2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
24 #include <netinet/in.h>
25 #include <sys/socket.h>
30 #include <sys/ioctl.h>
36 #include "dpif-provider.h"
38 #include "dynamic-string.h"
45 #include "ofp-print.h"
48 #include "poll-loop.h"
56 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
58 /* Configuration parameters. */
59 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
60 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
62 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
63 * headers to be aligned on a 4-byte boundary. */
64 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
67 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
68 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
69 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
70 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
72 struct dp_netdev_upcall {
73 struct dpif_upcall upcall; /* Queued upcall information. */
74 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
77 struct dp_netdev_queue {
78 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
79 unsigned int head, tail;
82 /* Datapath based on the network device interface from netdev.h. */
84 const struct dpif_class *class;
89 struct dp_netdev_queue queues[N_QUEUES];
90 struct hmap flow_table; /* Flow table. */
93 long long int n_hit; /* Number of flow table matches. */
94 long long int n_missed; /* Number of flow table misses. */
95 long long int n_lost; /* Number of misses not passed to client. */
98 struct dp_netdev_port *ports[MAX_PORTS];
99 struct list port_list;
103 /* A port in a netdev-based datapath. */
104 struct dp_netdev_port {
105 int port_no; /* Index into dp_netdev's 'ports'. */
106 struct list node; /* Element in dp_netdev's 'port_list'. */
107 struct netdev *netdev;
108 char *type; /* Port type as requested by user. */
111 /* A flow in dp_netdev's 'flow_table'. */
112 struct dp_netdev_flow {
113 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
117 long long int used; /* Last used time, in monotonic msecs. */
118 long long int packet_count; /* Number of packets matched. */
119 long long int byte_count; /* Number of bytes matched. */
120 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
123 struct nlattr *actions;
127 /* Interface to netdev-based datapath. */
130 struct dp_netdev *dp;
131 unsigned int dp_serial;
134 /* All netdev-based datapaths. */
135 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
137 /* Maximum port MTU seen so far. */
138 static int max_mtu = ETH_PAYLOAD_MAX;
140 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
141 struct dp_netdev_port **portp);
142 static int get_port_by_name(struct dp_netdev *, const char *devname,
143 struct dp_netdev_port **portp);
144 static void dp_netdev_free(struct dp_netdev *);
145 static void dp_netdev_flow_flush(struct dp_netdev *);
146 static int do_add_port(struct dp_netdev *, const char *devname,
147 const char *type, uint32_t port_no);
148 static int do_del_port(struct dp_netdev *, uint32_t port_no);
149 static int dpif_netdev_open(const struct dpif_class *, const char *name,
150 bool create, struct dpif **);
151 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
152 int queue_no, const struct flow *,
154 static void dp_netdev_execute_actions(struct dp_netdev *,
155 struct ofpbuf *, struct flow *,
156 const struct nlattr *actions,
159 static struct dpif_netdev *
160 dpif_netdev_cast(const struct dpif *dpif)
162 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
163 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
166 static struct dp_netdev *
167 get_dp_netdev(const struct dpif *dpif)
169 return dpif_netdev_cast(dpif)->dp;
173 dpif_netdev_enumerate(struct sset *all_dps)
175 struct shash_node *node;
177 SHASH_FOR_EACH(node, &dp_netdevs) {
178 sset_add(all_dps, node->name);
184 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
186 return strcmp(type, "internal") ? type
187 : class == &dpif_netdev_class ? "tap"
188 : class == &dpif_planetlab_class ? "pltap"
193 create_dpif_netdev(struct dp_netdev *dp)
195 uint16_t netflow_id = hash_string(dp->name, 0);
196 struct dpif_netdev *dpif;
200 dpif = xmalloc(sizeof *dpif);
201 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
203 dpif->dp_serial = dp->serial;
209 choose_port(struct dp_netdev *dp, const char *name)
213 if (dp->class != &dpif_netdev_class &&
214 dp->class != &dpif_planetlab_class) {
218 /* If the port name begins with "br", start the number search at
219 * 100 to make writing tests easier. */
220 if (!strncmp(name, "br", 2)) {
224 /* If the port name contains a number, try to assign that port number.
225 * This can make writing unit tests easier because port numbers are
227 for (p = name; *p != '\0'; p++) {
228 if (isdigit((unsigned char) *p)) {
229 port_no = start_no + strtol(p, NULL, 10);
230 if (port_no > 0 && port_no < MAX_PORTS
231 && !dp->ports[port_no]) {
239 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
240 if (!dp->ports[port_no]) {
249 create_dp_netdev(const char *name, const struct dpif_class *class,
250 struct dp_netdev **dpp)
252 struct dp_netdev *dp;
256 dp = xzalloc(sizeof *dp);
258 dp->name = xstrdup(name);
260 for (i = 0; i < N_QUEUES; i++) {
261 dp->queues[i].head = dp->queues[i].tail = 0;
263 hmap_init(&dp->flow_table);
264 list_init(&dp->port_list);
266 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
272 shash_add(&dp_netdevs, name, dp);
279 dpif_netdev_open(const struct dpif_class *class, const char *name,
280 bool create, struct dpif **dpifp)
282 struct dp_netdev *dp;
284 dp = shash_find_data(&dp_netdevs, name);
289 int error = create_dp_netdev(name, class, &dp);
293 ovs_assert(dp != NULL);
296 if (dp->class != class) {
303 *dpifp = create_dpif_netdev(dp);
308 dp_netdev_purge_queues(struct dp_netdev *dp)
312 for (i = 0; i < N_QUEUES; i++) {
313 struct dp_netdev_queue *q = &dp->queues[i];
315 while (q->tail != q->head) {
316 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
317 ofpbuf_uninit(&u->buf);
323 dp_netdev_free(struct dp_netdev *dp)
325 struct dp_netdev_port *port, *next;
327 dp_netdev_flow_flush(dp);
328 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
329 do_del_port(dp, port->port_no);
331 dp_netdev_purge_queues(dp);
332 hmap_destroy(&dp->flow_table);
338 dpif_netdev_close(struct dpif *dpif)
340 struct dp_netdev *dp = get_dp_netdev(dpif);
341 ovs_assert(dp->open_cnt > 0);
342 if (--dp->open_cnt == 0 && dp->destroyed) {
343 shash_find_and_delete(&dp_netdevs, dp->name);
350 dpif_netdev_destroy(struct dpif *dpif)
352 struct dp_netdev *dp = get_dp_netdev(dpif);
353 dp->destroyed = true;
358 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
360 struct dp_netdev *dp = get_dp_netdev(dpif);
361 stats->n_flows = hmap_count(&dp->flow_table);
362 stats->n_hit = dp->n_hit;
363 stats->n_missed = dp->n_missed;
364 stats->n_lost = dp->n_lost;
369 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
372 struct dp_netdev_port *port;
373 struct netdev *netdev;
374 const char *open_type;
378 /* XXX reject devices already in some dp_netdev. */
380 /* Open and validate network device. */
381 open_type = dpif_netdev_port_open_type(dp->class, type);
382 error = netdev_open(devname, open_type, &netdev);
386 /* XXX reject loopback devices */
387 /* XXX reject non-Ethernet devices */
389 error = netdev_listen(netdev);
391 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
392 devname, strerror(errno));
393 netdev_close(netdev);
397 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
399 netdev_close(netdev);
403 port = xmalloc(sizeof *port);
404 port->port_no = port_no;
405 port->netdev = netdev;
406 port->type = xstrdup(type);
408 error = netdev_get_mtu(netdev, &mtu);
413 list_push_back(&dp->port_list, &port->node);
414 dp->ports[port_no] = port;
421 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
424 struct dp_netdev *dp = get_dp_netdev(dpif);
427 if (*port_nop != UINT32_MAX) {
428 if (*port_nop >= MAX_PORTS) {
430 } else if (dp->ports[*port_nop]) {
435 port_no = choose_port(dp, netdev_get_name(netdev));
439 return do_add_port(dp, netdev_get_name(netdev),
440 netdev_get_type(netdev), port_no);
446 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
448 struct dp_netdev *dp = get_dp_netdev(dpif);
449 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
453 is_valid_port_number(uint32_t port_no)
455 return port_no < MAX_PORTS;
459 get_port_by_number(struct dp_netdev *dp,
460 uint32_t port_no, struct dp_netdev_port **portp)
462 if (!is_valid_port_number(port_no)) {
466 *portp = dp->ports[port_no];
467 return *portp ? 0 : ENOENT;
472 get_port_by_name(struct dp_netdev *dp,
473 const char *devname, struct dp_netdev_port **portp)
475 struct dp_netdev_port *port;
477 LIST_FOR_EACH (port, node, &dp->port_list) {
478 if (!strcmp(netdev_get_name(port->netdev), devname)) {
487 do_del_port(struct dp_netdev *dp, uint32_t port_no)
489 struct dp_netdev_port *port;
493 error = get_port_by_number(dp, port_no, &port);
498 list_remove(&port->node);
499 dp->ports[port->port_no] = NULL;
502 name = xstrdup(netdev_get_name(port->netdev));
503 netdev_close(port->netdev);
513 answer_port_query(const struct dp_netdev_port *port,
514 struct dpif_port *dpif_port)
516 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
517 dpif_port->type = xstrdup(port->type);
518 dpif_port->port_no = port->port_no;
522 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
523 struct dpif_port *dpif_port)
525 struct dp_netdev *dp = get_dp_netdev(dpif);
526 struct dp_netdev_port *port;
529 error = get_port_by_number(dp, port_no, &port);
530 if (!error && dpif_port) {
531 answer_port_query(port, dpif_port);
537 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
538 struct dpif_port *dpif_port)
540 struct dp_netdev *dp = get_dp_netdev(dpif);
541 struct dp_netdev_port *port;
544 error = get_port_by_name(dp, devname, &port);
545 if (!error && dpif_port) {
546 answer_port_query(port, dpif_port);
552 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
558 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
560 hmap_remove(&dp->flow_table, &flow->node);
566 dp_netdev_flow_flush(struct dp_netdev *dp)
568 struct dp_netdev_flow *flow, *next;
570 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
571 dp_netdev_free_flow(dp, flow);
576 dpif_netdev_flow_flush(struct dpif *dpif)
578 struct dp_netdev *dp = get_dp_netdev(dpif);
579 dp_netdev_flow_flush(dp);
583 struct dp_netdev_port_state {
589 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
591 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
596 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
597 struct dpif_port *dpif_port)
599 struct dp_netdev_port_state *state = state_;
600 struct dp_netdev *dp = get_dp_netdev(dpif);
603 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
604 struct dp_netdev_port *port = dp->ports[port_no];
607 state->name = xstrdup(netdev_get_name(port->netdev));
608 dpif_port->name = state->name;
609 dpif_port->type = port->type;
610 dpif_port->port_no = port->port_no;
611 state->port_no = port_no + 1;
619 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
621 struct dp_netdev_port_state *state = state_;
628 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
630 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
631 if (dpif->dp_serial != dpif->dp->serial) {
632 dpif->dp_serial = dpif->dp->serial;
640 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
642 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
643 if (dpif->dp_serial != dpif->dp->serial) {
644 poll_immediate_wake();
648 static struct dp_netdev_flow *
649 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
651 struct dp_netdev_flow *flow;
653 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
654 if (flow_equal(&flow->key, key)) {
662 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
664 stats->n_packets = flow->packet_count;
665 stats->n_bytes = flow->byte_count;
666 stats->used = flow->used;
667 stats->tcp_flags = flow->tcp_flags;
671 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
674 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
675 /* This should not happen: it indicates that odp_flow_key_from_flow()
676 * and odp_flow_key_to_flow() disagree on the acceptable form of a
677 * flow. Log the problem as an error, with enough details to enable
679 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
681 if (!VLOG_DROP_ERR(&rl)) {
685 odp_flow_key_format(key, key_len, &s);
686 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
693 if (flow->in_port < OFPP_MAX
694 ? flow->in_port >= MAX_PORTS
695 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
703 dpif_netdev_flow_get(const struct dpif *dpif,
704 const struct nlattr *nl_key, size_t nl_key_len,
705 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
707 struct dp_netdev *dp = get_dp_netdev(dpif);
708 struct dp_netdev_flow *flow;
712 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
717 flow = dp_netdev_lookup_flow(dp, &key);
723 get_dpif_flow_stats(flow, stats);
726 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
732 set_flow_actions(struct dp_netdev_flow *flow,
733 const struct nlattr *actions, size_t actions_len)
735 flow->actions = xrealloc(flow->actions, actions_len);
736 flow->actions_len = actions_len;
737 memcpy(flow->actions, actions, actions_len);
742 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
743 const struct nlattr *actions, size_t actions_len)
745 struct dp_netdev_flow *flow;
748 flow = xzalloc(sizeof *flow);
751 error = set_flow_actions(flow, actions, actions_len);
757 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
762 clear_stats(struct dp_netdev_flow *flow)
765 flow->packet_count = 0;
766 flow->byte_count = 0;
771 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
773 struct dp_netdev *dp = get_dp_netdev(dpif);
774 struct dp_netdev_flow *flow;
778 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
783 flow = dp_netdev_lookup_flow(dp, &key);
785 if (put->flags & DPIF_FP_CREATE) {
786 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
788 memset(put->stats, 0, sizeof *put->stats);
790 return dp_netdev_flow_add(dp, &key, put->actions,
799 if (put->flags & DPIF_FP_MODIFY) {
800 int error = set_flow_actions(flow, put->actions, put->actions_len);
803 get_dpif_flow_stats(flow, put->stats);
805 if (put->flags & DPIF_FP_ZERO_STATS) {
817 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
819 struct dp_netdev *dp = get_dp_netdev(dpif);
820 struct dp_netdev_flow *flow;
824 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
829 flow = dp_netdev_lookup_flow(dp, &key);
832 get_dpif_flow_stats(flow, del->stats);
834 dp_netdev_free_flow(dp, flow);
841 struct dp_netdev_flow_state {
844 struct nlattr *actions;
845 struct odputil_keybuf keybuf;
846 struct dpif_flow_stats stats;
850 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
852 struct dp_netdev_flow_state *state;
854 *statep = state = xmalloc(sizeof *state);
857 state->actions = NULL;
862 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
863 const struct nlattr **key, size_t *key_len,
864 const struct nlattr **actions, size_t *actions_len,
865 const struct dpif_flow_stats **stats)
867 struct dp_netdev_flow_state *state = state_;
868 struct dp_netdev *dp = get_dp_netdev(dpif);
869 struct dp_netdev_flow *flow;
870 struct hmap_node *node;
872 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
877 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
882 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
883 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
890 free(state->actions);
891 state->actions = xmemdup(flow->actions, flow->actions_len);
893 *actions = state->actions;
894 *actions_len = flow->actions_len;
898 get_dpif_flow_stats(flow, &state->stats);
899 *stats = &state->stats;
906 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
908 struct dp_netdev_flow_state *state = state_;
910 free(state->actions);
916 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
918 struct dp_netdev *dp = get_dp_netdev(dpif);
923 if (execute->packet->size < ETH_HEADER_LEN ||
924 execute->packet->size > UINT16_MAX) {
928 /* Make a deep copy of 'packet', because we might modify its data. */
929 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
930 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
931 ofpbuf_put(©, execute->packet->data, execute->packet->size);
933 flow_extract(©, 0, 0, NULL, -1, &key);
934 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
937 dp_netdev_execute_actions(dp, ©, &key,
938 execute->actions, execute->actions_len);
941 ofpbuf_uninit(©);
946 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
952 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
953 uint32_t queue_id, uint32_t *priority)
955 *priority = queue_id;
959 static struct dp_netdev_queue *
960 find_nonempty_queue(struct dpif *dpif)
962 struct dp_netdev *dp = get_dp_netdev(dpif);
965 for (i = 0; i < N_QUEUES; i++) {
966 struct dp_netdev_queue *q = &dp->queues[i];
967 if (q->head != q->tail) {
975 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
978 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
980 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
983 upcall->packet = buf;
995 dpif_netdev_recv_wait(struct dpif *dpif)
997 if (find_nonempty_queue(dpif)) {
998 poll_immediate_wake();
1000 /* No messages ready to be received, and dp_wait() will ensure that we
1001 * wake up to queue new messages, so there is nothing to do. */
1006 dpif_netdev_recv_purge(struct dpif *dpif)
1008 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1009 dp_netdev_purge_queues(dpif_netdev->dp);
1013 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1015 flow->used = time_msec();
1016 flow->packet_count++;
1017 flow->byte_count += packet->size;
1018 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1022 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1023 struct ofpbuf *packet)
1025 struct dp_netdev_flow *flow;
1028 if (packet->size < ETH_HEADER_LEN) {
1031 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1032 flow = dp_netdev_lookup_flow(dp, &key);
1034 dp_netdev_flow_used(flow, packet);
1035 dp_netdev_execute_actions(dp, packet, &key,
1036 flow->actions, flow->actions_len);
1040 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1045 dpif_netdev_run(struct dpif *dpif)
1047 struct dp_netdev *dp = get_dp_netdev(dpif);
1048 struct dp_netdev_port *port;
1049 struct ofpbuf packet;
1051 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1053 LIST_FOR_EACH (port, node, &dp->port_list) {
1056 /* Reset packet contents. */
1057 ofpbuf_clear(&packet);
1058 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1060 error = netdev_recv(port->netdev, &packet);
1062 dp_netdev_port_input(dp, port, &packet);
1063 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1064 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1065 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1066 netdev_get_name(port->netdev), strerror(error));
1069 ofpbuf_uninit(&packet);
1073 dpif_netdev_wait(struct dpif *dpif)
1075 struct dp_netdev *dp = get_dp_netdev(dpif);
1076 struct dp_netdev_port *port;
1078 LIST_FOR_EACH (port, node, &dp->port_list) {
1079 netdev_recv_wait(port->netdev);
1084 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1086 struct eth_header *eh = packet->l2;
1088 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1089 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1093 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1096 struct dp_netdev_port *p = dp->ports[out_port];
1098 netdev_send(p->netdev, packet);
1103 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1104 int queue_no, const struct flow *flow, uint64_t arg)
1106 struct dp_netdev_queue *q = &dp->queues[queue_no];
1107 struct dp_netdev_upcall *u;
1108 struct dpif_upcall *upcall;
1112 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1117 u = &q->upcalls[q->head++ & QUEUE_MASK];
1120 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1121 odp_flow_key_from_flow(buf, flow, flow->in_port);
1122 key_len = buf->size;
1123 ofpbuf_pull(buf, key_len);
1124 ofpbuf_reserve(buf, 2);
1125 ofpbuf_put(buf, packet->data, packet->size);
1127 upcall = &u->upcall;
1128 upcall->type = queue_no;
1129 upcall->packet = buf;
1130 upcall->key = buf->base;
1131 upcall->key_len = key_len;
1132 upcall->userdata = arg;
1138 dp_netdev_sample(struct dp_netdev *dp,
1139 struct ofpbuf *packet, struct flow *key,
1140 const struct nlattr *action)
1142 const struct nlattr *subactions = NULL;
1143 const struct nlattr *a;
1146 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1147 int type = nl_attr_type(a);
1149 switch ((enum ovs_sample_attr) type) {
1150 case OVS_SAMPLE_ATTR_PROBABILITY:
1151 if (random_uint32() >= nl_attr_get_u32(a)) {
1156 case OVS_SAMPLE_ATTR_ACTIONS:
1160 case OVS_SAMPLE_ATTR_UNSPEC:
1161 case __OVS_SAMPLE_ATTR_MAX:
1167 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1168 nl_attr_get_size(subactions));
1172 dp_netdev_action_userspace(struct dp_netdev *dp,
1173 struct ofpbuf *packet, struct flow *key,
1174 const struct nlattr *a)
1176 const struct nlattr *userdata_attr;
1179 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1180 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1181 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1185 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1187 enum ovs_key_attr type = nl_attr_type(a);
1188 const struct ovs_key_ipv4 *ipv4_key;
1189 const struct ovs_key_ipv6 *ipv6_key;
1190 const struct ovs_key_tcp *tcp_key;
1191 const struct ovs_key_udp *udp_key;
1194 case OVS_KEY_ATTR_TUN_ID:
1195 case OVS_KEY_ATTR_PRIORITY:
1196 case OVS_KEY_ATTR_SKB_MARK:
1197 case OVS_KEY_ATTR_TUNNEL:
1198 /* not implemented */
1201 case OVS_KEY_ATTR_ETHERNET:
1202 dp_netdev_set_dl(packet,
1203 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1206 case OVS_KEY_ATTR_IPV4:
1207 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1208 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1209 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1212 case OVS_KEY_ATTR_IPV6:
1213 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1214 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1215 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1216 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1219 case OVS_KEY_ATTR_TCP:
1220 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1221 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1224 case OVS_KEY_ATTR_UDP:
1225 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1226 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1229 case OVS_KEY_ATTR_UNSPEC:
1230 case OVS_KEY_ATTR_ENCAP:
1231 case OVS_KEY_ATTR_ETHERTYPE:
1232 case OVS_KEY_ATTR_IN_PORT:
1233 case OVS_KEY_ATTR_VLAN:
1234 case OVS_KEY_ATTR_ICMP:
1235 case OVS_KEY_ATTR_ICMPV6:
1236 case OVS_KEY_ATTR_ARP:
1237 case OVS_KEY_ATTR_ND:
1238 case __OVS_KEY_ATTR_MAX:
1245 dp_netdev_execute_actions(struct dp_netdev *dp,
1246 struct ofpbuf *packet, struct flow *key,
1247 const struct nlattr *actions,
1250 const struct nlattr *a;
1253 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1254 int type = nl_attr_type(a);
1256 switch ((enum ovs_action_attr) type) {
1257 case OVS_ACTION_ATTR_OUTPUT:
1258 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1261 case OVS_ACTION_ATTR_USERSPACE:
1262 dp_netdev_action_userspace(dp, packet, key, a);
1265 case OVS_ACTION_ATTR_PUSH_VLAN: {
1266 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1267 eth_push_vlan(packet, vlan->vlan_tci);
1271 case OVS_ACTION_ATTR_POP_VLAN:
1272 eth_pop_vlan(packet);
1275 case OVS_ACTION_ATTR_SET:
1276 execute_set_action(packet, nl_attr_get(a));
1279 case OVS_ACTION_ATTR_SAMPLE:
1280 dp_netdev_sample(dp, packet, key, a);
1283 case OVS_ACTION_ATTR_UNSPEC:
1284 case __OVS_ACTION_ATTR_MAX:
1290 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1291 dpif_netdev_enumerate, \
1292 dpif_netdev_port_open_type, \
1294 dpif_netdev_close, \
1295 dpif_netdev_destroy, \
1298 dpif_netdev_get_stats, \
1299 dpif_netdev_port_add, \
1300 dpif_netdev_port_del, \
1301 dpif_netdev_port_query_by_number, \
1302 dpif_netdev_port_query_by_name, \
1303 dpif_netdev_get_max_ports, \
1304 NULL, /* port_get_pid */ \
1305 dpif_netdev_port_dump_start, \
1306 dpif_netdev_port_dump_next, \
1307 dpif_netdev_port_dump_done, \
1308 dpif_netdev_port_poll, \
1309 dpif_netdev_port_poll_wait, \
1310 dpif_netdev_flow_get, \
1311 dpif_netdev_flow_put, \
1312 dpif_netdev_flow_del, \
1313 dpif_netdev_flow_flush, \
1314 dpif_netdev_flow_dump_start, \
1315 dpif_netdev_flow_dump_next, \
1316 dpif_netdev_flow_dump_done, \
1317 dpif_netdev_execute, \
1318 NULL, /* operate */ \
1319 dpif_netdev_recv_set, \
1320 dpif_netdev_queue_to_priority, \
1322 dpif_netdev_recv_wait, \
1323 dpif_netdev_recv_purge, \
1325 const struct dpif_class dpif_netdev_class = {
1327 DPIF_NETDEV_CLASS_FUNCTIONS
1330 const struct dpif_class dpif_planetlab_class = {
1332 DPIF_NETDEV_CLASS_FUNCTIONS
1336 dpif_dummy_register__(const char *type)
1338 struct dpif_class *class;
1340 class = xmalloc(sizeof *class);
1341 *class = dpif_netdev_class;
1342 class->type = xstrdup(type);
1343 dp_register_provider(class);
1347 dpif_dummy_register(bool override)
1354 dp_enumerate_types(&types);
1355 SSET_FOR_EACH (type, &types) {
1356 if (!dp_unregister_provider(type)) {
1357 dpif_dummy_register__(type);
1360 sset_destroy(&types);
1363 dpif_dummy_register__("dummy");