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"
56 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
58 /* Configuration parameters. */
59 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
60 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
62 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
63 * headers to be aligned on a 4-byte boundary. */
64 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
67 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
68 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
69 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
70 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
72 struct dp_netdev_queue {
73 struct dpif_upcall *upcalls[MAX_QUEUE_LEN];
74 unsigned int head, tail;
77 /* Datapath based on the network device interface from netdev.h. */
79 const struct dpif_class *class;
84 struct dp_netdev_queue queues[N_QUEUES];
85 struct hmap flow_table; /* Flow table. */
88 long long int n_hit; /* Number of flow table matches. */
89 long long int n_missed; /* Number of flow table misses. */
90 long long int n_lost; /* Number of misses not passed to client. */
93 struct dp_netdev_port *ports[MAX_PORTS];
94 struct list port_list;
98 /* A port in a netdev-based datapath. */
99 struct dp_netdev_port {
100 int port_no; /* Index into dp_netdev's 'ports'. */
101 struct list node; /* Element in dp_netdev's 'port_list'. */
102 struct netdev *netdev;
103 bool internal; /* Internal port? */
106 /* A flow in dp_netdev's 'flow_table'. */
107 struct dp_netdev_flow {
108 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
112 long long int used; /* Last used time, in monotonic msecs. */
113 long long int packet_count; /* Number of packets matched. */
114 long long int byte_count; /* Number of bytes matched. */
115 ovs_be16 tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */
118 struct nlattr *actions;
122 /* Interface to netdev-based datapath. */
125 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_class dpif_dummy_class;
157 static struct dpif_netdev *
158 dpif_netdev_cast(const struct dpif *dpif)
160 assert(dpif->dpif_class->open == dpif_netdev_open);
161 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
164 static struct dp_netdev *
165 get_dp_netdev(const struct dpif *dpif)
167 return dpif_netdev_cast(dpif)->dp;
171 create_dpif_netdev(struct dp_netdev *dp)
173 uint16_t netflow_id = hash_string(dp->name, 0);
174 struct dpif_netdev *dpif;
178 dpif = xmalloc(sizeof *dpif);
179 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
181 dpif->listen_mask = 0;
182 dpif->dp_serial = dp->serial;
188 create_dp_netdev(const char *name, const struct dpif_class *class,
189 struct dp_netdev **dpp)
191 struct dp_netdev *dp;
195 dp = xzalloc(sizeof *dp);
197 dp->name = xstrdup(name);
199 for (i = 0; i < N_QUEUES; i++) {
200 dp->queues[i].head = dp->queues[i].tail = 0;
202 hmap_init(&dp->flow_table);
203 list_init(&dp->port_list);
204 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
210 shash_add(&dp_netdevs, name, dp);
217 dpif_netdev_open(const struct dpif_class *class, const char *name,
218 bool create, struct dpif **dpifp)
220 struct dp_netdev *dp;
222 dp = shash_find_data(&dp_netdevs, name);
227 int error = create_dp_netdev(name, class, &dp);
234 if (dp->class != class) {
241 *dpifp = create_dpif_netdev(dp);
246 dp_netdev_purge_queues(struct dp_netdev *dp)
250 for (i = 0; i < N_QUEUES; i++) {
251 struct dp_netdev_queue *q = &dp->queues[i];
253 while (q->tail != q->head) {
254 struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK];
256 ofpbuf_delete(upcall->packet);
263 dp_netdev_free(struct dp_netdev *dp)
265 struct dp_netdev_port *port, *next;
267 dp_netdev_flow_flush(dp);
268 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
269 do_del_port(dp, port->port_no);
271 dp_netdev_purge_queues(dp);
272 hmap_destroy(&dp->flow_table);
278 dpif_netdev_close(struct dpif *dpif)
280 struct dp_netdev *dp = get_dp_netdev(dpif);
281 assert(dp->open_cnt > 0);
282 if (--dp->open_cnt == 0 && dp->destroyed) {
283 shash_find_and_delete(&dp_netdevs, dp->name);
290 dpif_netdev_destroy(struct dpif *dpif)
292 struct dp_netdev *dp = get_dp_netdev(dpif);
293 dp->destroyed = true;
298 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
300 struct dp_netdev *dp = get_dp_netdev(dpif);
301 stats->n_flows = hmap_count(&dp->flow_table);
302 stats->n_hit = dp->n_hit;
303 stats->n_missed = dp->n_missed;
304 stats->n_lost = dp->n_lost;
309 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
312 struct dp_netdev_port *port;
313 struct netdev *netdev;
318 /* XXX reject devices already in some dp_netdev. */
319 if (type[0] == '\0' || !strcmp(type, "system") || !strcmp(type, "dummy")) {
321 } else if (!strcmp(type, "internal")) {
324 VLOG_WARN("%s: unsupported port type %s", devname, type);
328 /* Open and validate network device. */
329 if (dp->class == &dpif_dummy_class) {
331 } else if (internal) {
335 error = netdev_open(devname, 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->internal = internal;
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_dummy_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 port_no = choose_port(dpif, netdev);
416 return do_add_port(dp, netdev_get_name(netdev),
417 netdev_get_type(netdev), port_no);
423 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
425 struct dp_netdev *dp = get_dp_netdev(dpif);
426 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
430 is_valid_port_number(uint16_t port_no)
432 return port_no < MAX_PORTS;
436 get_port_by_number(struct dp_netdev *dp,
437 uint16_t port_no, struct dp_netdev_port **portp)
439 if (!is_valid_port_number(port_no)) {
443 *portp = dp->ports[port_no];
444 return *portp ? 0 : ENOENT;
449 get_port_by_name(struct dp_netdev *dp,
450 const char *devname, struct dp_netdev_port **portp)
452 struct dp_netdev_port *port;
454 LIST_FOR_EACH (port, node, &dp->port_list) {
455 if (!strcmp(netdev_get_name(port->netdev), devname)) {
464 do_del_port(struct dp_netdev *dp, uint16_t port_no)
466 struct dp_netdev_port *port;
470 error = get_port_by_number(dp, port_no, &port);
475 list_remove(&port->node);
476 dp->ports[port->port_no] = NULL;
479 name = xstrdup(netdev_get_name(port->netdev));
480 netdev_close(port->netdev);
489 answer_port_query(const struct dp_netdev_port *port,
490 struct dpif_port *dpif_port)
492 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
493 dpif_port->type = xstrdup(port->internal ? "internal"
494 : netdev_get_type(port->netdev));
495 dpif_port->port_no = port->port_no;
499 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
500 struct dpif_port *dpif_port)
502 struct dp_netdev *dp = get_dp_netdev(dpif);
503 struct dp_netdev_port *port;
506 error = get_port_by_number(dp, port_no, &port);
508 answer_port_query(port, dpif_port);
514 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
515 struct dpif_port *dpif_port)
517 struct dp_netdev *dp = get_dp_netdev(dpif);
518 struct dp_netdev_port *port;
521 error = get_port_by_name(dp, devname, &port);
523 answer_port_query(port, dpif_port);
529 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
535 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
537 hmap_remove(&dp->flow_table, &flow->node);
543 dp_netdev_flow_flush(struct dp_netdev *dp)
545 struct dp_netdev_flow *flow, *next;
547 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
548 dp_netdev_free_flow(dp, flow);
553 dpif_netdev_flow_flush(struct dpif *dpif)
555 struct dp_netdev *dp = get_dp_netdev(dpif);
556 dp_netdev_flow_flush(dp);
560 struct dp_netdev_port_state {
566 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
568 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
573 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
574 struct dpif_port *dpif_port)
576 struct dp_netdev_port_state *state = state_;
577 struct dp_netdev *dp = get_dp_netdev(dpif);
580 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
581 struct dp_netdev_port *port = dp->ports[port_no];
584 state->name = xstrdup(netdev_get_name(port->netdev));
585 dpif_port->name = state->name;
586 dpif_port->type = (char *) (port->internal ? "internal"
587 : netdev_get_type(port->netdev));
588 dpif_port->port_no = port->port_no;
589 state->port_no = port_no + 1;
597 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
599 struct dp_netdev_port_state *state = state_;
606 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
608 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
609 if (dpif->dp_serial != dpif->dp->serial) {
610 dpif->dp_serial = dpif->dp->serial;
618 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
620 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
621 if (dpif->dp_serial != dpif->dp->serial) {
622 poll_immediate_wake();
626 static struct dp_netdev_flow *
627 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
629 struct dp_netdev_flow *flow;
631 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
632 if (flow_equal(&flow->key, key)) {
640 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
642 stats->n_packets = flow->packet_count;
643 stats->n_bytes = flow->byte_count;
644 stats->used = flow->used;
645 stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl);
649 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
652 if (odp_flow_key_to_flow(key, key_len, flow)) {
653 /* This should not happen: it indicates that odp_flow_key_from_flow()
654 * and odp_flow_key_to_flow() disagree on the acceptable form of a
655 * flow. Log the problem as an error, with enough details to enable
657 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
659 if (!VLOG_DROP_ERR(&rl)) {
663 odp_flow_key_format(key, key_len, &s);
664 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
671 if (flow->in_port < OFPP_MAX
672 ? flow->in_port >= MAX_PORTS
673 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
681 dpif_netdev_flow_get(const struct dpif *dpif,
682 const struct nlattr *nl_key, size_t nl_key_len,
683 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
685 struct dp_netdev *dp = get_dp_netdev(dpif);
686 struct dp_netdev_flow *flow;
690 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
695 flow = dp_netdev_lookup_flow(dp, &key);
701 get_dpif_flow_stats(flow, stats);
704 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
710 set_flow_actions(struct dp_netdev_flow *flow,
711 const struct nlattr *actions, size_t actions_len)
713 flow->actions = xrealloc(flow->actions, actions_len);
714 flow->actions_len = actions_len;
715 memcpy(flow->actions, actions, actions_len);
720 add_flow(struct dpif *dpif, const struct flow *key,
721 const struct nlattr *actions, size_t actions_len)
723 struct dp_netdev *dp = get_dp_netdev(dpif);
724 struct dp_netdev_flow *flow;
727 flow = xzalloc(sizeof *flow);
730 error = set_flow_actions(flow, actions, actions_len);
736 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
741 clear_stats(struct dp_netdev_flow *flow)
744 flow->packet_count = 0;
745 flow->byte_count = 0;
750 dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
751 const struct nlattr *nl_key, size_t nl_key_len,
752 const struct nlattr *actions, size_t actions_len,
753 struct dpif_flow_stats *stats)
755 struct dp_netdev *dp = get_dp_netdev(dpif);
756 struct dp_netdev_flow *flow;
760 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
765 flow = dp_netdev_lookup_flow(dp, &key);
767 if (flags & DPIF_FP_CREATE) {
768 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
770 memset(stats, 0, sizeof *stats);
772 return add_flow(dpif, &key, actions, actions_len);
780 if (flags & DPIF_FP_MODIFY) {
781 int error = set_flow_actions(flow, actions, actions_len);
784 get_dpif_flow_stats(flow, stats);
786 if (flags & DPIF_FP_ZERO_STATS) {
798 dpif_netdev_flow_del(struct dpif *dpif,
799 const struct nlattr *nl_key, size_t nl_key_len,
800 struct dpif_flow_stats *stats)
802 struct dp_netdev *dp = get_dp_netdev(dpif);
803 struct dp_netdev_flow *flow;
807 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
812 flow = dp_netdev_lookup_flow(dp, &key);
815 get_dpif_flow_stats(flow, stats);
817 dp_netdev_free_flow(dp, flow);
824 struct dp_netdev_flow_state {
827 struct nlattr *actions;
828 struct odputil_keybuf keybuf;
829 struct dpif_flow_stats stats;
833 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
835 struct dp_netdev_flow_state *state;
837 *statep = state = xmalloc(sizeof *state);
840 state->actions = NULL;
845 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
846 const struct nlattr **key, size_t *key_len,
847 const struct nlattr **actions, size_t *actions_len,
848 const struct dpif_flow_stats **stats)
850 struct dp_netdev_flow_state *state = state_;
851 struct dp_netdev *dp = get_dp_netdev(dpif);
852 struct dp_netdev_flow *flow;
853 struct hmap_node *node;
855 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
860 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
865 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
866 odp_flow_key_from_flow(&buf, &flow->key);
873 free(state->actions);
874 state->actions = xmemdup(flow->actions, flow->actions_len);
876 *actions = state->actions;
877 *actions_len = flow->actions_len;
881 get_dpif_flow_stats(flow, &state->stats);
882 *stats = &state->stats;
889 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
891 struct dp_netdev_flow_state *state = state_;
893 free(state->actions);
899 dpif_netdev_execute(struct dpif *dpif,
900 const struct nlattr *key_attrs, size_t key_len,
901 const struct nlattr *actions, size_t actions_len,
902 const struct ofpbuf *packet)
904 struct dp_netdev *dp = get_dp_netdev(dpif);
909 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
913 /* Make a deep copy of 'packet', because we might modify its data. */
914 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
915 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
916 ofpbuf_put(©, packet->data, packet->size);
918 flow_extract(©, 0, 0, -1, &key);
919 error = dpif_netdev_flow_from_nlattrs(key_attrs, key_len, &key);
921 dp_netdev_execute_actions(dp, ©, &key,
922 actions, actions_len);
925 ofpbuf_uninit(©);
930 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
932 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
933 *listen_mask = dpif_netdev->listen_mask;
938 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
940 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
941 dpif_netdev->listen_mask = listen_mask;
946 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
947 uint32_t queue_id, uint32_t *priority)
949 *priority = queue_id;
953 static struct dp_netdev_queue *
954 find_nonempty_queue(struct dpif *dpif)
956 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
957 struct dp_netdev *dp = get_dp_netdev(dpif);
958 int mask = dpif_netdev->listen_mask;
961 for (i = 0; i < N_QUEUES; i++) {
962 struct dp_netdev_queue *q = &dp->queues[i];
963 if (q->head != q->tail && mask & (1u << i)) {
971 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
973 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
975 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
986 dpif_netdev_recv_wait(struct dpif *dpif)
988 if (find_nonempty_queue(dpif)) {
989 poll_immediate_wake();
991 /* No messages ready to be received, and dp_wait() will ensure that we
992 * wake up to queue new messages, so there is nothing to do. */
997 dpif_netdev_recv_purge(struct dpif *dpif)
999 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1000 dp_netdev_purge_queues(dpif_netdev->dp);
1004 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
1005 const struct ofpbuf *packet)
1007 flow->used = time_msec();
1008 flow->packet_count++;
1009 flow->byte_count += packet->size;
1010 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1011 struct tcp_header *th = packet->l4;
1012 flow->tcp_ctl |= th->tcp_ctl;
1017 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1018 struct ofpbuf *packet)
1020 struct dp_netdev_flow *flow;
1023 if (packet->size < ETH_HEADER_LEN) {
1026 flow_extract(packet, 0, 0, port->port_no, &key);
1027 flow = dp_netdev_lookup_flow(dp, &key);
1029 dp_netdev_flow_used(flow, &key, packet);
1030 dp_netdev_execute_actions(dp, packet, &key,
1031 flow->actions, flow->actions_len);
1035 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1040 dpif_netdev_run(struct dpif *dpif)
1042 struct dp_netdev *dp = get_dp_netdev(dpif);
1043 struct dp_netdev_port *port;
1044 struct ofpbuf packet;
1046 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1048 LIST_FOR_EACH (port, node, &dp->port_list) {
1051 /* Reset packet contents. */
1052 ofpbuf_clear(&packet);
1053 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1055 error = netdev_recv(port->netdev, &packet);
1057 dp_netdev_port_input(dp, port, &packet);
1058 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1060 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1061 netdev_get_name(port->netdev), strerror(error));
1064 ofpbuf_uninit(&packet);
1068 dpif_netdev_wait(struct dpif *dpif)
1070 struct dp_netdev *dp = get_dp_netdev(dpif);
1071 struct dp_netdev_port *port;
1073 LIST_FOR_EACH (port, node, &dp->port_list) {
1074 netdev_recv_wait(port->netdev);
1079 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1081 struct eth_header *eh = packet->l2;
1083 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1084 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1088 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1091 struct dp_netdev_port *p = dp->ports[out_port];
1093 netdev_send(p->netdev, packet);
1098 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1099 int queue_no, const struct flow *flow, uint64_t arg)
1101 struct dp_netdev_queue *q = &dp->queues[queue_no];
1102 struct dpif_upcall *upcall;
1106 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1111 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1112 odp_flow_key_from_flow(buf, flow);
1113 key_len = buf->size;
1114 ofpbuf_pull(buf, key_len);
1115 ofpbuf_reserve(buf, 2);
1116 ofpbuf_put(buf, packet->data, packet->size);
1118 upcall = xzalloc(sizeof *upcall);
1119 upcall->type = queue_no;
1120 upcall->packet = buf;
1121 upcall->key = buf->base;
1122 upcall->key_len = key_len;
1123 upcall->userdata = arg;
1125 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
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_tcp *tcp_key;
1183 const struct ovs_key_udp *udp_key;
1186 case OVS_KEY_ATTR_TUN_ID:
1187 case OVS_KEY_ATTR_PRIORITY:
1188 case OVS_KEY_ATTR_IPV6:
1189 /* not implemented */
1192 case OVS_KEY_ATTR_ETHERNET:
1193 dp_netdev_set_dl(packet,
1194 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1197 case OVS_KEY_ATTR_IPV4:
1198 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1199 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1200 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1203 case OVS_KEY_ATTR_TCP:
1204 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1205 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1208 case OVS_KEY_ATTR_UDP:
1209 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1210 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1213 case OVS_KEY_ATTR_UNSPEC:
1214 case OVS_KEY_ATTR_ENCAP:
1215 case OVS_KEY_ATTR_ETHERTYPE:
1216 case OVS_KEY_ATTR_IN_PORT:
1217 case OVS_KEY_ATTR_VLAN:
1218 case OVS_KEY_ATTR_ICMP:
1219 case OVS_KEY_ATTR_ICMPV6:
1220 case OVS_KEY_ATTR_ARP:
1221 case OVS_KEY_ATTR_ND:
1222 case __OVS_KEY_ATTR_MAX:
1229 dp_netdev_execute_actions(struct dp_netdev *dp,
1230 struct ofpbuf *packet, struct flow *key,
1231 const struct nlattr *actions,
1234 const struct nlattr *a;
1237 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1238 const struct ovs_action_push_vlan *vlan;
1239 int type = nl_attr_type(a);
1241 switch ((enum ovs_action_attr) type) {
1242 case OVS_ACTION_ATTR_OUTPUT:
1243 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1246 case OVS_ACTION_ATTR_USERSPACE:
1247 dp_netdev_action_userspace(dp, packet, key, a);
1250 case OVS_ACTION_ATTR_PUSH_VLAN:
1251 vlan = nl_attr_get(a);
1252 eth_push_vlan(packet, vlan->vlan_tci);
1255 case OVS_ACTION_ATTR_POP_VLAN:
1256 eth_pop_vlan(packet);
1259 case OVS_ACTION_ATTR_SET:
1260 execute_set_action(packet, nl_attr_get(a));
1263 case OVS_ACTION_ATTR_SAMPLE:
1264 dp_netdev_sample(dp, packet, key, a);
1267 case OVS_ACTION_ATTR_UNSPEC:
1268 case __OVS_ACTION_ATTR_MAX:
1274 const struct dpif_class dpif_netdev_class = {
1276 NULL, /* enumerate */
1279 dpif_netdev_destroy,
1282 dpif_netdev_get_stats,
1283 dpif_netdev_port_add,
1284 dpif_netdev_port_del,
1285 dpif_netdev_port_query_by_number,
1286 dpif_netdev_port_query_by_name,
1287 dpif_netdev_get_max_ports,
1288 NULL, /* port_get_pid */
1289 dpif_netdev_port_dump_start,
1290 dpif_netdev_port_dump_next,
1291 dpif_netdev_port_dump_done,
1292 dpif_netdev_port_poll,
1293 dpif_netdev_port_poll_wait,
1294 dpif_netdev_flow_get,
1295 dpif_netdev_flow_put,
1296 dpif_netdev_flow_del,
1297 dpif_netdev_flow_flush,
1298 dpif_netdev_flow_dump_start,
1299 dpif_netdev_flow_dump_next,
1300 dpif_netdev_flow_dump_done,
1301 dpif_netdev_execute,
1303 dpif_netdev_recv_get_mask,
1304 dpif_netdev_recv_set_mask,
1305 dpif_netdev_queue_to_priority,
1307 dpif_netdev_recv_wait,
1308 dpif_netdev_recv_purge,
1312 dpif_dummy_register(void)
1314 if (!dpif_dummy_class.type) {
1315 dpif_dummy_class = dpif_netdev_class;
1316 dpif_dummy_class.type = "dummy";
1317 dp_register_provider(&dpif_dummy_class);