2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 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"
43 #include "netdev-vport.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 struct netdev_saved_flags *sf;
110 struct netdev_rx *rx;
111 char *type; /* Port type as requested by user. */
114 /* A flow in dp_netdev's 'flow_table'. */
115 struct dp_netdev_flow {
116 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
120 long long int used; /* Last used time, in monotonic msecs. */
121 long long int packet_count; /* Number of packets matched. */
122 long long int byte_count; /* Number of bytes matched. */
123 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
126 struct nlattr *actions;
130 /* Interface to netdev-based datapath. */
133 struct dp_netdev *dp;
134 unsigned int dp_serial;
137 /* All netdev-based datapaths. */
138 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
140 /* Maximum port MTU seen so far. */
141 static int max_mtu = ETH_PAYLOAD_MAX;
143 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
144 struct dp_netdev_port **portp);
145 static int get_port_by_name(struct dp_netdev *, const char *devname,
146 struct dp_netdev_port **portp);
147 static void dp_netdev_free(struct dp_netdev *);
148 static void dp_netdev_flow_flush(struct dp_netdev *);
149 static int do_add_port(struct dp_netdev *, const char *devname,
150 const char *type, uint32_t port_no);
151 static int do_del_port(struct dp_netdev *, uint32_t port_no);
152 static int dpif_netdev_open(const struct dpif_class *, const char *name,
153 bool create, struct dpif **);
154 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
155 int queue_no, const struct flow *,
156 const struct nlattr *userdata);
157 static void dp_netdev_execute_actions(struct dp_netdev *,
158 struct ofpbuf *, struct flow *,
159 const struct nlattr *actions,
162 static struct dpif_netdev *
163 dpif_netdev_cast(const struct dpif *dpif)
165 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
166 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
169 static struct dp_netdev *
170 get_dp_netdev(const struct dpif *dpif)
172 return dpif_netdev_cast(dpif)->dp;
176 dpif_netdev_enumerate(struct sset *all_dps)
178 struct shash_node *node;
180 SHASH_FOR_EACH(node, &dp_netdevs) {
181 sset_add(all_dps, node->name);
187 dpif_netdev_class_is_dummy(const struct dpif_class *class)
189 return class != &dpif_netdev_class;
193 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
195 return strcmp(type, "internal") ? type
196 : dpif_netdev_class_is_dummy(class) ? "dummy"
201 create_dpif_netdev(struct dp_netdev *dp)
203 uint16_t netflow_id = hash_string(dp->name, 0);
204 struct dpif_netdev *dpif;
208 dpif = xmalloc(sizeof *dpif);
209 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
211 dpif->dp_serial = dp->serial;
217 choose_port(struct dp_netdev *dp, const char *name)
221 if (dp->class != &dpif_netdev_class) {
225 /* If the port name begins with "br", start the number search at
226 * 100 to make writing tests easier. */
227 if (!strncmp(name, "br", 2)) {
231 /* If the port name contains a number, try to assign that port number.
232 * This can make writing unit tests easier because port numbers are
234 for (p = name; *p != '\0'; p++) {
235 if (isdigit((unsigned char) *p)) {
236 port_no = start_no + strtol(p, NULL, 10);
237 if (port_no > 0 && port_no < MAX_PORTS
238 && !dp->ports[port_no]) {
246 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
247 if (!dp->ports[port_no]) {
256 create_dp_netdev(const char *name, const struct dpif_class *class,
257 struct dp_netdev **dpp)
259 struct dp_netdev *dp;
263 dp = xzalloc(sizeof *dp);
265 dp->name = xstrdup(name);
267 for (i = 0; i < N_QUEUES; i++) {
268 dp->queues[i].head = dp->queues[i].tail = 0;
270 hmap_init(&dp->flow_table);
271 list_init(&dp->port_list);
273 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
279 shash_add(&dp_netdevs, name, dp);
286 dpif_netdev_open(const struct dpif_class *class, const char *name,
287 bool create, struct dpif **dpifp)
289 struct dp_netdev *dp;
291 dp = shash_find_data(&dp_netdevs, name);
296 int error = create_dp_netdev(name, class, &dp);
300 ovs_assert(dp != NULL);
303 if (dp->class != class) {
310 *dpifp = create_dpif_netdev(dp);
315 dp_netdev_purge_queues(struct dp_netdev *dp)
319 for (i = 0; i < N_QUEUES; i++) {
320 struct dp_netdev_queue *q = &dp->queues[i];
322 while (q->tail != q->head) {
323 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
324 ofpbuf_uninit(&u->buf);
330 dp_netdev_free(struct dp_netdev *dp)
332 struct dp_netdev_port *port, *next;
334 dp_netdev_flow_flush(dp);
335 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
336 do_del_port(dp, port->port_no);
338 dp_netdev_purge_queues(dp);
339 hmap_destroy(&dp->flow_table);
345 dpif_netdev_close(struct dpif *dpif)
347 struct dp_netdev *dp = get_dp_netdev(dpif);
348 ovs_assert(dp->open_cnt > 0);
349 if (--dp->open_cnt == 0 && dp->destroyed) {
350 shash_find_and_delete(&dp_netdevs, dp->name);
357 dpif_netdev_destroy(struct dpif *dpif)
359 struct dp_netdev *dp = get_dp_netdev(dpif);
360 dp->destroyed = true;
365 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
367 struct dp_netdev *dp = get_dp_netdev(dpif);
368 stats->n_flows = hmap_count(&dp->flow_table);
369 stats->n_hit = dp->n_hit;
370 stats->n_missed = dp->n_missed;
371 stats->n_lost = dp->n_lost;
376 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
379 struct netdev_saved_flags *sf;
380 struct dp_netdev_port *port;
381 struct netdev *netdev;
382 struct netdev_rx *rx;
383 const char *open_type;
387 /* XXX reject devices already in some dp_netdev. */
389 /* Open and validate network device. */
390 open_type = dpif_netdev_port_open_type(dp->class, type);
391 error = netdev_open(devname, open_type, &netdev);
395 /* XXX reject loopback devices */
396 /* XXX reject non-Ethernet devices */
398 error = netdev_rx_open(netdev, &rx);
400 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
401 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
402 devname, strerror(errno));
403 netdev_close(netdev);
407 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
410 netdev_close(netdev);
414 port = xmalloc(sizeof *port);
415 port->port_no = port_no;
416 port->netdev = netdev;
419 port->type = xstrdup(type);
421 error = netdev_get_mtu(netdev, &mtu);
422 if (!error && mtu > max_mtu) {
426 list_push_back(&dp->port_list, &port->node);
427 dp->ports[port_no] = port;
434 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
437 struct dp_netdev *dp = get_dp_netdev(dpif);
440 if (*port_nop != UINT32_MAX) {
441 if (*port_nop >= MAX_PORTS) {
443 } else if (dp->ports[*port_nop]) {
448 port_no = choose_port(dp, netdev_vport_get_dpif_port(netdev));
452 return do_add_port(dp, netdev_vport_get_dpif_port(netdev),
453 netdev_get_type(netdev), port_no);
459 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
461 struct dp_netdev *dp = get_dp_netdev(dpif);
462 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
466 is_valid_port_number(uint32_t port_no)
468 return port_no < MAX_PORTS;
472 get_port_by_number(struct dp_netdev *dp,
473 uint32_t port_no, struct dp_netdev_port **portp)
475 if (!is_valid_port_number(port_no)) {
479 *portp = dp->ports[port_no];
480 return *portp ? 0 : ENOENT;
485 get_port_by_name(struct dp_netdev *dp,
486 const char *devname, struct dp_netdev_port **portp)
488 struct dp_netdev_port *port;
490 LIST_FOR_EACH (port, node, &dp->port_list) {
491 if (!strcmp(netdev_vport_get_dpif_port(port->netdev), devname)) {
500 do_del_port(struct dp_netdev *dp, uint32_t port_no)
502 struct dp_netdev_port *port;
505 error = get_port_by_number(dp, port_no, &port);
510 list_remove(&port->node);
511 dp->ports[port->port_no] = NULL;
514 netdev_close(port->netdev);
515 netdev_restore_flags(port->sf);
516 netdev_rx_close(port->rx);
524 answer_port_query(const struct dp_netdev_port *port,
525 struct dpif_port *dpif_port)
527 dpif_port->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
528 dpif_port->type = xstrdup(port->type);
529 dpif_port->port_no = port->port_no;
533 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
534 struct dpif_port *dpif_port)
536 struct dp_netdev *dp = get_dp_netdev(dpif);
537 struct dp_netdev_port *port;
540 error = get_port_by_number(dp, port_no, &port);
541 if (!error && dpif_port) {
542 answer_port_query(port, dpif_port);
548 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
549 struct dpif_port *dpif_port)
551 struct dp_netdev *dp = get_dp_netdev(dpif);
552 struct dp_netdev_port *port;
555 error = get_port_by_name(dp, devname, &port);
556 if (!error && dpif_port) {
557 answer_port_query(port, dpif_port);
563 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
569 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
571 hmap_remove(&dp->flow_table, &flow->node);
577 dp_netdev_flow_flush(struct dp_netdev *dp)
579 struct dp_netdev_flow *flow, *next;
581 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
582 dp_netdev_free_flow(dp, flow);
587 dpif_netdev_flow_flush(struct dpif *dpif)
589 struct dp_netdev *dp = get_dp_netdev(dpif);
590 dp_netdev_flow_flush(dp);
594 struct dp_netdev_port_state {
600 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
602 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
607 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
608 struct dpif_port *dpif_port)
610 struct dp_netdev_port_state *state = state_;
611 struct dp_netdev *dp = get_dp_netdev(dpif);
614 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
615 struct dp_netdev_port *port = dp->ports[port_no];
618 state->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
619 dpif_port->name = state->name;
620 dpif_port->type = port->type;
621 dpif_port->port_no = port->port_no;
622 state->port_no = port_no + 1;
630 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
632 struct dp_netdev_port_state *state = state_;
639 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
641 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
642 if (dpif->dp_serial != dpif->dp->serial) {
643 dpif->dp_serial = dpif->dp->serial;
651 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
653 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
654 if (dpif->dp_serial != dpif->dp->serial) {
655 poll_immediate_wake();
659 static struct dp_netdev_flow *
660 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
662 struct dp_netdev_flow *flow;
664 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
665 if (flow_equal(&flow->key, key)) {
673 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
675 stats->n_packets = flow->packet_count;
676 stats->n_bytes = flow->byte_count;
677 stats->used = flow->used;
678 stats->tcp_flags = flow->tcp_flags;
682 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
685 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
686 /* This should not happen: it indicates that odp_flow_key_from_flow()
687 * and odp_flow_key_to_flow() disagree on the acceptable form of a
688 * flow. Log the problem as an error, with enough details to enable
690 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
692 if (!VLOG_DROP_ERR(&rl)) {
696 odp_flow_key_format(key, key_len, &s);
697 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
704 if (flow->in_port < OFPP_MAX
705 ? flow->in_port >= MAX_PORTS
706 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
714 dpif_netdev_flow_get(const struct dpif *dpif,
715 const struct nlattr *nl_key, size_t nl_key_len,
716 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
718 struct dp_netdev *dp = get_dp_netdev(dpif);
719 struct dp_netdev_flow *flow;
723 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
728 flow = dp_netdev_lookup_flow(dp, &key);
734 get_dpif_flow_stats(flow, stats);
737 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
743 set_flow_actions(struct dp_netdev_flow *flow,
744 const struct nlattr *actions, size_t actions_len)
746 flow->actions = xrealloc(flow->actions, actions_len);
747 flow->actions_len = actions_len;
748 memcpy(flow->actions, actions, actions_len);
753 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
754 const struct nlattr *actions, size_t actions_len)
756 struct dp_netdev_flow *flow;
759 flow = xzalloc(sizeof *flow);
762 error = set_flow_actions(flow, actions, actions_len);
768 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
773 clear_stats(struct dp_netdev_flow *flow)
776 flow->packet_count = 0;
777 flow->byte_count = 0;
782 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
784 struct dp_netdev *dp = get_dp_netdev(dpif);
785 struct dp_netdev_flow *flow;
789 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
794 flow = dp_netdev_lookup_flow(dp, &key);
796 if (put->flags & DPIF_FP_CREATE) {
797 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
799 memset(put->stats, 0, sizeof *put->stats);
801 return dp_netdev_flow_add(dp, &key, put->actions,
810 if (put->flags & DPIF_FP_MODIFY) {
811 int error = set_flow_actions(flow, put->actions, put->actions_len);
814 get_dpif_flow_stats(flow, put->stats);
816 if (put->flags & DPIF_FP_ZERO_STATS) {
828 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
830 struct dp_netdev *dp = get_dp_netdev(dpif);
831 struct dp_netdev_flow *flow;
835 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
840 flow = dp_netdev_lookup_flow(dp, &key);
843 get_dpif_flow_stats(flow, del->stats);
845 dp_netdev_free_flow(dp, flow);
852 struct dp_netdev_flow_state {
855 struct nlattr *actions;
856 struct odputil_keybuf keybuf;
857 struct dpif_flow_stats stats;
861 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
863 struct dp_netdev_flow_state *state;
865 *statep = state = xmalloc(sizeof *state);
868 state->actions = NULL;
873 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
874 const struct nlattr **key, size_t *key_len,
875 const struct nlattr **actions, size_t *actions_len,
876 const struct dpif_flow_stats **stats)
878 struct dp_netdev_flow_state *state = state_;
879 struct dp_netdev *dp = get_dp_netdev(dpif);
880 struct dp_netdev_flow *flow;
881 struct hmap_node *node;
883 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
888 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
893 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
894 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
901 free(state->actions);
902 state->actions = xmemdup(flow->actions, flow->actions_len);
904 *actions = state->actions;
905 *actions_len = flow->actions_len;
909 get_dpif_flow_stats(flow, &state->stats);
910 *stats = &state->stats;
917 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
919 struct dp_netdev_flow_state *state = state_;
921 free(state->actions);
927 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
929 struct dp_netdev *dp = get_dp_netdev(dpif);
934 if (execute->packet->size < ETH_HEADER_LEN ||
935 execute->packet->size > UINT16_MAX) {
939 /* Make a deep copy of 'packet', because we might modify its data. */
940 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
941 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
942 ofpbuf_put(©, execute->packet->data, execute->packet->size);
944 flow_extract(©, 0, 0, NULL, -1, &key);
945 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
948 dp_netdev_execute_actions(dp, ©, &key,
949 execute->actions, execute->actions_len);
952 ofpbuf_uninit(©);
957 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
963 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
964 uint32_t queue_id, uint32_t *priority)
966 *priority = queue_id;
970 static struct dp_netdev_queue *
971 find_nonempty_queue(struct dpif *dpif)
973 struct dp_netdev *dp = get_dp_netdev(dpif);
976 for (i = 0; i < N_QUEUES; i++) {
977 struct dp_netdev_queue *q = &dp->queues[i];
978 if (q->head != q->tail) {
986 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
989 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
991 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
994 upcall->packet = buf;
1006 dpif_netdev_recv_wait(struct dpif *dpif)
1008 if (find_nonempty_queue(dpif)) {
1009 poll_immediate_wake();
1011 /* No messages ready to be received, and dp_wait() will ensure that we
1012 * wake up to queue new messages, so there is nothing to do. */
1017 dpif_netdev_recv_purge(struct dpif *dpif)
1019 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1020 dp_netdev_purge_queues(dpif_netdev->dp);
1024 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1026 flow->used = time_msec();
1027 flow->packet_count++;
1028 flow->byte_count += packet->size;
1029 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1033 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1034 struct ofpbuf *packet)
1036 struct dp_netdev_flow *flow;
1039 if (packet->size < ETH_HEADER_LEN) {
1042 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1043 flow = dp_netdev_lookup_flow(dp, &key);
1045 dp_netdev_flow_used(flow, packet);
1046 dp_netdev_execute_actions(dp, packet, &key,
1047 flow->actions, flow->actions_len);
1051 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1056 dpif_netdev_run(struct dpif *dpif)
1058 struct dp_netdev *dp = get_dp_netdev(dpif);
1059 struct dp_netdev_port *port;
1060 struct ofpbuf packet;
1062 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1064 LIST_FOR_EACH (port, node, &dp->port_list) {
1067 /* Reset packet contents. */
1068 ofpbuf_clear(&packet);
1069 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1071 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1073 dp_netdev_port_input(dp, port, &packet);
1074 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1075 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1076 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1077 netdev_vport_get_dpif_port(port->netdev),
1081 ofpbuf_uninit(&packet);
1085 dpif_netdev_wait(struct dpif *dpif)
1087 struct dp_netdev *dp = get_dp_netdev(dpif);
1088 struct dp_netdev_port *port;
1090 LIST_FOR_EACH (port, node, &dp->port_list) {
1092 netdev_rx_wait(port->rx);
1098 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1100 struct eth_header *eh = packet->l2;
1102 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1103 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1107 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1110 struct dp_netdev_port *p = dp->ports[out_port];
1112 netdev_send(p->netdev, packet);
1117 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1118 int queue_no, const struct flow *flow,
1119 const struct nlattr *userdata)
1121 struct dp_netdev_queue *q = &dp->queues[queue_no];
1122 if (q->head - q->tail < MAX_QUEUE_LEN) {
1123 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1124 struct dpif_upcall *upcall = &u->upcall;
1125 struct ofpbuf *buf = &u->buf;
1128 upcall->type = queue_no;
1130 /* Allocate buffer big enough for everything. */
1131 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1133 buf_size += NLA_ALIGN(userdata->nla_len);
1135 ofpbuf_init(buf, buf_size);
1138 odp_flow_key_from_flow(buf, flow, flow->in_port);
1139 upcall->key = buf->data;
1140 upcall->key_len = buf->size;
1144 upcall->userdata = ofpbuf_put(buf, userdata,
1145 NLA_ALIGN(userdata->nla_len));
1150 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1151 * is visible in 'upcall->packet'. The ODP flow and (if present)
1152 * userdata become part of the headroom. */
1153 ofpbuf_put_zeros(buf, 2);
1154 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1155 buf->size = packet->size;
1156 upcall->packet = buf;
1166 dp_netdev_sample(struct dp_netdev *dp,
1167 struct ofpbuf *packet, struct flow *key,
1168 const struct nlattr *action)
1170 const struct nlattr *subactions = NULL;
1171 const struct nlattr *a;
1174 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1175 int type = nl_attr_type(a);
1177 switch ((enum ovs_sample_attr) type) {
1178 case OVS_SAMPLE_ATTR_PROBABILITY:
1179 if (random_uint32() >= nl_attr_get_u32(a)) {
1184 case OVS_SAMPLE_ATTR_ACTIONS:
1188 case OVS_SAMPLE_ATTR_UNSPEC:
1189 case __OVS_SAMPLE_ATTR_MAX:
1195 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1196 nl_attr_get_size(subactions));
1200 dp_netdev_action_userspace(struct dp_netdev *dp,
1201 struct ofpbuf *packet, const struct flow *key,
1202 const struct nlattr *userdata)
1204 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1208 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1210 enum ovs_key_attr type = nl_attr_type(a);
1211 const struct ovs_key_ipv4 *ipv4_key;
1212 const struct ovs_key_ipv6 *ipv6_key;
1213 const struct ovs_key_tcp *tcp_key;
1214 const struct ovs_key_udp *udp_key;
1217 case OVS_KEY_ATTR_PRIORITY:
1218 case OVS_KEY_ATTR_SKB_MARK:
1219 case OVS_KEY_ATTR_TUNNEL:
1220 /* not implemented */
1223 case OVS_KEY_ATTR_ETHERNET:
1224 dp_netdev_set_dl(packet,
1225 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1228 case OVS_KEY_ATTR_IPV4:
1229 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1230 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1231 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1234 case OVS_KEY_ATTR_IPV6:
1235 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1236 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1237 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1238 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1241 case OVS_KEY_ATTR_TCP:
1242 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1243 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1246 case OVS_KEY_ATTR_UDP:
1247 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1248 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1251 case OVS_KEY_ATTR_MPLS:
1252 set_mpls_lse(packet, nl_attr_get_be32(a));
1255 case OVS_KEY_ATTR_UNSPEC:
1256 case OVS_KEY_ATTR_ENCAP:
1257 case OVS_KEY_ATTR_ETHERTYPE:
1258 case OVS_KEY_ATTR_IN_PORT:
1259 case OVS_KEY_ATTR_VLAN:
1260 case OVS_KEY_ATTR_ICMP:
1261 case OVS_KEY_ATTR_ICMPV6:
1262 case OVS_KEY_ATTR_ARP:
1263 case OVS_KEY_ATTR_ND:
1264 case __OVS_KEY_ATTR_MAX:
1271 dp_netdev_execute_actions(struct dp_netdev *dp,
1272 struct ofpbuf *packet, struct flow *key,
1273 const struct nlattr *actions,
1276 const struct nlattr *a;
1279 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1280 int type = nl_attr_type(a);
1282 switch ((enum ovs_action_attr) type) {
1283 case OVS_ACTION_ATTR_OUTPUT:
1284 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1287 case OVS_ACTION_ATTR_USERSPACE: {
1288 const struct nlattr *userdata;
1290 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1291 dp_netdev_action_userspace(dp, packet, key, userdata);
1295 case OVS_ACTION_ATTR_PUSH_VLAN: {
1296 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1297 eth_push_vlan(packet, vlan->vlan_tci);
1301 case OVS_ACTION_ATTR_POP_VLAN:
1302 eth_pop_vlan(packet);
1305 case OVS_ACTION_ATTR_PUSH_MPLS: {
1306 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
1307 push_mpls(packet, mpls->mpls_ethertype, mpls->mpls_lse);
1311 case OVS_ACTION_ATTR_POP_MPLS:
1312 pop_mpls(packet, nl_attr_get_be16(a));
1315 case OVS_ACTION_ATTR_SET:
1316 execute_set_action(packet, nl_attr_get(a));
1319 case OVS_ACTION_ATTR_SAMPLE:
1320 dp_netdev_sample(dp, packet, key, a);
1323 case OVS_ACTION_ATTR_UNSPEC:
1324 case __OVS_ACTION_ATTR_MAX:
1330 const struct dpif_class dpif_netdev_class = {
1332 dpif_netdev_enumerate,
1333 dpif_netdev_port_open_type,
1336 dpif_netdev_destroy,
1339 dpif_netdev_get_stats,
1340 dpif_netdev_port_add,
1341 dpif_netdev_port_del,
1342 dpif_netdev_port_query_by_number,
1343 dpif_netdev_port_query_by_name,
1344 dpif_netdev_get_max_ports,
1345 NULL, /* port_get_pid */
1346 dpif_netdev_port_dump_start,
1347 dpif_netdev_port_dump_next,
1348 dpif_netdev_port_dump_done,
1349 dpif_netdev_port_poll,
1350 dpif_netdev_port_poll_wait,
1351 dpif_netdev_flow_get,
1352 dpif_netdev_flow_put,
1353 dpif_netdev_flow_del,
1354 dpif_netdev_flow_flush,
1355 dpif_netdev_flow_dump_start,
1356 dpif_netdev_flow_dump_next,
1357 dpif_netdev_flow_dump_done,
1358 dpif_netdev_execute,
1360 dpif_netdev_recv_set,
1361 dpif_netdev_queue_to_priority,
1363 dpif_netdev_recv_wait,
1364 dpif_netdev_recv_purge,
1368 dpif_dummy_register__(const char *type)
1370 struct dpif_class *class;
1372 class = xmalloc(sizeof *class);
1373 *class = dpif_netdev_class;
1374 class->type = xstrdup(type);
1375 dp_register_provider(class);
1379 dpif_dummy_register(bool override)
1386 dp_enumerate_types(&types);
1387 SSET_FOR_EACH (type, &types) {
1388 if (!dp_unregister_provider(type)) {
1389 dpif_dummy_register__(type);
1392 sset_destroy(&types);
1395 dpif_dummy_register__("dummy");