hash = flow_hash(&rule->flow, 0);
+ hash = flow_hash_headers(&rule->flow, 0);
old_rule = search_exact_table(cls, hash, &rule->flow);
if (old_rule) {
hmap_remove(&cls->exact_table, &old_rule->node.hmap);
find_bucket(struct hmap *table, size_t hash, const struct cls_rule *rule)
{
struct cls_bucket *bucket;
- HMAP_FOR_EACH_WITH_HASH (bucket, struct cls_bucket, hmap_node, hash,
- table) {
+ HMAP_FOR_EACH_WITH_HASH (bucket, hmap_node, hash, table) {
if (equal_fields(&bucket->fixed, &rule->flow, rule->table_idx)) {
return bucket;
}
rules_match_1wild(const struct cls_rule *fixed, const struct cls_rule *wild,
int field_idx)
{
- return rules_match(fixed, wild, wild->wc.wildcards, wild->wc.nw_src_mask,
+ return rules_match(fixed, wild, wild->flow.wildcards, wild->wc.nw_src_mask,
wild->wc.nw_dst_mask, field_idx);
}
int field_idx)
{
return rules_match(wild1, wild2,
- wild1->wc.wildcards | wild2->wc.wildcards,
+ wild1->flow.wildcards | wild2->flow.wildcards,
wild1->wc.nw_src_mask & wild2->wc.nw_src_mask,
wild1->wc.nw_dst_mask & wild2->wc.nw_dst_mask,
field_idx);
return NULL;
}
- LIST_FOR_EACH (pos, struct cls_rule, node.list, &bucket->rules) {
+ LIST_FOR_EACH (pos, node.list, &bucket->rules) {
if (rules_match_1wild(target, pos, field_idx)) {
return pos;
}
return search_bucket(bucket, field_idx, target);
}
- HMAP_FOR_EACH_WITH_HASH (bucket, struct cls_bucket, hmap_node,
+ HMAP_FOR_EACH_WITH_HASH (bucket, hmap_node,
hash_fields(&target->flow, field_idx), table) {
struct cls_rule *rule = search_bucket(bucket, field_idx, target);
if (rule) {
{
struct cls_rule *rule;
- HMAP_FOR_EACH_WITH_HASH (rule, struct cls_rule, node.hmap,
- hash, &cls->exact_table) {
+ HMAP_FOR_EACH_WITH_HASH (rule, node.hmap, hash, &cls->exact_table) {
- if (flow_equal(&rule->flow, target)) {
+ if (flow_equal_headers(&rule->flow, target)) {
return rule;
}
}
/* Initialization. */
void hmap_init(struct hmap *);
void hmap_destroy(struct hmap *);
+ void hmap_clear(struct hmap *);
void hmap_swap(struct hmap *a, struct hmap *b);
void hmap_moved(struct hmap *hmap);
static inline size_t hmap_count(const struct hmap *);
void hmap_node_moved(struct hmap *, struct hmap_node *, struct hmap_node *);
static inline void hmap_replace(struct hmap *, const struct hmap_node *old,
- struct hmap_node *new);
+ struct hmap_node *new_node);
struct hmap_node *hmap_random_node(const struct hmap *);
*
* HMAP_FOR_EACH_WITH_HASH iterates NODE over all of the nodes in HMAP that
* have hash value equal to HASH. HMAP_FOR_EACH_IN_BUCKET iterates NODE over
- * all of the nodes in HMAP that would fall in the same bucket as HASH. STRUCT
- * and MEMBER must be the name of the struct that contains the 'struct
- * hmap_node' and the name of the 'struct hmap_node' member, respectively.
+ * all of the nodes in HMAP that would fall in the same bucket as HASH. MEMBER
+ * must be the name of the 'struct hmap_node' member within NODE.
*
* These macros may be used interchangeably to search for a particular value in
* an hmap, see, e.g. shash_find() for an example. Usually, using
*
* HASH is only evaluated once.
*/
- #define HMAP_FOR_EACH_WITH_HASH(NODE, STRUCT, MEMBER, HASH, HMAP) \
- for ((NODE) = CONTAINER_OF(hmap_first_with_hash(HMAP, HASH), \
- STRUCT, MEMBER); \
+ #define HMAP_FOR_EACH_WITH_HASH(NODE, MEMBER, HASH, HMAP) \
+ for ((NODE) = OBJECT_CONTAINING(hmap_first_with_hash(HMAP, HASH), \
+ NODE, MEMBER); \
&(NODE)->MEMBER != NULL; \
- (NODE) = CONTAINER_OF(hmap_next_with_hash(&(NODE)->MEMBER), \
- STRUCT, MEMBER))
- #define HMAP_FOR_EACH_IN_BUCKET(NODE, STRUCT, MEMBER, HASH, HMAP) \
- for ((NODE) = CONTAINER_OF(hmap_first_in_bucket(HMAP, HASH), \
- STRUCT, MEMBER); \
+ (NODE) = OBJECT_CONTAINING(hmap_next_with_hash(&(NODE)->MEMBER), \
+ NODE, MEMBER))
+ #define HMAP_FOR_EACH_IN_BUCKET(NODE, MEMBER, HASH, HMAP) \
+ for ((NODE) = OBJECT_CONTAINING(hmap_first_in_bucket(HMAP, HASH), \
+ NODE, MEMBER); \
&(NODE)->MEMBER != NULL; \
- (NODE) = CONTAINER_OF(hmap_next_in_bucket(&(NODE)->MEMBER), \
- STRUCT, MEMBER))
+ (NODE) = OBJECT_CONTAINING(hmap_next_in_bucket(&(NODE)->MEMBER), \
+ NODE, MEMBER))
static inline struct hmap_node *hmap_first_with_hash(const struct hmap *,
size_t hash);
* The _SAFE version is needed when NODE may be freed. It is not needed when
* NODE may be removed from the hash map but its members remain accessible and
* intact. */
- #define HMAP_FOR_EACH(NODE, STRUCT, MEMBER, HMAP) \
- for ((NODE) = CONTAINER_OF(hmap_first(HMAP), STRUCT, MEMBER); \
- &(NODE)->MEMBER != NULL; \
- (NODE) = CONTAINER_OF(hmap_next(HMAP, &(NODE)->MEMBER), \
- STRUCT, MEMBER))
-
- #define HMAP_FOR_EACH_SAFE(NODE, NEXT, STRUCT, MEMBER, HMAP) \
- for ((NODE) = CONTAINER_OF(hmap_first(HMAP), STRUCT, MEMBER); \
- (&(NODE)->MEMBER != NULL \
- ? (NEXT) = CONTAINER_OF(hmap_next(HMAP, &(NODE)->MEMBER), \
- STRUCT, MEMBER), 1 \
- : 0); \
+ #define HMAP_FOR_EACH(NODE, MEMBER, HMAP) \
+ for ((NODE) = OBJECT_CONTAINING(hmap_first(HMAP), NODE, MEMBER); \
+ &(NODE)->MEMBER != NULL; \
+ (NODE) = OBJECT_CONTAINING(hmap_next(HMAP, &(NODE)->MEMBER), \
+ NODE, MEMBER))
+
+ #define HMAP_FOR_EACH_SAFE(NODE, NEXT, MEMBER, HMAP) \
+ for ((NODE) = OBJECT_CONTAINING(hmap_first(HMAP), NODE, MEMBER); \
+ (&(NODE)->MEMBER != NULL \
+ ? (NEXT) = OBJECT_CONTAINING(hmap_next(HMAP, &(NODE)->MEMBER), \
+ NODE, MEMBER), 1 \
+ : 0); \
(NODE) = (NEXT))
static inline struct hmap_node *hmap_first(const struct hmap *);
#include <time.h>
#include "flow.h"
+ #include "hmap.h"
#include "mac-learning.h"
#include "ofpbuf.h"
#include "ofp-parse.h"
#include "poll-loop.h"
#include "queue.h"
#include "rconn.h"
+ #include "shash.h"
#include "timeval.h"
#include "vconn.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(learning_switch)
+ struct lswitch_port {
+ struct hmap_node hmap_node; /* Hash node for port number. */
+ uint16_t port_no; /* OpenFlow port number, in host byte order. */
+ uint32_t queue_id; /* OpenFlow queue number. */
+ };
+
struct lswitch {
/* If nonnegative, the switch sets up flows that expire after the given
* number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
struct mac_learning *ml; /* NULL to act as hub instead of switch. */
uint32_t wildcards; /* Wildcards to apply to flows. */
bool action_normal; /* Use OFPP_NORMAL? */
- uint32_t queue; /* OpenFlow queue to use, or UINT32_MAX. */
+
+ /* Queue distribution. */
+ uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */
+ struct hmap queue_numbers; /* Map from port number to lswitch_port. */
+ struct shash queue_names; /* Map from port name to lswitch_port. */
/* Number of outgoing queued packets on the rconn. */
struct rconn_packet_counter *queued;
static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
static void send_features_request(struct lswitch *, struct rconn *);
- static void send_default_flows(struct lswitch *sw, struct rconn *rconn,
- FILE *default_flows);
typedef void packet_handler_func(struct lswitch *, struct rconn *, void *);
static packet_handler_func process_switch_features;
static packet_handler_func process_packet_in;
static packet_handler_func process_echo_request;
- /* Creates and returns a new learning switch.
- *
- * If 'learn_macs' is true, the new switch will learn the ports on which MAC
- * addresses appear. Otherwise, the new switch will flood all packets.
- *
- * If 'max_idle' is nonnegative, the new switch will set up flows that expire
- * after the given number of seconds (or never expire, if 'max_idle' is
- * OFP_FLOW_PERMANENT). Otherwise, the new switch will process every packet.
- *
- * The caller may provide the file stream 'default_flows' that defines
- * default flows that should be pushed when a switch connects. Each
- * line is a flow entry in the format described for "add-flows" command
- * in the Flow Syntax section of the ovs-ofct(8) man page. The caller
- * is responsible for closing the stream.
+ /* Creates and returns a new learning switch whose configuration is given by
+ * 'cfg'.
*
* 'rconn' is used to send out an OpenFlow features request. */
struct lswitch *
- lswitch_create(struct rconn *rconn, bool learn_macs,
- bool exact_flows, int max_idle, bool action_normal,
- FILE *default_flows)
+ lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
{
+ const struct ofpbuf *b;
struct lswitch *sw;
sw = xzalloc(sizeof *sw);
- sw->max_idle = max_idle;
+ sw->max_idle = cfg->max_idle;
sw->datapath_id = 0;
sw->last_features_request = time_now() - 1;
- sw->ml = learn_macs ? mac_learning_create() : NULL;
- sw->action_normal = action_normal;
- if (exact_flows) {
+ sw->ml = cfg->mode == LSW_LEARN ? mac_learning_create() : NULL;
+ sw->action_normal = cfg->mode == LSW_NORMAL;
+ if (cfg->exact_flows) {
/* Exact match. */
sw->wildcards = 0;
} else {
sw->wildcards = (OFPFW_DL_TYPE | OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK
| OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST);
}
- sw->queue = UINT32_MAX;
+
+ sw->default_queue = cfg->default_queue;
+ hmap_init(&sw->queue_numbers);
+ shash_init(&sw->queue_names);
+ if (cfg->port_queues) {
+ struct shash_node *node;
+
+ SHASH_FOR_EACH (node, cfg->port_queues) {
+ struct lswitch_port *port = xmalloc(sizeof *port);
+ hmap_node_nullify(&port->hmap_node);
+ port->queue_id = (uintptr_t) node->data;
+ shash_add(&sw->queue_names, node->name, port);
+ }
+ }
+
sw->queued = rconn_packet_counter_create();
send_features_request(sw, rconn);
- if (default_flows) {
- send_default_flows(sw, rconn, default_flows);
+
+ for (b = cfg->default_flows; b; b = b->next) {
+ queue_tx(sw, rconn, ofpbuf_clone(b));
}
+
return sw;
}
lswitch_destroy(struct lswitch *sw)
{
if (sw) {
+ struct lswitch_port *node, *next;
+
+ HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
+ hmap_remove(&sw->queue_numbers, &node->hmap_node);
+ free(node);
+ }
+ shash_destroy(&sw->queue_names);
mac_learning_destroy(sw->ml);
rconn_packet_counter_destroy(sw->queued);
free(sw);
}
}
- /* Sets 'queue' as the OpenFlow queue used by packets and flows set up by 'sw'.
- * Specify UINT32_MAX to avoid specifying a particular queue, which is also the
- * default if this function is never called for 'sw'. */
- void
- lswitch_set_queue(struct lswitch *sw, uint32_t queue)
- {
- sw->queue = queue;
- }
-
/* Takes care of necessary 'sw' activity, except for receiving packets (which
* the caller must do). */
void
lswitch_run(struct lswitch *sw)
{
if (sw->ml) {
- mac_learning_run(sw->ml, NULL);
+ mac_learning_run(sw->ml);
}
}
}
}
if (VLOG_IS_DBG_ENABLED()) {
- char *p = ofp_to_string(msg->data, msg->size, 2);
+ char *s = ofp_to_string(msg->data, msg->size, 2);
VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
- sw->datapath_id, p);
- free(p);
+ sw->datapath_id, s);
+ free(s);
}
}
\f
}
}
- static void
- send_default_flows(struct lswitch *sw, struct rconn *rconn,
- FILE *default_flows)
- {
- char line[1024];
-
- while (fgets(line, sizeof line, default_flows)) {
- struct ofpbuf *b;
- struct ofp_flow_mod *ofm;
- uint16_t priority, idle_timeout, hard_timeout;
- uint64_t cookie;
- struct ofp_match match;
-
- char *comment;
-
- /* Delete comments. */
- comment = strchr(line, '#');
- if (comment) {
- *comment = '\0';
- }
-
- /* Drop empty lines. */
- if (line[strspn(line, " \t\n")] == '\0') {
- continue;
- }
-
- /* Parse and send. str_to_flow() will expand and reallocate the data
- * in 'buffer', so we can't keep pointers to across the str_to_flow()
- * call. */
- make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &b);
- parse_ofp_str(line, &match, b,
- NULL, NULL, &priority, &idle_timeout, &hard_timeout,
- &cookie);
- ofm = b->data;
- ofm->match = match;
- ofm->command = htons(OFPFC_ADD);
- ofm->cookie = htonll(cookie);
- ofm->idle_timeout = htons(idle_timeout);
- ofm->hard_timeout = htons(hard_timeout);
- ofm->buffer_id = htonl(UINT32_MAX);
- ofm->priority = htons(priority);
-
- update_openflow_length(b);
- queue_tx(sw, rconn, b);
- }
- }
-
static void
queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
{
void *osf_)
{
struct ofp_switch_features *osf = osf_;
+ size_t n_ports;
+ size_t i;
+
+ if (check_ofp_message_array(&osf->header, OFPT_FEATURES_REPLY,
+ sizeof *osf, sizeof *osf->ports, &n_ports)) {
+ return;
+ }
sw->datapath_id = ntohll(osf->datapath_id);
+
+ for (i = 0; i < n_ports; i++) {
+ struct ofp_phy_port *opp = &osf->ports[i];
+ struct lswitch_port *lp;
+
+ opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0';
+ lp = shash_find_data(&sw->queue_names, (char *) opp->name);
+ if (lp && hmap_node_is_null(&lp->hmap_node)) {
+ lp->port_no = ntohs(opp->port_no);
+ hmap_insert(&sw->queue_numbers, &lp->hmap_node,
+ hash_int(lp->port_no, 0));
+ }
+ }
}
static uint16_t
return out_port;
}
+ static uint32_t
+ get_queue_id(const struct lswitch *sw, uint16_t in_port)
+ {
+ const struct lswitch_port *port;
+
+ HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
+ &sw->queue_numbers) {
+ if (port->port_no == in_port) {
+ return port->queue_id;
+ }
+ }
+
+ return sw->default_queue;
+ }
+
static void
process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_)
{
struct ofp_packet_in *opi = opi_;
uint16_t in_port = ntohs(opi->in_port);
+ uint32_t queue_id;
uint16_t out_port;
struct ofp_action_header actions[2];
out_port = lswitch_choose_destination(sw, &flow);
/* Make actions. */
+ queue_id = get_queue_id(sw, in_port);
if (out_port == OFPP_NONE) {
actions_len = 0;
- } else if (sw->queue == UINT32_MAX || out_port >= OFPP_MAX) {
+ } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
struct ofp_action_output oao;
memset(&oao, 0, sizeof oao);
oae.type = htons(OFPAT_ENQUEUE);
oae.len = htons(sizeof oae);
oae.port = htons(out_port);
- oae.queue_id = htonl(sw->queue);
+ oae.queue_id = htonl(queue_id);
memcpy(actions, &oae, sizeof oae);
actions_len = sizeof oae;
#include <stdbool.h>
#include <stddef.h>
#include "util.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
/* Doubly linked list head or element. */
struct list {
size_t list_size(const struct list *);
bool list_is_empty(const struct list *);
- #define LIST_FOR_EACH(ITER, STRUCT, MEMBER, LIST) \
- for (ITER = CONTAINER_OF((LIST)->next, STRUCT, MEMBER); \
- &(ITER)->MEMBER != (LIST); \
- ITER = CONTAINER_OF((ITER)->MEMBER.next, STRUCT, MEMBER))
- #define LIST_FOR_EACH_REVERSE(ITER, STRUCT, MEMBER, LIST) \
- for (ITER = CONTAINER_OF((LIST)->prev, STRUCT, MEMBER); \
- &(ITER)->MEMBER != (LIST); \
- ITER = CONTAINER_OF((ITER)->MEMBER.prev, STRUCT, MEMBER))
- #define LIST_FOR_EACH_SAFE(ITER, NEXT, STRUCT, MEMBER, LIST) \
- for (ITER = CONTAINER_OF((LIST)->next, STRUCT, MEMBER); \
- (NEXT = CONTAINER_OF((ITER)->MEMBER.next, STRUCT, MEMBER), \
- &(ITER)->MEMBER != (LIST)); \
+ #define LIST_FOR_EACH(ITER, MEMBER, LIST) \
+ for (ITER = OBJECT_CONTAINING((LIST)->next, ITER, MEMBER); \
+ &(ITER)->MEMBER != (LIST); \
+ ITER = OBJECT_CONTAINING((ITER)->MEMBER.next, ITER, MEMBER))
+ #define LIST_FOR_EACH_REVERSE(ITER, MEMBER, LIST) \
+ for (ITER = OBJECT_CONTAINING((LIST)->prev, ITER, MEMBER); \
+ &(ITER)->MEMBER != (LIST); \
+ ITER = OBJECT_CONTAINING((ITER)->MEMBER.prev, ITER, MEMBER))
+ #define LIST_FOR_EACH_SAFE(ITER, NEXT, MEMBER, LIST) \
+ for (ITER = OBJECT_CONTAINING((LIST)->next, ITER, MEMBER); \
+ (NEXT = OBJECT_CONTAINING((ITER)->MEMBER.next, ITER, MEMBER), \
+ &(ITER)->MEMBER != (LIST)); \
ITER = NEXT)
+
+#ifdef __cplusplus
+}
+#endif
#endif /* list.h */
uint16_t vlan)
{
struct mac_entry *e;
- LIST_FOR_EACH (e, struct mac_entry, hash_node, bucket) {
+ LIST_FOR_EACH (e, hash_node, bucket) {
if (eth_addr_equals(e->mac, mac) && e->vlan == vlan) {
return e;
}
}
}
-void
-mac_learning_run(struct mac_learning *ml, struct tag_set *set)
+tag_type
+mac_learning_run(struct mac_learning *ml)
{
struct mac_entry *e;
+ tag_type tags = 0;
+
while (get_lru(ml, &e) && time_now() >= e->expires) {
COVERAGE_INC(mac_learning_expired);
- if (set) {
- tag_set_add(set, e->tag);
- }
+ tags |= e->tag;
free_mac_entry(ml, e);
}
+ return tags;
}
void
#include "netdev-provider.h"
#include "netdev-vport.h"
#include "openflow/openflow.h"
-#include "openvswitch/datapath-protocol.h"
#include "openvswitch/tunnel.h"
+#include "openvswitch/xflow.h"
#include "packets.h"
#include "socket-util.h"
#include "vlog.h"
struct tnl_port_config *config)
{
struct shash_node *node;
+ bool ipsec_ip_set = false;
+ bool ipsec_mech_set = false;
memset(config, 0, sizeof *config);
config->flags |= TNL_F_PMTUD;
+ config->flags |= TNL_F_HDR_CACHE;
SHASH_FOR_EACH (node, args) {
if (!strcmp(node->name, "remote_ip")) {
if (!strcmp(node->data, "false")) {
config->flags &= ~TNL_F_PMTUD;
}
+ } else if (!strcmp(node->name, "header_cache")) {
+ if (!strcmp(node->data, "false")) {
+ config->flags &= ~TNL_F_HDR_CACHE;
+ }
+ } else if (!strcmp(node->name, "ipsec_local_ip")) {
+ ipsec_ip_set = true;
+ } else if (!strcmp(node->name, "ipsec_cert")
+ || !strcmp(node->name, "ipsec_psk")) {
+ ipsec_mech_set = true;
} else {
VLOG_WARN("%s: unknown %s argument '%s'", name, type, node->name);
}
}
+ /* IPsec doesn't work when header caching is enabled. Disable it if
+ * the IPsec local IP address and authentication mechanism have been
+ * defined. */
+ if (ipsec_ip_set && ipsec_mech_set) {
+ VLOG_INFO("%s: header caching disabled due to use of IPsec", name);
+ config->flags &= ~TNL_F_HDR_CACHE;
+ }
+
if (!config->daddr) {
VLOG_WARN("%s: %s type requires valid 'remote_ip' argument", name, type);
return EINVAL;
const struct shash *args, struct netdev_dev **netdev_devp)
{
int err;
- struct odp_vport_add ova;
+ struct xflow_vport_add ova;
struct tnl_port_config port_config;
struct netdev_dev_tunnel *netdev_dev;
return err;
}
- err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_ADD, &ova);
if (err == EBUSY) {
VLOG_WARN("%s: destroying existing device", name);
- err = netdev_vport_do_ioctl(ODP_VPORT_DEL, ova.devname);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_DEL, ova.devname);
if (err) {
return err;
}
- err = netdev_vport_do_ioctl(ODP_VPORT_ADD, &ova);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_ADD, &ova);
}
if (err) {
netdev_tunnel_reconfigure(struct netdev_dev *netdev_dev_, const struct shash *args)
{
const char *name = netdev_dev_get_name(netdev_dev_);
- struct odp_vport_mod ovm;
+ struct xflow_vport_mod ovm;
struct tnl_port_config port_config;
int err;
return err;
}
- return netdev_vport_do_ioctl(ODP_VPORT_MOD, &ovm);
+ return netdev_vport_do_ioctl(XFLOW_VPORT_MOD, &ovm);
}
static void
{
struct netdev_dev_tunnel *netdev_dev = netdev_dev_tunnel_cast(netdev_dev_);
- netdev_vport_do_ioctl(ODP_VPORT_DEL, (char *)netdev_dev_get_name(netdev_dev_));
+ netdev_vport_do_ioctl(XFLOW_VPORT_DEL, (char *)netdev_dev_get_name(netdev_dev_));
free(netdev_dev);
}
#include "list.h"
#include "netdev-vport.h"
-#include "openvswitch/datapath-protocol.h"
+#include "openvswitch/xflow.h"
#include "shash.h"
#include "socket-util.h"
#include "vlog.h"
netdev_vport_set_etheraddr(struct netdev *netdev,
const uint8_t mac[ETH_ADDR_LEN])
{
- struct odp_vport_ether vport_ether;
+ struct xflow_vport_ether vport_ether;
int err;
ovs_strlcpy(vport_ether.devname, netdev_get_name(netdev),
memcpy(vport_ether.ether_addr, mac, ETH_ADDR_LEN);
- err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_SET, &vport_ether);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_ETHER_SET, &vport_ether);
if (err) {
return err;
}
netdev_vport_get_etheraddr(const struct netdev *netdev,
uint8_t mac[ETH_ADDR_LEN])
{
- struct odp_vport_ether vport_ether;
+ struct xflow_vport_ether vport_ether;
int err;
ovs_strlcpy(vport_ether.devname, netdev_get_name(netdev),
sizeof vport_ether.devname);
- err = netdev_vport_do_ioctl(ODP_VPORT_ETHER_GET, &vport_ether);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_ETHER_GET, &vport_ether);
if (err) {
return err;
}
int
netdev_vport_get_mtu(const struct netdev *netdev, int *mtup)
{
- struct odp_vport_mtu vport_mtu;
+ struct xflow_vport_mtu vport_mtu;
int err;
ovs_strlcpy(vport_mtu.devname, netdev_get_name(netdev),
sizeof vport_mtu.devname);
- err = netdev_vport_do_ioctl(ODP_VPORT_MTU_GET, &vport_mtu);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_MTU_GET, &vport_mtu);
if (err) {
return err;
}
netdev_vport_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
{
const char *name = netdev_get_name(netdev);
- struct odp_vport_stats_req ovsr;
+ struct xflow_vport_stats_req ovsr;
int err;
ovs_strlcpy(ovsr.devname, name, sizeof ovsr.devname);
- err = netdev_vport_do_ioctl(ODP_VPORT_STATS_GET, &ovsr);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_STATS_GET, &ovsr);
if (err) {
return err;
}
int
netdev_vport_set_stats(struct netdev *netdev, const struct netdev_stats *stats)
{
- struct odp_vport_stats_req ovsr;
+ struct xflow_vport_stats_req ovsr;
int err;
ovs_strlcpy(ovsr.devname, netdev_get_name(netdev), sizeof ovsr.devname);
ovsr.stats.rx_crc_err = stats->rx_crc_errors;
ovsr.stats.rx_frame_err = stats->rx_frame_errors;
- err = netdev_vport_do_ioctl(ODP_VPORT_STATS_SET, &ovsr);
+ err = netdev_vport_do_ioctl(XFLOW_VPORT_STATS_SET, &ovsr);
/* If the vport layer doesn't know about the device, that doesn't mean it
* doesn't exist (after all were able to open it when netdev_open() was
if (list) {
struct netdev_vport_notifier *notifier;
- LIST_FOR_EACH (notifier, struct netdev_vport_notifier,
- list_node, list) {
+ LIST_FOR_EACH (notifier, list_node, list) {
struct netdev_notifier *n = ¬ifier->notifier;
n->cb(n);
}
* to the new network device, otherwise to null.
*
* If this is the first time the device has been opened, then create is called
- * before opening. The device is created using the given type and arguments.
+ * before opening. The device is created using the given type and arguments.
*
* 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to
* capture frames of that type received on the device. It may also be one of
return netdev_open(&options, netdevp);
}
+/* Increments the reference count on 'netdev'. Returns 'netdev'. */
+struct netdev *
+netdev_reopen(struct netdev *netdev)
+{
+ netdev->ref_cnt++;
+ return netdev;
+}
+
/* Reconfigures the device 'netdev' with 'args'. 'args' may be empty
* or NULL if none are needed. */
int
return 0;
}
-/* Closes and destroys 'netdev'. */
+/* Decrements the reference count on 'netdev'. If the reference count reaches
+ * zero, closes and destroys 'netdev'. */
void
netdev_close(struct netdev *netdev)
{
- if (netdev) {
+ assert(!netdev || netdev->ref_cnt > 0);
+ if (netdev && !--netdev->ref_cnt) {
struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
assert(netdev_dev->ref_cnt);
* be returned.
*
* Some network devices may not implement support for this function. In such
- * cases this function will always return EOPNOTSUPP.
- */
+ * cases this function will always return EOPNOTSUPP. */
int
netdev_recv(struct netdev *netdev, struct ofpbuf *buffer)
{
* passed-in values are set to 0.
*
* Some network devices may not implement support for this function. In such
- * cases this function will always return EOPNOTSUPP.
- */
+ * cases this function will always return EOPNOTSUPP. */
int
netdev_get_features(struct netdev *netdev,
uint32_t *current, uint32_t *advertised,
*
* - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'.
*
- * 'address' or 'netmask' or both may be null, in which case the address or netmask
- * is not reported. */
+ * 'address' or 'netmask' or both may be null, in which case the address or
+ * netmask is not reported. */
int
netdev_get_in4(const struct netdev *netdev,
struct in_addr *address_, struct in_addr *netmask_)
* the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)).
*
* This function does not modify 'details', and the caller retains ownership of
- * it.
- */
+ * it. */
int
netdev_set_queue(struct netdev *netdev,
unsigned int queue_id, const struct shash *details)
{
memset(netdev, 0, sizeof *netdev);
netdev->netdev_dev = netdev_dev;
+ netdev->ref_cnt = 1;
list_push_back(&netdev_list, &netdev->node);
}
*
* Normally this function only needs to be called from netdev_close().
* However, it may be called by providers due to an error on opening
- * that occurs after initialization. It this case netdev_close() would
+ * that occurs after initialization. In this case netdev_close() would
* never be called. */
void
netdev_uninit(struct netdev *netdev, bool close)
* sets '*devnamep' to the name of a device that has changed and returns 0.
* The caller is responsible for freeing '*devnamep' (with free()).
*
- * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN.
- */
+ * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. */
int
netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep)
{
*devnamep = NULL;
return EAGAIN;
} else {
- *devnamep = xstrdup(node->name);
- shash_delete(&monitor->changed_netdevs, node);
+ *devnamep = shash_steal(&monitor->changed_netdevs, node);
return 0;
}
}
close_all_netdevs(void *aux OVS_UNUSED)
{
struct netdev *netdev, *next;
- LIST_FOR_EACH_SAFE(netdev, next, struct netdev, node, &netdev_list) {
+ LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) {
netdev_close(netdev);
}
}
#include <errno.h>
#include <stdlib.h>
+ #include "dynamic-string.h"
#include "netdev.h"
#include "ofp-util.h"
#include "ofpbuf.h"
#include "socket-util.h"
#include "vconn.h"
#include "vlog.h"
-
+ #include "xtoxll.h"
VLOG_DEFINE_THIS_MODULE(ofp_parse)
- #define DEFAULT_IDLE_TIMEOUT 60
-
static uint32_t
str_to_u32(const char *str)
{
nast->vendor = htonl(NX_VENDOR_ID);
nast->subtype = htons(NXAST_SET_TUNNEL);
nast->tun_id = htonl(str_to_u32(arg));
+ } else if (!strcasecmp(act, "drop_spoofed_arp")) {
+ struct nx_action_header *nah;
+ nah = put_action(b, sizeof *nah, OFPAT_VENDOR);
+ nah->vendor = htonl(NX_VENDOR_ID);
+ nah->subtype = htons(NXAST_DROP_SPOOFED_ARP);
+ } else if (!strcasecmp(act, "set_queue")) {
+ struct nx_action_set_queue *nasq;
+ nasq = put_action(b, sizeof *nasq, OFPAT_VENDOR);
+ nasq->vendor = htonl(NX_VENDOR_ID);
+ nasq->subtype = htons(NXAST_SET_QUEUE);
+ nasq->queue_id = htonl(str_to_u32(arg));
+ } else if (!strcasecmp(act, "pop_queue")) {
+ struct nx_action_header *nah;
+ nah = put_action(b, sizeof *nah, OFPAT_VENDOR);
+ nah->vendor = htonl(NX_VENDOR_ID);
+ nah->subtype = htons(NXAST_POP_QUEUE);
} else if (!strcasecmp(act, "output")) {
put_output_action(b, str_to_u32(arg));
} else if (!strcasecmp(act, "enqueue")) {
char *sp = NULL;
- char *port = strtok_r(arg, ":q", &sp);
+ char *port_s = strtok_r(arg, ":q", &sp);
char *queue = strtok_r(NULL, "", &sp);
- if (port == NULL || queue == NULL) {
+ if (port_s == NULL || queue == NULL) {
ovs_fatal(0, "\"enqueue\" syntax is \"enqueue:PORT:QUEUE\"");
}
- put_enqueue_action(b, str_to_u32(port), str_to_u32(queue));
+ put_enqueue_action(b, str_to_u32(port_s), str_to_u32(queue));
} else if (!strcasecmp(act, "drop")) {
/* A drop action in OpenFlow occurs by just not setting
* an action. */
*priority = OFP_DEFAULT_PRIORITY;
}
if (idle_timeout) {
- *idle_timeout = DEFAULT_IDLE_TIMEOUT;
+ *idle_timeout = OFP_FLOW_PERMANENT;
}
if (hard_timeout) {
*hard_timeout = OFP_FLOW_PERMANENT;
if (table_idx && !strcmp(name, "table")) {
*table_idx = atoi(value);
+ if (*table_idx > 31) {
+ ovs_fatal(0, "table %s is invalid, "
+ "must be between 0 and 31", value);
+ }
} else if (out_port && !strcmp(name, "out_port")) {
*out_port = atoi(value);
} else if (priority && !strcmp(name, "priority")) {
if (!strcmp(value, "*") || !strcmp(value, "ANY")) {
wildcards |= f->wildcard;
} else {
+ uint16_t port_no;
+
wildcards &= ~f->wildcard;
if (f->wildcard == OFPFW_IN_PORT
- && parse_port_name(value, (uint16_t *) data)) {
- /* Nothing to do. */
+ && parse_port_name(value, &port_no)) {
+ match->in_port = htons(port_no);
} else if (f->type == F_U8) {
*(uint8_t *) data = str_to_u32(value);
} else if (f->type == F_U16) {
free(new);
}
}
- * ofpbuf that contains it. */
+
+ /* Parses 'string' as a OFPT_FLOW_MOD with subtype OFPFC_ADD and returns an
-parse_ofp_add_flow_str(char *string)
++ * ofpbuf that contains it. Sets '*table_idx' to the index of the table to
++ * which the flow should be added, or to 0xff if none was specified. */
+ struct ofpbuf *
- parse_ofp_str(string, &match, buffer,
- NULL, NULL, &priority, &idle_timeout, &hard_timeout,
- &cookie);
++parse_ofp_add_flow_str(char *string, uint8_t *table_idx)
+ {
+ struct ofpbuf *buffer;
+ struct ofp_flow_mod *ofm;
+ uint16_t priority, idle_timeout, hard_timeout;
+ uint64_t cookie;
+ struct ofp_match match;
+
+ /* parse_ofp_str() will expand and reallocate the data in 'buffer', so we
+ * can't keep pointers to across the parse_ofp_str() call. */
+ make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer);
-parse_ofp_add_flow_file(FILE *stream)
++ parse_ofp_str(string, &match, buffer, table_idx, NULL, &priority,
++ &idle_timeout, &hard_timeout, &cookie);
+ ofm = buffer->data;
+ ofm->match = match;
+ ofm->command = htons(OFPFC_ADD);
++ if (*table_idx != 0xff) {
++ ofm->command |= htons(*table_idx << 8);
++ }
+ ofm->cookie = htonll(cookie);
+ ofm->idle_timeout = htons(idle_timeout);
+ ofm->hard_timeout = htons(hard_timeout);
+ ofm->buffer_id = htonl(UINT32_MAX);
+ ofm->priority = htons(priority);
+ update_openflow_length(buffer);
+
+ return buffer;
+ }
+
+ /* Parses an OFPT_FLOW_MOD with subtype OFPFC_ADD from 'stream' and returns an
+ * ofpbuf that contains it. Returns a null pointer if end-of-file is reached
+ * before reading a flow. */
+ struct ofpbuf *
- b = parse_ofp_add_flow_str(line);
++parse_ofp_add_flow_file(FILE *stream, uint8_t *table_idx)
+ {
+ struct ofpbuf *b = NULL;
+ struct ds s = DS_EMPTY_INITIALIZER;
+
+ while (!ds_get_line(&s, stream)) {
+ char *line = ds_cstr(&s);
+ char *comment;
+
+ /* Delete comments. */
+ comment = strchr(line, '#');
+ if (comment) {
+ *comment = '\0';
+ }
+
+ /* Drop empty lines. */
+ if (line[strspn(line, " \t\n")] == '\0') {
+ continue;
+ }
+
++ b = parse_ofp_add_flow_str(line, table_idx);
+ break;
+ }
+ ds_destroy(&s);
+
+ return b;
+ }
#define OFP_PARSE_H 1
#include <stdint.h>
+ #include <stdio.h>
struct ofp_match;
struct ofpbuf;
uint16_t *out_port, uint16_t *priority,
uint16_t *idle_timeout, uint16_t *hard_timeout,
uint64_t *cookie);
-struct ofpbuf *parse_ofp_add_flow_str(char *string);
-struct ofpbuf *parse_ofp_add_flow_file(FILE *);
++struct ofpbuf *parse_ofp_add_flow_str(char *string, uint8_t *table_idx);
++struct ofpbuf *parse_ofp_add_flow_file(FILE *, uint8_t *table_idx);
#endif /* ofp-parse.h */
#include "xtoxll.h"
static void ofp_print_port_name(struct ds *string, uint16_t port);
+ static void ofp_print_queue_name(struct ds *string, uint32_t port);
/* Returns a string that represents the contents of the Ethernet frame in the
* 'len' bytes starting at 'data' to 'stream' as output by tcpdump.
packet.data = (void *) op->data;
packet.size = data_len;
flow_extract(&packet, 0, ntohs(op->in_port), &flow);
- flow_to_match(&flow, 0, false, &match);
+ flow_to_match(&flow, false, &match);
ofp_print_match(string, &match, verbosity);
ds_put_char(string, '\n');
}
break;
}
+ case NXAST_DROP_SPOOFED_ARP:
+ ds_put_cstr(string, "drop_spoofed_arp");
+ break;
+
+ case NXAST_SET_QUEUE: {
+ const struct nx_action_set_queue *nasq =
+ (struct nx_action_set_queue *)nah;
+ ds_put_format(string, "set_queue:%u", ntohl(nasq->queue_id));
+ break;
+ }
+
+ case NXAST_POP_QUEUE:
+ ds_put_cstr(string, "pop_queue");
+ break;
+
default:
- ds_put_format(string, "***unknown Nicira action:%d***\n",
+ ds_put_format(string, "***unknown Nicira action:%d***",
ntohs(nah->subtype));
}
}
int verbosity)
{
const struct ofp_flow_mod *ofm = oh;
+ unsigned int command = ntohs(ofm->command);
+ ds_put_char(string, ' ');
ofp_print_match(string, &ofm->match, verbosity);
- switch (ntohs(ofm->command)) {
+ if (ds_last(string) != ' ') {
+ ds_put_char(string, ' ');
+ }
+
+ switch (command & 0xff) {
case OFPFC_ADD:
- ds_put_cstr(string, " ADD: ");
+ ds_put_cstr(string, "ADD:");
break;
case OFPFC_MODIFY:
- ds_put_cstr(string, " MOD: ");
+ ds_put_cstr(string, "MOD:");
break;
case OFPFC_MODIFY_STRICT:
- ds_put_cstr(string, " MOD_STRICT: ");
+ ds_put_cstr(string, "MOD_STRICT:");
break;
case OFPFC_DELETE:
- ds_put_cstr(string, " DEL: ");
+ ds_put_cstr(string, "DEL:");
break;
case OFPFC_DELETE_STRICT:
- ds_put_cstr(string, " DEL_STRICT: ");
+ ds_put_cstr(string, "DEL_STRICT:");
break;
default:
- ds_put_format(string, " cmd:%u ", command);
- ds_put_format(string, "cmd:%d", ntohs(ofm->command));
++ ds_put_format(string, "cmd:%u", command & 0xff);
+ }
+ if (command & 0xff00) {
- ds_put_format(string, "table_id:%u ", command >> 8);
++ ds_put_format(string, " table_id:%u", command >> 8);
+ }
+ if (ofm->cookie != htonll(0)) {
+ ds_put_format(string, " cookie:0x%"PRIx64, ntohll(ofm->cookie));
+ }
+ if (ofm->idle_timeout != htons(OFP_FLOW_PERMANENT)) {
+ ds_put_format(string, " idle:%d", ntohs(ofm->idle_timeout));
}
- ds_put_format(string, "cookie:0x%"PRIx64" idle:%d hard:%d pri:%d "
- "buf:%#x flags:%"PRIx16" ", ntohll(ofm->cookie),
- ntohs(ofm->idle_timeout), ntohs(ofm->hard_timeout),
- ofm->match.wildcards ? ntohs(ofm->priority) : (uint16_t)-1,
- ntohl(ofm->buffer_id), ntohs(ofm->flags));
+ if (ofm->hard_timeout != htons(OFP_FLOW_PERMANENT)) {
+ ds_put_format(string, " hard:%d", ntohs(ofm->hard_timeout));
+ }
+ if (ofm->priority != htons(32768)) {
+ ds_put_format(string, " pri:%"PRIu16, ntohs(ofm->priority));
+ }
+ if (ofm->buffer_id != htonl(UINT32_MAX)) {
+ ds_put_format(string, " buf:%#"PRIx32, ntohl(ofm->buffer_id));
+ }
+ if (ofm->flags != htons(0)) {
+ ds_put_format(string, " flags:%"PRIx16, ntohs(ofm->flags));
+ }
+ ds_put_cstr(string, " ");
ofp_print_actions(string, ofm->actions,
len - offsetof(struct ofp_flow_mod, actions));
ds_put_char(string, '\n');
ds_put_format(string, "**%"PRIu8"**", ofr->reason);
break;
}
- ds_put_format(string,
- " cookie0x%"PRIx64" pri%"PRIu16" secs%"PRIu32" nsecs%"PRIu32
+
+ if (ofr->cookie != htonll(0)) {
+ ds_put_format(string, " cookie:0x%"PRIx64, ntohll(ofr->cookie));
+ }
+ if (ofr->priority != htons(32768)) {
+ ds_put_format(string, " pri:%"PRIu16, ntohs(ofr->priority));
+ }
+ ds_put_format(string, " secs%"PRIu32" nsecs%"PRIu32
" idle%"PRIu16" pkts%"PRIu64" bytes%"PRIu64"\n",
- ntohll(ofr->cookie),
- ofr->match.wildcards ? ntohs(ofr->priority) : (uint16_t)-1,
ntohl(ofr->duration_sec), ntohl(ofr->duration_nsec),
ntohs(ofr->idle_timeout), ntohll(ofr->packet_count),
ntohll(ofr->byte_count));
}
}
+ static void
+ ofp_print_queue_name(struct ds *string, uint32_t queue_id)
+ {
+ if (queue_id == OFPQ_ALL) {
+ ds_put_cstr(string, "ALL");
+ } else {
+ ds_put_format(string, "%"PRIu32, queue_id);
+ }
+ }
+
+ static void
+ ofp_queue_stats_request(struct ds *string, const void *body_,
+ size_t len OVS_UNUSED, int verbosity OVS_UNUSED)
+ {
+ const struct ofp_queue_stats_request *qsr = body_;
+
+ ds_put_cstr(string, "port=");
+ ofp_print_port_name(string, ntohs(qsr->port_no));
+
+ ds_put_cstr(string, " queue=");
+ ofp_print_queue_name(string, ntohl(qsr->queue_id));
+ }
+
+ static void
+ ofp_queue_stats_reply(struct ds *string, const void *body, size_t len,
+ int verbosity)
+ {
+ const struct ofp_queue_stats *qs = body;
+ size_t n = len / sizeof *qs;
+ ds_put_format(string, " %zu queues\n", n);
+ if (verbosity < 1) {
+ return;
+ }
+
+ for (; n--; qs++) {
+ ds_put_cstr(string, " port ");
+ ofp_print_port_name(string, ntohs(qs->port_no));
+ ds_put_cstr(string, " queue ");
+ ofp_print_queue_name(string, ntohl(qs->queue_id));
+ ds_put_cstr(string, ": ");
+
+ print_port_stat(string, "bytes=", ntohll(qs->tx_bytes), 1);
+ print_port_stat(string, "pkts=", ntohll(qs->tx_packets), 1);
+ print_port_stat(string, "errors=", ntohll(qs->tx_errors), 0);
+ }
+ }
+
static void
vendor_stat(struct ds *string, const void *body, size_t len,
int verbosity OVS_UNUSED)
{ 0, SIZE_MAX, ofp_port_stats_reply },
},
{
+ OFPST_QUEUE,
+ "queue",
+ { sizeof(struct ofp_queue_stats_request),
+ sizeof(struct ofp_queue_stats_request),
+ ofp_queue_stats_request },
+ { 0, SIZE_MAX, ofp_queue_stats_reply },
+ },
+ {
OFPST_VENDOR,
"vendor-specific",
{ sizeof(uint32_t), SIZE_MAX, vendor_stat },
ofm->header.length = htons(size);
ofm->cookie = 0;
ofm->match.wildcards = htonl(0);
- ofm->match.in_port = htons(flow->in_port == ODPP_LOCAL ? OFPP_LOCAL
+ ofm->match.in_port = htons(flow->in_port == XFLOWP_LOCAL ? OFPP_LOCAL
: flow->in_port);
memcpy(ofm->match.dl_src, flow->dl_src, sizeof ofm->match.dl_src);
memcpy(ofm->match.dl_dst, flow->dl_dst, sizeof ofm->match.dl_dst);
opo->header.length = htons(size);
opo->header.xid = htonl(0);
opo->buffer_id = htonl(buffer_id);
- opo->in_port = htons(in_port == ODPP_LOCAL ? OFPP_LOCAL : in_port);
+ opo->in_port = htons(in_port == XFLOWP_LOCAL ? XFLOWP_LOCAL : in_port);
opo->actions_len = htons(actions_len);
ofpbuf_put(out, actions, actions_len);
if (packet) {
return 0;
}
++struct ofpbuf *
++make_nxt_flow_mod_table_id(bool enable)
++{
++ struct nxt_flow_mod_table_id *flow_mod_table_id;
++ struct ofpbuf *buffer;
++
++ flow_mod_table_id = make_openflow(sizeof *flow_mod_table_id, OFPT_VENDOR,
++ &buffer);
++
++ flow_mod_table_id->vendor = htonl(NX_VENDOR_ID);
++ flow_mod_table_id->subtype = htonl(NXT_FLOW_MOD_TABLE_ID);
++ flow_mod_table_id->set = enable;
++ return buffer;
++}
++
const struct ofp_flow_stats *
flow_stats_first(struct flow_stats_iterator *iter,
const struct ofp_stats_reply *osr)
switch (ntohs(nah->subtype)) {
case NXAST_RESUBMIT:
case NXAST_SET_TUNNEL:
+ case NXAST_DROP_SPOOFED_ARP:
+ case NXAST_SET_QUEUE:
+ case NXAST_POP_QUEUE:
return check_action_exact_len(a, len, 16);
default:
return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_VENDOR_TYPE);
m->wildcards = htonl(wc);
}
+/* Converts all of the fields in 'opp' from host to native byte-order. */
+void
+hton_ofp_phy_port(struct ofp_phy_port *opp)
+{
+ opp->port_no = htons(opp->port_no);
+ opp->config = htonl(opp->config);
+ opp->state = htonl(opp->state);
+ opp->curr = htonl(opp->curr);
+ opp->advertised = htonl(opp->advertised);
+ opp->supported = htonl(opp->supported);
+ opp->peer = htonl(opp->peer);
+}
+
+/* Converts all of the fields in 'opp' from native to host byte-order. */
+void
+ntoh_ofp_phy_port(struct ofp_phy_port *opp)
+{
+ /* ntohX and htonX are really the same functions. */
+ hton_ofp_phy_port(opp);
+}
+
/* Returns a string that describes 'match' in a very literal way, without
* interpreting its contents except in a very basic fashion. The returned
* string is intended to be fixed-length, so that it is easy to see differences
ntohs(match->tp_src),
ntohs(match->tp_dst));
}
+
+static uint32_t
+vendor_code_to_id(uint8_t code)
+{
+ switch (code) {
+#define OFPUTIL_VENDOR(NAME, VENDOR_ID) case NAME: return VENDOR_ID;
+ default:
+ return UINT32_MAX;
+ }
+}
+
+/* Creates and returns an OpenFlow message of type OFPT_ERROR with the error
+ * information taken from 'error', whose encoding must be as described in the
+ * large comment in ofp-util.h. If 'oh' is nonnull, then the error will use
+ * oh->xid as its transaction ID, and it will include up to the first 64 bytes
+ * of 'oh'.
+ *
+ * Returns NULL if 'error' is not an OpenFlow error code. */
+struct ofpbuf *
+make_ofp_error_msg(int error, const struct ofp_header *oh)
+{
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ struct ofpbuf *buf;
+ const void *data;
+ size_t len;
+ uint8_t vendor;
+ uint16_t type;
+ uint16_t code;
+ uint32_t xid;
+
+ if (!is_ofp_error(error)) {
+ /* We format 'error' with strerror() here since it seems likely to be
+ * a system errno value. */
+ VLOG_WARN_RL(&rl, "invalid OpenFlow error code %d (%s)",
+ error, strerror(error));
+ return NULL;
+ }
+
+ if (oh) {
+ xid = oh->xid;
+ data = oh;
+ len = ntohs(oh->length);
+ if (len > 64) {
+ len = 64;
+ }
+ } else {
+ xid = 0;
+ data = NULL;
+ len = 0;
+ }
+
+ vendor = get_ofp_err_vendor(error);
+ type = get_ofp_err_type(error);
+ code = get_ofp_err_code(error);
+ if (vendor == OFPUTIL_VENDOR_OPENFLOW) {
+ struct ofp_error_msg *oem;
+
+ oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, xid, &buf);
+ oem->type = htons(type);
+ oem->code = htons(code);
+ } else {
+ struct ofp_error_msg *oem;
+ struct nx_vendor_error *ove;
+ uint32_t vendor_id;
+
+ vendor_id = vendor_code_to_id(vendor);
+ if (vendor_id == UINT32_MAX) {
+ VLOG_WARN_RL(&rl, "error %x contains invalid vendor code %d",
+ error, vendor);
+ return NULL;
+ }
+
+ oem = make_openflow_xid(len + sizeof *oem + sizeof *ove,
+ OFPT_ERROR, xid, &buf);
+ oem->type = htons(NXET_VENDOR);
+ oem->code = htons(NXVC_VENDOR_ERROR);
+
+ ove = ofpbuf_put_uninit(buf, sizeof *ove);
+ ove->vendor = htonl(vendor_id);
+ ove->type = htons(type);
+ ove->code = htons(code);
+ }
+
+ if (len) {
+ ofpbuf_put(buf, data, len);
+ }
+
+ return buf;
+}
int check_ofp_packet_out(const struct ofp_header *, struct ofpbuf *data,
int *n_actions, int max_ports);
++struct ofpbuf *make_nxt_flow_mod_table_id(bool enable);
++
struct flow_stats_iterator {
const uint8_t *pos, *end;
};
void normalize_match(struct ofp_match *);
char *ofp_match_to_literal_string(const struct ofp_match *match);
+void hton_ofp_phy_port(struct ofp_phy_port *);
+void ntoh_ofp_phy_port(struct ofp_phy_port *);
+\f
+/* OpenFlow vendors.
+ *
+ * These functions map vendor */
+/* Vendor error numbers currently used in Open vSwitch. */
+#define OFPUTIL_VENDORS \
+ /* vendor name vendor value */ \
+ OFPUTIL_VENDOR(OFPUTIL_VENDOR_OPENFLOW, 0x00000000) \
+ OFPUTIL_VENDOR(OFPUTIL_VENDOR_NICIRA, 0x00002320)
+
+/* OFPUTIL_VENDOR_* definitions. */
+enum ofputil_vendor_codes {
+#define OFPUTIL_VENDOR(NAME, VENDOR_ID) NAME,
+ OFPUTIL_VENDORS
+ OFPUTIL_N_VENDORS
+#undef OFPUTIL_VENDOR
+};
+\f
+/* Error codes.
+ *
+ * We embed system errno values and OpenFlow standard and vendor extension
+ * error codes into a single 31-bit space using the following encoding.
+ * (Bit 31 is unused and assumed 0 to avoid negative "int" values.)
+ *
+ * 31 0
+ * +------------------------------------------------------+
+ * | 0 | success
+ * +------------------------------------------------------+
+ *
+ * 30 29 0
+ * +--+---------------------------------------------------+
+ * |0 | errno value | errno value
+ * +--+---------------------------------------------------+
+ *
+ * 30 29 26 25 16 15 0
+ * +--+-------+----------------+--------------------------+
+ * |1 | 0 | type | code | standard OpenFlow
+ * +--+-------+----------------+--------------------------+ error
+ *
+ * 30 29 26 25 16 15 0
+ * +--+-------+----------------+--------------------------+ Nicira
+ * | 1| vendor| type | code | NXET_VENDOR
+ * +--+-------+----------------+--------------------------+ error extension
+ *
+ * C and POSIX say that errno values are positive. We assume that they are
+ * less than 2**29. They are actually less than 65536 on at least Linux,
+ * FreeBSD, OpenBSD, and Windows.
+ *
+ * The 'vendor' field holds one of the OFPUTIL_VENDOR_* codes defined above.
+ * It must be nonzero.
+ *
+ * Negative values are not defined.
+ */
+
+/* Currently 4 bits are allocated to the "vendor" field. Make sure that all
+ * the vendor codes can fit. */
+BUILD_ASSERT_DECL(OFPUTIL_N_VENDORS <= 16);
+
+/* Returns the standard OpenFlow error with the specified 'type' and 'code' as
+ * an integer. */
static inline int
ofp_mkerr(uint16_t type, uint16_t code)
{
- assert(type > 0 && type <= 0x7fff);
- return (type << 16) | code;
+ return (1 << 30) | (type << 16) | code;
}
+/* Returns the OpenFlow vendor error with the specified 'vendor', 'type', and
+ * 'code' as an integer. 'vendor' must be an OFPUTIL_VENDOR_* constant. */
+static inline int
+ofp_mkerr_vendor(uint8_t vendor, uint16_t type, uint16_t code)
+{
+ assert(vendor < OFPUTIL_N_VENDORS);
+ return (1 << 30) | (vendor << 26) | (type << 16) | code;
+}
+
+/* Returns the OpenFlow vendor error with Nicira as vendor, with the specific
+ * 'type' and 'code', as an integer. */
+static inline int
+ofp_mkerr_nicira(uint16_t type, uint16_t code)
+{
+ return ofp_mkerr_vendor(OFPUTIL_VENDOR_NICIRA, type, code);
+}
+
+/* Returns true if 'error' encodes an OpenFlow standard or vendor extension
+ * error codes as documented above. */
+static inline bool
+is_ofp_error(int error)
+{
+ return (error & (1 << 30)) != 0;
+}
+
+/* Returns true if 'error' appears to be a system errno value. */
+static inline bool
+is_errno(int error)
+{
+ return !is_ofp_error(error);
+}
+
+/* Returns the "vendor" part of the OpenFlow error code 'error' (which must be
+ * in the format explained above). This is normally one of the
+ * OFPUTIL_VENDOR_* constants. Returns OFPUTIL_VENDOR_OPENFLOW (0) for a
+ * standard OpenFlow error. */
+static inline uint8_t
+get_ofp_err_vendor(int error)
+{
+ return (error >> 26) & 0xf;
+}
+
+/* Returns the "type" part of the OpenFlow error code 'error' (which must be in
+ * the format explained above). */
+static inline uint16_t
+get_ofp_err_type(int error)
+{
+ return (error >> 16) & 0x3ff;
+}
+
+/* Returns the "code" part of the OpenFlow error code 'error' (which must be in
+ * the format explained above). */
+static inline uint16_t
+get_ofp_err_code(int error)
+{
+ return error & 0xffff;
+}
+
+struct ofpbuf *make_ofp_error_msg(int error, const struct ofp_header *);
+
#endif /* ofp-util.h */
* caller must not modify or free.
*
* This is probably not the function that you want. You should probably be
- * using dpif_port_poll() or netdev_monitor_create(), which unlike this
+ * using xfif_port_poll() or netdev_monitor_create(), which unlike this
* function are not Linux-specific.
*
* Returns 0 if successful, otherwise a positive errno value. */
change.master_ifindex = (attrs[IFLA_MASTER]
? nl_attr_get_u32(attrs[IFLA_MASTER]) : 0);
- LIST_FOR_EACH (notifier, struct rtnetlink_notifier, node,
- &all_notifiers) {
+ LIST_FOR_EACH (notifier, node, &all_notifiers) {
notifier->cb(&change, notifier->aux);
}
}
{
struct rtnetlink_notifier *notifier;
- LIST_FOR_EACH (notifier, struct rtnetlink_notifier, node,
- &all_notifiers) {
+ LIST_FOR_EACH (notifier, node, &all_notifiers) {
notifier->cb(NULL, notifier->aux);
}
}
VLOG_MODULE(dhcp)
VLOG_MODULE(dhcp_client)
VLOG_MODULE(discovery)
-VLOG_MODULE(dpif)
-VLOG_MODULE(dpif_linux)
-VLOG_MODULE(dpif_netdev)
VLOG_MODULE(dpctl)
VLOG_MODULE(entropy)
VLOG_MODULE(fail_open)
VLOG_MODULE(ofctl)
VLOG_MODULE(ofp_parse)
VLOG_MODULE(ofp_util)
-VLOG_MODULE(ovs_discover)
VLOG_MODULE(ofproto)
VLOG_MODULE(openflowd)
+VLOG_MODULE(ovs_discover)
VLOG_MODULE(ovsdb_client)
VLOG_MODULE(ovsdb_error)
VLOG_MODULE(ovsdb_file)
VLOG_MODULE(ovsdb_idl)
-VLOG_MODULE(ovsdb_log)
VLOG_MODULE(ovsdb_jsonrpc_server)
+VLOG_MODULE(ovsdb_log)
VLOG_MODULE(ovsdb_server)
VLOG_MODULE(ovsdb_tool)
-VLOG_MODULE(pktbuf)
VLOG_MODULE(pcap)
+VLOG_MODULE(pktbuf)
VLOG_MODULE(poll_loop)
VLOG_MODULE(proc_net_compat)
VLOG_MODULE(process)
VLOG_MODULE(reconnect)
VLOG_MODULE(rtnetlink)
VLOG_MODULE(sflow)
+VLOG_MODULE(socket_util)
+VLOG_MODULE(status)
+VLOG_MODULE(stream)
VLOG_MODULE(stream_fd)
VLOG_MODULE(stream_ssl)
VLOG_MODULE(stream_tcp)
VLOG_MODULE(stream_unix)
-VLOG_MODULE(stream)
-VLOG_MODULE(status)
VLOG_MODULE(svec)
VLOG_MODULE(timeval)
VLOG_MODULE(socket_util)
+ VLOG_MODULE(system_stats)
VLOG_MODULE(unixctl)
VLOG_MODULE(util)
-VLOG_MODULE(vconn_stream)
VLOG_MODULE(vconn)
-VLOG_MODULE(vsctl)
+VLOG_MODULE(vconn_stream)
VLOG_MODULE(vlog)
+VLOG_MODULE(vsctl)
VLOG_MODULE(vswitchd)
+VLOG_MODULE(wdp)
+VLOG_MODULE(wdp_xflow)
VLOG_MODULE(xenserver)
+VLOG_MODULE(xfif)
+VLOG_MODULE(xfif_linux)
+VLOG_MODULE(xfif_netdev)
#undef VLOG_MODULE
*/
#include <config.h>
-#include "dpif.h"
+#include "xfif.h"
#include <assert.h>
#include <ctype.h>
#include <sys/stat.h>
#include <unistd.h>
-#include "dpif-provider.h"
#include "netdev.h"
#include "ofpbuf.h"
#include "poll-loop.h"
#include "svec.h"
#include "util.h"
#include "vlog.h"
+#include "xfif-provider.h"
-VLOG_DEFINE_THIS_MODULE(dpif_linux)
+VLOG_DEFINE_THIS_MODULE(xfif_linux)
/* Datapath interface for the openvswitch Linux kernel module. */
-struct dpif_linux {
- struct dpif dpif;
+struct xfif_linux {
+ struct xfif xfif;
int fd;
- /* Used by dpif_linux_get_all_names(). */
+ /* Used by xfif_linux_get_all_names(). */
char *local_ifname;
int minor;
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
-static int do_ioctl(const struct dpif *, int cmd, const void *arg);
+static int do_ioctl(const struct xfif *, int cmd, const void *arg);
static int lookup_minor(const char *name, int *minor);
-static int finish_open(struct dpif *, const char *local_ifname);
+static int finish_open(struct xfif *, const char *local_ifname);
static int get_openvswitch_major(void);
-static int create_minor(const char *name, int minor, struct dpif **dpifp);
-static int open_minor(int minor, struct dpif **dpifp);
+static int create_minor(const char *name, int minor, struct xfif **xfifp);
+static int open_minor(int minor, struct xfif **xfifp);
static int make_openvswitch_device(int minor, char **fnp);
-static void dpif_linux_port_changed(const struct rtnetlink_change *,
- void *dpif);
+static void xfif_linux_port_changed(const struct rtnetlink_change *,
+ void *xfif);
-static struct dpif_linux *
-dpif_linux_cast(const struct dpif *dpif)
+static struct xfif_linux *
+xfif_linux_cast(const struct xfif *xfif)
{
- dpif_assert_class(dpif, &dpif_linux_class);
- return CONTAINER_OF(dpif, struct dpif_linux, dpif);
+ xfif_assert_class(xfif, &xfif_linux_class);
+ return CONTAINER_OF(xfif, struct xfif_linux, xfif);
}
static int
-dpif_linux_enumerate(struct svec *all_dps)
+xfif_linux_enumerate(struct svec *all_dps)
{
int major;
int error;
}
error = 0;
- for (i = 0; i < ODP_MAX; i++) {
- struct dpif *dpif;
+ for (i = 0; i < XFLOW_MAX; i++) {
+ struct xfif *xfif;
char devname[16];
int retval;
sprintf(devname, "dp%d", i);
- retval = dpif_open(devname, "system", &dpif);
+ retval = xfif_open(devname, "system", &xfif);
if (!retval) {
svec_add(all_dps, devname);
- dpif_uninit(dpif, true);
+ xfif_uninit(xfif, true);
} else if (retval != ENODEV && !error) {
error = retval;
}
}
static int
-dpif_linux_open(const char *name, const char *type OVS_UNUSED, bool create,
- struct dpif **dpifp)
+xfif_linux_open(const char *name, const char *type OVS_UNUSED, bool create,
+ struct xfif **xfifp)
{
int minor;
&& isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
if (create) {
if (minor >= 0) {
- return create_minor(name, minor, dpifp);
+ return create_minor(name, minor, xfifp);
} else {
/* Scan for unused minor number. */
- for (minor = 0; minor < ODP_MAX; minor++) {
- int error = create_minor(name, minor, dpifp);
+ for (minor = 0; minor < XFLOW_MAX; minor++) {
+ int error = create_minor(name, minor, xfifp);
if (error != EBUSY) {
return error;
}
return ENOBUFS;
}
} else {
- struct dpif_linux *dpif;
- struct odp_port port;
+ struct xfif_linux *xfif;
+ struct xflow_port port;
int error;
if (minor < 0) {
}
}
- error = open_minor(minor, dpifp);
+ error = open_minor(minor, xfifp);
if (error) {
return error;
}
- dpif = dpif_linux_cast(*dpifp);
+ xfif = xfif_linux_cast(*xfifp);
/* We need the local port's ifindex for the poll function. Start by
* getting the local port's name. */
memset(&port, 0, sizeof port);
- port.port = ODPP_LOCAL;
- if (ioctl(dpif->fd, ODP_PORT_QUERY, &port)) {
+ port.port = XFLOWP_LOCAL;
+ if (ioctl(xfif->fd, XFLOW_PORT_QUERY, &port)) {
error = errno;
if (error != ENODEV) {
VLOG_WARN("%s: probe returned unexpected error: %s",
- dpif_name(*dpifp), strerror(error));
+ xfif_name(*xfifp), strerror(error));
}
- dpif_uninit(*dpifp, true);
+ xfif_uninit(*xfifp, true);
return error;
}
/* Then use that to finish up opening. */
- return finish_open(&dpif->dpif, port.devname);
+ return finish_open(&xfif->xfif, port.devname);
}
}
static void
-dpif_linux_close(struct dpif *dpif_)
+xfif_linux_close(struct xfif *xfif_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- rtnetlink_notifier_unregister(&dpif->port_notifier);
- shash_destroy(&dpif->changed_ports);
- free(dpif->local_ifname);
- close(dpif->fd);
- free(dpif);
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
+ rtnetlink_notifier_unregister(&xfif->port_notifier);
+ shash_destroy(&xfif->changed_ports);
+ free(xfif->local_ifname);
+ close(xfif->fd);
+ free(xfif);
}
static int
-dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names)
+xfif_linux_get_all_names(const struct xfif *xfif_, struct svec *all_names)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
- svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor));
- svec_add(all_names, dpif->local_ifname);
+ svec_add_nocopy(all_names, xasprintf("dp%d", xfif->minor));
+ svec_add(all_names, xfif->local_ifname);
return 0;
}
static int
-dpif_linux_destroy(struct dpif *dpif_)
+xfif_linux_destroy(struct xfif *xfif_)
{
- struct odp_port *ports;
+ struct xflow_port *ports;
size_t n_ports;
int err;
int i;
- err = dpif_port_list(dpif_, &ports, &n_ports);
+ err = xfif_port_list(xfif_, &ports, &n_ports);
if (err) {
return err;
}
for (i = 0; i < n_ports; i++) {
- if (ports[i].port != ODPP_LOCAL) {
- err = do_ioctl(dpif_, ODP_VPORT_DEL, ports[i].devname);
+ if (ports[i].port != XFLOWP_LOCAL) {
+ err = do_ioctl(xfif_, XFLOW_VPORT_DEL, ports[i].devname);
if (err) {
VLOG_WARN_RL(&error_rl, "%s: error deleting port %s (%s)",
- dpif_name(dpif_), ports[i].devname, strerror(err));
+ xfif_name(xfif_), ports[i].devname, strerror(err));
}
}
}
free(ports);
- return do_ioctl(dpif_, ODP_DP_DESTROY, NULL);
+ return do_ioctl(xfif_, XFLOW_DP_DESTROY, NULL);
}
static int
-dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
+xfif_linux_get_stats(const struct xfif *xfif_, struct xflow_stats *stats)
{
memset(stats, 0, sizeof *stats);
- return do_ioctl(dpif_, ODP_DP_STATS, stats);
+ return do_ioctl(xfif_, XFLOW_DP_STATS, stats);
}
static int
-dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
+xfif_linux_get_drop_frags(const struct xfif *xfif_, bool *drop_fragsp)
{
int drop_frags;
int error;
- error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags);
+ error = do_ioctl(xfif_, XFLOW_GET_DROP_FRAGS, &drop_frags);
if (!error) {
*drop_fragsp = drop_frags & 1;
}
}
static int
-dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
+xfif_linux_set_drop_frags(struct xfif *xfif_, bool drop_frags)
{
int drop_frags_int = drop_frags;
- return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
+ return do_ioctl(xfif_, XFLOW_SET_DROP_FRAGS, &drop_frags_int);
}
static int
-dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags,
+xfif_linux_port_add(struct xfif *xfif_, const char *devname, uint16_t flags,
uint16_t *port_no)
{
- struct odp_port port;
+ struct xflow_port port;
int error;
memset(&port, 0, sizeof port);
strncpy(port.devname, devname, sizeof port.devname);
port.flags = flags;
- error = do_ioctl(dpif_, ODP_PORT_ATTACH, &port);
+ error = do_ioctl(xfif_, XFLOW_PORT_ATTACH, &port);
if (!error) {
*port_no = port.port;
}
}
static int
-dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
+xfif_linux_port_del(struct xfif *xfif_, uint16_t port_no)
{
int tmp = port_no;
int err;
- struct odp_port port;
+ struct xflow_port port;
- err = dpif_port_query_by_number(dpif_, port_no, &port);
+ err = xfif_port_query_by_number(xfif_, port_no, &port);
if (err) {
return err;
}
- err = do_ioctl(dpif_, ODP_PORT_DETACH, &tmp);
+ err = do_ioctl(xfif_, XFLOW_PORT_DETACH, &tmp);
if (err) {
return err;
}
/* Try deleting the port if no one has it open. This shouldn't
* actually be necessary unless the config changed while we weren't
* running but it won't hurt anything if the port is already gone. */
- do_ioctl(dpif_, ODP_VPORT_DEL, port.devname);
+ do_ioctl(xfif_, XFLOW_VPORT_DEL, port.devname);
}
return 0;
}
static int
-dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no,
- struct odp_port *port)
+xfif_linux_port_query_by_number(const struct xfif *xfif_, uint16_t port_no,
+ struct xflow_port *port)
{
memset(port, 0, sizeof *port);
port->port = port_no;
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
+ return do_ioctl(xfif_, XFLOW_PORT_QUERY, port);
}
static int
-dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname,
- struct odp_port *port)
+xfif_linux_port_query_by_name(const struct xfif *xfif_, const char *devname,
+ struct xflow_port *port)
{
memset(port, 0, sizeof *port);
strncpy(port->devname, devname, sizeof port->devname);
- return do_ioctl(dpif_, ODP_PORT_QUERY, port);
+ return do_ioctl(xfif_, XFLOW_PORT_QUERY, port);
}
static int
-dpif_linux_flow_flush(struct dpif *dpif_)
+xfif_linux_flow_flush(struct xfif *xfif_)
{
- return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
+ return do_ioctl(xfif_, XFLOW_FLOW_FLUSH, NULL);
}
static int
-dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n)
+xfif_linux_port_list(const struct xfif *xfif_, struct xflow_port *ports, int n)
{
- struct odp_portvec pv;
+ struct xflow_portvec pv;
int error;
pv.ports = ports;
pv.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_LIST, &pv);
+ error = do_ioctl(xfif_, XFLOW_PORT_LIST, &pv);
return error ? -error : pv.n_ports;
}
static int
-dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
+xfif_linux_port_poll(const struct xfif *xfif_, char **devnamep)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
- if (dpif->change_error) {
- dpif->change_error = false;
- shash_clear(&dpif->changed_ports);
+ if (xfif->change_error) {
+ xfif->change_error = false;
+ shash_clear(&xfif->changed_ports);
return ENOBUFS;
- } else if (!shash_is_empty(&dpif->changed_ports)) {
- struct shash_node *node = shash_first(&dpif->changed_ports);
- *devnamep = shash_steal(&dpif->changed_ports, node);
+ } else if (!shash_is_empty(&xfif->changed_ports)) {
+ struct shash_node *node = shash_first(&xfif->changed_ports);
- *devnamep = xstrdup(node->name);
- shash_delete(&xfif->changed_ports, node);
++ *devnamep = shash_steal(&xfif->changed_ports, node);
return 0;
} else {
return EAGAIN;
}
static void
-dpif_linux_port_poll_wait(const struct dpif *dpif_)
+xfif_linux_port_poll_wait(const struct xfif *xfif_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) {
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
+ if (!shash_is_empty(&xfif->changed_ports) || xfif->change_error) {
poll_immediate_wake();
} else {
rtnetlink_notifier_wait();
}
static int
-dpif_linux_port_group_get(const struct dpif *dpif_, int group,
+xfif_linux_port_group_get(const struct xfif *xfif_, int group,
uint16_t ports[], int n)
{
- struct odp_port_group pg;
+ struct xflow_port_group pg;
int error;
assert(n <= UINT16_MAX);
pg.group = group;
pg.ports = ports;
pg.n_ports = n;
- error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg);
+ error = do_ioctl(xfif_, XFLOW_PORT_GROUP_GET, &pg);
return error ? -error : pg.n_ports;
}
static int
-dpif_linux_port_group_set(struct dpif *dpif_, int group,
+xfif_linux_port_group_set(struct xfif *xfif_, int group,
const uint16_t ports[], int n)
{
- struct odp_port_group pg;
+ struct xflow_port_group pg;
assert(n <= UINT16_MAX);
pg.group = group;
pg.ports = (uint16_t *) ports;
pg.n_ports = n;
- return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg);
+ return do_ioctl(xfif_, XFLOW_PORT_GROUP_SET, &pg);
}
static int
-dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
+xfif_linux_flow_get(const struct xfif *xfif_, struct xflow_flow flows[], int n)
{
- struct odp_flowvec fv;
+ struct xflow_flowvec fv;
fv.flows = flows;
fv.n_flows = n;
- return do_ioctl(dpif_, ODP_FLOW_GET, &fv);
+ return do_ioctl(xfif_, XFLOW_FLOW_GET, &fv);
}
static int
-dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put)
+xfif_linux_flow_put(struct xfif *xfif_, struct xflow_flow_put *put)
{
- return do_ioctl(dpif_, ODP_FLOW_PUT, put);
+ return do_ioctl(xfif_, XFLOW_FLOW_PUT, put);
}
static int
-dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow)
+xfif_linux_flow_del(struct xfif *xfif_, struct xflow_flow *flow)
{
- return do_ioctl(dpif_, ODP_FLOW_DEL, flow);
+ return do_ioctl(xfif_, XFLOW_FLOW_DEL, flow);
}
static int
-dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n)
+xfif_linux_flow_list(const struct xfif *xfif_, struct xflow_flow flows[], int n)
{
- struct odp_flowvec fv;
+ struct xflow_flowvec fv;
int error;
fv.flows = flows;
fv.n_flows = n;
- error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv);
+ error = do_ioctl(xfif_, XFLOW_FLOW_LIST, &fv);
return error ? -error : fv.n_flows;
}
static int
-dpif_linux_execute(struct dpif *dpif_, uint16_t in_port,
- const union odp_action actions[], int n_actions,
+xfif_linux_execute(struct xfif *xfif_, uint16_t in_port,
+ const union xflow_action actions[], int n_actions,
const struct ofpbuf *buf)
{
- struct odp_execute execute;
+ struct xflow_execute execute;
memset(&execute, 0, sizeof execute);
execute.in_port = in_port;
- execute.actions = (union odp_action *) actions;
+ execute.actions = (union xflow_action *) actions;
execute.n_actions = n_actions;
execute.data = buf->data;
execute.length = buf->size;
- return do_ioctl(dpif_, ODP_EXECUTE, &execute);
+ return do_ioctl(xfif_, XFLOW_EXECUTE, &execute);
}
static int
-dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
+xfif_linux_recv_get_mask(const struct xfif *xfif_, int *listen_mask)
{
- return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
+ return do_ioctl(xfif_, XFLOW_GET_LISTEN_MASK, listen_mask);
}
static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+xfif_linux_recv_set_mask(struct xfif *xfif_, int listen_mask)
{
- return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
+ return do_ioctl(xfif_, XFLOW_SET_LISTEN_MASK, &listen_mask);
}
static int
-dpif_linux_get_sflow_probability(const struct dpif *dpif_,
+xfif_linux_get_sflow_probability(const struct xfif *xfif_,
uint32_t *probability)
{
- return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability);
+ return do_ioctl(xfif_, XFLOW_GET_SFLOW_PROBABILITY, probability);
}
static int
-dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
+xfif_linux_set_sflow_probability(struct xfif *xfif_, uint32_t probability)
{
- return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability);
+ return do_ioctl(xfif_, XFLOW_SET_SFLOW_PROBABILITY, &probability);
}
static int
-dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
+xfif_linux_queue_to_priority(const struct xfif *xfif OVS_UNUSED,
uint32_t queue_id, uint32_t *priority)
{
if (queue_id < 0xf000) {
}
static int
-dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp)
+xfif_linux_recv(struct xfif *xfif_, struct ofpbuf **bufp)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
struct ofpbuf *buf;
int retval;
int error;
- buf = ofpbuf_new(65536 + XFIF_RECV_MSG_PADDING);
- ofpbuf_reserve(buf, XFIF_RECV_MSG_PADDING);
- buf = ofpbuf_new_with_headroom(65536, DPIF_RECV_MSG_PADDING);
- retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
++ buf = ofpbuf_new_with_headroom(65536, XFIF_RECV_MSG_PADDING);
+ retval = read(xfif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
if (retval < 0) {
error = errno;
if (error != EAGAIN) {
VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
- dpif_name(dpif_), strerror(error));
+ xfif_name(xfif_), strerror(error));
}
- } else if (retval >= sizeof(struct odp_msg)) {
- struct odp_msg *msg = buf->data;
+ } else if (retval >= sizeof(struct xflow_msg)) {
+ struct xflow_msg *msg = buf->data;
if (msg->length <= retval) {
buf->size += retval;
*bufp = buf;
} else {
VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
"from %"PRIu32" bytes to %d",
- dpif_name(dpif_), msg->length, retval);
+ xfif_name(xfif_), msg->length, retval);
error = ERANGE;
}
} else if (!retval) {
- VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_));
+ VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", xfif_name(xfif_));
error = EPROTO;
} else {
VLOG_WARN_RL(&error_rl,
"%s: discarding too-short message (%d bytes)",
- dpif_name(dpif_), retval);
+ xfif_name(xfif_), retval);
error = ERANGE;
}
}
static void
-dpif_linux_recv_wait(struct dpif *dpif_)
+xfif_linux_recv_wait(struct xfif *xfif_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- poll_fd_wait(dpif->fd, POLLIN);
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
+ poll_fd_wait(xfif->fd, POLLIN);
}
-const struct dpif_class dpif_linux_class = {
+const struct xfif_class xfif_linux_class = {
"system",
NULL,
NULL,
- dpif_linux_enumerate,
- dpif_linux_open,
- dpif_linux_close,
- dpif_linux_get_all_names,
- dpif_linux_destroy,
- dpif_linux_get_stats,
- dpif_linux_get_drop_frags,
- dpif_linux_set_drop_frags,
- dpif_linux_port_add,
- dpif_linux_port_del,
- dpif_linux_port_query_by_number,
- dpif_linux_port_query_by_name,
- dpif_linux_port_list,
- dpif_linux_port_poll,
- dpif_linux_port_poll_wait,
- dpif_linux_port_group_get,
- dpif_linux_port_group_set,
- dpif_linux_flow_get,
- dpif_linux_flow_put,
- dpif_linux_flow_del,
- dpif_linux_flow_flush,
- dpif_linux_flow_list,
- dpif_linux_execute,
- dpif_linux_recv_get_mask,
- dpif_linux_recv_set_mask,
- dpif_linux_get_sflow_probability,
- dpif_linux_set_sflow_probability,
- dpif_linux_queue_to_priority,
- dpif_linux_recv,
- dpif_linux_recv_wait,
+ xfif_linux_enumerate,
+ xfif_linux_open,
+ xfif_linux_close,
+ xfif_linux_get_all_names,
+ xfif_linux_destroy,
+ xfif_linux_get_stats,
+ xfif_linux_get_drop_frags,
+ xfif_linux_set_drop_frags,
+ xfif_linux_port_add,
+ xfif_linux_port_del,
+ xfif_linux_port_query_by_number,
+ xfif_linux_port_query_by_name,
+ xfif_linux_port_list,
+ xfif_linux_port_poll,
+ xfif_linux_port_poll_wait,
+ xfif_linux_port_group_get,
+ xfif_linux_port_group_set,
+ xfif_linux_flow_get,
+ xfif_linux_flow_put,
+ xfif_linux_flow_del,
+ xfif_linux_flow_flush,
+ xfif_linux_flow_list,
+ xfif_linux_execute,
+ xfif_linux_recv_get_mask,
+ xfif_linux_recv_set_mask,
+ xfif_linux_get_sflow_probability,
+ xfif_linux_set_sflow_probability,
+ xfif_linux_queue_to_priority,
+ xfif_linux_recv,
+ xfif_linux_recv_wait,
};
\f
static int get_openvswitch_major(void);
static int get_major(const char *target);
static int
-do_ioctl(const struct dpif *dpif_, int cmd, const void *arg)
+do_ioctl(const struct xfif *xfif_, int cmd, const void *arg)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, cmd, arg) ? errno : 0;
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
+ return ioctl(xfif->fd, cmd, arg) ? errno : 0;
}
static int
VLOG_WARN("%s ethtool bus_info has unexpected format", name);
error = EPROTOTYPE;
goto error_close_sock;
- } else if (port_no != ODPP_LOCAL) {
+ } else if (port_no != XFLOWP_LOCAL) {
/* This is an Open vSwitch device but not the local port. We
* intentionally support only using the name of the local port as the
* name of a datapath; otherwise, it would be too difficult to
return major;
}
} else {
- static bool warned;
- if (!warned) {
- VLOG_WARN("%s:%d: syntax error", fn, ln);
- }
- warned = true;
+ VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln);
}
}
}
static int
-finish_open(struct dpif *dpif_, const char *local_ifname)
+finish_open(struct xfif *xfif_, const char *local_ifname)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- dpif->local_ifname = xstrdup(local_ifname);
- dpif->local_ifindex = if_nametoindex(local_ifname);
- if (!dpif->local_ifindex) {
+ struct xfif_linux *xfif = xfif_linux_cast(xfif_);
+ xfif->local_ifname = xstrdup(local_ifname);
+ xfif->local_ifindex = if_nametoindex(local_ifname);
+ if (!xfif->local_ifindex) {
int error = errno;
- dpif_uninit(dpif_, true);
+ xfif_uninit(xfif_, true);
VLOG_WARN("could not get ifindex of %s device: %s",
local_ifname, strerror(errno));
return error;
}
static int
-create_minor(const char *name, int minor, struct dpif **dpifp)
+create_minor(const char *name, int minor, struct xfif **xfifp)
{
- int error = open_minor(minor, dpifp);
+ int error = open_minor(minor, xfifp);
if (!error) {
- error = do_ioctl(*dpifp, ODP_DP_CREATE, name);
+ error = do_ioctl(*xfifp, XFLOW_DP_CREATE, name);
if (!error) {
- error = finish_open(*dpifp, name);
+ error = finish_open(*xfifp, name);
} else {
- dpif_uninit(*dpifp, true);
+ xfif_uninit(*xfifp, true);
}
}
return error;
}
static int
-open_minor(int minor, struct dpif **dpifp)
+open_minor(int minor, struct xfif **xfifp)
{
int error;
char *fn;
fd = open(fn, O_RDONLY | O_NONBLOCK);
if (fd >= 0) {
- struct dpif_linux *dpif = xmalloc(sizeof *dpif);
- error = rtnetlink_notifier_register(&dpif->port_notifier,
- dpif_linux_port_changed, dpif);
+ struct xfif_linux *xfif = xmalloc(sizeof *xfif);
+ error = rtnetlink_notifier_register(&xfif->port_notifier,
+ xfif_linux_port_changed, xfif);
if (!error) {
char *name;
name = xasprintf("dp%d", minor);
- dpif_init(&dpif->dpif, &dpif_linux_class, name, minor, minor);
+ xfif_init(&xfif->xfif, &xfif_linux_class, name, minor, minor);
free(name);
- dpif->fd = fd;
- dpif->local_ifname = NULL;
- dpif->minor = minor;
- dpif->local_ifindex = 0;
- shash_init(&dpif->changed_ports);
- dpif->change_error = false;
- *dpifp = &dpif->dpif;
+ xfif->fd = fd;
+ xfif->local_ifname = NULL;
+ xfif->minor = minor;
+ xfif->local_ifindex = 0;
+ shash_init(&xfif->changed_ports);
+ xfif->change_error = false;
+ *xfifp = &xfif->xfif;
} else {
- free(dpif);
+ free(xfif);
}
} else {
error = errno;
}
static void
-dpif_linux_port_changed(const struct rtnetlink_change *change, void *dpif_)
+xfif_linux_port_changed(const struct rtnetlink_change *change, void *xfif_)
{
- struct dpif_linux *dpif = dpif_;
+ struct xfif_linux *xfif = xfif_;
if (change) {
- if (change->master_ifindex == dpif->local_ifindex
+ if (change->master_ifindex == xfif->local_ifindex
&& (change->nlmsg_type == RTM_NEWLINK
|| change->nlmsg_type == RTM_DELLINK))
{
/* Our datapath changed, either adding a new port or deleting an
* existing one. */
- shash_add_once(&dpif->changed_ports, change->ifname, NULL);
+ shash_add_once(&xfif->changed_ports, change->ifname, NULL);
}
} else {
- dpif->change_error = true;
+ xfif->change_error = true;
}
}
--- /dev/null
- LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) {
+/*
+ * Copyright (c) 2009, 2010 Nicira Networks.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+#include "xfif.h"
+
+#include <assert.h>
+#include <ctype.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <netinet/in.h>
+#include <sys/socket.h>
+#include <net/if.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#include "csum.h"
+#include "flow.h"
+#include "hmap.h"
+#include "list.h"
+#include "netdev.h"
+#include "xflow-util.h"
+#include "ofp-print.h"
+#include "ofpbuf.h"
+#include "packets.h"
+#include "poll-loop.h"
+#include "queue.h"
+#include "timeval.h"
+#include "util.h"
+#include "vlog.h"
+#include "xfif-provider.h"
+
+VLOG_DEFINE_THIS_MODULE(xfif_netdev)
+
+/* Configuration parameters. */
+enum { N_QUEUES = 2 }; /* Number of queues for xfif_recv(). */
+enum { MAX_QUEUE_LEN = 100 }; /* Maximum number of packets per queue. */
+enum { N_GROUPS = 16 }; /* Number of port groups. */
+enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
+enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
+
+/* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
+ * headers to be aligned on a 4-byte boundary. */
+enum { XF_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
+
+/* Datapath based on the network device interface from netdev.h. */
+struct xf_netdev {
+ struct list node;
+ int xf_idx;
+ int open_cnt;
+ bool destroyed;
+
+ bool drop_frags; /* Drop all IP fragments, if true. */
+ struct ovs_queue queues[N_QUEUES]; /* Messages queued for xfif_recv(). */
+ struct hmap flow_table; /* Flow table. */
+ struct xflow_port_group groups[N_GROUPS];
+
+ /* Statistics. */
+ long long int n_frags; /* Number of dropped IP fragments. */
+ long long int n_hit; /* Number of flow table matches. */
+ long long int n_missed; /* Number of flow table misses. */
+ long long int n_lost; /* Number of misses not passed to client. */
+
+ /* Ports. */
+ int n_ports;
+ struct xf_netdev_port *ports[MAX_PORTS];
+ struct list port_list;
+ unsigned int serial;
+};
+
+/* A port in a netdev-based datapath. */
+struct xf_netdev_port {
+ int port_no; /* Index into xf_netdev's 'ports'. */
+ struct list node; /* Element in xf_netdev's 'port_list'. */
+ struct netdev *netdev;
+ bool internal; /* Internal port (as XFLOW_PORT_INTERNAL)? */
+};
+
+/* A flow in xf_netdev's 'flow_table'. */
+struct xf_netdev_flow {
+ struct hmap_node node; /* Element in xf_netdev's 'flow_table'. */
+ struct xflow_key key;
+
+ /* Statistics. */
+ struct timespec used; /* Last used time. */
+ long long int packet_count; /* Number of packets matched. */
+ long long int byte_count; /* Number of bytes matched. */
+ uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */
+
+ /* Actions. */
+ union xflow_action *actions;
+ unsigned int n_actions;
+};
+
+/* Interface to netdev-based datapath. */
+struct xfif_netdev {
+ struct xfif xfif;
+ struct xf_netdev *xf;
+ int listen_mask;
+ unsigned int xf_serial;
+};
+
+/* All netdev-based datapaths. */
+static struct xf_netdev *xf_netdevs[256];
+struct list xf_netdev_list = LIST_INITIALIZER(&xf_netdev_list);
+enum { N_XF_NETDEVS = ARRAY_SIZE(xf_netdevs) };
+
+/* Maximum port MTU seen so far. */
+static int max_mtu = ETH_PAYLOAD_MAX;
+
+static int get_port_by_number(struct xf_netdev *, uint16_t port_no,
+ struct xf_netdev_port **portp);
+static int get_port_by_name(struct xf_netdev *, const char *devname,
+ struct xf_netdev_port **portp);
+static void xf_netdev_free(struct xf_netdev *);
+static void xf_netdev_flow_flush(struct xf_netdev *);
+static int do_add_port(struct xf_netdev *, const char *devname, uint16_t flags,
+ uint16_t port_no);
+static int do_del_port(struct xf_netdev *, uint16_t port_no);
+static int xf_netdev_output_control(struct xf_netdev *, const struct ofpbuf *,
+ int queue_no, int port_no, uint32_t arg);
+static int xf_netdev_execute_actions(struct xf_netdev *,
+ struct ofpbuf *, struct xflow_key *,
+ const union xflow_action *, int n);
+
+static struct xfif_netdev *
+xfif_netdev_cast(const struct xfif *xfif)
+{
+ xfif_assert_class(xfif, &xfif_netdev_class);
+ return CONTAINER_OF(xfif, struct xfif_netdev, xfif);
+}
+
+static struct xf_netdev *
+get_xf_netdev(const struct xfif *xfif)
+{
+ return xfif_netdev_cast(xfif)->xf;
+}
+
+static int
+name_to_xf_idx(const char *name)
+{
+ if (!strncmp(name, "xf", 2) && isdigit((unsigned char)name[2])) {
+ int xf_idx = atoi(name + 2);
+ if (xf_idx >= 0 && xf_idx < N_XF_NETDEVS) {
+ return xf_idx;
+ }
+ }
+ return -1;
+}
+
+static struct xf_netdev *
+find_xf_netdev(const char *name)
+{
+ int xf_idx;
+ size_t i;
+
+ xf_idx = name_to_xf_idx(name);
+ if (xf_idx >= 0) {
+ return xf_netdevs[xf_idx];
+ }
+
+ for (i = 0; i < N_XF_NETDEVS; i++) {
+ struct xf_netdev *xf = xf_netdevs[i];
+ if (xf) {
+ struct xf_netdev_port *port;
+ if (!get_port_by_name(xf, name, &port)) {
+ return xf;
+ }
+ }
+ }
+ return NULL;
+}
+
+static struct xfif *
+create_xfif_netdev(struct xf_netdev *xf)
+{
+ struct xfif_netdev *xfif;
+ char *xfname;
+
+ xf->open_cnt++;
+
+ xfname = xasprintf("xf%d", xf->xf_idx);
+ xfif = xmalloc(sizeof *xfif);
+ xfif_init(&xfif->xfif, &xfif_netdev_class, xfname, xf->xf_idx, xf->xf_idx);
+ xfif->xf = xf;
+ xfif->listen_mask = 0;
+ xfif->xf_serial = xf->serial;
+ free(xfname);
+
+ return &xfif->xfif;
+}
+
+static int
+create_xf_netdev(const char *name, int xf_idx, struct xfif **xfifp)
+{
+ struct xf_netdev *xf;
+ int error;
+ int i;
+
+ if (xf_netdevs[xf_idx]) {
+ return EBUSY;
+ }
+
+ /* Create datapath. */
+ xf_netdevs[xf_idx] = xf = xzalloc(sizeof *xf);
+ list_push_back(&xf_netdev_list, &xf->node);
+ xf->xf_idx = xf_idx;
+ xf->open_cnt = 0;
+ xf->drop_frags = false;
+ for (i = 0; i < N_QUEUES; i++) {
+ queue_init(&xf->queues[i]);
+ }
+ hmap_init(&xf->flow_table);
+ for (i = 0; i < N_GROUPS; i++) {
+ xf->groups[i].ports = NULL;
+ xf->groups[i].n_ports = 0;
+ xf->groups[i].group = i;
+ }
+ list_init(&xf->port_list);
+ error = do_add_port(xf, name, XFLOW_PORT_INTERNAL, XFLOWP_LOCAL);
+ if (error) {
+ xf_netdev_free(xf);
+ return ENODEV;
+ }
+
+ *xfifp = create_xfif_netdev(xf);
+ return 0;
+}
+
+static int
+xfif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create,
+ struct xfif **xfifp)
+{
+ if (create) {
+ if (find_xf_netdev(name)) {
+ return EEXIST;
+ } else {
+ int xf_idx = name_to_xf_idx(name);
+ if (xf_idx >= 0) {
+ return create_xf_netdev(name, xf_idx, xfifp);
+ } else {
+ /* Scan for unused xf_idx number. */
+ for (xf_idx = 0; xf_idx < N_XF_NETDEVS; xf_idx++) {
+ int error = create_xf_netdev(name, xf_idx, xfifp);
+ if (error != EBUSY) {
+ return error;
+ }
+ }
+
+ /* All datapath numbers in use. */
+ return ENOBUFS;
+ }
+ }
+ } else {
+ struct xf_netdev *xf = find_xf_netdev(name);
+ if (xf) {
+ *xfifp = create_xfif_netdev(xf);
+ return 0;
+ } else {
+ return ENODEV;
+ }
+ }
+}
+
+static void
+xf_netdev_free(struct xf_netdev *xf)
+{
+ int i;
+
+ xf_netdev_flow_flush(xf);
+ while (xf->n_ports > 0) {
+ struct xf_netdev_port *port = CONTAINER_OF(
+ xf->port_list.next, struct xf_netdev_port, node);
+ do_del_port(xf, port->port_no);
+ }
+ for (i = 0; i < N_QUEUES; i++) {
+ queue_destroy(&xf->queues[i]);
+ }
+ hmap_destroy(&xf->flow_table);
+ for (i = 0; i < N_GROUPS; i++) {
+ free(xf->groups[i].ports);
+ }
+ xf_netdevs[xf->xf_idx] = NULL;
+ list_remove(&xf->node);
+ free(xf);
+}
+
+static void
+xfif_netdev_close(struct xfif *xfif)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ assert(xf->open_cnt > 0);
+ if (--xf->open_cnt == 0 && xf->destroyed) {
+ xf_netdev_free(xf);
+ }
+ free(xfif);
+}
+
+static int
+xfif_netdev_destroy(struct xfif *xfif)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ xf->destroyed = true;
+ return 0;
+}
+
+static int
+xfif_netdev_get_stats(const struct xfif *xfif, struct xflow_stats *stats)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ memset(stats, 0, sizeof *stats);
+ stats->n_flows = hmap_count(&xf->flow_table);
+ stats->cur_capacity = hmap_capacity(&xf->flow_table);
+ stats->max_capacity = MAX_FLOWS;
+ stats->n_ports = xf->n_ports;
+ stats->max_ports = MAX_PORTS;
+ stats->max_groups = N_GROUPS;
+ stats->n_frags = xf->n_frags;
+ stats->n_hit = xf->n_hit;
+ stats->n_missed = xf->n_missed;
+ stats->n_lost = xf->n_lost;
+ stats->max_miss_queue = MAX_QUEUE_LEN;
+ stats->max_action_queue = MAX_QUEUE_LEN;
+ return 0;
+}
+
+static int
+xfif_netdev_get_drop_frags(const struct xfif *xfif, bool *drop_fragsp)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ *drop_fragsp = xf->drop_frags;
+ return 0;
+}
+
+static int
+xfif_netdev_set_drop_frags(struct xfif *xfif, bool drop_frags)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ xf->drop_frags = drop_frags;
+ return 0;
+}
+
+static int
+do_add_port(struct xf_netdev *xf, const char *devname, uint16_t flags,
+ uint16_t port_no)
+{
+ bool internal = (flags & XFLOW_PORT_INTERNAL) != 0;
+ struct xf_netdev_port *port;
+ struct netdev_options netdev_options;
+ struct netdev *netdev;
+ int mtu;
+ int error;
+
+ /* XXX reject devices already in some xf_netdev. */
+
+ /* Open and validate network device. */
+ memset(&netdev_options, 0, sizeof netdev_options);
+ netdev_options.name = devname;
+ netdev_options.ethertype = NETDEV_ETH_TYPE_ANY;
+ if (internal) {
+ netdev_options.type = "tap";
+ }
+
+ error = netdev_open(&netdev_options, &netdev);
+ if (error) {
+ return error;
+ }
+ /* XXX reject loopback devices */
+ /* XXX reject non-Ethernet devices */
+
+ error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
+ if (error) {
+ netdev_close(netdev);
+ return error;
+ }
+
+ port = xmalloc(sizeof *port);
+ port->port_no = port_no;
+ port->netdev = netdev;
+ port->internal = internal;
+
+ netdev_get_mtu(netdev, &mtu);
+ if (mtu > max_mtu) {
+ max_mtu = mtu;
+ }
+
+ list_push_back(&xf->port_list, &port->node);
+ xf->ports[port_no] = port;
+ xf->n_ports++;
+ xf->serial++;
+
+ return 0;
+}
+
+static int
+xfif_netdev_port_add(struct xfif *xfif, const char *devname, uint16_t flags,
+ uint16_t *port_nop)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ int port_no;
+
+ for (port_no = 0; port_no < MAX_PORTS; port_no++) {
+ if (!xf->ports[port_no]) {
+ *port_nop = port_no;
+ return do_add_port(xf, devname, flags, port_no);
+ }
+ }
+ return EFBIG;
+}
+
+static int
+xfif_netdev_port_del(struct xfif *xfif, uint16_t port_no)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ return port_no == XFLOWP_LOCAL ? EINVAL : do_del_port(xf, port_no);
+}
+
+static bool
+is_valid_port_number(uint16_t port_no)
+{
+ return port_no < MAX_PORTS;
+}
+
+static int
+get_port_by_number(struct xf_netdev *xf,
+ uint16_t port_no, struct xf_netdev_port **portp)
+{
+ if (!is_valid_port_number(port_no)) {
+ *portp = NULL;
+ return EINVAL;
+ } else {
+ *portp = xf->ports[port_no];
+ return *portp ? 0 : ENOENT;
+ }
+}
+
+static int
+get_port_by_name(struct xf_netdev *xf,
+ const char *devname, struct xf_netdev_port **portp)
+{
+ struct xf_netdev_port *port;
+
- HMAP_FOR_EACH_SAFE (flow, next, struct xf_netdev_flow, node,
- &xf->flow_table) {
++ LIST_FOR_EACH (port, node, &xf->port_list) {
+ if (!strcmp(netdev_get_name(port->netdev), devname)) {
+ *portp = port;
+ return 0;
+ }
+ }
+ return ENOENT;
+}
+
+static int
+do_del_port(struct xf_netdev *xf, uint16_t port_no)
+{
+ struct xf_netdev_port *port;
+ char *name;
+ int error;
+
+ error = get_port_by_number(xf, port_no, &port);
+ if (error) {
+ return error;
+ }
+
+ list_remove(&port->node);
+ xf->ports[port->port_no] = NULL;
+ xf->n_ports--;
+ xf->serial++;
+
+ name = xstrdup(netdev_get_name(port->netdev));
+ netdev_close(port->netdev);
+
+ free(name);
+ free(port);
+
+ return 0;
+}
+
+static void
+answer_port_query(const struct xf_netdev_port *port, struct xflow_port *xflow_port)
+{
+ memset(xflow_port, 0, sizeof *xflow_port);
+ ovs_strlcpy(xflow_port->devname, netdev_get_name(port->netdev),
+ sizeof xflow_port->devname);
+ xflow_port->port = port->port_no;
+ xflow_port->flags = port->internal ? XFLOW_PORT_INTERNAL : 0;
+}
+
+static int
+xfif_netdev_port_query_by_number(const struct xfif *xfif, uint16_t port_no,
+ struct xflow_port *xflow_port)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_port *port;
+ int error;
+
+ error = get_port_by_number(xf, port_no, &port);
+ if (!error) {
+ answer_port_query(port, xflow_port);
+ }
+ return error;
+}
+
+static int
+xfif_netdev_port_query_by_name(const struct xfif *xfif, const char *devname,
+ struct xflow_port *xflow_port)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_port *port;
+ int error;
+
+ error = get_port_by_name(xf, devname, &port);
+ if (!error) {
+ answer_port_query(port, xflow_port);
+ }
+ return error;
+}
+
+static void
+xf_netdev_free_flow(struct xf_netdev *xf, struct xf_netdev_flow *flow)
+{
+ hmap_remove(&xf->flow_table, &flow->node);
+ free(flow->actions);
+ free(flow);
+}
+
+static void
+xf_netdev_flow_flush(struct xf_netdev *xf)
+{
+ struct xf_netdev_flow *flow, *next;
+
- LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) {
++ HMAP_FOR_EACH_SAFE (flow, next, node, &xf->flow_table) {
+ xf_netdev_free_flow(xf, flow);
+ }
+}
+
+static int
+xfif_netdev_flow_flush(struct xfif *xfif)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ xf_netdev_flow_flush(xf);
+ return 0;
+}
+
+static int
+xfif_netdev_port_list(const struct xfif *xfif, struct xflow_port *ports, int n)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_port *port;
+ int i;
+
+ i = 0;
- HMAP_FOR_EACH_WITH_HASH (flow, struct xf_netdev_flow, node,
++ LIST_FOR_EACH (port, node, &xf->port_list) {
+ struct xflow_port *xflow_port = &ports[i];
+ if (i >= n) {
+ break;
+ }
+ answer_port_query(port, xflow_port);
+ i++;
+ }
+ return xf->n_ports;
+}
+
+static int
+xfif_netdev_port_poll(const struct xfif *xfif_, char **devnamep OVS_UNUSED)
+{
+ struct xfif_netdev *xfif = xfif_netdev_cast(xfif_);
+ if (xfif->xf_serial != xfif->xf->serial) {
+ xfif->xf_serial = xfif->xf->serial;
+ return ENOBUFS;
+ } else {
+ return EAGAIN;
+ }
+}
+
+static void
+xfif_netdev_port_poll_wait(const struct xfif *xfif_)
+{
+ struct xfif_netdev *xfif = xfif_netdev_cast(xfif_);
+ if (xfif->xf_serial != xfif->xf->serial) {
+ poll_immediate_wake();
+ }
+}
+
+static int
+get_port_group(const struct xfif *xfif, int group_no,
+ struct xflow_port_group **groupp)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+
+ if (group_no >= 0 && group_no < N_GROUPS) {
+ *groupp = &xf->groups[group_no];
+ return 0;
+ } else {
+ *groupp = NULL;
+ return EINVAL;
+ }
+}
+
+static int
+xfif_netdev_port_group_get(const struct xfif *xfif, int group_no,
+ uint16_t ports[], int n)
+{
+ struct xflow_port_group *group;
+ int error;
+
+ if (n < 0) {
+ return -EINVAL;
+ }
+
+ error = get_port_group(xfif, group_no, &group);
+ if (!error) {
+ memcpy(ports, group->ports, MIN(n, group->n_ports) * sizeof *ports);
+ return group->n_ports;
+ } else {
+ return -error;
+ }
+}
+
+static int
+xfif_netdev_port_group_set(struct xfif *xfif, int group_no,
+ const uint16_t ports[], int n)
+{
+ struct xflow_port_group *group;
+ int error;
+
+ if (n < 0 || n > MAX_PORTS) {
+ return EINVAL;
+ }
+
+ error = get_port_group(xfif, group_no, &group);
+ if (!error) {
+ free(group->ports);
+ group->ports = xmemdup(ports, n * sizeof *group->ports);
+ group->n_ports = n;
+ group->group = group_no;
+ }
+ return error;
+}
+
+static struct xf_netdev_flow *
+xf_netdev_lookup_flow(const struct xf_netdev *xf,
+ const struct xflow_key *key)
+{
+ struct xf_netdev_flow *flow;
+
- HMAP_FOR_EACH (flow, struct xf_netdev_flow, node, &xf->flow_table) {
++ HMAP_FOR_EACH_WITH_HASH (flow, node,
+ xflow_key_hash(key, 0), &xf->flow_table) {
+ if (xflow_key_equal(&flow->key, key)) {
+ return flow;
+ }
+ }
+ return NULL;
+}
+
+static void
+answer_flow_query(struct xf_netdev_flow *flow, uint32_t query_flags,
+ struct xflow_flow *xflow_flow)
+{
+ if (flow) {
+ xflow_flow->key = flow->key;
+ xflow_flow->stats.n_packets = flow->packet_count;
+ xflow_flow->stats.n_bytes = flow->byte_count;
+ xflow_flow->stats.used_sec = flow->used.tv_sec;
+ xflow_flow->stats.used_nsec = flow->used.tv_nsec;
+ xflow_flow->stats.tcp_flags = TCP_FLAGS(flow->tcp_ctl);
+ xflow_flow->stats.reserved = 0;
+ xflow_flow->stats.error = 0;
+ if (xflow_flow->n_actions > 0) {
+ unsigned int n = MIN(xflow_flow->n_actions, flow->n_actions);
+ memcpy(xflow_flow->actions, flow->actions,
+ n * sizeof *xflow_flow->actions);
+ xflow_flow->n_actions = flow->n_actions;
+ }
+
+ if (query_flags & XFLOWFF_ZERO_TCP_FLAGS) {
+ flow->tcp_ctl = 0;
+ }
+
+ } else {
+ xflow_flow->stats.error = ENOENT;
+ }
+}
+
+static int
+xfif_netdev_flow_get(const struct xfif *xfif, struct xflow_flow flows[], int n)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ int i;
+
+ for (i = 0; i < n; i++) {
+ struct xflow_flow *xflow_flow = &flows[i];
+ answer_flow_query(xf_netdev_lookup_flow(xf, &xflow_flow->key),
+ xflow_flow->flags, xflow_flow);
+ }
+ return 0;
+}
+
+static int
+xfif_netdev_validate_actions(const union xflow_action *actions, int n_actions,
+ bool *mutates)
+{
+ unsigned int i;
+
+ *mutates = false;
+ for (i = 0; i < n_actions; i++) {
+ const union xflow_action *a = &actions[i];
+ switch (a->type) {
+ case XFLOWAT_OUTPUT:
+ if (a->output.port >= MAX_PORTS) {
+ return EINVAL;
+ }
+ break;
+
+ case XFLOWAT_OUTPUT_GROUP:
+ *mutates = true;
+ if (a->output_group.group >= N_GROUPS) {
+ return EINVAL;
+ }
+ break;
+
+ case XFLOWAT_CONTROLLER:
+ break;
+
+ case XFLOWAT_SET_DL_TCI:
+ *mutates = true;
+ if (a->dl_tci.mask != htons(VLAN_VID_MASK)
+ && a->dl_tci.mask != htons(VLAN_PCP_MASK)
+ && a->dl_tci.mask != htons(VLAN_VID_MASK | VLAN_PCP_MASK)) {
+ return EINVAL;
+ }
+ if (a->dl_tci.tci & ~a->dl_tci.mask){
+ return EINVAL;
+ }
+ break;
+
+ case XFLOWAT_SET_NW_TOS:
+ *mutates = true;
+ if (a->nw_tos.nw_tos & IP_ECN_MASK) {
+ return EINVAL;
+ }
+ break;
+
+ case XFLOWAT_STRIP_VLAN:
+ case XFLOWAT_SET_DL_SRC:
+ case XFLOWAT_SET_DL_DST:
+ case XFLOWAT_SET_NW_SRC:
+ case XFLOWAT_SET_NW_DST:
+ case XFLOWAT_SET_TP_SRC:
+ case XFLOWAT_SET_TP_DST:
+ *mutates = true;
+ break;
+
+ default:
+ return EOPNOTSUPP;
+ }
+ }
+ return 0;
+}
+
+static int
+set_flow_actions(struct xf_netdev_flow *flow, struct xflow_flow *xflow_flow)
+{
+ size_t n_bytes;
+ bool mutates;
+ int error;
+
+ if (xflow_flow->n_actions >= 4096 / sizeof *xflow_flow->actions) {
+ return EINVAL;
+ }
+ error = xfif_netdev_validate_actions(xflow_flow->actions,
+ xflow_flow->n_actions, &mutates);
+ if (error) {
+ return error;
+ }
+
+ n_bytes = xflow_flow->n_actions * sizeof *flow->actions;
+ flow->actions = xrealloc(flow->actions, n_bytes);
+ flow->n_actions = xflow_flow->n_actions;
+ memcpy(flow->actions, xflow_flow->actions, n_bytes);
+ return 0;
+}
+
+static int
+add_flow(struct xfif *xfif, struct xflow_flow *xflow_flow)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_flow *flow;
+ int error;
+
+ flow = xzalloc(sizeof *flow);
+ flow->key = xflow_flow->key;
+
+ error = set_flow_actions(flow, xflow_flow);
+ if (error) {
+ free(flow);
+ return error;
+ }
+
+ hmap_insert(&xf->flow_table, &flow->node,
+ xflow_key_hash(&flow->key, 0));
+ return 0;
+}
+
+static void
+clear_stats(struct xf_netdev_flow *flow)
+{
+ flow->used.tv_sec = 0;
+ flow->used.tv_nsec = 0;
+ flow->packet_count = 0;
+ flow->byte_count = 0;
+ flow->tcp_ctl = 0;
+}
+
+static int
+xfif_netdev_flow_put(struct xfif *xfif, struct xflow_flow_put *put)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_flow *flow;
+
+ flow = xf_netdev_lookup_flow(xf, &put->flow.key);
+ if (!flow) {
+ if (put->flags & XFLOWPF_CREATE) {
+ if (hmap_count(&xf->flow_table) < MAX_FLOWS) {
+ return add_flow(xfif, &put->flow);
+ } else {
+ return EFBIG;
+ }
+ } else {
+ return ENOENT;
+ }
+ } else {
+ if (put->flags & XFLOWPF_MODIFY) {
+ int error = set_flow_actions(flow, &put->flow);
+ if (!error && put->flags & XFLOWPF_ZERO_STATS) {
+ clear_stats(flow);
+ }
+ return error;
+ } else {
+ return EEXIST;
+ }
+ }
+}
+
+
+static int
+xfif_netdev_flow_del(struct xfif *xfif, struct xflow_flow *xflow_flow)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_flow *flow;
+
+ flow = xf_netdev_lookup_flow(xf, &xflow_flow->key);
+ if (flow) {
+ answer_flow_query(flow, 0, xflow_flow);
+ xf_netdev_free_flow(xf, flow);
+ return 0;
+ } else {
+ return ENOENT;
+ }
+}
+
+static int
+xfif_netdev_flow_list(const struct xfif *xfif, struct xflow_flow flows[], int n)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct xf_netdev_flow *flow;
+ int i;
+
+ i = 0;
- LIST_FOR_EACH (xf, struct xf_netdev, node, &xf_netdev_list) {
++ HMAP_FOR_EACH (flow, node, &xf->flow_table) {
+ if (i >= n) {
+ break;
+ }
+ answer_flow_query(flow, 0, &flows[i++]);
+ }
+ return hmap_count(&xf->flow_table);
+}
+
+static int
+xfif_netdev_execute(struct xfif *xfif, uint16_t in_port,
+ const union xflow_action actions[], int n_actions,
+ const struct ofpbuf *packet)
+{
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ struct ofpbuf copy;
+ bool mutates;
+ struct xflow_key key;
+ flow_t flow;
+ int error;
+
+ if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
+ return EINVAL;
+ }
+
+ error = xfif_netdev_validate_actions(actions, n_actions, &mutates);
+ if (error) {
+ return error;
+ }
+
+ if (mutates) {
+ /* We need a deep copy of 'packet' since we're going to modify its
+ * data. */
+ ofpbuf_init(©, XF_NETDEV_HEADROOM + packet->size);
+ copy.data = (char*)copy.base + XF_NETDEV_HEADROOM;
+ ofpbuf_put(©, packet->data, packet->size);
+ } else {
+ /* We still need a shallow copy of 'packet', even though we won't
+ * modify its data, because flow_extract() modifies packet->l2, etc.
+ * We could probably get away with modifying those but it's more polite
+ * if we don't. */
+ copy = *packet;
+ }
+ flow_extract(©, 0, in_port, &flow);
+ xflow_key_from_flow(&key, &flow);
+ error = xf_netdev_execute_actions(xf, ©, &key, actions, n_actions);
+ if (mutates) {
+ ofpbuf_uninit(©);
+ }
+ return error;
+}
+
+static int
+xfif_netdev_recv_get_mask(const struct xfif *xfif, int *listen_mask)
+{
+ struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif);
+ *listen_mask = xfif_netdev->listen_mask;
+ return 0;
+}
+
+static int
+xfif_netdev_recv_set_mask(struct xfif *xfif, int listen_mask)
+{
+ struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif);
+ if (!(listen_mask & ~XFLOWL_ALL)) {
+ xfif_netdev->listen_mask = listen_mask;
+ return 0;
+ } else {
+ return EINVAL;
+ }
+}
+
+static struct ovs_queue *
+find_nonempty_queue(struct xfif *xfif)
+{
+ struct xfif_netdev *xfif_netdev = xfif_netdev_cast(xfif);
+ struct xf_netdev *xf = get_xf_netdev(xfif);
+ int mask = xfif_netdev->listen_mask;
+ int i;
+
+ for (i = 0; i < N_QUEUES; i++) {
+ struct ovs_queue *q = &xf->queues[i];
+ if (q->n && mask & (1u << i)) {
+ return q;
+ }
+ }
+ return NULL;
+}
+
+static int
+xfif_netdev_recv(struct xfif *xfif, struct ofpbuf **bufp)
+{
+ struct ovs_queue *q = find_nonempty_queue(xfif);
+ if (q) {
+ *bufp = queue_pop_head(q);
+ return 0;
+ } else {
+ return EAGAIN;
+ }
+}
+
+static void
+xfif_netdev_recv_wait(struct xfif *xfif)
+{
+ struct ovs_queue *q = find_nonempty_queue(xfif);
+ if (q) {
+ poll_immediate_wake();
+ } else {
+ /* No messages ready to be received, and xf_wait() will ensure that we
+ * wake up to queue new messages, so there is nothing to do. */
+ }
+}
+\f
+static void
+xf_netdev_flow_used(struct xf_netdev_flow *flow,
+ const struct xflow_key *key,
+ const struct ofpbuf *packet)
+{
+ time_timespec(&flow->used);
+ flow->packet_count++;
+ flow->byte_count += packet->size;
+ if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
+ struct tcp_header *th = packet->l4;
+ flow->tcp_ctl |= th->tcp_ctl;
+ }
+}
+
+static void
+xf_netdev_port_input(struct xf_netdev *xf, struct xf_netdev_port *port,
+ struct ofpbuf *packet)
+{
+ struct xf_netdev_flow *flow;
+ struct xflow_key key;
+ flow_t f;
+
+ if (packet->size < ETH_HEADER_LEN) {
+ return;
+ }
+ if (flow_extract(packet, 0, port->port_no, &f) && xf->drop_frags) {
+ xf->n_frags++;
+ return;
+ }
+ xflow_key_from_flow(&key, &f);
+
+ flow = xf_netdev_lookup_flow(xf, &key);
+ if (flow) {
+ xf_netdev_flow_used(flow, &key, packet);
+ xf_netdev_execute_actions(xf, packet, &key,
+ flow->actions, flow->n_actions);
+ xf->n_hit++;
+ } else {
+ xf->n_missed++;
+ xf_netdev_output_control(xf, packet, _XFLOWL_MISS_NR, port->port_no, 0);
+ }
+}
+
+static void
+xf_netdev_run(void)
+{
+ struct ofpbuf packet;
+ struct xf_netdev *xf;
+
+ ofpbuf_init(&packet, XF_NETDEV_HEADROOM + max_mtu);
- LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) {
++ LIST_FOR_EACH (xf, node, &xf_netdev_list) {
+ struct xf_netdev_port *port;
+
- LIST_FOR_EACH (xf, struct xf_netdev, node, &xf_netdev_list) {
++ LIST_FOR_EACH (port, node, &xf->port_list) {
+ int error;
+
+ /* Reset packet contents. */
+ packet.data = (char*)packet.base + XF_NETDEV_HEADROOM;
+ packet.size = 0;
+
+ error = netdev_recv(port->netdev, &packet);
+ if (!error) {
+ xf_netdev_port_input(xf, port, &packet);
+ } else if (error != EAGAIN) {
+ struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+ VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
+ netdev_get_name(port->netdev), strerror(error));
+ }
+ }
+ }
+ ofpbuf_uninit(&packet);
+}
+
+static void
+xf_netdev_wait(void)
+{
+ struct xf_netdev *xf;
+
- LIST_FOR_EACH (port, struct xf_netdev_port, node, &xf->port_list) {
++ LIST_FOR_EACH (xf, node, &xf_netdev_list) {
+ struct xf_netdev_port *port;
- struct eth_header *eh = packet->l2;
++ LIST_FOR_EACH (port, node, &xf->port_list) {
+ netdev_recv_wait(port->netdev);
+ }
+ }
+}
+
+
+/* Modify or add a 802.1Q header in 'packet' according to 'a'. */
+static void
+xf_netdev_set_dl_tci(struct ofpbuf *packet,
+ const struct xflow_action_dl_tci *a)
+{
+ struct vlan_eth_header *veh;
+ struct eth_header *eh;
+
+ eh = packet->l2;
+ if (packet->size >= sizeof(struct vlan_eth_header)
+ && eh->eth_type == htons(ETH_TYPE_VLAN)) {
+ veh = packet->l2;
+ veh->veth_tci = (veh->veth_tci & ~a->mask) | a->tci;
+ } else {
+ /* Insert new 802.1Q header. */
+ struct vlan_eth_header tmp;
+ memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
+ memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
+ tmp.veth_type = htons(ETH_TYPE_VLAN);
+ tmp.veth_tci = htons(a->tci);
+ tmp.veth_next_type = eh->eth_type;
+
+ veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN);
+ memcpy(veh, &tmp, sizeof tmp);
+ packet->l2 = (char*)packet->l2 - VLAN_HEADER_LEN;
+ }
+}
+
+static void
+xf_netdev_strip_vlan(struct ofpbuf *packet, struct xflow_key *key)
+{
+ struct vlan_eth_header *veh = packet->l2;
+ if (packet->size >= sizeof *veh
+ && veh->veth_type == htons(ETH_TYPE_VLAN)) {
+ struct eth_header tmp;
+
+ memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
+ memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
+ tmp.eth_type = veh->veth_next_type;
+
+ packet->size -= VLAN_HEADER_LEN;
+ packet->data = (char*)packet->data + VLAN_HEADER_LEN;
+ packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
+ memcpy(packet->data, &tmp, sizeof tmp);
+
+ key->dl_tci = htons(0);
+ }
+}
+
+static void
+xf_netdev_set_dl_src(struct ofpbuf *packet,
+ const uint8_t dl_addr[ETH_ADDR_LEN])
+{
+ struct eth_header *eh = packet->l2;
+ memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
+}
+
+static void
+xf_netdev_set_dl_dst(struct ofpbuf *packet,
+ const uint8_t dl_addr[ETH_ADDR_LEN])
+{
+ struct eth_header *eh = packet->l2;
+ memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
+}
+
+static bool
+is_ip(const struct ofpbuf *packet, const struct xflow_key *key)
+{
+ return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
+}
+
+static void
+xf_netdev_set_nw_addr(struct ofpbuf *packet, const struct xflow_key *key,
+ const struct xflow_action_nw_addr *a)
+{
+ if (is_ip(packet, key)) {
+ struct ip_header *nh = packet->l3;
+ uint32_t *field;
+
+ field = a->type == XFLOWAT_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
+ if (key->nw_proto == IP_TYPE_TCP && packet->l7) {
+ struct tcp_header *th = packet->l4;
+ th->tcp_csum = recalc_csum32(th->tcp_csum, *field, a->nw_addr);
+ } else if (key->nw_proto == IP_TYPE_UDP && packet->l7) {
+ struct udp_header *uh = packet->l4;
+ if (uh->udp_csum) {
+ uh->udp_csum = recalc_csum32(uh->udp_csum, *field, a->nw_addr);
+ if (!uh->udp_csum) {
+ uh->udp_csum = 0xffff;
+ }
+ }
+ }
+ nh->ip_csum = recalc_csum32(nh->ip_csum, *field, a->nw_addr);
+ *field = a->nw_addr;
+ }
+}
+
+static void
+xf_netdev_set_nw_tos(struct ofpbuf *packet, const struct xflow_key *key,
+ const struct xflow_action_nw_tos *a)
+{
+ if (is_ip(packet, key)) {
+ struct ip_header *nh = packet->l3;
+ uint8_t *field = &nh->ip_tos;
+
+ /* Set the DSCP bits and preserve the ECN bits. */
+ uint8_t new = a->nw_tos | (nh->ip_tos & IP_ECN_MASK);
+
+ nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
+ htons((uint16_t)a->nw_tos));
+ *field = new;
+ }
+}
+
+static void
+xf_netdev_set_tp_port(struct ofpbuf *packet, const struct xflow_key *key,
+ const struct xflow_action_tp_port *a)
+{
+ if (is_ip(packet, key)) {
+ uint16_t *field;
+ if (key->nw_proto == IPPROTO_TCP && packet->l7) {
+ struct tcp_header *th = packet->l4;
+ field = a->type == XFLOWAT_SET_TP_SRC ? &th->tcp_src : &th->tcp_dst;
+ th->tcp_csum = recalc_csum16(th->tcp_csum, *field, a->tp_port);
+ *field = a->tp_port;
+ } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
+ struct udp_header *uh = packet->l4;
+ field = a->type == XFLOWAT_SET_TP_SRC ? &uh->udp_src : &uh->udp_dst;
+ uh->udp_csum = recalc_csum16(uh->udp_csum, *field, a->tp_port);
+ *field = a->tp_port;
+ } else {
+ return;
+ }
+ }
+}
+
+static void
+xf_netdev_output_port(struct xf_netdev *xf, struct ofpbuf *packet,
+ uint16_t out_port)
+{
+ struct xf_netdev_port *p = xf->ports[out_port];
+ if (p) {
+ netdev_send(p->netdev, packet);
+ }
+}
+
+static void
+xf_netdev_output_group(struct xf_netdev *xf, uint16_t group, uint16_t in_port,
+ struct ofpbuf *packet)
+{
+ struct xflow_port_group *g = &xf->groups[group];
+ int i;
+
+ for (i = 0; i < g->n_ports; i++) {
+ uint16_t out_port = g->ports[i];
+ if (out_port != in_port) {
+ xf_netdev_output_port(xf, packet, out_port);
+ }
+ }
+}
+
+static int
+xf_netdev_output_control(struct xf_netdev *xf, const struct ofpbuf *packet,
+ int queue_no, int port_no, uint32_t arg)
+{
+ struct ovs_queue *q = &xf->queues[queue_no];
+ struct xflow_msg *header;
+ struct ofpbuf *msg;
+ size_t msg_size;
+
+ if (q->n >= MAX_QUEUE_LEN) {
+ xf->n_lost++;
+ return ENOBUFS;
+ }
+
+ msg_size = sizeof *header + packet->size;
+ msg = ofpbuf_new(msg_size + XFIF_RECV_MSG_PADDING);
+ header = ofpbuf_put_uninit(msg, sizeof *header);
+ ofpbuf_reserve(msg, XFIF_RECV_MSG_PADDING);
+ header->type = queue_no;
+ header->length = msg_size;
+ header->port = port_no;
+ header->arg = arg;
+ ofpbuf_put(msg, packet->data, packet->size);
+ queue_push_tail(q, msg);
+
+ return 0;
+}
+
+/* Returns true if 'packet' is an invalid Ethernet+IPv4 ARP packet: one with
+ * screwy or truncated header fields or one whose inner and outer Ethernet
+ * address differ. */
+static bool
+xf_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct xflow_key *key)
+{
+ struct arp_eth_header *arp;
+ struct eth_header *eth;
+ ptrdiff_t l3_size;
+
+ if (key->dl_type != htons(ETH_TYPE_ARP)) {
+ return false;
+ }
+
+ l3_size = (char *) ofpbuf_end(packet) - (char *) packet->l3;
+ if (l3_size < sizeof(struct arp_eth_header)) {
+ return true;
+ }
+
+ eth = packet->l2;
+ arp = packet->l3;
+ return (arp->ar_hrd != htons(ARP_HRD_ETHERNET)
+ || arp->ar_pro != htons(ARP_PRO_IP)
+ || arp->ar_hln != ETH_HEADER_LEN
+ || arp->ar_pln != 4
+ || !eth_addr_equals(arp->ar_sha, eth->eth_src));
+}
+
+static int
+xf_netdev_execute_actions(struct xf_netdev *xf,
+ struct ofpbuf *packet, struct xflow_key *key,
+ const union xflow_action *actions, int n_actions)
+{
+ int i;
+ for (i = 0; i < n_actions; i++) {
+ const union xflow_action *a = &actions[i];
+
+ switch (a->type) {
+ case XFLOWAT_OUTPUT:
+ xf_netdev_output_port(xf, packet, a->output.port);
+ break;
+
+ case XFLOWAT_OUTPUT_GROUP:
+ xf_netdev_output_group(xf, a->output_group.group, key->in_port,
+ packet);
+ break;
+
+ case XFLOWAT_CONTROLLER:
+ xf_netdev_output_control(xf, packet, _XFLOWL_ACTION_NR,
+ key->in_port, a->controller.arg);
+ break;
+
+ case XFLOWAT_SET_DL_TCI:
+ xf_netdev_set_dl_tci(packet, &a->dl_tci);
+ break;
+
+ case XFLOWAT_STRIP_VLAN:
+ xf_netdev_strip_vlan(packet, key);
+ break;
+
+ case XFLOWAT_SET_DL_SRC:
+ xf_netdev_set_dl_src(packet, a->dl_addr.dl_addr);
+ break;
+
+ case XFLOWAT_SET_DL_DST:
+ xf_netdev_set_dl_dst(packet, a->dl_addr.dl_addr);
+ break;
+
+ case XFLOWAT_SET_NW_SRC:
+ case XFLOWAT_SET_NW_DST:
+ xf_netdev_set_nw_addr(packet, key, &a->nw_addr);
+ break;
+
+ case XFLOWAT_SET_NW_TOS:
+ xf_netdev_set_nw_tos(packet, key, &a->nw_tos);
+ break;
+
+ case XFLOWAT_SET_TP_SRC:
+ case XFLOWAT_SET_TP_DST:
+ xf_netdev_set_tp_port(packet, key, &a->tp_port);
+ break;
+
+ case XFLOWAT_DROP_SPOOFED_ARP:
+ if (xf_netdev_is_spoofed_arp(packet, key)) {
+ return 0;
+ }
+ }
+ }
+ return 0;
+}
+
+const struct xfif_class xfif_netdev_class = {
+ "netdev",
+ xf_netdev_run,
+ xf_netdev_wait,
+ NULL, /* enumerate */
+ xfif_netdev_open,
+ xfif_netdev_close,
+ NULL, /* get_all_names */
+ xfif_netdev_destroy,
+ xfif_netdev_get_stats,
+ xfif_netdev_get_drop_frags,
+ xfif_netdev_set_drop_frags,
+ xfif_netdev_port_add,
+ xfif_netdev_port_del,
+ xfif_netdev_port_query_by_number,
+ xfif_netdev_port_query_by_name,
+ xfif_netdev_port_list,
+ xfif_netdev_port_poll,
+ xfif_netdev_port_poll_wait,
+ xfif_netdev_port_group_get,
+ xfif_netdev_port_group_set,
+ xfif_netdev_flow_get,
+ xfif_netdev_flow_put,
+ xfif_netdev_flow_del,
+ xfif_netdev_flow_flush,
+ xfif_netdev_flow_list,
+ xfif_netdev_execute,
+ xfif_netdev_recv_get_mask,
+ xfif_netdev_recv_set_mask,
+ NULL, /* get_sflow_probability */
+ NULL, /* set_sflow_probability */
+ NULL, /* queue_to_priority */
+ xfif_netdev_recv,
+ xfif_netdev_recv_wait,
+};
--- /dev/null
+/*
+ * Copyright (c) 2009, 2010 Nicira Networks.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+#include "xflow-util.h"
+#include <inttypes.h>
+#include <stdlib.h>
+#include <string.h>
+#include "coverage.h"
+#include "dynamic-string.h"
+#include "flow.h"
+#include "packets.h"
+#include "timeval.h"
+#include "util.h"
+
+union xflow_action *
+xflow_actions_add(struct xflow_actions *actions, uint16_t type)
+{
+ union xflow_action *a;
+ size_t idx;
+
+ idx = actions->n_actions++ & (MAX_XFLOW_ACTIONS - 1);
+ a = &actions->actions[idx];
+ memset(a, 0, sizeof *a);
+ a->type = type;
+ return a;
+}
+
+void
+format_xflow_key(struct ds *ds, const struct xflow_key *key)
+{
+ ds_put_format(ds, "tunnel%"PRIx32":in_port%04x",
+ key->tun_id, key->in_port);
+ if (key->dl_tci) {
+ ds_put_format(ds, ":vlan%"PRIu16":pcp%d",
+ vlan_tci_to_vid(key->dl_tci),
+ vlan_tci_to_pcp(key->dl_tci));
+ }
+ ds_put_format(ds, " mac"ETH_ADDR_FMT"->"ETH_ADDR_FMT" type%04x "
+ "proto%"PRId8" tos%"PRIu8" ip"IP_FMT"->"IP_FMT" port%d->%d",
+ ETH_ADDR_ARGS(key->dl_src), ETH_ADDR_ARGS(key->dl_dst),
+ ntohs(key->dl_type), key->nw_proto, key->nw_tos,
+ IP_ARGS(&key->nw_src), IP_ARGS(&key->nw_dst),
+ ntohs(key->tp_src), ntohs(key->tp_dst));
+}
+
+void
+format_xflow_action(struct ds *ds, const union xflow_action *a)
+{
+ switch (a->type) {
+ case XFLOWAT_OUTPUT:
+ ds_put_format(ds, "%"PRIu16, a->output.port);
+ break;
+ case XFLOWAT_OUTPUT_GROUP:
+ ds_put_format(ds, "g%"PRIu16, a->output_group.group);
+ break;
+ case XFLOWAT_CONTROLLER:
+ ds_put_format(ds, "ctl(%"PRIu32")", a->controller.arg);
+ break;
+ case XFLOWAT_SET_TUNNEL:
+ ds_put_format(ds, "set_tunnel(0x%08"PRIx32")", ntohl(a->tunnel.tun_id));
+ break;
+ case XFLOWAT_SET_DL_TCI:
+ ds_put_format(ds, "set_tci(%04"PRIx16",mask=%04"PRIx16")",
+ ntohs(a->dl_tci.tci), ntohs(a->dl_tci.mask));
+ break;
+ case XFLOWAT_STRIP_VLAN:
+ ds_put_format(ds, "strip_vlan");
+ break;
+ case XFLOWAT_SET_DL_SRC:
+ ds_put_format(ds, "set_dl_src("ETH_ADDR_FMT")",
+ ETH_ADDR_ARGS(a->dl_addr.dl_addr));
+ break;
+ case XFLOWAT_SET_DL_DST:
+ ds_put_format(ds, "set_dl_dst("ETH_ADDR_FMT")",
+ ETH_ADDR_ARGS(a->dl_addr.dl_addr));
+ break;
+ case XFLOWAT_SET_NW_SRC:
+ ds_put_format(ds, "set_nw_src("IP_FMT")",
+ IP_ARGS(&a->nw_addr.nw_addr));
+ break;
+ case XFLOWAT_SET_NW_DST:
+ ds_put_format(ds, "set_nw_dst("IP_FMT")",
+ IP_ARGS(&a->nw_addr.nw_addr));
+ break;
+ case XFLOWAT_SET_NW_TOS:
+ ds_put_format(ds, "set_nw_tos(%"PRIu8")", a->nw_tos.nw_tos);
+ break;
+ case XFLOWAT_SET_TP_SRC:
+ ds_put_format(ds, "set_tp_src(%"PRIu16")", ntohs(a->tp_port.tp_port));
+ break;
+ case XFLOWAT_SET_TP_DST:
+ ds_put_format(ds, "set_tp_dst(%"PRIu16")", ntohs(a->tp_port.tp_port));
+ break;
+ case XFLOWAT_SET_PRIORITY:
+ ds_put_format(ds, "set_priority(0x%"PRIx32")", a->priority.priority);
+ break;
+ case XFLOWAT_POP_PRIORITY:
+ ds_put_cstr(ds, "pop_priority");
+ break;
++ case XFLOWAT_DROP_SPOOFED_ARP:
++ ds_put_cstr(ds, "drop_spoofed_arp");
++ break;
+ default:
+ ds_put_format(ds, "***bad action 0x%"PRIx16"***", a->type);
+ break;
+ }
+}
+
+void
+format_xflow_actions(struct ds *ds, const union xflow_action *actions,
+ size_t n_actions)
+{
+ size_t i;
+ for (i = 0; i < n_actions; i++) {
+ if (i) {
+ ds_put_char(ds, ',');
+ }
+ format_xflow_action(ds, &actions[i]);
+ }
+ if (!n_actions) {
+ ds_put_cstr(ds, "drop");
+ }
+}
+
+void
+format_xflow_flow_stats(struct ds *ds, const struct xflow_flow_stats *s)
+{
+ ds_put_format(ds, "packets:%llu, bytes:%llu, used:",
+ (unsigned long long int) s->n_packets,
+ (unsigned long long int) s->n_bytes);
+ if (s->used_sec) {
+ long long int used = s->used_sec * 1000 + s->used_nsec / 1000000;
+ ds_put_format(ds, "%.3fs", (time_msec() - used) / 1000.0);
+ } else {
+ ds_put_format(ds, "never");
+ }
+}
+
+void
+format_xflow_flow(struct ds *ds, const struct xflow_flow *f)
+{
+ format_xflow_key(ds, &f->key);
+ ds_put_cstr(ds, ", ");
+ format_xflow_flow_stats(ds, &f->stats);
+ ds_put_cstr(ds, ", actions:");
+ format_xflow_actions(ds, f->actions, f->n_actions);
+}
+\f
+void
+xflow_key_from_flow(struct xflow_key *key, const struct flow *flow)
+{
+ key->tun_id = flow->tun_id;
+ key->nw_src = flow->nw_src;
+ key->nw_dst = flow->nw_dst;
+ key->in_port = ofp_port_to_xflow_port(flow->in_port);
+ if (flow->dl_vlan == htons(OFP_VLAN_NONE)) {
+ key->dl_tci = htons(0);
+ } else {
+ uint16_t vid = flow->dl_vlan & htons(VLAN_VID_MASK);
+ uint16_t pcp = htons((flow->dl_vlan_pcp << VLAN_PCP_SHIFT)
+ & VLAN_PCP_MASK);
+ key->dl_tci = vid | pcp | htons(XFLOW_TCI_PRESENT);
+ }
+ key->dl_type = flow->dl_type;
+ key->tp_src = flow->tp_src;
+ key->tp_dst = flow->tp_dst;
+ memcpy(key->dl_src, flow->dl_src, ETH_ADDR_LEN);
+ memcpy(key->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
+ key->nw_proto = flow->nw_proto;
+ key->nw_tos = flow->nw_tos;
+}
+
+void
+xflow_key_to_flow(const struct xflow_key *key, struct flow *flow)
+{
+ flow->wildcards = 0;
+ flow->priority = 0xffff;
+ flow->tun_id = key->tun_id;
+ flow->nw_src = key->nw_src;
+ flow->nw_dst = key->nw_dst;
+ flow->in_port = xflow_port_to_ofp_port(key->in_port);
+ if (key->dl_tci) {
+ flow->dl_vlan = htons(vlan_tci_to_vid(key->dl_tci));
+ flow->dl_vlan_pcp = vlan_tci_to_pcp(key->dl_tci);
+ } else {
+ flow->dl_vlan = htons(OFP_VLAN_NONE);
+ flow->dl_vlan_pcp = 0;
+ }
+ flow->dl_type = key->dl_type;
+ flow->tp_src = key->tp_src;
+ flow->tp_dst = key->tp_dst;
+ memcpy(flow->dl_src, key->dl_src, ETH_ADDR_LEN);
+ memcpy(flow->dl_dst, key->dl_dst, ETH_ADDR_LEN);
+ flow->nw_proto = key->nw_proto;
+ flow->nw_tos = key->nw_tos;
+}
#include <inttypes.h>
#include <stdlib.h>
#include "collectors.h"
-#include "dpif.h"
#include "compiler.h"
+ #include "hash.h"
+ #include "hmap.h"
#include "netdev.h"
#include "ofpbuf.h"
#include "ofproto.h"
+#include "packets.h"
#include "poll-loop.h"
- #include "port-array.h"
#include "sflow_api.h"
#include "socket-util.h"
#include "timeval.h"
#include "vlog.h"
+#include "wdp.h"
+#include "xfif.h"
VLOG_DEFINE_THIS_MODULE(sflow)
struct ofproto_sflow_port {
+ struct hmap_node hmap_node; /* In struct ofproto_sflow's "ports" hmap. */
struct netdev *netdev; /* Underlying network device, for stats. */
SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
- uint16_t odp_port; /* ODP port number. */
++ uint16_t xflow_port; /* xflow port number. */
};
struct ofproto_sflow {
struct collectors *collectors;
SFLAgent *sflow_agent;
struct ofproto_sflow_options *options;
- struct dpif *dpif;
+ struct wdp *wdp;
time_t next_tick;
size_t n_flood, n_all;
- struct port_array ports; /* Indexed by XFLOW port number. */
+ struct hmap ports; /* Contains "struct ofproto_sflow_port"s. */
};
+ static void ofproto_sflow_del_port__(struct ofproto_sflow *,
+ struct ofproto_sflow_port *);
+
#define RECEIVER_INDEX 1
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
collectors_send(os->collectors, pkt, pktLen);
}
-ofproto_sflow_find_port(const struct ofproto_sflow *os, uint16_t odp_port)
+ static struct ofproto_sflow_port *
- hash_int(odp_port, 0), &os->ports) {
- if (osp->odp_port == odp_port) {
++ofproto_sflow_find_port(const struct ofproto_sflow *os, uint16_t xflow_port)
+ {
+ struct ofproto_sflow_port *osp;
+
+ HMAP_FOR_EACH_IN_BUCKET (osp, hmap_node,
++ hash_int(xflow_port, 0), &os->ports) {
++ if (osp->xflow_port == xflow_port) {
+ return osp;
+ }
+ }
+ return NULL;
+ }
+
static void
sflow_agent_get_counters(void *os_, SFLPoller *poller,
SFL_COUNTERS_SAMPLE_TYPE *cs)
enum netdev_flags flags;
uint32_t current;
- osp = port_array_get(&os->ports, poller->bridgePort);
+ osp = ofproto_sflow_find_port(os, poller->bridgePort);
if (!osp) {
return;
}
os->options = NULL;
/* Turn off sampling to save CPU cycles. */
- dpif_set_sflow_probability(os->dpif, 0);
+ wdp_set_sflow_probability(os->wdp, 0);
}
bool
}
struct ofproto_sflow *
-ofproto_sflow_create(struct dpif *dpif)
+ofproto_sflow_create(struct wdp *wdp)
{
struct ofproto_sflow *os;
os = xcalloc(1, sizeof *os);
- os->dpif = dpif;
+ os->wdp = wdp;
os->next_tick = time_now() + 1;
- port_array_init(&os->ports);
+ hmap_init(&os->ports);
return os;
}
ofproto_sflow_destroy(struct ofproto_sflow *os)
{
if (os) {
- struct ofproto_sflow_port *osp;
- unsigned int xflow_port;
+ struct ofproto_sflow_port *osp, *next;
ofproto_sflow_clear(os);
- PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
- ofproto_sflow_del_port(os, xflow_port);
+ HMAP_FOR_EACH_SAFE (osp, next, hmap_node, &os->ports) {
+ ofproto_sflow_del_port__(os, osp);
}
- port_array_destroy(&os->ports);
+ hmap_destroy(&os->ports);
free(os);
}
}
static void
ofproto_sflow_add_poller(struct ofproto_sflow *os,
- struct ofproto_sflow_port *osp, uint16_t odp_port)
+ struct ofproto_sflow_port *osp, uint16_t xflow_port)
{
SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
sflow_agent_get_counters);
sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
- sfl_poller_set_bridgePort(poller, odp_port);
+ sfl_poller_set_bridgePort(poller, xflow_port);
}
static void
}
void
-ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
+ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t xflow_port,
const char *netdev_name)
{
struct ofproto_sflow_port *osp;
uint32_t ifindex;
int error;
- ofproto_sflow_del_port(os, odp_port);
+ ofproto_sflow_del_port(os, xflow_port);
/* Open network device. */
error = netdev_open_default(netdev_name, &netdev);
osp->netdev = netdev;
ifindex = netdev_get_ifindex(netdev);
if (ifindex <= 0) {
- ifindex = (os->sflow_agent->subId << 16) + odp_port;
+ ifindex = (os->sflow_agent->subId << 16) + xflow_port;
}
SFL_DS_SET(osp->dsi, 0, ifindex, 0);
- port_array_set(&os->ports, xflow_port, osp);
- osp->odp_port = odp_port;
- hmap_insert(&os->ports, &osp->hmap_node, hash_int(odp_port, 0));
++ osp->xflow_port = xflow_port;
++ hmap_insert(&os->ports, &osp->hmap_node, hash_int(xflow_port, 0));
/* Add poller and sampler. */
if (os->sflow_agent) {
- ofproto_sflow_add_poller(os, osp, odp_port);
+ ofproto_sflow_add_poller(os, osp, xflow_port);
ofproto_sflow_add_sampler(os, osp);
}
}
+ static void
+ ofproto_sflow_del_port__(struct ofproto_sflow *os,
+ struct ofproto_sflow_port *osp)
+ {
+ if (os->sflow_agent) {
+ sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
+ sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
+ }
+ netdev_close(osp->netdev);
+ hmap_remove(&os->ports, &osp->hmap_node);
+ free(osp);
+ }
+
void
-ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
+ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t xflow_port)
{
- struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
- struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port);
++ struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, xflow_port);
if (osp) {
- if (os->sflow_agent) {
- sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
- sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
- }
- netdev_close(osp->netdev);
- free(osp);
- port_array_delete(&os->ports, xflow_port);
+ ofproto_sflow_del_port__(os, osp);
}
}
struct ofproto_sflow_port *osp;
bool options_changed;
SFLReceiver *receiver;
- unsigned int xflow_port;
SFLAddress agentIP;
time_t now;
sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
/* Set the sampling_rate down in the datapath. */
- dpif_set_sflow_probability(os->dpif,
- MAX(1, UINT32_MAX / options->sampling_rate));
+ wdp_set_sflow_probability(os->wdp,
+ MAX(1, UINT32_MAX / options->sampling_rate));
/* Add samplers and pollers for the currently known ports. */
- PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
- ofproto_sflow_add_poller(os, osp, xflow_port);
+ HMAP_FOR_EACH (osp, hmap_node, &os->ports) {
- ofproto_sflow_add_poller(os, osp, osp->odp_port);
++ ofproto_sflow_add_poller(os, osp, osp->xflow_port);
ofproto_sflow_add_sampler(os, osp);
}
}
static int
-ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os,
- uint16_t odp_port)
+ofproto_sflow_xflow_port_to_ifindex(const struct ofproto_sflow *os,
- uint16_t xflow_port)
++ uint16_t xflow_port)
{
- struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
- struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port);
++ struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, xflow_port);
return osp ? SFL_DS_INDEX(osp->dsi) : 0;
}
void
-ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
+ofproto_sflow_received(struct ofproto_sflow *os, struct xflow_msg *msg)
{
SFL_FLOW_SAMPLE_TYPE fs;
SFLFlow_sample_element hdrElem;
SFLSampled_header *header;
SFLFlow_sample_element switchElem;
SFLSampler *sampler;
- const struct odp_sflow_sample_header *hdr;
- const union odp_action *actions;
+ const struct xflow_sflow_sample_header *hdr;
+ const union xflow_action *actions;
struct ofpbuf payload;
size_t n_actions, n_outputs;
size_t min_size;
flow_t flow;
size_t i;
- /* Get odp_sflow_sample_header. */
+ /* Get xflow_sflow_sample_header. */
min_size = sizeof *msg + sizeof *hdr;
if (min_size > msg->length) {
VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
msg->length, min_size);
return;
}
- hdr = (const struct odp_sflow_sample_header *) (msg + 1);
+ hdr = (const struct xflow_sflow_sample_header *) (msg + 1);
/* Get actions. */
n_actions = hdr->n_actions;
n_actions, msg->length, min_size);
return;
}
- actions = (const union odp_action *) (hdr + 1);
+ actions = (const union xflow_action *) (hdr + 1);
/* Get packet payload and extract flow. */
- payload.data = (union odp_action *) (actions + n_actions);
+ payload.data = (union xflow_action *) (actions + n_actions);
payload.size = msg->length - min_size;
flow_extract(&payload, 0, msg->port, &flow);
/* Build a flow sample */
memset(&fs, 0, sizeof fs);
- fs.input = ofproto_sflow_odp_port_to_ifindex(os, msg->port);
+ fs.input = ofproto_sflow_xflow_port_to_ifindex(os, msg->port);
fs.output = 0; /* Filled in correctly below. */
fs.sample_pool = hdr->sample_pool;
/* Figure out the output ports. */
n_outputs = 0;
for (i = 0; i < n_actions; i++) {
- const union odp_action *a = &actions[i];
+ const union xflow_action *a = &actions[i];
switch (a->type) {
- case ODPAT_OUTPUT:
- fs.output = ofproto_sflow_odp_port_to_ifindex(os, a->output.port);
+ case XFLOWAT_OUTPUT:
+ fs.output = ofproto_sflow_xflow_port_to_ifindex(os, a->output.port);
n_outputs++;
break;
- case ODPAT_OUTPUT_GROUP:
+ case XFLOWAT_OUTPUT_GROUP:
+#if 0
n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
: a->output_group.group == DP_GROUP_ALL ? os->n_all
: 0);
+#endif
break;
- case ODPAT_SET_VLAN_VID:
- switchElem.flowType.sw.dst_vlan = ntohs(a->vlan_vid.vlan_vid);
- break;
-
- case ODPAT_SET_VLAN_PCP:
- switchElem.flowType.sw.dst_priority = a->vlan_pcp.vlan_pcp;
+ case XFLOWAT_SET_DL_TCI:
+ if (a->dl_tci.mask & htons(VLAN_VID_MASK)) {
+ switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(a->dl_tci.tci);
+ }
+ if (a->dl_tci.mask & htons(VLAN_PCP_MASK)) {
+ switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(a->dl_tci.tci);
+ }
break;
default:
#include "classifier.h"
#include "coverage.h"
#include "discovery.h"
-#include "dpif.h"
#include "dynamic-string.h"
#include "fail-open.h"
+ #include "hash.h"
+ #include "hmap.h"
#include "in-band.h"
#include "mac-learning.h"
#include "netdev.h"
#include "netflow.h"
-#include "odp-util.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "ofproto-sflow.h"
#include "ofpbuf.h"
#include "openflow/nicira-ext.h"
#include "openflow/openflow.h"
-#include "openvswitch/datapath-protocol.h"
+#include "openvswitch/xflow.h"
#include "packets.h"
#include "pinsched.h"
#include "pktbuf.h"
#include "poll-loop.h"
- #include "port-array.h"
#include "rconn.h"
#include "shash.h"
#include "status.h"
#include "unixctl.h"
#include "vconn.h"
#include "vlog.h"
+#include "wdp.h"
+#include "xfif.h"
#include "xtoxll.h"
VLOG_DEFINE_THIS_MODULE(ofproto)
#include "sflow_api.h"
-enum {
- TABLEID_HASH = 0,
- TABLEID_CLASSIFIER = 1
-};
-
-
-struct ofport {
- struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
- struct netdev *netdev;
- struct ofp_phy_port opp; /* In host byte order. */
- uint16_t odp_port;
-};
-
-static void ofport_free(struct ofport *);
-static void hton_ofp_phy_port(struct ofp_phy_port *);
-
-static int xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags,
- bool *may_set_up_flow, uint16_t *nf_output_iface);
-
-struct rule {
- struct cls_rule cr;
-
+struct ofproto_rule {
uint64_t flow_cookie; /* Controller-issued identifier.
(Kept in network-byte order.) */
- uint16_t idle_timeout; /* In seconds from time of last use. */
- uint16_t hard_timeout; /* In seconds from time of creation. */
bool send_flow_removed; /* Send a flow removed message? */
- long long int used; /* Last-used time (0 if never used). */
- long long int created; /* Creation time. */
- uint64_t packet_count; /* Number of packets received. */
- uint64_t byte_count; /* Number of bytes received. */
- uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
tag_type tags; /* Tags (set only by hooks). */
- struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
-
- /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
- * exact-match rule (having cr.wc.wildcards of 0) generated from the
- * wildcard rule 'super'. In this case, 'list' is an element of the
- * super-rule's list.
- *
- * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
- * a list of subrules. A super-rule with no wildcards (where
- * cr.wc.wildcards is 0) will never have any subrules. */
- struct rule *super;
- struct list list;
+};
- /* OpenFlow actions.
- *
- * 'n_actions' is the number of elements in the 'actions' array. A single
- * action may take up more more than one element's worth of space.
- *
- * A subrule has no actions (it uses the super-rule's actions). */
- int n_actions;
- union ofp_action *actions;
+static struct ofproto_rule *
+ofproto_rule_cast(const struct wdp_rule *wdp_rule)
+{
+ return wdp_rule->client_data;
+}
- /* Datapath actions.
- *
- * A super-rule with wildcard fields never has ODP actions (since the
- * datapath only supports exact-match flows). */
- bool installed; /* Installed in datapath? */
- bool may_install; /* True ordinarily; false if actions must
- * be reassessed for every packet. */
- int n_odp_actions;
- union odp_action *odp_actions;
-};
+static void
+ofproto_rule_init(struct wdp_rule *wdp_rule)
+{
+ wdp_rule->client_data = xzalloc(sizeof(struct ofproto_rule));
+}
static inline bool
-rule_is_hidden(const struct rule *rule)
+rule_is_hidden(const struct wdp_rule *rule)
{
- /* Subrules are merely an implementation detail, so hide them from the
- * controller. */
- if (rule->super != NULL) {
- return true;
- }
-
/* Rules with priority higher than UINT16_MAX are set up by ofproto itself
* (e.g. by in-band control) and are intentionally hidden from the
* controller. */
- if (rule->cr.priority > UINT16_MAX) {
+ if (rule->cr.flow.priority > UINT16_MAX) {
return true;
}
return false;
}
-static struct rule *rule_create(struct ofproto *, struct rule *super,
- const union ofp_action *, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed);
-static void rule_free(struct rule *);
-static void rule_destroy(struct ofproto *, struct rule *);
-static struct rule *rule_from_cls_rule(const struct cls_rule *);
-static void rule_insert(struct ofproto *, struct rule *,
- struct ofpbuf *packet, uint16_t in_port);
-static void rule_remove(struct ofproto *, struct rule *);
-static bool rule_make_actions(struct ofproto *, struct rule *,
- const struct ofpbuf *packet);
-static void rule_install(struct ofproto *, struct rule *,
- struct rule *displaced_rule);
-static void rule_uninstall(struct ofproto *, struct rule *);
-static void rule_post_uninstall(struct ofproto *, struct rule *);
-static void send_flow_removed(struct ofproto *p, struct rule *rule,
- long long int now, uint8_t reason);
+static int delete_flow(struct ofproto *, struct wdp_rule *, uint8_t reason);
/* ofproto supports two kinds of OpenFlow connections:
*
struct list node; /* In struct ofproto's "all_conns" list. */
struct rconn *rconn; /* OpenFlow connection. */
enum ofconn_type type; /* Type. */
+ bool flow_mod_table_id; /* NXT_FLOW_MOD_TABLE_ID enabled? */
/* OFPT_PACKET_IN related data. */
struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
/* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
* "schedulers" array. Their values are 0 and 1, and their meanings and values
- * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
+ * coincide with WDP_CHAN_MISS and WDP_CHAN_ACTION, so this is convenient. In
* case anything ever changes, check their values here. */
#define N_SCHEDULERS 2
BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
-BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
+BUILD_ASSERT_DECL(OFPR_NO_MATCH == WDP_CHAN_MISS);
BUILD_ASSERT_DECL(OFPR_ACTION == 1);
-BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
+BUILD_ASSERT_DECL(OFPR_ACTION == WDP_CHAN_ACTION);
static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
enum ofconn_type);
static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
struct rconn_packet_counter *counter);
-static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
-static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
+static void send_packet_in(struct ofproto *, struct wdp_packet *);
+static void do_send_packet_in(struct wdp_packet *, void *ofconn);
struct ofproto {
/* Settings. */
char *dp_desc; /* Datapath description. */
/* Datapath. */
- struct dpif *dpif;
- struct netdev_monitor *netdev_monitor;
- struct hmap ports; /* Contains "struct ofport"s. */
- struct shash port_by_name;
+ struct wdp *wdp;
uint32_t max_ports;
/* Configuration. */
struct fail_open *fail_open;
struct netflow *netflow;
struct ofproto_sflow *sflow;
+ bool tun_id_from_cookie; /* NXT_TUN_ID_FROM_COOKIE enabled? */
/* In-band control. */
struct in_band *in_band;
struct sockaddr_in *extra_in_band_remotes;
size_t n_extra_remotes;
- /* Flow table. */
- struct classifier cls;
- bool need_revalidate;
- long long int next_expiration;
- struct tag_set revalidate_set;
- bool tun_id_from_cookie;
-
/* OpenFlow connections. */
struct hmap controllers; /* Controller "struct ofconn"s. */
struct list all_conns; /* Contains "struct ofconn"s. */
struct hmap services; /* Contains "struct ofservice"s. */
struct pvconn **snoops;
size_t n_snoops;
-
- /* Hooks for ovs-vswitchd. */
- const struct ofhooks *ofhooks;
- void *aux;
-
- /* Used by default ofhooks. */
- struct mac_learning *ml;
};
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
-static const struct ofhooks default_ofhooks;
-
static uint64_t pick_datapath_id(const struct ofproto *);
static uint64_t pick_fallback_dpid(void);
-static void update_used(struct ofproto *);
-static void update_stats(struct ofproto *, struct rule *,
- const struct odp_flow_stats *);
-static void expire_rule(struct cls_rule *, void *ofproto);
-static void active_timeout(struct ofproto *ofproto, struct rule *rule);
-static bool revalidate_rule(struct ofproto *p, struct rule *rule);
-static void revalidate_cb(struct cls_rule *rule_, void *p_);
-
-static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
+static void handle_wdp_packet(struct ofproto *, struct wdp_packet *);
static void handle_openflow(struct ofconn *, struct ofproto *,
struct ofpbuf *);
-static void refresh_port_groups(struct ofproto *);
-
-static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
-static void update_port(struct ofproto *, const char *devname);
-static int init_ports(struct ofproto *);
-static void reinit_ports(struct ofproto *);
-
int
ofproto_create(const char *datapath, const char *datapath_type,
const struct ofhooks *ofhooks, void *aux,
struct ofproto **ofprotop)
{
- struct odp_stats stats;
+ struct wdp_stats stats;
struct ofproto *p;
- struct dpif *dpif;
+ struct wdp *wdp;
int error;
*ofprotop = NULL;
/* Connect to datapath and start listening for messages. */
- error = dpif_open(datapath, datapath_type, &dpif);
+ error = wdp_open(datapath, datapath_type, &wdp);
if (error) {
VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
return error;
}
- error = dpif_get_dp_stats(dpif, &stats);
+ error = wdp_get_wdp_stats(wdp, &stats);
if (error) {
VLOG_ERR("failed to obtain stats for datapath %s: %s",
datapath, strerror(error));
- dpif_close(dpif);
+ wdp_close(wdp);
return error;
}
- error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
+ error = wdp_recv_set_mask(wdp, ((1 << WDP_CHAN_MISS)
+ | (1 << WDP_CHAN_ACTION)
+ | (1 << WDP_CHAN_SFLOW)));
if (error) {
VLOG_ERR("failed to listen on datapath %s: %s",
datapath, strerror(error));
- dpif_close(dpif);
+ wdp_close(wdp);
return error;
}
- dpif_flow_flush(dpif);
- dpif_recv_purge(dpif);
+ wdp_flow_flush(wdp);
+ wdp_recv_purge(wdp);
+ wdp_set_ofhooks(wdp, ofhooks, aux);
/* Initialize settings. */
p = xzalloc(sizeof *p);
p->dp_desc = xstrdup(DEFAULT_DP_DESC);
/* Initialize datapath. */
- p->dpif = dpif;
- p->netdev_monitor = netdev_monitor_create();
- hmap_init(&p->ports);
- shash_init(&p->port_by_name);
+ p->wdp = wdp;
p->max_ports = stats.max_ports;
/* Initialize submodules. */
p->netflow = NULL;
p->sflow = NULL;
- /* Initialize flow table. */
- classifier_init(&p->cls);
- p->need_revalidate = false;
- p->next_expiration = time_msec() + 1000;
- tag_set_init(&p->revalidate_set);
-
/* Initialize OpenFlow connections. */
list_init(&p->all_conns);
hmap_init(&p->controllers);
p->snoops = NULL;
p->n_snoops = 0;
- /* Initialize hooks. */
- if (ofhooks) {
- p->ofhooks = ofhooks;
- p->aux = aux;
- p->ml = NULL;
- } else {
- p->ofhooks = &default_ofhooks;
- p->aux = p;
- p->ml = mac_learning_create();
- }
-
/* Pick final datapath ID. */
p->datapath_id = pick_datapath_id(p);
VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
if (is_discovery_controller(c)) {
int error = discovery_create(c->accept_re, c->update_resolv_conf,
- ofproto->dpif, ofproto->switch_status,
+ ofproto->wdp, ofproto->switch_status,
&discovery);
if (error) {
return;
{
struct ofconn *ofconn;
- HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
+ HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
hash_string(target, 0), &ofproto->controllers) {
if (!strcmp(ofconn_get_target(ofconn), target)) {
return ofconn;
/* Add all the remotes. */
discovery = false;
- HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
+ HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
struct sockaddr_in *sin = &addrs[n_addrs];
if (ofconn->band == OFPROTO_OUT_OF_BAND) {
* even before we know any remote addresses. */
if (n_addrs || discovery) {
if (!ofproto->in_band) {
- in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
+ in_band_create(ofproto, ofproto->wdp, ofproto->switch_status,
&ofproto->in_band);
}
if (ofproto->in_band) {
n = 0;
rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
- HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
+ HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
rconns[n++] = ofconn->rconn;
}
}
} else {
VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
- dpif_name(p->dpif), c->target);
+ wdp_name(p->wdp), c->target);
continue;
}
/* Delete controllers that are no longer configured.
* Update configuration of all now-existing controllers. */
ss_exists = false;
- HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
- &p->controllers) {
+ HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
struct ofproto_controller *c;
c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
/* Delete services that are no longer configured.
* Update configuration of all now-existing services. */
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
- &p->services) {
+ HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
struct ofproto_controller *c;
c = shash_find_data(&new_controllers,
{
struct ofconn *ofconn;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
+ LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
rconn_reconnect(ofconn->rconn);
}
}
struct ofproto_sflow *os = ofproto->sflow;
if (oso) {
if (!os) {
- struct ofport *ofport;
-
- os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
- refresh_port_groups(ofproto);
- HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
- ofproto_sflow_add_port(os, ofport->odp_port,
- netdev_get_name(ofport->netdev));
- }
+ os = ofproto->sflow = ofproto_sflow_create(ofproto->wdp);
+ /* XXX ofport */
}
ofproto_sflow_set_options(os, oso);
} else {
{
struct ofservice *ofservice, *next_ofservice;
struct ofconn *ofconn, *next_ofconn;
- struct ofport *ofport, *next_ofport;
size_t i;
if (!p) {
free(p->extra_in_band_remotes);
ofproto_flush_flows(p);
- classifier_destroy(&p->cls);
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
+ LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
ofconn_destroy(ofconn);
}
hmap_destroy(&p->controllers);
- dpif_close(p->dpif);
- netdev_monitor_destroy(p->netdev_monitor);
- HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
- hmap_remove(&p->ports, &ofport->hmap_node);
- ofport_free(ofport);
- }
- shash_destroy(&p->port_by_name);
+ wdp_close(p->wdp);
switch_status_destroy(p->switch_status);
netflow_destroy(p->netflow);
ofproto_sflow_destroy(p->sflow);
- HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
- &p->services) {
+ HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
ofservice_destroy(p, ofservice);
}
hmap_destroy(&p->services);
}
free(p->snoops);
- mac_learning_destroy(p->ml);
-
free(p->mfr_desc);
free(p->hw_desc);
free(p->sw_desc);
free(p->serial_desc);
free(p->dp_desc);
- hmap_destroy(&p->ports);
-
free(p);
}
return error;
}
-static void
-process_port_change(struct ofproto *ofproto, int error, char *devname)
-{
- if (error == ENOBUFS) {
- reinit_ports(ofproto);
- } else if (!error) {
- update_port(ofproto, devname);
- free(devname);
- }
-}
-
/* Returns a "preference level" for snooping 'ofconn'. A higher return value
* means that 'ofconn' is more interesting for monitoring than a lower return
* value. */
/* Pick a controller for monitoring. */
best = NULL;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
+ LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
if (ofconn->type == OFCONN_PRIMARY
&& (!best || snoop_preference(ofconn) > snoop_preference(best))) {
best = ofconn;
}
}
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
+static void
+ofproto_port_poll_cb(const struct ofp_phy_port *opp, uint8_t reason,
+ void *ofproto_)
+{
+ /* XXX Should limit the number of queued port status change messages. */
+ struct ofproto *ofproto = ofproto_;
+ struct ofconn *ofconn;
+
++ LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
+ struct ofp_port_status *ops;
+ struct ofpbuf *b;
+
+ if (!ofconn_receives_async_msgs(ofconn)) {
+ continue;
+ }
+
+ ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
+ ops->reason = reason;
+ ops->desc = *opp;
+ hton_ofp_phy_port(&ops->desc);
+ queue_tx(b, ofconn, NULL);
+ }
+}
+
int
ofproto_run1(struct ofproto *p)
{
struct ofconn *ofconn, *next_ofconn;
struct ofservice *ofservice;
- char *devname;
- int error;
int i;
- if (shash_is_empty(&p->port_by_name)) {
- init_ports(p);
- }
-
for (i = 0; i < 50; i++) {
- struct ofpbuf *buf;
+ struct wdp_packet packet;
+ int error;
- error = dpif_recv(p->dpif, &buf);
+ error = wdp_recv(p->wdp, &packet);
if (error) {
if (error == ENODEV) {
/* Someone destroyed the datapath behind our back. The caller
* spin from here on out. */
static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
- dpif_name(p->dpif));
+ wdp_name(p->wdp));
return ENODEV;
}
break;
}
- handle_odp_msg(p, buf);
+ handle_wdp_packet(p, xmemdup(&packet, sizeof packet));
}
- while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
- }
- while ((error = netdev_monitor_poll(p->netdev_monitor,
- &devname)) != EAGAIN) {
- process_port_change(p, error, devname);
- }
+ wdp_port_poll(p->wdp, ofproto_port_poll_cb, p);
if (p->in_band) {
if (time_msec() >= p->next_in_band_update) {
in_band_run(p->in_band);
}
- LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
- &p->all_conns) {
+ LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
ofconn_run(ofconn, p);
}
fail_open_run(p->fail_open);
}
- HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
+ HMAP_FOR_EACH (ofservice, node, &p->services) {
struct vconn *vconn;
int retval;
retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
if (!retval) {
- struct ofconn *ofconn;
struct rconn *rconn;
char *name;
}
}
- if (time_msec() >= p->next_expiration) {
- COVERAGE_INC(ofproto_expiration);
- p->next_expiration = time_msec() + 1000;
- update_used(p);
-
- classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
-
- /* Let the hook know that we're at a stable point: all outstanding data
- * in existing flows has been accounted to the account_cb. Thus, the
- * hook can now reasonably do operations that depend on having accurate
- * flow volume accounting (currently, that's just bond rebalancing). */
- if (p->ofhooks->account_checkpoint_cb) {
- p->ofhooks->account_checkpoint_cb(p->aux);
- }
- }
-
if (p->netflow) {
netflow_run(p->netflow);
}
return 0;
}
-struct revalidate_cbdata {
- struct ofproto *ofproto;
- bool revalidate_all; /* Revalidate all exact-match rules? */
- bool revalidate_subrules; /* Revalidate all exact-match subrules? */
- struct tag_set revalidate_set; /* Set of tags to revalidate. */
-};
-
int
-ofproto_run2(struct ofproto *p, bool revalidate_all)
+ofproto_run2(struct ofproto *p OVS_UNUSED, bool revalidate_all OVS_UNUSED)
{
- if (p->need_revalidate || revalidate_all
- || !tag_set_is_empty(&p->revalidate_set)) {
- struct revalidate_cbdata cbdata;
- cbdata.ofproto = p;
- cbdata.revalidate_all = revalidate_all;
- cbdata.revalidate_subrules = p->need_revalidate;
- cbdata.revalidate_set = p->revalidate_set;
- tag_set_init(&p->revalidate_set);
- COVERAGE_INC(ofproto_revalidate);
- classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
- p->need_revalidate = false;
- }
-
return 0;
}
struct ofconn *ofconn;
size_t i;
- dpif_recv_wait(p->dpif);
- dpif_port_poll_wait(p->dpif);
- netdev_monitor_poll_wait(p->netdev_monitor);
+ wdp_recv_wait(p->wdp);
+ wdp_port_poll_wait(p->wdp);
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
+ LIST_FOR_EACH (ofconn, node, &p->all_conns) {
ofconn_wait(ofconn);
}
if (p->in_band) {
if (p->sflow) {
ofproto_sflow_wait(p->sflow);
}
- HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
- if (!tag_set_is_empty(&p->revalidate_set)) {
- poll_immediate_wake();
- }
- if (p->need_revalidate) {
- /* Shouldn't happen, but if it does just go around again. */
- VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
- poll_immediate_wake();
- } else if (p->next_expiration != LLONG_MAX) {
- poll_timer_wait_until(p->next_expiration);
- }
+ HMAP_FOR_EACH (ofservice, node, &p->services) {
pvconn_wait(ofservice->pvconn);
}
for (i = 0; i < p->n_snoops; i++) {
void
ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
{
- tag_set_add(&ofproto->revalidate_set, tag);
+ wdp_revalidate(ofproto->wdp, tag);
}
-struct tag_set *
-ofproto_get_revalidate_set(struct ofproto *ofproto)
+void
+ofproto_revalidate_all(struct ofproto *ofproto)
{
- return &ofproto->revalidate_set;
+ wdp_revalidate_all(ofproto->wdp);
}
bool
const union ofp_action *actions, size_t n_actions,
const struct ofpbuf *packet)
{
- struct odp_actions odp_actions;
- int error;
-
- error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
- NULL, NULL, NULL);
- if (error) {
- return error;
- }
-
- /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
+ /* XXX Should we translate the wdp_execute() errno value into an OpenFlow
* error code? */
- dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
- odp_actions.n_actions, packet);
+ wdp_execute(p->wdp, flow->in_port, actions, n_actions, packet);
return 0;
}
+/* Intended for used by ofproto clients and ofproto submodules to add flows to
+ * the flow table. */
void
-ofproto_add_flow(struct ofproto *p,
- const flow_t *flow, uint32_t wildcards, unsigned int priority,
+ofproto_add_flow(struct ofproto *p, const flow_t *flow,
const union ofp_action *actions, size_t n_actions,
int idle_timeout)
{
- struct rule *rule;
- rule = rule_create(p, NULL, actions, n_actions,
- idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
- 0, 0, false);
- cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
- rule_insert(p, rule, NULL, 0);
-}
+ struct wdp_flow_put put;
+ struct wdp_rule *rule;
-void
-ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
- uint32_t wildcards, unsigned int priority)
-{
- struct rule *rule;
+ put.flags = WDP_PUT_CREATE | WDP_PUT_MODIFY | WDP_PUT_ALL;
+ put.flow = flow;
+ put.actions = actions;
+ put.n_actions = n_actions;
+ put.idle_timeout = idle_timeout;
+ put.hard_timeout = 0;
+ put.ofp_table_id = 0xff;
+ put.cookie = htonll(0);
+ put.xid = htonl(0);
- rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
- flow, wildcards,
- priority));
- if (rule) {
- rule_remove(ofproto, rule);
+ if (!wdp_flow_put(p->wdp, &put, NULL, &rule)) {
+ ofproto_rule_init(rule);
}
}
-static void
-destroy_rule(struct cls_rule *rule_, void *ofproto_)
+/* Intended for used by ofproto clients and ofproto submodules to delete flows
+ * that they earlier added to the flow table. */
+void
+ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow)
{
- struct rule *rule = rule_from_cls_rule(rule_);
- struct ofproto *ofproto = ofproto_;
-
- /* Mark the flow as not installed, even though it might really be
- * installed, so that rule_remove() doesn't bother trying to uninstall it.
- * There is no point in uninstalling it individually since we are about to
- * blow away all the flows with dpif_flow_flush(). */
- rule->installed = false;
-
- rule_remove(ofproto, rule);
+ struct wdp_rule *rule = wdp_flow_get(ofproto->wdp, flow, UINT_MAX);
+ if (rule) {
+ delete_flow(ofproto, rule, OFPRR_DELETE);
+ }
}
void
ofproto_flush_flows(struct ofproto *ofproto)
{
COVERAGE_INC(ofproto_flush);
- classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
- dpif_flow_flush(ofproto->dpif);
+ wdp_flow_flush(ofproto->wdp);
if (ofproto->in_band) {
in_band_flushed(ofproto->in_band);
}
}
}
\f
-static void
-reinit_ports(struct ofproto *p)
-{
- struct svec devnames;
- struct ofport *ofport;
- struct odp_port *odp_ports;
- size_t n_odp_ports;
- size_t i;
-
- svec_init(&devnames);
- HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
- svec_add (&devnames, (char *) ofport->opp.name);
- }
- dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
- for (i = 0; i < n_odp_ports; i++) {
- svec_add (&devnames, odp_ports[i].devname);
- }
- free(odp_ports);
-
- svec_sort_unique(&devnames);
- for (i = 0; i < devnames.n; i++) {
- update_port(p, devnames.names[i]);
- }
- svec_destroy(&devnames);
-}
-
-static size_t
-refresh_port_group(struct ofproto *p, unsigned int group)
-{
- uint16_t *ports;
- size_t n_ports;
- struct ofport *port;
-
- assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
-
- ports = xmalloc(hmap_count(&p->ports) * sizeof *ports);
- n_ports = 0;
- HMAP_FOR_EACH (port, hmap_node, &p->ports) {
- if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
- ports[n_ports++] = port->odp_port;
- }
- }
- dpif_port_group_set(p->dpif, group, ports, n_ports);
- free(ports);
-
- return n_ports;
-}
-
-static void
-refresh_port_groups(struct ofproto *p)
-{
- size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
- size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
- if (p->sflow) {
- ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
- }
-}
-
-static struct ofport *
-make_ofport(const struct odp_port *odp_port)
-{
- struct netdev_options netdev_options;
- enum netdev_flags flags;
- struct ofport *ofport;
- struct netdev *netdev;
- bool carrier;
- int error;
-
- memset(&netdev_options, 0, sizeof netdev_options);
- netdev_options.name = odp_port->devname;
- netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
-
- error = netdev_open(&netdev_options, &netdev);
- if (error) {
- VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
- "cannot be opened (%s)",
- odp_port->devname, odp_port->port,
- odp_port->devname, strerror(error));
- return NULL;
- }
-
- ofport = xmalloc(sizeof *ofport);
- ofport->netdev = netdev;
- ofport->odp_port = odp_port->port;
- ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
- netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
- memcpy(ofport->opp.name, odp_port->devname,
- MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
- ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
-
- netdev_get_flags(netdev, &flags);
- ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
-
- netdev_get_carrier(netdev, &carrier);
- ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
-
- netdev_get_features(netdev,
- &ofport->opp.curr, &ofport->opp.advertised,
- &ofport->opp.supported, &ofport->opp.peer);
- return ofport;
-}
-
-static bool
-ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
-{
- if (get_port(p, odp_port->port)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
- odp_port->port);
- return true;
- } else if (shash_find(&p->port_by_name, odp_port->devname)) {
- VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
- odp_port->devname);
- return true;
- } else {
- return false;
- }
-}
-
-static int
-ofport_equal(const struct ofport *a_, const struct ofport *b_)
-{
- const struct ofp_phy_port *a = &a_->opp;
- const struct ofp_phy_port *b = &b_->opp;
-
- BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
- return (a->port_no == b->port_no
- && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
- && !strcmp((char *) a->name, (char *) b->name)
- && a->state == b->state
- && a->config == b->config
- && a->curr == b->curr
- && a->advertised == b->advertised
- && a->supported == b->supported
- && a->peer == b->peer);
-}
-
-static void
-send_port_status(struct ofproto *p, const struct ofport *ofport,
- uint8_t reason)
-{
- /* XXX Should limit the number of queued port status change messages. */
- struct ofconn *ofconn;
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- struct ofp_port_status *ops;
- struct ofpbuf *b;
-
- if (!ofconn_receives_async_msgs(ofconn)) {
- continue;
- }
-
- ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
- ops->reason = reason;
- ops->desc = ofport->opp;
- hton_ofp_phy_port(&ops->desc);
- queue_tx(b, ofconn, NULL);
- }
- if (p->ofhooks->port_changed_cb) {
- p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
- }
-}
-
-static void
-ofport_install(struct ofproto *p, struct ofport *ofport)
-{
- const char *netdev_name = (const char *) ofport->opp.name;
-
- netdev_monitor_add(p->netdev_monitor, ofport->netdev);
- hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
- shash_add(&p->port_by_name, netdev_name, ofport);
- if (p->sflow) {
- ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
- }
-}
-
-static void
-ofport_remove(struct ofproto *p, struct ofport *ofport)
-{
- netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
- hmap_remove(&p->ports, &ofport->hmap_node);
- shash_delete(&p->port_by_name,
- shash_find(&p->port_by_name, (char *) ofport->opp.name));
- if (p->sflow) {
- ofproto_sflow_del_port(p->sflow, ofport->odp_port);
- }
-}
-
-static void
-ofport_free(struct ofport *ofport)
-{
- if (ofport) {
- netdev_close(ofport->netdev);
- free(ofport);
- }
-}
-
-static struct ofport *
-get_port(const struct ofproto *ofproto, uint16_t odp_port)
-{
- struct ofport *port;
-
- HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
- hash_int(odp_port, 0), &ofproto->ports) {
- if (port->odp_port == odp_port) {
- return port;
- }
- }
- return NULL;
-}
-
-static void
-update_port(struct ofproto *p, const char *devname)
-{
- struct odp_port odp_port;
- struct ofport *old_ofport;
- struct ofport *new_ofport;
- int error;
-
- COVERAGE_INC(ofproto_update_port);
-
- /* Query the datapath for port information. */
- error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
-
- /* Find the old ofport. */
- old_ofport = shash_find_data(&p->port_by_name, devname);
- if (!error) {
- if (!old_ofport) {
- /* There's no port named 'devname' but there might be a port with
- * the same port number. This could happen if a port is deleted
- * and then a new one added in its place very quickly, or if a port
- * is renamed. In the former case we want to send an OFPPR_DELETE
- * and an OFPPR_ADD, and in the latter case we want to send a
- * single OFPPR_MODIFY. We can distinguish the cases by comparing
- * the old port's ifindex against the new port, or perhaps less
- * reliably but more portably by comparing the old port's MAC
- * against the new port's MAC. However, this code isn't that smart
- * and always sends an OFPPR_MODIFY (XXX). */
- old_ofport = get_port(p, odp_port.port);
- }
- } else if (error != ENOENT && error != ENODEV) {
- VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
- "%s", strerror(error));
- return;
- }
-
- /* Create a new ofport. */
- new_ofport = !error ? make_ofport(&odp_port) : NULL;
-
- /* Eliminate a few pathological cases. */
- if (!old_ofport && !new_ofport) {
- return;
- } else if (old_ofport && new_ofport) {
- /* Most of the 'config' bits are OpenFlow soft state, but
- * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
- * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
- * leaves the other bits 0.) */
- new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
-
- if (ofport_equal(old_ofport, new_ofport)) {
- /* False alarm--no change. */
- ofport_free(new_ofport);
- return;
- }
- }
-
- /* Now deal with the normal cases. */
- if (old_ofport) {
- ofport_remove(p, old_ofport);
- }
- if (new_ofport) {
- ofport_install(p, new_ofport);
- }
- send_port_status(p, new_ofport ? new_ofport : old_ofport,
- (!old_ofport ? OFPPR_ADD
- : !new_ofport ? OFPPR_DELETE
- : OFPPR_MODIFY));
- ofport_free(old_ofport);
-
- /* Update port groups. */
- refresh_port_groups(p);
-}
-
-static int
-init_ports(struct ofproto *p)
-{
- struct odp_port *ports;
- size_t n_ports;
- size_t i;
- int error;
-
- error = dpif_port_list(p->dpif, &ports, &n_ports);
- if (error) {
- return error;
- }
-
- for (i = 0; i < n_ports; i++) {
- const struct odp_port *odp_port = &ports[i];
- if (!ofport_conflicts(p, odp_port)) {
- struct ofport *ofport = make_ofport(odp_port);
- if (ofport) {
- ofport_install(p, ofport);
- }
- }
- }
- free(ports);
- refresh_port_groups(p);
- return 0;
-}
-\f
static struct ofconn *
ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
{
static char *
ofconn_make_name(const struct ofproto *ofproto, const char *target)
{
- return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
+ return xasprintf("%s<->%s", wdp_base_name(ofproto->wdp), target);
}
static void
{
struct ofservice *ofservice;
- HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
- hash_string(target, 0), &ofproto->services) {
+ HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
+ &ofproto->services) {
if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
return ofservice;
}
return NULL;
}
\f
-/* Caller is responsible for initializing the 'cr' member of the returned
- * rule. */
-static struct rule *
-rule_create(struct ofproto *ofproto, struct rule *super,
- const union ofp_action *actions, size_t n_actions,
- uint16_t idle_timeout, uint16_t hard_timeout,
- uint64_t flow_cookie, bool send_flow_removed)
-{
- struct rule *rule = xzalloc(sizeof *rule);
- rule->idle_timeout = idle_timeout;
- rule->hard_timeout = hard_timeout;
- rule->flow_cookie = flow_cookie;
- rule->used = rule->created = time_msec();
- rule->send_flow_removed = send_flow_removed;
- rule->super = super;
- if (super) {
- list_push_back(&super->list, &rule->list);
- } else {
- list_init(&rule->list);
- }
- rule->n_actions = n_actions;
- rule->actions = xmemdup(actions, n_actions * sizeof *actions);
- netflow_flow_clear(&rule->nf_flow);
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
-
- return rule;
-}
-
-static struct rule *
-rule_from_cls_rule(const struct cls_rule *cls_rule)
-{
- return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
-}
-
-static void
-rule_free(struct rule *rule)
-{
- free(rule->actions);
- free(rule->odp_actions);
- free(rule);
-}
-
-/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
- * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
- * through all of its subrules and revalidates them, destroying any that no
- * longer has a super-rule (which is probably all of them).
- *
- * Before calling this function, the caller must make have removed 'rule' from
- * the classifier. If 'rule' is an exact-match rule, the caller is also
- * responsible for ensuring that it has been uninstalled from the datapath. */
-static void
-rule_destroy(struct ofproto *ofproto, struct rule *rule)
-{
- if (!rule->super) {
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
- revalidate_rule(ofproto, subrule);
- }
- } else {
- list_remove(&rule->list);
- }
- rule_free(rule);
-}
-
static bool
-rule_has_out_port(const struct rule *rule, uint16_t out_port)
+rule_has_out_port(const struct wdp_rule *rule, uint16_t out_port)
{
const union ofp_action *oa;
struct actions_iterator i;
}
return false;
}
-
-/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
- * 'packet', which arrived on 'in_port'.
- *
- * Takes ownership of 'packet'. */
-static bool
-execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
- const union odp_action *actions, size_t n_actions,
- struct ofpbuf *packet)
-{
- if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
- /* As an optimization, avoid a round-trip from userspace to kernel to
- * userspace. This also avoids possibly filling up kernel packet
- * buffers along the way. */
- struct odp_msg *msg;
-
- msg = ofpbuf_push_uninit(packet, sizeof *msg);
- msg->type = _ODPL_ACTION_NR;
- msg->length = sizeof(struct odp_msg) + packet->size;
- msg->port = in_port;
- msg->reserved = 0;
- msg->arg = actions[0].controller.arg;
-
- send_packet_in(ofproto, packet);
-
- return true;
- } else {
- int error;
-
- error = dpif_execute(ofproto->dpif, in_port,
- actions, n_actions, packet);
- ofpbuf_delete(packet);
- return !error;
- }
-}
-
-/* Executes the actions indicated by 'rule' on 'packet', which is in flow
- * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
- * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
- *
- * The flow that 'packet' actually contains does not need to actually match
- * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
- * the packet and byte counters for 'rule' will be credited for the packet sent
- * out whether or not the packet actually matches 'rule'.
- *
- * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
- * the caller must already have accurately composed ODP actions for it given
- * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
- * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
- * function will compose a set of ODP actions based on 'rule''s OpenFlow
- * actions and apply them to 'packet'.
- *
- * Takes ownership of 'packet'. */
-static void
-rule_execute(struct ofproto *ofproto, struct rule *rule,
- struct ofpbuf *packet, const flow_t *flow)
-{
- const union odp_action *actions;
- struct odp_flow_stats stats;
- size_t n_actions;
- struct odp_actions a;
-
- assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
-
- /* Grab or compose the ODP actions.
- *
- * The special case for an exact-match 'rule' where 'flow' is not the
- * rule's flow is important to avoid, e.g., sending a packet out its input
- * port simply because the ODP actions were composed for the wrong
- * scenario. */
- if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
- struct rule *super = rule->super ? rule->super : rule;
- if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
- packet, &a, NULL, 0, NULL)) {
- ofpbuf_delete(packet);
- return;
- }
- actions = a.actions;
- n_actions = a.n_actions;
- } else {
- actions = rule->odp_actions;
- n_actions = rule->n_odp_actions;
- }
-
- /* Execute the ODP actions. */
- flow_extract_stats(flow, packet, &stats);
- if (execute_odp_actions(ofproto, flow->in_port,
- actions, n_actions, packet)) {
- update_stats(ofproto, rule, &stats);
- rule->used = time_msec();
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
- }
-}
-
-/* Inserts 'rule' into 'p''s flow table.
- *
- * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
- * actions on it and credits the statistics for sending the packet to 'rule'.
- * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
- * headroom. */
-static void
-rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
- uint16_t in_port)
-{
- struct rule *displaced_rule;
-
- /* Insert the rule in the classifier. */
- displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
- if (!rule->cr.wc.wildcards) {
- rule_make_actions(p, rule, packet);
- }
-
- /* Send the packet and credit it to the rule. */
- if (packet) {
- flow_t flow;
- flow_extract(packet, 0, in_port, &flow);
- rule_execute(p, rule, packet, &flow);
- }
-
- /* Install the rule in the datapath only after sending the packet, to
- * avoid packet reordering. */
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_add_wc_flow);
- p->need_revalidate = true;
- } else {
- rule_install(p, rule, displaced_rule);
- }
-
- /* Free the rule that was displaced, if any. */
- if (displaced_rule) {
- rule_destroy(p, displaced_rule);
- }
-}
-
-static struct rule *
-rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
- const flow_t *flow)
-{
- struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
- rule->idle_timeout, rule->hard_timeout,
- 0, false);
- COVERAGE_INC(ofproto_subrule_create);
- cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
- : rule->cr.priority), &subrule->cr);
- classifier_insert_exact(&ofproto->cls, &subrule->cr);
-
- return subrule;
-}
-
-static void
-rule_remove(struct ofproto *ofproto, struct rule *rule)
-{
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_del_wc_flow);
- ofproto->need_revalidate = true;
- } else {
- rule_uninstall(ofproto, rule);
- }
- classifier_remove(&ofproto->cls, &rule->cr);
- rule_destroy(ofproto, rule);
-}
-
-/* Returns true if the actions changed, false otherwise. */
-static bool
-rule_make_actions(struct ofproto *p, struct rule *rule,
- const struct ofpbuf *packet)
-{
- const struct rule *super;
- struct odp_actions a;
- size_t actions_len;
-
- assert(!rule->cr.wc.wildcards);
-
- super = rule->super ? rule->super : rule;
- rule->tags = 0;
- xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
- packet, &a, &rule->tags, &rule->may_install,
- &rule->nf_flow.output_iface);
-
- actions_len = a.n_actions * sizeof *a.actions;
- if (rule->n_odp_actions != a.n_actions
- || memcmp(rule->odp_actions, a.actions, actions_len)) {
- COVERAGE_INC(ofproto_odp_unchanged);
- free(rule->odp_actions);
- rule->n_odp_actions = a.n_actions;
- rule->odp_actions = xmemdup(a.actions, actions_len);
- return true;
- } else {
- return false;
- }
-}
-
-static int
-do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
- struct odp_flow_put *put)
-{
- memset(&put->flow.stats, 0, sizeof put->flow.stats);
- put->flow.key = rule->cr.flow;
- put->flow.actions = rule->odp_actions;
- put->flow.n_actions = rule->n_odp_actions;
- put->flow.flags = 0;
- put->flags = flags;
- return dpif_flow_put(ofproto->dpif, put);
-}
-
-static void
-rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
-{
- assert(!rule->cr.wc.wildcards);
-
- if (rule->may_install) {
- struct odp_flow_put put;
- if (!do_put_flow(p, rule,
- ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
- &put)) {
- rule->installed = true;
- if (displaced_rule) {
- update_stats(p, displaced_rule, &put.flow.stats);
- rule_post_uninstall(p, displaced_rule);
- }
- }
- } else if (displaced_rule) {
- rule_uninstall(p, displaced_rule);
- }
-}
-
-static void
-rule_reinstall(struct ofproto *ofproto, struct rule *rule)
-{
- if (rule->installed) {
- struct odp_flow_put put;
- COVERAGE_INC(ofproto_dp_missed);
- do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
- } else {
- rule_install(ofproto, rule, NULL);
- }
-}
-
-static void
-rule_update_actions(struct ofproto *ofproto, struct rule *rule)
-{
- bool actions_changed;
- uint16_t new_out_iface, old_out_iface;
-
- old_out_iface = rule->nf_flow.output_iface;
- actions_changed = rule_make_actions(ofproto, rule, NULL);
-
- if (rule->may_install) {
- if (rule->installed) {
- if (actions_changed) {
- struct odp_flow_put put;
- do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
- | ODPPF_ZERO_STATS, &put);
- update_stats(ofproto, rule, &put.flow.stats);
-
- /* Temporarily set the old output iface so that NetFlow
- * messages have the correct output interface for the old
- * stats. */
- new_out_iface = rule->nf_flow.output_iface;
- rule->nf_flow.output_iface = old_out_iface;
- rule_post_uninstall(ofproto, rule);
- rule->nf_flow.output_iface = new_out_iface;
- }
- } else {
- rule_install(ofproto, rule, NULL);
- }
- } else {
- rule_uninstall(ofproto, rule);
- }
-}
-
-static void
-rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
-{
- uint64_t total_bytes = rule->byte_count + extra_bytes;
-
- if (ofproto->ofhooks->account_flow_cb
- && total_bytes > rule->accounted_bytes)
- {
- ofproto->ofhooks->account_flow_cb(
- &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
- total_bytes - rule->accounted_bytes, ofproto->aux);
- rule->accounted_bytes = total_bytes;
- }
-}
-
-static void
-rule_uninstall(struct ofproto *p, struct rule *rule)
-{
- assert(!rule->cr.wc.wildcards);
- if (rule->installed) {
- struct odp_flow odp_flow;
-
- odp_flow.key = rule->cr.flow;
- odp_flow.actions = NULL;
- odp_flow.n_actions = 0;
- odp_flow.flags = 0;
- if (!dpif_flow_del(p->dpif, &odp_flow)) {
- update_stats(p, rule, &odp_flow.stats);
- }
- rule->installed = false;
-
- rule_post_uninstall(p, rule);
- }
-}
-
-static bool
-is_controller_rule(struct rule *rule)
-{
- /* If the only action is send to the controller then don't report
- * NetFlow expiration messages since it is just part of the control
- * logic for the network and not real traffic. */
-
- return (rule
- && rule->super
- && rule->super->n_actions == 1
- && action_outputs_to_port(&rule->super->actions[0],
- htons(OFPP_CONTROLLER)));
-}
-
-static void
-rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
-{
- struct rule *super = rule->super;
-
- rule_account(ofproto, rule, 0);
-
- if (ofproto->netflow && !is_controller_rule(rule)) {
- struct ofexpired expired;
- expired.flow = rule->cr.flow;
- expired.packet_count = rule->packet_count;
- expired.byte_count = rule->byte_count;
- expired.used = rule->used;
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
- }
- if (super) {
- super->packet_count += rule->packet_count;
- super->byte_count += rule->byte_count;
-
- /* Reset counters to prevent double counting if the rule ever gets
- * reinstalled. */
- rule->packet_count = 0;
- rule->byte_count = 0;
- rule->accounted_bytes = 0;
-
- netflow_flow_clear(&rule->nf_flow);
- }
-}
\f
static void
queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
}
}
-static void
-send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
- int error, const void *data, size_t len)
-{
- struct ofpbuf *buf;
- struct ofp_error_msg *oem;
-
- if (!(error >> 16)) {
- VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
- error);
- return;
- }
-
- COVERAGE_INC(ofproto_error);
- oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
- oh ? oh->xid : 0, &buf);
- oem->type = htons((unsigned int) error >> 16);
- oem->code = htons(error & 0xffff);
- memcpy(oem->data, data, len);
- queue_tx(buf, ofconn, ofconn->reply_counter);
-}
-
static void
send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
int error)
{
- size_t oh_length = ntohs(oh->length);
- send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
-}
-
-static void
-hton_ofp_phy_port(struct ofp_phy_port *opp)
-{
- opp->port_no = htons(opp->port_no);
- opp->config = htonl(opp->config);
- opp->state = htonl(opp->state);
- opp->curr = htonl(opp->curr);
- opp->advertised = htonl(opp->advertised);
- opp->supported = htonl(opp->supported);
- opp->peer = htonl(opp->peer);
+ struct ofpbuf *buf = make_ofp_error_msg(error, oh);
+ if (buf) {
+ COVERAGE_INC(ofproto_error);
+ queue_tx(buf, ofconn, ofconn->reply_counter);
+ }
}
static int
handle_features_request(struct ofproto *p, struct ofconn *ofconn,
struct ofp_header *oh)
{
- struct ofp_switch_features *osf;
- struct ofpbuf *buf;
- struct ofport *port;
-
- osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
- osf->datapath_id = htonll(p->datapath_id);
- osf->n_buffers = htonl(pktbuf_capacity());
- osf->n_tables = 2;
- osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
- OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
- osf->actions = htonl((1u << OFPAT_OUTPUT) |
- (1u << OFPAT_SET_VLAN_VID) |
- (1u << OFPAT_SET_VLAN_PCP) |
- (1u << OFPAT_STRIP_VLAN) |
- (1u << OFPAT_SET_DL_SRC) |
- (1u << OFPAT_SET_DL_DST) |
- (1u << OFPAT_SET_NW_SRC) |
- (1u << OFPAT_SET_NW_DST) |
- (1u << OFPAT_SET_NW_TOS) |
- (1u << OFPAT_SET_TP_SRC) |
- (1u << OFPAT_SET_TP_DST) |
- (1u << OFPAT_ENQUEUE));
-
- HMAP_FOR_EACH (port, hmap_node, &p->ports) {
- hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
- }
+ struct ofpbuf *features;
+ int error;
- queue_tx(buf, ofconn, ofconn->reply_counter);
- return 0;
+ error = wdp_get_features(p->wdp, &features);
+ if (!error) {
+ struct ofp_switch_features *osf = features->data;
+
+ update_openflow_length(features);
+ osf->header.version = OFP_VERSION;
+ osf->header.type = OFPT_FEATURES_REPLY;
+ osf->header.xid = oh->xid;
+
+ osf->datapath_id = htonll(p->datapath_id);
+ osf->n_buffers = htonl(pktbuf_capacity());
+ memset(osf->pad, 0, sizeof osf->pad);
+
+ /* Turn on capabilities implemented by ofproto. */
+ osf->capabilities |= htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
+ OFPC_PORT_STATS);
+
+ queue_tx(features, ofconn, ofconn->reply_counter);
+ }
+ return error;
}
static int
bool drop_frags;
/* Figure out flags. */
- dpif_get_drop_frags(p->dpif, &drop_frags);
+ wdp_get_drop_frags(p->wdp, &drop_frags);
flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
/* Send reply. */
if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
switch (flags & OFPC_FRAG_MASK) {
case OFPC_FRAG_NORMAL:
- dpif_set_drop_frags(p->dpif, false);
+ wdp_set_drop_frags(p->wdp, false);
break;
case OFPC_FRAG_DROP:
- dpif_set_drop_frags(p->dpif, true);
+ wdp_set_drop_frags(p->wdp, true);
break;
default:
VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
return 0;
}
-static void
-add_output_group_action(struct odp_actions *actions, uint16_t group,
- uint16_t *nf_output_iface)
-{
- odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
-
- if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
- *nf_output_iface = NF_OUT_FLOOD;
- }
-}
-
-static void
-add_controller_action(struct odp_actions *actions, uint16_t max_len)
-{
- union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
- a->controller.arg = max_len;
-}
-
-struct action_xlate_ctx {
- /* Input. */
- flow_t flow; /* Flow to which these actions correspond. */
- int recurse; /* Recursion level, via xlate_table_action. */
- struct ofproto *ofproto;
- const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
- * null pointer if we are revalidating
- * without a packet to refer to. */
-
- /* Output. */
- struct odp_actions *out; /* Datapath actions. */
- tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
- bool may_set_up_flow; /* True ordinarily; false if the actions must
- * be reassessed for every packet. */
- uint16_t nf_output_iface; /* Output interface index for NetFlow. */
-};
-
-/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
- * flow translation. */
-#define MAX_RESUBMIT_RECURSION 8
-
-static void do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx);
-
-static void
-add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
-{
- const struct ofport *ofport = get_port(ctx->ofproto, port);
-
- if (ofport) {
- if (ofport->opp.config & OFPPC_NO_FWD) {
- /* Forwarding disabled on port. */
- return;
- }
- } else {
- /*
- * We don't have an ofport record for this port, but it doesn't hurt to
- * allow forwarding to it anyhow. Maybe such a port will appear later
- * and we're pre-populating the flow table.
- */
- }
-
- odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
- ctx->nf_output_iface = port;
-}
-
-static struct rule *
-lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
-{
- struct rule *rule;
- rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
-
- /* The rule we found might not be valid, since we could be in need of
- * revalidation. If it is not valid, don't return it. */
- if (rule
- && rule->super
- && ofproto->need_revalidate
- && !revalidate_rule(ofproto, rule)) {
- COVERAGE_INC(ofproto_invalidated);
- return NULL;
- }
-
- return rule;
-}
-
-static void
-xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
-{
- if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
- uint16_t old_in_port;
- struct rule *rule;
-
- /* Look up a flow with 'in_port' as the input port. Then restore the
- * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
- * have surprising behavior). */
- old_in_port = ctx->flow.in_port;
- ctx->flow.in_port = in_port;
- rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
- ctx->flow.in_port = old_in_port;
-
- if (rule) {
- if (rule->super) {
- rule = rule->super;
- }
-
- ctx->recurse++;
- do_xlate_actions(rule->actions, rule->n_actions, ctx);
- ctx->recurse--;
- }
- } else {
- struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
-
- VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
- MAX_RESUBMIT_RECURSION);
- }
-}
-
-static void
-xlate_output_action__(struct action_xlate_ctx *ctx,
- uint16_t port, uint16_t max_len)
-{
- uint16_t odp_port;
- uint16_t prev_nf_output_iface = ctx->nf_output_iface;
-
- ctx->nf_output_iface = NF_OUT_DROP;
-
- switch (port) {
- case OFPP_IN_PORT:
- add_output_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_TABLE:
- xlate_table_action(ctx, ctx->flow.in_port);
- break;
- case OFPP_NORMAL:
- if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
- ctx->out, ctx->tags,
- &ctx->nf_output_iface,
- ctx->ofproto->aux)) {
- COVERAGE_INC(ofproto_uninstallable);
- ctx->may_set_up_flow = false;
- }
- break;
- case OFPP_FLOOD:
- add_output_group_action(ctx->out, DP_GROUP_FLOOD,
- &ctx->nf_output_iface);
- break;
- case OFPP_ALL:
- add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
- break;
- case OFPP_CONTROLLER:
- add_controller_action(ctx->out, max_len);
- break;
- case OFPP_LOCAL:
- add_output_action(ctx, ODPP_LOCAL);
- break;
- default:
- odp_port = ofp_port_to_odp_port(port);
- if (odp_port != ctx->flow.in_port) {
- add_output_action(ctx, odp_port);
- }
- break;
- }
-
- if (prev_nf_output_iface == NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_FLOOD;
- } else if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = prev_nf_output_iface;
- } else if (prev_nf_output_iface != NF_OUT_DROP &&
- ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
- }
-}
-
-static void
-xlate_output_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_output *oao)
-{
- xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
-}
-
-/* If the final ODP action in 'ctx' is "pop priority", drop it, as an
- * optimization, because we're going to add another action that sets the
- * priority immediately after, or because there are no actions following the
- * pop. */
-static void
-remove_pop_action(struct action_xlate_ctx *ctx)
-{
- size_t n = ctx->out->n_actions;
- if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
- ctx->out->n_actions--;
- }
-}
-
-static void
-xlate_enqueue_action(struct action_xlate_ctx *ctx,
- const struct ofp_action_enqueue *oae)
-{
- uint16_t ofp_port, odp_port;
- uint32_t priority;
- int error;
-
- error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
- &priority);
- if (error) {
- /* Fall back to ordinary output action. */
- xlate_output_action__(ctx, ntohs(oae->port), 0);
- return;
- }
-
- /* Figure out ODP output port. */
- ofp_port = ntohs(oae->port);
- if (ofp_port != OFPP_IN_PORT) {
- odp_port = ofp_port_to_odp_port(ofp_port);
- } else {
- odp_port = ctx->flow.in_port;
- }
-
- /* Add ODP actions. */
- remove_pop_action(ctx);
- odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
- = priority;
- add_output_action(ctx, odp_port);
- odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
-
- /* Update NetFlow output port. */
- if (ctx->nf_output_iface == NF_OUT_DROP) {
- ctx->nf_output_iface = odp_port;
- } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
- ctx->nf_output_iface = NF_OUT_MULTI;
- }
-}
-
-static void
-xlate_set_queue_action(struct action_xlate_ctx *ctx,
- const struct nx_action_set_queue *nasq)
-{
- uint32_t priority;
- int error;
-
- error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
- &priority);
- if (error) {
- /* Couldn't translate queue to a priority, so ignore. A warning
- * has already been logged. */
- return;
- }
-
- remove_pop_action(ctx);
- odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
- = priority;
-}
-
-static void
-xlate_nicira_action(struct action_xlate_ctx *ctx,
- const struct nx_action_header *nah)
-{
- const struct nx_action_resubmit *nar;
- const struct nx_action_set_tunnel *nast;
- const struct nx_action_set_queue *nasq;
- union odp_action *oa;
- int subtype = ntohs(nah->subtype);
-
- assert(nah->vendor == htonl(NX_VENDOR_ID));
- switch (subtype) {
- case NXAST_RESUBMIT:
- nar = (const struct nx_action_resubmit *) nah;
- xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
- break;
-
- case NXAST_SET_TUNNEL:
- nast = (const struct nx_action_set_tunnel *) nah;
- oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
- ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
- break;
-
- case NXAST_DROP_SPOOFED_ARP:
- if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
- odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
- }
- break;
-
- case NXAST_SET_QUEUE:
- nasq = (const struct nx_action_set_queue *) nah;
- xlate_set_queue_action(ctx, nasq);
- break;
-
- case NXAST_POP_QUEUE:
- odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
- break;
-
- /* If you add a new action here that modifies flow data, don't forget to
- * update the flow key in ctx->flow at the same time. */
-
- default:
- VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
- break;
- }
-}
-
-static void
-do_xlate_actions(const union ofp_action *in, size_t n_in,
- struct action_xlate_ctx *ctx)
-{
- struct actions_iterator iter;
- const union ofp_action *ia;
- const struct ofport *port;
-
- port = get_port(ctx->ofproto, ctx->flow.in_port);
- if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
- port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
- ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
- /* Drop this flow. */
- return;
- }
-
- for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
- uint16_t type = ntohs(ia->type);
- union odp_action *oa;
-
- switch (type) {
- case OFPAT_OUTPUT:
- xlate_output_action(ctx, &ia->output);
- break;
-
- case OFPAT_SET_VLAN_VID:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
- ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
- break;
-
- case OFPAT_SET_VLAN_PCP:
- oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
- ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
- break;
-
- case OFPAT_STRIP_VLAN:
- odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
- ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
- ctx->flow.dl_vlan_pcp = 0;
- break;
-
- case OFPAT_SET_DL_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_src,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
-
- case OFPAT_SET_DL_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
- memcpy(oa->dl_addr.dl_addr,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- memcpy(ctx->flow.dl_dst,
- ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
- break;
-
- case OFPAT_SET_NW_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
- ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
-
- case OFPAT_SET_NW_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
- ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
- break;
-
- case OFPAT_SET_NW_TOS:
- oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
- ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
- break;
-
- case OFPAT_SET_TP_SRC:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
- ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
-
- case OFPAT_SET_TP_DST:
- oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
- ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
- break;
-
- case OFPAT_VENDOR:
- xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
- break;
-
- case OFPAT_ENQUEUE:
- xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
- break;
-
- default:
- VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
- break;
- }
- }
-}
-
-static int
-xlate_actions(const union ofp_action *in, size_t n_in,
- const flow_t *flow, struct ofproto *ofproto,
- const struct ofpbuf *packet,
- struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
- uint16_t *nf_output_iface)
-{
- tag_type no_tags = 0;
- struct action_xlate_ctx ctx;
- COVERAGE_INC(ofproto_ofp2odp);
- odp_actions_init(out);
- ctx.flow = *flow;
- ctx.recurse = 0;
- ctx.ofproto = ofproto;
- ctx.packet = packet;
- ctx.out = out;
- ctx.tags = tags ? tags : &no_tags;
- ctx.may_set_up_flow = true;
- ctx.nf_output_iface = NF_OUT_DROP;
- do_xlate_actions(in, n_in, &ctx);
- remove_pop_action(&ctx);
-
- /* Check with in-band control to see if we're allowed to set up this
- * flow. */
- if (!in_band_rule_check(ofproto->in_band, flow, out)) {
- ctx.may_set_up_flow = false;
- }
-
- if (may_set_up_flow) {
- *may_set_up_flow = ctx.may_set_up_flow;
- }
- if (nf_output_iface) {
- *nf_output_iface = ctx.nf_output_iface;
- }
- if (odp_actions_overflow(out)) {
- COVERAGE_INC(odp_overflow);
- odp_actions_init(out);
- return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
- }
- return 0;
-}
-
/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
* error message code (composed with ofp_mkerr()) for the caller to propagate
* upward. Otherwise, returns 0.
{
struct ofp_packet_out *opo;
struct ofpbuf payload, *buffer;
- struct odp_actions actions;
+ struct ofp_action_header *actions;
int n_actions;
uint16_t in_port;
flow_t flow;
return error;
}
opo = (struct ofp_packet_out *) oh;
+ actions = opo->actions;
COVERAGE_INC(ofproto_packet_out);
if (opo->buffer_id != htonl(UINT32_MAX)) {
buffer = NULL;
}
- flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
- error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
- &flow, p, &payload, &actions, NULL, NULL, NULL);
- if (error) {
- return error;
- }
-
- dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
- &payload);
+ flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
+ wdp_execute(p->wdp, flow.in_port, (const union ofp_action *) actions,
+ n_actions, &payload);
ofpbuf_delete(buffer);
return 0;
}
-static void
-update_port_config(struct ofproto *p, struct ofport *port,
- uint32_t config, uint32_t mask)
-{
- mask &= config ^ port->opp.config;
- if (mask & OFPPC_PORT_DOWN) {
- if (config & OFPPC_PORT_DOWN) {
- netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
- } else {
- netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
- }
- }
-#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
- if (mask & REVALIDATE_BITS) {
- COVERAGE_INC(ofproto_costly_flags);
- port->opp.config ^= mask & REVALIDATE_BITS;
- p->need_revalidate = true;
- }
-#undef REVALIDATE_BITS
- if (mask & OFPPC_NO_FLOOD) {
- port->opp.config ^= OFPPC_NO_FLOOD;
- refresh_port_groups(p);
- }
- if (mask & OFPPC_NO_PACKET_IN) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
- }
-}
-
static int
handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
struct ofp_header *oh)
{
const struct ofp_port_mod *opm;
- struct ofport *port;
+ struct wdp_port port;
int error;
error = reject_slave_controller(ofconn, oh);
}
opm = (struct ofp_port_mod *) oh;
- port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
- if (!port) {
- return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
- } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
- return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
+ if (wdp_port_query_by_number(p->wdp, ntohs(opm->port_no), &port)) {
+ error = ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
+ } else if (memcmp(port.opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
+ error = ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
} else {
- update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
+ uint32_t mask, new_config;
+
+ mask = ntohl(opm->mask) & (OFPPC_PORT_DOWN | OFPPC_NO_STP
+ | OFPPC_NO_RECV | OFPPC_NO_RECV_STP
+ | OFPPC_NO_FLOOD | OFPPC_NO_FWD
+ | OFPPC_NO_PACKET_IN);
+ new_config = (port.opp.config & ~mask) | (ntohl(opm->config) & mask);
+ if (new_config != port.opp.config) {
+ error = wdp_port_set_config(p->wdp, ntohs(opm->port_no),
+ new_config);
+ }
if (opm->advertise) {
- netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
+ netdev_set_advertisements(port.netdev, ntohl(opm->advertise));
}
}
- return 0;
+ wdp_port_free(&port);
+
+ return error;
}
static struct ofpbuf *
return 0;
}
-static void
-count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
-{
- struct rule *rule = rule_from_cls_rule(cls_rule);
- int *n_subrules = n_subrules_;
-
- if (rule->super) {
- (*n_subrules)++;
- }
-}
-
static int
handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
struct ofp_stats_request *request)
{
- struct ofp_table_stats *ots;
struct ofpbuf *msg;
- struct odp_stats dpstats;
- int n_exact, n_subrules, n_wild;
-
- msg = start_stats_reply(request, sizeof *ots * 2);
-
- /* Count rules of various kinds. */
- n_subrules = 0;
- classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
- n_exact = classifier_count_exact(&p->cls) - n_subrules;
- n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
-
- /* Hash table. */
- dpif_get_dp_stats(p->dpif, &dpstats);
- ots = append_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_HASH;
- strcpy(ots->name, "hash");
- ots->wildcards = htonl(0);
- ots->max_entries = htonl(dpstats.max_capacity);
- ots->active_count = htonl(n_exact);
- ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
- dpstats.n_missed);
- ots->matched_count = htonll(dpstats.n_hit); /* XXX */
-
- /* Classifier table. */
- ots = append_stats_reply(sizeof *ots, ofconn, &msg);
- memset(ots, 0, sizeof *ots);
- ots->table_id = TABLEID_CLASSIFIER;
- strcpy(ots->name, "classifier");
- ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
- : htonl(OFPFW_ALL);
- ots->max_entries = htonl(65536);
- ots->active_count = htonl(n_wild);
- ots->lookup_count = htonll(0); /* XXX */
- ots->matched_count = htonll(0); /* XXX */
+ int error;
- queue_tx(msg, ofconn, ofconn->reply_counter);
- return 0;
+ msg = start_stats_reply(request, sizeof(struct ofp_table_stats) * 3);
+ error = wdp_get_table_stats(p->wdp, msg);
+ if (!error) {
+ queue_tx(msg, ofconn, ofconn->reply_counter);
+ } else {
+ ofpbuf_delete(msg);
+ }
+ return error;
}
static void
-append_port_stat(struct ofport *port, struct ofconn *ofconn,
+append_port_stat(struct wdp_port *port, struct ofconn *ofconn,
struct ofpbuf **msgp)
{
struct netdev_stats stats;
struct ofp_port_stats_request *psr;
struct ofp_port_stats *ops;
struct ofpbuf *msg;
- struct ofport *port;
if (arg_size != sizeof *psr) {
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
msg = start_stats_reply(osr, sizeof *ops * 16);
if (psr->port_no != htons(OFPP_NONE)) {
- port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
- if (port) {
- append_port_stat(port, ofconn, &msg);
+ struct wdp_port port;
+
+ if (!wdp_port_query_by_number(p->wdp, ntohs(psr->port_no), &port)) {
+ append_port_stat(&port, ofconn, &msg);
+ wdp_port_free(&port);
}
} else {
- HMAP_FOR_EACH (port, hmap_node, &p->ports) {
- append_port_stat(port, ofconn, &msg);
+ struct wdp_port *ports;
+ size_t n_ports;
+ size_t i;
+
+ wdp_port_list(p->wdp, &ports, &n_ports);
+ for (i = 0; i < n_ports; i++) {
+ append_port_stat(&ports[i], ofconn, &msg);
}
+ wdp_port_array_free(ports, n_ports);
}
queue_tx(msg, ofconn, ofconn->reply_counter);
* '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
* returned statistic include statistics for all of 'rule''s subrules. */
static void
-query_stats(struct ofproto *p, struct rule *rule,
+query_stats(struct ofproto *p, struct wdp_rule *rule,
uint64_t *packet_countp, uint64_t *byte_countp)
{
- uint64_t packet_count, byte_count;
- struct rule *subrule;
- struct odp_flow *odp_flows;
- size_t n_odp_flows;
-
- /* Start from historical data for 'rule' itself that are no longer tracked
- * by the datapath. This counts, for example, subrules that have
- * expired. */
- packet_count = rule->packet_count;
- byte_count = rule->byte_count;
-
- /* Prepare to ask the datapath for statistics on 'rule', or if it is
- * wildcarded then on all of its subrules.
- *
- * Also, add any statistics that are not tracked by the datapath for each
- * subrule. This includes, for example, statistics for packets that were
- * executed "by hand" by ofproto via dpif_execute() but must be accounted
- * to a flow. */
- n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
- odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
- if (rule->cr.wc.wildcards) {
- size_t i = 0;
- LIST_FOR_EACH (subrule, list, &rule->list) {
- odp_flows[i++].key = subrule->cr.flow;
- packet_count += subrule->packet_count;
- byte_count += subrule->byte_count;
- }
- } else {
- odp_flows[0].key = rule->cr.flow;
- }
+ struct wdp_flow_stats stats;
- /* Fetch up-to-date statistics from the datapath and add them in. */
- if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
- size_t i;
- for (i = 0; i < n_odp_flows; i++) {
- struct odp_flow *odp_flow = &odp_flows[i];
- packet_count += odp_flow->stats.n_packets;
- byte_count += odp_flow->stats.n_bytes;
- }
+ if (!wdp_flow_get_stats(p->wdp, rule, &stats)) {
+ *packet_countp = stats.n_packets;
+ *byte_countp = stats.n_bytes;
+ } else {
+ *packet_countp = 0;
+ *byte_countp = 0;
}
- free(odp_flows);
-
- /* Return the stats to the caller. */
- *packet_countp = packet_count;
- *byte_countp = byte_count;
}
-static void
-flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
+static int
+flow_stats_cb(struct wdp_rule *rule, void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(rule_);
struct flow_stats_cbdata *cbdata = cbdata_;
struct ofp_flow_stats *ofs;
uint64_t packet_count, byte_count;
uint32_t sec = tdiff / 1000;
uint32_t msec = tdiff - (sec * 1000);
- if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
- return;
+ if (rule_is_hidden(rule)
+ || !rule_has_out_port(rule, cbdata->out_port)) {
+ return 0;
}
act_len = sizeof *rule->actions * rule->n_actions;
ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
ofs->length = htons(len);
- ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
+ ofs->table_id = rule->ofp_table_id;
ofs->pad = 0;
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
- cbdata->ofproto->tun_id_from_cookie, &ofs->match);
+ flow_to_match(&rule->cr.flow, cbdata->ofproto->tun_id_from_cookie,
+ &ofs->match);
ofs->duration_sec = htonl(sec);
ofs->duration_nsec = htonl(msec * 1000000);
- ofs->cookie = rule->flow_cookie;
- ofs->priority = htons(rule->cr.priority);
+ ofs->cookie = ofproto_rule_cast(rule)->flow_cookie;
+ ofs->priority = htons(rule->cr.flow.priority);
ofs->idle_timeout = htons(rule->idle_timeout);
ofs->hard_timeout = htons(rule->hard_timeout);
memset(ofs->pad2, 0, sizeof ofs->pad2);
ofs->packet_count = htonll(packet_count);
ofs->byte_count = htonll(byte_count);
memcpy(ofs->actions, rule->actions, act_len);
+
+ return 0;
}
-static int
+static unsigned int
table_id_to_include(uint8_t table_id)
{
- return (table_id == TABLEID_HASH ? CLS_INC_EXACT
- : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
- : table_id == 0xff ? CLS_INC_ALL
+ return (table_id == 0xff ? UINT_MAX
+ : table_id < 32 ? 1u << table_id
: 0);
}
+static unsigned int
+flow_mod_table_id(const struct ofconn *ofconn, const struct ofp_flow_mod *ofm)
+{
+ return ofconn->flow_mod_table_id ? ntohs(ofm->command) >> 8 : 0xff;
+}
+
static int
handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
const struct ofp_stats_request *osr,
{
struct ofp_flow_stats_request *fsr;
struct flow_stats_cbdata cbdata;
- struct cls_rule target;
+ flow_t target;
if (arg_size != sizeof *fsr) {
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
cbdata.ofconn = ofconn;
cbdata.out_port = fsr->out_port;
cbdata.msg = start_stats_reply(osr, 1024);
- cls_rule_from_match(&fsr->match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target,
- table_id_to_include(fsr->table_id),
- flow_stats_cb, &cbdata);
+ flow_from_match(&fsr->match, 0, false, 0, &target);
+ wdp_flow_for_each_match(p->wdp, &target,
+ table_id_to_include(fsr->table_id),
+ flow_stats_cb, &cbdata);
queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
return 0;
}
struct ds *results;
};
-static void
-flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
+static int
+flow_stats_ds_cb(struct wdp_rule *rule, void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(rule_);
struct flow_stats_ds_cbdata *cbdata = cbdata_;
struct ds *results = cbdata->results;
struct ofp_match match;
uint64_t packet_count, byte_count;
size_t act_len = sizeof *rule->actions * rule->n_actions;
- /* Don't report on subrules. */
- if (rule->super != NULL) {
- return;
- }
-
query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
- cbdata->ofproto->tun_id_from_cookie, &match);
+ flow_to_match(&rule->cr.flow, cbdata->ofproto->tun_id_from_cookie,
+ &match);
ds_put_format(results, "duration=%llds, ",
(time_msec() - rule->created) / 1000);
- ds_put_format(results, "priority=%u, ", rule->cr.priority);
+ ds_put_format(results, "priority=%u, ", rule->cr.flow.priority);
ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
ofp_print_match(results, &match, true);
ofp_print_actions(results, &rule->actions->header, act_len);
ds_put_cstr(results, "\n");
+
+ return 0;
}
/* Adds a pretty-printed description of all flows to 'results', including
void
ofproto_get_all_flows(struct ofproto *p, struct ds *results)
{
- struct ofp_match match;
- struct cls_rule target;
struct flow_stats_ds_cbdata cbdata;
+ struct ofp_match match;
+ flow_t target;
memset(&match, 0, sizeof match);
match.wildcards = htonl(OVSFW_ALL);
cbdata.ofproto = p;
cbdata.results = results;
- cls_rule_from_match(&match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- flow_stats_ds_cb, &cbdata);
+ flow_from_match(&match, 0, false, 0, &target);
+ wdp_flow_for_each_match(p->wdp, &target, UINT_MAX,
+ flow_stats_ds_cb, &cbdata);
}
struct aggregate_stats_cbdata {
uint32_t n_flows;
};
-static void
-aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
+static int
+aggregate_stats_cb(struct wdp_rule *rule, void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(rule_);
struct aggregate_stats_cbdata *cbdata = cbdata_;
uint64_t packet_count, byte_count;
if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
- return;
+ return 0;
}
query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
cbdata->packet_count += packet_count;
cbdata->byte_count += byte_count;
cbdata->n_flows++;
+
+ return 0;
}
static int
struct ofp_aggregate_stats_request *asr;
struct ofp_aggregate_stats_reply *reply;
struct aggregate_stats_cbdata cbdata;
- struct cls_rule target;
struct ofpbuf *msg;
+ flow_t target;
if (arg_size != sizeof *asr) {
return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
cbdata.packet_count = 0;
cbdata.byte_count = 0;
cbdata.n_flows = 0;
- cls_rule_from_match(&asr->match, 0, false, 0, &target);
- classifier_for_each_match(&p->cls, &target,
- table_id_to_include(asr->table_id),
- aggregate_stats_cb, &cbdata);
+ flow_from_match(&asr->match, 0, false, 0, &target);
+ wdp_flow_for_each_match(p->wdp, &target,
+ table_id_to_include(asr->table_id),
+ aggregate_stats_cb, &cbdata);
msg = start_stats_reply(osr, sizeof *reply);
reply = append_stats_reply(sizeof *reply, ofconn, &msg);
struct queue_stats_cbdata {
struct ofconn *ofconn;
- struct ofport *ofport;
++ struct wdp_port *wdp_port;
struct ofpbuf *msg;
- uint16_t port_no;
};
static void
struct ofp_queue_stats *reply;
reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
- reply->port_no = htons(cbdata->port_no);
- reply->port_no = htons(cbdata->ofport->opp.port_no);
++ reply->port_no = htons(cbdata->wdp_port->opp.port_no);
memset(reply->pad, 0, sizeof reply->pad);
reply->queue_id = htonl(queue_id);
reply->tx_bytes = htonll(stats->tx_bytes);
}
static void
-handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
+handle_queue_stats_for_port(struct wdp_port *port, uint32_t queue_id,
struct queue_stats_cbdata *cbdata)
{
- cbdata->port_no = port->opp.port_no;
- cbdata->ofport = port;
++ cbdata->wdp_port = port;
if (queue_id == OFPQ_ALL) {
netdev_dump_queue_stats(port->netdev,
handle_queue_stats_dump_cb, cbdata);
} else {
struct netdev_queue_stats stats;
- netdev_get_queue_stats(port->netdev, queue_id, &stats);
- put_queue_stats(cbdata, queue_id, &stats);
+ if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
+ put_queue_stats(cbdata, queue_id, &stats);
+ }
}
}
{
struct ofp_queue_stats_request *qsr;
struct queue_stats_cbdata cbdata;
- struct ofport *port;
unsigned int port_no;
uint32_t queue_id;
port_no = ntohs(qsr->port_no);
queue_id = ntohl(qsr->queue_id);
if (port_no == OFPP_ALL) {
- HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
- handle_queue_stats_for_port(port, queue_id, &cbdata);
+ struct wdp_port *ports;
+ size_t n_ports, i;
+
+ wdp_port_list(ofproto->wdp, &ports, &n_ports);
+ /* XXX deal with wdp_port_list() errors */
+ for (i = 0; i < n_ports; i++) {
+ handle_queue_stats_for_port(&ports[i], queue_id, &cbdata);
}
+ wdp_port_array_free(ports, n_ports);
} else if (port_no < ofproto->max_ports) {
- port = get_port(ofproto, ofp_port_to_odp_port(port_no));
- if (port) {
- handle_queue_stats_for_port(port, queue_id, &cbdata);
+ struct wdp_port port;
+ int error;
+
+ error = wdp_port_query_by_number(ofproto->wdp, port_no, &port);
+ if (!error) {
+ handle_queue_stats_for_port(&port, queue_id, &cbdata);
+ } else {
+ /* XXX deal with wdp_port_query_by_number() errors */
}
+ wdp_port_free(&port);
} else {
ofpbuf_delete(cbdata.msg);
return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
}
}
-static long long int
-msec_from_nsec(uint64_t sec, uint32_t nsec)
-{
- return !sec ? 0 : sec * 1000 + nsec / 1000000;
-}
-
-static void
-update_time(struct ofproto *ofproto, struct rule *rule,
- const struct odp_flow_stats *stats)
-{
- long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
- if (used > rule->used) {
- rule->used = used;
- if (rule->super && used > rule->super->used) {
- rule->super->used = used;
- }
- netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
- }
-}
-
-static void
-update_stats(struct ofproto *ofproto, struct rule *rule,
- const struct odp_flow_stats *stats)
-{
- if (stats->n_packets) {
- update_time(ofproto, rule, stats);
- rule->packet_count += stats->n_packets;
- rule->byte_count += stats->n_bytes;
- netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
- }
-}
-
/* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
* in which no matching flow already exists in the flow table.
*
add_flow(struct ofproto *p, struct ofconn *ofconn,
const struct ofp_flow_mod *ofm, size_t n_actions)
{
+ struct wdp_rule *rule;
+ struct wdp_flow_put put;
struct ofpbuf *packet;
- struct rule *rule;
uint16_t in_port;
+ flow_t flow;
int error;
- if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
- flow_t flow;
- uint32_t wildcards;
-
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
- ntohs(ofm->priority))) {
- return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
- }
+ flow_from_match(&ofm->match, ntohs(ofm->priority), p->tun_id_from_cookie,
+ ofm->cookie, &flow);
+ if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)
+ && wdp_flow_overlaps(p->wdp, &flow)) {
+ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
+ }
+
+ put.flags = WDP_PUT_CREATE | WDP_PUT_MODIFY | WDP_PUT_ALL;
+ put.flow = &flow;
+ put.actions = (const union ofp_action *) ofm->actions;
+ put.n_actions = n_actions;
+ put.idle_timeout = ntohs(ofm->idle_timeout);
+ put.hard_timeout = ntohs(ofm->hard_timeout);
+ put.ofp_table_id = flow_mod_table_id(ofconn, ofm);
+ put.cookie = ofm->cookie;
+ put.xid = ofm->header.xid;
+ error = wdp_flow_put(p->wdp, &put, NULL, &rule);
+ if (error) {
+ /* XXX wdp_flow_put should return OpenFlow error code. */
+ return error;
}
+ ofproto_rule_init(rule);
- rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
- n_actions, ntohs(ofm->idle_timeout),
- ntohs(ofm->hard_timeout), ofm->cookie,
- ofm->flags & htons(OFPFF_SEND_FLOW_REM));
- cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
- p->tun_id_from_cookie, ofm->cookie, &rule->cr);
-
- error = 0;
if (ofm->buffer_id != htonl(UINT32_MAX)) {
error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
&packet, &in_port);
- } else {
- packet = NULL;
- in_port = UINT16_MAX;
+ if (!error) {
+ error = wdp_flow_inject(p->wdp, rule, in_port, packet);
+ ofpbuf_delete(packet);
+ }
}
- rule_insert(p, rule, packet, in_port);
return error;
}
-static struct rule *
-find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
+static struct wdp_rule *
+find_flow_strict(struct ofproto *p, const struct ofconn *ofconn,
+ const struct ofp_flow_mod *ofm)
{
- uint32_t wildcards;
+ uint8_t table_id;
flow_t flow;
- flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
- &flow, &wildcards);
- return rule_from_cls_rule(classifier_find_rule_exactly(
- &p->cls, &flow, wildcards,
- ntohs(ofm->priority)));
+ flow_from_match(&ofm->match, ntohs(ofm->priority),
+ p->tun_id_from_cookie, ofm->cookie, &flow);
+ table_id = flow_mod_table_id(ofconn, ofm);
+ return wdp_flow_get(p->wdp, &flow, table_id_to_include(table_id));
}
static int
send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
- struct rule *rule, const struct ofp_flow_mod *ofm)
+ struct wdp_rule *rule, const struct ofp_flow_mod *ofm)
{
struct ofpbuf *packet;
uint16_t in_port;
- flow_t flow;
int error;
if (ofm->buffer_id == htonl(UINT32_MAX)) {
return error;
}
- flow_extract(packet, 0, in_port, &flow);
- rule_execute(ofproto, rule, packet, &flow);
+ error = wdp_flow_inject(ofproto->wdp, rule, in_port, packet);
+ ofpbuf_delete(packet);
- return 0;
+ return error;
}
\f
/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
struct ofproto *ofproto;
const struct ofp_flow_mod *ofm;
size_t n_actions;
- struct rule *match;
+ struct wdp_rule *match;
};
static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
- size_t n_actions, struct rule *);
-static void modify_flows_cb(struct cls_rule *, void *cbdata_);
+ size_t n_actions, struct wdp_rule *)
+ WARN_UNUSED_RESULT;
+static int modify_flows_cb(struct wdp_rule *, void *cbdata_);
-/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
- * encoded by ofp_mkerr() on failure.
+/* Implements OFPFC_ADD and OFPFC_MODIFY. Returns 0 on success or an OpenFlow
+ * error code as encoded by ofp_mkerr() on failure.
*
* 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
* if any. */
const struct ofp_flow_mod *ofm, size_t n_actions)
{
struct modify_flows_cbdata cbdata;
- struct cls_rule target;
+ uint8_t table_id;
+ flow_t target;
+ int error;
cbdata.ofproto = p;
cbdata.ofm = ofm;
cbdata.n_actions = n_actions;
cbdata.match = NULL;
- cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
- &target);
+ flow_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
+ &target);
+
+ table_id = flow_mod_table_id(ofconn, ofm);
+ error = wdp_flow_for_each_match(p->wdp, &target,
+ table_id_to_include(table_id),
+ modify_flows_cb, &cbdata);
+ if (error) {
+ return error;
+ }
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- modify_flows_cb, &cbdata);
if (cbdata.match) {
- /* This credits the packet to whichever flow happened to happened to
- * match last. That's weird. Maybe we should do a lookup for the
- * flow that actually matches the packet? Who knows. */
- send_buffered_packet(p, ofconn, cbdata.match, ofm);
- return 0;
+ /* This credits the packet to whichever flow happened to match last.
+ * That's weird. Maybe we should do a lookup for the flow that
+ * actually matches the packet? Who knows. */
+ return send_buffered_packet(p, ofconn, cbdata.match, ofm);
} else {
return add_flow(p, ofconn, ofm, n_actions);
}
modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
struct ofp_flow_mod *ofm, size_t n_actions)
{
- struct rule *rule = find_flow_strict(p, ofm);
+ struct wdp_rule *rule = find_flow_strict(p, ofconn, ofm);
if (rule && !rule_is_hidden(rule)) {
- modify_flow(p, ofm, n_actions, rule);
- return send_buffered_packet(p, ofconn, rule, ofm);
+ int error = modify_flow(p, ofm, n_actions, rule);
+ return error ? error : send_buffered_packet(p, ofconn, rule, ofm);
} else {
return add_flow(p, ofconn, ofm, n_actions);
}
}
/* Callback for modify_flows_loose(). */
-static void
-modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
+static int
+modify_flows_cb(struct wdp_rule *rule, void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(rule_);
struct modify_flows_cbdata *cbdata = cbdata_;
if (!rule_is_hidden(rule)) {
cbdata->match = rule;
- modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
+ return modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions,
+ rule);
}
+ return 0;
}
/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
* ofp_action[] structures). */
static int
modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
- size_t n_actions, struct rule *rule)
+ size_t n_actions, struct wdp_rule *rule)
{
- size_t actions_len = n_actions * sizeof *rule->actions;
+ const struct ofp_action_header *actions = ofm->actions;
+ struct ofproto_rule *ofproto_rule = ofproto_rule_cast(rule);
+ struct wdp_flow_put put;
- rule->flow_cookie = ofm->cookie;
+ ofproto_rule->flow_cookie = ofm->cookie;
/* If the actions are the same, do nothing. */
if (n_actions == rule->n_actions
- && !memcmp(ofm->actions, rule->actions, actions_len))
+ && !memcmp(ofm->actions, rule->actions, sizeof *actions * n_actions))
{
return 0;
}
- /* Replace actions. */
- free(rule->actions);
- rule->actions = xmemdup(ofm->actions, actions_len);
- rule->n_actions = n_actions;
-
- /* Make sure that the datapath gets updated properly. */
- if (rule->cr.wc.wildcards) {
- COVERAGE_INC(ofproto_mod_wc_flow);
- p->need_revalidate = true;
- } else {
- rule_update_actions(p, rule);
- }
-
- return 0;
+ put.flags = WDP_PUT_MODIFY | WDP_PUT_ACTIONS;
+ put.flow = &rule->cr.flow;
+ put.actions = (const union ofp_action *) actions;
+ put.n_actions = n_actions;
+ put.idle_timeout = put.hard_timeout = 0;
+ put.ofp_table_id = rule->ofp_table_id;
+ put.cookie = ofm->cookie;
+ put.xid = ofm->header.xid;
+ return wdp_flow_put(p->wdp, &put, NULL, NULL);
}
\f
/* OFPFC_DELETE implementation. */
uint16_t out_port;
};
-static void delete_flows_cb(struct cls_rule *, void *cbdata_);
-static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
+static int delete_flows_cb(struct wdp_rule *, void *cbdata_);
+static int delete_flow_core(struct ofproto *, struct wdp_rule *,
+ uint16_t out_port);
/* Implements OFPFC_DELETE. */
-static void
-delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
+static int
+delete_flows_loose(struct ofproto *p, const struct ofconn *ofconn,
+ const struct ofp_flow_mod *ofm)
{
struct delete_flows_cbdata cbdata;
- struct cls_rule target;
+ uint8_t table_id;
+ flow_t target;
cbdata.ofproto = p;
cbdata.out_port = ofm->out_port;
- cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
- &target);
+ flow_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
+ &target);
+ table_id = flow_mod_table_id(ofconn, ofm);
- classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
- delete_flows_cb, &cbdata);
+ return wdp_flow_for_each_match(p->wdp, &target,
+ table_id_to_include(table_id),
+ delete_flows_cb, &cbdata);
}
/* Implements OFPFC_DELETE_STRICT. */
-static void
-delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
+static int
+delete_flow_strict(struct ofproto *p, const struct ofconn *ofconn,
+ struct ofp_flow_mod *ofm)
{
- struct rule *rule = find_flow_strict(p, ofm);
+ struct wdp_rule *rule = find_flow_strict(p, ofconn, ofm);
if (rule) {
- delete_flow(p, rule, ofm->out_port);
+ return delete_flow_core(p, rule, ofm->out_port);
}
+ return 0;
}
/* Callback for delete_flows_loose(). */
-static void
-delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
+static int
+delete_flows_cb(struct wdp_rule *rule, void *cbdata_)
{
- struct rule *rule = rule_from_cls_rule(rule_);
struct delete_flows_cbdata *cbdata = cbdata_;
- delete_flow(cbdata->ofproto, rule, cbdata->out_port);
+ return delete_flow_core(cbdata->ofproto, rule, cbdata->out_port);
}
/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
* Will not delete 'rule' if it is hidden. Will delete 'rule' only if
* 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
* specified 'out_port'. */
-static void
-delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
+static int
+delete_flow_core(struct ofproto *p, struct wdp_rule *rule, uint16_t out_port)
{
if (rule_is_hidden(rule)) {
- return;
+ return 0;
}
if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
- return;
+ return 0;
}
- send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
- rule_remove(p, rule);
+ return delete_flow(p, rule, OFPRR_DELETE);
}
\f
static int
return error;
}
- switch (ntohs(ofm->command)) {
+ if (!ofconn->flow_mod_table_id && ofm->command & htons(0xff00)) {
+ static struct vlog_rate_limit table_id_rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_WARN_RL(&table_id_rl, "%s: flow_mod table_id feature must be "
+ "enabled with NXT_FLOW_MOD_TABLE_ID",
+ rconn_get_name(ofconn->rconn));
+ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
+ }
+
+ switch (ntohs(ofm->command) & 0xff) {
case OFPFC_ADD:
return add_flow(p, ofconn, ofm, n_actions);
return modify_flow_strict(p, ofconn, ofm, n_actions);
case OFPFC_DELETE:
- delete_flows_loose(p, ofm);
- return 0;
+ return delete_flows_loose(p, ofconn, ofm);
case OFPFC_DELETE_STRICT:
- delete_flow_strict(p, ofm);
- return 0;
+ return delete_flow_strict(p, ofconn, ofm);
default:
return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
return 0;
}
+static int
+handle_flow_mod_table_id(struct ofconn *ofconn,
+ struct nxt_flow_mod_table_id *msg)
+{
+ int error;
+
+ error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
+ if (error) {
+ return error;
+ }
+
+ ofconn->flow_mod_table_id = !!msg->set;
+ return 0;
+}
+
static int
handle_role_request(struct ofproto *ofproto,
struct ofconn *ofconn, struct nicira_header *msg)
if (role == NX_ROLE_MASTER) {
struct ofconn *other;
- HMAP_FOR_EACH (other, struct ofconn, hmap_node,
- &ofproto->controllers) {
+ HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
if (other->role == NX_ROLE_MASTER) {
other->role = NX_ROLE_SLAVE;
}
case NXT_TUN_ID_FROM_COOKIE:
return handle_tun_id_from_cookie(p, msg);
+ case NXT_FLOW_MOD_TABLE_ID:
+ return handle_flow_mod_table_id(ofconn, msg);
+
case NXT_ROLE_REQUEST:
return handle_role_request(p, ofconn, msg);
}
}
\f
static void
-handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
+handle_flow_miss(struct ofproto *p, struct wdp_packet *packet)
{
- struct odp_msg *msg = packet->data;
- struct rule *rule;
- struct ofpbuf payload;
+ struct wdp_rule *rule;
flow_t flow;
- payload.data = msg + 1;
- payload.size = msg->length - sizeof *msg;
- flow_extract(&payload, msg->arg, msg->port, &flow);
-
- /* Check with in-band control to see if this packet should be sent
- * to the local port regardless of the flow table. */
- if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
- union odp_action action;
-
- memset(&action, 0, sizeof(action));
- action.output.type = ODPAT_OUTPUT;
- action.output.port = ODPP_LOCAL;
- dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
- }
-
- rule = lookup_valid_rule(p, &flow);
+ flow_extract(packet->payload, packet->tun_id, packet->in_port, &flow);
+ rule = wdp_flow_match(p->wdp, &flow);
if (!rule) {
/* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
- struct ofport *port = get_port(p, msg->port);
- if (port) {
- if (port->opp.config & OFPPC_NO_PACKET_IN) {
+ struct wdp_port port;
+
+ if (!wdp_port_query_by_number(p->wdp, packet->in_port, &port)) {
+ bool no_packet_in = (port.opp.config & OFPPC_NO_PACKET_IN) != 0;
+ wdp_port_free(&port);
+ if (no_packet_in) {
COVERAGE_INC(ofproto_no_packet_in);
- /* XXX install 'drop' flow entry */
- ofpbuf_delete(packet);
+ wdp_packet_destroy(packet);
return;
}
} else {
- VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
+ VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
+ packet->in_port);
}
COVERAGE_INC(ofproto_packet_in);
return;
}
- if (rule->cr.wc.wildcards) {
- rule = rule_create_subrule(p, rule, &flow);
- rule_make_actions(p, rule, packet);
- } else {
- if (!rule->may_install) {
- /* The rule is not installable, that is, we need to process every
- * packet, so process the current packet and set its actions into
- * 'subrule'. */
- rule_make_actions(p, rule, packet);
- } else {
- /* XXX revalidate rule if it needs it */
- }
- }
+ wdp_flow_inject(p->wdp, rule, packet->in_port, packet->payload);
- if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
+ if (rule->cr.flow.priority == FAIL_OPEN_PRIORITY) {
/*
* Extra-special case for fail-open mode.
*
*
* See the top-level comment in fail-open.c for more information.
*/
- send_packet_in(p, ofpbuf_clone_with_headroom(packet,
- DPIF_RECV_MSG_PADDING));
+ send_packet_in(p, packet);
+ } else {
+ wdp_packet_destroy(packet);
}
-
- ofpbuf_pull(packet, sizeof *msg);
- rule_execute(p, rule, packet, &flow);
- rule_reinstall(p, rule);
}
static void
-handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
+handle_wdp_packet(struct ofproto *p, struct wdp_packet *packet)
{
- struct odp_msg *msg = packet->data;
-
- switch (msg->type) {
- case _ODPL_ACTION_NR:
+ switch (packet->channel) {
+ case WDP_CHAN_ACTION:
COVERAGE_INC(ofproto_ctlr_action);
send_packet_in(p, packet);
break;
- case _ODPL_SFLOW_NR:
- if (p->sflow) {
- ofproto_sflow_received(p->sflow, msg);
- }
- ofpbuf_delete(packet);
+ case WDP_CHAN_SFLOW:
+ /* XXX */
+ wdp_packet_destroy(packet);
break;
- case _ODPL_MISS_NR:
- handle_odp_miss_msg(p, packet);
+ case WDP_CHAN_MISS:
+ handle_flow_miss(p, packet);
break;
+ case WDP_N_CHANS:
default:
- VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
- msg->type);
+ wdp_packet_destroy(packet);
+ VLOG_WARN_RL(&rl, "received message on unexpected channel %d",
+ (int) packet->channel);
break;
}
}
\f
-static void
-revalidate_cb(struct cls_rule *sub_, void *cbdata_)
-{
- struct rule *sub = rule_from_cls_rule(sub_);
- struct revalidate_cbdata *cbdata = cbdata_;
-
- if (cbdata->revalidate_all
- || (cbdata->revalidate_subrules && sub->super)
- || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
- revalidate_rule(cbdata->ofproto, sub);
- }
-}
-
-static bool
-revalidate_rule(struct ofproto *p, struct rule *rule)
-{
- const flow_t *flow = &rule->cr.flow;
-
- COVERAGE_INC(ofproto_revalidate_rule);
- if (rule->super) {
- struct rule *super;
- super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
- if (!super) {
- rule_remove(p, rule);
- return false;
- } else if (super != rule->super) {
- COVERAGE_INC(ofproto_revalidate_moved);
- list_remove(&rule->list);
- list_push_back(&super->list, &rule->list);
- rule->super = super;
- rule->hard_timeout = super->hard_timeout;
- rule->idle_timeout = super->idle_timeout;
- rule->created = super->created;
- rule->used = 0;
- }
- }
-
- rule_update_actions(p, rule);
- return true;
-}
-
static struct ofpbuf *
-compose_flow_removed(struct ofproto *p, const struct rule *rule,
- long long int now, uint8_t reason)
+compose_flow_removed(struct ofproto *p, const struct wdp_rule *rule,
+ uint8_t reason)
{
- struct ofp_flow_removed *ofr;
- struct ofpbuf *buf;
- long long int tdiff = now - rule->created;
+ long long int tdiff = time_msec() - rule->created;
uint32_t sec = tdiff / 1000;
uint32_t msec = tdiff - (sec * 1000);
+ struct ofp_flow_removed *ofr;
+ struct ofpbuf *buf;
ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
- flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
- &ofr->match);
- ofr->cookie = rule->flow_cookie;
- ofr->priority = htons(rule->cr.priority);
+ flow_to_match(&rule->cr.flow, p->tun_id_from_cookie, &ofr->match);
+ ofr->cookie = ofproto_rule_cast(rule)->flow_cookie;
+ ofr->priority = htons(rule->cr.flow.priority);
ofr->reason = reason;
ofr->duration_sec = htonl(sec);
ofr->duration_nsec = htonl(msec * 1000000);
ofr->idle_timeout = htons(rule->idle_timeout);
- ofr->packet_count = htonll(rule->packet_count);
- ofr->byte_count = htonll(rule->byte_count);
return buf;
}
-static void
-uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
-{
- assert(rule->installed);
- assert(!rule->cr.wc.wildcards);
-
- if (rule->super) {
- rule_remove(ofproto, rule);
- } else {
- rule_uninstall(ofproto, rule);
- }
-}
-
-static void
-send_flow_removed(struct ofproto *p, struct rule *rule,
- long long int now, uint8_t reason)
+static int
+delete_flow(struct ofproto *p, struct wdp_rule *rule, uint8_t reason)
{
- struct ofconn *ofconn;
- struct ofconn *prev;
- struct ofpbuf *buf = NULL;
-
/* We limit the maximum number of queued flow expirations it by accounting
* them under the counter for replies. That works because preventing
* OpenFlow requests from being processed also prevents new flows from
* being added (and expiring). (It also prevents processing OpenFlow
* requests that would not add new flows, so it is imperfect.) */
- prev = NULL;
- LIST_FOR_EACH (ofconn, node, &p->all_conns) {
- if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
- && ofconn_receives_async_msgs(ofconn)) {
- if (prev) {
- queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
- } else {
- buf = compose_flow_removed(p, rule, now, reason);
- }
- prev = ofconn;
- }
- }
- if (prev) {
- queue_tx(buf, prev, prev->reply_counter);
- }
-}
-
-
-static void
-expire_rule(struct cls_rule *cls_rule, void *p_)
-{
- struct ofproto *p = p_;
- struct rule *rule = rule_from_cls_rule(cls_rule);
- long long int hard_expire, idle_expire, expire, now;
-
- hard_expire = (rule->hard_timeout
- ? rule->created + rule->hard_timeout * 1000
- : LLONG_MAX);
- idle_expire = (rule->idle_timeout
- && (rule->super || list_is_empty(&rule->list))
- ? rule->used + rule->idle_timeout * 1000
- : LLONG_MAX);
- expire = MIN(hard_expire, idle_expire);
-
- now = time_msec();
- if (now < expire) {
- if (rule->installed && now >= rule->used + 5000) {
- uninstall_idle_flow(p, rule);
- } else if (!rule->cr.wc.wildcards) {
- active_timeout(p, rule);
- }
-
- return;
- }
-
- COVERAGE_INC(ofproto_expired);
+ struct ofproto_rule *ofproto_rule = ofproto_rule_cast(rule);
+ struct wdp_flow_stats stats;
+ struct ofpbuf *buf;
+ int error;
- /* Update stats. This code will be a no-op if the rule expired
- * due to an idle timeout. */
- if (rule->cr.wc.wildcards) {
- struct rule *subrule, *next;
- LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
- rule_remove(p, subrule);
- }
+ if (ofproto_rule->send_flow_removed) {
+ /* Compose most of the ofp_flow_removed before 'rule' is destroyed. */
+ buf = compose_flow_removed(p, rule, reason);
} else {
- rule_uninstall(p, rule);
- }
-
- if (!rule_is_hidden(rule)) {
- send_flow_removed(p, rule, now,
- (now >= hard_expire
- ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
- }
- rule_remove(p, rule);
-}
-
-static void
-active_timeout(struct ofproto *ofproto, struct rule *rule)
-{
- if (ofproto->netflow && !is_controller_rule(rule) &&
- netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
- struct ofexpired expired;
- struct odp_flow odp_flow;
-
- /* Get updated flow stats. */
- memset(&odp_flow, 0, sizeof odp_flow);
- if (rule->installed) {
- odp_flow.key = rule->cr.flow;
- odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
- dpif_flow_get(ofproto->dpif, &odp_flow);
-
- if (odp_flow.stats.n_packets) {
- update_time(ofproto, rule, &odp_flow.stats);
- netflow_flow_update_flags(&rule->nf_flow,
- odp_flow.stats.tcp_flags);
- }
- }
-
- expired.flow = rule->cr.flow;
- expired.packet_count = rule->packet_count +
- odp_flow.stats.n_packets;
- expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
- expired.used = rule->used;
-
- netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
-
- /* Schedule us to send the accumulated records once we have
- * collected all of them. */
- poll_immediate_wake();
+ buf = NULL;
}
-}
-
-static void
-update_used(struct ofproto *p)
-{
- struct odp_flow *flows;
- size_t n_flows;
- size_t i;
- int error;
- error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
+ error = wdp_flow_delete(p->wdp, rule, &stats);
if (error) {
- return;
+ return error;
}
- for (i = 0; i < n_flows; i++) {
- struct odp_flow *f = &flows[i];
- struct rule *rule;
+ if (buf) {
+ struct ofp_flow_removed *ofr;
+ struct ofconn *prev = NULL;
+ struct ofconn *ofconn;
- rule = rule_from_cls_rule(
- classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
- if (!rule || !rule->installed) {
- COVERAGE_INC(ofproto_unexpected_rule);
- dpif_flow_del(p->dpif, f);
- continue;
- }
+ /* Compose the parts of the ofp_flow_removed that require stats. */
+ ofr = buf->data;
+ ofr->packet_count = htonll(stats.n_packets);
+ ofr->byte_count = htonll(stats.n_bytes);
- LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
- update_time(p, rule, &f->stats);
- rule_account(p, rule, f->stats.n_bytes);
++ LIST_FOR_EACH (ofconn, node, &p->all_conns) {
+ if (rconn_is_connected(ofconn->rconn)) {
+ if (prev) {
+ queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
+ }
+ prev = ofconn;
+ }
+ }
+ if (prev) {
+ queue_tx(buf, prev, prev->reply_counter);
+ } else {
+ ofpbuf_delete(buf);
+ }
}
- free(flows);
+ free(ofproto_rule);
+
+ return 0;
}
/* pinsched callback for sending 'packet' on 'ofconn'. */
static void
-do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
+do_send_packet_in(struct wdp_packet *packet, void *ofconn_)
{
struct ofconn *ofconn = ofconn_;
- rconn_send_with_limit(ofconn->rconn, packet,
+ rconn_send_with_limit(ofconn->rconn, packet->payload,
ofconn->packet_in_counter, 100);
+ packet->payload = NULL;
+ wdp_packet_destroy(packet);
}
/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
* If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
* ownership is transferred to this function. */
static void
-schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
- bool clone)
+schedule_packet_in(struct ofconn *ofconn, struct wdp_packet *packet,
+ int max_len, bool clone)
{
struct ofproto *ofproto = ofconn->ofproto;
- struct ofp_packet_in *opi = packet->data;
- uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
+ struct ofp_packet_in *opi = packet->payload->data;
int send_len, trim_size;
uint32_t buffer_id;
} else {
struct ofpbuf payload;
payload.data = opi->data;
- payload.size = packet->size - offsetof(struct ofp_packet_in, data);
- buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
+ payload.size = (packet->payload->size
+ - offsetof(struct ofp_packet_in, data));
+ buffer_id = pktbuf_save(ofconn->pktbuf, &payload, packet->in_port);
}
/* Figure out how much of the packet to send. */
/* Adjust packet length and clone if necessary. */
trim_size = offsetof(struct ofp_packet_in, data) + send_len;
if (clone) {
- packet = ofpbuf_clone_data(packet->data, trim_size);
- opi = packet->data;
+ packet = wdp_packet_clone(packet, trim_size);
+ opi = packet->payload->data;
} else {
- packet->size = trim_size;
+ packet->payload->size = trim_size;
}
/* Update packet headers. */
opi->buffer_id = htonl(buffer_id);
- update_openflow_length(packet);
+ update_openflow_length(packet->payload);
/* Hand over to packet scheduler. It might immediately call into
* do_send_packet_in() or it might buffer it for a while (until a later
* call to pinsched_run()). */
- pinsched_send(ofconn->schedulers[opi->reason], in_port,
+ pinsched_send(ofconn->schedulers[opi->reason], packet->in_port,
packet, do_send_packet_in, ofconn);
}
-/* Replace struct odp_msg header in 'packet' by equivalent struct
- * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
- * returned by dpif_recv()).
+/* Converts 'packet->payload' to a struct ofp_packet_in. It must have
+ * sufficient headroom to do so (e.g. as returned by xfif_recv()).
*
* The conversion is not complete: the caller still needs to trim any unneeded
* payload off the end of the buffer, set the length in the OpenFlow header,
* Returns the maximum number of bytes of the packet that should be sent to
* the controller (INT_MAX if no limit). */
static int
-do_convert_to_packet_in(struct ofpbuf *packet)
+do_convert_to_packet_in(struct wdp_packet *packet)
{
- struct odp_msg *msg = packet->data;
+ uint16_t total_len = packet->payload->size;
struct ofp_packet_in *opi;
- uint8_t reason;
- uint16_t total_len;
- uint16_t in_port;
- int max_len;
-
- /* Extract relevant header fields */
- if (msg->type == _ODPL_ACTION_NR) {
- reason = OFPR_ACTION;
- max_len = msg->arg;
- } else {
- reason = OFPR_NO_MATCH;
- max_len = INT_MAX;
- }
- total_len = msg->length - sizeof *msg;
- in_port = odp_port_to_ofp_port(msg->port);
/* Repurpose packet buffer by overwriting header. */
- ofpbuf_pull(packet, sizeof(struct odp_msg));
- opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
+ opi = ofpbuf_push_zeros(packet->payload,
+ offsetof(struct ofp_packet_in, data));
opi->header.version = OFP_VERSION;
opi->header.type = OFPT_PACKET_IN;
opi->total_len = htons(total_len);
- opi->in_port = htons(in_port);
- opi->reason = reason;
-
- return max_len;
+ opi->in_port = htons(packet->in_port);
+ if (packet->channel == WDP_CHAN_MISS) {
+ opi->reason = OFPR_NO_MATCH;
+ return INT_MAX;
+ } else {
+ opi->reason = OFPR_ACTION;
+ return packet->send_len;
+ }
}
-/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
- * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
- * as necessary according to their individual configurations.
+/* Given 'packet' with channel WDP_CHAN_ACTION or WDP_CHAN_MISS, sends an
+ * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
+ * their individual configurations.
*
- * 'packet' must have sufficient headroom to convert it into a struct
+ * 'packet->payload' must have sufficient headroom to convert it into a struct
* ofp_packet_in (e.g. as returned by dpif_recv()).
*
* Takes ownership of 'packet'. */
static void
-send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
+send_packet_in(struct ofproto *ofproto, struct wdp_packet *packet)
{
struct ofconn *ofconn, *prev;
int max_len;
max_len = do_convert_to_packet_in(packet);
prev = NULL;
- LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
+ LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
if (ofconn_receives_async_msgs(ofconn)) {
if (prev) {
schedule_packet_in(prev, packet, max_len, true);
if (prev) {
schedule_packet_in(prev, packet, max_len, false);
} else {
- ofpbuf_delete(packet);
+ wdp_packet_destroy(packet);
}
}
static uint64_t
pick_datapath_id(const struct ofproto *ofproto)
{
- const struct ofport *port;
+ struct wdp_port port;
- port = get_port(ofproto, ODPP_LOCAL);
- if (port) {
+ if (!wdp_port_query_by_number(ofproto->wdp, OFPP_LOCAL, &port)) {
uint8_t ea[ETH_ADDR_LEN];
int error;
- error = netdev_get_etheraddr(port->netdev, ea);
+ error = netdev_get_etheraddr(port.netdev, ea);
if (!error) {
+ wdp_port_free(&port);
return eth_addr_to_uint64(ea);
}
VLOG_WARN("could not get MAC address for %s (%s)",
- netdev_get_name(port->netdev), strerror(error));
+ netdev_get_name(port.netdev), strerror(error));
+ wdp_port_free(&port);
}
+
return ofproto->fallback_dpid;
}
eth_addr_nicira_random(ea);
return eth_addr_to_uint64(ea);
}
-\f
-static bool
-default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
- struct odp_actions *actions, tag_type *tags,
- uint16_t *nf_output_iface, void *ofproto_)
-{
- struct ofproto *ofproto = ofproto_;
- int out_port;
-
- /* Drop frames for reserved multicast addresses. */
- if (eth_addr_is_reserved(flow->dl_dst)) {
- return true;
- }
-
- /* Learn source MAC (but don't try to learn from revalidation). */
- if (packet != NULL) {
- tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
- 0, flow->in_port,
- GRAT_ARP_LOCK_NONE);
- if (rev_tag) {
- /* The log messages here could actually be useful in debugging,
- * so keep the rate limit relatively high. */
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
- VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
- ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
- ofproto_revalidate(ofproto, rev_tag);
- }
- }
-
- /* Determine output port. */
- out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
- NULL);
- if (out_port < 0) {
- add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
- } else if (out_port != flow->in_port) {
- odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
- *nf_output_iface = out_port;
- } else {
- /* Drop. */
- }
-
- return true;
-}
-
-static const struct ofhooks default_ofhooks = {
- NULL,
- default_normal_ofhook_cb,
- NULL,
- NULL
-};
#include "coverage.h"
#include "ofp-util.h"
#include "ofpbuf.h"
+#include "openflow/openflow.h"
#include "timeval.h"
#include "util.h"
#include "vconn.h"
if (++p->cookie >= COOKIE_MAX) {
p->cookie = 0;
}
- p->buffer = ofpbuf_new(sizeof(struct ofp_packet_in) + buffer->size);
- ofpbuf_reserve(p->buffer, sizeof(struct ofp_packet_in));
+ p->buffer = ofpbuf_new_with_headroom(buffer->size,
+ sizeof(struct ofp_packet_in));
ofpbuf_put(p->buffer, buffer->data, buffer->size);
p->timeout = time_msec() + OVERWRITE_MSECS;
p->in_port = in_port;
* datapath port number on which the packet was received in '*in_port'. The
* caller becomes responsible for freeing the buffer. However, if 'id'
* identifies a "null" packet buffer (created with pktbuf_get_null()), stores
- * NULL in '*bufferp' and UINT16_max in '*in_port'.
+ * NULL in '*bufferp' and OFPP_NONE in '*in_port'.
*
* A returned packet will have at least sizeof(struct ofp_packet_in) bytes of
* headroom.
error = 0;
}
*bufferp = NULL;
- *in_port = UINT16_MAX;
+ *in_port = OFPP_NONE;
return error;
}
--- /dev/null
- struct port_array ports; /* Index is xflow port nr;
- * wdp_port->opp.port_no is OFP port nr. */
+/*
+ * Copyright (c) 2010 Nicira Networks.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+
+#include "wdp-xflow.h"
+
+#include <errno.h>
+#include <inttypes.h>
+
+#include "coverage.h"
+#include "dhcp.h"
+#include "mac-learning.h"
+#include "netdev.h"
+#include "netflow.h"
+#include "ofp-util.h"
+#include "ofpbuf.h"
+#include "ofproto.h"
+#include "openflow/nicira-ext.h"
+#include "openflow/openflow.h"
+#include "packets.h"
+#include "poll-loop.h"
+#include "port-array.h"
+#include "queue.h"
+#include "shash.h"
+#include "svec.h"
+#include "timeval.h"
+#include "util.h"
+#include "vconn.h"
+#include "wdp-provider.h"
+#include "xfif.h"
+#include "xflow-util.h"
+#include "vlog.h"
+#include "xtoxll.h"
+
+VLOG_DEFINE_THIS_MODULE(wdp_xflow)
+
+enum {
+ TABLEID_HASH = 0,
+ TABLEID_CLASSIFIER = 1
+};
+
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+\f
+/* Maximum numbers of rules. */
+#define WX_MAX_WILD 65536 /* Wildcarded rules. */
+#define WX_MAX_EXACT 1048576 /* Exact-match rules. */
+
++struct wx_port {
++ struct hmap_node hmap_node;
++ struct wdp_port wdp_port;
++ uint16_t xflow_port;
++};
++
+struct wx {
+ struct list list_node;
+ struct wdp wdp;
+ struct xfif *xfif;
+ struct classifier cls;
+ struct netdev_monitor *netdev_monitor;
- LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) {
++ struct hmap ports; /* Contains "struct wx_port"s. */
+ struct shash port_by_name;
+ long long int next_expiration;
+ int wdp_listen_mask;
+
+ /* Rules that might need to be revalidated. */
+ bool need_revalidate; /* Revalidate all subrules? */
+ bool revalidate_all; /* Revalidate all subrules and other rules? */
+ struct tag_set revalidate_set; /* Tag set of (sub)rules to revalidate. */
+
+ /* Hooks for ovs-vswitchd. */
+ const struct ofhooks *ofhooks;
+ void *aux;
+
+ /* Used by default ofhooks. */
+ struct mac_learning *ml;
+
+ /* List of "struct wdp_packets" queued for the controller by
+ * execute_xflow_actions(). */
+#define MAX_CTL_PACKETS 50
+ struct list ctl_packets;
+ int n_ctl_packets;
+};
+
+static const struct ofhooks default_ofhooks;
+
+static struct list all_wx = LIST_INITIALIZER(&all_wx);
+
+static int wx_port_init(struct wx *);
++static struct wx_port *wx_port_get(const struct wx *, uint16_t xflow_port);
+static void wx_port_process_change(struct wx *wx, int error, char *devname,
+ wdp_port_poll_cb_func *cb, void *aux);
+static void wx_port_refresh_groups(struct wx *);
+
+static void wx_purge_ctl_packets__(struct wx *);
+
+enum {
+ WX_GROUP_FLOOD = 0,
+ WX_GROUP_ALL = 1
+};
+
+static struct wx *
+wx_cast(const struct wdp *wdp)
+{
+ return CONTAINER_OF(wdp, struct wx, wdp);
+}
+
+static int
+wx_xlate_actions(struct wx *, const union ofp_action *, size_t n,
+ const flow_t *flow, const struct ofpbuf *packet,
+ tag_type *tags, struct xflow_actions *out,
+ bool *may_set_up_flow);
+\f
+struct wx_rule {
+ struct wdp_rule wr;
+
+ uint64_t packet_count; /* Number of packets received. */
+ uint64_t byte_count; /* Number of bytes received. */
+ uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
+ long long int used; /* Last-used time (0 if never used). */
+ tag_type tags; /* Tags (set only by hooks). */
+
+ /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
+ * exact-match rule (having cr.wc.wildcards of 0) generated from the
+ * wildcard rule 'super'. In this case, 'list' is an element of the
+ * super-rule's list.
+ *
+ * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
+ * a list of subrules. A super-rule with no wildcards (where
+ * cr.wc.wildcards is 0) will never have any subrules. */
+ struct wx_rule *super;
+ struct list list;
+
+ /* Datapath actions.
+ *
+ * A super-rule with wildcard fields never has xflow actions (since the
+ * datapath only supports exact-match flows). */
+ bool installed; /* Installed in datapath? */
+ bool may_install; /* True ordinarily; false if actions must
+ * be reassessed for every packet. */
+ int n_xflow_actions;
+ union xflow_action *xflow_actions;
+};
+
+static void wx_rule_destroy(struct wx *, struct wx_rule *);
+static void wx_rule_update_actions(struct wx *, struct wx_rule *);
+static void wx_rule_execute(struct wx *, struct wx_rule *,
+ struct ofpbuf *packet, const flow_t *);
+static bool wx_rule_make_actions(struct wx *, struct wx_rule *,
+ const struct ofpbuf *packet);
+static void wx_rule_install(struct wx *, struct wx_rule *,
+ struct wx_rule *displaced_rule);
+
+static struct wx_rule *
+wx_rule_cast(const struct cls_rule *cls_rule)
+{
+ return cls_rule ? CONTAINER_OF(cls_rule, struct wx_rule, wr.cr) : NULL;
+}
+
+/* Returns true if 'rule' is merely an implementation detail that should be
+ * hidden from the client. */
+static inline bool
+wx_rule_is_hidden(const struct wx_rule *rule)
+{
+ return rule->super != NULL;
+}
+
+static void
+wx_rule_free(struct wx_rule *rule)
+{
+ wdp_rule_uninit(&rule->wr);
+ free(rule->xflow_actions);
+ free(rule);
+}
+
+static void
+wx_rule_account(struct wx *wx OVS_UNUSED, struct wx_rule *rule OVS_UNUSED,
+ uint64_t extra_bytes OVS_UNUSED)
+{
+ /* XXX call account_cb hook */
+}
+
+static void
+wx_rule_post_uninstall(struct wx *wx, struct wx_rule *rule)
+{
+ struct wx_rule *super = rule->super;
+
+ wx_rule_account(wx, rule, 0);
+
+ /* XXX netflow expiration */
+
+ if (super) {
+ super->packet_count += rule->packet_count;
+ super->byte_count += rule->byte_count;
+
+ /* Reset counters to prevent double counting if the rule ever gets
+ * reinstalled. */
+ rule->packet_count = 0;
+ rule->byte_count = 0;
+ rule->accounted_bytes = 0;
+
+ //XXX netflow_flow_clear(&rule->nf_flow);
+ }
+}
+
+static long long int
+xflow_flow_stats_to_msec(const struct xflow_flow_stats *stats)
+{
+ return (stats->used_sec
+ ? stats->used_sec * 1000 + stats->used_nsec / 1000000
+ : 0);
+}
+
+static void
+wx_rule_update_time(struct wx *wx OVS_UNUSED, struct wx_rule *rule,
+ const struct xflow_flow_stats *stats)
+{
+ long long int used = xflow_flow_stats_to_msec(stats);
+ if (used > rule->used) {
+ rule->used = used;
+ if (rule->super && used > rule->super->used) {
+ rule->super->used = used;
+ }
+ //XXX netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
+ }
+}
+
+static void
+wx_rule_update_stats(struct wx *wx, struct wx_rule *rule,
+ const struct xflow_flow_stats *stats)
+{
+ if (stats->n_packets) {
+ wx_rule_update_time(wx, rule, stats);
+ rule->packet_count += stats->n_packets;
+ rule->byte_count += stats->n_bytes;
+ /* XXX netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); */
+ }
+}
+
+static void
+wx_rule_uninstall(struct wx *wx, struct wx_rule *rule)
+{
+ assert(!rule->wr.cr.flow.wildcards);
+ if (rule->installed) {
+ struct xflow_flow xflow_flow;
+
+ xflow_key_from_flow(&xflow_flow.key, &rule->wr.cr.flow);
+ xflow_flow.actions = NULL;
+ xflow_flow.n_actions = 0;
+ xflow_flow.flags = 0;
+ if (!xfif_flow_del(wx->xfif, &xflow_flow)) {
+ wx_rule_update_stats(wx, rule, &xflow_flow.stats);
+ }
+ rule->installed = false;
+
+ wx_rule_post_uninstall(wx, rule);
+ }
+}
+
+#if 0
+static bool
+is_controller_rule(struct wx_rule *rule)
+{
+ /* If the only action is send to the controller then don't report
+ * NetFlow expiration messages since it is just part of the control
+ * logic for the network and not real traffic. */
+
+ return (rule
+ && rule->super
+ && rule->super->n_actions == 1
+ && action_outputs_to_port(&rule->super->actions[0],
+ htons(OFPP_CONTROLLER)));
+}
+#endif
+
+static void
+wx_rule_remove(struct wx *wx, struct wx_rule *rule)
+{
+ if (rule->wr.cr.flow.wildcards) {
+ COVERAGE_INC(wx_del_wc_flow);
+ wx->need_revalidate = true;
+ } else {
+ wx_rule_uninstall(wx, rule);
+ }
+ classifier_remove(&wx->cls, &rule->wr.cr);
+ wx_rule_destroy(wx, rule);
+}
+
+static bool
+wx_rule_revalidate(struct wx *wx, struct wx_rule *rule)
+{
+ const flow_t *flow = &rule->wr.cr.flow;
+
+ COVERAGE_INC(wx_rule_revalidate);
+ if (rule->super) {
+ struct wx_rule *super;
+ super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow));
+ if (!super) {
+ wx_rule_remove(wx, rule);
+ return false;
+ } else if (super != rule->super) {
+ COVERAGE_INC(wx_revalidate_moved);
+ list_remove(&rule->list);
+ list_push_back(&super->list, &rule->list);
+ rule->super = super;
+ rule->wr.hard_timeout = super->wr.hard_timeout;
+ rule->wr.idle_timeout = super->wr.idle_timeout;
+ rule->wr.created = super->wr.created;
+ rule->used = 0;
+ }
+ }
+
+ wx_rule_update_actions(wx, rule);
+ return true;
+}
+
+/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
+ * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
+ * through all of its subrules and revalidates them, destroying any that no
+ * longer has a super-rule (which is probably all of them).
+ *
+ * Before calling this function, the caller must make have removed 'rule' from
+ * the classifier. If 'rule' is an exact-match rule, the caller is also
+ * responsible for ensuring that it has been uninstalled from the datapath. */
+static void
+wx_rule_destroy(struct wx *wx, struct wx_rule *rule)
+{
+ if (!rule->super) {
+ struct wx_rule *subrule, *next;
- const struct wdp_port *wdp_port = port_array_get(&ctx->wx->ports, port);
++ LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
+ wx_rule_revalidate(wx, subrule);
+ }
+ } else {
+ list_remove(&rule->list);
+ }
+ wx_rule_free(rule);
+}
+
+#if 0
+static bool
+wx_rule_has_out_port(const struct wx_rule *rule, uint16_t out_port)
+{
+ const union ofp_action *oa;
+ struct actions_iterator i;
+
+ if (out_port == htons(OFPP_NONE)) {
+ return true;
+ }
+ for (oa = actions_first(&i, rule->wr.actions,
+ rule->wr.n_actions);
+ oa;
+ oa = actions_next(&i)) {
+ if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
+ return true;
+ }
+ }
+ return false;
+}
+#endif
+
+/* Caller is responsible for initializing the 'cr' and ofp_table_id members of
+ * the returned rule. */
+static struct wx_rule *
+wx_rule_create(struct wx_rule *super,
+ const union ofp_action *actions, size_t n_actions,
+ uint16_t idle_timeout, uint16_t hard_timeout)
+{
+ struct wx_rule *rule = xzalloc(sizeof *rule);
+ wdp_rule_init(&rule->wr, actions, n_actions);
+ rule->wr.idle_timeout = idle_timeout;
+ rule->wr.hard_timeout = hard_timeout;
+ rule->used = rule->wr.created;
+ rule->super = super;
+ if (super) {
+ list_push_back(&super->list, &rule->list);
+ } else {
+ list_init(&rule->list);
+ }
+#if 0
+ netflow_flow_clear(&rule->nf_flow);
+ netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
+#endif
+
+ return rule;
+}
+
+/* Executes, within 'wx', the 'n_actions' actions in 'actions' on 'packet',
+ * which arrived on 'in_port'.
+ *
+ * Takes ownership of 'packet'. */
+static bool
+execute_xflow_actions(struct wx *wx, uint16_t in_port,
+ const union xflow_action *actions, size_t n_actions,
+ struct ofpbuf *packet)
+{
+ if (n_actions == 1 && actions[0].type == XFLOWAT_CONTROLLER
+ && wx->n_ctl_packets < MAX_CTL_PACKETS) {
+ /* As an optimization, avoid a round-trip from userspace to kernel to
+ * userspace. This also avoids possibly filling up kernel packet
+ * buffers along the way. */
+ struct wdp_packet *wdp_packet;
+
+ if (!(wx->wdp_listen_mask & WDP_CHAN_ACTION)) {
+ return true;
+ }
+
+ wdp_packet = xmalloc(sizeof *wdp_packet);
+ wdp_packet->channel = WDP_CHAN_ACTION;
+ wdp_packet->tun_id = 0;
+ wdp_packet->in_port = in_port;
+ wdp_packet->send_len = actions[0].controller.arg;
+ wdp_packet->payload = packet;
+
+ list_push_back(&wx->ctl_packets, &wdp_packet->list);
+
+ return true;
+ } else {
+ int error;
+
+ error = xfif_execute(wx->xfif, in_port, actions, n_actions, packet);
+ ofpbuf_delete(packet);
+ return !error;
+ }
+}
+
+/* Executes the actions indicated by 'rule' on 'packet', which is in flow
+ * 'flow' and is considered to have arrived on xflow port 'in_port'. 'packet'
+ * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
+ *
+ * The flow that 'packet' actually contains does not need to actually match
+ * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
+ * the packet and byte counters for 'rule' will be credited for the packet sent
+ * out whether or not the packet actually matches 'rule'.
+ *
+ * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
+ * the caller must already have accurately composed xflow actions for it given
+ * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
+ * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
+ * function will compose a set of xflow actions based on 'rule''s OpenFlow
+ * actions and apply them to 'packet'.
+ *
+ * Takes ownership of 'packet'. */
+static void
+wx_rule_execute(struct wx *wx, struct wx_rule *rule,
+ struct ofpbuf *packet, const flow_t *flow)
+{
+ const union xflow_action *actions;
+ struct xflow_flow_stats stats;
+ size_t n_actions;
+ struct xflow_actions a;
+
+ assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
+
+ /* Grab or compose the xflow actions.
+ *
+ * The special case for an exact-match 'rule' where 'flow' is not the
+ * rule's flow is important to avoid, e.g., sending a packet out its input
+ * port simply because the xflow actions were composed for the wrong
+ * scenario. */
+ if (rule->wr.cr.flow.wildcards
+ || !flow_equal_headers(flow, &rule->wr.cr.flow))
+ {
+ struct wx_rule *super = rule->super ? rule->super : rule;
+ if (wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions, flow,
+ packet, NULL, &a, NULL)) {
+ ofpbuf_delete(packet);
+ return;
+ }
+ actions = a.actions;
+ n_actions = a.n_actions;
+ } else {
+ actions = rule->xflow_actions;
+ n_actions = rule->n_xflow_actions;
+ }
+
+ /* Execute the xflow actions. */
+ flow_extract_stats(flow, packet, &stats);
+ if (!execute_xflow_actions(wx, flow->in_port,
+ actions, n_actions, packet)) {
+ wx_rule_update_stats(wx, rule, &stats);
+ rule->used = time_msec();
+ //XXX netflow_flow_update_time(wx->netflow, &rule->nf_flow, rule->used);
+ }
+}
+
+/* Inserts 'rule' into 'p''s flow table.
+ *
+ * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
+ * actions on it and credits the statistics for sending the packet to 'rule'.
+ * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
+ * headroom. */
+static void
+wx_rule_insert(struct wx *wx, struct wx_rule *rule, struct ofpbuf *packet,
+ uint16_t in_port)
+{
+ struct wx_rule *displaced_rule;
+
+ /* Insert the rule in the classifier. */
+ displaced_rule = wx_rule_cast(classifier_insert(&wx->cls, &rule->wr.cr));
+ if (!rule->wr.cr.flow.wildcards) {
+ wx_rule_make_actions(wx, rule, packet);
+ }
+
+ /* Send the packet and credit it to the rule. */
+ if (packet) {
+ flow_t flow;
+ flow_extract(packet, 0, in_port, &flow);
+ wx_rule_execute(wx, rule, packet, &flow);
+ }
+
+ /* Install the rule in the datapath only after sending the packet, to
+ * avoid packet reordering. */
+ if (rule->wr.cr.flow.wildcards) {
+ COVERAGE_INC(wx_add_wc_flow);
+ wx->need_revalidate = true;
+ } else {
+ wx_rule_install(wx, rule, displaced_rule);
+ }
+
+ /* Free the rule that was displaced, if any. */
+ if (displaced_rule) {
+ rule->wr.client_data = displaced_rule->wr.client_data;
+ wx_rule_destroy(wx, displaced_rule);
+ }
+}
+
+static struct wx_rule *
+wx_rule_create_subrule(struct wx *wx, struct wx_rule *rule, const flow_t *flow)
+{
+ struct wx_rule *subrule;
+
+ subrule = wx_rule_create(rule, NULL, 0,
+ rule->wr.idle_timeout,
+ rule->wr.hard_timeout);
+ /* Subrules aren't really in any OpenFlow table, so don't bother with
+ * subrule->wr.ofp_table_id. */
+ COVERAGE_INC(wx_subrule_create);
+ cls_rule_from_flow(flow, &subrule->wr.cr);
+ classifier_insert_exact(&wx->cls, &subrule->wr.cr);
+
+ return subrule;
+}
+
+/* Returns true if the actions changed, false otherwise. */
+static bool
+wx_rule_make_actions(struct wx *wx, struct wx_rule *rule,
+ const struct ofpbuf *packet)
+{
+ const struct wx_rule *super;
+ struct xflow_actions a;
+ size_t actions_len;
+
+ assert(!rule->wr.cr.flow.wildcards);
+
+ super = rule->super ? rule->super : rule;
+ wx_xlate_actions(wx, super->wr.actions, super->wr.n_actions,
+ &rule->wr.cr.flow, packet,
+ &rule->tags, &a, &rule->may_install);
+
+ actions_len = a.n_actions * sizeof *a.actions;
+ if (rule->n_xflow_actions != a.n_actions
+ || memcmp(rule->xflow_actions, a.actions, actions_len)) {
+ COVERAGE_INC(wx_xflow_unchanged);
+ free(rule->xflow_actions);
+ rule->n_xflow_actions = a.n_actions;
+ rule->xflow_actions = xmemdup(a.actions, actions_len);
+ return true;
+ } else {
+ return false;
+ }
+}
+
+static int
+do_put_flow(struct wx *wx, struct wx_rule *rule, int flags,
+ struct xflow_flow_put *put)
+{
+ memset(&put->flow.stats, 0, sizeof put->flow.stats);
+ xflow_key_from_flow(&put->flow.key, &rule->wr.cr.flow);
+ put->flow.actions = rule->xflow_actions;
+ put->flow.n_actions = rule->n_xflow_actions;
+ put->flow.flags = 0;
+ put->flags = flags;
+ return xfif_flow_put(wx->xfif, put);
+}
+
+static void
+wx_rule_install(struct wx *wx, struct wx_rule *rule, struct wx_rule *displaced_rule)
+{
+ assert(!rule->wr.cr.flow.wildcards);
+
+ if (rule->may_install) {
+ struct xflow_flow_put put;
+ if (!do_put_flow(wx, rule,
+ XFLOWPF_CREATE | XFLOWPF_MODIFY | XFLOWPF_ZERO_STATS,
+ &put)) {
+ rule->installed = true;
+ if (displaced_rule) {
+ wx_rule_update_stats(wx, displaced_rule, &put.flow.stats);
+ wx_rule_post_uninstall(wx, displaced_rule);
+ }
+ }
+ } else if (displaced_rule) {
+ wx_rule_uninstall(wx, displaced_rule);
+ }
+}
+
+static void
+wx_rule_reinstall(struct wx *wx, struct wx_rule *rule)
+{
+ if (rule->installed) {
+ struct xflow_flow_put put;
+ COVERAGE_INC(wx_dp_missed);
+ do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY, &put);
+ } else {
+ wx_rule_install(wx, rule, NULL);
+ }
+}
+
+static void
+wx_rule_update_actions(struct wx *wx, struct wx_rule *rule)
+{
+ bool actions_changed;
+#if 0
+ uint16_t new_out_iface, old_out_iface;
+
+ old_out_iface = rule->nf_flow.output_iface;
+#endif
+ actions_changed = wx_rule_make_actions(wx, rule, NULL);
+
+ if (rule->may_install) {
+ if (rule->installed) {
+ if (actions_changed) {
+ struct xflow_flow_put put;
+ do_put_flow(wx, rule, XFLOWPF_CREATE | XFLOWPF_MODIFY
+ | XFLOWPF_ZERO_STATS, &put);
+ wx_rule_update_stats(wx, rule, &put.flow.stats);
+#if 0
+ /* Temporarily set the old output iface so that NetFlow
+ * messages have the correct output interface for the old
+ * stats. */
+ new_out_iface = rule->nf_flow.output_iface;
+ rule->nf_flow.output_iface = old_out_iface;
+#endif
+ wx_rule_post_uninstall(wx, rule);
+ //rule->nf_flow.output_iface = new_out_iface;
+ }
+ } else {
+ wx_rule_install(wx, rule, NULL);
+ }
+ } else {
+ wx_rule_uninstall(wx, rule);
+ }
+}
+\f
+static void
+add_output_group_action(struct xflow_actions *actions, uint16_t group,
+ uint16_t *nf_output_iface)
+{
+ xflow_actions_add(actions, XFLOWAT_OUTPUT_GROUP)->output_group.group = group;
+
+ if (group == WX_GROUP_ALL || group == WX_GROUP_FLOOD) {
+ *nf_output_iface = NF_OUT_FLOOD;
+ }
+}
+
+static void
+add_controller_action(struct xflow_actions *actions, uint16_t max_len)
+{
+ union xflow_action *a = xflow_actions_add(actions, XFLOWAT_CONTROLLER);
+ a->controller.arg = max_len;
+}
+
+struct wx_xlate_ctx {
+ /* Input. */
+ flow_t flow; /* Flow to which these actions correspond. */
+ int recurse; /* Recursion level, via xlate_table_action. */
+ struct wx *wx;
+ const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
+ * null pointer if we are revalidating
+ * without a packet to refer to. */
+
+ /* Output. */
+ struct xflow_actions *out; /* Datapath actions. */
+ tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
+ bool may_set_up_flow; /* True ordinarily; false if the actions must
+ * be reassessed for every packet. */
+ uint16_t nf_output_iface; /* Output interface index for NetFlow. */
+};
+
+static void do_xlate_actions(const union ofp_action *in, size_t n_in,
+ struct wx_xlate_ctx *ctx);
+
+static void
+add_output_action(struct wx_xlate_ctx *ctx, uint16_t port)
+{
- if (wdp_port) {
- if (wdp_port->opp.config & OFPPC_NO_FWD) {
++ const struct wx_port *wx_port = wx_port_get(ctx->wx, port);
+
- const struct wdp_port *port;
++ if (wx_port) {
++ if (wx_port->wdp_port.opp.config & OFPPC_NO_FWD) {
+ /* Forwarding disabled on port. */
+ return;
+ }
+ } else {
+ /*
+ * We don't have an ofport record for this port, but it doesn't hurt to
+ * allow forwarding to it anyhow. Maybe such a port will appear later
+ * and we're pre-populating the flow table.
+ */
+ }
+
+ xflow_actions_add(ctx->out, XFLOWAT_OUTPUT)->output.port = port;
+ //ctx->nf_output_iface = port;
+}
+
+static struct wx_rule *
+wx_rule_lookup_valid(struct wx *wx, const flow_t *flow)
+{
+ struct wx_rule *rule = wx_rule_cast(classifier_lookup(&wx->cls, flow));
+
+ /* The rule we found might not be valid, since we could be in need of
+ * revalidation. If it is not valid, don't return it. */
+ if (rule
+ && rule->super
+ && wx->need_revalidate
+ && !wx_rule_revalidate(wx, rule)) {
+ COVERAGE_INC(wx_invalidated);
+ return NULL;
+ }
+
+ return rule;
+}
+
+static void
+xlate_table_action(struct wx_xlate_ctx *ctx, uint16_t in_port)
+{
+ if (!ctx->recurse) {
+ uint16_t old_in_port;
+ struct wx_rule *rule;
+
+ /* Look up a flow with 'in_port' as the input port. Then restore the
+ * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
+ * have surprising behavior). */
+ old_in_port = ctx->flow.in_port;
+ ctx->flow.in_port = in_port;
+ rule = wx_rule_lookup_valid(ctx->wx, &ctx->flow);
+ ctx->flow.in_port = old_in_port;
+
+ if (rule) {
+ if (rule->super) {
+ rule = rule->super;
+ }
+
+ ctx->recurse++;
+ do_xlate_actions(rule->wr.actions, rule->wr.n_actions, ctx);
+ ctx->recurse--;
+ }
+ }
+}
+
+static void
+xlate_output_action__(struct wx_xlate_ctx *ctx,
+ uint16_t port, uint16_t max_len)
+{
+ uint16_t xflow_port;
+ uint16_t prev_nf_output_iface = ctx->nf_output_iface;
+
+ ctx->nf_output_iface = NF_OUT_DROP;
+
+ switch (port) {
+ case OFPP_IN_PORT:
+ add_output_action(ctx, ctx->flow.in_port);
+ break;
+ case OFPP_TABLE:
+ xlate_table_action(ctx, ctx->flow.in_port);
+ break;
+ case OFPP_NORMAL:
+ if (!ctx->wx->ofhooks->normal_cb(&ctx->flow, ctx->packet,
+ ctx->out, ctx->tags,
+ &ctx->nf_output_iface,
+ ctx->wx->aux)) {
+ COVERAGE_INC(wx_uninstallable);
+ ctx->may_set_up_flow = false;
+ }
+ break;
+
+ case OFPP_FLOOD:
+ add_output_group_action(ctx->out, WX_GROUP_FLOOD,
+ &ctx->nf_output_iface);
+ break;
+ case OFPP_ALL:
+ add_output_group_action(ctx->out, WX_GROUP_ALL, &ctx->nf_output_iface);
+ break;
+ case OFPP_CONTROLLER:
+ add_controller_action(ctx->out, max_len);
+ break;
+ case OFPP_LOCAL:
+ add_output_action(ctx, XFLOWP_LOCAL);
+ break;
+ default:
+ xflow_port = ofp_port_to_xflow_port(port);
+ if (xflow_port != ctx->flow.in_port) {
+ add_output_action(ctx, xflow_port);
+ }
+ break;
+ }
+
+ if (prev_nf_output_iface == NF_OUT_FLOOD) {
+ ctx->nf_output_iface = NF_OUT_FLOOD;
+ } else if (ctx->nf_output_iface == NF_OUT_DROP) {
+ ctx->nf_output_iface = prev_nf_output_iface;
+ } else if (prev_nf_output_iface != NF_OUT_DROP &&
+ ctx->nf_output_iface != NF_OUT_FLOOD) {
+ ctx->nf_output_iface = NF_OUT_MULTI;
+ }
+}
+
+static void
+xlate_output_action(struct wx_xlate_ctx *ctx,
+ const struct ofp_action_output *oao)
+{
+ xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
+}
+
+/* If the final xflow action in 'ctx' is "pop priority", drop it, as an
+ * optimization, because we're going to add another action that sets the
+ * priority immediately after, or because there are no actions following the
+ * pop. */
+static void
+remove_pop_action(struct wx_xlate_ctx *ctx)
+{
+ size_t n = ctx->out->n_actions;
+ if (n > 0 && ctx->out->actions[n - 1].type == XFLOWAT_POP_PRIORITY) {
+ ctx->out->n_actions--;
+ }
+}
+
+static void
+xlate_enqueue_action(struct wx_xlate_ctx *ctx,
+ const struct ofp_action_enqueue *oae)
+{
+ uint16_t ofp_port, xflow_port;
+ uint32_t priority;
+ int error;
+
+ error = xfif_queue_to_priority(ctx->wx->xfif, ntohl(oae->queue_id),
+ &priority);
+ if (error) {
+ /* Fall back to ordinary output action. */
+ xlate_output_action__(ctx, ntohs(oae->port), 0);
+ return;
+ }
+
+ /* Figure out xflow output port. */
+ ofp_port = ntohs(oae->port);
+ if (ofp_port != OFPP_IN_PORT) {
+ xflow_port = ofp_port_to_xflow_port(ofp_port);
+ } else {
+ xflow_port = ctx->flow.in_port;
+ }
+
+ /* Add xflow actions. */
+ remove_pop_action(ctx);
+ xflow_actions_add(ctx->out, XFLOWAT_SET_PRIORITY)->priority.priority
+ = priority;
+ add_output_action(ctx, xflow_port);
+ xflow_actions_add(ctx->out, XFLOWAT_POP_PRIORITY);
+
+ /* Update NetFlow output port. */
+ if (ctx->nf_output_iface == NF_OUT_DROP) {
+ ctx->nf_output_iface = xflow_port;
+ } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
+ ctx->nf_output_iface = NF_OUT_MULTI;
+ }
+}
+
++static void
++xlate_set_queue_action(struct wx_xlate_ctx *ctx,
++ const struct nx_action_set_queue *nasq)
++{
++ uint32_t priority;
++ int error;
++
++ error = xfif_queue_to_priority(ctx->wx->xfif, ntohl(nasq->queue_id),
++ &priority);
++ if (error) {
++ /* Couldn't translate queue to a priority, so ignore. A warning
++ * has already been logged. */
++ return;
++ }
++
++ remove_pop_action(ctx);
++ xflow_actions_add(ctx->out, XFLOWAT_SET_PRIORITY)->priority.priority
++ = priority;
++}
++
+static void
+xlate_nicira_action(struct wx_xlate_ctx *ctx,
+ const struct nx_action_header *nah)
+{
+ const struct nx_action_resubmit *nar;
+ const struct nx_action_set_tunnel *nast;
++ const struct nx_action_set_queue *nasq;
+ union xflow_action *oa;
+ int subtype = ntohs(nah->subtype);
+
+ assert(nah->vendor == htonl(NX_VENDOR_ID));
+ switch (subtype) {
+ case NXAST_RESUBMIT:
+ nar = (const struct nx_action_resubmit *) nah;
+ xlate_table_action(ctx, ofp_port_to_xflow_port(ntohs(nar->in_port)));
+ break;
+
+ case NXAST_SET_TUNNEL:
+ nast = (const struct nx_action_set_tunnel *) nah;
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TUNNEL);
+ ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
+ break;
+
+ case NXAST_DROP_SPOOFED_ARP:
+ if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
+ xflow_actions_add(ctx->out, XFLOWAT_DROP_SPOOFED_ARP);
+ }
+ break;
+
++ case NXAST_SET_QUEUE:
++ nasq = (const struct nx_action_set_queue *) nah;
++ xlate_set_queue_action(ctx, nasq);
++ break;
++
++ case NXAST_POP_QUEUE:
++ xflow_actions_add(ctx->out, XFLOWAT_POP_PRIORITY);
++ break;
++
+ /* If you add a new action here that modifies flow data, don't forget to
+ * update the flow key in ctx->flow at the same time. */
+
+ default:
+ VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
+ break;
+ }
+}
+
+static void
+do_xlate_actions(const union ofp_action *in, size_t n_in,
+ struct wx_xlate_ctx *ctx)
+{
+ struct actions_iterator iter;
+ const union ofp_action *ia;
- port = port_array_get(&ctx->wx->ports, ctx->flow.in_port);
- if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
- port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
- ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
- /* Drop this flow. */
- return;
++ const struct wx_port *port;
+
- LIST_FOR_EACH_SAFE (subrule, next, struct wx_rule, list, &rule->list) {
++ port = wx_port_get(ctx->wx, ctx->flow.in_port);
++ if (port) {
++ const struct ofp_phy_port *opp = &port->wdp_port.opp;
++ if (opp->config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
++ opp->config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
++ ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
++ /* Drop this flow. */
++ return;
++ }
+ }
+
+ for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
+ uint16_t type = ntohs(ia->type);
+ union xflow_action *oa;
+
+ switch (type) {
+ case OFPAT_OUTPUT:
+ xlate_output_action(ctx, &ia->output);
+ break;
+
+ case OFPAT_SET_VLAN_VID:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
+ oa->dl_tci.tci = ia->vlan_vid.vlan_vid & htons(VLAN_VID_MASK);
+ oa->dl_tci.mask = htons(VLAN_VID_MASK);
+ ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
+ break;
+
+ case OFPAT_SET_VLAN_PCP:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_TCI);
+ oa->dl_tci.tci = htons((ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT)
+ & VLAN_PCP_MASK);
+ oa->dl_tci.mask = htons(VLAN_PCP_MASK);
+
+ if (ctx->flow.dl_vlan == htons(OFP_VLAN_NONE)) {
+ ctx->flow.dl_vlan = htons(0);
+ }
+ ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
+ break;
+
+ case OFPAT_STRIP_VLAN:
+ xflow_actions_add(ctx->out, XFLOWAT_STRIP_VLAN);
+ ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
+ ctx->flow.dl_vlan_pcp = 0;
+ break;
+
+ case OFPAT_SET_DL_SRC:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_SRC);
+ memcpy(oa->dl_addr.dl_addr,
+ ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
+ memcpy(ctx->flow.dl_src,
+ ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
+ break;
+
+ case OFPAT_SET_DL_DST:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_DL_DST);
+ memcpy(oa->dl_addr.dl_addr,
+ ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
+ memcpy(ctx->flow.dl_dst,
+ ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
+ break;
+
+ case OFPAT_SET_NW_SRC:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_SRC);
+ ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
+ break;
+
+ case OFPAT_SET_NW_DST:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_DST);
+ ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
+ break;
+
+ case OFPAT_SET_NW_TOS:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_NW_TOS);
+ ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
+ break;
+
+ case OFPAT_SET_TP_SRC:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_SRC);
+ ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
+ break;
+
+ case OFPAT_SET_TP_DST:
+ oa = xflow_actions_add(ctx->out, XFLOWAT_SET_TP_DST);
+ ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
+ break;
+
+ case OFPAT_ENQUEUE:
+ xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
+ break;
+
+ case OFPAT_VENDOR:
+ xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
+ break;
+
+ default:
+ VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
+ break;
+ }
+ }
+}
+
+/* Returns true if 'flow' and 'actions' may be set up as a flow in the kernel.
+ * This is true most of the time, but we don't allow flows that would prevent
+ * DHCP replies from being seen by the local port to be set up in the
+ * kernel.
+ *
+ * We only need this, strictly speaking, when in-band control is turned on. */
+static bool
+wx_may_set_up(const flow_t *flow, const struct xflow_actions *actions)
+{
+ if (flow->dl_type == htons(ETH_TYPE_IP)
+ && flow->nw_proto == IP_TYPE_UDP
+ && flow->tp_src == htons(DHCP_SERVER_PORT)
+ && flow->tp_dst == htons(DHCP_CLIENT_PORT)) {
+ int i;
+
+ for (i = 0; i < actions->n_actions; i++) {
+ const struct xflow_action_output *oao = &actions->actions[i].output;
+ if (oao->type == XFLOWAT_OUTPUT && oao->port == XFLOWP_LOCAL) {
+ return true;
+ }
+ }
+ return false;
+ }
+
+ return true;
+}
+
+static int
+wx_xlate_actions(struct wx *wx, const union ofp_action *in, size_t n_in,
+ const flow_t *flow, const struct ofpbuf *packet,
+ tag_type *tags, struct xflow_actions *out,
+ bool *may_set_up_flow)
+{
+ tag_type no_tags = 0;
+ struct wx_xlate_ctx ctx;
+ COVERAGE_INC(wx_ofp2xflow);
+ xflow_actions_init(out);
+ ctx.flow = *flow;
+ ctx.recurse = 0;
+ ctx.wx = wx;
+ ctx.packet = packet;
+ ctx.out = out;
+ ctx.tags = tags ? tags : &no_tags;
+ ctx.may_set_up_flow = true;
+ ctx.nf_output_iface = NF_OUT_DROP;
+ do_xlate_actions(in, n_in, &ctx);
+ remove_pop_action(&ctx);
+
+ if (may_set_up_flow) {
+ *may_set_up_flow = ctx.may_set_up_flow && wx_may_set_up(flow, out);
+ }
+#if 0
+ if (nf_output_iface) {
+ *nf_output_iface = ctx.nf_output_iface;
+ }
+#endif
+ if (xflow_actions_overflow(out)) {
+ COVERAGE_INC(xflow_overflow);
+ xflow_actions_init(out);
+ return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
+ }
+ return 0;
+}
+\f
+static void
+update_used(struct wx *wx)
+{
+ struct xflow_flow *flows;
+ size_t n_flows;
+ size_t i;
+ int error;
+
+ error = xfif_flow_list_all(wx->xfif, &flows, &n_flows);
+ if (error) {
+ return;
+ }
+
+ for (i = 0; i < n_flows; i++) {
+ struct xflow_flow *f = &flows[i];
+ struct wx_rule *rule;
+ flow_t flow;
+
+ xflow_key_to_flow(&f->key, &flow);
+ rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, &flow));
+ if (!rule || !rule->installed) {
+ COVERAGE_INC(wx_unexpected_rule);
+ xfif_flow_del(wx->xfif, f);
+ continue;
+ }
+
+ wx_rule_update_time(wx, rule, &f->stats);
+ wx_rule_account(wx, rule, f->stats.n_bytes);
+ }
+ free(flows);
+}
+
+static void
+uninstall_idle_flow(struct wx *wx, struct wx_rule *rule)
+{
+ assert(rule->installed);
+ assert(!rule->wr.cr.flow.wildcards);
+
+ if (rule->super) {
+ wx_rule_remove(wx, rule);
+ } else {
+ wx_rule_uninstall(wx, rule);
+ }
+}
+
+static int
+expire_rule(struct cls_rule *cls_rule, void *wx_)
+{
+ struct wx *wx = wx_;
+ struct wx_rule *rule = wx_rule_cast(cls_rule);
+ long long int hard_expire, idle_expire, expire, now;
+
+ hard_expire = (rule->wr.hard_timeout
+ ? rule->wr.created + rule->wr.hard_timeout * 1000
+ : LLONG_MAX);
+ idle_expire = (rule->wr.idle_timeout
+ && (rule->super || list_is_empty(&rule->list))
+ ? rule->used + rule->wr.idle_timeout * 1000
+ : LLONG_MAX);
+ expire = MIN(hard_expire, idle_expire);
+
+ now = time_msec();
+ if (now < expire) {
+ if (rule->installed && now >= rule->used + 5000) {
+ uninstall_idle_flow(wx, rule);
+ } else if (!rule->wr.cr.flow.wildcards) {
+ //XXX active_timeout(wx, rule);
+ }
+
+ return 0;
+ }
+
+ COVERAGE_INC(wx_expired);
+
+ /* Update stats. This code will be a no-op if the rule expired
+ * due to an idle timeout. */
+ if (rule->wr.cr.flow.wildcards) {
+ struct wx_rule *subrule, *next;
- LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
++ LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
+ wx_rule_remove(wx, subrule);
+ }
+ } else {
+ wx_rule_uninstall(wx, rule);
+ }
+
+#if 0 /* XXX */
+ if (!wx_rule_is_hidden(rule)) {
+ send_flow_removed(wx, rule, now,
+ (now >= hard_expire
+ ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
+ }
+#endif
+ wx_rule_remove(wx, rule);
+
+ return 0;
+}
+
+struct revalidate_cbdata {
+ struct wx *wx;
+ bool revalidate_all; /* Revalidate all exact-match rules? */
+ bool revalidate_subrules; /* Revalidate all exact-match subrules? */
+ struct tag_set revalidate_set; /* Set of tags to revalidate. */
+};
+
+static bool
+revalidate_rule(struct wx *wx, struct wx_rule *rule)
+{
+ const flow_t *flow = &rule->wr.cr.flow;
+
+ COVERAGE_INC(wx_revalidate_rule);
+ if (rule->super) {
+ struct wx_rule *super;
+ super = wx_rule_cast(classifier_lookup_wild(&wx->cls, flow));
+ if (!super) {
+ wx_rule_remove(wx, rule);
+ return false;
+ } else if (super != rule->super) {
+ COVERAGE_INC(wx_revalidate_moved);
+ list_remove(&rule->list);
+ list_push_back(&super->list, &rule->list);
+ rule->super = super;
+ rule->wr.hard_timeout = super->wr.hard_timeout;
+ rule->wr.idle_timeout = super->wr.idle_timeout;
+ rule->wr.created = super->wr.created;
+ rule->used = 0;
+ }
+ }
+
+ wx_rule_update_actions(wx, rule);
+ return true;
+}
+
+static int
+revalidate_cb(struct cls_rule *sub_, void *cbdata_)
+{
+ struct wx_rule *sub = wx_rule_cast(sub_);
+ struct revalidate_cbdata *cbdata = cbdata_;
+
+ if (cbdata->revalidate_all
+ || (cbdata->revalidate_subrules && sub->super)
+ || tag_set_intersects(&cbdata->revalidate_set, sub->tags)) {
+ revalidate_rule(cbdata->wx, sub);
+ }
+ return 0;
+}
+
+static void
+wx_run_one(struct wx *wx)
+{
+ if (time_msec() >= wx->next_expiration) {
+ COVERAGE_INC(wx_expiration);
+ wx->next_expiration = time_msec() + 1000;
+ update_used(wx);
+
+ classifier_for_each(&wx->cls, CLS_INC_ALL, expire_rule, wx);
+
+ /* XXX account_checkpoint_cb */
+ }
+
+ if (wx->need_revalidate || !tag_set_is_empty(&wx->revalidate_set)) {
+ struct revalidate_cbdata cbdata;
+ cbdata.wx = wx;
+ cbdata.revalidate_all = wx->revalidate_all;
+ cbdata.revalidate_subrules = wx->need_revalidate;
+ cbdata.revalidate_set = wx->revalidate_set;
+ tag_set_init(&wx->revalidate_set);
+ COVERAGE_INC(wx_revalidate);
+ classifier_for_each(&wx->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
+ wx->need_revalidate = false;
+ }
+}
+
+static void
+wx_run(void)
+{
+ struct wx *wx;
+
- LIST_FOR_EACH (wx, struct wx, list_node, &all_wx) {
++ LIST_FOR_EACH (wx, list_node, &all_wx) {
+ wx_run_one(wx);
+ }
+ xf_run();
+}
+
+static void
+wx_wait_one(struct wx *wx)
+{
+ if (wx->need_revalidate || !tag_set_is_empty(&wx->revalidate_set)) {
+ poll_immediate_wake();
+ } else if (wx->next_expiration != LLONG_MAX) {
+ poll_timer_wait_until(wx->next_expiration);
+ }
+}
+
+static void
+wx_wait(void)
+{
+ struct wx *wx;
+
- port_array_init(&wx->ports);
++ LIST_FOR_EACH (wx, list_node, &all_wx) {
+ wx_wait_one(wx);
+ }
+ xf_wait();
+}
+\f
+static int wx_flow_flush(struct wdp *);
+
+static int
+wx_enumerate(const struct wdp_class *wdp_class, struct svec *all_wdps)
+{
+ struct svec names = SVEC_EMPTY_INITIALIZER;
+ int error = xf_enumerate_names(wdp_class->type, &names);
+ svec_move(all_wdps, &names);
+ return error;
+}
+
+static int
+wx_open(const struct wdp_class *wdp_class, const char *name, bool create,
+ struct wdp **wdpp)
+{
+ struct xfif *xfif;
+ int error;
+
+ error = (create
+ ? xfif_create_and_open(name, wdp_class->type, &xfif)
+ : xfif_open(name, wdp_class->type, &xfif));
+ if (!error) {
+ struct wx *wx;
+
+ wx = xzalloc(sizeof *wx);
+ list_push_back(&all_wx, &wx->list_node);
+ wdp_init(&wx->wdp, wdp_class, name, 0, 0);
+ wx->xfif = xfif;
+ classifier_init(&wx->cls);
+ wx->netdev_monitor = netdev_monitor_create();
- unsigned int port_no;
- struct wdp_port *port;
++ hmap_init(&wx->ports);
+ shash_init(&wx->port_by_name);
+ wx->next_expiration = time_msec() + 1000;
+ tag_set_init(&wx->revalidate_set);
+
+ wx_port_init(wx);
+
+ wx->ofhooks = &default_ofhooks;
+ wx->aux = wx;
+ wx->ml = mac_learning_create();
+
+ list_init(&wx->ctl_packets);
+
+ *wdpp = &wx->wdp;
+ }
+
+ return error;
+}
+
+static void
+wx_close(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ wx_flow_flush(wdp);
+ xfif_close(wx->xfif);
+ classifier_destroy(&wx->cls);
+ netdev_monitor_destroy(wx->netdev_monitor);
+ list_remove(&wx->list_node);
+ mac_learning_destroy(wx->ml);
++ hmap_destroy(&wx->ports);
++ shash_destroy(&wx->port_by_name);
+ free(wx);
+}
+
+static int
+wx_get_all_names(const struct wdp *wdp, struct svec *all_names)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_get_all_names(wx->xfif, all_names);
+}
+
+static int
+wx_destroy(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_delete(wx->xfif);
+}
+
+static int
+wx_get_features(const struct wdp *wdp, struct ofpbuf **featuresp)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct ofp_switch_features *osf;
+ struct ofpbuf *buf;
- PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
- hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
++ struct wx_port *port;
+
+ buf = ofpbuf_new(sizeof *osf);
+ osf = ofpbuf_put_zeros(buf, sizeof *osf);
+ osf->n_tables = 2;
+ osf->capabilities = htonl(OFPC_ARP_MATCH_IP);
+ osf->actions = htonl((1u << OFPAT_OUTPUT) |
+ (1u << OFPAT_SET_VLAN_VID) |
+ (1u << OFPAT_SET_VLAN_PCP) |
+ (1u << OFPAT_STRIP_VLAN) |
+ (1u << OFPAT_SET_DL_SRC) |
+ (1u << OFPAT_SET_DL_DST) |
+ (1u << OFPAT_SET_NW_SRC) |
+ (1u << OFPAT_SET_NW_DST) |
+ (1u << OFPAT_SET_NW_TOS) |
+ (1u << OFPAT_SET_TP_SRC) |
+ (1u << OFPAT_SET_TP_DST) |
+ (1u << OFPAT_ENQUEUE));
+
- wx_answer_port_query(const struct wdp_port *port, struct wdp_port *portp)
++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) {
++ const struct ofp_phy_port *opp = &port->wdp_port.opp;
++ hton_ofp_phy_port(ofpbuf_put(buf, opp, sizeof *opp));
+ }
+
+ *featuresp = buf;
+ return 0;
+}
+
+static int
+count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
+{
+ struct wx_rule *rule = wx_rule_cast(cls_rule);
+ int *n_subrules = n_subrules_;
+
+ if (rule->super) {
+ (*n_subrules)++;
+ }
+ return 0;
+}
+
+static int
+wx_get_stats(const struct wdp *wdp, struct wdp_stats *stats)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct xflow_stats xflow_stats;
+ int error;
+
+ error = xfif_get_xf_stats(wx->xfif, &xflow_stats);
+ stats->max_ports = xflow_stats.max_ports;
+ return error;
+}
+
+static int
+wx_get_table_stats(const struct wdp *wdp, struct ofpbuf *stats)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct xflow_stats xflow_stats;
+ struct ofp_table_stats *exact, *wild;
+ int n_subrules;
+
+ xfif_get_xf_stats(wx->xfif, &xflow_stats);
+ /* XXX should pass up errors, but there are no appropriate OpenFlow error
+ * codes. */
+
+ n_subrules = 0;
+ classifier_for_each(&wx->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
+
+ exact = ofpbuf_put_zeros(stats, sizeof *exact);
+ exact->table_id = TABLEID_HASH;
+ strcpy(exact->name, "exact");
+ exact->wildcards = htonl(0);
+ exact->max_entries = htonl(MIN(WX_MAX_EXACT, xflow_stats.max_capacity));
+ exact->active_count = htonl(classifier_count_exact(&wx->cls) - n_subrules);
+ exact->lookup_count = htonll(xflow_stats.n_hit + xflow_stats.n_missed);
+ exact->matched_count = htonll(xflow_stats.n_hit);
+
+ wild = ofpbuf_put_zeros(stats, sizeof *exact);
+ wild->table_id = TABLEID_CLASSIFIER;
+ strcpy(wild->name, "classifier");
+ wild->wildcards = htonl(OVSFW_ALL);
+ wild->max_entries = htonl(WX_MAX_WILD);
+ wild->active_count = htonl(classifier_count_wild(&wx->cls));
+ wild->lookup_count = htonll(0); /* XXX */
+ wild->matched_count = htonll(0); /* XXX */
+
+ return 0;
+}
+
+static int
+wx_get_drop_frags(const struct wdp *wdp, bool *drop_frags)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_get_drop_frags(wx->xfif, drop_frags);
+}
+
+static int
+wx_set_drop_frags(struct wdp *wdp, bool drop_frags)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_set_drop_frags(wx->xfif, drop_frags);
+}
+
+static int
+wx_port_add(struct wdp *wdp, const char *devname,
+ bool internal, uint16_t *port_no)
+{
+ struct wx *wx = wx_cast(wdp);
+ uint16_t xflow_flags = internal ? XFLOW_PORT_INTERNAL : 0;
+ return xfif_port_add(wx->xfif, devname, xflow_flags, port_no);
+}
+
+static int
+wx_port_del(struct wdp *wdp, uint16_t port_no)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_port_del(wx->xfif, port_no);
+}
+
+static int
- wdp_port_copy(portp, port);
++wx_answer_port_query(const struct wx_port *port, struct wdp_port *portp)
+{
+ if (port) {
- const struct wdp_port *port;
++ wdp_port_copy(portp, &port->wdp_port);
+ return 0;
+ } else {
+ return ENOENT;
+ }
+}
+
+static int
+wx_port_query_by_number(const struct wdp *wdp, uint16_t port_no,
+ struct wdp_port *portp)
+{
+ struct wx *wx = wx_cast(wdp);
- port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
- return wx_answer_port_query(port, portp);
++ struct wx_port *wx_port = wx_port_get(wx, ofp_port_to_xflow_port(port_no));
+
- struct wdp_port *port;
++ return wx_answer_port_query(wx_port, portp);
+}
+
+static int
+wx_port_query_by_name(const struct wdp *wdp, const char *devname,
+ struct wdp_port *portp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return wx_answer_port_query(shash_find_data(&wx->port_by_name, devname),
+ portp);
+}
+
+static int
+wx_port_set_config(struct wdp *wdp, uint16_t port_no, uint32_t config)
+{
+ struct wx *wx = wx_cast(wdp);
- port = port_array_get(&wx->ports, ofp_port_to_xflow_port(port_no));
++ struct wx_port *port;
++ struct ofp_phy_port *opp;
+ uint32_t changes;
+
- changes = config ^ port->opp.config;
++ port = wx_port_get(wx, ofp_port_to_xflow_port(port_no));
+ if (!port) {
+ return ENOENT;
+ }
- error = netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
++ opp = &port->wdp_port.opp;
++ changes = config ^ opp->config;
+
+ if (changes & OFPPC_PORT_DOWN) {
++ struct netdev *netdev = port->wdp_port.netdev;
+ int error;
++
+ if (config & OFPPC_PORT_DOWN) {
- error = netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
++ error = netdev_turn_flags_off(netdev, NETDEV_UP, true);
+ } else {
- port->opp.config ^= OFPPC_PORT_DOWN;
++ error = netdev_turn_flags_on(netdev, NETDEV_UP, true);
+ }
+ if (!error) {
- port->opp.config ^= changes & REVALIDATE_BITS;
++ opp->config ^= OFPPC_PORT_DOWN;
+ }
+ }
+
+#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
+ if (changes & REVALIDATE_BITS) {
+ COVERAGE_INC(wx_costly_flags);
- port->opp.config ^= OFPPC_NO_FLOOD;
++ opp->config ^= changes & REVALIDATE_BITS;
+ wx->need_revalidate = true;
+ }
+#undef REVALIDATE_BITS
+
+ if (changes & OFPPC_NO_FLOOD) {
- port->opp.config ^= OFPPC_NO_PACKET_IN;
++ opp->config ^= OFPPC_NO_FLOOD;
+ wx_port_refresh_groups(wx);
+ }
+
+ if (changes & OFPPC_NO_PACKET_IN) {
- struct wdp_port *ports, *port;
- unsigned int port_no;
++ opp->config ^= OFPPC_NO_PACKET_IN;
+ }
+
+ return 0;
+}
+
+static int
+wx_port_list(const struct wdp *wdp, struct wdp_port **portsp, size_t *n_portsp)
+{
+ struct wx *wx = wx_cast(wdp);
- *n_portsp = n_ports = port_array_count(&wx->ports);
++ struct wdp_port *ports;
++ struct wx_port *port;
+ size_t n_ports, i;
+
- PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
- wdp_port_copy(&ports[i++], port);
++ *n_portsp = n_ports = hmap_count(&wx->ports);
+ *portsp = ports = xmalloc(n_ports * sizeof *ports);
+ i = 0;
- LIST_FOR_EACH (subrule, struct wx_rule, list, &rule->list) {
++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) {
++ wdp_port_copy(&ports[i++], &port->wdp_port);
+ }
+ assert(i == n_ports);
+
+ return 0;
+}
+
+static int
+wx_port_poll(struct wdp *wdp, wdp_port_poll_cb_func *cb, void *aux)
+{
+ struct wx *wx = wx_cast(wdp);
+ char *devname;
+ int retval;
+ int error;
+
+ retval = 0;
+ while ((error = xfif_port_poll(wx->xfif, &devname)) != EAGAIN) {
+ wx_port_process_change(wx, error, devname, cb, aux);
+ if (error && error != ENOBUFS) {
+ retval = error;
+ }
+ }
+ while ((error = netdev_monitor_poll(wx->netdev_monitor,
+ &devname)) != EAGAIN) {
+ wx_port_process_change(wx, error, devname, cb, aux);
+ if (error && error != ENOBUFS) {
+ retval = error;
+ }
+ }
+ return retval;
+}
+
+static int
+wx_port_poll_wait(const struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ xfif_port_poll_wait(wx->xfif);
+ netdev_monitor_poll_wait(wx->netdev_monitor);
+ return 0;
+}
+
+static struct wdp_rule *
+wx_flow_get(const struct wdp *wdp, const flow_t *flow, unsigned int include)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_rule *rule;
+ int table_id;
+
+ table_id = flow->wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
+ if (!(include & (1u << table_id))) {
+ return NULL;
+ }
+
+ rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, flow));
+ return rule && !wx_rule_is_hidden(rule) ? &rule->wr : NULL;
+}
+
+static struct wdp_rule *
+wx_flow_match(const struct wdp *wdp, const flow_t *flow)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_rule *rule;
+
+ rule = wx_rule_cast(classifier_lookup(&wx->cls, flow));
+ if (rule) {
+ if (wx_rule_is_hidden(rule)) {
+ rule = rule->super;
+ }
+ return &rule->wr;
+ } else {
+ return NULL;
+ }
+}
+
+struct wx_for_each_thunk_aux {
+ wdp_flow_cb_func *client_callback;
+ void *client_aux;
+};
+
+static int
+wx_for_each_thunk(struct cls_rule *cls_rule, void *aux_)
+{
+ struct wx_for_each_thunk_aux *aux = aux_;
+ struct wx_rule *rule = wx_rule_cast(cls_rule);
+
+ if (!wx_rule_is_hidden(rule)) {
+ return aux->client_callback(&rule->wr, aux->client_aux);
+ }
+ return 0;
+}
+
+static int
+wx_flow_for_each_match(const struct wdp *wdp, const flow_t *target,
+ unsigned int include,
+ wdp_flow_cb_func *client_callback, void *client_aux)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_for_each_thunk_aux aux;
+ int cls_include;
+
+ cls_include = 0;
+ if (include & (1u << TABLEID_HASH)) {
+ cls_include |= CLS_INC_EXACT;
+ }
+ if (include & (1u << TABLEID_CLASSIFIER)) {
+ cls_include |= CLS_INC_WILD;
+ }
+
+ aux.client_callback = client_callback;
+ aux.client_aux = client_aux;
+ return classifier_for_each_match(&wx->cls, target, cls_include,
+ wx_for_each_thunk, &aux);
+}
+
+/* Obtains statistic counters for 'rule' within 'wx' and stores them into
+ * '*stats'. If 'rule' is a wildcarded rule, the returned statistic include
+ * statistics for all of 'rule''s subrules. */
+static void
+query_stats(struct wx *wx, struct wx_rule *rule, struct wdp_flow_stats *stats)
+{
+ struct wx_rule *subrule;
+ struct xflow_flow *xflow_flows;
+ size_t n_xflow_flows;
+
+ /* Start from historical data for 'rule' itself that are no longer tracked
+ * by the datapath. This counts, for example, subrules that have
+ * expired. */
+ stats->n_packets = rule->packet_count;
+ stats->n_bytes = rule->byte_count;
+ stats->inserted = rule->wr.created;
+ stats->used = LLONG_MIN;
+ stats->tcp_flags = 0;
+ stats->ip_tos = 0;
+
+ /* Prepare to ask the datapath for statistics on 'rule', or if it is
+ * wildcarded then on all of its subrules.
+ *
+ * Also, add any statistics that are not tracked by the datapath for each
+ * subrule. This includes, for example, statistics for packets that were
+ * executed "by hand" by ofproto via xfif_execute() but must be accounted
+ * to a flow. */
+ n_xflow_flows = rule->wr.cr.flow.wildcards ? list_size(&rule->list) : 1;
+ xflow_flows = xzalloc(n_xflow_flows * sizeof *xflow_flows);
+ if (rule->wr.cr.flow.wildcards) {
+ size_t i = 0;
- const struct wdp_port *port = port_array_get(&wx->ports, XFLOWP_LOCAL);
- return port ? port->opp.hw_addr : NULL;
++ LIST_FOR_EACH (subrule, list, &rule->list) {
+ xflow_key_from_flow(&xflow_flows[i++].key, &subrule->wr.cr.flow);
+ stats->n_packets += subrule->packet_count;
+ stats->n_bytes += subrule->byte_count;
+ }
+ } else {
+ xflow_key_from_flow(&xflow_flows[0].key, &rule->wr.cr.flow);
+ }
+
+ /* Fetch up-to-date statistics from the datapath and add them in. */
+ if (!xfif_flow_get_multiple(wx->xfif, xflow_flows, n_xflow_flows)) {
+ size_t i;
+ for (i = 0; i < n_xflow_flows; i++) {
+ struct xflow_flow *xflow_flow = &xflow_flows[i];
+ long long int used;
+
+ stats->n_packets += xflow_flow->stats.n_packets;
+ stats->n_bytes += xflow_flow->stats.n_bytes;
+ used = xflow_flow_stats_to_msec(&xflow_flow->stats);
+ if (used > stats->used) {
+ stats->used = used;
+ }
+ stats->tcp_flags |= xflow_flow->stats.tcp_flags;
+ }
+ }
+ free(xflow_flows);
+}
+
+static int
+wx_flow_get_stats(const struct wdp *wdp,
+ const struct wdp_rule *wdp_rule,
+ struct wdp_flow_stats *stats)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
+
+ query_stats(wx, rule, stats);
+ return 0;
+}
+
+static bool
+wx_flow_overlaps(const struct wdp *wdp, const flow_t *flow)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ /* XXX overlap with a subrule? */
+ return classifier_rule_overlaps(&wx->cls, flow);
+}
+
+static int
+wx_flow_put(struct wdp *wdp, const struct wdp_flow_put *put,
+ struct wdp_flow_stats *old_stats, struct wdp_rule **rulep)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_rule *rule;
+ uint8_t ofp_table_id;
+
+ ofp_table_id = put->flow->wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
+ if (put->ofp_table_id != 0xff && put->ofp_table_id != ofp_table_id) {
+ return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
+ }
+
+ rule = wx_rule_cast(classifier_find_rule_exactly(&wx->cls, put->flow));
+ if (rule && wx_rule_is_hidden(rule)) {
+ rule = NULL;
+ }
+
+ if (rule) {
+ if (!(put->flags & WDP_PUT_MODIFY)) {
+ return EEXIST;
+ }
+ } else {
+ if (!(put->flags & WDP_PUT_CREATE)) {
+ return EINVAL;
+ }
+ if ((put->flow->wildcards
+ ? classifier_count_wild(&wx->cls) >= WX_MAX_WILD
+ : classifier_count_exact(&wx->cls) >= WX_MAX_EXACT)) {
+ /* XXX subrules should not count against exact-match limit */
+ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
+ }
+ }
+
+ rule = wx_rule_create(NULL, put->actions, put->n_actions,
+ put->idle_timeout, put->hard_timeout);
+ cls_rule_from_flow(put->flow, &rule->wr.cr);
+ rule->wr.ofp_table_id = ofp_table_id;
+ wx_rule_insert(wx, rule, NULL, 0);
+
+ if (old_stats) {
+ /* XXX */
+ memset(old_stats, 0, sizeof *old_stats);
+ }
+ if (rulep) {
+ *rulep = &rule->wr;
+ }
+
+ return 0;
+}
+
+static int
+wx_flow_delete(struct wdp *wdp, struct wdp_rule *wdp_rule,
+ struct wdp_flow_stats *final_stats)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
+
+ wx_rule_remove(wx, rule);
+ if (final_stats) {
+ memset(final_stats, 0, sizeof *final_stats); /* XXX */
+ }
+ return 0;
+}
+
+static int
+wx_flush_rule(struct cls_rule *cls_rule, void *wx_)
+{
+ struct wx_rule *rule = wx_rule_cast(cls_rule);
+ struct wx *wx = wx_;
+
+ /* Mark the flow as not installed, even though it might really be
+ * installed, so that wx_rule_remove() doesn't bother trying to uninstall
+ * it. There is no point in uninstalling it individually since we are
+ * about to blow away all the flows with xfif_flow_flush(). */
+ rule->installed = false;
+
+ wx_rule_remove(wx, rule);
+
+ return 0;
+}
+
+static int
+wx_flow_flush(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ COVERAGE_INC(wx_flow_flush);
+ classifier_for_each(&wx->cls, CLS_INC_ALL, wx_flush_rule, wx);
+ xfif_flow_flush(wx->xfif);
+ return 0;
+}
+
+static int
+wx_execute(struct wdp *wdp, uint16_t in_port,
+ const union ofp_action actions[], int n_actions,
+ const struct ofpbuf *packet)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct xflow_actions xflow_actions;
+ flow_t flow;
+ int error;
+
+ flow_extract((struct ofpbuf *) packet, 0, in_port, &flow);
+ error = wx_xlate_actions(wx, actions, n_actions, &flow, packet,
+ NULL, &xflow_actions, NULL);
+ if (error) {
+ return error;
+ }
+ return xfif_execute(wx->xfif, ofp_port_to_xflow_port(in_port),
+ xflow_actions.actions, xflow_actions.n_actions,
+ packet);
+}
+
+static int
+wx_flow_inject(struct wdp *wdp, struct wdp_rule *wdp_rule,
+ uint16_t in_port, const struct ofpbuf *packet)
+{
+ struct wx_rule *rule = wx_rule_cast(&wdp_rule->cr);
+ int error;
+
+ error = wx_execute(wdp, in_port, rule->wr.actions, rule->wr.n_actions,
+ packet);
+ if (!error) {
+ rule->packet_count++;
+ rule->byte_count += packet->size;
+ rule->used = time_msec();
+ }
+ return error;
+}
+
+static int
+wx_recv_get_mask(const struct wdp *wdp, int *listen_mask)
+{
+ struct wx *wx = wx_cast(wdp);
+ int xflow_listen_mask;
+ int error;
+
+ error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask);
+ if (!error) {
+ *listen_mask = 0;
+ if (xflow_listen_mask & XFLOWL_MISS) {
+ *listen_mask |= 1 << WDP_CHAN_MISS;
+ }
+ if (xflow_listen_mask & XFLOWL_ACTION) {
+ *listen_mask |= 1 << WDP_CHAN_ACTION;
+ }
+ if (xflow_listen_mask & XFLOWL_SFLOW) {
+ *listen_mask |= 1 << WDP_CHAN_SFLOW;
+ }
+ }
+ return error;
+}
+
+static int
+wx_recv_set_mask(struct wdp *wdp, int listen_mask)
+{
+ struct wx *wx = wx_cast(wdp);
+ int xflow_listen_mask;
+
+ wx->wdp_listen_mask = listen_mask;
+
+ xflow_listen_mask = 0;
+ if (listen_mask & (1 << WDP_CHAN_MISS)) {
+ xflow_listen_mask |= XFLOWL_MISS;
+ }
+ if (listen_mask & (1 << WDP_CHAN_ACTION)) {
+ xflow_listen_mask |= XFLOWL_ACTION;
+ } else {
+ wx_purge_ctl_packets__(wx);
+ }
+ if (listen_mask & (1 << WDP_CHAN_SFLOW)) {
+ xflow_listen_mask |= XFLOWL_SFLOW;
+ }
+
+ return xfif_recv_set_mask(wx->xfif, xflow_listen_mask);
+}
+
+static int
+wx_get_sflow_probability(const struct wdp *wdp, uint32_t *probability)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_get_sflow_probability(wx->xfif, probability);
+}
+
+static int
+wx_set_sflow_probability(struct wdp *wdp, uint32_t probability)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ return xfif_set_sflow_probability(wx->xfif, probability);
+}
+
+static int
+wx_translate_xflow_msg(struct xflow_msg *msg, struct ofpbuf *payload,
+ struct wdp_packet *packet)
+{
+ packet->in_port = xflow_port_to_ofp_port(msg->port);
+ packet->send_len = 0;
+ packet->tun_id = 0;
+
+ switch (msg->type) {
+ case _XFLOWL_MISS_NR:
+ packet->channel = WDP_CHAN_MISS;
+ packet->payload = payload;
+ packet->tun_id = msg->arg;
+ return 0;
+
+ case _XFLOWL_ACTION_NR:
+ packet->channel = WDP_CHAN_ACTION;
+ packet->payload = payload;
+ packet->send_len = msg->arg;
+ return 0;
+
+ case _XFLOWL_SFLOW_NR:
+ /* XXX */
+ ofpbuf_delete(payload);
+ return ENOSYS;
+
+ default:
+ VLOG_WARN_RL(&rl, "received XFLOW message of unexpected type %"PRIu32,
+ msg->type);
+ ofpbuf_delete(payload);
+ return ENOSYS;
+ }
+}
+
+static const uint8_t *
+get_local_mac(const struct wx *wx)
+{
- LIST_FOR_EACH_SAFE (this, next, struct wdp_packet, list,
- &wx->ctl_packets) {
++ const struct wx_port *port = wx_port_get(wx, XFLOWP_LOCAL);
++ return port ? port->wdp_port.opp.hw_addr : NULL;
+}
+
+/* Returns true if 'packet' is a DHCP reply to the local port. Such a reply
+ * should be sent to the local port regardless of the flow table.
+ *
+ * We only need this, strictly speaking, when in-band control is turned on. */
+static bool
+wx_is_local_dhcp_reply(const struct wx *wx,
+ const flow_t *flow, const struct ofpbuf *packet)
+{
+ if (flow->dl_type == htons(ETH_TYPE_IP)
+ && flow->nw_proto == IP_TYPE_UDP
+ && flow->tp_src == htons(DHCP_SERVER_PORT)
+ && flow->tp_dst == htons(DHCP_CLIENT_PORT)
+ && packet->l7)
+ {
+ const uint8_t *local_mac = get_local_mac(wx);
+ struct dhcp_header *dhcp = ofpbuf_at(
+ packet, (char *)packet->l7 - (char *)packet->data, sizeof *dhcp);
+ return dhcp && local_mac && eth_addr_equals(dhcp->chaddr, local_mac);
+ }
+
+ return false;
+}
+
+/* Determines whether 'payload' that arrived on 'in_port' is included in any of
+ * the flows in 'wx''s OpenFlow flow table. If so, then it adds a
+ * corresponding flow to the xfif's exact-match flow table, taking ownership of
+ * 'payload', and returns true. If not, it returns false and the caller
+ * retains ownership of 'payload'. */
+static bool
+wx_explode_rule(struct wx *wx, uint16_t in_port, struct ofpbuf *payload)
+{
+ struct wx_rule *rule;
+ flow_t flow;
+
+ flow_extract(payload, 0, xflow_port_to_ofp_port(in_port), &flow);
+
+ if (wx_is_local_dhcp_reply(wx, &flow, payload)) {
+ union xflow_action action;
+
+ memset(&action, 0, sizeof(action));
+ action.output.type = XFLOWAT_OUTPUT;
+ action.output.port = XFLOWP_LOCAL;
+ xfif_execute(wx->xfif, in_port, &action, 1, payload);
+ }
+
+ rule = wx_rule_lookup_valid(wx, &flow);
+ if (!rule) {
+ return false;
+ }
+
+ if (rule->wr.cr.flow.wildcards) {
+ rule = wx_rule_create_subrule(wx, rule, &flow);
+ wx_rule_make_actions(wx, rule, payload);
+ } else {
+ if (!rule->may_install) {
+ /* The rule is not installable, that is, we need to process every
+ * packet, so process the current packet and set its actions into
+ * 'subrule'. */
+ wx_rule_make_actions(wx, rule, payload);
+ } else {
+ /* XXX revalidate rule if it needs it */
+ }
+ }
+
+ wx_rule_execute(wx, rule, payload, &flow);
+ wx_rule_reinstall(wx, rule);
+
+ return true;
+}
+
+static int
+wx_recv(struct wdp *wdp, struct wdp_packet *packet)
+{
+ struct wx *wx = wx_cast(wdp);
+ int i;
+
+ if (wx->n_ctl_packets) {
+ struct wdp_packet *wdp_packet;
+
+ wdp_packet = CONTAINER_OF(list_pop_front(&wx->ctl_packets),
+ struct wdp_packet, list);
+ wx->n_ctl_packets--;
+
+ *packet = *wdp_packet;
+ free(wdp_packet);
+
+ return 0;
+ }
+
+ /* XXX need to avoid 50*50 potential cost for caller. */
+ for (i = 0; i < 50; i++) {
+ struct xflow_msg *msg;
+ struct ofpbuf *buf;
+ int error;
+
+ error = xfif_recv(wx->xfif, &buf);
+ if (error) {
+ return error;
+ }
+
+ msg = ofpbuf_pull(buf, sizeof *msg);
+ if (msg->type != _XFLOWL_MISS_NR
+ || !wx_explode_rule(wx, msg->port, buf)) {
+ return wx_translate_xflow_msg(msg, buf, packet);
+ }
+ }
+ return EAGAIN;
+}
+
+static void
+wx_recv_purge_queue__(struct wx *wx, int max, int xflow_listen_mask,
+ int *errorp)
+{
+ int error;
+
+ error = xfif_recv_set_mask(wx->xfif, xflow_listen_mask);
+ if (!error) {
+ struct ofpbuf *buf;
+
+ while (max > 0 && (error = xfif_recv(wx->xfif, &buf)) == 0) {
+ ofpbuf_delete(buf);
+ max--;
+ }
+ }
+ if (error && error != EAGAIN) {
+ *errorp = error;
+ }
+}
+
+static void
+wx_purge_ctl_packets__(struct wx *wx)
+{
+ struct wdp_packet *this, *next;
+
- struct wdp_port *port;
- unsigned int port_no;
++ LIST_FOR_EACH_SAFE (this, next, list, &wx->ctl_packets) {
+ list_remove(&this->list);
+ ofpbuf_delete(this->payload);
+ free(this);
+ }
+ wx->n_ctl_packets = 0;
+}
+
+static int
+wx_recv_purge(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+ struct xflow_stats xflow_stats;
+ int xflow_listen_mask;
+ int retval, error;
+
+ xfif_get_xf_stats(wx->xfif, &xflow_stats);
+
+ error = xfif_recv_get_mask(wx->xfif, &xflow_listen_mask);
+ if (error || !(xflow_listen_mask & XFLOWL_ALL)) {
+ return error;
+ }
+
+ if (xflow_listen_mask & XFLOWL_MISS) {
+ wx_recv_purge_queue__(wx, xflow_stats.max_miss_queue, XFLOWL_MISS,
+ &error);
+ }
+ if (xflow_listen_mask & XFLOWL_ACTION) {
+ wx_recv_purge_queue__(wx, xflow_stats.max_action_queue, XFLOWL_ACTION,
+ &error);
+ wx_purge_ctl_packets__(wx);
+ }
+ if (xflow_listen_mask & XFLOWL_SFLOW) {
+ wx_recv_purge_queue__(wx, xflow_stats.max_sflow_queue, XFLOWL_SFLOW,
+ &error);
+ }
+
+ retval = xfif_recv_set_mask(wx->xfif, xflow_listen_mask);
+ return retval ? retval : error;
+}
+
+
+static void
+wx_recv_wait(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ if (wx->n_ctl_packets) {
+ poll_immediate_wake();
+ } else {
+ xfif_recv_wait(wx->xfif);
+ }
+}
+
+static int
+wx_set_ofhooks(struct wdp *wdp, const struct ofhooks *ofhooks, void *aux)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ if (wx->ofhooks == &default_ofhooks) {
+ mac_learning_destroy(wx->ml);
+ wx->ml = NULL;
+ }
+
+ wx->ofhooks = ofhooks;
+ wx->aux = aux;
+ return 0;
+}
+
+static void
+wx_revalidate(struct wdp *wdp, tag_type tag)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ tag_set_add(&wx->revalidate_set, tag);
+}
+
+static void
+wx_revalidate_all(struct wdp *wdp)
+{
+ struct wx *wx = wx_cast(wdp);
+
+ wx->revalidate_all = true;
+}
+\f
+static void wx_port_update(struct wx *, const char *devname,
+ wdp_port_poll_cb_func *cb, void *aux);
+static void wx_port_reinit(struct wx *, wdp_port_poll_cb_func *cb, void *aux);
+
+static void
+wx_port_process_change(struct wx *wx, int error, char *devname,
+ wdp_port_poll_cb_func *cb, void *aux)
+{
+ if (error == ENOBUFS) {
+ wx_port_reinit(wx, cb, aux);
+ } else if (!error) {
+ wx_port_update(wx, devname, cb, aux);
+ free(devname);
+ }
+}
+
+static size_t
+wx_port_refresh_group(struct wx *wx, unsigned int group)
+{
+ uint16_t *ports;
+ size_t n_ports;
- ports = xmalloc(port_array_count(&wx->ports) * sizeof *ports);
++ struct wx_port *port;
+
+ assert(group == WX_GROUP_ALL || group == WX_GROUP_FLOOD);
+
- PORT_ARRAY_FOR_EACH (port, &wx->ports, port_no) {
- if (group == WX_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
- ports[n_ports++] = port_no;
++ ports = xmalloc(hmap_count(&wx->ports) * sizeof *ports);
+ n_ports = 0;
- struct wdp_port *wdp_port;
- unsigned int port_no;
++ HMAP_FOR_EACH (port, hmap_node, &wx->ports) {
++ const struct ofp_phy_port *opp = &port->wdp_port.opp;
++ if (group == WX_GROUP_ALL || !(opp->config & OFPPC_NO_FLOOD)) {
++ ports[n_ports++] = port->xflow_port;
+ }
+ }
+ xfif_port_group_set(wx->xfif, group, ports, n_ports);
+ free(ports);
+
+ return n_ports;
+}
+
+static void
+wx_port_refresh_groups(struct wx *wx)
+{
+ wx_port_refresh_group(wx, WX_GROUP_FLOOD);
+ wx_port_refresh_group(wx, WX_GROUP_ALL);
+}
+
+static void
+wx_port_reinit(struct wx *wx, wdp_port_poll_cb_func *cb, void *aux)
+{
+ struct svec devnames;
- PORT_ARRAY_FOR_EACH (wdp_port, &wx->ports, port_no) {
- svec_add (&devnames, (char *) wdp_port->opp.name);
++ struct wx_port *wx_port;
+ struct xflow_port *xflow_ports;
+ size_t n_xflow_ports;
+ size_t i;
+
+ svec_init(&devnames);
- static struct wdp_port *
- make_wdp_port(const struct xflow_port *xflow_port)
++ HMAP_FOR_EACH (wx_port, hmap_node, &wx->ports) {
++ svec_add (&devnames, (char *) wx_port->wdp_port.opp.name);
+ }
+ xfif_port_list(wx->xfif, &xflow_ports, &n_xflow_ports);
+ for (i = 0; i < n_xflow_ports; i++) {
+ svec_add(&devnames, xflow_ports[i].devname);
+ }
+ free(xflow_ports);
+
+ svec_sort_unique(&devnames);
+ for (i = 0; i < devnames.n; i++) {
+ wx_port_update(wx, devnames.names[i], cb, aux);
+ }
+ svec_destroy(&devnames);
+
+ wx_port_refresh_groups(wx);
+}
+
- wdp_port = xmalloc(sizeof *wdp_port);
++static struct wx_port *
++make_wx_port(const struct xflow_port *xflow_port)
+{
+ struct netdev_options netdev_options;
+ enum netdev_flags flags;
++ struct wx_port *wx_port;
+ struct wdp_port *wdp_port;
+ struct netdev *netdev;
+ bool carrier;
+ int error;
+
+ memset(&netdev_options, 0, sizeof netdev_options);
+ netdev_options.name = xflow_port->devname;
+ netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
+
+ error = netdev_open(&netdev_options, &netdev);
+ if (error) {
+ VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
+ "cannot be opened (%s)",
+ xflow_port->devname, xflow_port->port,
+ xflow_port->devname, strerror(error));
+ return NULL;
+ }
+
- return wdp_port;
++ wx_port = xmalloc(sizeof *wx_port);
++ wx_port->xflow_port = xflow_port->port;
++ wdp_port = &wx_port->wdp_port;
+ wdp_port->netdev = netdev;
+ wdp_port->opp.port_no = xflow_port_to_ofp_port(xflow_port->port);
+ netdev_get_etheraddr(netdev, wdp_port->opp.hw_addr);
+ strncpy((char *) wdp_port->opp.name, xflow_port->devname,
+ sizeof wdp_port->opp.name);
+ wdp_port->opp.name[sizeof wdp_port->opp.name - 1] = '\0';
+
+ netdev_get_flags(netdev, &flags);
+ wdp_port->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
+
+ netdev_get_carrier(netdev, &carrier);
+ wdp_port->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
+
+ netdev_get_features(netdev,
+ &wdp_port->opp.curr, &wdp_port->opp.advertised,
+ &wdp_port->opp.supported, &wdp_port->opp.peer);
+
+ wdp_port->devname = xstrdup(xflow_port->devname);
+ wdp_port->internal = (xflow_port->flags & XFLOW_PORT_INTERNAL) != 0;
- if (port_array_get(&wx->ports, xflow_port->port)) {
++ return wx_port;
+}
+
+static bool
+wx_port_conflicts(const struct wx *wx, const struct xflow_port *xflow_port)
+{
- wdp_port_equal(const struct wdp_port *a_, const struct wdp_port *b_)
++ if (wx_port_get(wx, xflow_port->port)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
+ xflow_port->port);
+ return true;
+ } else if (shash_find(&wx->port_by_name, xflow_port->devname)) {
+ VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
+ xflow_port->devname);
+ return true;
+ } else {
+ return false;
+ }
+}
+
+static int
- const struct ofp_phy_port *a = &a_->opp;
- const struct ofp_phy_port *b = &b_->opp;
++wx_port_equal(const struct wx_port *a_, const struct wx_port *b_)
+{
- wx_port_install(struct wx *wx, struct wdp_port *wdp_port)
++ const struct ofp_phy_port *a = &a_->wdp_port.opp;
++ const struct ofp_phy_port *b = &b_->wdp_port.opp;
+
+ BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
+ return (a->port_no == b->port_no
+ && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
+ && !strcmp((char *) a->name, (char *) b->name)
+ && a->state == b->state
+ && a->config == b->config
+ && a->curr == b->curr
+ && a->advertised == b->advertised
+ && a->supported == b->supported
+ && a->peer == b->peer);
+}
+
+static void
- uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
- const char *netdev_name = (const char *) wdp_port->opp.name;
++wx_port_install(struct wx *wx, struct wx_port *wx_port)
+{
- netdev_monitor_add(wx->netdev_monitor, wdp_port->netdev);
- port_array_set(&wx->ports, xflow_port, wdp_port);
- shash_add(&wx->port_by_name, netdev_name, wdp_port);
++ const struct ofp_phy_port *opp = &wx_port->wdp_port.opp;
++ uint16_t xflow_port = ofp_port_to_xflow_port(opp->port_no);
++ const char *name = (const char *) opp->name;
+
- wx_port_remove(struct wx *wx, struct wdp_port *wdp_port)
++ netdev_monitor_add(wx->netdev_monitor, wx_port->wdp_port.netdev);
++ hmap_insert(&wx->ports, &wx_port->hmap_node, hash_int(xflow_port, 0));
++ shash_add(&wx->port_by_name, name, wx_port);
+}
+
+static void
- uint16_t xflow_port = ofp_port_to_xflow_port(wdp_port->opp.port_no);
++wx_port_remove(struct wx *wx, struct wx_port *wx_port)
+{
- netdev_monitor_remove(wx->netdev_monitor, wdp_port->netdev);
- port_array_delete(&wx->ports, xflow_port);
- shash_delete(&wx->port_by_name,
- shash_find(&wx->port_by_name, (char *) wdp_port->opp.name));
++ const struct ofp_phy_port *opp = &wx_port->wdp_port.opp;
++ const char *name = (const char *) opp->name;
+
- wx_port_free(struct wdp_port *wdp_port)
++ netdev_monitor_remove(wx->netdev_monitor, wx_port->wdp_port.netdev);
++ hmap_remove(&wx->ports, &wx_port->hmap_node);
++ shash_delete(&wx->port_by_name, shash_find(&wx->port_by_name, name));
+}
+
+static void
- wdp_port_free(wdp_port);
- free(wdp_port);
++wx_port_free(struct wx_port *wx_port)
+{
- struct wdp_port *old_wdp_port;
- struct wdp_port *new_wdp_port;
++ if (wx_port) {
++ wdp_port_free(&wx_port->wdp_port);
++ free(wx_port);
++ }
+}
+
+static void
+wx_port_update(struct wx *wx, const char *devname,
+ wdp_port_poll_cb_func *cb, void *aux)
+{
+ struct xflow_port xflow_port;
- /* Find the old wdp_port. */
- old_wdp_port = shash_find_data(&wx->port_by_name, devname);
++ struct wx_port *old_wx_port;
++ struct wx_port *new_wx_port;
+ int error;
+
+ COVERAGE_INC(wx_update_port);
+
+ /* Query the datapath for port information. */
+ error = xfif_port_query_by_name(wx->xfif, devname, &xflow_port);
+
- if (!old_wdp_port) {
++ /* Find the old wx_port. */
++ old_wx_port = shash_find_data(&wx->port_by_name, devname);
+ if (!error) {
- old_wdp_port = port_array_get(&wx->ports, xflow_port.port);
++ if (!old_wx_port) {
+ /* There's no port named 'devname' but there might be a port with
+ * the same port number. This could happen if a port is deleted
+ * and then a new one added in its place very quickly, or if a port
+ * is renamed. In the former case we want to send an OFPPR_DELETE
+ * and an OFPPR_ADD, and in the latter case we want to send a
+ * single OFPPR_MODIFY. We can distinguish the cases by comparing
+ * the old port's ifindex against the new port, or perhaps less
+ * reliably but more portably by comparing the old port's MAC
+ * against the new port's MAC. However, this code isn't that smart
+ * and always sends an OFPPR_MODIFY (XXX). */
- /* Create a new wdp_port. */
- new_wdp_port = !error ? make_wdp_port(&xflow_port) : NULL;
++ old_wx_port = wx_port_get(wx, xflow_port.port);
+ }
+ } else if (error != ENOENT && error != ENODEV) {
+ VLOG_WARN_RL(&rl, "xfif_port_query_by_name returned unexpected error "
+ "%s", strerror(error));
+ return;
+ }
+
- if (!old_wdp_port && !new_wdp_port) {
++ /* Create a new wx_port. */
++ new_wx_port = !error ? make_wx_port(&xflow_port) : NULL;
+
+ /* Eliminate a few pathological cases. */
- } else if (old_wdp_port && new_wdp_port) {
++ if (!old_wx_port && !new_wx_port) {
+ return;
- * OpenFlow bits from old_wdp_port. (make_wdp_port() only sets
++ } else if (old_wx_port && new_wx_port) {
+ /* Most of the 'config' bits are OpenFlow soft state, but
+ * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
- new_wdp_port->opp.config |= old_wdp_port->opp.config & ~OFPPC_PORT_DOWN;
++ * OpenFlow bits from old_wx_port. (make_wx_port() only sets
+ * OFPPC_PORT_DOWN and leaves the other bits 0.) */
- if (wdp_port_equal(old_wdp_port, new_wdp_port)) {
++ struct ofp_phy_port *new_opp = &new_wx_port->wdp_port.opp;
++ struct ofp_phy_port *old_opp = &old_wx_port->wdp_port.opp;
++ new_opp->config |= old_opp->config & ~OFPPC_PORT_DOWN;
+
- wx_port_free(new_wdp_port);
++ if (wx_port_equal(old_wx_port, new_wx_port)) {
+ /* False alarm--no change. */
- if (old_wdp_port) {
- wx_port_remove(wx, old_wdp_port);
++ wx_port_free(new_wx_port);
+ return;
+ }
+ }
+
+ /* Now deal with the normal cases. */
- if (new_wdp_port) {
- wx_port_install(wx, new_wdp_port);
++ if (old_wx_port) {
++ wx_port_remove(wx, old_wx_port);
+ }
- if (!old_wdp_port) {
- (*cb)(&new_wdp_port->opp, OFPPR_ADD, aux);
- } else if (!new_wdp_port) {
- (*cb)(&old_wdp_port->opp, OFPPR_DELETE, aux);
++ if (new_wx_port) {
++ wx_port_install(wx, new_wx_port);
+ }
+
+ /* Call back. */
- (*cb)(&new_wdp_port->opp, OFPPR_MODIFY, aux);
++ if (!old_wx_port) {
++ (*cb)(&new_wx_port->wdp_port.opp, OFPPR_ADD, aux);
++ } else if (!new_wx_port) {
++ (*cb)(&old_wx_port->wdp_port.opp, OFPPR_DELETE, aux);
+ } else {
- wx_port_free(old_wdp_port);
++ (*cb)(&new_wx_port->wdp_port.opp, OFPPR_MODIFY, aux);
+ }
+
+ /* Update port groups. */
+ wx_port_refresh_groups(wx);
+
+ /* Clean up. */
- struct wdp_port *wdp_port = make_wdp_port(xflow_port);
- if (wdp_port) {
- wx_port_install(wx, wdp_port);
++ wx_port_free(old_wx_port);
+}
+
+static int
+wx_port_init(struct wx *wx)
+{
+ struct xflow_port *ports;
+ size_t n_ports;
+ size_t i;
+ int error;
+
+ error = xfif_port_list(wx->xfif, &ports, &n_ports);
+ if (error) {
+ return error;
+ }
+
+ for (i = 0; i < n_ports; i++) {
+ const struct xflow_port *xflow_port = &ports[i];
+ if (!wx_port_conflicts(wx, xflow_port)) {
++ struct wx_port *wx_port = make_wx_port(xflow_port);
++ if (wx_port) {
++ wx_port_install(wx, wx_port);
+ }
+ }
+ }
+ free(ports);
+ wx_port_refresh_groups(wx);
+ return 0;
+}
++
++/* Returns the port in 'wx' with xflow port number 'xflow_port'. */
++static struct wx_port *
++wx_port_get(const struct wx *wx, uint16_t xflow_port)
++{
++ struct wx_port *port;
++
++ HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_int(xflow_port, 0),
++ &wx->ports) {
++ if (port->xflow_port == xflow_port) {
++ return port;
++ }
++ }
++ return NULL;
++}
+\f
+void
+wdp_xflow_register(void)
+{
+ static const struct wdp_class wdp_xflow_class = {
+ NULL, /* name */
+ wx_run,
+ wx_wait,
+ wx_enumerate,
+ wx_open,
+ wx_close,
+ wx_get_all_names,
+ wx_destroy,
+ wx_get_features,
+ wx_get_stats,
+ wx_get_table_stats,
+ wx_get_drop_frags,
+ wx_set_drop_frags,
+ wx_port_add,
+ wx_port_del,
+ wx_port_query_by_number,
+ wx_port_query_by_name,
+ wx_port_list,
+ wx_port_set_config,
+ wx_port_poll,
+ wx_port_poll_wait,
+ wx_flow_get,
+ wx_flow_match,
+ wx_flow_for_each_match,
+ wx_flow_get_stats,
+ wx_flow_overlaps,
+ wx_flow_put,
+ wx_flow_delete,
+ wx_flow_flush,
+ wx_flow_inject,
+ wx_execute,
+ wx_recv_get_mask,
+ wx_recv_set_mask,
+ wx_get_sflow_probability,
+ wx_set_sflow_probability,
+ wx_recv,
+ wx_recv_purge,
+ wx_recv_wait,
+ wx_set_ofhooks,
+ wx_revalidate,
+ wx_revalidate_all,
+ };
+
+ static bool inited = false;
+
+ struct svec types;
+ const char *type;
+ bool registered;
+ int i;
+
+ if (inited) {
+ return;
+ }
+ inited = true;
+
+ svec_init(&types);
+ xf_enumerate_types(&types);
+
+ registered = false;
+ SVEC_FOR_EACH (i, type, &types) {
+ struct wdp_class *class;
+
+ class = xmalloc(sizeof *class);
+ *class = wdp_xflow_class;
+ class->type = xstrdup(type);
+ if (registered) {
+ class->run = NULL;
+ class->wait = NULL;
+ }
+ if (!wdp_register_provider(class)) {
+ registered = true;
+ }
+ }
+
+ svec_destroy(&types);
+}
+\f
+static bool
+default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
+ struct xflow_actions *actions, tag_type *tags,
+ uint16_t *nf_output_iface, void *wx_)
+{
+ struct wx *wx = wx_;
+ int out_port;
+
+ /* Drop frames for reserved multicast addresses. */
+ if (eth_addr_is_reserved(flow->dl_dst)) {
+ return true;
+ }
+
+ /* Learn source MAC (but don't try to learn from revalidation). */
+ if (packet != NULL) {
+ tag_type rev_tag = mac_learning_learn(wx->ml, flow->dl_src,
+ 0, flow->in_port,
+ GRAT_ARP_LOCK_NONE);
+ if (rev_tag) {
+ /* The log messages here could actually be useful in debugging,
+ * so keep the rate limit relatively high. */
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
+ VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
+ ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
+ tag_set_add(&wx->revalidate_set, rev_tag);
+ }
+ }
+
+ /* Determine output port. */
+ out_port = mac_learning_lookup_tag(wx->ml, flow->dl_dst, 0, tags,
+ NULL);
+ if (out_port < 0) {
+ add_output_group_action(actions, WX_GROUP_FLOOD, nf_output_iface);
+ } else if (out_port != flow->in_port) {
+ xflow_actions_add(actions, XFLOWAT_OUTPUT)->output.port = out_port;
+ *nf_output_iface = out_port;
+ } else {
+ /* Drop. */
+ }
+
+ return true;
+}
+
+static const struct ofhooks default_ofhooks = {
+ NULL,
+ default_normal_ofhook_cb,
+ NULL,
+ NULL
+};
tests/check-structs.at \
tests/daemon.at \
tests/daemon-py.at \
+ tests/ovs-ofctl.at \
tests/vconn.at \
tests/dir_name.at \
tests/aes128.at \
tests_test_aes128_LDADD = lib/libopenvswitch.a
noinst_PROGRAMS += tests/test-classifier
-tests_test_classifier_SOURCES = tests/test-classifier.c
+tests_test_classifier_SOURCES = \
+ tests/test-classifier.c \
+ tests/test-command-line.c \
+ tests/test-command-line.h
tests_test_classifier_LDADD = lib/libopenvswitch.a
noinst_PROGRAMS += tests/test-csum
#include "command-line.h"
#include "flow.h"
#include "packets.h"
+#include "test-command-line.h"
#undef NDEBUG
#include <assert.h>
n_exact = 0;
for (i = 0; i < tcls->n_rules; i++) {
- n_exact += tcls->rules[i]->cls_rule.wc.wildcards == 0;
+ n_exact += tcls->rules[i]->cls_rule.flow.wildcards == 0;
}
return n_exact;
}
return tcls->n_rules == 0;
}
+static unsigned int
+effective_priority(const flow_t *flow)
+{
+ return flow->wildcards ? flow->priority : MAX(flow->priority, UINT16_MAX);
+}
+
static struct test_rule *
tcls_insert(struct tcls *tcls, const struct test_rule *rule)
{
+ unsigned int priority = effective_priority(&rule->cls_rule.flow);
size_t i;
- assert(rule->cls_rule.wc.wildcards || rule->cls_rule.priority == UINT_MAX);
for (i = 0; i < tcls->n_rules; i++) {
const struct cls_rule *pos = &tcls->rules[i]->cls_rule;
- if (pos->priority == rule->cls_rule.priority
- && pos->wc.wildcards == rule->cls_rule.wc.wildcards
- && flow_equal(&pos->flow, &rule->cls_rule.flow)) {
- /* Exact match.
- * XXX flow_equal should ignore wildcarded fields */
+ if (pos->flow.priority == priority
+ && pos->flow.wildcards == rule->cls_rule.flow.wildcards
+ && flow_equal_headers(&pos->flow, &rule->cls_rule.flow)) {
+ /* Exact match. */
free(tcls->rules[i]);
tcls->rules[i] = xmemdup(rule, sizeof *rule);
return tcls->rules[i];
- } else if (pos->priority < rule->cls_rule.priority) {
+ } else if (pos->flow.priority < priority) {
break;
}
}
void *wild_field = (char *) &wild->flow + f->ofs;
void *fixed_field = (char *) fixed + f->ofs;
- if ((wild->wc.wildcards & f->wildcards) == f->wildcards ||
+ if ((wild->flow.wildcards & f->wildcards) == f->wildcards ||
!memcmp(wild_field, fixed_field, f->len)) {
/* Definite match. */
continue;
}
- if (wild->wc.wildcards & f->wildcards) {
+ if (wild->flow.wildcards & f->wildcards) {
uint32_t test = read_uint32(wild_field);
uint32_t ip = read_uint32(fixed_field);
int shift = (f_idx == CLS_F_IDX_NW_SRC
? OFPFW_NW_SRC_SHIFT : OFPFW_NW_DST_SHIFT);
- uint32_t mask = flow_nw_bits_to_mask(wild->wc.wildcards, shift);
+ uint32_t mask = flow_nw_bits_to_mask(wild->flow.wildcards, shift);
if (!((test ^ ip) & mask)) {
continue;
}
for (i = 0; i < cls->n_rules; i++) {
struct test_rule *pos = cls->rules[i];
- uint32_t wildcards = pos->cls_rule.wc.wildcards;
+ uint32_t wildcards = pos->cls_rule.flow.wildcards;
if (include & (wildcards ? CLS_INC_WILD : CLS_INC_EXACT)
&& match(&pos->cls_rule, flow)) {
return &pos->cls_rule;
for (i = 0; i < cls->n_rules; ) {
struct test_rule *pos = cls->rules[i];
- uint32_t wildcards = pos->cls_rule.wc.wildcards;
+ uint32_t wildcards = pos->cls_rule.flow.wildcards;
if (include & (wildcards ? CLS_INC_WILD : CLS_INC_EXACT)
&& match(target, &pos->cls_rule.flow)) {
tcls_remove(cls, pos);
ETH_ADDR_LEN);
flow.nw_proto = nw_proto_values[get_value(&x, N_NW_PROTO_VALUES)];
flow.nw_tos = nw_tos_values[get_value(&x, N_NW_TOS_VALUES)];
- memset(flow.reserved, 0, sizeof flow.reserved);
for (include = 1; include <= 3; include++) {
cr0 = lookup_with_include_bits(cls, &flow, include);
const struct test_rule *tr0 = test_rule_from_cls_rule(cr0);
const struct test_rule *tr1 = test_rule_from_cls_rule(cr1);
- assert(flow_equal(&cr0->flow, &cr1->flow));
- assert(cr0->wc.wildcards == cr1->wc.wildcards);
- assert(cr0->priority == cr1->priority);
- /* Skip nw_src_mask and nw_dst_mask, because they are derived
- * members whose values are used only for optimization. */
+ assert(flow_equal_headers(&cr0->flow, &cr1->flow));
+ assert(cr0->flow.wildcards == cr1->flow.wildcards);
+ assert(cr0->flow.priority == cr1->flow.priority);
+ /* Skip nw_src_mask, nw_dst_mask, and dl_tci_mask, because they
+ * are derived members used only for optimization. */
assert(tr0->aux == tr1->aux);
}
}
}
}
-static void
+static int
free_rule(struct cls_rule *cls_rule, void *cls)
{
classifier_remove(cls, cls_rule);
free(test_rule_from_cls_rule(cls_rule));
+ return 0;
}
static void
if (!hmap_is_empty(&cls->tables[i])) {
found_tables++;
}
- HMAP_FOR_EACH (bucket, struct cls_bucket, hmap_node, &cls->tables[i]) {
+ HMAP_FOR_EACH (bucket, hmap_node, &cls->tables[i]) {
found_buckets++;
assert(!list_is_empty(&bucket->rules));
found_rules += list_size(&bucket->rules);
{
const struct cls_field *f;
struct test_rule *rule;
- uint32_t wildcards;
flow_t flow;
- wildcards = 0;
memset(&flow, 0, sizeof flow);
+ flow.priority = priority;
for (f = &cls_fields[0]; f < &cls_fields[CLS_N_FIELDS]; f++) {
int f_idx = f - cls_fields;
if (wc_fields & (1u << f_idx)) {
- wildcards |= f->wildcards;
+ flow.wildcards |= f->wildcards;
} else {
int value_idx = (value_pat & (1u << f_idx)) != 0;
memcpy((char *) &flow + f->ofs, values[f_idx][value_idx], f->len);
}
rule = xzalloc(sizeof *rule);
- cls_rule_from_flow(&flow, wildcards, !wildcards ? UINT_MAX : priority,
- &rule->cls_rule);
+ cls_rule_from_flow(&flow, &rule->cls_rule);
return rule;
}
struct test_rule *rule = xmemdup(tcls.rules[rand() % tcls.n_rules],
sizeof(struct test_rule));
int include = rand() % 2 ? CLS_INC_WILD : CLS_INC_EXACT;
- include |= (rule->cls_rule.wc.wildcards
+ include |= (rule->cls_rule.flow.wildcards
? CLS_INC_WILD : CLS_INC_EXACT);
- classifier_for_each_match(&cls, &rule->cls_rule, include,
+ classifier_for_each_match(&cls, &rule->cls_rule.flow, include,
free_rule, &cls);
tcls_delete_matches(&tcls, &rule->cls_rule, include);
compare_classifiers(&cls, &tcls);
}
}
\f
-static const struct command commands[] = {
- {"empty", 0, 0, test_empty},
- {"destroy-null", 0, 0, test_destroy_null},
- {"single-rule", 0, 0, test_single_rule},
- {"rule-replacement", 0, 0, test_rule_replacement},
- {"two-rules-in-one-bucket", 0, 0, test_two_rules_in_one_bucket},
- {"two-rules-in-one-table", 0, 0, test_two_rules_in_one_table},
- {"two-rules-in-different-tables", 0, 0,
- test_two_rules_in_different_tables},
- {"many-rules-in-one-bucket", 0, 0, test_many_rules_in_one_bucket},
- {"many-rules-in-one-table", 0, 0, test_many_rules_in_one_table},
- {"many-rules-in-different-tables", 0, 0,
- test_many_rules_in_different_tables},
- {NULL, 0, 0, NULL},
-};
-
int
main(int argc, char *argv[])
{
+ static const struct command all_commands[] = {
+ { "empty", 0, 0, test_empty },
+ { "destroy-null", 0, 0, test_destroy_null },
+ { "single-rule", 0, 0, test_single_rule },
+ { "rule-replacement", 0, 0, test_rule_replacement },
+ { "two-rules-in-one-bucket", 0, 0, test_two_rules_in_one_bucket },
+ { "two-rules-in-one-table", 0, 0, test_two_rules_in_one_table },
+ { "two-rules-in-different-tables", 0, 0,
+ test_two_rules_in_different_tables },
+ { "many-rules-in-one-bucket", 0, 0,
+ test_many_rules_in_one_bucket },
+ { "many-rules-in-one-table", 0, 0, test_many_rules_in_one_table },
+ { "many-rules-in-different-tables", 0, 0,
+ test_many_rules_in_different_tables },
+ { NULL, 0, 0, NULL },
+ };
+
+ set_program_name(argv[0]);
init_values();
- run_command(argc - 1, argv + 1, commands);
+ parse_test_options(argc, argv, all_commands);
+ run_command(argc - 1, argv + 1, all_commands);
return 0;
}
#include "compiler.h"
#include "daemon.h"
#include "learning-switch.h"
+ #include "ofp-parse.h"
++#include "ofp-util.h"
#include "ofpbuf.h"
#include "openflow/openflow.h"
#include "poll-loop.h"
#include "rconn.h"
+ #include "shash.h"
#include "stream-ssl.h"
#include "timeval.h"
#include "unixctl.h"
struct rconn *rconn;
};
- /* Learn the ports on which MAC addresses appear? */
+ /* -H, --hub: Learn the ports on which MAC addresses appear? */
static bool learn_macs = true;
- /* Set up flows? (If not, every packet is processed at the controller.) */
+ /* -n, --noflow: Set up flows? (If not, every packet is processed at the
+ * controller.) */
static bool set_up_flows = true;
/* -N, --normal: Use "NORMAL" action instead of explicit port? */
* of their messages (for debugging fail-open mode). */
static bool mute = false;
- /* -q, --queue: OpenFlow queue to use, or the default queue if UINT32_MAX. */
- static uint32_t queue_id = UINT32_MAX;
+ /* -q, --queue: default OpenFlow queue, none if UINT32_MAX. */
+ static uint32_t default_queue = UINT32_MAX;
+
+ /* -Q, --port-queue: map from port name to port number (cast to void *). */
+ static struct shash port_queues = SHASH_INITIALIZER(&port_queues);
/* --with-flows: File with flows to send to switch, or null to not load
* any default flows. */
- static FILE *flow_file = NULL;
+ static struct ovs_queue default_flows = OVS_QUEUE_INITIALIZER;
/* --unixctl: Name of unixctl socket, or null to use the default. */
static char *unixctl_path = NULL;
for (i = optind; i < argc; i++) {
const char *name = argv[i];
struct vconn *vconn;
- int retval;
retval = vconn_open(name, OFP_VERSION, &vconn);
if (!retval) {
while (n_switches > 0 || n_listeners > 0) {
int iteration;
- int i;
/* Accept connections on listening vconns. */
for (i = 0; i < n_listeners && n_switches < MAX_SWITCHES; ) {
struct vconn *new_vconn;
- int retval;
retval = pvconn_accept(listeners[i], OFP_VERSION, &new_vconn);
if (!retval || retval == EAGAIN) {
bool progress = false;
for (i = 0; i < n_switches; ) {
struct switch_ *this = &switches[i];
- int retval = do_switching(this);
+
+ retval = do_switching(this);
if (!retval || retval == EAGAIN) {
if (!retval) {
progress = true;
static void
new_switch(struct switch_ *sw, struct vconn *vconn)
{
+ struct lswitch_config cfg;
+
sw->rconn = rconn_create(60, 0);
rconn_connect_unreliably(sw->rconn, vconn, NULL);
- /* If it was set, rewind 'flow_file' to the beginning, since a
- * previous call to lswitch_create() will leave the stream at the
- * end. */
- if (flow_file) {
- rewind(flow_file);
- }
- sw->lswitch = lswitch_create(sw->rconn, learn_macs, exact_flows,
- set_up_flows ? max_idle : -1,
- action_normal, flow_file);
-
- lswitch_set_queue(sw->lswitch, queue_id);
+ cfg.mode = (action_normal ? LSW_NORMAL
+ : learn_macs ? LSW_LEARN
+ : LSW_FLOOD);
+ cfg.max_idle = set_up_flows ? max_idle : -1;
+ cfg.default_flows = default_flows.head;
+ cfg.default_queue = default_queue;
+ cfg.port_queues = &port_queues;
+ sw->lswitch = lswitch_create(sw->rconn, &cfg);
}
static int
: EAGAIN);
}
- while ((b = parse_ofp_add_flow_file(stream)) != NULL) {
+ static void
+ read_flow_file(const char *name)
+ {
++ bool table_id_enabled = false;
++ uint8_t table_idx;
+ struct ofpbuf *b;
+ FILE *stream;
+
+ stream = fopen(optarg, "r");
+ if (!stream) {
+ ovs_fatal(errno, "%s: open", name);
+ }
+
++ while ((b = parse_ofp_add_flow_file(stream, &table_idx)) != NULL) {
++ if ((table_idx != 0xff) != table_id_enabled) {
++ table_id_enabled = table_idx != 0xff;
++ queue_push_tail(&default_flows,
++ make_nxt_flow_mod_table_id(table_id_enabled));
++ }
+ queue_push_tail(&default_flows, b);
+ }
+
+ fclose(stream);
+ }
+
+ static void
+ add_port_queue(char *s)
+ {
+ char *save_ptr = NULL;
+ char *port_name;
+ char *queue_id;
+
+ port_name = strtok_r(s, ":", &save_ptr);
+ queue_id = strtok_r(NULL, "", &save_ptr);
+ if (!queue_id) {
+ ovs_fatal(0, "argument to -Q or --port-queue should take the form "
+ "\"<port-name>:<queue-id>\"");
+ }
+
+ if (!shash_add_once(&port_queues, port_name,
+ (void *) (uintptr_t) atoi(queue_id))) {
+ ovs_fatal(0, "<port-name> arguments for -Q or --port-queue must "
+ "be unique");
+ }
+ }
+
static void
parse_options(int argc, char *argv[])
{
{"max-idle", required_argument, 0, OPT_MAX_IDLE},
{"mute", no_argument, 0, OPT_MUTE},
{"queue", required_argument, 0, 'q'},
+ {"port-queue", required_argument, 0, 'Q'},
{"with-flows", required_argument, 0, OPT_WITH_FLOWS},
{"unixctl", required_argument, 0, OPT_UNIXCTL},
{"help", no_argument, 0, 'h'},
break;
case 'q':
- queue_id = atoi(optarg);
+ default_queue = atoi(optarg);
+ break;
+
+ case 'Q':
+ add_port_queue(optarg);
break;
case OPT_WITH_FLOWS:
- flow_file = fopen(optarg, "r");
- if (flow_file == NULL) {
- ovs_fatal(errno, "%s: open", optarg);
- }
+ read_flow_file(optarg);
break;
case OPT_UNIXCTL:
}
}
free(short_options);
+
+ if (!shash_is_empty(&port_queues) || default_queue != UINT32_MAX) {
+ if (action_normal) {
+ ovs_error(0, "queue IDs are incompatible with -N or --normal; "
+ "not using OFPP_NORMAL");
+ action_normal = false;
+ }
+
+ if (!learn_macs) {
+ ovs_error(0, "queue IDs are incompatible with -H or --hub; "
+ "not acting as hub");
+ learn_macs = true;
+ }
+ }
}
static void
" -H, --hub act as hub instead of learning switch\n"
" -n, --noflow pass traffic, but don't add flows\n"
" --max-idle=SECS max idle time for new flows\n"
- " -N, --normal use OFPAT_NORMAL action\n"
+ " -N, --normal use OFPP_NORMAL action\n"
" -w, --wildcard use wildcards, not exact-match rules\n"
- " -q, --queue=QUEUE OpenFlow queue ID to use for output\n"
+ " -q, --queue=QUEUE-ID OpenFlow queue ID to use for output\n"
+ " -Q PORT-NAME:QUEUE-ID use QUEUE-ID for frames from PORT-NAME\n"
" --with-flows FILE use the flows from FILE\n"
" --unixctl=SOCKET override default control socket name\n"
" -h, --help display this help message\n"
tables. See \fBFlow Syntax\fR, below, for the syntax of \fIflows\fR.
The output format is descrbed in \fBTable Entry Output\fR.
.
+ .IP "\fBqueue\-stats \fIswitch \fR[\fIport \fR[\fIqueue\fR]]"
+ Prints to the console statistics for the specified \fIqueue\fR on
+ \fIport\fR within \fIswitch\fR. Either of \fIport\fR or \fIqueue\fR
+ or both may be omitted (or equivalently specified as \fBALL\fR). If
+ both are omitted, statistics are printed for all queues on all ports.
+ If only \fIqueue\fR is omitted, then statistics are printed for all
+ queues on \fIport\fR; if only \fIport\fR is omitted, then statistics
+ are printed for \fIqueue\fR on every port where it exists.
+ .
.TP
\fBadd\-flow \fIswitch flow\fR
Add the flow entry as described by \fIflow\fR to the \fIswitch\fR's
When \fBdl_type\fR and \fBnw_proto\fR take other values, the values of
these settings are ignored (see \fBFlow Syntax\fR above).
.
+.IP \fBtable=\fInumber\fR
+If specified, limits the flow manipulation and flow dump commands to
+only apply to the table with the given \fInumber\fR.
+\fInumber\fR is a number between 0 and 31, inclusive.
+.
+If this field is not specified, for \fBadd-flow\fR, \fBadd-flows\fR,
+\fBmod-flows\fR and \fBdel-flows\fR commands, the switch will choose
+the table for these commands to operate on; for \fBdump-flows\fR and
+\fBdump-aggregate\fR commands, statistics are gathered about flows
+from all tables.
+.IP
+When this field is specified in \fBadd-flow\fR, \fBadd-flows\fR,
+\fBmod-flows\fR and \fBdel-flows\fR commands, it will activate a
+Nicira extension to OpenFlow, which as of this writing is only
+known to be implemented by Open vSwitch.
+.
.PP
The following shorthand notations are also available:
.
If outputting to a port that encapsulates the packet in a tunnel and supports
an identifier (such as GRE), sets the identifier to \fBid\fR.
.
+ .IP \fBdrop_spoofed_arp\fR
+ Stops processing further actions, if the packet being processed is an
+ Ethernet+IPv4 ARP packet for which the source Ethernet address inside
+ the ARP packet differs from the source Ethernet address in the
+ Ethernet header.
+ .
+ This is useful because OpenFlow does not provide a way to match on the
+ Ethernet addresses inside ARP packets, so there is no other way to
+ drop spoofed ARPs other than sending every ARP packet to a controller.
+ .
+ .IP \fBset_queue\fB:\fIqueue\fR
+ Sets the queue that should be used to \fIqueue\fR when packets are
+ output. The number of supported queues depends on the switch; some
+ OpenFlow implementations do not support queuing at all.
+ .
+ .IP \fBpop_queue\fR
+ Restores the queue to the value it was before any \fBset_queue\fR
+ actions were applied.
+ .
.RE
.
.IP
.TP
\fBidle_timeout=\fIseconds\fR
Causes the flow to expire after the given number of seconds of
- inactivity. A value of 0 prevents a flow from expiring due to
- inactivity. The default is 60 seconds.
+ inactivity. A value of 0 (the default) prevents a flow from expiring due to
+ inactivity.
.
.IP \fBhard_timeout=\fIseconds\fR
Causes the flow to expire after the given number of seconds,
\fBout_port=\fIport\fR
If set, a matching flow must include an output action to \fIport\fR.
.
-.PP
-The \fBdump\-flows\fR and \fBdump\-aggregate\fR commands support an
-additional optional field:
-.
-.IP \fBtable=\fInumber\fR
-If specified, limits the flows about which statistics are gathered to
-those in the table with the given \fInumber\fR. Tables are numbered
-as shown by the \fBdump\-tables\fR command.
-.
-If this field is not specified, or if \fInumber\fR is given as
-\fB255\fR, statistics are gathered about flows from all tables.
.
.SS "Table Entry Output"
.
#include "command-line.h"
#include "compiler.h"
#include "dirs.h"
-#include "dpif.h"
+#include "dynamic-string.h"
+#include "netdev.h"
#include "netlink.h"
-#include "odp-util.h"
#include "ofp-parse.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "util.h"
#include "vconn.h"
#include "vlog.h"
+#include "xfif.h"
+#include "xflow-util.h"
#include "xtoxll.h"
VLOG_DEFINE_THIS_MODULE(ofctl)
" dump-flows SWITCH FLOW print matching FLOWs\n"
" dump-aggregate SWITCH print aggregate flow statistics\n"
" dump-aggregate SWITCH FLOW print aggregate stats for FLOWs\n"
+ " queue-stats SWITCH [PORT [QUEUE]] dump queue stats\n"
" add-flow SWITCH FLOW add flow described by FLOW\n"
" add-flows SWITCH FILE add flows from FILE\n"
" mod-flows SWITCH FLOW modify actions of matching FLOWs\n"
open_vconn__(const char *name, const char *default_suffix,
struct vconn **vconnp)
{
- struct dpif *dpif;
+ struct xfif *xfif;
struct stat s;
char *bridge_path, *datapath_name, *datapath_type;
bridge_path = xasprintf("%s/%s.%s", ovs_rundir, name, default_suffix);
- dp_parse_name(name, &datapath_name, &datapath_type);
+ xf_parse_name(name, &datapath_name, &datapath_type);
if (strstr(name, ":")) {
run(vconn_open_block(name, OFP_VERSION, vconnp),
open_vconn_socket(name, vconnp);
} else if (!stat(bridge_path, &s) && S_ISSOCK(s.st_mode)) {
open_vconn_socket(bridge_path, vconnp);
- } else if (!dpif_open(datapath_name, datapath_type, &dpif)) {
- char dpif_name[IF_NAMESIZE + 1];
+ } else if (!xfif_open(datapath_name, datapath_type, &xfif)) {
+ char xfif_name[IF_NAMESIZE + 1];
char *socket_name;
- run(dpif_port_get_name(dpif, ODPP_LOCAL, dpif_name, sizeof dpif_name),
- "obtaining name of %s", dpif_name);
- dpif_close(dpif);
- if (strcmp(dpif_name, name)) {
- VLOG_INFO("datapath %s is named %s", name, dpif_name);
+ run(xfif_port_get_name(xfif, XFLOWP_LOCAL, xfif_name, sizeof xfif_name),
+ "obtaining name of %s", xfif_name);
+ xfif_close(xfif);
+ if (strcmp(xfif_name, name)) {
+ VLOG_INFO("datapath %s is named %s", name, xfif_name);
}
socket_name = xasprintf("%s/%s.%s",
- ovs_rundir, dpif_name, default_suffix);
+ ovs_rundir, xfif_name, default_suffix);
if (stat(socket_name, &s)) {
ovs_fatal(errno, "cannot connect to %s: stat failed on %s",
name, socket_name);
dump_stats_transaction(argv[1], request);
}
- enable_flow_mod_table_id_ext(struct vconn *vconn, uint8_t enable)
+static void
- struct nxt_flow_mod_table_id *flow_mod_table_id;
- struct ofpbuf *buffer;
++enable_flow_mod_table_id_ext(struct vconn *vconn, bool enable)
+{
- flow_mod_table_id = make_openflow(sizeof *flow_mod_table_id, OFPT_VENDOR, &buffer);
++ send_openflow_buffer(vconn, make_nxt_flow_mod_table_id(enable));
++}
+
+ static void
+ do_queue_stats(int argc, char *argv[])
+ {
+ struct ofp_queue_stats_request *req;
+ struct ofpbuf *request;
- flow_mod_table_id->vendor = htonl(NX_VENDOR_ID);
- flow_mod_table_id->subtype = htonl(NXT_FLOW_MOD_TABLE_ID);
- flow_mod_table_id->set = enable;
+ req = alloc_stats_request(sizeof *req, OFPST_QUEUE, &request);
- send_openflow_buffer(vconn, buffer);
+ if (argc > 2 && argv[2][0] && strcasecmp(argv[2], "all")) {
+ req->port_no = htons(str_to_port_no(argv[1], argv[2]));
+ } else {
+ req->port_no = htons(OFPP_ALL);
+ }
+ if (argc > 3 && argv[3][0] && strcasecmp(argv[3], "all")) {
+ req->queue_id = htonl(atoi(argv[3]));
+ } else {
+ req->queue_id = htonl(OFPQ_ALL);
+ }
+
+ memset(req->pad, 0, sizeof req->pad);
+
+ dump_stats_transaction(argv[1], request);
}
static void
uint16_t priority, idle_timeout, hard_timeout;
uint64_t cookie;
struct ofp_match match;
+ uint8_t table_idx;
/* Parse and send. parse_ofp_str() will expand and reallocate the
* data in 'buffer', so we can't keep pointers to across the
* parse_ofp_str() call. */
make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer);
parse_ofp_str(argv[2], &match, buffer,
- NULL, NULL, &priority, &idle_timeout, &hard_timeout,
+ &table_idx, NULL, &priority, &idle_timeout, &hard_timeout,
&cookie);
ofm = buffer->data;
ofm->match = match;
ofm->priority = htons(priority);
open_vconn(argv[1], &vconn);
+ if (table_idx != 0xff) {
+ enable_flow_mod_table_id_ext(vconn, 1);
+ ofm->command = htons(ntohs(ofm->command) | (table_idx << 8));
+ }
send_openflow_buffer(vconn, buffer);
vconn_close(vconn);
}
do_add_flows(int argc OVS_UNUSED, char *argv[])
{
struct vconn *vconn;
+ struct ofpbuf *b;
FILE *file;
- char line[1024];
++ bool table_id_enabled = false;
+ uint8_t table_idx;
- int table_id_enabled = 0;
file = fopen(argv[2], "r");
if (file == NULL) {
}
open_vconn(argv[1], &vconn);
- while (fgets(line, sizeof line, file)) {
- struct ofpbuf *buffer;
- struct ofp_flow_mod *ofm;
- uint16_t priority, idle_timeout, hard_timeout;
- uint64_t cookie;
- struct ofp_match match;
-
- char *comment;
-
- /* Delete comments. */
- comment = strchr(line, '#');
- if (comment) {
- *comment = '\0';
- }
-
- /* Drop empty lines. */
- if (line[strspn(line, " \t\n")] == '\0') {
- continue;
- }
-
- /* Parse and send. parse_ofp_str() will expand and reallocate
- * the data in 'buffer', so we can't keep pointers to across the
- * parse_ofp_str() call. */
- make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer);
- parse_ofp_str(line, &match, buffer, &table_idx, NULL, &priority,
- &idle_timeout, &hard_timeout, &cookie);
- ofm = buffer->data;
- ofm->match = match;
- ofm->command = htons(OFPFC_ADD);
- ofm->cookie = htonll(cookie);
- ofm->idle_timeout = htons(idle_timeout);
- ofm->hard_timeout = htons(hard_timeout);
- ofm->buffer_id = htonl(UINT32_MAX);
- ofm->priority = htons(priority);
-
- if (table_idx != 0xff) {
- if (!table_id_enabled) {
- enable_flow_mod_table_id_ext(vconn, 1);
- table_id_enabled = 1;
- }
- ofm->command = htons(ntohs(ofm->command) | (table_idx << 8));
- } else {
- if (table_id_enabled) {
- enable_flow_mod_table_id_ext(vconn, 0);
- table_id_enabled = 0;
- }
- while ((b = parse_ofp_add_flow_file(file)) != NULL) {
++ while ((b = parse_ofp_add_flow_file(file, &table_idx)) != NULL) {
++ if ((table_idx != 0xff) != table_id_enabled) {
++ table_id_enabled = table_idx != 0xff;
++ enable_flow_mod_table_id_ext(vconn, table_id_enabled);
+ }
- send_openflow_buffer(vconn, buffer);
+ send_openflow_buffer(vconn, b);
}
vconn_close(vconn);
fclose(file);
struct ofpbuf *buffer;
struct ofp_flow_mod *ofm;
struct ofp_match match;
+ uint8_t table_idx;
/* Parse and send. parse_ofp_str() will expand and reallocate the
* data in 'buffer', so we can't keep pointers to across the
* parse_ofp_str() call. */
make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer);
parse_ofp_str(argv[2], &match, buffer,
- NULL, NULL, &priority, &idle_timeout, &hard_timeout,
+ &table_idx, NULL, &priority, &idle_timeout, &hard_timeout,
&cookie);
ofm = buffer->data;
ofm->match = match;
ofm->priority = htons(priority);
open_vconn(argv[1], &vconn);
+ if (table_idx != 0xff) {
+ enable_flow_mod_table_id_ext(vconn, 1);
+ ofm->command = htons(ntohs(ofm->command) | (table_idx << 8));
+ }
send_openflow_buffer(vconn, buffer);
vconn_close(vconn);
}
uint16_t out_port;
struct ofpbuf *buffer;
struct ofp_flow_mod *ofm;
+ uint8_t table_idx;
/* Parse and send. */
ofm = make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &buffer);
- parse_ofp_str(argc > 2 ? argv[2] : "", &ofm->match, NULL, NULL,
+ parse_ofp_str(argc > 2 ? argv[2] : "", &ofm->match, NULL, &table_idx,
&out_port, &priority, NULL, NULL, NULL);
if (strict) {
ofm->command = htons(OFPFC_DELETE_STRICT);
ofm->priority = htons(priority);
open_vconn(argv[1], &vconn);
+ if (table_idx != 0xff) {
+ enable_flow_mod_table_id_ext(vconn, 1);
+ ofm->command = htons(ntohs(ofm->command) | (table_idx << 8));
+ }
send_openflow_buffer(vconn, buffer);
vconn_close(vconn);
}
count * message_size / (duration / 1000.0));
}
- while ((b = parse_ofp_add_flow_file(file)) != NULL) {
+ /* This command is really only useful for testing the flow parser (ofp_parse),
+ * so it is undocumented. */
+ static void
+ do_parse_flows(int argc OVS_UNUSED, char *argv[])
+ {
++ uint8_t table_idx;
+ struct ofpbuf *b;
+ FILE *file;
+
+ file = fopen(argv[1], "r");
+ if (file == NULL) {
+ ovs_fatal(errno, "%s: open", argv[2]);
+ }
+
++ while ((b = parse_ofp_add_flow_file(file, &table_idx)) != NULL) {
+ ofp_print(stdout, b->data, b->size, 0);
+ ofpbuf_delete(b);
+ }
+ fclose(file);
+ }
+
static void
do_help(int argc OVS_UNUSED, char *argv[] OVS_UNUSED)
{
{ "dump-tables", 1, 1, do_dump_tables },
{ "dump-flows", 1, 2, do_dump_flows },
{ "dump-aggregate", 1, 2, do_dump_aggregate },
+ { "queue-stats", 1, 3, do_queue_stats },
{ "add-flow", 2, 2, do_add_flow },
{ "add-flows", 2, 2, do_add_flows },
{ "mod-flows", 2, 2, do_mod_flows },
{ "probe", 1, 1, do_probe },
{ "ping", 1, 2, do_ping },
{ "benchmark", 3, 3, do_benchmark },
+ { "parse-flows", 1, 1, do_parse_flows },
{ "help", 0, INT_MAX, do_help },
{ NULL, 0, 0, NULL },
};
#include "compiler.h"
#include "daemon.h"
#include "dirs.h"
-#include "dpif.h"
#include "leak-checker.h"
#include "list.h"
#include "netdev.h"
#include "ofpbuf.h"
#include "ofproto/ofproto.h"
+#include "ofproto/wdp.h"
#include "openflow/openflow.h"
#include "packets.h"
#include "poll-loop.h"
#include "util.h"
#include "vconn.h"
#include "vlog.h"
+#include "xfif.h"
VLOG_DEFINE_THIS_MODULE(openflowd)
struct ofproto *ofproto;
struct ofsettings s;
int error;
- struct dpif *dpif;
+ struct xfif *xfif;
struct netflow_options nf_options;
proctitle_init(argc, argv);
VLOG_INFO("Open vSwitch version %s", VERSION BUILDNR);
VLOG_INFO("OpenFlow protocol version 0x%02x", OFP_VERSION);
- error = dpif_create_and_open(s.dp_name, s.dp_type, &dpif);
+ error = xfif_create_and_open(s.dp_name, s.dp_type, &xfif);
if (error) {
ovs_fatal(error, "could not create datapath");
}
size_t i;
SVEC_FOR_EACH (i, port, &s.ports) {
- error = dpif_port_add(dpif, port, 0, NULL);
+ error = xfif_port_add(xfif, port, 0, NULL);
if (error) {
ovs_fatal(error, "failed to add %s as a port", port);
}
ovs_fatal(error, "unrecoverable datapath error");
}
unixctl_server_run(unixctl);
- dp_run();
+ wdp_run();
netdev_run();
ofproto_wait(ofproto);
unixctl_server_wait(unixctl);
- dp_wait();
+ wdp_wait();
netdev_wait();
poll_block();
}
- dpif_close(dpif);
+ xfif_close(xfif);
return 0;
}
}
/* Local vconns. */
- dp_parse_name(argv[0], &s->dp_name, &s->dp_type);
+ xf_parse_name(argv[0], &s->dp_name, &s->dp_type);
/* Figure out controller names. */
if (!controllers.n) {
s->n_controllers = controllers.n;
s->controllers = xmalloc(s->n_controllers * sizeof *s->controllers);
if (argc > 1) {
- size_t i;
-
for (i = 0; i < s->n_controllers; i++) {
s->controllers[i] = controller_opts;
s->controllers[i].target = controllers.names[i];
/* Sanity check. */
if (controller_opts.band == OFPROTO_OUT_OF_BAND) {
- size_t i;
-
for (i = 0; i < s->n_controllers; i++) {
if (!strcmp(s->controllers[i].target, "discover")) {
ovs_fatal(0, "Cannot perform discovery with out-of-band "
#include "bitmap.h"
#include "coverage.h"
#include "dirs.h"
-#include "dpif.h"
#include "dynamic-string.h"
#include "flow.h"
#include "hash.h"
+ #include "hmap.h"
#include "jsonrpc.h"
#include "list.h"
#include "mac-learning.h"
#include "netdev.h"
-#include "odp-util.h"
+#include "xflow-util.h"
#include "ofp-print.h"
#include "ofpbuf.h"
#include "ofproto/netflow.h"
#include "ovsdb-data.h"
#include "packets.h"
#include "poll-loop.h"
- #include "port-array.h"
#include "proc-net-compat.h"
#include "process.h"
#include "sha1.h"
#include "socket-util.h"
#include "stream-ssl.h"
#include "svec.h"
+ #include "system-stats.h"
#include "timeval.h"
#include "util.h"
#include "unixctl.h"
#include "vswitchd/vswitch-idl.h"
#include "xenserver.h"
#include "vlog.h"
+#include "xfif.h"
#include "xtoxll.h"
#include "sflow_api.h"
struct dst {
uint16_t vlan;
- uint16_t dp_ifidx;
+ uint16_t xf_ifidx;
};
struct iface {
/* These members are valid only after bridge_reconfigure() causes them to
* be initialized. */
- struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */
- int dp_ifidx; /* Index within kernel datapath. */
++ struct hmap_node xf_ifidx_node; /* In struct bridge's "ifaces" hmap. */
+ int xf_ifidx; /* Index within kernel datapath. */
struct netdev *netdev; /* Network device. */
bool enabled; /* May be chosen for flows? */
const struct ovsrec_interface *cfg;
struct ofproto *ofproto; /* OpenFlow switch. */
/* Kernel datapath information. */
- struct dpif *dpif; /* Datapath. */
+ struct xfif *xfif; /* Datapath. */
- struct port_array ifaces; /* Indexed by kernel datapath port number. */
+ struct hmap ifaces; /* Contains "struct iface"s. */
/* Bridge ports. */
struct port **ports;
/* OVSDB IDL used to obtain configuration. */
static struct ovsdb_idl *idl;
- /* Each time this timer expires, the bridge fetches statistics for every
- * interface and pushes them into the database. */
- #define IFACE_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
- static long long int iface_stats_timer = LLONG_MIN;
+ /* Each time this timer expires, the bridge fetches systems and interface
+ * statistics and pushes them into the database. */
+ #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
+ static long long int stats_timer = LLONG_MIN;
static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg);
static void bridge_destroy(struct bridge *);
static void port_destroy(struct port *);
static struct port *port_lookup(const struct bridge *, const char *name);
static struct iface *port_lookup_iface(const struct port *, const char *name);
-static struct port *port_from_dp_ifidx(const struct bridge *,
- uint16_t dp_ifidx);
+static struct port *port_from_xf_ifidx(const struct bridge *,
+ uint16_t xf_ifidx);
static void port_update_bond_compat(struct port *);
static void port_update_vlan_compat(struct port *);
static void port_update_bonding(struct port *);
const struct ovsrec_interface *if_cfg);
static void iface_destroy(struct iface *);
static struct iface *iface_lookup(const struct bridge *, const char *name);
-static struct iface *iface_from_dp_ifidx(const struct bridge *,
- uint16_t dp_ifidx);
+static struct iface *iface_from_xf_ifidx(const struct bridge *,
+ uint16_t xf_ifidx);
static bool iface_is_internal(const struct bridge *, const char *name);
static void iface_set_mac(struct iface *);
static void iface_update_qos(struct iface *, const struct ovsrec_qos *);
{
static bool already_configured_once;
struct svec bridge_names;
- struct svec dpif_names, dpif_types;
+ struct svec xfif_names, xfif_types;
size_t i;
/* Only do this once per ovs-vswitchd run. */
}
already_configured_once = true;
- iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
+ stats_timer = time_msec() + STATS_INTERVAL;
/* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
svec_init(&bridge_names);
}
svec_sort(&bridge_names);
- /* Iterate over all system dpifs and delete any of them that do not appear
+ /* Iterate over all system xfifs and delete any of them that do not appear
* in 'cfg'. */
- svec_init(&dpif_names);
- svec_init(&dpif_types);
- dp_enumerate_types(&dpif_types);
- for (i = 0; i < dpif_types.n; i++) {
- struct dpif *dpif;
+ svec_init(&xfif_names);
+ svec_init(&xfif_types);
+ xf_enumerate_types(&xfif_types);
+ for (i = 0; i < xfif_types.n; i++) {
+ struct xfif *xfif;
int retval;
size_t j;
- dp_enumerate_names(dpif_types.names[i], &dpif_names);
+ xf_enumerate_names(xfif_types.names[i], &xfif_names);
- /* For each dpif... */
- for (j = 0; j < dpif_names.n; j++) {
- retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif);
+ /* For each xfif... */
+ for (j = 0; j < xfif_names.n; j++) {
+ retval = xfif_open(xfif_names.names[j], xfif_types.names[i], &xfif);
if (!retval) {
struct svec all_names;
size_t k;
/* ...check whether any of its names is in 'bridge_names'. */
svec_init(&all_names);
- dpif_get_all_names(dpif, &all_names);
+ xfif_get_all_names(xfif, &all_names);
for (k = 0; k < all_names.n; k++) {
if (svec_contains(&bridge_names, all_names.names[k])) {
goto found;
}
}
- /* No. Delete the dpif. */
- dpif_delete(dpif);
+ /* No. Delete the xfif. */
+ xfif_delete(xfif);
found:
svec_destroy(&all_names);
- dpif_close(dpif);
+ xfif_close(xfif);
}
}
}
svec_destroy(&bridge_names);
- svec_destroy(&dpif_names);
- svec_destroy(&dpif_types);
+ svec_destroy(&xfif_names);
+ svec_destroy(&xfif_types);
}
/* Attempt to create the network device 'iface_name' through the netdev
xstrdup(iface_cfg->value_options[i]));
}
+ /* Include 'other_config' keys in hash of netdev options. The
+ * namespace of 'other_config' and 'options' must be disjoint.
+ * Prefer 'options' keys over 'other_config' keys. */
+ for (i = 0; i < iface_cfg->n_other_config; i++) {
+ char *value = xstrdup(iface_cfg->value_other_config[i]);
+ if (!shash_add_once(&options, iface_cfg->key_other_config[i],
+ value)) {
+ VLOG_WARN("%s: \"other_config\" key %s conflicts with existing "
+ "\"other_config\" or \"options\" entry...ignoring",
+ iface_cfg->name, iface_cfg->key_other_config[i]);
+ free(value);
+ }
+ }
+
if (create) {
struct netdev_options netdev_options;
}
netdev_options.args = &options;
netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
-
error = netdev_open(&netdev_options, &iface->netdev);
if (iface->netdev) {
}
static bool
-check_iface_dp_ifidx(struct bridge *br, struct iface *iface,
+check_iface_xf_ifidx(struct bridge *br, struct iface *iface,
void *aux OVS_UNUSED)
{
- if (iface->dp_ifidx >= 0) {
+ if (iface->xf_ifidx >= 0) {
VLOG_DBG("%s has interface %s on port %d",
- dpif_name(br->dpif),
- iface->name, iface->dp_ifidx);
+ xfif_name(br->xfif),
+ iface->name, iface->xf_ifidx);
return true;
} else {
VLOG_ERR("%s interface not in %s, dropping",
- iface->name, dpif_name(br->dpif));
+ iface->name, xfif_name(br->xfif));
return false;
}
}
/* Set MAC address of internal interfaces other than the local
* interface. */
- if (iface->dp_ifidx != ODPP_LOCAL
+ if (iface->xf_ifidx != XFLOWP_LOCAL
&& iface_is_internal(br, iface->name)) {
iface_set_mac(iface);
}
/* Collect old and new bridges. */
shash_init(&old_br);
shash_init(&new_br);
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
shash_add(&old_br, br->name, br);
}
for (i = 0; i < ovs_cfg->n_bridges; i++) {
}
/* Get rid of deleted bridges and add new bridges. */
- LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) {
struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name);
if (br_cfg) {
br->cfg = br_cfg;
shash_destroy(&new_br);
/* Reconfigure all bridges. */
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
bridge_reconfigure_one(br);
}
* The kernel will reject any attempt to add a given port to a datapath if
* that port already belongs to a different datapath, so we must do all
* port deletions before any port additions. */
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
- struct odp_port *dpif_ports;
- size_t n_dpif_ports;
+ struct xflow_port *xfif_ports;
+ size_t n_xfif_ports;
struct shash want_ifaces;
- dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
+ xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
bridge_get_all_ifaces(br, &want_ifaces);
- for (i = 0; i < n_dpif_ports; i++) {
- const struct odp_port *p = &dpif_ports[i];
+ for (i = 0; i < n_xfif_ports; i++) {
+ const struct xflow_port *p = &xfif_ports[i];
if (!shash_find(&want_ifaces, p->devname)
&& strcmp(p->devname, br->name)) {
- int retval = dpif_port_del(br->dpif, p->port);
+ int retval = xfif_port_del(br->xfif, p->port);
if (retval) {
VLOG_ERR("failed to remove %s interface from %s: %s",
- p->devname, dpif_name(br->dpif),
+ p->devname, xfif_name(br->xfif),
strerror(retval));
}
}
}
shash_destroy(&want_ifaces);
- free(dpif_ports);
+ free(xfif_ports);
}
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
- struct odp_port *dpif_ports;
- size_t n_dpif_ports;
+ struct xflow_port *xfif_ports;
+ size_t n_xfif_ports;
struct shash cur_ifaces, want_ifaces;
- struct shash_node *node;
/* Get the set of interfaces currently in this datapath. */
- dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
+ xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
shash_init(&cur_ifaces);
- for (i = 0; i < n_dpif_ports; i++) {
- const char *name = dpif_ports[i].devname;
+ for (i = 0; i < n_xfif_ports; i++) {
+ const char *name = xfif_ports[i].devname;
shash_add_once(&cur_ifaces, name, NULL);
}
- free(dpif_ports);
+ free(xfif_ports);
/* Get the set of interfaces we want on this datapath. */
bridge_get_all_ifaces(br, &want_ifaces);
/* Add to datapath. */
internal = iface_is_internal(br, if_name);
- error = dpif_port_add(br->dpif, if_name,
- internal ? ODP_PORT_INTERNAL : 0, NULL);
+ error = xfif_port_add(br->xfif, if_name,
+ internal ? XFLOW_PORT_INTERNAL : 0, NULL);
if (error == EFBIG) {
VLOG_ERR("ran out of valid port numbers on %s",
- dpif_name(br->dpif));
+ xfif_name(br->xfif));
break;
} else if (error) {
VLOG_ERR("failed to add %s interface to %s: %s",
- if_name, dpif_name(br->dpif), strerror(error));
+ if_name, xfif_name(br->xfif), strerror(error));
}
}
}
shash_destroy(&want_ifaces);
}
sflow_bridge_number = 0;
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
uint8_t ea[8];
uint64_t dpid;
struct iface *local_iface;
bridge_fetch_dp_ifaces(br);
iterate_and_prune_ifaces(br, check_iface_netdev, NULL);
- iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL);
+ iterate_and_prune_ifaces(br, check_iface_xf_ifidx, NULL);
/* Pick local port hardware address, datapath ID. */
bridge_pick_local_hw_addr(br, ea, &hw_addr_iface);
memset(&opts, 0, sizeof opts);
- dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id);
+ xfif_get_netflow_ids(br->xfif, &opts.engine_type, &opts.engine_id);
if (nf_cfg->engine_type) {
opts.engine_type = *nf_cfg->engine_type;
}
struct ovsrec_controller **controllers;
struct ofproto_sflow_options oso;
size_t n_controllers;
- size_t i;
memset(&oso, 0, sizeof oso);
/* Update the controller and related settings. It would be more
* straightforward to call this from bridge_reconfigure_one(), but we
* can't do it there for two reasons. First, and most importantly, at
- * that point we don't know the dp_ifidx of any interfaces that have
+ * that point we don't know the xf_ifidx of any interfaces that have
* been added to the bridge (because we haven't actually added them to
* the datapath). Second, at that point we haven't set the datapath ID
* yet; when a controller is configured, resetting the datapath ID will
* the datapath ID before the controller. */
bridge_reconfigure_remotes(br, managers, n_managers);
}
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
for (i = 0; i < br->n_ports; i++) {
struct port *port = br->ports[i];
int j;
}
}
}
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
iterate_and_prune_ifaces(br, set_iface_properties, NULL);
}
/* The local port doesn't count (since we're trying to choose its
* MAC address anyway). */
- if (iface->dp_ifidx == ODPP_LOCAL) {
+ if (iface->xf_ifidx == XFLOWP_LOCAL) {
continue;
}
ovsrec_interface_set_statistics(iface->cfg, keys, values, n);
}
+ static void
+ refresh_system_stats(const struct ovsrec_open_vswitch *cfg)
+ {
+ struct ovsdb_datum datum;
+ struct shash stats;
+
+ shash_init(&stats);
+ get_system_stats(&stats);
+
+ ovsdb_datum_from_shash(&datum, &stats);
+ ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics,
+ &datum);
+ }
+
void
bridge_run(void)
{
/* Let each bridge do the work that it needs to do. */
datapath_destroyed = false;
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
int error = bridge_run_one(br);
if (error) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
}
#endif
- /* Refresh interface stats if necessary. */
- if (time_msec() >= iface_stats_timer) {
- struct ovsdb_idl_txn *txn;
+ /* Refresh system and interface stats if necessary. */
+ if (time_msec() >= stats_timer) {
+ if (cfg) {
+ struct ovsdb_idl_txn *txn;
- txn = ovsdb_idl_txn_create(idl);
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
- size_t i;
+ txn = ovsdb_idl_txn_create(idl);
+ LIST_FOR_EACH (br, node, &all_bridges) {
+ size_t i;
- for (i = 0; i < br->n_ports; i++) {
- struct port *port = br->ports[i];
- size_t j;
+ for (i = 0; i < br->n_ports; i++) {
+ struct port *port = br->ports[i];
+ size_t j;
- for (j = 0; j < port->n_ifaces; j++) {
- struct iface *iface = port->ifaces[j];
- iface_refresh_stats(iface);
+ for (j = 0; j < port->n_ifaces; j++) {
+ struct iface *iface = port->ifaces[j];
+ iface_refresh_stats(iface);
+ }
}
}
+ refresh_system_stats(cfg);
+ ovsdb_idl_txn_commit(txn);
+ ovsdb_idl_txn_destroy(txn); /* XXX */
}
- ovsdb_idl_txn_commit(txn);
- ovsdb_idl_txn_destroy(txn); /* XXX */
- iface_stats_timer = time_msec() + IFACE_STATS_INTERVAL;
+ stats_timer = time_msec() + STATS_INTERVAL;
}
}
{
struct bridge *br;
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
ofproto_wait(br->ofproto);
if (ofproto_has_primary_controller(br->ofproto)) {
continue;
bond_wait(br);
}
ovsdb_idl_wait(idl);
- poll_timer_wait_until(iface_stats_timer);
+ poll_timer_wait_until(stats_timer);
}
/* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
bridge_flush(struct bridge *br)
{
COVERAGE_INC(bridge_flush);
- br->flush = true;
+ ofproto_revalidate_all(br->ofproto);
mac_learning_flush(br->ml);
}
-/* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
+/* Returns the 'br' interface for the XFLOWP_LOCAL port, or null if 'br' has no
* such interface. */
static struct iface *
bridge_get_local_iface(struct bridge *br)
struct port *port = br->ports[i];
for (j = 0; j < port->n_ifaces; j++) {
struct iface *iface = port->ifaces[j];
- if (iface->dp_ifidx == ODPP_LOCAL) {
+ if (iface->xf_ifidx == XFLOWP_LOCAL) {
return iface;
}
}
}
ds_put_cstr(&ds, " port VLAN MAC Age\n");
- LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
+ LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
if (e->port < 0 || e->port >= br->n_ports) {
continue;
}
ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
- br->ports[e->port]->ifaces[0]->dp_ifidx,
+ br->ports[e->port]->ifaces[0]->xf_ifidx,
e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
}
unixctl_command_reply(conn, 200, ds_cstr(&ds));
assert(!bridge_lookup(br_cfg->name));
br = xzalloc(sizeof *br);
- error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
- &br->dpif);
+ error = xfif_create_and_open(br_cfg->name, br_cfg->datapath_type,
+ &br->xfif);
if (error) {
free(br);
return NULL;
}
- dpif_flow_flush(br->dpif);
+ xfif_flow_flush(br->xfif);
error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
br, &br->ofproto);
if (error) {
VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
strerror(error));
- dpif_delete(br->dpif);
- dpif_close(br->dpif);
+ xfif_delete(br->xfif);
+ xfif_close(br->xfif);
free(br);
return NULL;
}
br->ml = mac_learning_create();
eth_addr_nicira_random(br->default_ea);
- port_array_init(&br->ifaces);
+ hmap_init(&br->ifaces);
shash_init(&br->port_by_name);
shash_init(&br->iface_by_name);
- br->flush = false;
-
list_push_back(&all_bridges, &br->node);
- VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
+ VLOG_INFO("created bridge %s on %s", br->name, xfif_name(br->xfif));
return br;
}
port_destroy(br->ports[br->n_ports - 1]);
}
list_remove(&br->node);
- error = dpif_delete(br->dpif);
+ error = xfif_delete(br->xfif);
if (error && error != ENOENT) {
VLOG_ERR("failed to delete %s: %s",
- dpif_name(br->dpif), strerror(error));
+ xfif_name(br->xfif), strerror(error));
}
- dpif_close(br->dpif);
+ xfif_close(br->xfif);
ofproto_destroy(br->ofproto);
mac_learning_destroy(br->ml);
- port_array_destroy(&br->ifaces);
+ hmap_destroy(&br->ifaces);
shash_destroy(&br->port_by_name);
shash_destroy(&br->iface_by_name);
free(br->ports);
{
struct bridge *br;
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
if (!strcmp(br->name, name)) {
return br;
}
}
ofproto_reconnect_controllers(br->ofproto);
} else {
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
ofproto_reconnect_controllers(br->ofproto);
}
}
static int
bridge_run_one(struct bridge *br)
{
- int error;
-
- error = ofproto_run1(br->ofproto);
- if (error) {
- return error;
- }
-
- mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
+ ofproto_revalidate(br->ofproto, mac_learning_run(br->ml));
bond_run(br);
-
- error = ofproto_run2(br->ofproto, br->flush);
- br->flush = false;
-
- return error;
+ return ofproto_run(br->ofproto);
}
static size_t
char local_name[IF_NAMESIZE];
int error;
- error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
+ error = xfif_port_get_name(br->xfif, XFLOWP_LOCAL,
local_name, sizeof local_name);
if (!error && !shash_find(&new_ports, local_name)) {
VLOG_WARN("bridge %s: controller specified but no local port "
* domain sockets and overwriting arbitrary local files. */
VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller "
"\"%s\" due to possibility for remote exploit",
- dpif_name(br->dpif), c->target);
+ br->name, c->target);
continue;
}
action.output.len = htons(sizeof action);
action.output.port = htons(OFPP_NORMAL);
memset(&flow, 0, sizeof flow);
- ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0);
+ flow.wildcards = OVSFW_ALL;
+ ofproto_add_flow(br->ofproto, &flow, &action, 1, 0);
}
}
/* For robustness, in case the administrator moves around datapath ports behind
* our back, we re-check all the datapath port numbers here.
*
- * This function will set the 'dp_ifidx' members of interfaces that have
+ * This function will set the 'xf_ifidx' members of interfaces that have
* disappeared to -1, so only call this function from a context where those
* 'struct iface's will be removed from the bridge. Otherwise, the -1
- * 'dp_ifidx'es will cause trouble later when we try to send them to the
+ * 'xf_ifidx'es will cause trouble later when we try to send them to the
* datapath, which doesn't support UINT16_MAX+1 ports. */
static void
bridge_fetch_dp_ifaces(struct bridge *br)
{
- struct odp_port *dpif_ports;
- size_t n_dpif_ports;
+ struct xflow_port *xfif_ports;
+ size_t n_xfif_ports;
size_t i, j;
/* Reset all interface numbers. */
struct port *port = br->ports[i];
for (j = 0; j < port->n_ifaces; j++) {
struct iface *iface = port->ifaces[j];
- iface->dp_ifidx = -1;
+ iface->xf_ifidx = -1;
}
}
- port_array_clear(&br->ifaces);
+ hmap_clear(&br->ifaces);
- dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
- for (i = 0; i < n_dpif_ports; i++) {
- struct odp_port *p = &dpif_ports[i];
+ xfif_port_list(br->xfif, &xfif_ports, &n_xfif_ports);
+ for (i = 0; i < n_xfif_ports; i++) {
+ struct xflow_port *p = &xfif_ports[i];
struct iface *iface = iface_lookup(br, p->devname);
if (iface) {
- if (iface->dp_ifidx >= 0) {
+ if (iface->xf_ifidx >= 0) {
VLOG_WARN("%s reported interface %s twice",
- dpif_name(br->dpif), p->devname);
- } else if (iface_from_dp_ifidx(br, p->port)) {
+ xfif_name(br->xfif), p->devname);
+ } else if (iface_from_xf_ifidx(br, p->port)) {
VLOG_WARN("%s reported interface %"PRIu16" twice",
- dpif_name(br->dpif), p->port);
+ xfif_name(br->xfif), p->port);
} else {
- port_array_set(&br->ifaces, p->port, iface);
- iface->dp_ifidx = p->port;
- hmap_insert(&br->ifaces, &iface->dp_ifidx_node,
- hash_int(iface->dp_ifidx, 0));
+ iface->xf_ifidx = p->port;
++ hmap_insert(&br->ifaces, &iface->xf_ifidx_node,
++ hash_int(iface->xf_ifidx, 0));
}
if (iface->cfg) {
- int64_t ofport = (iface->dp_ifidx >= 0
- ? odp_port_to_ofp_port(iface->dp_ifidx)
+ int64_t ofport = (iface->xf_ifidx >= 0
+ ? xflow_port_to_ofp_port(iface->xf_ifidx)
: -1);
ovsrec_interface_set_ofport(iface->cfg, &ofport, 1);
}
}
}
- free(dpif_ports);
+ free(xfif_ports);
}
\f
/* Bridge packet processing functions. */
static bool
choose_output_iface(const struct port *port, const uint8_t *dl_src,
- uint16_t *dp_ifidx, tag_type *tags)
+ uint16_t *xf_ifidx, tag_type *tags)
{
struct iface *iface;
*tags |= e->iface_tag;
iface = port->ifaces[e->iface_idx];
}
- *dp_ifidx = iface->dp_ifidx;
+ *xf_ifidx = iface->xf_ifidx;
*tags |= iface->tag; /* Currently only used for bonding. */
return true;
}
p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
: in_port->vlan >= 0 ? in_port->vlan
: ntohs(flow->dl_vlan));
- return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
+ return choose_output_iface(out_port, flow->dl_src, &p->xf_ifidx, tags);
}
static void
{
size_t i;
for (i = 0; i < n_dsts; i++) {
- if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
+ if (dsts[i].vlan == test->vlan && dsts[i].xf_ifidx == test->xf_ifidx) {
return true;
}
}
size_t i;
if (out_port == FLOOD_PORT) {
- /* XXX use ODP_FLOOD if no vlans or bonding. */
+ /* XXX use XFLOW_FLOOD if no vlans or bonding. */
/* XXX even better, define each VLAN as a datapath port group */
for (i = 0; i < br->n_ports; i++) {
struct port *port = br->ports[i];
}
*nf_output_iface = NF_OUT_FLOOD;
} else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
- *nf_output_iface = dst->dp_ifidx;
+ *nf_output_iface = dst->xf_ifidx;
mirrors |= out_port->dst_mirrors;
dst++;
}
print_dsts(const struct dst *dsts, size_t n)
{
for (; n--; dsts++) {
- printf(">p%"PRIu16, dsts->dp_ifidx);
+ printf(">p%"PRIu16, dsts->xf_ifidx);
if (dsts->vlan != OFP_VLAN_NONE) {
printf("v%"PRIu16, dsts->vlan);
}
static void
compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
const struct port *in_port, const struct port *out_port,
- tag_type *tags, struct odp_actions *actions,
+ tag_type *tags, struct xflow_actions *actions,
uint16_t *nf_output_iface)
{
struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
cur_vlan = ntohs(flow->dl_vlan);
for (p = dsts; p < &dsts[n_dsts]; p++) {
- union odp_action *a;
+ union xflow_action *a;
if (p->vlan != cur_vlan) {
if (p->vlan == OFP_VLAN_NONE) {
- odp_actions_add(actions, ODPAT_STRIP_VLAN);
+ xflow_actions_add(actions, XFLOWAT_STRIP_VLAN);
} else {
- a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
- a->vlan_vid.vlan_vid = htons(p->vlan);
+ a = xflow_actions_add(actions, XFLOWAT_SET_DL_TCI);
+ a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK);
+ a->dl_tci.mask = htons(VLAN_VID_MASK);
}
cur_vlan = p->vlan;
}
- a = odp_actions_add(actions, ODPAT_OUTPUT);
- a->output.port = p->dp_ifidx;
+ a = xflow_actions_add(actions, XFLOWAT_OUTPUT);
+ a->output.port = p->xf_ifidx;
}
}
int vlan;
/* Find the interface and port structure for the received packet. */
- in_iface = iface_from_dp_ifidx(br, flow->in_port);
+ in_iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(flow->in_port));
if (!in_iface) {
/* No interface? Something fishy... */
if (have_packet) {
* not at all, if 'packet' was NULL. */
static bool
process_flow(struct bridge *br, const flow_t *flow,
- const struct ofpbuf *packet, struct odp_actions *actions,
+ const struct ofpbuf *packet, struct xflow_actions *actions,
tag_type *tags, uint16_t *nf_output_iface)
{
struct port *in_port;
struct iface *iface;
struct port *port;
- iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
+ iface = iface_from_xf_ifidx(br, ofp_port_to_xflow_port(opp->port_no));
if (!iface) {
return;
}
static bool
bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
- struct odp_actions *actions, tag_type *tags,
+ struct xflow_actions *actions, tag_type *tags,
uint16_t *nf_output_iface, void *br_)
{
struct bridge *br = br_;
static void
bridge_account_flow_ofhook_cb(const flow_t *flow, tag_type tags,
- const union odp_action *actions,
+ const union xflow_action *actions,
size_t n_actions, unsigned long long int n_bytes,
void *br_)
{
struct bridge *br = br_;
- const union odp_action *a;
+ const union xflow_action *a;
struct port *in_port;
tag_type dummy = 0;
int vlan;
return;
}
for (a = actions; a < &actions[n_actions]; a++) {
- if (a->type == ODPAT_OUTPUT) {
- struct port *out_port = port_from_dp_ifidx(br, a->output.port);
+ if (a->type == XFLOWAT_OUTPUT) {
+ struct port *out_port = port_from_xf_ifidx(br, a->output.port);
if (out_port && out_port->n_ifaces >= 2) {
struct bond_entry *e = lookup_bond_entry(out_port,
flow->dl_src);
* smallest hashes instead of the biggest ones. There is little
* reason behind this decision; we could use the opposite sort
* order to shift away big hashes ahead of small ones. */
- size_t i;
bool order_swapped;
for (i = 0; i < from->n_hashes; i++) {
ofpbuf_init(&packet, 128);
error = n_packets = n_errors = 0;
- LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
+ LIST_FOR_EACH (e, lru_node, &br->ml->lrus) {
union ofp_action actions[2], *a;
- uint16_t dp_ifidx;
+ uint16_t xf_ifidx;
tag_type tags = 0;
flow_t flow;
int retval;
if (e->port == port->port_idx
- || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
+ || !choose_output_iface(port, e->mac, &xf_ifidx, &tags)) {
continue;
}
}
a->output.type = htons(OFPAT_OUTPUT);
a->output.len = htons(sizeof *a);
- a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
+ a->output.port = htons(xflow_port_to_ofp_port(xf_ifidx));
a++;
/* Send packet. */
n_packets++;
compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177,
e->mac);
- flow_extract(&packet, 0, ODPP_NONE, &flow);
+ flow_extract(&packet, 0, OFPP_NONE, &flow);
retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
&packet);
if (retval) {
ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
size_t i;
for (i = 0; i < br->n_ports; i++) {
{
const struct bridge *br;
- LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
+ LIST_FOR_EACH (br, node, &all_bridges) {
size_t i;
for (i = 0; i < br->n_ports; i++) {
hash, be->tx_bytes / 1024);
/* MACs. */
- LIST_FOR_EACH (me, struct mac_entry, lru_node,
- &port->bridge->ml->lrus) {
+ LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) {
- uint16_t dp_ifidx;
+ uint16_t xf_ifidx;
tag_type tags = 0;
if (bond_hash(me->mac) == hash
&& me->port != port->port_idx
- && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
- && dp_ifidx == iface->dp_ifidx)
+ && choose_output_iface(port, me->mac, &xf_ifidx, &tags)
+ && xf_ifidx == iface->xf_ifidx)
{
ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
ETH_ADDR_ARGS(me->mac));
trunks = NULL;
if (vlan < 0 && cfg->n_trunks) {
size_t n_errors;
- size_t i;
trunks = bitmap_allocate(4096);
n_errors = 0;
}
static struct port *
-port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
+port_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
{
- struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
+ struct iface *iface = iface_from_xf_ifidx(br, xf_ifidx);
return iface ? iface->port : NULL;
}
iface->port = port;
iface->port_ifidx = port->n_ifaces;
iface->name = xstrdup(name);
- iface->dp_ifidx = -1;
+ iface->xf_ifidx = -1;
iface->tag = tag_create_random();
iface->delay_expires = LLONG_MAX;
iface->netdev = NULL;
shash_find_and_delete_assert(&br->iface_by_name, iface->name);
- if (iface->dp_ifidx >= 0) {
- hmap_remove(&br->ifaces, &iface->dp_ifidx_node);
+ if (iface->xf_ifidx >= 0) {
- port_array_set(&br->ifaces, iface->xf_ifidx, NULL);
++ hmap_remove(&br->ifaces, &iface->xf_ifidx_node);
}
del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
}
static struct iface *
-iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
+iface_from_xf_ifidx(const struct bridge *br, uint16_t xf_ifidx)
{
- return port_array_get(&br->ifaces, xf_ifidx);
+ struct iface *iface;
+
- HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node,
- hash_int(dp_ifidx, 0), &br->ifaces) {
- if (iface->dp_ifidx == dp_ifidx) {
++ HMAP_FOR_EACH_IN_BUCKET (iface, xf_ifidx_node,
++ hash_int(xf_ifidx, 0), &br->ifaces) {
++ if (iface->xf_ifidx == xf_ifidx) {
+ return iface;
+ }
+ }
+ return NULL;
}
/* Returns true if 'iface' is the name of an "internal" interface on bridge
* 'br', that is, an interface that is entirely simulated within the datapath.
- * The local port (ODPP_LOCAL) is always an internal interface. Other local
+ * The local port (XFLOWP_LOCAL) is always an internal interface. Other local
* interfaces are created by setting "iface.<iface>.internal = true".
*
* In addition, we have a kluge-y feature that creates an internal port with
if (eth_addr_is_multicast(ea)) {
VLOG_ERR("interface %s: cannot set MAC to multicast address",
iface->name);
- } else if (iface->dp_ifidx == ODPP_LOCAL) {
+ } else if (iface->xf_ifidx == XFLOWP_LOCAL) {
VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
iface->name, iface->name);
} else {
git://git.onelab.eu / sliver-openvswitch.git / commitdiff