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.
25 #include <netinet/in.h>
26 #include <sys/socket.h>
31 #include <sys/ioctl.h>
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
46 #include "ofp-print.h"
49 #include "poll-loop.h"
57 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
59 /* Configuration parameters. */
60 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
61 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
63 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
64 * headers to be aligned on a 4-byte boundary. */
65 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
68 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
69 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
70 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
71 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
73 struct dp_netdev_upcall {
74 struct dpif_upcall upcall; /* Queued upcall information. */
75 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
78 struct dp_netdev_queue {
79 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
80 unsigned int head, tail;
83 /* Datapath based on the network device interface from netdev.h. */
85 const struct dpif_class *class;
90 struct dp_netdev_queue queues[N_QUEUES];
91 struct hmap flow_table; /* Flow table. */
94 long long int n_hit; /* Number of flow table matches. */
95 long long int n_missed; /* Number of flow table misses. */
96 long long int n_lost; /* Number of misses not passed to client. */
99 struct dp_netdev_port *ports[MAX_PORTS];
100 struct list port_list;
104 /* A port in a netdev-based datapath. */
105 struct dp_netdev_port {
106 int port_no; /* Index into dp_netdev's 'ports'. */
107 struct list node; /* Element in dp_netdev's 'port_list'. */
108 struct netdev *netdev;
109 char *type; /* Port type as requested by user. */
112 /* A flow in dp_netdev's 'flow_table'. */
113 struct dp_netdev_flow {
114 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
118 long long int used; /* Last used time, in monotonic msecs. */
119 long long int packet_count; /* Number of packets matched. */
120 long long int byte_count; /* Number of bytes matched. */
121 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
124 struct nlattr *actions;
128 /* Interface to netdev-based datapath. */
131 struct dp_netdev *dp;
132 unsigned int dp_serial;
135 /* All netdev-based datapaths. */
136 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
138 /* Maximum port MTU seen so far. */
139 static int max_mtu = ETH_PAYLOAD_MAX;
141 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
142 struct dp_netdev_port **portp);
143 static int get_port_by_name(struct dp_netdev *, const char *devname,
144 struct dp_netdev_port **portp);
145 static void dp_netdev_free(struct dp_netdev *);
146 static void dp_netdev_flow_flush(struct dp_netdev *);
147 static int do_add_port(struct dp_netdev *, const char *devname,
148 const char *type, uint32_t port_no);
149 static int do_del_port(struct dp_netdev *, uint32_t port_no);
150 static int dpif_netdev_open(const struct dpif_class *, const char *name,
151 bool create, struct dpif **);
152 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
153 int queue_no, const struct flow *,
155 static void dp_netdev_execute_actions(struct dp_netdev *,
156 struct ofpbuf *, struct flow *,
157 const struct nlattr *actions,
160 static struct dpif_netdev *
161 dpif_netdev_cast(const struct dpif *dpif)
163 assert(dpif->dpif_class->open == dpif_netdev_open);
164 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
167 static struct dp_netdev *
168 get_dp_netdev(const struct dpif *dpif)
170 return dpif_netdev_cast(dpif)->dp;
174 dpif_netdev_enumerate(struct sset *all_dps)
176 struct shash_node *node;
178 SHASH_FOR_EACH(node, &dp_netdevs) {
179 sset_add(all_dps, node->name);
185 create_dpif_netdev(struct dp_netdev *dp)
187 uint16_t netflow_id = hash_string(dp->name, 0);
188 struct dpif_netdev *dpif;
192 dpif = xmalloc(sizeof *dpif);
193 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
195 dpif->dp_serial = dp->serial;
201 choose_port(struct dp_netdev *dp, const char *name)
205 if (dp->class != &dpif_netdev_class) {
209 /* If the port name begins with "br", start the number search at
210 * 100 to make writing tests easier. */
211 if (!strncmp(name, "br", 2)) {
215 /* If the port name contains a number, try to assign that port number.
216 * This can make writing unit tests easier because port numbers are
218 for (p = name; *p != '\0'; p++) {
219 if (isdigit((unsigned char) *p)) {
220 port_no = start_no + strtol(p, NULL, 10);
221 if (port_no > 0 && port_no < MAX_PORTS
222 && !dp->ports[port_no]) {
230 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
231 if (!dp->ports[port_no]) {
240 create_dp_netdev(const char *name, const struct dpif_class *class,
241 struct dp_netdev **dpp)
243 struct dp_netdev *dp;
248 dp = xzalloc(sizeof *dp);
250 dp->name = xstrdup(name);
252 for (i = 0; i < N_QUEUES; i++) {
253 dp->queues[i].head = dp->queues[i].tail = 0;
255 hmap_init(&dp->flow_table);
256 list_init(&dp->port_list);
258 port_no = !strncmp(name, "br", 2) ? choose_port(dp, name) : OVSP_LOCAL;
259 error = do_add_port(dp, name, "internal", port_no);
265 shash_add(&dp_netdevs, name, dp);
272 dpif_netdev_open(const struct dpif_class *class, const char *name,
273 bool create, struct dpif **dpifp)
275 struct dp_netdev *dp;
277 dp = shash_find_data(&dp_netdevs, name);
282 int error = create_dp_netdev(name, class, &dp);
289 if (dp->class != class) {
296 *dpifp = create_dpif_netdev(dp);
301 dp_netdev_purge_queues(struct dp_netdev *dp)
305 for (i = 0; i < N_QUEUES; i++) {
306 struct dp_netdev_queue *q = &dp->queues[i];
308 while (q->tail != q->head) {
309 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
310 ofpbuf_uninit(&u->buf);
316 dp_netdev_free(struct dp_netdev *dp)
318 struct dp_netdev_port *port, *next;
320 dp_netdev_flow_flush(dp);
321 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
322 do_del_port(dp, port->port_no);
324 dp_netdev_purge_queues(dp);
325 hmap_destroy(&dp->flow_table);
331 dpif_netdev_close(struct dpif *dpif)
333 struct dp_netdev *dp = get_dp_netdev(dpif);
334 assert(dp->open_cnt > 0);
335 if (--dp->open_cnt == 0 && dp->destroyed) {
336 shash_find_and_delete(&dp_netdevs, dp->name);
343 dpif_netdev_destroy(struct dpif *dpif)
345 struct dp_netdev *dp = get_dp_netdev(dpif);
346 dp->destroyed = true;
351 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
353 struct dp_netdev *dp = get_dp_netdev(dpif);
354 stats->n_flows = hmap_count(&dp->flow_table);
355 stats->n_hit = dp->n_hit;
356 stats->n_missed = dp->n_missed;
357 stats->n_lost = dp->n_lost;
362 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
365 struct dp_netdev_port *port;
366 struct netdev *netdev;
367 const char *open_type;
371 /* XXX reject devices already in some dp_netdev. */
373 /* Open and validate network device. */
374 open_type = (strcmp(type, "internal") ? type
375 : dp->class != &dpif_netdev_class ? "dummy"
377 error = netdev_open(devname, open_type, &netdev);
381 /* XXX reject loopback devices */
382 /* XXX reject non-Ethernet devices */
384 error = netdev_listen(netdev);
386 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
387 devname, strerror(errno));
388 netdev_close(netdev);
392 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
394 netdev_close(netdev);
398 port = xmalloc(sizeof *port);
399 port->port_no = port_no;
400 port->netdev = netdev;
401 port->type = xstrdup(type);
403 error = netdev_get_mtu(netdev, &mtu);
408 list_push_back(&dp->port_list, &port->node);
409 dp->ports[port_no] = port;
416 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
419 struct dp_netdev *dp = get_dp_netdev(dpif);
422 if (*port_nop != UINT32_MAX) {
423 if (*port_nop >= MAX_PORTS) {
425 } else if (dp->ports[*port_nop]) {
430 port_no = choose_port(dp, netdev_get_name(netdev));
434 return do_add_port(dp, netdev_get_name(netdev),
435 netdev_get_type(netdev), port_no);
441 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
443 struct dp_netdev *dp = get_dp_netdev(dpif);
444 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
448 is_valid_port_number(uint32_t port_no)
450 return port_no < MAX_PORTS;
454 get_port_by_number(struct dp_netdev *dp,
455 uint32_t port_no, struct dp_netdev_port **portp)
457 if (!is_valid_port_number(port_no)) {
461 *portp = dp->ports[port_no];
462 return *portp ? 0 : ENOENT;
467 get_port_by_name(struct dp_netdev *dp,
468 const char *devname, struct dp_netdev_port **portp)
470 struct dp_netdev_port *port;
472 LIST_FOR_EACH (port, node, &dp->port_list) {
473 if (!strcmp(netdev_get_name(port->netdev), devname)) {
482 do_del_port(struct dp_netdev *dp, uint32_t port_no)
484 struct dp_netdev_port *port;
488 error = get_port_by_number(dp, port_no, &port);
493 list_remove(&port->node);
494 dp->ports[port->port_no] = NULL;
497 name = xstrdup(netdev_get_name(port->netdev));
498 netdev_close(port->netdev);
508 answer_port_query(const struct dp_netdev_port *port,
509 struct dpif_port *dpif_port)
511 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
512 dpif_port->type = xstrdup(port->type);
513 dpif_port->port_no = port->port_no;
517 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
518 struct dpif_port *dpif_port)
520 struct dp_netdev *dp = get_dp_netdev(dpif);
521 struct dp_netdev_port *port;
524 error = get_port_by_number(dp, port_no, &port);
525 if (!error && dpif_port) {
526 answer_port_query(port, dpif_port);
532 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
533 struct dpif_port *dpif_port)
535 struct dp_netdev *dp = get_dp_netdev(dpif);
536 struct dp_netdev_port *port;
539 error = get_port_by_name(dp, devname, &port);
540 if (!error && dpif_port) {
541 answer_port_query(port, dpif_port);
547 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
553 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
555 hmap_remove(&dp->flow_table, &flow->node);
561 dp_netdev_flow_flush(struct dp_netdev *dp)
563 struct dp_netdev_flow *flow, *next;
565 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
566 dp_netdev_free_flow(dp, flow);
571 dpif_netdev_flow_flush(struct dpif *dpif)
573 struct dp_netdev *dp = get_dp_netdev(dpif);
574 dp_netdev_flow_flush(dp);
578 struct dp_netdev_port_state {
584 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
586 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
591 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
592 struct dpif_port *dpif_port)
594 struct dp_netdev_port_state *state = state_;
595 struct dp_netdev *dp = get_dp_netdev(dpif);
598 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
599 struct dp_netdev_port *port = dp->ports[port_no];
602 state->name = xstrdup(netdev_get_name(port->netdev));
603 dpif_port->name = state->name;
604 dpif_port->type = port->type;
605 dpif_port->port_no = port->port_no;
606 state->port_no = port_no + 1;
614 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
616 struct dp_netdev_port_state *state = state_;
623 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
625 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
626 if (dpif->dp_serial != dpif->dp->serial) {
627 dpif->dp_serial = dpif->dp->serial;
635 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
637 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
638 if (dpif->dp_serial != dpif->dp->serial) {
639 poll_immediate_wake();
643 static struct dp_netdev_flow *
644 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
646 struct dp_netdev_flow *flow;
648 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
649 if (flow_equal(&flow->key, key)) {
657 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
659 stats->n_packets = flow->packet_count;
660 stats->n_bytes = flow->byte_count;
661 stats->used = flow->used;
662 stats->tcp_flags = flow->tcp_flags;
666 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
669 if (odp_flow_key_to_flow(key, key_len, flow)) {
670 /* This should not happen: it indicates that odp_flow_key_from_flow()
671 * and odp_flow_key_to_flow() disagree on the acceptable form of a
672 * flow. Log the problem as an error, with enough details to enable
674 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
676 if (!VLOG_DROP_ERR(&rl)) {
680 odp_flow_key_format(key, key_len, &s);
681 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
688 if (flow->in_port < OFPP_MAX
689 ? flow->in_port >= MAX_PORTS
690 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
698 dpif_netdev_flow_get(const struct dpif *dpif,
699 const struct nlattr *nl_key, size_t nl_key_len,
700 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
702 struct dp_netdev *dp = get_dp_netdev(dpif);
703 struct dp_netdev_flow *flow;
707 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
712 flow = dp_netdev_lookup_flow(dp, &key);
718 get_dpif_flow_stats(flow, stats);
721 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
727 set_flow_actions(struct dp_netdev_flow *flow,
728 const struct nlattr *actions, size_t actions_len)
730 flow->actions = xrealloc(flow->actions, actions_len);
731 flow->actions_len = actions_len;
732 memcpy(flow->actions, actions, actions_len);
737 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
738 const struct nlattr *actions, size_t actions_len)
740 struct dp_netdev_flow *flow;
743 flow = xzalloc(sizeof *flow);
746 error = set_flow_actions(flow, actions, actions_len);
752 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
757 clear_stats(struct dp_netdev_flow *flow)
760 flow->packet_count = 0;
761 flow->byte_count = 0;
766 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
768 struct dp_netdev *dp = get_dp_netdev(dpif);
769 struct dp_netdev_flow *flow;
773 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
778 flow = dp_netdev_lookup_flow(dp, &key);
780 if (put->flags & DPIF_FP_CREATE) {
781 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
783 memset(put->stats, 0, sizeof *put->stats);
785 return dp_netdev_flow_add(dp, &key, put->actions,
794 if (put->flags & DPIF_FP_MODIFY) {
795 int error = set_flow_actions(flow, put->actions, put->actions_len);
798 get_dpif_flow_stats(flow, put->stats);
800 if (put->flags & DPIF_FP_ZERO_STATS) {
812 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
814 struct dp_netdev *dp = get_dp_netdev(dpif);
815 struct dp_netdev_flow *flow;
819 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
824 flow = dp_netdev_lookup_flow(dp, &key);
827 get_dpif_flow_stats(flow, del->stats);
829 dp_netdev_free_flow(dp, flow);
836 struct dp_netdev_flow_state {
839 struct nlattr *actions;
840 struct odputil_keybuf keybuf;
841 struct dpif_flow_stats stats;
845 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
847 struct dp_netdev_flow_state *state;
849 *statep = state = xmalloc(sizeof *state);
852 state->actions = NULL;
857 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
858 const struct nlattr **key, size_t *key_len,
859 const struct nlattr **actions, size_t *actions_len,
860 const struct dpif_flow_stats **stats)
862 struct dp_netdev_flow_state *state = state_;
863 struct dp_netdev *dp = get_dp_netdev(dpif);
864 struct dp_netdev_flow *flow;
865 struct hmap_node *node;
867 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
872 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
877 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
878 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
885 free(state->actions);
886 state->actions = xmemdup(flow->actions, flow->actions_len);
888 *actions = state->actions;
889 *actions_len = flow->actions_len;
893 get_dpif_flow_stats(flow, &state->stats);
894 *stats = &state->stats;
901 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
903 struct dp_netdev_flow_state *state = state_;
905 free(state->actions);
911 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
913 struct dp_netdev *dp = get_dp_netdev(dpif);
918 if (execute->packet->size < ETH_HEADER_LEN ||
919 execute->packet->size > UINT16_MAX) {
923 /* Make a deep copy of 'packet', because we might modify its data. */
924 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
925 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
926 ofpbuf_put(©, execute->packet->data, execute->packet->size);
928 flow_extract(©, 0, NULL, -1, &key);
929 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
932 dp_netdev_execute_actions(dp, ©, &key,
933 execute->actions, execute->actions_len);
936 ofpbuf_uninit(©);
941 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
947 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
948 uint32_t queue_id, uint32_t *priority)
950 *priority = queue_id;
954 static struct dp_netdev_queue *
955 find_nonempty_queue(struct dpif *dpif)
957 struct dp_netdev *dp = get_dp_netdev(dpif);
960 for (i = 0; i < N_QUEUES; i++) {
961 struct dp_netdev_queue *q = &dp->queues[i];
962 if (q->head != q->tail) {
970 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
973 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
975 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
978 upcall->packet = buf;
990 dpif_netdev_recv_wait(struct dpif *dpif)
992 if (find_nonempty_queue(dpif)) {
993 poll_immediate_wake();
995 /* No messages ready to be received, and dp_wait() will ensure that we
996 * wake up to queue new messages, so there is nothing to do. */
1001 dpif_netdev_recv_purge(struct dpif *dpif)
1003 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1004 dp_netdev_purge_queues(dpif_netdev->dp);
1008 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1010 flow->used = time_msec();
1011 flow->packet_count++;
1012 flow->byte_count += packet->size;
1013 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1017 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1018 struct ofpbuf *packet)
1020 struct dp_netdev_flow *flow;
1023 if (packet->size < ETH_HEADER_LEN) {
1026 flow_extract(packet, 0, NULL, port->port_no, &key);
1027 flow = dp_netdev_lookup_flow(dp, &key);
1029 dp_netdev_flow_used(flow, packet);
1030 dp_netdev_execute_actions(dp, packet, &key,
1031 flow->actions, flow->actions_len);
1035 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1040 dpif_netdev_run(struct dpif *dpif)
1042 struct dp_netdev *dp = get_dp_netdev(dpif);
1043 struct dp_netdev_port *port;
1044 struct ofpbuf packet;
1046 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1048 LIST_FOR_EACH (port, node, &dp->port_list) {
1051 /* Reset packet contents. */
1052 ofpbuf_clear(&packet);
1053 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1055 error = netdev_recv(port->netdev, &packet);
1057 dp_netdev_port_input(dp, port, &packet);
1058 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1060 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1061 netdev_get_name(port->netdev), strerror(error));
1064 ofpbuf_uninit(&packet);
1068 dpif_netdev_wait(struct dpif *dpif)
1070 struct dp_netdev *dp = get_dp_netdev(dpif);
1071 struct dp_netdev_port *port;
1073 LIST_FOR_EACH (port, node, &dp->port_list) {
1074 netdev_recv_wait(port->netdev);
1079 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1081 struct eth_header *eh = packet->l2;
1083 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1084 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1088 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1091 struct dp_netdev_port *p = dp->ports[out_port];
1093 netdev_send(p->netdev, packet);
1098 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1099 int queue_no, const struct flow *flow, uint64_t arg)
1101 struct dp_netdev_queue *q = &dp->queues[queue_no];
1102 struct dp_netdev_upcall *u;
1103 struct dpif_upcall *upcall;
1107 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1112 u = &q->upcalls[q->head++ & QUEUE_MASK];
1115 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1116 odp_flow_key_from_flow(buf, flow, flow->in_port);
1117 key_len = buf->size;
1118 ofpbuf_pull(buf, key_len);
1119 ofpbuf_reserve(buf, 2);
1120 ofpbuf_put(buf, packet->data, packet->size);
1122 upcall = &u->upcall;
1123 upcall->type = queue_no;
1124 upcall->packet = buf;
1125 upcall->key = buf->base;
1126 upcall->key_len = key_len;
1127 upcall->userdata = arg;
1133 dp_netdev_sample(struct dp_netdev *dp,
1134 struct ofpbuf *packet, struct flow *key,
1135 const struct nlattr *action)
1137 const struct nlattr *subactions = NULL;
1138 const struct nlattr *a;
1141 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1142 int type = nl_attr_type(a);
1144 switch ((enum ovs_sample_attr) type) {
1145 case OVS_SAMPLE_ATTR_PROBABILITY:
1146 if (random_uint32() >= nl_attr_get_u32(a)) {
1151 case OVS_SAMPLE_ATTR_ACTIONS:
1155 case OVS_SAMPLE_ATTR_UNSPEC:
1156 case __OVS_SAMPLE_ATTR_MAX:
1162 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1163 nl_attr_get_size(subactions));
1167 dp_netdev_action_userspace(struct dp_netdev *dp,
1168 struct ofpbuf *packet, struct flow *key,
1169 const struct nlattr *a)
1171 const struct nlattr *userdata_attr;
1174 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1175 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1176 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1180 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1182 enum ovs_key_attr type = nl_attr_type(a);
1183 const struct ovs_key_ipv4 *ipv4_key;
1184 const struct ovs_key_ipv6 *ipv6_key;
1185 const struct ovs_key_tcp *tcp_key;
1186 const struct ovs_key_udp *udp_key;
1189 case OVS_KEY_ATTR_TUN_ID:
1190 case OVS_KEY_ATTR_PRIORITY:
1191 case OVS_KEY_ATTR_IPV4_TUNNEL:
1192 /* not implemented */
1195 case OVS_KEY_ATTR_ETHERNET:
1196 dp_netdev_set_dl(packet,
1197 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1200 case OVS_KEY_ATTR_IPV4:
1201 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1202 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1203 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1206 case OVS_KEY_ATTR_IPV6:
1207 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1208 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1209 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1210 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1213 case OVS_KEY_ATTR_TCP:
1214 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1215 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1218 case OVS_KEY_ATTR_UDP:
1219 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1220 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1223 case OVS_KEY_ATTR_UNSPEC:
1224 case OVS_KEY_ATTR_ENCAP:
1225 case OVS_KEY_ATTR_ETHERTYPE:
1226 case OVS_KEY_ATTR_IN_PORT:
1227 case OVS_KEY_ATTR_VLAN:
1228 case OVS_KEY_ATTR_ICMP:
1229 case OVS_KEY_ATTR_ICMPV6:
1230 case OVS_KEY_ATTR_ARP:
1231 case OVS_KEY_ATTR_ND:
1232 case __OVS_KEY_ATTR_MAX:
1239 dp_netdev_execute_actions(struct dp_netdev *dp,
1240 struct ofpbuf *packet, struct flow *key,
1241 const struct nlattr *actions,
1244 const struct nlattr *a;
1247 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1248 const struct ovs_action_push_vlan *vlan;
1249 int type = nl_attr_type(a);
1251 switch ((enum ovs_action_attr) type) {
1252 case OVS_ACTION_ATTR_OUTPUT:
1253 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1256 case OVS_ACTION_ATTR_USERSPACE:
1257 dp_netdev_action_userspace(dp, packet, key, a);
1260 case OVS_ACTION_ATTR_PUSH_VLAN:
1261 vlan = nl_attr_get(a);
1262 eth_push_vlan(packet, vlan->vlan_tci);
1265 case OVS_ACTION_ATTR_POP_VLAN:
1266 eth_pop_vlan(packet);
1269 case OVS_ACTION_ATTR_SET:
1270 execute_set_action(packet, nl_attr_get(a));
1273 case OVS_ACTION_ATTR_SAMPLE:
1274 dp_netdev_sample(dp, packet, key, a);
1277 case OVS_ACTION_ATTR_UNSPEC:
1278 case __OVS_ACTION_ATTR_MAX:
1284 const struct dpif_class dpif_netdev_class = {
1286 dpif_netdev_enumerate,
1289 dpif_netdev_destroy,
1292 dpif_netdev_get_stats,
1293 dpif_netdev_port_add,
1294 dpif_netdev_port_del,
1295 dpif_netdev_port_query_by_number,
1296 dpif_netdev_port_query_by_name,
1297 dpif_netdev_get_max_ports,
1298 NULL, /* port_get_pid */
1299 dpif_netdev_port_dump_start,
1300 dpif_netdev_port_dump_next,
1301 dpif_netdev_port_dump_done,
1302 dpif_netdev_port_poll,
1303 dpif_netdev_port_poll_wait,
1304 dpif_netdev_flow_get,
1305 dpif_netdev_flow_put,
1306 dpif_netdev_flow_del,
1307 dpif_netdev_flow_flush,
1308 dpif_netdev_flow_dump_start,
1309 dpif_netdev_flow_dump_next,
1310 dpif_netdev_flow_dump_done,
1311 dpif_netdev_execute,
1313 dpif_netdev_recv_set,
1314 dpif_netdev_queue_to_priority,
1316 dpif_netdev_recv_wait,
1317 dpif_netdev_recv_purge,
1321 dpif_dummy_register__(const char *type)
1323 struct dpif_class *class;
1325 class = xmalloc(sizeof *class);
1326 *class = dpif_netdev_class;
1327 class->type = xstrdup(type);
1328 dp_register_provider(class);
1332 dpif_dummy_register(bool override)
1339 dp_enumerate_types(&types);
1340 SSET_FOR_EACH (type, &types) {
1341 if (!dp_unregister_provider(type)) {
1342 dpif_dummy_register__(type);
1345 sset_destroy(&types);
1348 dpif_dummy_register__("dummy");