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 *, uint16_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, uint16_t port_no);
149 static int do_del_port(struct dp_netdev *, uint16_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 create_dp_netdev(const char *name, const struct dpif_class *class,
202 struct dp_netdev **dpp)
204 struct dp_netdev *dp;
208 dp = xzalloc(sizeof *dp);
210 dp->name = xstrdup(name);
212 for (i = 0; i < N_QUEUES; i++) {
213 dp->queues[i].head = dp->queues[i].tail = 0;
215 hmap_init(&dp->flow_table);
216 list_init(&dp->port_list);
217 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
223 shash_add(&dp_netdevs, name, dp);
230 dpif_netdev_open(const struct dpif_class *class, const char *name,
231 bool create, struct dpif **dpifp)
233 struct dp_netdev *dp;
235 dp = shash_find_data(&dp_netdevs, name);
240 int error = create_dp_netdev(name, class, &dp);
247 if (dp->class != class) {
254 *dpifp = create_dpif_netdev(dp);
259 dp_netdev_purge_queues(struct dp_netdev *dp)
263 for (i = 0; i < N_QUEUES; i++) {
264 struct dp_netdev_queue *q = &dp->queues[i];
266 while (q->tail != q->head) {
267 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
268 ofpbuf_uninit(&u->buf);
274 dp_netdev_free(struct dp_netdev *dp)
276 struct dp_netdev_port *port, *next;
278 dp_netdev_flow_flush(dp);
279 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
280 do_del_port(dp, port->port_no);
282 dp_netdev_purge_queues(dp);
283 hmap_destroy(&dp->flow_table);
289 dpif_netdev_close(struct dpif *dpif)
291 struct dp_netdev *dp = get_dp_netdev(dpif);
292 assert(dp->open_cnt > 0);
293 if (--dp->open_cnt == 0 && dp->destroyed) {
294 shash_find_and_delete(&dp_netdevs, dp->name);
301 dpif_netdev_destroy(struct dpif *dpif)
303 struct dp_netdev *dp = get_dp_netdev(dpif);
304 dp->destroyed = true;
309 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
311 struct dp_netdev *dp = get_dp_netdev(dpif);
312 stats->n_flows = hmap_count(&dp->flow_table);
313 stats->n_hit = dp->n_hit;
314 stats->n_missed = dp->n_missed;
315 stats->n_lost = dp->n_lost;
320 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
323 struct dp_netdev_port *port;
324 struct netdev *netdev;
325 const char *open_type;
329 /* XXX reject devices already in some dp_netdev. */
331 /* Open and validate network device. */
332 open_type = (strcmp(type, "internal") ? type
333 : dp->class != &dpif_netdev_class ? "dummy"
335 error = netdev_open(devname, open_type, &netdev);
339 /* XXX reject loopback devices */
340 /* XXX reject non-Ethernet devices */
342 error = netdev_listen(netdev);
344 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
345 devname, strerror(errno));
346 netdev_close(netdev);
350 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
352 netdev_close(netdev);
356 port = xmalloc(sizeof *port);
357 port->port_no = port_no;
358 port->netdev = netdev;
359 port->type = xstrdup(type);
361 error = netdev_get_mtu(netdev, &mtu);
366 list_push_back(&dp->port_list, &port->node);
367 dp->ports[port_no] = port;
374 choose_port(struct dpif *dpif, struct netdev *netdev)
376 struct dp_netdev *dp = get_dp_netdev(dpif);
379 if (dpif->dpif_class != &dpif_netdev_class) {
380 /* If the port name contains a number, try to assign that port number.
381 * This can make writing unit tests easier because port numbers are
385 for (p = netdev_get_name(netdev); *p != '\0'; p++) {
386 if (isdigit((unsigned char) *p)) {
387 port_no = strtol(p, NULL, 10);
388 if (port_no > 0 && port_no < MAX_PORTS
389 && !dp->ports[port_no]) {
397 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
398 if (!dp->ports[port_no]) {
407 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
410 struct dp_netdev *dp = get_dp_netdev(dpif);
413 if (*port_nop != UINT16_MAX) {
414 if (*port_nop >= MAX_PORTS) {
416 } else if (dp->ports[*port_nop]) {
421 port_no = choose_port(dpif, netdev);
425 return do_add_port(dp, netdev_get_name(netdev),
426 netdev_get_type(netdev), port_no);
432 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
434 struct dp_netdev *dp = get_dp_netdev(dpif);
435 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
439 is_valid_port_number(uint16_t port_no)
441 return port_no < MAX_PORTS;
445 get_port_by_number(struct dp_netdev *dp,
446 uint16_t port_no, struct dp_netdev_port **portp)
448 if (!is_valid_port_number(port_no)) {
452 *portp = dp->ports[port_no];
453 return *portp ? 0 : ENOENT;
458 get_port_by_name(struct dp_netdev *dp,
459 const char *devname, struct dp_netdev_port **portp)
461 struct dp_netdev_port *port;
463 LIST_FOR_EACH (port, node, &dp->port_list) {
464 if (!strcmp(netdev_get_name(port->netdev), devname)) {
473 do_del_port(struct dp_netdev *dp, uint16_t port_no)
475 struct dp_netdev_port *port;
479 error = get_port_by_number(dp, port_no, &port);
484 list_remove(&port->node);
485 dp->ports[port->port_no] = NULL;
488 name = xstrdup(netdev_get_name(port->netdev));
489 netdev_close(port->netdev);
499 answer_port_query(const struct dp_netdev_port *port,
500 struct dpif_port *dpif_port)
502 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
503 dpif_port->type = xstrdup(port->type);
504 dpif_port->port_no = port->port_no;
508 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
509 struct dpif_port *dpif_port)
511 struct dp_netdev *dp = get_dp_netdev(dpif);
512 struct dp_netdev_port *port;
515 error = get_port_by_number(dp, port_no, &port);
517 answer_port_query(port, dpif_port);
523 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
524 struct dpif_port *dpif_port)
526 struct dp_netdev *dp = get_dp_netdev(dpif);
527 struct dp_netdev_port *port;
530 error = get_port_by_name(dp, devname, &port);
532 answer_port_query(port, dpif_port);
538 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
544 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
546 hmap_remove(&dp->flow_table, &flow->node);
552 dp_netdev_flow_flush(struct dp_netdev *dp)
554 struct dp_netdev_flow *flow, *next;
556 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
557 dp_netdev_free_flow(dp, flow);
562 dpif_netdev_flow_flush(struct dpif *dpif)
564 struct dp_netdev *dp = get_dp_netdev(dpif);
565 dp_netdev_flow_flush(dp);
569 struct dp_netdev_port_state {
575 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
577 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
582 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
583 struct dpif_port *dpif_port)
585 struct dp_netdev_port_state *state = state_;
586 struct dp_netdev *dp = get_dp_netdev(dpif);
589 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
590 struct dp_netdev_port *port = dp->ports[port_no];
593 state->name = xstrdup(netdev_get_name(port->netdev));
594 dpif_port->name = state->name;
595 dpif_port->type = port->type;
596 dpif_port->port_no = port->port_no;
597 state->port_no = port_no + 1;
605 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
607 struct dp_netdev_port_state *state = state_;
614 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
616 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
617 if (dpif->dp_serial != dpif->dp->serial) {
618 dpif->dp_serial = dpif->dp->serial;
626 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
628 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
629 if (dpif->dp_serial != dpif->dp->serial) {
630 poll_immediate_wake();
634 static struct dp_netdev_flow *
635 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
637 struct dp_netdev_flow *flow;
639 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
640 if (flow_equal(&flow->key, key)) {
648 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
650 stats->n_packets = flow->packet_count;
651 stats->n_bytes = flow->byte_count;
652 stats->used = flow->used;
653 stats->tcp_flags = flow->tcp_flags;
657 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
660 if (odp_flow_key_to_flow(key, key_len, flow)) {
661 /* This should not happen: it indicates that odp_flow_key_from_flow()
662 * and odp_flow_key_to_flow() disagree on the acceptable form of a
663 * flow. Log the problem as an error, with enough details to enable
665 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
667 if (!VLOG_DROP_ERR(&rl)) {
671 odp_flow_key_format(key, key_len, &s);
672 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
679 if (flow->in_port < OFPP_MAX
680 ? flow->in_port >= MAX_PORTS
681 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
689 dpif_netdev_flow_get(const struct dpif *dpif,
690 const struct nlattr *nl_key, size_t nl_key_len,
691 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
693 struct dp_netdev *dp = get_dp_netdev(dpif);
694 struct dp_netdev_flow *flow;
698 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
703 flow = dp_netdev_lookup_flow(dp, &key);
709 get_dpif_flow_stats(flow, stats);
712 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
718 set_flow_actions(struct dp_netdev_flow *flow,
719 const struct nlattr *actions, size_t actions_len)
721 flow->actions = xrealloc(flow->actions, actions_len);
722 flow->actions_len = actions_len;
723 memcpy(flow->actions, actions, actions_len);
728 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
729 const struct nlattr *actions, size_t actions_len)
731 struct dp_netdev_flow *flow;
734 flow = xzalloc(sizeof *flow);
737 error = set_flow_actions(flow, actions, actions_len);
743 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
748 clear_stats(struct dp_netdev_flow *flow)
751 flow->packet_count = 0;
752 flow->byte_count = 0;
757 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
759 struct dp_netdev *dp = get_dp_netdev(dpif);
760 struct dp_netdev_flow *flow;
764 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
769 flow = dp_netdev_lookup_flow(dp, &key);
771 if (put->flags & DPIF_FP_CREATE) {
772 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
774 memset(put->stats, 0, sizeof *put->stats);
776 return dp_netdev_flow_add(dp, &key, put->actions,
785 if (put->flags & DPIF_FP_MODIFY) {
786 int error = set_flow_actions(flow, put->actions, put->actions_len);
789 get_dpif_flow_stats(flow, put->stats);
791 if (put->flags & DPIF_FP_ZERO_STATS) {
803 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
805 struct dp_netdev *dp = get_dp_netdev(dpif);
806 struct dp_netdev_flow *flow;
810 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
815 flow = dp_netdev_lookup_flow(dp, &key);
818 get_dpif_flow_stats(flow, del->stats);
820 dp_netdev_free_flow(dp, flow);
827 struct dp_netdev_flow_state {
830 struct nlattr *actions;
831 struct odputil_keybuf keybuf;
832 struct dpif_flow_stats stats;
836 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
838 struct dp_netdev_flow_state *state;
840 *statep = state = xmalloc(sizeof *state);
843 state->actions = NULL;
848 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
849 const struct nlattr **key, size_t *key_len,
850 const struct nlattr **actions, size_t *actions_len,
851 const struct dpif_flow_stats **stats)
853 struct dp_netdev_flow_state *state = state_;
854 struct dp_netdev *dp = get_dp_netdev(dpif);
855 struct dp_netdev_flow *flow;
856 struct hmap_node *node;
858 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
863 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
868 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
869 odp_flow_key_from_flow(&buf, &flow->key);
876 free(state->actions);
877 state->actions = xmemdup(flow->actions, flow->actions_len);
879 *actions = state->actions;
880 *actions_len = flow->actions_len;
884 get_dpif_flow_stats(flow, &state->stats);
885 *stats = &state->stats;
892 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
894 struct dp_netdev_flow_state *state = state_;
896 free(state->actions);
902 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
904 struct dp_netdev *dp = get_dp_netdev(dpif);
909 if (execute->packet->size < ETH_HEADER_LEN ||
910 execute->packet->size > UINT16_MAX) {
914 /* Make a deep copy of 'packet', because we might modify its data. */
915 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
916 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
917 ofpbuf_put(©, execute->packet->data, execute->packet->size);
919 flow_extract(©, 0, NULL, -1, &key);
920 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
923 dp_netdev_execute_actions(dp, ©, &key,
924 execute->actions, execute->actions_len);
927 ofpbuf_uninit(©);
932 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
938 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
939 uint32_t queue_id, uint32_t *priority)
941 *priority = queue_id;
945 static struct dp_netdev_queue *
946 find_nonempty_queue(struct dpif *dpif)
948 struct dp_netdev *dp = get_dp_netdev(dpif);
951 for (i = 0; i < N_QUEUES; i++) {
952 struct dp_netdev_queue *q = &dp->queues[i];
953 if (q->head != q->tail) {
961 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
964 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
966 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
969 upcall->packet = buf;
981 dpif_netdev_recv_wait(struct dpif *dpif)
983 if (find_nonempty_queue(dpif)) {
984 poll_immediate_wake();
986 /* No messages ready to be received, and dp_wait() will ensure that we
987 * wake up to queue new messages, so there is nothing to do. */
992 dpif_netdev_recv_purge(struct dpif *dpif)
994 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
995 dp_netdev_purge_queues(dpif_netdev->dp);
999 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1001 flow->used = time_msec();
1002 flow->packet_count++;
1003 flow->byte_count += packet->size;
1004 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1008 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1009 struct ofpbuf *packet)
1011 struct dp_netdev_flow *flow;
1014 if (packet->size < ETH_HEADER_LEN) {
1017 flow_extract(packet, 0, NULL, odp_port_to_ofp_port(port->port_no), &key);
1018 flow = dp_netdev_lookup_flow(dp, &key);
1020 dp_netdev_flow_used(flow, packet);
1021 dp_netdev_execute_actions(dp, packet, &key,
1022 flow->actions, flow->actions_len);
1026 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1031 dpif_netdev_run(struct dpif *dpif)
1033 struct dp_netdev *dp = get_dp_netdev(dpif);
1034 struct dp_netdev_port *port;
1035 struct ofpbuf packet;
1037 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1039 LIST_FOR_EACH (port, node, &dp->port_list) {
1042 /* Reset packet contents. */
1043 ofpbuf_clear(&packet);
1044 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1046 error = netdev_recv(port->netdev, &packet);
1048 dp_netdev_port_input(dp, port, &packet);
1049 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1050 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1051 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1052 netdev_get_name(port->netdev), strerror(error));
1055 ofpbuf_uninit(&packet);
1059 dpif_netdev_wait(struct dpif *dpif)
1061 struct dp_netdev *dp = get_dp_netdev(dpif);
1062 struct dp_netdev_port *port;
1064 LIST_FOR_EACH (port, node, &dp->port_list) {
1065 netdev_recv_wait(port->netdev);
1070 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1072 struct eth_header *eh = packet->l2;
1074 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1075 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1079 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1082 struct dp_netdev_port *p = dp->ports[out_port];
1084 netdev_send(p->netdev, packet);
1089 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1090 int queue_no, const struct flow *flow, uint64_t arg)
1092 struct dp_netdev_queue *q = &dp->queues[queue_no];
1093 struct dp_netdev_upcall *u;
1094 struct dpif_upcall *upcall;
1098 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1103 u = &q->upcalls[q->head++ & QUEUE_MASK];
1106 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1107 odp_flow_key_from_flow(buf, flow);
1108 key_len = buf->size;
1109 ofpbuf_pull(buf, key_len);
1110 ofpbuf_reserve(buf, 2);
1111 ofpbuf_put(buf, packet->data, packet->size);
1113 upcall = &u->upcall;
1114 upcall->type = queue_no;
1115 upcall->packet = buf;
1116 upcall->key = buf->base;
1117 upcall->key_len = key_len;
1118 upcall->userdata = arg;
1124 dp_netdev_sample(struct dp_netdev *dp,
1125 struct ofpbuf *packet, struct flow *key,
1126 const struct nlattr *action)
1128 const struct nlattr *subactions = NULL;
1129 const struct nlattr *a;
1132 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1133 int type = nl_attr_type(a);
1135 switch ((enum ovs_sample_attr) type) {
1136 case OVS_SAMPLE_ATTR_PROBABILITY:
1137 if (random_uint32() >= nl_attr_get_u32(a)) {
1142 case OVS_SAMPLE_ATTR_ACTIONS:
1146 case OVS_SAMPLE_ATTR_UNSPEC:
1147 case __OVS_SAMPLE_ATTR_MAX:
1153 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1154 nl_attr_get_size(subactions));
1158 dp_netdev_action_userspace(struct dp_netdev *dp,
1159 struct ofpbuf *packet, struct flow *key,
1160 const struct nlattr *a)
1162 const struct nlattr *userdata_attr;
1165 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1166 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1167 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1171 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1173 enum ovs_key_attr type = nl_attr_type(a);
1174 const struct ovs_key_ipv4 *ipv4_key;
1175 const struct ovs_key_tcp *tcp_key;
1176 const struct ovs_key_udp *udp_key;
1179 case OVS_KEY_ATTR_TUN_ID:
1180 case OVS_KEY_ATTR_PRIORITY:
1181 case OVS_KEY_ATTR_IPV6:
1182 case OVS_KEY_ATTR_IPV4_TUNNEL:
1183 /* not implemented */
1186 case OVS_KEY_ATTR_ETHERNET:
1187 dp_netdev_set_dl(packet,
1188 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1191 case OVS_KEY_ATTR_IPV4:
1192 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1193 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1194 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1197 case OVS_KEY_ATTR_TCP:
1198 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1199 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1202 case OVS_KEY_ATTR_UDP:
1203 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1204 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1207 case OVS_KEY_ATTR_UNSPEC:
1208 case OVS_KEY_ATTR_ENCAP:
1209 case OVS_KEY_ATTR_ETHERTYPE:
1210 case OVS_KEY_ATTR_IN_PORT:
1211 case OVS_KEY_ATTR_VLAN:
1212 case OVS_KEY_ATTR_ICMP:
1213 case OVS_KEY_ATTR_ICMPV6:
1214 case OVS_KEY_ATTR_ARP:
1215 case OVS_KEY_ATTR_ND:
1216 case __OVS_KEY_ATTR_MAX:
1223 dp_netdev_execute_actions(struct dp_netdev *dp,
1224 struct ofpbuf *packet, struct flow *key,
1225 const struct nlattr *actions,
1228 const struct nlattr *a;
1231 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1232 const struct ovs_action_push_vlan *vlan;
1233 int type = nl_attr_type(a);
1235 switch ((enum ovs_action_attr) type) {
1236 case OVS_ACTION_ATTR_OUTPUT:
1237 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1240 case OVS_ACTION_ATTR_USERSPACE:
1241 dp_netdev_action_userspace(dp, packet, key, a);
1244 case OVS_ACTION_ATTR_PUSH_VLAN:
1245 vlan = nl_attr_get(a);
1246 eth_push_vlan(packet, vlan->vlan_tci);
1249 case OVS_ACTION_ATTR_POP_VLAN:
1250 eth_pop_vlan(packet);
1253 case OVS_ACTION_ATTR_SET:
1254 execute_set_action(packet, nl_attr_get(a));
1257 case OVS_ACTION_ATTR_SAMPLE:
1258 dp_netdev_sample(dp, packet, key, a);
1261 case OVS_ACTION_ATTR_UNSPEC:
1262 case __OVS_ACTION_ATTR_MAX:
1268 const struct dpif_class dpif_netdev_class = {
1270 dpif_netdev_enumerate,
1273 dpif_netdev_destroy,
1276 dpif_netdev_get_stats,
1277 dpif_netdev_port_add,
1278 dpif_netdev_port_del,
1279 dpif_netdev_port_query_by_number,
1280 dpif_netdev_port_query_by_name,
1281 dpif_netdev_get_max_ports,
1282 NULL, /* port_get_pid */
1283 dpif_netdev_port_dump_start,
1284 dpif_netdev_port_dump_next,
1285 dpif_netdev_port_dump_done,
1286 dpif_netdev_port_poll,
1287 dpif_netdev_port_poll_wait,
1288 dpif_netdev_flow_get,
1289 dpif_netdev_flow_put,
1290 dpif_netdev_flow_del,
1291 dpif_netdev_flow_flush,
1292 dpif_netdev_flow_dump_start,
1293 dpif_netdev_flow_dump_next,
1294 dpif_netdev_flow_dump_done,
1295 dpif_netdev_execute,
1297 dpif_netdev_recv_set,
1298 dpif_netdev_queue_to_priority,
1300 dpif_netdev_recv_wait,
1301 dpif_netdev_recv_purge,
1305 dpif_dummy_register__(const char *type)
1307 struct dpif_class *class;
1309 class = xmalloc(sizeof *class);
1310 *class = dpif_netdev_class;
1311 class->type = xstrdup(type);
1312 dp_register_provider(class);
1316 dpif_dummy_register(bool override)
1323 dp_enumerate_types(&types);
1324 SSET_FOR_EACH (type, &types) {
1325 if (!dp_unregister_provider(type)) {
1326 dpif_dummy_register__(type);
1329 sset_destroy(&types);
1332 dpif_dummy_register__("dummy");