#include <sys/stat.h>
#include <unistd.h>
+ #include "classifier.h"
#include "csum.h"
#include "dpif.h"
#include "dpif-provider.h"
VLOG_DEFINE_THIS_MODULE(dpif_netdev);
+ /* By default, choose a priority in the middle. */
+ #define NETDEV_RULE_PRIORITY 0x8000
+
/* Configuration parameters. */
enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
int max_mtu; /* Maximum MTU of any port added so far. */
struct dp_netdev_queue queues[N_QUEUES];
+ struct classifier cls; /* Classifier. */
struct hmap flow_table; /* Flow table. */
struct seq *queue_seq; /* Incremented whenever a packet is queued. */
/* A flow in dp_netdev's 'flow_table'. */
struct dp_netdev_flow {
- struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
- struct flow key;
+ /* Packet classification. */
+ struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
+
+ /* Hash table index by unmasked flow.*/
+ struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
+ struct flow flow; /* The flow that created this entry. */
/* Statistics. */
long long int used; /* Last used time, in monotonic msecs. */
static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
int queue_no, const struct flow *,
const struct nlattr *userdata);
- static void dp_netdev_execute_actions(struct dp_netdev *,
- struct ofpbuf *, struct flow *,
+ static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
+ struct ofpbuf *,
const struct nlattr *actions,
size_t actions_len);
static void dp_netdev_port_input(struct dp_netdev *dp,
return class != &dpif_netdev_class;
}
+static bool
+dpif_netdev_class_is_planetlab(const struct dpif_class *class)
+{
+ return class == &dpif_planetlab_class;
+}
+
static const char *
dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
{
return strcmp(type, "internal") ? type
+ : dpif_netdev_class_is_planetlab(class) ? "pltap"
: dpif_netdev_class_is_dummy(class) ? "dummy"
: "tap";
}
{
uint32_t port_no;
- if (dp->class != &dpif_netdev_class) {
+ if (dp->class != &dpif_netdev_class &&
+ dp->class != &dpif_planetlab_class) {
const char *p;
int start_no = 0;
dp->queues[i].head = dp->queues[i].tail = 0;
}
dp->queue_seq = seq_create();
+ classifier_init(&dp->cls, NULL);
hmap_init(&dp->flow_table);
list_init(&dp->port_list);
dp->port_seq = seq_create();
}
dp_netdev_purge_queues(dp);
seq_destroy(dp->queue_seq);
+ classifier_destroy(&dp->cls);
hmap_destroy(&dp->flow_table);
seq_destroy(dp->port_seq);
free(dp->name);
static void
dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
{
+ ovs_rwlock_wrlock(&dp->cls.rwlock);
+ classifier_remove(&dp->cls, &netdev_flow->cr);
+ ovs_rwlock_unlock(&dp->cls.rwlock);
+ cls_rule_destroy(&netdev_flow->cr);
+
hmap_remove(&dp->flow_table, &netdev_flow->node);
free(netdev_flow->actions);
free(netdev_flow);
}
static struct dp_netdev_flow *
- dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
+ dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
+ {
+ struct cls_rule *cr;
+
+ ovs_rwlock_wrlock(&dp->cls.rwlock);
+ cr = classifier_lookup(&dp->cls, flow, NULL);
+ ovs_rwlock_unlock(&dp->cls.rwlock);
+
+ return (cr
+ ? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
+ : NULL);
+ }
+
+ static struct dp_netdev_flow *
+ dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
{
struct dp_netdev_flow *netdev_flow;
- HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(key, 0),
+ HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
&dp->flow_table) {
- if (flow_equal(&netdev_flow->key, key)) {
+ if (flow_equal(&netdev_flow->flow, flow)) {
return netdev_flow;
}
}
}
static int
- dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
- struct flow *flow)
+ dpif_netdev_flow_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ const struct nlattr *mask_key,
+ uint32_t mask_key_len, struct flow *flow,
+ struct flow *mask)
{
odp_port_t in_port;
- if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
+ if (odp_flow_key_to_flow(key, key_len, flow)
+ || (mask_key
+ && odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow))) {
/* This should not happen: it indicates that odp_flow_key_from_flow()
- * and odp_flow_key_to_flow() disagree on the acceptable form of a
- * flow. Log the problem as an error, with enough details to enable
- * debugging. */
+ * and odp_flow_key_to_flow() disagree on the acceptable form of a flow
+ * or odp_flow_key_from_mask() and odp_flow_key_to_mask() disagree on
+ * the acceptable form of a mask. Log the problem as an error, with
+ * enough details to enable debugging. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
if (!VLOG_DROP_ERR(&rl)) {
struct ds s;
ds_init(&s);
- odp_flow_key_format(key, key_len, &s);
+ odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
+ true);
VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
ds_destroy(&s);
}
return EINVAL;
}
+ if (mask_key) {
+ /* Force unwildcard the in_port. */
+ mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
+ }
+
in_port = flow->in_port.odp_port;
if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
return EINVAL;
return 0;
}
+ static int
+ dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ struct flow *flow)
+ {
+ return dpif_netdev_flow_mask_from_nlattrs(key, key_len, NULL, 0, flow,
+ NULL);
+ }
+
static int
dpif_netdev_flow_get(const struct dpif *dpif,
const struct nlattr *nl_key, size_t nl_key_len,
}
ovs_mutex_lock(&dp_netdev_mutex);
- netdev_flow = dp_netdev_lookup_flow(dp, &key);
+ netdev_flow = dp_netdev_find_flow(dp, &key);
if (netdev_flow) {
if (stats) {
get_dpif_flow_stats(netdev_flow, stats);
}
static int
- dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
- const struct nlattr *actions, size_t actions_len)
+ dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
+ const struct flow_wildcards *wc,
+ const struct nlattr *actions,
+ size_t actions_len)
{
struct dp_netdev_flow *netdev_flow;
+ struct match match;
int error;
netdev_flow = xzalloc(sizeof *netdev_flow);
- netdev_flow->key = *key;
+ netdev_flow->flow = *flow;
+
+ match_init(&match, flow, wc);
+ cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
+ ovs_rwlock_wrlock(&dp->cls.rwlock);
+ classifier_insert(&dp->cls, &netdev_flow->cr);
+ ovs_rwlock_unlock(&dp->cls.rwlock);
error = set_flow_actions(netdev_flow, actions, actions_len);
if (error) {
+ ovs_rwlock_wrlock(&dp->cls.rwlock);
+ classifier_remove(&dp->cls, &netdev_flow->cr);
+ ovs_rwlock_unlock(&dp->cls.rwlock);
+ cls_rule_destroy(&netdev_flow->cr);
+
free(netdev_flow);
return error;
}
- hmap_insert(&dp->flow_table, &netdev_flow->node,
- flow_hash(&netdev_flow->key, 0));
+ hmap_insert(&dp->flow_table, &netdev_flow->node, flow_hash(flow, 0));
return 0;
}
{
struct dp_netdev *dp = get_dp_netdev(dpif);
struct dp_netdev_flow *netdev_flow;
- struct flow key;
+ struct flow flow;
+ struct flow_wildcards wc;
int error;
- error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
+ error = dpif_netdev_flow_mask_from_nlattrs(put->key, put->key_len,
+ put->mask, put->mask_len, &flow, &wc.masks);
if (error) {
return error;
}
ovs_mutex_lock(&dp_netdev_mutex);
- netdev_flow = dp_netdev_lookup_flow(dp, &key);
+ netdev_flow = dp_netdev_lookup_flow(dp, &flow);
if (!netdev_flow) {
if (put->flags & DPIF_FP_CREATE) {
if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
if (put->stats) {
memset(put->stats, 0, sizeof *put->stats);
}
- error = dp_netdev_flow_add(dp, &key, put->actions,
+ error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
put->actions_len);
} else {
error = EFBIG;
error = ENOENT;
}
} else {
- if (put->flags & DPIF_FP_MODIFY) {
+ if (put->flags & DPIF_FP_MODIFY
+ && flow_equal(&flow, &netdev_flow->flow)) {
error = set_flow_actions(netdev_flow, put->actions,
put->actions_len);
if (!error) {
clear_stats(netdev_flow);
}
}
- } else {
+ } else if (put->flags & DPIF_FP_CREATE) {
error = EEXIST;
+ } else {
+ /* Overlapping flow. */
+ error = EINVAL;
}
}
ovs_mutex_unlock(&dp_netdev_mutex);
}
ovs_mutex_lock(&dp_netdev_mutex);
- netdev_flow = dp_netdev_lookup_flow(dp, &key);
+ netdev_flow = dp_netdev_find_flow(dp, &key);
if (netdev_flow) {
if (del->stats) {
get_dpif_flow_stats(netdev_flow, del->stats);
uint32_t offset;
struct nlattr *actions;
struct odputil_keybuf keybuf;
+ struct odputil_keybuf maskbuf;
struct dpif_flow_stats stats;
};
struct ofpbuf buf;
ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
- odp_flow_key_from_flow(&buf, &netdev_flow->key,
- netdev_flow->key.in_port.odp_port);
+ odp_flow_key_from_flow(&buf, &netdev_flow->flow,
+ netdev_flow->flow.in_port.odp_port);
*key = buf.data;
*key_len = buf.size;
}
- if (mask) {
- *mask = NULL;
- *mask_len = 0;
+ if (key && mask) {
+ struct ofpbuf buf;
+ struct flow_wildcards wc;
+
+ ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
+ minimask_expand(&netdev_flow->cr.match.mask, &wc);
+ odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
+ odp_to_u32(wc.masks.in_port.odp_port));
+
+ *mask = buf.data;
+ *mask_len = buf.size;
}
if (actions) {
dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- struct ofpbuf copy;
- struct flow key;
+ struct flow md;
int error;
if (execute->packet->size < ETH_HEADER_LEN ||
return EINVAL;
}
- /* Make a deep copy of 'packet', because we might modify its data. */
- ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
- ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
- ofpbuf_put(©, execute->packet->data, execute->packet->size);
-
- flow_extract(©, 0, 0, NULL, NULL, &key);
- error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
- &key);
+ /* Get packet metadata. */
+ error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, &md);
if (!error) {
+ struct ofpbuf *copy;
+ struct flow key;
+
+ /* Make a deep copy of 'packet', because we might modify its data. */
+ copy = ofpbuf_clone_with_headroom(execute->packet, DP_NETDEV_HEADROOM);
+
+ /* Extract flow key. */
+ flow_extract(copy, md.skb_priority, md.pkt_mark, &md.tunnel,
+ &md.in_port, &key);
ovs_mutex_lock(&dp_netdev_mutex);
- dp_netdev_execute_actions(dp, ©, &key,
+ dp_netdev_execute_actions(dp, &key, copy,
execute->actions, execute->actions_len);
ovs_mutex_unlock(&dp_netdev_mutex);
+ ofpbuf_delete(copy);
}
-
- ofpbuf_uninit(©);
return error;
}
netdev_flow->used = time_msec();
netdev_flow->packet_count++;
netdev_flow->byte_count += packet->size;
- netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->key);
+ netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
}
static void
netdev_flow = dp_netdev_lookup_flow(dp, &key);
if (netdev_flow) {
dp_netdev_flow_used(netdev_flow, packet);
- dp_netdev_execute_actions(dp, packet, &key,
+ dp_netdev_execute_actions(dp, &key, packet,
netdev_flow->actions,
netdev_flow->actions_len);
dp->n_hit++;
ovs_mutex_unlock(&dp_netdev_mutex);
}
- static void
- dp_netdev_output_port(void *dp_, struct ofpbuf *packet,
- const struct flow *flow OVS_UNUSED, odp_port_t out_port)
- {
- struct dp_netdev *dp = dp_;
- struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
- if (p) {
- netdev_send(p->netdev, packet);
- }
- }
-
static int
dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
int queue_no, const struct flow *flow,
}
}
+ struct dp_netdev_execute_aux {
+ struct dp_netdev *dp;
+ const struct flow *key;
+ };
+
static void
- dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
- const struct flow *key,
+ dp_netdev_action_output(void *aux_, struct ofpbuf *packet,
+ const struct flow *flow OVS_UNUSED,
+ odp_port_t out_port)
+ {
+ struct dp_netdev_execute_aux *aux = aux_;
+ struct dp_netdev_port *p = aux->dp->ports[odp_to_u32(out_port)];
+ if (p) {
+ netdev_send(p->netdev, packet);
+ }
+ }
+
+ static void
+ dp_netdev_action_userspace(void *aux_, struct ofpbuf *packet,
+ const struct flow *flow OVS_UNUSED,
const struct nlattr *a)
{
+ struct dp_netdev_execute_aux *aux = aux_;
const struct nlattr *userdata;
userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
- dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
+ dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
+ userdata);
}
static void
- dp_netdev_execute_actions(struct dp_netdev *dp,
- struct ofpbuf *packet, struct flow *key,
- const struct nlattr *actions,
- size_t actions_len)
+ dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
+ struct ofpbuf *packet,
+ const struct nlattr *actions, size_t actions_len)
{
- odp_execute_actions(dp, packet, key, actions, actions_len,
- dp_netdev_output_port, dp_netdev_action_userspace);
+ struct dp_netdev_execute_aux aux = {dp, key};
+ struct flow md = *key; /* Packet metadata, may be modified by actions. */
+
+ odp_execute_actions(&aux, packet, &md, actions, actions_len,
+ dp_netdev_action_output, dp_netdev_action_userspace);
}
+#define DPIF_NETDEV_CLASS_FUNCTIONS \
+ dpif_netdev_enumerate, \
+ dpif_netdev_port_open_type, \
+ dpif_netdev_open, \
+ dpif_netdev_close, \
+ dpif_netdev_destroy, \
+ dpif_netdev_run, \
+ dpif_netdev_wait, \
+ dpif_netdev_get_stats, \
+ dpif_netdev_port_add, \
+ dpif_netdev_port_del, \
+ dpif_netdev_port_query_by_number, \
+ dpif_netdev_port_query_by_name, \
+ dpif_netdev_get_max_ports, \
+ NULL, /* port_get_pid */ \
+ dpif_netdev_port_dump_start, \
+ dpif_netdev_port_dump_next, \
+ dpif_netdev_port_dump_done, \
+ dpif_netdev_port_poll, \
+ dpif_netdev_port_poll_wait, \
+ dpif_netdev_flow_get, \
+ dpif_netdev_flow_put, \
+ dpif_netdev_flow_del, \
+ dpif_netdev_flow_flush, \
+ dpif_netdev_flow_dump_start, \
+ dpif_netdev_flow_dump_next, \
+ dpif_netdev_flow_dump_done, \
+ dpif_netdev_execute, \
+ NULL, /* operate */ \
+ dpif_netdev_recv_set, \
+ dpif_netdev_queue_to_priority, \
+ dpif_netdev_recv, \
+ dpif_netdev_recv_wait, \
+ dpif_netdev_recv_purge, \
+
const struct dpif_class dpif_netdev_class = {
"netdev",
- dpif_netdev_enumerate,
- dpif_netdev_port_open_type,
- dpif_netdev_open,
- dpif_netdev_close,
- dpif_netdev_destroy,
- dpif_netdev_run,
- dpif_netdev_wait,
- dpif_netdev_get_stats,
- dpif_netdev_port_add,
- dpif_netdev_port_del,
- dpif_netdev_port_query_by_number,
- dpif_netdev_port_query_by_name,
- dpif_netdev_get_max_ports,
- NULL, /* port_get_pid */
- dpif_netdev_port_dump_start,
- dpif_netdev_port_dump_next,
- dpif_netdev_port_dump_done,
- dpif_netdev_port_poll,
- dpif_netdev_port_poll_wait,
- dpif_netdev_flow_get,
- dpif_netdev_flow_put,
- dpif_netdev_flow_del,
- dpif_netdev_flow_flush,
- dpif_netdev_flow_dump_start,
- dpif_netdev_flow_dump_next,
- dpif_netdev_flow_dump_done,
- dpif_netdev_execute,
- NULL, /* operate */
- dpif_netdev_recv_set,
- dpif_netdev_queue_to_priority,
- dpif_netdev_recv,
- dpif_netdev_recv_wait,
- dpif_netdev_recv_purge,
+ DPIF_NETDEV_CLASS_FUNCTIONS
+};
+
+const struct dpif_class dpif_planetlab_class = {
+ "planetlab",
+ DPIF_NETDEV_CLASS_FUNCTIONS
};
static void
"DP PORT NEW-NUMBER",
3, 3, dpif_dummy_change_port_number, NULL);
}
+
git://git.onelab.eu / sliver-openvswitch.git / commitdiff