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 ? "dummy"
192 create_dpif_netdev(struct dp_netdev *dp)
194 uint16_t netflow_id = hash_string(dp->name, 0);
195 struct dpif_netdev *dpif;
199 dpif = xmalloc(sizeof *dpif);
200 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
202 dpif->dp_serial = dp->serial;
208 choose_port(struct dp_netdev *dp, const char *name)
212 if (dp->class != &dpif_netdev_class) {
216 /* If the port name begins with "br", start the number search at
217 * 100 to make writing tests easier. */
218 if (!strncmp(name, "br", 2)) {
222 /* If the port name contains a number, try to assign that port number.
223 * This can make writing unit tests easier because port numbers are
225 for (p = name; *p != '\0'; p++) {
226 if (isdigit((unsigned char) *p)) {
227 port_no = start_no + strtol(p, NULL, 10);
228 if (port_no > 0 && port_no < MAX_PORTS
229 && !dp->ports[port_no]) {
237 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
238 if (!dp->ports[port_no]) {
247 create_dp_netdev(const char *name, const struct dpif_class *class,
248 struct dp_netdev **dpp)
250 struct dp_netdev *dp;
254 dp = xzalloc(sizeof *dp);
256 dp->name = xstrdup(name);
258 for (i = 0; i < N_QUEUES; i++) {
259 dp->queues[i].head = dp->queues[i].tail = 0;
261 hmap_init(&dp->flow_table);
262 list_init(&dp->port_list);
264 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
270 shash_add(&dp_netdevs, name, dp);
277 dpif_netdev_open(const struct dpif_class *class, const char *name,
278 bool create, struct dpif **dpifp)
280 struct dp_netdev *dp;
282 dp = shash_find_data(&dp_netdevs, name);
287 int error = create_dp_netdev(name, class, &dp);
291 ovs_assert(dp != NULL);
294 if (dp->class != class) {
301 *dpifp = create_dpif_netdev(dp);
306 dp_netdev_purge_queues(struct dp_netdev *dp)
310 for (i = 0; i < N_QUEUES; i++) {
311 struct dp_netdev_queue *q = &dp->queues[i];
313 while (q->tail != q->head) {
314 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
315 ofpbuf_uninit(&u->buf);
321 dp_netdev_free(struct dp_netdev *dp)
323 struct dp_netdev_port *port, *next;
325 dp_netdev_flow_flush(dp);
326 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
327 do_del_port(dp, port->port_no);
329 dp_netdev_purge_queues(dp);
330 hmap_destroy(&dp->flow_table);
336 dpif_netdev_close(struct dpif *dpif)
338 struct dp_netdev *dp = get_dp_netdev(dpif);
339 ovs_assert(dp->open_cnt > 0);
340 if (--dp->open_cnt == 0 && dp->destroyed) {
341 shash_find_and_delete(&dp_netdevs, dp->name);
348 dpif_netdev_destroy(struct dpif *dpif)
350 struct dp_netdev *dp = get_dp_netdev(dpif);
351 dp->destroyed = true;
356 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
358 struct dp_netdev *dp = get_dp_netdev(dpif);
359 stats->n_flows = hmap_count(&dp->flow_table);
360 stats->n_hit = dp->n_hit;
361 stats->n_missed = dp->n_missed;
362 stats->n_lost = dp->n_lost;
367 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
370 struct dp_netdev_port *port;
371 struct netdev *netdev;
372 const char *open_type;
376 /* XXX reject devices already in some dp_netdev. */
378 /* Open and validate network device. */
379 open_type = dpif_netdev_port_open_type(dp->class, type);
380 error = netdev_open(devname, open_type, &netdev);
384 /* XXX reject loopback devices */
385 /* XXX reject non-Ethernet devices */
387 error = netdev_listen(netdev);
389 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
390 devname, strerror(errno));
391 netdev_close(netdev);
395 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
397 netdev_close(netdev);
401 port = xmalloc(sizeof *port);
402 port->port_no = port_no;
403 port->netdev = netdev;
404 port->type = xstrdup(type);
406 error = netdev_get_mtu(netdev, &mtu);
411 list_push_back(&dp->port_list, &port->node);
412 dp->ports[port_no] = port;
419 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
422 struct dp_netdev *dp = get_dp_netdev(dpif);
425 if (*port_nop != UINT32_MAX) {
426 if (*port_nop >= MAX_PORTS) {
428 } else if (dp->ports[*port_nop]) {
433 port_no = choose_port(dp, netdev_get_name(netdev));
437 return do_add_port(dp, netdev_get_name(netdev),
438 netdev_get_type(netdev), port_no);
444 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
446 struct dp_netdev *dp = get_dp_netdev(dpif);
447 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
451 is_valid_port_number(uint32_t port_no)
453 return port_no < MAX_PORTS;
457 get_port_by_number(struct dp_netdev *dp,
458 uint32_t port_no, struct dp_netdev_port **portp)
460 if (!is_valid_port_number(port_no)) {
464 *portp = dp->ports[port_no];
465 return *portp ? 0 : ENOENT;
470 get_port_by_name(struct dp_netdev *dp,
471 const char *devname, struct dp_netdev_port **portp)
473 struct dp_netdev_port *port;
475 LIST_FOR_EACH (port, node, &dp->port_list) {
476 if (!strcmp(netdev_get_name(port->netdev), devname)) {
485 do_del_port(struct dp_netdev *dp, uint32_t port_no)
487 struct dp_netdev_port *port;
491 error = get_port_by_number(dp, port_no, &port);
496 list_remove(&port->node);
497 dp->ports[port->port_no] = NULL;
500 name = xstrdup(netdev_get_name(port->netdev));
501 netdev_close(port->netdev);
511 answer_port_query(const struct dp_netdev_port *port,
512 struct dpif_port *dpif_port)
514 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
515 dpif_port->type = xstrdup(port->type);
516 dpif_port->port_no = port->port_no;
520 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
521 struct dpif_port *dpif_port)
523 struct dp_netdev *dp = get_dp_netdev(dpif);
524 struct dp_netdev_port *port;
527 error = get_port_by_number(dp, port_no, &port);
528 if (!error && dpif_port) {
529 answer_port_query(port, dpif_port);
535 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
536 struct dpif_port *dpif_port)
538 struct dp_netdev *dp = get_dp_netdev(dpif);
539 struct dp_netdev_port *port;
542 error = get_port_by_name(dp, devname, &port);
543 if (!error && dpif_port) {
544 answer_port_query(port, dpif_port);
550 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
556 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
558 hmap_remove(&dp->flow_table, &flow->node);
564 dp_netdev_flow_flush(struct dp_netdev *dp)
566 struct dp_netdev_flow *flow, *next;
568 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
569 dp_netdev_free_flow(dp, flow);
574 dpif_netdev_flow_flush(struct dpif *dpif)
576 struct dp_netdev *dp = get_dp_netdev(dpif);
577 dp_netdev_flow_flush(dp);
581 struct dp_netdev_port_state {
587 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
589 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
594 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
595 struct dpif_port *dpif_port)
597 struct dp_netdev_port_state *state = state_;
598 struct dp_netdev *dp = get_dp_netdev(dpif);
601 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
602 struct dp_netdev_port *port = dp->ports[port_no];
605 state->name = xstrdup(netdev_get_name(port->netdev));
606 dpif_port->name = state->name;
607 dpif_port->type = port->type;
608 dpif_port->port_no = port->port_no;
609 state->port_no = port_no + 1;
617 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
619 struct dp_netdev_port_state *state = state_;
626 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
628 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
629 if (dpif->dp_serial != dpif->dp->serial) {
630 dpif->dp_serial = dpif->dp->serial;
638 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
640 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
641 if (dpif->dp_serial != dpif->dp->serial) {
642 poll_immediate_wake();
646 static struct dp_netdev_flow *
647 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
649 struct dp_netdev_flow *flow;
651 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
652 if (flow_equal(&flow->key, key)) {
660 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
662 stats->n_packets = flow->packet_count;
663 stats->n_bytes = flow->byte_count;
664 stats->used = flow->used;
665 stats->tcp_flags = flow->tcp_flags;
669 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
672 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
673 /* This should not happen: it indicates that odp_flow_key_from_flow()
674 * and odp_flow_key_to_flow() disagree on the acceptable form of a
675 * flow. Log the problem as an error, with enough details to enable
677 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
679 if (!VLOG_DROP_ERR(&rl)) {
683 odp_flow_key_format(key, key_len, &s);
684 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
691 if (flow->in_port < OFPP_MAX
692 ? flow->in_port >= MAX_PORTS
693 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
701 dpif_netdev_flow_get(const struct dpif *dpif,
702 const struct nlattr *nl_key, size_t nl_key_len,
703 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
705 struct dp_netdev *dp = get_dp_netdev(dpif);
706 struct dp_netdev_flow *flow;
710 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
715 flow = dp_netdev_lookup_flow(dp, &key);
721 get_dpif_flow_stats(flow, stats);
724 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
730 set_flow_actions(struct dp_netdev_flow *flow,
731 const struct nlattr *actions, size_t actions_len)
733 flow->actions = xrealloc(flow->actions, actions_len);
734 flow->actions_len = actions_len;
735 memcpy(flow->actions, actions, actions_len);
740 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
741 const struct nlattr *actions, size_t actions_len)
743 struct dp_netdev_flow *flow;
746 flow = xzalloc(sizeof *flow);
749 error = set_flow_actions(flow, actions, actions_len);
755 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
760 clear_stats(struct dp_netdev_flow *flow)
763 flow->packet_count = 0;
764 flow->byte_count = 0;
769 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
771 struct dp_netdev *dp = get_dp_netdev(dpif);
772 struct dp_netdev_flow *flow;
776 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
781 flow = dp_netdev_lookup_flow(dp, &key);
783 if (put->flags & DPIF_FP_CREATE) {
784 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
786 memset(put->stats, 0, sizeof *put->stats);
788 return dp_netdev_flow_add(dp, &key, put->actions,
797 if (put->flags & DPIF_FP_MODIFY) {
798 int error = set_flow_actions(flow, put->actions, put->actions_len);
801 get_dpif_flow_stats(flow, put->stats);
803 if (put->flags & DPIF_FP_ZERO_STATS) {
815 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
817 struct dp_netdev *dp = get_dp_netdev(dpif);
818 struct dp_netdev_flow *flow;
822 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
827 flow = dp_netdev_lookup_flow(dp, &key);
830 get_dpif_flow_stats(flow, del->stats);
832 dp_netdev_free_flow(dp, flow);
839 struct dp_netdev_flow_state {
842 struct nlattr *actions;
843 struct odputil_keybuf keybuf;
844 struct dpif_flow_stats stats;
848 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
850 struct dp_netdev_flow_state *state;
852 *statep = state = xmalloc(sizeof *state);
855 state->actions = NULL;
860 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
861 const struct nlattr **key, size_t *key_len,
862 const struct nlattr **actions, size_t *actions_len,
863 const struct dpif_flow_stats **stats)
865 struct dp_netdev_flow_state *state = state_;
866 struct dp_netdev *dp = get_dp_netdev(dpif);
867 struct dp_netdev_flow *flow;
868 struct hmap_node *node;
870 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
875 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
880 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
881 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
888 free(state->actions);
889 state->actions = xmemdup(flow->actions, flow->actions_len);
891 *actions = state->actions;
892 *actions_len = flow->actions_len;
896 get_dpif_flow_stats(flow, &state->stats);
897 *stats = &state->stats;
904 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
906 struct dp_netdev_flow_state *state = state_;
908 free(state->actions);
914 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
916 struct dp_netdev *dp = get_dp_netdev(dpif);
921 if (execute->packet->size < ETH_HEADER_LEN ||
922 execute->packet->size > UINT16_MAX) {
926 /* Make a deep copy of 'packet', because we might modify its data. */
927 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
928 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
929 ofpbuf_put(©, execute->packet->data, execute->packet->size);
931 flow_extract(©, 0, 0, NULL, -1, &key);
932 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
935 dp_netdev_execute_actions(dp, ©, &key,
936 execute->actions, execute->actions_len);
939 ofpbuf_uninit(©);
944 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
950 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
951 uint32_t queue_id, uint32_t *priority)
953 *priority = queue_id;
957 static struct dp_netdev_queue *
958 find_nonempty_queue(struct dpif *dpif)
960 struct dp_netdev *dp = get_dp_netdev(dpif);
963 for (i = 0; i < N_QUEUES; i++) {
964 struct dp_netdev_queue *q = &dp->queues[i];
965 if (q->head != q->tail) {
973 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
976 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
978 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
981 upcall->packet = buf;
993 dpif_netdev_recv_wait(struct dpif *dpif)
995 if (find_nonempty_queue(dpif)) {
996 poll_immediate_wake();
998 /* No messages ready to be received, and dp_wait() will ensure that we
999 * wake up to queue new messages, so there is nothing to do. */
1004 dpif_netdev_recv_purge(struct dpif *dpif)
1006 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1007 dp_netdev_purge_queues(dpif_netdev->dp);
1011 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1013 flow->used = time_msec();
1014 flow->packet_count++;
1015 flow->byte_count += packet->size;
1016 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1020 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1021 struct ofpbuf *packet)
1023 struct dp_netdev_flow *flow;
1026 if (packet->size < ETH_HEADER_LEN) {
1029 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1030 flow = dp_netdev_lookup_flow(dp, &key);
1032 dp_netdev_flow_used(flow, packet);
1033 dp_netdev_execute_actions(dp, packet, &key,
1034 flow->actions, flow->actions_len);
1038 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1043 dpif_netdev_run(struct dpif *dpif)
1045 struct dp_netdev *dp = get_dp_netdev(dpif);
1046 struct dp_netdev_port *port;
1047 struct ofpbuf packet;
1049 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1051 LIST_FOR_EACH (port, node, &dp->port_list) {
1054 /* Reset packet contents. */
1055 ofpbuf_clear(&packet);
1056 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1058 error = netdev_recv(port->netdev, &packet);
1060 dp_netdev_port_input(dp, port, &packet);
1061 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1062 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1063 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1064 netdev_get_name(port->netdev), strerror(error));
1067 ofpbuf_uninit(&packet);
1071 dpif_netdev_wait(struct dpif *dpif)
1073 struct dp_netdev *dp = get_dp_netdev(dpif);
1074 struct dp_netdev_port *port;
1076 LIST_FOR_EACH (port, node, &dp->port_list) {
1077 netdev_recv_wait(port->netdev);
1082 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1084 struct eth_header *eh = packet->l2;
1086 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1087 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1091 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1094 struct dp_netdev_port *p = dp->ports[out_port];
1096 netdev_send(p->netdev, packet);
1101 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1102 int queue_no, const struct flow *flow, uint64_t arg)
1104 struct dp_netdev_queue *q = &dp->queues[queue_no];
1105 struct dp_netdev_upcall *u;
1106 struct dpif_upcall *upcall;
1110 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1115 u = &q->upcalls[q->head++ & QUEUE_MASK];
1118 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1119 odp_flow_key_from_flow(buf, flow, flow->in_port);
1120 key_len = buf->size;
1121 ofpbuf_pull(buf, key_len);
1122 ofpbuf_reserve(buf, 2);
1123 ofpbuf_put(buf, packet->data, packet->size);
1125 upcall = &u->upcall;
1126 upcall->type = queue_no;
1127 upcall->packet = buf;
1128 upcall->key = buf->base;
1129 upcall->key_len = key_len;
1130 upcall->userdata = arg;
1136 dp_netdev_sample(struct dp_netdev *dp,
1137 struct ofpbuf *packet, struct flow *key,
1138 const struct nlattr *action)
1140 const struct nlattr *subactions = NULL;
1141 const struct nlattr *a;
1144 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1145 int type = nl_attr_type(a);
1147 switch ((enum ovs_sample_attr) type) {
1148 case OVS_SAMPLE_ATTR_PROBABILITY:
1149 if (random_uint32() >= nl_attr_get_u32(a)) {
1154 case OVS_SAMPLE_ATTR_ACTIONS:
1158 case OVS_SAMPLE_ATTR_UNSPEC:
1159 case __OVS_SAMPLE_ATTR_MAX:
1165 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1166 nl_attr_get_size(subactions));
1170 dp_netdev_action_userspace(struct dp_netdev *dp,
1171 struct ofpbuf *packet, struct flow *key,
1172 const struct nlattr *a)
1174 const struct nlattr *userdata_attr;
1177 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1178 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1179 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1183 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1185 enum ovs_key_attr type = nl_attr_type(a);
1186 const struct ovs_key_ipv4 *ipv4_key;
1187 const struct ovs_key_ipv6 *ipv6_key;
1188 const struct ovs_key_tcp *tcp_key;
1189 const struct ovs_key_udp *udp_key;
1192 case OVS_KEY_ATTR_TUN_ID:
1193 case OVS_KEY_ATTR_PRIORITY:
1194 case OVS_KEY_ATTR_SKB_MARK:
1195 case OVS_KEY_ATTR_IPV4_TUNNEL:
1196 /* not implemented */
1199 case OVS_KEY_ATTR_ETHERNET:
1200 dp_netdev_set_dl(packet,
1201 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1204 case OVS_KEY_ATTR_IPV4:
1205 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1206 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1207 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1210 case OVS_KEY_ATTR_IPV6:
1211 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1212 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1213 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1214 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1217 case OVS_KEY_ATTR_TCP:
1218 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1219 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1222 case OVS_KEY_ATTR_UDP:
1223 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1224 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1227 case OVS_KEY_ATTR_UNSPEC:
1228 case OVS_KEY_ATTR_ENCAP:
1229 case OVS_KEY_ATTR_ETHERTYPE:
1230 case OVS_KEY_ATTR_IN_PORT:
1231 case OVS_KEY_ATTR_VLAN:
1232 case OVS_KEY_ATTR_ICMP:
1233 case OVS_KEY_ATTR_ICMPV6:
1234 case OVS_KEY_ATTR_ARP:
1235 case OVS_KEY_ATTR_ND:
1236 case __OVS_KEY_ATTR_MAX:
1243 dp_netdev_execute_actions(struct dp_netdev *dp,
1244 struct ofpbuf *packet, struct flow *key,
1245 const struct nlattr *actions,
1248 const struct nlattr *a;
1251 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1252 const struct ovs_action_push_vlan *vlan;
1253 int type = nl_attr_type(a);
1255 switch ((enum ovs_action_attr) type) {
1256 case OVS_ACTION_ATTR_OUTPUT:
1257 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1260 case OVS_ACTION_ATTR_USERSPACE:
1261 dp_netdev_action_userspace(dp, packet, key, a);
1264 case OVS_ACTION_ATTR_PUSH_VLAN:
1265 vlan = nl_attr_get(a);
1266 eth_push_vlan(packet, vlan->vlan_tci);
1269 case OVS_ACTION_ATTR_POP_VLAN:
1270 eth_pop_vlan(packet);
1273 case OVS_ACTION_ATTR_SET:
1274 execute_set_action(packet, nl_attr_get(a));
1277 case OVS_ACTION_ATTR_SAMPLE:
1278 dp_netdev_sample(dp, packet, key, a);
1281 case OVS_ACTION_ATTR_UNSPEC:
1282 case __OVS_ACTION_ATTR_MAX:
1288 const struct dpif_class dpif_netdev_class = {
1290 dpif_netdev_enumerate,
1291 dpif_netdev_port_open_type,
1294 dpif_netdev_destroy,
1297 dpif_netdev_get_stats,
1298 dpif_netdev_port_add,
1299 dpif_netdev_port_del,
1300 dpif_netdev_port_query_by_number,
1301 dpif_netdev_port_query_by_name,
1302 dpif_netdev_get_max_ports,
1303 NULL, /* port_get_pid */
1304 dpif_netdev_port_dump_start,
1305 dpif_netdev_port_dump_next,
1306 dpif_netdev_port_dump_done,
1307 dpif_netdev_port_poll,
1308 dpif_netdev_port_poll_wait,
1309 dpif_netdev_flow_get,
1310 dpif_netdev_flow_put,
1311 dpif_netdev_flow_del,
1312 dpif_netdev_flow_flush,
1313 dpif_netdev_flow_dump_start,
1314 dpif_netdev_flow_dump_next,
1315 dpif_netdev_flow_dump_done,
1316 dpif_netdev_execute,
1318 dpif_netdev_recv_set,
1319 dpif_netdev_queue_to_priority,
1321 dpif_netdev_recv_wait,
1322 dpif_netdev_recv_purge,
1326 dpif_dummy_register__(const char *type)
1328 struct dpif_class *class;
1330 class = xmalloc(sizeof *class);
1331 *class = dpif_netdev_class;
1332 class->type = xstrdup(type);
1333 dp_register_provider(class);
1337 dpif_dummy_register(bool override)
1344 dp_enumerate_types(&types);
1345 SSET_FOR_EACH (type, &types) {
1346 if (!dp_unregister_provider(type)) {
1347 dpif_dummy_register__(type);
1350 sset_destroy(&types);
1353 dpif_dummy_register__("dummy");