#include "datapath.h"
#include "table.h"
#include "chain.h"
+#include "dp_dev.h"
#include "forward.h"
#include "flow.h"
-#include "datapath_t.h"
#include "compat.h"
#define BRIDGE_PORT_NO_FLOOD 0x00000001
#define UINT32_MAX 4294967295U
+#define UINT16_MAX 65535
#define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
struct net_bridge_port {
static struct genl_family dp_genl_family;
static struct genl_multicast_group mc_group;
-int dp_dev_setup(struct net_device *dev);
-
/* It's hard to imagine wanting more than one datapath, but... */
#define DP_MAX 32
static int dp_maint_func(void *data);
static int send_port_status(struct net_bridge_port *p, uint8_t status);
-
+static int dp_genl_openflow_done(struct netlink_callback *);
+static struct net_bridge_port *new_nbp(struct datapath *,
+ struct net_device *, int port_no);
+static int del_switch_port(struct net_bridge_port *);
/* nla_shrink - reduce amount of space reserved by nla_reserve
* @skb: socket buffer from which to recover room
struct sk_buff *skb;
int max_openflow_len;
+ if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
+ if (net_ratelimit())
+ printk("alloc_openflow_skb: openflow message too large: %zu\n",
+ openflow_len);
+ return NULL;
+ }
+
genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
genl_len += nla_total_size(openflow_len); /* DP_GENL_A_OPENFLOW */
static int
send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
{
- int err = (sender
- ? genlmsg_unicast(skb, sender->pid)
- : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
- if (err && net_ratelimit())
- printk(KERN_WARNING "send_openflow_skb: send failed: %d\n",
- err);
- return err;
+ return (sender
+ ? genlmsg_unicast(skb, sender->pid)
+ : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
}
/* Generates a unique datapath id. It incorporates the datapath index
if (dp == NULL)
goto err_unlock;
+ /* Setup our "of" device */
+ err = dp_dev_setup(dp);
+ if (err)
+ goto err_free_dp;
+
dp->dp_idx = dp_idx;
dp->id = gen_datapath_id(dp_idx);
dp->chain = chain_create(dp);
if (dp->chain == NULL)
- goto err_free_dp;
+ goto err_destroy_dp_dev;
INIT_LIST_HEAD(&dp->port_list);
-#if 0
- /* Setup our "of" device */
- dp->dev.priv = dp;
- rtnl_lock();
- err = dp_dev_setup(&dp->dev);
- rtnl_unlock();
- if (err != 0)
- printk("datapath: problem setting up 'of' device\n");
-#endif
+ dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
+ if (IS_ERR(dp->local_port)) {
+ err = PTR_ERR(dp->local_port);
+ goto err_destroy_local_port;
+ }
dp->flags = 0;
dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
if (IS_ERR(dp->dp_task))
- goto err_free_dp;
+ goto err_destroy_chain;
rcu_assign_pointer(dps[dp_idx], dp);
mutex_unlock(&dp_mutex);
return 0;
+err_destroy_local_port:
+ del_switch_port(dp->local_port);
+err_destroy_chain:
+ chain_destroy(dp->chain);
+err_destroy_dp_dev:
+ dp_dev_destroy(dp);
err_free_dp:
kfree(dp);
err_unlock:
}
static struct net_bridge_port *new_nbp(struct datapath *dp,
- struct net_device *dev)
+ struct net_device *dev, int port_no)
{
struct net_bridge_port *p;
- int port_no;
- port_no = find_portno(dp);
- if (port_no < 0)
- return ERR_PTR(port_no);
+ if (dev->br_port != NULL)
+ return ERR_PTR(-EBUSY);
p = kzalloc(sizeof(*p), GFP_KERNEL);
if (p == NULL)
return ERR_PTR(-ENOMEM);
- p->dp = dp;
+ rtnl_lock();
+ dev_set_promiscuity(dev, 1);
+ rtnl_unlock();
dev_hold(dev);
+ p->dp = dp;
p->dev = dev;
p->port_no = port_no;
+ if (port_no != OFPP_LOCAL)
+ rcu_assign_pointer(dev->br_port, p);
+ if (port_no < OFPP_MAX)
+ rcu_assign_pointer(dp->ports[port_no], p);
+ list_add_rcu(&p->node, &dp->port_list);
return p;
}
int add_switch_port(struct datapath *dp, struct net_device *dev)
{
struct net_bridge_port *p;
+ int port_no;
- if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
+ if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
+ || is_dp_dev(dev))
return -EINVAL;
- if (dev->br_port != NULL)
- return -EBUSY;
+ port_no = find_portno(dp);
+ if (port_no < 0)
+ return port_no;
- p = new_nbp(dp, dev);
+ p = new_nbp(dp, dev, port_no);
if (IS_ERR(p))
return PTR_ERR(p);
- dev_hold(dev);
- rcu_assign_pointer(dev->br_port, p);
- rtnl_lock();
- dev_set_promiscuity(dev, 1);
- rtnl_unlock();
-
- rcu_assign_pointer(dp->ports[p->port_no], p);
- list_add_rcu(&p->node, &dp->port_list);
-
/* Notify the ctlpath that this port has been added */
send_port_status(p, OFPPR_ADD);
dev_set_promiscuity(p->dev, -1);
rtnl_unlock();
list_del_rcu(&p->node);
- rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
+ if (p->port_no != OFPP_LOCAL)
+ rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
rcu_assign_pointer(p->dev->br_port, NULL);
/* Then wait until no one is still using it, and destroy it. */
{
struct net_bridge_port *p, *n;
-#if 0
- /* Unregister the "of" device of this dp */
- rtnl_lock();
- unregister_netdevice(&dp->dev);
- rtnl_unlock();
-#endif
-
kthread_stop(dp->dp_task);
/* Drop references to DP. */
del_switch_port(p);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
+ /* Kill off local_port dev references from buffered packets that have
+ * associated dst entries. */
+ synchronize_rcu();
+ fwd_discard_all();
+
+ /* Destroy dp->netdev. (Must follow deleting switch ports since
+ * dp->local_port has a reference to it.) */
+ dp_dev_destroy(dp);
+
/* Wait until no longer in use, then destroy it. */
synchronize_rcu();
chain_destroy(dp->chain);
struct datapath *dp = (struct datapath *) data;
while (!kthread_should_stop()) {
-#if 1
chain_timeout(dp->chain);
-#else
- int count = chain_timeout(dp->chain);
- chain_print_stats(dp->chain);
- if (count)
- printk("%d flows timed out\n", count);
-#endif
msleep_interruptible(MAINT_SLEEP_MSECS);
}
return 0;
}
+static void
+do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
+{
+ /* Push the Ethernet header back on. */
+ skb_push(skb, ETH_HLEN);
+ fwd_port_input(p->dp->chain, skb, p->port_no);
+}
+
/*
* Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
- * different set of devices!) Returns 0 if *pskb should be processed further,
- * 1 if *pskb is handled. */
+ * different set of devices!)
+ */
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
/* Called with rcu_read_lock. */
static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
struct sk_buff *skb)
{
- struct ethhdr *eh = eth_hdr(skb);
- struct sk_buff *skb_local = NULL;
-
-
- if (compare_ether_addr(eh->h_dest, skb->dev->dev_addr) == 0)
- return skb;
-
- if (is_broadcast_ether_addr(eh->h_dest)
- || is_multicast_ether_addr(eh->h_dest)
- || is_local_ether_addr(eh->h_dest))
- skb_local = skb_clone(skb, GFP_ATOMIC);
-
- /* Push the Ethernet header back on. */
- if (skb->protocol == htons(ETH_P_8021Q))
- skb_push(skb, VLAN_ETH_HLEN);
- else
- skb_push(skb, ETH_HLEN);
-
- fwd_port_input(p->dp->chain, skb, p->port_no);
-
- return skb_local;
+ do_port_input(p, skb);
+ return NULL;
}
#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
{
- /* Push the Ethernet header back on. */
- if ((*pskb)->protocol == htons(ETH_P_8021Q))
- skb_push(*pskb, VLAN_ETH_HLEN);
- else
- skb_push(*pskb, ETH_HLEN);
-
- fwd_port_input(p->dp->chain, *pskb, p->port_no);
+ do_port_input(p, *pskb);
return 1;
}
-#else
+#else
/* NB: This has only been tested on 2.4.35 */
/* Called without any locks (?) */
static void dp_frame_hook(struct sk_buff *skb)
{
struct net_bridge_port *p = skb->dev->br_port;
-
- /* Push the Ethernet header back on. */
- if (skb->protocol == htons(ETH_P_8021Q))
- skb_push(skb, VLAN_ETH_HLEN);
- else
- skb_push(skb, ETH_HLEN);
-
if (p) {
rcu_read_lock();
- fwd_port_input(p->dp->chain, skb, p->port_no);
+ do_port_input(p, skb);
rcu_read_unlock();
} else
kfree_skb(skb);
/* Forwarding output path.
* Based on net/bridge/br_forward.c. */
-/* Don't forward packets to originating port or with flooding disabled */
-static inline int should_deliver(const struct net_bridge_port *p,
- const struct sk_buff *skb)
-{
- if ((skb->dev == p->dev) || (p->flags & BRIDGE_PORT_NO_FLOOD)) {
- return 0;
- }
-
- return 1;
-}
-
static inline unsigned packet_length(const struct sk_buff *skb)
{
int length = skb->len - ETH_HLEN;
return length;
}
+/* Send packets out all the ports except the originating one. If the
+ * "flood" argument is set, only send along the minimum spanning tree.
+ */
static int
-flood(struct datapath *dp, struct sk_buff *skb)
+output_all(struct datapath *dp, struct sk_buff *skb, int flood)
{
+ u32 disable = flood ? BRIDGE_PORT_NO_FLOOD : 0;
struct net_bridge_port *p;
- int prev_port;
+ int prev_port = -1;
- prev_port = -1;
list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (!should_deliver(p, skb))
+ if (skb->dev == p->dev || p->flags & disable)
continue;
if (prev_port != -1) {
struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
int dp_set_origin(struct datapath *dp, uint16_t in_port,
struct sk_buff *skb)
{
- if (in_port < OFPP_MAX && dp->ports[in_port]) {
- skb->dev = dp->ports[in_port]->dev;
+ struct net_bridge_port *p = (in_port < OFPP_MAX ? dp->ports[in_port]
+ : in_port == OFPP_LOCAL ? dp->local_port
+ : NULL);
+ if (p) {
+ skb->dev = p->dev;
return 0;
}
return -ENOENT;
*/
int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
{
- struct net_bridge_port *p;
- int len = skb->len;
-
BUG_ON(!skb);
if (out_port == OFPP_FLOOD)
- return flood(dp, skb);
+ return output_all(dp, skb, 1);
+ else if (out_port == OFPP_ALL)
+ return output_all(dp, skb, 0);
else if (out_port == OFPP_CONTROLLER)
return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
OFPR_ACTION);
else if (out_port == OFPP_TABLE) {
+ struct net_bridge_port *p = skb->dev->br_port;
struct sw_flow_key key;
struct sw_flow *flow;
- flow_extract(skb, skb->dev->br_port->port_no, &key);
+ flow_extract(skb, p ? p->port_no : OFPP_LOCAL, &key);
flow = chain_lookup(dp->chain, &key);
if (likely(flow != NULL)) {
flow_used(flow, skb);
return 0;
}
return -ESRCH;
- } else if (out_port >= OFPP_MAX)
- goto bad_port;
+ } else if (out_port == OFPP_LOCAL) {
+ struct net_device *dev = dp->netdev;
+ return dev ? dp_dev_recv(dev, skb) : -ESRCH;
+ } else if (out_port >= 0 && out_port < OFPP_MAX) {
+ struct net_bridge_port *p = dp->ports[out_port];
+ int len = skb->len;
+ if (p == NULL)
+ goto bad_port;
+ skb->dev = p->dev;
+ if (packet_length(skb) > skb->dev->mtu) {
+ printk("dropped over-mtu packet: %d > %d\n",
+ packet_length(skb), skb->dev->mtu);
+ kfree_skb(skb);
+ return -E2BIG;
+ }
- p = dp->ports[out_port];
- if (p == NULL)
- goto bad_port;
+ dev_queue_xmit(skb);
- skb->dev = p->dev;
- if (packet_length(skb) > skb->dev->mtu) {
- printk("dropped over-mtu packet: %d > %d\n",
- packet_length(skb), skb->dev->mtu);
- kfree_skb(skb);
- return -E2BIG;
+ return len;
}
- dev_queue_xmit(skb);
-
- return len;
-
bad_port:
kfree_skb(skb);
if (net_ratelimit())
* forward the whole packet? */
struct sk_buff *f_skb;
struct ofp_packet_in *opi;
+ struct net_bridge_port *p;
size_t fwd_len, opi_len;
int err;
}
opi->buffer_id = htonl(buffer_id);
opi->total_len = htons(skb->len);
- opi->in_port = htons(skb->dev->br_port->port_no);
+ p = skb->dev->br_port;
+ opi->in_port = htons(p ? p->port_no : OFPP_LOCAL);
opi->reason = reason;
opi->pad = 0;
memcpy(opi->data, skb_mac_header(skb), fwd_len);
int
dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
{
- struct net_bridge_port *p;
-
- p = dp->ports[htons(opp->port_no)];
-
+ int port_no = ntohs(opp->port_no);
+ struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no]
+ : port_no == OFPP_LOCAL ? dp->local_port
+ : NULL);
/* Make sure the port id hasn't changed since this was sent */
- if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN) != 0)
+ if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN))
return -1;
-
p->flags = htonl(opp->flags);
-
return 0;
}
if (!ops)
return -ENOMEM;
ops->reason = status;
+ memset(ops->pad, 0, sizeof ops->pad);
fill_port_desc(p, &ops->desc);
return send_openflow_skb(skb, NULL);
return -ENOMEM;
flow_fill_match(&ofe->match, &flow->key);
- duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
- ofe->duration = htonl(duration_j / HZ);
- ofe->packet_count = cpu_to_be64(flow->packet_count);
- ofe->byte_count = cpu_to_be64(flow->byte_count);
- return send_openflow_skb(skb, NULL);
-}
-static void
-fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
- int table_idx)
-{
- ofs->match.wildcards = htons(flow->key.wildcards);
- ofs->match.in_port = flow->key.in_port;
- memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
- memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
- ofs->match.dl_vlan = flow->key.dl_vlan;
- ofs->match.dl_type = flow->key.dl_type;
- ofs->match.nw_src = flow->key.nw_src;
- ofs->match.nw_dst = flow->key.nw_dst;
- ofs->match.nw_proto = flow->key.nw_proto;
- memset(ofs->match.pad, 0, sizeof ofs->match.pad);
- ofs->match.tp_src = flow->key.tp_src;
- ofs->match.tp_dst = flow->key.tp_dst;
- ofs->duration = htonl((jiffies - flow->init_time) / HZ);
- ofs->priority = htons(flow->priority);
- ofs->table_id = table_idx;
- ofs->packet_count = cpu_to_be64(flow->packet_count);
- ofs->byte_count = cpu_to_be64(flow->byte_count);
-}
+ memset(ofe->pad, 0, sizeof ofe->pad);
+ ofe->priority = htons(flow->priority);
-static int
-fill_port_stats_reply(struct datapath *dp, struct ofp_port_stats_reply *psr)
-{
- struct net_bridge_port *p;
- int port_count = 0;
-
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- struct ofp_port_stats *ps = &psr->ports[port_count++];
- struct net_device_stats *stats = p->dev->get_stats(p->dev);
- ps->port_no = htons(p->port_no);
- memset(ps->pad, 0, sizeof ps->pad);
- ps->rx_count = cpu_to_be64(stats->rx_packets);
- ps->tx_count = cpu_to_be64(stats->tx_packets);
- ps->drop_count = cpu_to_be64(stats->rx_dropped
- + stats->tx_dropped);
- }
+ duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
+ ofe->duration = htonl(duration_j / HZ);
+ ofe->packet_count = cpu_to_be64(flow->packet_count);
+ ofe->byte_count = cpu_to_be64(flow->byte_count);
- return port_count;
+ return send_openflow_skb(skb, NULL);
}
+EXPORT_SYMBOL(dp_send_flow_expired);
int
-dp_send_port_stats(struct datapath *dp, const struct sender *sender)
+dp_send_error_msg(struct datapath *dp, const struct sender *sender,
+ uint16_t type, uint16_t code, const uint8_t *data, size_t len)
{
struct sk_buff *skb;
- struct ofp_port_stats_reply *psr;
- size_t psr_len, port_max_len;
- int port_count;
+ struct ofp_error_msg *oem;
- /* Overallocate. */
- port_max_len = sizeof(struct ofp_port_stats) * OFPP_MAX;
- psr = alloc_openflow_skb(dp, sizeof *psr + port_max_len,
- OFPT_PORT_STATS_REPLY, sender, &skb);
- if (!psr)
+
+ oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR_MSG,
+ sender, &skb);
+ if (!oem)
return -ENOMEM;
- /* Fill. */
- port_count = fill_port_stats_reply(dp, psr);
+ oem->type = htons(type);
+ oem->code = htons(code);
+ memcpy(oem->data, data, len);
- /* Shrink to fit. */
- psr_len = sizeof *psr + sizeof(struct ofp_port_stats) * port_count;
- resize_openflow_skb(skb, &psr->header, psr_len);
return send_openflow_skb(skb, sender);
}
int
-dp_send_table_stats(struct datapath *dp, const struct sender *sender)
+dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
+ const struct ofp_header *rq)
{
struct sk_buff *skb;
- struct ofp_table_stats_reply *tsr;
- int i, n_tables;
-
- n_tables = dp->chain->n_tables;
- tsr = alloc_openflow_skb(dp, (offsetof(struct ofp_table_stats_reply,
- tables)
- + sizeof tsr->tables[0] * n_tables),
- OFPT_TABLE_STATS_REPLY, sender, &skb);
- if (!tsr)
+ struct ofp_header *reply;
+
+ reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
+ sender, &skb);
+ if (!reply)
return -ENOMEM;
- for (i = 0; i < n_tables; i++) {
- struct ofp_table_stats *ots = &tsr->tables[i];
- struct sw_table_stats stats;
- dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
- strncpy(ots->name, stats.name, sizeof ots->name);
- ots->table_id = i;
- ots->pad[0] = ots->pad[1] = 0;
- ots->max_entries = htonl(stats.max_flows);
- ots->active_count = htonl(stats.n_flows);
- ots->matched_count = cpu_to_be64(0); /* FIXME */
- }
+
+ memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
return send_openflow_skb(skb, sender);
}
[DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
};
-struct flow_stats_cb_state {
- int dp_idx;
+struct flow_stats_state {
int table_idx;
struct sw_table_position position;
- struct ofp_flow_stats_request *rq;
- int sent_terminator;
+ const struct ofp_flow_stats_request *rq;
- struct ofp_flow_stats *flows;
- int n_flows, max_flows;
+ void *body;
+ int bytes_used, bytes_allocated;
};
-static int muster_callback(struct sw_flow *flow, void *private)
+static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
+ void **state)
{
- struct flow_stats_cb_state *s = private;
+ const struct ofp_flow_stats_request *fsr = body;
+ struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
+ if (!s)
+ return -ENOMEM;
+ s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
+ memset(&s->position, 0, sizeof s->position);
+ s->rq = fsr;
+ *state = s;
+ return 0;
+}
+
+static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+ struct flow_stats_state *s = private;
+ struct ofp_flow_stats *ofs;
+ int actions_length;
+ int length;
+
+ actions_length = sizeof *ofs->actions * flow->n_actions;
+ length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
+ if (length + s->bytes_used > s->bytes_allocated)
+ return 1;
+
+ ofs = s->body + s->bytes_used;
+ ofs->length = htons(length);
+ ofs->table_id = s->table_idx;
+ ofs->pad = 0;
+ ofs->match.wildcards = htons(flow->key.wildcards);
+ ofs->match.in_port = flow->key.in_port;
+ memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
+ memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
+ ofs->match.dl_vlan = flow->key.dl_vlan;
+ ofs->match.dl_type = flow->key.dl_type;
+ ofs->match.nw_src = flow->key.nw_src;
+ ofs->match.nw_dst = flow->key.nw_dst;
+ ofs->match.nw_proto = flow->key.nw_proto;
+ memset(ofs->match.pad, 0, sizeof ofs->match.pad);
+ ofs->match.tp_src = flow->key.tp_src;
+ ofs->match.tp_dst = flow->key.tp_dst;
+ ofs->duration = htonl((jiffies - flow->init_time) / HZ);
+ ofs->packet_count = cpu_to_be64(flow->packet_count);
+ ofs->byte_count = cpu_to_be64(flow->byte_count);
+ ofs->priority = htons(flow->priority);
+ ofs->max_idle = htons(flow->max_idle);
+ memcpy(ofs->actions, flow->actions, actions_length);
- fill_flow_stats(&s->flows[s->n_flows], flow, s->table_idx);
- return ++s->n_flows >= s->max_flows;
+ s->bytes_used += length;
+ return 0;
}
-int
-muster_flow_stats(struct datapath *dp, struct flow_stats_cb_state *s,
- const struct sender *sender, struct sk_buff *skb)
+static int flow_stats_dump(struct datapath *dp, void *state,
+ void *body, int *body_len)
{
- struct ofp_flow_stats_reply *fsr;
- size_t header_size, flow_size;
+ struct flow_stats_state *s = state;
struct sw_flow_key match_key;
- int max_openflow_len;
- size_t size;
+ int error = 0;
- fsr = put_openflow_headers(dp, skb, OFPT_FLOW_STATS_REPLY, sender,
- &max_openflow_len);
- if (IS_ERR(fsr))
- return PTR_ERR(fsr);
- resize_openflow_skb(skb, &fsr->header, max_openflow_len);
-
- header_size = offsetof(struct ofp_flow_stats_reply, flows);
- flow_size = sizeof fsr->flows[0];
- s->max_flows = (max_openflow_len - header_size) / flow_size;
- if (s->max_flows <= 0)
- return -ENOMEM;
- s->flows = fsr->flows;
+ s->bytes_used = 0;
+ s->bytes_allocated = *body_len;
+ s->body = body;
flow_extract_match(&match_key, &s->rq->match);
- s->n_flows = 0;
while (s->table_idx < dp->chain->n_tables
&& (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
{
struct sw_table *table = dp->chain->tables[s->table_idx];
- if (table->iterate(table, &match_key, &s->position,
- muster_callback, s))
+ error = table->iterate(table, &match_key, &s->position,
+ flow_stats_dump_callback, s);
+ if (error)
break;
s->table_idx++;
memset(&s->position, 0, sizeof s->position);
}
- if (!s->n_flows) {
- /* Signal dump completion. */
- if (s->sent_terminator) {
- return 0;
- }
- s->sent_terminator = 1;
+ *body_len = s->bytes_used;
+
+ /* If error is 0, we're done.
+ * Otherwise, if some bytes were used, there are more flows to come.
+ * Otherwise, we were not able to fit even a single flow in the body,
+ * which indicates that we have a single flow with too many actions to
+ * fit. We won't ever make any progress at that rate, so give up. */
+ return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
+}
+
+static void flow_stats_done(void *state)
+{
+ kfree(state);
+}
+
+static int aggregate_stats_init(struct datapath *dp,
+ const void *body, int body_len,
+ void **state)
+{
+ *state = (void *)body;
+ return 0;
+}
+
+static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+ struct ofp_aggregate_stats_reply *rpy = private;
+ rpy->packet_count += flow->packet_count;
+ rpy->byte_count += flow->byte_count;
+ rpy->flow_count++;
+ return 0;
+}
+
+static int aggregate_stats_dump(struct datapath *dp, void *state,
+ void *body, int *body_len)
+{
+ struct ofp_aggregate_stats_request *rq = state;
+ struct ofp_aggregate_stats_reply *rpy;
+ struct sw_table_position position;
+ struct sw_flow_key match_key;
+ int table_idx;
+
+ if (*body_len < sizeof *rpy)
+ return -ENOBUFS;
+ rpy = body;
+ *body_len = sizeof *rpy;
+
+ memset(rpy, 0, sizeof *rpy);
+
+ flow_extract_match(&match_key, &rq->match);
+ table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
+ memset(&position, 0, sizeof position);
+ while (table_idx < dp->chain->n_tables
+ && (rq->table_id == 0xff || rq->table_id == table_idx))
+ {
+ struct sw_table *table = dp->chain->tables[table_idx];
+ int error;
+
+ error = table->iterate(table, &match_key, &position,
+ aggregate_stats_dump_callback, rpy);
+ if (error)
+ return error;
+
+ table_idx++;
+ memset(&position, 0, sizeof position);
+ }
+
+ rpy->packet_count = cpu_to_be64(rpy->packet_count);
+ rpy->byte_count = cpu_to_be64(rpy->byte_count);
+ rpy->flow_count = htonl(rpy->flow_count);
+ return 0;
+}
+
+static int table_stats_dump(struct datapath *dp, void *state,
+ void *body, int *body_len)
+{
+ struct ofp_table_stats *ots;
+ int nbytes = dp->chain->n_tables * sizeof *ots;
+ int i;
+ if (nbytes > *body_len)
+ return -ENOBUFS;
+ *body_len = nbytes;
+ for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
+ struct sw_table_stats stats;
+ dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
+ strncpy(ots->name, stats.name, sizeof ots->name);
+ ots->table_id = i;
+ memset(ots->pad, 0, sizeof ots->pad);
+ ots->max_entries = htonl(stats.max_flows);
+ ots->active_count = htonl(stats.n_flows);
+ ots->matched_count = cpu_to_be64(0); /* FIXME */
+ }
+ return 0;
+}
+
+struct port_stats_state {
+ int port;
+};
+
+static int port_stats_init(struct datapath *dp, const void *body, int body_len,
+ void **state)
+{
+ struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
+ if (!s)
+ return -ENOMEM;
+ s->port = 0;
+ *state = s;
+ return 0;
+}
+
+static int port_stats_dump(struct datapath *dp, void *state,
+ void *body, int *body_len)
+{
+ struct port_stats_state *s = state;
+ struct ofp_port_stats *ops;
+ int n_ports, max_ports;
+ int i;
+
+ max_ports = *body_len / sizeof *ops;
+ if (!max_ports)
+ return -ENOMEM;
+ ops = body;
+
+ n_ports = 0;
+ for (i = s->port; i < OFPP_MAX && n_ports < max_ports; i++) {
+ struct net_bridge_port *p = dp->ports[i];
+ struct net_device_stats *stats;
+ if (!p)
+ continue;
+ stats = p->dev->get_stats(p->dev);
+ ops->port_no = htons(p->port_no);
+ memset(ops->pad, 0, sizeof ops->pad);
+ ops->rx_count = cpu_to_be64(stats->rx_packets);
+ ops->tx_count = cpu_to_be64(stats->tx_packets);
+ ops->drop_count = cpu_to_be64(stats->rx_dropped
+ + stats->tx_dropped);
+ n_ports++;
+ ops++;
}
- size = header_size + flow_size * s->n_flows;
- resize_openflow_skb(skb, &fsr->header, size);
- return skb->len;
+ s->port = i;
+ *body_len = n_ports * sizeof *ops;
+ return n_ports >= max_ports;
}
+static void port_stats_done(void *state)
+{
+ kfree(state);
+}
+
+struct stats_type {
+ /* Minimum and maximum acceptable number of bytes in body member of
+ * struct ofp_stats_request. */
+ size_t min_body, max_body;
+
+ /* Prepares to dump some kind of statistics on 'dp'. 'body' and
+ * 'body_len' are the 'body' member of the struct ofp_stats_request.
+ * Returns zero if successful, otherwise a negative error code.
+ * May initialize '*state' to state information. May be null if no
+ * initialization is required.*/
+ int (*init)(struct datapath *dp, const void *body, int body_len,
+ void **state);
+
+ /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
+ * modifies '*body_len' to reflect the number of bytes actually used.
+ * ('body' will be transmitted as the 'body' member of struct
+ * ofp_stats_reply.) */
+ int (*dump)(struct datapath *dp, void *state,
+ void *body, int *body_len);
+
+ /* Cleans any state created by the init or dump functions. May be null
+ * if no cleanup is required. */
+ void (*done)(void *state);
+};
+
+static const struct stats_type stats[] = {
+ [OFPST_FLOW] = {
+ sizeof(struct ofp_flow_stats_request),
+ sizeof(struct ofp_flow_stats_request),
+ flow_stats_init,
+ flow_stats_dump,
+ flow_stats_done
+ },
+ [OFPST_AGGREGATE] = {
+ sizeof(struct ofp_aggregate_stats_request),
+ sizeof(struct ofp_aggregate_stats_request),
+ aggregate_stats_init,
+ aggregate_stats_dump,
+ NULL
+ },
+ [OFPST_TABLE] = {
+ 0,
+ 0,
+ NULL,
+ table_stats_dump,
+ NULL
+ },
+ [OFPST_PORT] = {
+ 0,
+ 0,
+ port_stats_init,
+ port_stats_dump,
+ port_stats_done
+ },
+};
+
static int
dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
{
struct datapath *dp;
struct sender sender;
- struct flow_stats_cb_state *state;
+ const struct stats_type *s;
+ struct ofp_stats_reply *osr;
+ int dp_idx;
+ int max_openflow_len, body_len;
+ void *body;
int err;
+ /* Set up the cleanup function for this dump. Linux 2.6.20 and later
+ * support setting up cleanup functions via the .doneit member of
+ * struct genl_ops. This kluge supports earlier versions also. */
+ cb->done = dp_genl_openflow_done;
+
+ rcu_read_lock();
if (!cb->args[0]) {
struct nlattr *attrs[DP_GENL_A_MAX + 1];
- struct ofp_flow_stats_request *rq;
+ struct ofp_stats_request *rq;
struct nlattr *va;
+ size_t len, body_len;
+ int type;
err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
dp_genl_openflow_policy);
if (err < 0)
return err;
+ err = -EINVAL;
+
if (!attrs[DP_GENL_A_DP_IDX])
- return -EINVAL;
+ goto out;
+ dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
+ dp = dp_get(dp_idx);
+ if (!dp) {
+ err = -ENOENT;
+ goto out;
+ }
va = attrs[DP_GENL_A_OPENFLOW];
- if (!va || nla_len(va) != sizeof *state->rq)
- return -EINVAL;
+ len = nla_len(va);
+ if (!va || len < sizeof *rq)
+ goto out;
rq = nla_data(va);
+ type = ntohs(rq->type);
if (rq->header.version != OFP_VERSION
- || rq->header.type != OFPT_FLOW_STATS_REQUEST
- || ntohs(rq->header.length) != sizeof *rq)
- return -EINVAL;
-
- state = kmalloc(sizeof *state, GFP_KERNEL);
- if (!state)
- return -ENOMEM;
- state->dp_idx = nla_get_u32(attrs[DP_GENL_A_DP_IDX]);
- state->table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
- memset(&state->position, 0, sizeof state->position);
- state->rq = rq;
- state->sent_terminator = 0;
-
- cb->args[0] = (long) state;
- } else {
- state = (struct flow_stats_cb_state *) cb->args[0];
- }
-
- if (state->rq->type != OFPFS_INDIV) {
- return -ENOTSUPP;
- }
+ || rq->header.type != OFPT_STATS_REQUEST
+ || ntohs(rq->header.length) != len
+ || type >= ARRAY_SIZE(stats)
+ || !stats[type].dump)
+ goto out;
+
+ s = &stats[type];
+ body_len = len - offsetof(struct ofp_stats_request, body);
+ if (body_len < s->min_body || body_len > s->max_body)
+ goto out;
+
+ cb->args[0] = 1;
+ cb->args[1] = dp_idx;
+ cb->args[2] = type;
+ cb->args[3] = rq->header.xid;
+ if (s->init) {
+ void *state;
+ err = s->init(dp, rq->body, body_len, &state);
+ if (err)
+ goto out;
+ cb->args[4] = (long) state;
+ }
+ } else if (cb->args[0] == 1) {
+ dp_idx = cb->args[1];
+ s = &stats[cb->args[2]];
- rcu_read_lock();
- dp = dp_get(state->dp_idx);
- if (!dp) {
- err = -ENOENT;
+ dp = dp_get(dp_idx);
+ if (!dp) {
+ err = -ENOENT;
+ goto out;
+ }
+ } else {
+ err = 0;
goto out;
}
- sender.xid = state->rq->header.xid;
+ sender.xid = cb->args[3];
sender.pid = NETLINK_CB(cb->skb).pid;
sender.seq = cb->nlh->nlmsg_seq;
- err = muster_flow_stats(dp, state, &sender, skb);
+
+ osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
+ &max_openflow_len);
+ if (IS_ERR(osr)) {
+ err = PTR_ERR(osr);
+ goto out;
+ }
+ osr->type = htons(s - stats);
+ osr->flags = 0;
+ resize_openflow_skb(skb, &osr->header, max_openflow_len);
+ body = osr->body;
+ body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
+
+ err = s->dump(dp, (void *) cb->args[4], body, &body_len);
+ if (err >= 0) {
+ if (!err)
+ cb->args[0] = 2;
+ else
+ osr->flags = ntohs(OFPSF_REPLY_MORE);
+ resize_openflow_skb(skb, &osr->header,
+ (offsetof(struct ofp_stats_reply, body)
+ + body_len));
+ err = skb->len;
+ }
out:
rcu_read_unlock();
static int
dp_genl_openflow_done(struct netlink_callback *cb)
{
- struct flow_stats_cb_state *state;
- state = (struct flow_stats_cb_state *) cb->args[0];
- kfree(state);
+ if (cb->args[0]) {
+ const struct stats_type *s = &stats[cb->args[2]];
+ if (s->done)
+ s->done((void *) cb->args[4]);
+ }
return 0;
}
.policy = dp_genl_openflow_policy,
.doit = dp_genl_openflow,
.dumpit = dp_genl_openflow_dumpit,
- .done = dp_genl_openflow_done,
-};
-
-static struct nla_policy dp_genl_benchmark_policy[DP_GENL_A_MAX + 1] = {
- [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
- [DP_GENL_A_NPACKETS] = { .type = NLA_U32 },
- [DP_GENL_A_PSIZE] = { .type = NLA_U32 },
-};
-
-static struct genl_ops dp_genl_ops_benchmark_nl = {
- .cmd = DP_GENL_C_BENCHMARK_NL,
- .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
- .policy = dp_genl_benchmark_policy,
- .doit = dp_genl_benchmark_nl,
- .dumpit = NULL,
};
static struct genl_ops *dp_genl_all_ops[] = {
&dp_genl_ops_query_dp,
&dp_genl_ops_add_port,
&dp_genl_ops_del_port,
- &dp_genl_ops_benchmark_nl,
};
static int dp_init_netlink(void)
git://git.onelab.eu / sliver-openvswitch.git / blobdiff