2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
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_queue {
74 struct dpif_upcall *upcalls[MAX_QUEUE_LEN];
75 unsigned int head, tail;
78 /* Datapath based on the network device interface from netdev.h. */
80 const struct dpif_class *class;
85 struct dp_netdev_queue queues[N_QUEUES];
86 struct hmap flow_table; /* Flow table. */
89 long long int n_hit; /* Number of flow table matches. */
90 long long int n_missed; /* Number of flow table misses. */
91 long long int n_lost; /* Number of misses not passed to client. */
94 struct dp_netdev_port *ports[MAX_PORTS];
95 struct list port_list;
99 /* A port in a netdev-based datapath. */
100 struct dp_netdev_port {
101 int port_no; /* Index into dp_netdev's 'ports'. */
102 struct list node; /* Element in dp_netdev's 'port_list'. */
103 struct netdev *netdev;
104 char *type; /* Port type as requested by user. */
107 /* A flow in dp_netdev's 'flow_table'. */
108 struct dp_netdev_flow {
109 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
113 long long int used; /* Last used time, in monotonic msecs. */
114 long long int packet_count; /* Number of packets matched. */
115 long long int byte_count; /* Number of bytes matched. */
116 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
119 struct nlattr *actions;
123 /* Interface to netdev-based datapath. */
126 struct dp_netdev *dp;
127 unsigned int dp_serial;
130 /* All netdev-based datapaths. */
131 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
133 /* Maximum port MTU seen so far. */
134 static int max_mtu = ETH_PAYLOAD_MAX;
136 static int get_port_by_number(struct dp_netdev *, uint16_t port_no,
137 struct dp_netdev_port **portp);
138 static int get_port_by_name(struct dp_netdev *, const char *devname,
139 struct dp_netdev_port **portp);
140 static void dp_netdev_free(struct dp_netdev *);
141 static void dp_netdev_flow_flush(struct dp_netdev *);
142 static int do_add_port(struct dp_netdev *, const char *devname,
143 const char *type, uint16_t port_no);
144 static int do_del_port(struct dp_netdev *, uint16_t port_no);
145 static int dpif_netdev_open(const struct dpif_class *, const char *name,
146 bool create, struct dpif **);
147 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
148 int queue_no, const struct flow *,
150 static void dp_netdev_execute_actions(struct dp_netdev *,
151 struct ofpbuf *, struct flow *,
152 const struct nlattr *actions,
155 static struct dpif_netdev *
156 dpif_netdev_cast(const struct dpif *dpif)
158 assert(dpif->dpif_class->open == dpif_netdev_open);
159 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
162 static struct dp_netdev *
163 get_dp_netdev(const struct dpif *dpif)
165 return dpif_netdev_cast(dpif)->dp;
169 create_dpif_netdev(struct dp_netdev *dp)
171 uint16_t netflow_id = hash_string(dp->name, 0);
172 struct dpif_netdev *dpif;
176 dpif = xmalloc(sizeof *dpif);
177 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
179 dpif->dp_serial = dp->serial;
185 create_dp_netdev(const char *name, const struct dpif_class *class,
186 struct dp_netdev **dpp)
188 struct dp_netdev *dp;
192 dp = xzalloc(sizeof *dp);
194 dp->name = xstrdup(name);
196 for (i = 0; i < N_QUEUES; i++) {
197 dp->queues[i].head = dp->queues[i].tail = 0;
199 hmap_init(&dp->flow_table);
200 list_init(&dp->port_list);
201 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
207 shash_add(&dp_netdevs, name, dp);
214 dpif_netdev_open(const struct dpif_class *class, const char *name,
215 bool create, struct dpif **dpifp)
217 struct dp_netdev *dp;
219 dp = shash_find_data(&dp_netdevs, name);
224 int error = create_dp_netdev(name, class, &dp);
231 if (dp->class != class) {
238 *dpifp = create_dpif_netdev(dp);
243 dp_netdev_purge_queues(struct dp_netdev *dp)
247 for (i = 0; i < N_QUEUES; i++) {
248 struct dp_netdev_queue *q = &dp->queues[i];
250 while (q->tail != q->head) {
251 struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK];
253 ofpbuf_delete(upcall->packet);
260 dp_netdev_free(struct dp_netdev *dp)
262 struct dp_netdev_port *port, *next;
264 dp_netdev_flow_flush(dp);
265 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
266 do_del_port(dp, port->port_no);
268 dp_netdev_purge_queues(dp);
269 hmap_destroy(&dp->flow_table);
275 dpif_netdev_close(struct dpif *dpif)
277 struct dp_netdev *dp = get_dp_netdev(dpif);
278 assert(dp->open_cnt > 0);
279 if (--dp->open_cnt == 0 && dp->destroyed) {
280 shash_find_and_delete(&dp_netdevs, dp->name);
287 dpif_netdev_destroy(struct dpif *dpif)
289 struct dp_netdev *dp = get_dp_netdev(dpif);
290 dp->destroyed = true;
295 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
297 struct dp_netdev *dp = get_dp_netdev(dpif);
298 stats->n_flows = hmap_count(&dp->flow_table);
299 stats->n_hit = dp->n_hit;
300 stats->n_missed = dp->n_missed;
301 stats->n_lost = dp->n_lost;
305 static const char* internal_port_type(const struct dp_netdev* dp)
307 if (dp->class == &dpif_netdev_class)
309 if (dp->class == &dpif_planetlab_class)
315 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
318 struct dp_netdev_port *port;
319 struct netdev *netdev;
320 const char *open_type;
324 /* XXX reject devices already in some dp_netdev. */
326 /* Open and validate network device. */
327 open_type = (strcmp(type, "internal") ? type : internal_port_type(dp));
328 error = netdev_open(devname, open_type, &netdev);
332 /* XXX reject loopback devices */
333 /* XXX reject non-Ethernet devices */
335 error = netdev_listen(netdev);
337 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
338 devname, strerror(errno));
339 netdev_close(netdev);
343 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
345 netdev_close(netdev);
349 port = xmalloc(sizeof *port);
350 port->port_no = port_no;
351 port->netdev = netdev;
352 port->type = xstrdup(type);
354 error = netdev_get_mtu(netdev, &mtu);
359 list_push_back(&dp->port_list, &port->node);
360 dp->ports[port_no] = port;
367 choose_port(struct dpif *dpif, struct netdev *netdev)
369 struct dp_netdev *dp = get_dp_netdev(dpif);
372 if (dpif->dpif_class != &dpif_netdev_class &&
373 dpif->dpif_class != &dpif_planetlab_class)
375 /* If the port name contains a number, try to assign that port number.
376 * This can make writing unit tests easier because port numbers are
380 for (p = netdev_get_name(netdev); *p != '\0'; p++) {
381 if (isdigit((unsigned char) *p)) {
382 port_no = strtol(p, NULL, 10);
383 if (port_no > 0 && port_no < MAX_PORTS
384 && !dp->ports[port_no]) {
392 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
393 if (!dp->ports[port_no]) {
402 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
405 struct dp_netdev *dp = get_dp_netdev(dpif);
408 port_no = choose_port(dpif, netdev);
411 return do_add_port(dp, netdev_get_name(netdev),
412 netdev_get_type(netdev), port_no);
418 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
420 struct dp_netdev *dp = get_dp_netdev(dpif);
421 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
425 is_valid_port_number(uint16_t port_no)
427 return port_no < MAX_PORTS;
431 get_port_by_number(struct dp_netdev *dp,
432 uint16_t port_no, struct dp_netdev_port **portp)
434 if (!is_valid_port_number(port_no)) {
438 *portp = dp->ports[port_no];
439 return *portp ? 0 : ENOENT;
444 get_port_by_name(struct dp_netdev *dp,
445 const char *devname, struct dp_netdev_port **portp)
447 struct dp_netdev_port *port;
449 LIST_FOR_EACH (port, node, &dp->port_list) {
450 if (!strcmp(netdev_get_name(port->netdev), devname)) {
459 do_del_port(struct dp_netdev *dp, uint16_t port_no)
461 struct dp_netdev_port *port;
465 error = get_port_by_number(dp, port_no, &port);
470 list_remove(&port->node);
471 dp->ports[port->port_no] = NULL;
474 name = xstrdup(netdev_get_name(port->netdev));
475 netdev_close(port->netdev);
485 answer_port_query(const struct dp_netdev_port *port,
486 struct dpif_port *dpif_port)
488 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
489 dpif_port->type = xstrdup(port->type);
490 dpif_port->port_no = port->port_no;
494 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
495 struct dpif_port *dpif_port)
497 struct dp_netdev *dp = get_dp_netdev(dpif);
498 struct dp_netdev_port *port;
501 error = get_port_by_number(dp, port_no, &port);
503 answer_port_query(port, dpif_port);
509 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
510 struct dpif_port *dpif_port)
512 struct dp_netdev *dp = get_dp_netdev(dpif);
513 struct dp_netdev_port *port;
516 error = get_port_by_name(dp, devname, &port);
518 answer_port_query(port, dpif_port);
524 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
530 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
532 hmap_remove(&dp->flow_table, &flow->node);
538 dp_netdev_flow_flush(struct dp_netdev *dp)
540 struct dp_netdev_flow *flow, *next;
542 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
543 dp_netdev_free_flow(dp, flow);
548 dpif_netdev_flow_flush(struct dpif *dpif)
550 struct dp_netdev *dp = get_dp_netdev(dpif);
551 dp_netdev_flow_flush(dp);
555 struct dp_netdev_port_state {
561 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
563 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
568 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
569 struct dpif_port *dpif_port)
571 struct dp_netdev_port_state *state = state_;
572 struct dp_netdev *dp = get_dp_netdev(dpif);
575 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
576 struct dp_netdev_port *port = dp->ports[port_no];
579 state->name = xstrdup(netdev_get_name(port->netdev));
580 dpif_port->name = state->name;
581 dpif_port->type = port->type;
582 dpif_port->port_no = port->port_no;
583 state->port_no = port_no + 1;
591 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
593 struct dp_netdev_port_state *state = state_;
600 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
602 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
603 if (dpif->dp_serial != dpif->dp->serial) {
604 dpif->dp_serial = dpif->dp->serial;
612 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
614 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
615 if (dpif->dp_serial != dpif->dp->serial) {
616 poll_immediate_wake();
620 static struct dp_netdev_flow *
621 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
623 struct dp_netdev_flow *flow;
625 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
626 if (flow_equal(&flow->key, key)) {
634 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
636 stats->n_packets = flow->packet_count;
637 stats->n_bytes = flow->byte_count;
638 stats->used = flow->used;
639 stats->tcp_flags = flow->tcp_flags;
643 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
646 if (odp_flow_key_to_flow(key, key_len, flow)) {
647 /* This should not happen: it indicates that odp_flow_key_from_flow()
648 * and odp_flow_key_to_flow() disagree on the acceptable form of a
649 * flow. Log the problem as an error, with enough details to enable
651 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
653 if (!VLOG_DROP_ERR(&rl)) {
657 odp_flow_key_format(key, key_len, &s);
658 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
665 if (flow->in_port < OFPP_MAX
666 ? flow->in_port >= MAX_PORTS
667 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
675 dpif_netdev_flow_get(const struct dpif *dpif,
676 const struct nlattr *nl_key, size_t nl_key_len,
677 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
679 struct dp_netdev *dp = get_dp_netdev(dpif);
680 struct dp_netdev_flow *flow;
684 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
689 flow = dp_netdev_lookup_flow(dp, &key);
695 get_dpif_flow_stats(flow, stats);
698 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
704 set_flow_actions(struct dp_netdev_flow *flow,
705 const struct nlattr *actions, size_t actions_len)
707 flow->actions = xrealloc(flow->actions, actions_len);
708 flow->actions_len = actions_len;
709 memcpy(flow->actions, actions, actions_len);
714 add_flow(struct dpif *dpif, const struct flow *key,
715 const struct nlattr *actions, size_t actions_len)
717 struct dp_netdev *dp = get_dp_netdev(dpif);
718 struct dp_netdev_flow *flow;
721 flow = xzalloc(sizeof *flow);
724 error = set_flow_actions(flow, actions, actions_len);
730 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
735 clear_stats(struct dp_netdev_flow *flow)
738 flow->packet_count = 0;
739 flow->byte_count = 0;
744 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
746 struct dp_netdev *dp = get_dp_netdev(dpif);
747 struct dp_netdev_flow *flow;
751 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
756 flow = dp_netdev_lookup_flow(dp, &key);
758 if (put->flags & DPIF_FP_CREATE) {
759 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
761 memset(put->stats, 0, sizeof *put->stats);
763 return add_flow(dpif, &key, put->actions, put->actions_len);
771 if (put->flags & DPIF_FP_MODIFY) {
772 int error = set_flow_actions(flow, put->actions, put->actions_len);
775 get_dpif_flow_stats(flow, put->stats);
777 if (put->flags & DPIF_FP_ZERO_STATS) {
789 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
791 struct dp_netdev *dp = get_dp_netdev(dpif);
792 struct dp_netdev_flow *flow;
796 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
801 flow = dp_netdev_lookup_flow(dp, &key);
804 get_dpif_flow_stats(flow, del->stats);
806 dp_netdev_free_flow(dp, flow);
813 struct dp_netdev_flow_state {
816 struct nlattr *actions;
817 struct odputil_keybuf keybuf;
818 struct dpif_flow_stats stats;
822 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
824 struct dp_netdev_flow_state *state;
826 *statep = state = xmalloc(sizeof *state);
829 state->actions = NULL;
834 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
835 const struct nlattr **key, size_t *key_len,
836 const struct nlattr **actions, size_t *actions_len,
837 const struct dpif_flow_stats **stats)
839 struct dp_netdev_flow_state *state = state_;
840 struct dp_netdev *dp = get_dp_netdev(dpif);
841 struct dp_netdev_flow *flow;
842 struct hmap_node *node;
844 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
849 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
854 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
855 odp_flow_key_from_flow(&buf, &flow->key);
862 free(state->actions);
863 state->actions = xmemdup(flow->actions, flow->actions_len);
865 *actions = state->actions;
866 *actions_len = flow->actions_len;
870 get_dpif_flow_stats(flow, &state->stats);
871 *stats = &state->stats;
878 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
880 struct dp_netdev_flow_state *state = state_;
882 free(state->actions);
888 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
890 struct dp_netdev *dp = get_dp_netdev(dpif);
895 if (execute->packet->size < ETH_HEADER_LEN ||
896 execute->packet->size > UINT16_MAX) {
900 /* Make a deep copy of 'packet', because we might modify its data. */
901 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
902 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
903 ofpbuf_put(©, execute->packet->data, execute->packet->size);
905 flow_extract(©, 0, 0, -1, &key);
906 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
909 dp_netdev_execute_actions(dp, ©, &key,
910 execute->actions, execute->actions_len);
913 ofpbuf_uninit(©);
918 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
924 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
925 uint32_t queue_id, uint32_t *priority)
927 *priority = queue_id;
931 static struct dp_netdev_queue *
932 find_nonempty_queue(struct dpif *dpif)
934 struct dp_netdev *dp = get_dp_netdev(dpif);
937 for (i = 0; i < N_QUEUES; i++) {
938 struct dp_netdev_queue *q = &dp->queues[i];
939 if (q->head != q->tail) {
947 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
950 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
952 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
966 dpif_netdev_recv_wait(struct dpif *dpif)
968 if (find_nonempty_queue(dpif)) {
969 poll_immediate_wake();
971 /* No messages ready to be received, and dp_wait() will ensure that we
972 * wake up to queue new messages, so there is nothing to do. */
977 dpif_netdev_recv_purge(struct dpif *dpif)
979 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
980 dp_netdev_purge_queues(dpif_netdev->dp);
984 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
985 const struct ofpbuf *packet)
987 flow->used = time_msec();
988 flow->packet_count++;
989 flow->byte_count += packet->size;
990 flow->tcp_flags |= packet_get_tcp_flags(packet, key);
994 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
995 struct ofpbuf *packet)
997 struct dp_netdev_flow *flow;
1000 if (packet->size < ETH_HEADER_LEN) {
1003 flow_extract(packet, 0, 0, port->port_no, &key);
1004 flow = dp_netdev_lookup_flow(dp, &key);
1006 dp_netdev_flow_used(flow, &key, packet);
1007 dp_netdev_execute_actions(dp, packet, &key,
1008 flow->actions, flow->actions_len);
1012 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1017 dpif_netdev_run(struct dpif *dpif)
1019 struct dp_netdev *dp = get_dp_netdev(dpif);
1020 struct dp_netdev_port *port;
1021 struct ofpbuf packet;
1023 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1025 LIST_FOR_EACH (port, node, &dp->port_list) {
1028 /* Reset packet contents. */
1029 ofpbuf_clear(&packet);
1030 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1032 error = netdev_recv(port->netdev, &packet);
1034 dp_netdev_port_input(dp, port, &packet);
1035 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1036 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1037 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1038 netdev_get_name(port->netdev), strerror(error));
1041 ofpbuf_uninit(&packet);
1045 dpif_netdev_wait(struct dpif *dpif)
1047 struct dp_netdev *dp = get_dp_netdev(dpif);
1048 struct dp_netdev_port *port;
1050 LIST_FOR_EACH (port, node, &dp->port_list) {
1051 netdev_recv_wait(port->netdev);
1056 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1058 struct eth_header *eh = packet->l2;
1060 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1061 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1065 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1068 struct dp_netdev_port *p = dp->ports[out_port];
1070 netdev_send(p->netdev, packet);
1075 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1076 int queue_no, const struct flow *flow, uint64_t arg)
1078 struct dp_netdev_queue *q = &dp->queues[queue_no];
1079 struct dpif_upcall *upcall;
1083 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1088 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1089 odp_flow_key_from_flow(buf, flow);
1090 key_len = buf->size;
1091 ofpbuf_pull(buf, key_len);
1092 ofpbuf_reserve(buf, 2);
1093 ofpbuf_put(buf, packet->data, packet->size);
1095 upcall = xzalloc(sizeof *upcall);
1096 upcall->type = queue_no;
1097 upcall->packet = buf;
1098 upcall->key = buf->base;
1099 upcall->key_len = key_len;
1100 upcall->userdata = arg;
1102 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1108 dp_netdev_sample(struct dp_netdev *dp,
1109 struct ofpbuf *packet, struct flow *key,
1110 const struct nlattr *action)
1112 const struct nlattr *subactions = NULL;
1113 const struct nlattr *a;
1116 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1117 int type = nl_attr_type(a);
1119 switch ((enum ovs_sample_attr) type) {
1120 case OVS_SAMPLE_ATTR_PROBABILITY:
1121 if (random_uint32() >= nl_attr_get_u32(a)) {
1126 case OVS_SAMPLE_ATTR_ACTIONS:
1130 case OVS_SAMPLE_ATTR_UNSPEC:
1131 case __OVS_SAMPLE_ATTR_MAX:
1137 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1138 nl_attr_get_size(subactions));
1142 dp_netdev_action_userspace(struct dp_netdev *dp,
1143 struct ofpbuf *packet, struct flow *key,
1144 const struct nlattr *a)
1146 const struct nlattr *userdata_attr;
1149 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1150 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1151 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1155 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1157 enum ovs_key_attr type = nl_attr_type(a);
1158 const struct ovs_key_ipv4 *ipv4_key;
1159 const struct ovs_key_tcp *tcp_key;
1160 const struct ovs_key_udp *udp_key;
1163 case OVS_KEY_ATTR_TUN_ID:
1164 case OVS_KEY_ATTR_PRIORITY:
1165 case OVS_KEY_ATTR_IPV6:
1166 /* not implemented */
1169 case OVS_KEY_ATTR_ETHERNET:
1170 dp_netdev_set_dl(packet,
1171 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1174 case OVS_KEY_ATTR_IPV4:
1175 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1176 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1177 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1180 case OVS_KEY_ATTR_TCP:
1181 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1182 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1185 case OVS_KEY_ATTR_UDP:
1186 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1187 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1190 case OVS_KEY_ATTR_UNSPEC:
1191 case OVS_KEY_ATTR_ENCAP:
1192 case OVS_KEY_ATTR_ETHERTYPE:
1193 case OVS_KEY_ATTR_IN_PORT:
1194 case OVS_KEY_ATTR_VLAN:
1195 case OVS_KEY_ATTR_ICMP:
1196 case OVS_KEY_ATTR_ICMPV6:
1197 case OVS_KEY_ATTR_ARP:
1198 case OVS_KEY_ATTR_ND:
1199 case __OVS_KEY_ATTR_MAX:
1206 dp_netdev_execute_actions(struct dp_netdev *dp,
1207 struct ofpbuf *packet, struct flow *key,
1208 const struct nlattr *actions,
1211 const struct nlattr *a;
1214 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1215 const struct ovs_action_push_vlan *vlan;
1216 int type = nl_attr_type(a);
1218 switch ((enum ovs_action_attr) type) {
1219 case OVS_ACTION_ATTR_OUTPUT:
1220 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1223 case OVS_ACTION_ATTR_USERSPACE:
1224 dp_netdev_action_userspace(dp, packet, key, a);
1227 case OVS_ACTION_ATTR_PUSH_VLAN:
1228 vlan = nl_attr_get(a);
1229 eth_push_vlan(packet, vlan->vlan_tci);
1232 case OVS_ACTION_ATTR_POP_VLAN:
1233 eth_pop_vlan(packet);
1236 case OVS_ACTION_ATTR_SET:
1237 execute_set_action(packet, nl_attr_get(a));
1240 case OVS_ACTION_ATTR_SAMPLE:
1241 dp_netdev_sample(dp, packet, key, a);
1244 case OVS_ACTION_ATTR_UNSPEC:
1245 case __OVS_ACTION_ATTR_MAX:
1251 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1252 NULL, /* enumerate */ \
1254 dpif_netdev_close, \
1255 dpif_netdev_destroy, \
1258 dpif_netdev_get_stats, \
1259 dpif_netdev_port_add, \
1260 dpif_netdev_port_del, \
1261 dpif_netdev_port_query_by_number, \
1262 dpif_netdev_port_query_by_name, \
1263 dpif_netdev_get_max_ports, \
1264 NULL, /* port_get_pid */ \
1265 dpif_netdev_port_dump_start, \
1266 dpif_netdev_port_dump_next, \
1267 dpif_netdev_port_dump_done, \
1268 dpif_netdev_port_poll, \
1269 dpif_netdev_port_poll_wait, \
1270 dpif_netdev_flow_get, \
1271 dpif_netdev_flow_put, \
1272 dpif_netdev_flow_del, \
1273 dpif_netdev_flow_flush, \
1274 dpif_netdev_flow_dump_start, \
1275 dpif_netdev_flow_dump_next, \
1276 dpif_netdev_flow_dump_done, \
1277 dpif_netdev_execute, \
1278 NULL, /* operate */ \
1279 dpif_netdev_recv_set, \
1280 dpif_netdev_queue_to_priority, \
1282 dpif_netdev_recv_wait, \
1283 dpif_netdev_recv_purge, \
1285 const struct dpif_class dpif_netdev_class = {
1287 DPIF_NETDEV_CLASS_FUNCTIONS
1290 const struct dpif_class dpif_planetlab_class = {
1292 DPIF_NETDEV_CLASS_FUNCTIONS
1296 dpif_dummy_register__(const char *type)
1298 struct dpif_class *class;
1300 class = xmalloc(sizeof *class);
1301 *class = dpif_netdev_class;
1302 class->type = xstrdup(type);
1303 dp_register_provider(class);
1307 dpif_dummy_register(bool override)
1314 dp_enumerate_types(&types);
1315 SSET_FOR_EACH (type, &types) {
1316 if (!dp_unregister_provider(type)) {
1317 dpif_dummy_register__(type);
1320 sset_destroy(&types);
1323 dpif_dummy_register__("dummy");