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_class_is_dummy(const struct dpif_class *class)
186 return class != &dpif_netdev_class;
190 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
192 return strcmp(type, "internal") ? type
193 : dpif_netdev_class_is_dummy(class) ? "dummy"
198 create_dpif_netdev(struct dp_netdev *dp)
200 uint16_t netflow_id = hash_string(dp->name, 0);
201 struct dpif_netdev *dpif;
205 dpif = xmalloc(sizeof *dpif);
206 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
208 dpif->dp_serial = dp->serial;
214 choose_port(struct dp_netdev *dp, const char *name)
218 if (dp->class != &dpif_netdev_class) {
222 /* If the port name begins with "br", start the number search at
223 * 100 to make writing tests easier. */
224 if (!strncmp(name, "br", 2)) {
228 /* If the port name contains a number, try to assign that port number.
229 * This can make writing unit tests easier because port numbers are
231 for (p = name; *p != '\0'; p++) {
232 if (isdigit((unsigned char) *p)) {
233 port_no = start_no + strtol(p, NULL, 10);
234 if (port_no > 0 && port_no < MAX_PORTS
235 && !dp->ports[port_no]) {
243 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
244 if (!dp->ports[port_no]) {
253 create_dp_netdev(const char *name, const struct dpif_class *class,
254 struct dp_netdev **dpp)
256 struct dp_netdev *dp;
260 dp = xzalloc(sizeof *dp);
262 dp->name = xstrdup(name);
264 for (i = 0; i < N_QUEUES; i++) {
265 dp->queues[i].head = dp->queues[i].tail = 0;
267 hmap_init(&dp->flow_table);
268 list_init(&dp->port_list);
270 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
276 shash_add(&dp_netdevs, name, dp);
283 dpif_netdev_open(const struct dpif_class *class, const char *name,
284 bool create, struct dpif **dpifp)
286 struct dp_netdev *dp;
288 dp = shash_find_data(&dp_netdevs, name);
293 int error = create_dp_netdev(name, class, &dp);
297 ovs_assert(dp != NULL);
300 if (dp->class != class) {
307 *dpifp = create_dpif_netdev(dp);
312 dp_netdev_purge_queues(struct dp_netdev *dp)
316 for (i = 0; i < N_QUEUES; i++) {
317 struct dp_netdev_queue *q = &dp->queues[i];
319 while (q->tail != q->head) {
320 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
321 ofpbuf_uninit(&u->buf);
327 dp_netdev_free(struct dp_netdev *dp)
329 struct dp_netdev_port *port, *next;
331 dp_netdev_flow_flush(dp);
332 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
333 do_del_port(dp, port->port_no);
335 dp_netdev_purge_queues(dp);
336 hmap_destroy(&dp->flow_table);
342 dpif_netdev_close(struct dpif *dpif)
344 struct dp_netdev *dp = get_dp_netdev(dpif);
345 ovs_assert(dp->open_cnt > 0);
346 if (--dp->open_cnt == 0 && dp->destroyed) {
347 shash_find_and_delete(&dp_netdevs, dp->name);
354 dpif_netdev_destroy(struct dpif *dpif)
356 struct dp_netdev *dp = get_dp_netdev(dpif);
357 dp->destroyed = true;
362 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
364 struct dp_netdev *dp = get_dp_netdev(dpif);
365 stats->n_flows = hmap_count(&dp->flow_table);
366 stats->n_hit = dp->n_hit;
367 stats->n_missed = dp->n_missed;
368 stats->n_lost = dp->n_lost;
373 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
376 struct dp_netdev_port *port;
377 struct netdev *netdev;
378 const char *open_type;
382 /* XXX reject devices already in some dp_netdev. */
384 /* Open and validate network device. */
385 open_type = dpif_netdev_port_open_type(dp->class, type);
386 error = netdev_open(devname, open_type, &netdev);
390 /* XXX reject loopback devices */
391 /* XXX reject non-Ethernet devices */
393 error = netdev_listen(netdev);
395 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
396 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
397 devname, strerror(errno));
398 netdev_close(netdev);
402 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
404 netdev_close(netdev);
408 port = xmalloc(sizeof *port);
409 port->port_no = port_no;
410 port->netdev = netdev;
411 port->type = xstrdup(type);
413 error = netdev_get_mtu(netdev, &mtu);
418 list_push_back(&dp->port_list, &port->node);
419 dp->ports[port_no] = port;
426 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
429 struct dp_netdev *dp = get_dp_netdev(dpif);
432 if (*port_nop != UINT32_MAX) {
433 if (*port_nop >= MAX_PORTS) {
435 } else if (dp->ports[*port_nop]) {
440 port_no = choose_port(dp, netdev_get_name(netdev));
444 return do_add_port(dp, netdev_get_name(netdev),
445 netdev_get_type(netdev), port_no);
451 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
453 struct dp_netdev *dp = get_dp_netdev(dpif);
454 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
458 is_valid_port_number(uint32_t port_no)
460 return port_no < MAX_PORTS;
464 get_port_by_number(struct dp_netdev *dp,
465 uint32_t port_no, struct dp_netdev_port **portp)
467 if (!is_valid_port_number(port_no)) {
471 *portp = dp->ports[port_no];
472 return *portp ? 0 : ENOENT;
477 get_port_by_name(struct dp_netdev *dp,
478 const char *devname, struct dp_netdev_port **portp)
480 struct dp_netdev_port *port;
482 LIST_FOR_EACH (port, node, &dp->port_list) {
483 if (!strcmp(netdev_get_name(port->netdev), devname)) {
492 do_del_port(struct dp_netdev *dp, uint32_t port_no)
494 struct dp_netdev_port *port;
498 error = get_port_by_number(dp, port_no, &port);
503 list_remove(&port->node);
504 dp->ports[port->port_no] = NULL;
507 name = xstrdup(netdev_get_name(port->netdev));
508 netdev_close(port->netdev);
518 answer_port_query(const struct dp_netdev_port *port,
519 struct dpif_port *dpif_port)
521 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
522 dpif_port->type = xstrdup(port->type);
523 dpif_port->port_no = port->port_no;
527 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
528 struct dpif_port *dpif_port)
530 struct dp_netdev *dp = get_dp_netdev(dpif);
531 struct dp_netdev_port *port;
534 error = get_port_by_number(dp, port_no, &port);
535 if (!error && dpif_port) {
536 answer_port_query(port, dpif_port);
542 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
543 struct dpif_port *dpif_port)
545 struct dp_netdev *dp = get_dp_netdev(dpif);
546 struct dp_netdev_port *port;
549 error = get_port_by_name(dp, devname, &port);
550 if (!error && dpif_port) {
551 answer_port_query(port, dpif_port);
557 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
563 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
565 hmap_remove(&dp->flow_table, &flow->node);
571 dp_netdev_flow_flush(struct dp_netdev *dp)
573 struct dp_netdev_flow *flow, *next;
575 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
576 dp_netdev_free_flow(dp, flow);
581 dpif_netdev_flow_flush(struct dpif *dpif)
583 struct dp_netdev *dp = get_dp_netdev(dpif);
584 dp_netdev_flow_flush(dp);
588 struct dp_netdev_port_state {
594 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
596 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
601 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
602 struct dpif_port *dpif_port)
604 struct dp_netdev_port_state *state = state_;
605 struct dp_netdev *dp = get_dp_netdev(dpif);
608 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
609 struct dp_netdev_port *port = dp->ports[port_no];
612 state->name = xstrdup(netdev_get_name(port->netdev));
613 dpif_port->name = state->name;
614 dpif_port->type = port->type;
615 dpif_port->port_no = port->port_no;
616 state->port_no = port_no + 1;
624 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
626 struct dp_netdev_port_state *state = state_;
633 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
635 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
636 if (dpif->dp_serial != dpif->dp->serial) {
637 dpif->dp_serial = dpif->dp->serial;
645 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
647 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
648 if (dpif->dp_serial != dpif->dp->serial) {
649 poll_immediate_wake();
653 static struct dp_netdev_flow *
654 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
656 struct dp_netdev_flow *flow;
658 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
659 if (flow_equal(&flow->key, key)) {
667 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
669 stats->n_packets = flow->packet_count;
670 stats->n_bytes = flow->byte_count;
671 stats->used = flow->used;
672 stats->tcp_flags = flow->tcp_flags;
676 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
679 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
680 /* This should not happen: it indicates that odp_flow_key_from_flow()
681 * and odp_flow_key_to_flow() disagree on the acceptable form of a
682 * flow. Log the problem as an error, with enough details to enable
684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
686 if (!VLOG_DROP_ERR(&rl)) {
690 odp_flow_key_format(key, key_len, &s);
691 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
698 if (flow->in_port < OFPP_MAX
699 ? flow->in_port >= MAX_PORTS
700 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
708 dpif_netdev_flow_get(const struct dpif *dpif,
709 const struct nlattr *nl_key, size_t nl_key_len,
710 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
712 struct dp_netdev *dp = get_dp_netdev(dpif);
713 struct dp_netdev_flow *flow;
717 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
722 flow = dp_netdev_lookup_flow(dp, &key);
728 get_dpif_flow_stats(flow, stats);
731 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
737 set_flow_actions(struct dp_netdev_flow *flow,
738 const struct nlattr *actions, size_t actions_len)
740 flow->actions = xrealloc(flow->actions, actions_len);
741 flow->actions_len = actions_len;
742 memcpy(flow->actions, actions, actions_len);
747 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
748 const struct nlattr *actions, size_t actions_len)
750 struct dp_netdev_flow *flow;
753 flow = xzalloc(sizeof *flow);
756 error = set_flow_actions(flow, actions, actions_len);
762 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
767 clear_stats(struct dp_netdev_flow *flow)
770 flow->packet_count = 0;
771 flow->byte_count = 0;
776 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
778 struct dp_netdev *dp = get_dp_netdev(dpif);
779 struct dp_netdev_flow *flow;
783 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
788 flow = dp_netdev_lookup_flow(dp, &key);
790 if (put->flags & DPIF_FP_CREATE) {
791 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
793 memset(put->stats, 0, sizeof *put->stats);
795 return dp_netdev_flow_add(dp, &key, put->actions,
804 if (put->flags & DPIF_FP_MODIFY) {
805 int error = set_flow_actions(flow, put->actions, put->actions_len);
808 get_dpif_flow_stats(flow, put->stats);
810 if (put->flags & DPIF_FP_ZERO_STATS) {
822 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
824 struct dp_netdev *dp = get_dp_netdev(dpif);
825 struct dp_netdev_flow *flow;
829 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
834 flow = dp_netdev_lookup_flow(dp, &key);
837 get_dpif_flow_stats(flow, del->stats);
839 dp_netdev_free_flow(dp, flow);
846 struct dp_netdev_flow_state {
849 struct nlattr *actions;
850 struct odputil_keybuf keybuf;
851 struct dpif_flow_stats stats;
855 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
857 struct dp_netdev_flow_state *state;
859 *statep = state = xmalloc(sizeof *state);
862 state->actions = NULL;
867 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
868 const struct nlattr **key, size_t *key_len,
869 const struct nlattr **actions, size_t *actions_len,
870 const struct dpif_flow_stats **stats)
872 struct dp_netdev_flow_state *state = state_;
873 struct dp_netdev *dp = get_dp_netdev(dpif);
874 struct dp_netdev_flow *flow;
875 struct hmap_node *node;
877 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
882 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
887 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
888 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
895 free(state->actions);
896 state->actions = xmemdup(flow->actions, flow->actions_len);
898 *actions = state->actions;
899 *actions_len = flow->actions_len;
903 get_dpif_flow_stats(flow, &state->stats);
904 *stats = &state->stats;
911 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
913 struct dp_netdev_flow_state *state = state_;
915 free(state->actions);
921 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
923 struct dp_netdev *dp = get_dp_netdev(dpif);
928 if (execute->packet->size < ETH_HEADER_LEN ||
929 execute->packet->size > UINT16_MAX) {
933 /* Make a deep copy of 'packet', because we might modify its data. */
934 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
935 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
936 ofpbuf_put(©, execute->packet->data, execute->packet->size);
938 flow_extract(©, 0, 0, NULL, -1, &key);
939 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
942 dp_netdev_execute_actions(dp, ©, &key,
943 execute->actions, execute->actions_len);
946 ofpbuf_uninit(©);
951 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
957 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
958 uint32_t queue_id, uint32_t *priority)
960 *priority = queue_id;
964 static struct dp_netdev_queue *
965 find_nonempty_queue(struct dpif *dpif)
967 struct dp_netdev *dp = get_dp_netdev(dpif);
970 for (i = 0; i < N_QUEUES; i++) {
971 struct dp_netdev_queue *q = &dp->queues[i];
972 if (q->head != q->tail) {
980 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
983 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
985 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
988 upcall->packet = buf;
1000 dpif_netdev_recv_wait(struct dpif *dpif)
1002 if (find_nonempty_queue(dpif)) {
1003 poll_immediate_wake();
1005 /* No messages ready to be received, and dp_wait() will ensure that we
1006 * wake up to queue new messages, so there is nothing to do. */
1011 dpif_netdev_recv_purge(struct dpif *dpif)
1013 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1014 dp_netdev_purge_queues(dpif_netdev->dp);
1018 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1020 flow->used = time_msec();
1021 flow->packet_count++;
1022 flow->byte_count += packet->size;
1023 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1027 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1028 struct ofpbuf *packet)
1030 struct dp_netdev_flow *flow;
1033 if (packet->size < ETH_HEADER_LEN) {
1036 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1037 flow = dp_netdev_lookup_flow(dp, &key);
1039 dp_netdev_flow_used(flow, packet);
1040 dp_netdev_execute_actions(dp, packet, &key,
1041 flow->actions, flow->actions_len);
1045 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1050 dpif_netdev_run(struct dpif *dpif)
1052 struct dp_netdev *dp = get_dp_netdev(dpif);
1053 struct dp_netdev_port *port;
1054 struct ofpbuf packet;
1056 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1058 LIST_FOR_EACH (port, node, &dp->port_list) {
1061 /* Reset packet contents. */
1062 ofpbuf_clear(&packet);
1063 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1065 error = netdev_recv(port->netdev, &packet);
1067 dp_netdev_port_input(dp, port, &packet);
1068 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1069 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1070 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1071 netdev_get_name(port->netdev), strerror(error));
1074 ofpbuf_uninit(&packet);
1078 dpif_netdev_wait(struct dpif *dpif)
1080 struct dp_netdev *dp = get_dp_netdev(dpif);
1081 struct dp_netdev_port *port;
1083 LIST_FOR_EACH (port, node, &dp->port_list) {
1084 netdev_recv_wait(port->netdev);
1089 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1091 struct eth_header *eh = packet->l2;
1093 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1094 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1098 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1101 struct dp_netdev_port *p = dp->ports[out_port];
1103 netdev_send(p->netdev, packet);
1108 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1109 int queue_no, const struct flow *flow, uint64_t arg)
1111 struct dp_netdev_queue *q = &dp->queues[queue_no];
1112 struct dp_netdev_upcall *u;
1113 struct dpif_upcall *upcall;
1117 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1122 u = &q->upcalls[q->head++ & QUEUE_MASK];
1125 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1126 odp_flow_key_from_flow(buf, flow, flow->in_port);
1127 key_len = buf->size;
1128 ofpbuf_pull(buf, key_len);
1129 ofpbuf_reserve(buf, 2);
1130 ofpbuf_put(buf, packet->data, packet->size);
1132 upcall = &u->upcall;
1133 upcall->type = queue_no;
1134 upcall->packet = buf;
1135 upcall->key = buf->base;
1136 upcall->key_len = key_len;
1137 upcall->userdata = arg;
1143 dp_netdev_sample(struct dp_netdev *dp,
1144 struct ofpbuf *packet, struct flow *key,
1145 const struct nlattr *action)
1147 const struct nlattr *subactions = NULL;
1148 const struct nlattr *a;
1151 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1152 int type = nl_attr_type(a);
1154 switch ((enum ovs_sample_attr) type) {
1155 case OVS_SAMPLE_ATTR_PROBABILITY:
1156 if (random_uint32() >= nl_attr_get_u32(a)) {
1161 case OVS_SAMPLE_ATTR_ACTIONS:
1165 case OVS_SAMPLE_ATTR_UNSPEC:
1166 case __OVS_SAMPLE_ATTR_MAX:
1172 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1173 nl_attr_get_size(subactions));
1177 dp_netdev_action_userspace(struct dp_netdev *dp,
1178 struct ofpbuf *packet, struct flow *key,
1179 const struct nlattr *a)
1181 const struct nlattr *userdata_attr;
1184 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1185 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1186 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1190 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1192 enum ovs_key_attr type = nl_attr_type(a);
1193 const struct ovs_key_ipv4 *ipv4_key;
1194 const struct ovs_key_ipv6 *ipv6_key;
1195 const struct ovs_key_tcp *tcp_key;
1196 const struct ovs_key_udp *udp_key;
1199 case OVS_KEY_ATTR_TUN_ID:
1200 case OVS_KEY_ATTR_PRIORITY:
1201 case OVS_KEY_ATTR_SKB_MARK:
1202 case OVS_KEY_ATTR_TUNNEL:
1203 /* not implemented */
1206 case OVS_KEY_ATTR_ETHERNET:
1207 dp_netdev_set_dl(packet,
1208 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1211 case OVS_KEY_ATTR_IPV4:
1212 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1213 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1214 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1217 case OVS_KEY_ATTR_IPV6:
1218 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1219 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1220 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1221 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1224 case OVS_KEY_ATTR_TCP:
1225 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1226 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1229 case OVS_KEY_ATTR_UDP:
1230 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1231 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1234 case OVS_KEY_ATTR_UNSPEC:
1235 case OVS_KEY_ATTR_ENCAP:
1236 case OVS_KEY_ATTR_ETHERTYPE:
1237 case OVS_KEY_ATTR_IN_PORT:
1238 case OVS_KEY_ATTR_VLAN:
1239 case OVS_KEY_ATTR_ICMP:
1240 case OVS_KEY_ATTR_ICMPV6:
1241 case OVS_KEY_ATTR_ARP:
1242 case OVS_KEY_ATTR_ND:
1243 case __OVS_KEY_ATTR_MAX:
1250 dp_netdev_execute_actions(struct dp_netdev *dp,
1251 struct ofpbuf *packet, struct flow *key,
1252 const struct nlattr *actions,
1255 const struct nlattr *a;
1258 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1259 int type = nl_attr_type(a);
1261 switch ((enum ovs_action_attr) type) {
1262 case OVS_ACTION_ATTR_OUTPUT:
1263 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1266 case OVS_ACTION_ATTR_USERSPACE:
1267 dp_netdev_action_userspace(dp, packet, key, a);
1270 case OVS_ACTION_ATTR_PUSH_VLAN: {
1271 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1272 eth_push_vlan(packet, vlan->vlan_tci);
1276 case OVS_ACTION_ATTR_POP_VLAN:
1277 eth_pop_vlan(packet);
1280 case OVS_ACTION_ATTR_SET:
1281 execute_set_action(packet, nl_attr_get(a));
1284 case OVS_ACTION_ATTR_SAMPLE:
1285 dp_netdev_sample(dp, packet, key, a);
1288 case OVS_ACTION_ATTR_UNSPEC:
1289 case __OVS_ACTION_ATTR_MAX:
1295 const struct dpif_class dpif_netdev_class = {
1297 dpif_netdev_enumerate,
1298 dpif_netdev_port_open_type,
1301 dpif_netdev_destroy,
1304 dpif_netdev_get_stats,
1305 dpif_netdev_port_add,
1306 dpif_netdev_port_del,
1307 dpif_netdev_port_query_by_number,
1308 dpif_netdev_port_query_by_name,
1309 dpif_netdev_get_max_ports,
1310 NULL, /* port_get_pid */
1311 dpif_netdev_port_dump_start,
1312 dpif_netdev_port_dump_next,
1313 dpif_netdev_port_dump_done,
1314 dpif_netdev_port_poll,
1315 dpif_netdev_port_poll_wait,
1316 dpif_netdev_flow_get,
1317 dpif_netdev_flow_put,
1318 dpif_netdev_flow_del,
1319 dpif_netdev_flow_flush,
1320 dpif_netdev_flow_dump_start,
1321 dpif_netdev_flow_dump_next,
1322 dpif_netdev_flow_dump_done,
1323 dpif_netdev_execute,
1325 dpif_netdev_recv_set,
1326 dpif_netdev_queue_to_priority,
1328 dpif_netdev_recv_wait,
1329 dpif_netdev_recv_purge,
1333 dpif_dummy_register__(const char *type)
1335 struct dpif_class *class;
1337 class = xmalloc(sizeof *class);
1338 *class = dpif_netdev_class;
1339 class->type = xstrdup(type);
1340 dp_register_provider(class);
1344 dpif_dummy_register(bool override)
1351 dp_enumerate_types(&types);
1352 SSET_FOR_EACH (type, &types) {
1353 if (!dp_unregister_provider(type)) {
1354 dpif_dummy_register__(type);
1357 sset_destroy(&types);
1360 dpif_dummy_register__("dummy");