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;
247 dp = xzalloc(sizeof *dp);
249 dp->name = xstrdup(name);
251 for (i = 0; i < N_QUEUES; i++) {
252 dp->queues[i].head = dp->queues[i].tail = 0;
254 hmap_init(&dp->flow_table);
255 list_init(&dp->port_list);
257 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
263 shash_add(&dp_netdevs, name, dp);
270 dpif_netdev_open(const struct dpif_class *class, const char *name,
271 bool create, struct dpif **dpifp)
273 struct dp_netdev *dp;
275 dp = shash_find_data(&dp_netdevs, name);
280 int error = create_dp_netdev(name, class, &dp);
287 if (dp->class != class) {
294 *dpifp = create_dpif_netdev(dp);
299 dp_netdev_purge_queues(struct dp_netdev *dp)
303 for (i = 0; i < N_QUEUES; i++) {
304 struct dp_netdev_queue *q = &dp->queues[i];
306 while (q->tail != q->head) {
307 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
308 ofpbuf_uninit(&u->buf);
314 dp_netdev_free(struct dp_netdev *dp)
316 struct dp_netdev_port *port, *next;
318 dp_netdev_flow_flush(dp);
319 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
320 do_del_port(dp, port->port_no);
322 dp_netdev_purge_queues(dp);
323 hmap_destroy(&dp->flow_table);
329 dpif_netdev_close(struct dpif *dpif)
331 struct dp_netdev *dp = get_dp_netdev(dpif);
332 assert(dp->open_cnt > 0);
333 if (--dp->open_cnt == 0 && dp->destroyed) {
334 shash_find_and_delete(&dp_netdevs, dp->name);
341 dpif_netdev_destroy(struct dpif *dpif)
343 struct dp_netdev *dp = get_dp_netdev(dpif);
344 dp->destroyed = true;
349 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
351 struct dp_netdev *dp = get_dp_netdev(dpif);
352 stats->n_flows = hmap_count(&dp->flow_table);
353 stats->n_hit = dp->n_hit;
354 stats->n_missed = dp->n_missed;
355 stats->n_lost = dp->n_lost;
360 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
363 struct dp_netdev_port *port;
364 struct netdev *netdev;
365 const char *open_type;
369 /* XXX reject devices already in some dp_netdev. */
371 /* Open and validate network device. */
372 open_type = (strcmp(type, "internal") ? type
373 : dp->class != &dpif_netdev_class ? "dummy"
375 error = netdev_open(devname, open_type, &netdev);
379 /* XXX reject loopback devices */
380 /* XXX reject non-Ethernet devices */
382 error = netdev_listen(netdev);
384 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
385 devname, strerror(errno));
386 netdev_close(netdev);
390 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
392 netdev_close(netdev);
396 port = xmalloc(sizeof *port);
397 port->port_no = port_no;
398 port->netdev = netdev;
399 port->type = xstrdup(type);
401 error = netdev_get_mtu(netdev, &mtu);
406 list_push_back(&dp->port_list, &port->node);
407 dp->ports[port_no] = port;
414 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
417 struct dp_netdev *dp = get_dp_netdev(dpif);
420 if (*port_nop != UINT32_MAX) {
421 if (*port_nop >= MAX_PORTS) {
423 } else if (dp->ports[*port_nop]) {
428 port_no = choose_port(dp, netdev_get_name(netdev));
432 return do_add_port(dp, netdev_get_name(netdev),
433 netdev_get_type(netdev), port_no);
439 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
441 struct dp_netdev *dp = get_dp_netdev(dpif);
442 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
446 is_valid_port_number(uint32_t port_no)
448 return port_no < MAX_PORTS;
452 get_port_by_number(struct dp_netdev *dp,
453 uint32_t port_no, struct dp_netdev_port **portp)
455 if (!is_valid_port_number(port_no)) {
459 *portp = dp->ports[port_no];
460 return *portp ? 0 : ENOENT;
465 get_port_by_name(struct dp_netdev *dp,
466 const char *devname, struct dp_netdev_port **portp)
468 struct dp_netdev_port *port;
470 LIST_FOR_EACH (port, node, &dp->port_list) {
471 if (!strcmp(netdev_get_name(port->netdev), devname)) {
480 do_del_port(struct dp_netdev *dp, uint32_t port_no)
482 struct dp_netdev_port *port;
486 error = get_port_by_number(dp, port_no, &port);
491 list_remove(&port->node);
492 dp->ports[port->port_no] = NULL;
495 name = xstrdup(netdev_get_name(port->netdev));
496 netdev_close(port->netdev);
506 answer_port_query(const struct dp_netdev_port *port,
507 struct dpif_port *dpif_port)
509 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
510 dpif_port->type = xstrdup(port->type);
511 dpif_port->port_no = port->port_no;
515 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
516 struct dpif_port *dpif_port)
518 struct dp_netdev *dp = get_dp_netdev(dpif);
519 struct dp_netdev_port *port;
522 error = get_port_by_number(dp, port_no, &port);
523 if (!error && dpif_port) {
524 answer_port_query(port, dpif_port);
530 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
531 struct dpif_port *dpif_port)
533 struct dp_netdev *dp = get_dp_netdev(dpif);
534 struct dp_netdev_port *port;
537 error = get_port_by_name(dp, devname, &port);
538 if (!error && dpif_port) {
539 answer_port_query(port, dpif_port);
545 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
551 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
553 hmap_remove(&dp->flow_table, &flow->node);
559 dp_netdev_flow_flush(struct dp_netdev *dp)
561 struct dp_netdev_flow *flow, *next;
563 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
564 dp_netdev_free_flow(dp, flow);
569 dpif_netdev_flow_flush(struct dpif *dpif)
571 struct dp_netdev *dp = get_dp_netdev(dpif);
572 dp_netdev_flow_flush(dp);
576 struct dp_netdev_port_state {
582 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
584 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
589 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
590 struct dpif_port *dpif_port)
592 struct dp_netdev_port_state *state = state_;
593 struct dp_netdev *dp = get_dp_netdev(dpif);
596 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
597 struct dp_netdev_port *port = dp->ports[port_no];
600 state->name = xstrdup(netdev_get_name(port->netdev));
601 dpif_port->name = state->name;
602 dpif_port->type = port->type;
603 dpif_port->port_no = port->port_no;
604 state->port_no = port_no + 1;
612 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
614 struct dp_netdev_port_state *state = state_;
621 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
623 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
624 if (dpif->dp_serial != dpif->dp->serial) {
625 dpif->dp_serial = dpif->dp->serial;
633 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
635 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
636 if (dpif->dp_serial != dpif->dp->serial) {
637 poll_immediate_wake();
641 static struct dp_netdev_flow *
642 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
644 struct dp_netdev_flow *flow;
646 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
647 if (flow_equal(&flow->key, key)) {
655 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
657 stats->n_packets = flow->packet_count;
658 stats->n_bytes = flow->byte_count;
659 stats->used = flow->used;
660 stats->tcp_flags = flow->tcp_flags;
664 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
667 if (odp_flow_key_to_flow(key, key_len, flow)) {
668 /* This should not happen: it indicates that odp_flow_key_from_flow()
669 * and odp_flow_key_to_flow() disagree on the acceptable form of a
670 * flow. Log the problem as an error, with enough details to enable
672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
674 if (!VLOG_DROP_ERR(&rl)) {
678 odp_flow_key_format(key, key_len, &s);
679 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
686 if (flow->in_port < OFPP_MAX
687 ? flow->in_port >= MAX_PORTS
688 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
696 dpif_netdev_flow_get(const struct dpif *dpif,
697 const struct nlattr *nl_key, size_t nl_key_len,
698 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
700 struct dp_netdev *dp = get_dp_netdev(dpif);
701 struct dp_netdev_flow *flow;
705 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
710 flow = dp_netdev_lookup_flow(dp, &key);
716 get_dpif_flow_stats(flow, stats);
719 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
725 set_flow_actions(struct dp_netdev_flow *flow,
726 const struct nlattr *actions, size_t actions_len)
728 flow->actions = xrealloc(flow->actions, actions_len);
729 flow->actions_len = actions_len;
730 memcpy(flow->actions, actions, actions_len);
735 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
736 const struct nlattr *actions, size_t actions_len)
738 struct dp_netdev_flow *flow;
741 flow = xzalloc(sizeof *flow);
744 error = set_flow_actions(flow, actions, actions_len);
750 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
755 clear_stats(struct dp_netdev_flow *flow)
758 flow->packet_count = 0;
759 flow->byte_count = 0;
764 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
766 struct dp_netdev *dp = get_dp_netdev(dpif);
767 struct dp_netdev_flow *flow;
771 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
776 flow = dp_netdev_lookup_flow(dp, &key);
778 if (put->flags & DPIF_FP_CREATE) {
779 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
781 memset(put->stats, 0, sizeof *put->stats);
783 return dp_netdev_flow_add(dp, &key, put->actions,
792 if (put->flags & DPIF_FP_MODIFY) {
793 int error = set_flow_actions(flow, put->actions, put->actions_len);
796 get_dpif_flow_stats(flow, put->stats);
798 if (put->flags & DPIF_FP_ZERO_STATS) {
810 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
812 struct dp_netdev *dp = get_dp_netdev(dpif);
813 struct dp_netdev_flow *flow;
817 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
822 flow = dp_netdev_lookup_flow(dp, &key);
825 get_dpif_flow_stats(flow, del->stats);
827 dp_netdev_free_flow(dp, flow);
834 struct dp_netdev_flow_state {
837 struct nlattr *actions;
838 struct odputil_keybuf keybuf;
839 struct dpif_flow_stats stats;
843 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
845 struct dp_netdev_flow_state *state;
847 *statep = state = xmalloc(sizeof *state);
850 state->actions = NULL;
855 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
856 const struct nlattr **key, size_t *key_len,
857 const struct nlattr **actions, size_t *actions_len,
858 const struct dpif_flow_stats **stats)
860 struct dp_netdev_flow_state *state = state_;
861 struct dp_netdev *dp = get_dp_netdev(dpif);
862 struct dp_netdev_flow *flow;
863 struct hmap_node *node;
865 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
870 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
875 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
876 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
883 free(state->actions);
884 state->actions = xmemdup(flow->actions, flow->actions_len);
886 *actions = state->actions;
887 *actions_len = flow->actions_len;
891 get_dpif_flow_stats(flow, &state->stats);
892 *stats = &state->stats;
899 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
901 struct dp_netdev_flow_state *state = state_;
903 free(state->actions);
909 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
911 struct dp_netdev *dp = get_dp_netdev(dpif);
916 if (execute->packet->size < ETH_HEADER_LEN ||
917 execute->packet->size > UINT16_MAX) {
921 /* Make a deep copy of 'packet', because we might modify its data. */
922 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
923 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
924 ofpbuf_put(©, execute->packet->data, execute->packet->size);
926 flow_extract(©, 0, NULL, -1, &key);
927 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
930 dp_netdev_execute_actions(dp, ©, &key,
931 execute->actions, execute->actions_len);
934 ofpbuf_uninit(©);
939 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
945 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
946 uint32_t queue_id, uint32_t *priority)
948 *priority = queue_id;
952 static struct dp_netdev_queue *
953 find_nonempty_queue(struct dpif *dpif)
955 struct dp_netdev *dp = get_dp_netdev(dpif);
958 for (i = 0; i < N_QUEUES; i++) {
959 struct dp_netdev_queue *q = &dp->queues[i];
960 if (q->head != q->tail) {
968 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
971 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
973 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
976 upcall->packet = buf;
988 dpif_netdev_recv_wait(struct dpif *dpif)
990 if (find_nonempty_queue(dpif)) {
991 poll_immediate_wake();
993 /* No messages ready to be received, and dp_wait() will ensure that we
994 * wake up to queue new messages, so there is nothing to do. */
999 dpif_netdev_recv_purge(struct dpif *dpif)
1001 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1002 dp_netdev_purge_queues(dpif_netdev->dp);
1006 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1008 flow->used = time_msec();
1009 flow->packet_count++;
1010 flow->byte_count += packet->size;
1011 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1015 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1016 struct ofpbuf *packet)
1018 struct dp_netdev_flow *flow;
1021 if (packet->size < ETH_HEADER_LEN) {
1024 flow_extract(packet, 0, NULL, port->port_no, &key);
1025 flow = dp_netdev_lookup_flow(dp, &key);
1027 dp_netdev_flow_used(flow, packet);
1028 dp_netdev_execute_actions(dp, packet, &key,
1029 flow->actions, flow->actions_len);
1033 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1038 dpif_netdev_run(struct dpif *dpif)
1040 struct dp_netdev *dp = get_dp_netdev(dpif);
1041 struct dp_netdev_port *port;
1042 struct ofpbuf packet;
1044 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1046 LIST_FOR_EACH (port, node, &dp->port_list) {
1049 /* Reset packet contents. */
1050 ofpbuf_clear(&packet);
1051 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1053 error = netdev_recv(port->netdev, &packet);
1055 dp_netdev_port_input(dp, port, &packet);
1056 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1057 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1058 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1059 netdev_get_name(port->netdev), strerror(error));
1062 ofpbuf_uninit(&packet);
1066 dpif_netdev_wait(struct dpif *dpif)
1068 struct dp_netdev *dp = get_dp_netdev(dpif);
1069 struct dp_netdev_port *port;
1071 LIST_FOR_EACH (port, node, &dp->port_list) {
1072 netdev_recv_wait(port->netdev);
1077 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1079 struct eth_header *eh = packet->l2;
1081 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1082 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1086 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1089 struct dp_netdev_port *p = dp->ports[out_port];
1091 netdev_send(p->netdev, packet);
1096 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1097 int queue_no, const struct flow *flow, uint64_t arg)
1099 struct dp_netdev_queue *q = &dp->queues[queue_no];
1100 struct dp_netdev_upcall *u;
1101 struct dpif_upcall *upcall;
1105 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1110 u = &q->upcalls[q->head++ & QUEUE_MASK];
1113 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1114 odp_flow_key_from_flow(buf, flow, flow->in_port);
1115 key_len = buf->size;
1116 ofpbuf_pull(buf, key_len);
1117 ofpbuf_reserve(buf, 2);
1118 ofpbuf_put(buf, packet->data, packet->size);
1120 upcall = &u->upcall;
1121 upcall->type = queue_no;
1122 upcall->packet = buf;
1123 upcall->key = buf->base;
1124 upcall->key_len = key_len;
1125 upcall->userdata = arg;
1131 dp_netdev_sample(struct dp_netdev *dp,
1132 struct ofpbuf *packet, struct flow *key,
1133 const struct nlattr *action)
1135 const struct nlattr *subactions = NULL;
1136 const struct nlattr *a;
1139 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1140 int type = nl_attr_type(a);
1142 switch ((enum ovs_sample_attr) type) {
1143 case OVS_SAMPLE_ATTR_PROBABILITY:
1144 if (random_uint32() >= nl_attr_get_u32(a)) {
1149 case OVS_SAMPLE_ATTR_ACTIONS:
1153 case OVS_SAMPLE_ATTR_UNSPEC:
1154 case __OVS_SAMPLE_ATTR_MAX:
1160 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1161 nl_attr_get_size(subactions));
1165 dp_netdev_action_userspace(struct dp_netdev *dp,
1166 struct ofpbuf *packet, struct flow *key,
1167 const struct nlattr *a)
1169 const struct nlattr *userdata_attr;
1172 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1173 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1174 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1178 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1180 enum ovs_key_attr type = nl_attr_type(a);
1181 const struct ovs_key_ipv4 *ipv4_key;
1182 const struct ovs_key_ipv6 *ipv6_key;
1183 const struct ovs_key_tcp *tcp_key;
1184 const struct ovs_key_udp *udp_key;
1187 case OVS_KEY_ATTR_TUN_ID:
1188 case OVS_KEY_ATTR_PRIORITY:
1189 case OVS_KEY_ATTR_IPV4_TUNNEL:
1190 /* not implemented */
1193 case OVS_KEY_ATTR_ETHERNET:
1194 dp_netdev_set_dl(packet,
1195 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1198 case OVS_KEY_ATTR_IPV4:
1199 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1200 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1201 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1204 case OVS_KEY_ATTR_IPV6:
1205 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1206 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1207 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1208 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1211 case OVS_KEY_ATTR_TCP:
1212 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1213 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1216 case OVS_KEY_ATTR_UDP:
1217 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1218 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1221 case OVS_KEY_ATTR_UNSPEC:
1222 case OVS_KEY_ATTR_ENCAP:
1223 case OVS_KEY_ATTR_ETHERTYPE:
1224 case OVS_KEY_ATTR_IN_PORT:
1225 case OVS_KEY_ATTR_VLAN:
1226 case OVS_KEY_ATTR_ICMP:
1227 case OVS_KEY_ATTR_ICMPV6:
1228 case OVS_KEY_ATTR_ARP:
1229 case OVS_KEY_ATTR_ND:
1230 case __OVS_KEY_ATTR_MAX:
1237 dp_netdev_execute_actions(struct dp_netdev *dp,
1238 struct ofpbuf *packet, struct flow *key,
1239 const struct nlattr *actions,
1242 const struct nlattr *a;
1245 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1246 const struct ovs_action_push_vlan *vlan;
1247 int type = nl_attr_type(a);
1249 switch ((enum ovs_action_attr) type) {
1250 case OVS_ACTION_ATTR_OUTPUT:
1251 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1254 case OVS_ACTION_ATTR_USERSPACE:
1255 dp_netdev_action_userspace(dp, packet, key, a);
1258 case OVS_ACTION_ATTR_PUSH_VLAN:
1259 vlan = nl_attr_get(a);
1260 eth_push_vlan(packet, vlan->vlan_tci);
1263 case OVS_ACTION_ATTR_POP_VLAN:
1264 eth_pop_vlan(packet);
1267 case OVS_ACTION_ATTR_SET:
1268 execute_set_action(packet, nl_attr_get(a));
1271 case OVS_ACTION_ATTR_SAMPLE:
1272 dp_netdev_sample(dp, packet, key, a);
1275 case OVS_ACTION_ATTR_UNSPEC:
1276 case __OVS_ACTION_ATTR_MAX:
1282 const struct dpif_class dpif_netdev_class = {
1284 dpif_netdev_enumerate,
1287 dpif_netdev_destroy,
1290 dpif_netdev_get_stats,
1291 dpif_netdev_port_add,
1292 dpif_netdev_port_del,
1293 dpif_netdev_port_query_by_number,
1294 dpif_netdev_port_query_by_name,
1295 dpif_netdev_get_max_ports,
1296 NULL, /* port_get_pid */
1297 dpif_netdev_port_dump_start,
1298 dpif_netdev_port_dump_next,
1299 dpif_netdev_port_dump_done,
1300 dpif_netdev_port_poll,
1301 dpif_netdev_port_poll_wait,
1302 dpif_netdev_flow_get,
1303 dpif_netdev_flow_put,
1304 dpif_netdev_flow_del,
1305 dpif_netdev_flow_flush,
1306 dpif_netdev_flow_dump_start,
1307 dpif_netdev_flow_dump_next,
1308 dpif_netdev_flow_dump_done,
1309 dpif_netdev_execute,
1311 dpif_netdev_recv_set,
1312 dpif_netdev_queue_to_priority,
1314 dpif_netdev_recv_wait,
1315 dpif_netdev_recv_purge,
1319 dpif_dummy_register__(const char *type)
1321 struct dpif_class *class;
1323 class = xmalloc(sizeof *class);
1324 *class = dpif_netdev_class;
1325 class->type = xstrdup(type);
1326 dp_register_provider(class);
1330 dpif_dummy_register(bool override)
1337 dp_enumerate_types(&types);
1338 SSET_FOR_EACH (type, &types) {
1339 if (!dp_unregister_provider(type)) {
1340 dpif_dummy_register__(type);
1343 sset_destroy(&types);
1346 dpif_dummy_register__("dummy");