2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
25 #include <netinet/in.h>
26 #include <sys/socket.h>
31 #include <sys/ioctl.h>
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
46 #include "ofp-print.h"
49 #include "poll-loop.h"
57 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
59 /* Configuration parameters. */
60 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
61 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
63 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
64 * headers to be aligned on a 4-byte boundary. */
65 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
68 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
69 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
70 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
71 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
73 struct dp_netdev_upcall {
74 struct dpif_upcall upcall; /* Queued upcall information. */
75 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
78 struct dp_netdev_queue {
79 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
80 unsigned int head, tail;
83 /* Datapath based on the network device interface from netdev.h. */
85 const struct dpif_class *class;
90 struct dp_netdev_queue queues[N_QUEUES];
91 struct hmap flow_table; /* Flow table. */
94 long long int n_hit; /* Number of flow table matches. */
95 long long int n_missed; /* Number of flow table misses. */
96 long long int n_lost; /* Number of misses not passed to client. */
99 struct dp_netdev_port *ports[MAX_PORTS];
100 struct list port_list;
104 /* A port in a netdev-based datapath. */
105 struct dp_netdev_port {
106 int port_no; /* Index into dp_netdev's 'ports'. */
107 struct list node; /* Element in dp_netdev's 'port_list'. */
108 struct netdev *netdev;
109 char *type; /* Port type as requested by user. */
112 /* A flow in dp_netdev's 'flow_table'. */
113 struct dp_netdev_flow {
114 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
118 long long int used; /* Last used time, in monotonic msecs. */
119 long long int packet_count; /* Number of packets matched. */
120 long long int byte_count; /* Number of bytes matched. */
121 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
124 struct nlattr *actions;
128 /* Interface to netdev-based datapath. */
131 struct dp_netdev *dp;
132 unsigned int dp_serial;
135 /* All netdev-based datapaths. */
136 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
138 /* Maximum port MTU seen so far. */
139 static int max_mtu = ETH_PAYLOAD_MAX;
141 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
142 struct dp_netdev_port **portp);
143 static int get_port_by_name(struct dp_netdev *, const char *devname,
144 struct dp_netdev_port **portp);
145 static void dp_netdev_free(struct dp_netdev *);
146 static void dp_netdev_flow_flush(struct dp_netdev *);
147 static int do_add_port(struct dp_netdev *, const char *devname,
148 const char *type, uint32_t port_no);
149 static int do_del_port(struct dp_netdev *, uint32_t port_no);
150 static int dpif_netdev_open(const struct dpif_class *, const char *name,
151 bool create, struct dpif **);
152 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
153 int queue_no, const struct flow *,
155 static void dp_netdev_execute_actions(struct dp_netdev *,
156 struct ofpbuf *, struct flow *,
157 const struct nlattr *actions,
160 static struct dpif_netdev *
161 dpif_netdev_cast(const struct dpif *dpif)
163 assert(dpif->dpif_class->open == dpif_netdev_open);
164 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
167 static struct dp_netdev *
168 get_dp_netdev(const struct dpif *dpif)
170 return dpif_netdev_cast(dpif)->dp;
174 dpif_netdev_enumerate(struct sset *all_dps)
176 struct shash_node *node;
178 SHASH_FOR_EACH(node, &dp_netdevs) {
179 sset_add(all_dps, node->name);
185 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
187 return strcmp(type, "internal") ? type
188 : class != &dpif_netdev_class ? "dummy"
193 dpif_planetlab_port_open_type(const struct dpif_class *class, const char *type)
195 return strcmp(type, "internal") ? type
196 : class != &dpif_planetlab_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 &&
222 dp->class != &dpif_planetlab_class) {
226 /* If the port name begins with "br", start the number search at
227 * 100 to make writing tests easier. */
228 if (!strncmp(name, "br", 2)) {
232 /* If the port name contains a number, try to assign that port number.
233 * This can make writing unit tests easier because port numbers are
235 for (p = name; *p != '\0'; p++) {
236 if (isdigit((unsigned char) *p)) {
237 port_no = start_no + strtol(p, NULL, 10);
238 if (port_no > 0 && port_no < MAX_PORTS
239 && !dp->ports[port_no]) {
247 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
248 if (!dp->ports[port_no]) {
257 create_dp_netdev(const char *name, const struct dpif_class *class,
258 struct dp_netdev **dpp)
260 struct dp_netdev *dp;
264 dp = xzalloc(sizeof *dp);
266 dp->name = xstrdup(name);
268 for (i = 0; i < N_QUEUES; i++) {
269 dp->queues[i].head = dp->queues[i].tail = 0;
271 hmap_init(&dp->flow_table);
272 list_init(&dp->port_list);
274 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
280 shash_add(&dp_netdevs, name, dp);
287 dpif_netdev_open(const struct dpif_class *class, const char *name,
288 bool create, struct dpif **dpifp)
290 struct dp_netdev *dp;
292 dp = shash_find_data(&dp_netdevs, name);
297 int error = create_dp_netdev(name, class, &dp);
304 if (dp->class != class) {
311 *dpifp = create_dpif_netdev(dp);
316 dp_netdev_purge_queues(struct dp_netdev *dp)
320 for (i = 0; i < N_QUEUES; i++) {
321 struct dp_netdev_queue *q = &dp->queues[i];
323 while (q->tail != q->head) {
324 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
325 ofpbuf_uninit(&u->buf);
331 dp_netdev_free(struct dp_netdev *dp)
333 struct dp_netdev_port *port, *next;
335 dp_netdev_flow_flush(dp);
336 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
337 do_del_port(dp, port->port_no);
339 dp_netdev_purge_queues(dp);
340 hmap_destroy(&dp->flow_table);
346 dpif_netdev_close(struct dpif *dpif)
348 struct dp_netdev *dp = get_dp_netdev(dpif);
349 assert(dp->open_cnt > 0);
350 if (--dp->open_cnt == 0 && dp->destroyed) {
351 shash_find_and_delete(&dp_netdevs, dp->name);
358 dpif_netdev_destroy(struct dpif *dpif)
360 struct dp_netdev *dp = get_dp_netdev(dpif);
361 dp->destroyed = true;
366 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
368 struct dp_netdev *dp = get_dp_netdev(dpif);
369 stats->n_flows = hmap_count(&dp->flow_table);
370 stats->n_hit = dp->n_hit;
371 stats->n_missed = dp->n_missed;
372 stats->n_lost = dp->n_lost;
377 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
380 struct dp_netdev_port *port;
381 struct netdev *netdev;
382 const char *open_type;
386 /* XXX reject devices already in some dp_netdev. */
388 /* Open and validate network device. */
389 open_type = dpif_netdev_port_open_type(dp->class, type);
390 error = netdev_open(devname, open_type, &netdev);
394 /* XXX reject loopback devices */
395 /* XXX reject non-Ethernet devices */
397 error = netdev_listen(netdev);
399 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
400 devname, strerror(errno));
401 netdev_close(netdev);
405 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
407 netdev_close(netdev);
411 port = xmalloc(sizeof *port);
412 port->port_no = port_no;
413 port->netdev = netdev;
414 port->type = xstrdup(type);
416 error = netdev_get_mtu(netdev, &mtu);
421 list_push_back(&dp->port_list, &port->node);
422 dp->ports[port_no] = port;
429 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
432 struct dp_netdev *dp = get_dp_netdev(dpif);
435 if (*port_nop != UINT32_MAX) {
436 if (*port_nop >= MAX_PORTS) {
438 } else if (dp->ports[*port_nop]) {
443 port_no = choose_port(dp, netdev_get_name(netdev));
447 return do_add_port(dp, netdev_get_name(netdev),
448 netdev_get_type(netdev), port_no);
454 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
456 struct dp_netdev *dp = get_dp_netdev(dpif);
457 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
461 is_valid_port_number(uint32_t port_no)
463 return port_no < MAX_PORTS;
467 get_port_by_number(struct dp_netdev *dp,
468 uint32_t port_no, struct dp_netdev_port **portp)
470 if (!is_valid_port_number(port_no)) {
474 *portp = dp->ports[port_no];
475 return *portp ? 0 : ENOENT;
480 get_port_by_name(struct dp_netdev *dp,
481 const char *devname, struct dp_netdev_port **portp)
483 struct dp_netdev_port *port;
485 LIST_FOR_EACH (port, node, &dp->port_list) {
486 if (!strcmp(netdev_get_name(port->netdev), devname)) {
495 do_del_port(struct dp_netdev *dp, uint32_t port_no)
497 struct dp_netdev_port *port;
501 error = get_port_by_number(dp, port_no, &port);
506 list_remove(&port->node);
507 dp->ports[port->port_no] = NULL;
510 name = xstrdup(netdev_get_name(port->netdev));
511 netdev_close(port->netdev);
521 answer_port_query(const struct dp_netdev_port *port,
522 struct dpif_port *dpif_port)
524 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
525 dpif_port->type = xstrdup(port->type);
526 dpif_port->port_no = port->port_no;
530 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
531 struct dpif_port *dpif_port)
533 struct dp_netdev *dp = get_dp_netdev(dpif);
534 struct dp_netdev_port *port;
537 error = get_port_by_number(dp, port_no, &port);
538 if (!error && dpif_port) {
539 answer_port_query(port, dpif_port);
545 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
546 struct dpif_port *dpif_port)
548 struct dp_netdev *dp = get_dp_netdev(dpif);
549 struct dp_netdev_port *port;
552 error = get_port_by_name(dp, devname, &port);
553 if (!error && dpif_port) {
554 answer_port_query(port, dpif_port);
560 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
566 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
568 hmap_remove(&dp->flow_table, &flow->node);
574 dp_netdev_flow_flush(struct dp_netdev *dp)
576 struct dp_netdev_flow *flow, *next;
578 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
579 dp_netdev_free_flow(dp, flow);
584 dpif_netdev_flow_flush(struct dpif *dpif)
586 struct dp_netdev *dp = get_dp_netdev(dpif);
587 dp_netdev_flow_flush(dp);
591 struct dp_netdev_port_state {
597 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
599 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
604 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
605 struct dpif_port *dpif_port)
607 struct dp_netdev_port_state *state = state_;
608 struct dp_netdev *dp = get_dp_netdev(dpif);
611 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
612 struct dp_netdev_port *port = dp->ports[port_no];
615 state->name = xstrdup(netdev_get_name(port->netdev));
616 dpif_port->name = state->name;
617 dpif_port->type = port->type;
618 dpif_port->port_no = port->port_no;
619 state->port_no = port_no + 1;
627 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
629 struct dp_netdev_port_state *state = state_;
636 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
638 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
639 if (dpif->dp_serial != dpif->dp->serial) {
640 dpif->dp_serial = dpif->dp->serial;
648 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
650 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
651 if (dpif->dp_serial != dpif->dp->serial) {
652 poll_immediate_wake();
656 static struct dp_netdev_flow *
657 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
659 struct dp_netdev_flow *flow;
661 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
662 if (flow_equal(&flow->key, key)) {
670 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
672 stats->n_packets = flow->packet_count;
673 stats->n_bytes = flow->byte_count;
674 stats->used = flow->used;
675 stats->tcp_flags = flow->tcp_flags;
679 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
682 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
683 /* This should not happen: it indicates that odp_flow_key_from_flow()
684 * and odp_flow_key_to_flow() disagree on the acceptable form of a
685 * flow. Log the problem as an error, with enough details to enable
687 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
689 if (!VLOG_DROP_ERR(&rl)) {
693 odp_flow_key_format(key, key_len, &s);
694 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
701 if (flow->in_port < OFPP_MAX
702 ? flow->in_port >= MAX_PORTS
703 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
711 dpif_netdev_flow_get(const struct dpif *dpif,
712 const struct nlattr *nl_key, size_t nl_key_len,
713 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
715 struct dp_netdev *dp = get_dp_netdev(dpif);
716 struct dp_netdev_flow *flow;
720 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
725 flow = dp_netdev_lookup_flow(dp, &key);
731 get_dpif_flow_stats(flow, stats);
734 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
740 set_flow_actions(struct dp_netdev_flow *flow,
741 const struct nlattr *actions, size_t actions_len)
743 flow->actions = xrealloc(flow->actions, actions_len);
744 flow->actions_len = actions_len;
745 memcpy(flow->actions, actions, actions_len);
750 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
751 const struct nlattr *actions, size_t actions_len)
753 struct dp_netdev_flow *flow;
756 flow = xzalloc(sizeof *flow);
759 error = set_flow_actions(flow, actions, actions_len);
765 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
770 clear_stats(struct dp_netdev_flow *flow)
773 flow->packet_count = 0;
774 flow->byte_count = 0;
779 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
781 struct dp_netdev *dp = get_dp_netdev(dpif);
782 struct dp_netdev_flow *flow;
786 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
791 flow = dp_netdev_lookup_flow(dp, &key);
793 if (put->flags & DPIF_FP_CREATE) {
794 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
796 memset(put->stats, 0, sizeof *put->stats);
798 return dp_netdev_flow_add(dp, &key, put->actions,
807 if (put->flags & DPIF_FP_MODIFY) {
808 int error = set_flow_actions(flow, put->actions, put->actions_len);
811 get_dpif_flow_stats(flow, put->stats);
813 if (put->flags & DPIF_FP_ZERO_STATS) {
825 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
827 struct dp_netdev *dp = get_dp_netdev(dpif);
828 struct dp_netdev_flow *flow;
832 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
837 flow = dp_netdev_lookup_flow(dp, &key);
840 get_dpif_flow_stats(flow, del->stats);
842 dp_netdev_free_flow(dp, flow);
849 struct dp_netdev_flow_state {
852 struct nlattr *actions;
853 struct odputil_keybuf keybuf;
854 struct dpif_flow_stats stats;
858 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
860 struct dp_netdev_flow_state *state;
862 *statep = state = xmalloc(sizeof *state);
865 state->actions = NULL;
870 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
871 const struct nlattr **key, size_t *key_len,
872 const struct nlattr **actions, size_t *actions_len,
873 const struct dpif_flow_stats **stats)
875 struct dp_netdev_flow_state *state = state_;
876 struct dp_netdev *dp = get_dp_netdev(dpif);
877 struct dp_netdev_flow *flow;
878 struct hmap_node *node;
880 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
885 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
890 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
891 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
898 free(state->actions);
899 state->actions = xmemdup(flow->actions, flow->actions_len);
901 *actions = state->actions;
902 *actions_len = flow->actions_len;
906 get_dpif_flow_stats(flow, &state->stats);
907 *stats = &state->stats;
914 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
916 struct dp_netdev_flow_state *state = state_;
918 free(state->actions);
924 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
926 struct dp_netdev *dp = get_dp_netdev(dpif);
931 if (execute->packet->size < ETH_HEADER_LEN ||
932 execute->packet->size > UINT16_MAX) {
936 /* Make a deep copy of 'packet', because we might modify its data. */
937 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
938 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
939 ofpbuf_put(©, execute->packet->data, execute->packet->size);
941 flow_extract(©, 0, 0, NULL, -1, &key);
942 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
945 dp_netdev_execute_actions(dp, ©, &key,
946 execute->actions, execute->actions_len);
949 ofpbuf_uninit(©);
954 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
960 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
961 uint32_t queue_id, uint32_t *priority)
963 *priority = queue_id;
967 static struct dp_netdev_queue *
968 find_nonempty_queue(struct dpif *dpif)
970 struct dp_netdev *dp = get_dp_netdev(dpif);
973 for (i = 0; i < N_QUEUES; i++) {
974 struct dp_netdev_queue *q = &dp->queues[i];
975 if (q->head != q->tail) {
983 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
986 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
988 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
991 upcall->packet = buf;
1003 dpif_netdev_recv_wait(struct dpif *dpif)
1005 if (find_nonempty_queue(dpif)) {
1006 poll_immediate_wake();
1008 /* No messages ready to be received, and dp_wait() will ensure that we
1009 * wake up to queue new messages, so there is nothing to do. */
1014 dpif_netdev_recv_purge(struct dpif *dpif)
1016 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1017 dp_netdev_purge_queues(dpif_netdev->dp);
1021 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1023 flow->used = time_msec();
1024 flow->packet_count++;
1025 flow->byte_count += packet->size;
1026 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1030 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1031 struct ofpbuf *packet)
1033 struct dp_netdev_flow *flow;
1036 if (packet->size < ETH_HEADER_LEN) {
1039 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1040 flow = dp_netdev_lookup_flow(dp, &key);
1042 dp_netdev_flow_used(flow, packet);
1043 dp_netdev_execute_actions(dp, packet, &key,
1044 flow->actions, flow->actions_len);
1048 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1053 dpif_netdev_run(struct dpif *dpif)
1055 struct dp_netdev *dp = get_dp_netdev(dpif);
1056 struct dp_netdev_port *port;
1057 struct ofpbuf packet;
1059 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1061 LIST_FOR_EACH (port, node, &dp->port_list) {
1064 /* Reset packet contents. */
1065 ofpbuf_clear(&packet);
1066 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1068 error = netdev_recv(port->netdev, &packet);
1070 dp_netdev_port_input(dp, port, &packet);
1071 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1073 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1074 netdev_get_name(port->netdev), strerror(error));
1077 ofpbuf_uninit(&packet);
1081 dpif_netdev_wait(struct dpif *dpif)
1083 struct dp_netdev *dp = get_dp_netdev(dpif);
1084 struct dp_netdev_port *port;
1086 LIST_FOR_EACH (port, node, &dp->port_list) {
1087 netdev_recv_wait(port->netdev);
1092 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1094 struct eth_header *eh = packet->l2;
1096 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1097 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1101 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1104 struct dp_netdev_port *p = dp->ports[out_port];
1106 netdev_send(p->netdev, packet);
1111 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1112 int queue_no, const struct flow *flow, uint64_t arg)
1114 struct dp_netdev_queue *q = &dp->queues[queue_no];
1115 struct dp_netdev_upcall *u;
1116 struct dpif_upcall *upcall;
1120 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1125 u = &q->upcalls[q->head++ & QUEUE_MASK];
1128 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1129 odp_flow_key_from_flow(buf, flow, flow->in_port);
1130 key_len = buf->size;
1131 ofpbuf_pull(buf, key_len);
1132 ofpbuf_reserve(buf, 2);
1133 ofpbuf_put(buf, packet->data, packet->size);
1135 upcall = &u->upcall;
1136 upcall->type = queue_no;
1137 upcall->packet = buf;
1138 upcall->key = buf->base;
1139 upcall->key_len = key_len;
1140 upcall->userdata = arg;
1146 dp_netdev_sample(struct dp_netdev *dp,
1147 struct ofpbuf *packet, struct flow *key,
1148 const struct nlattr *action)
1150 const struct nlattr *subactions = NULL;
1151 const struct nlattr *a;
1154 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1155 int type = nl_attr_type(a);
1157 switch ((enum ovs_sample_attr) type) {
1158 case OVS_SAMPLE_ATTR_PROBABILITY:
1159 if (random_uint32() >= nl_attr_get_u32(a)) {
1164 case OVS_SAMPLE_ATTR_ACTIONS:
1168 case OVS_SAMPLE_ATTR_UNSPEC:
1169 case __OVS_SAMPLE_ATTR_MAX:
1175 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1176 nl_attr_get_size(subactions));
1180 dp_netdev_action_userspace(struct dp_netdev *dp,
1181 struct ofpbuf *packet, struct flow *key,
1182 const struct nlattr *a)
1184 const struct nlattr *userdata_attr;
1187 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1188 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1189 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1193 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1195 enum ovs_key_attr type = nl_attr_type(a);
1196 const struct ovs_key_ipv4 *ipv4_key;
1197 const struct ovs_key_ipv6 *ipv6_key;
1198 const struct ovs_key_tcp *tcp_key;
1199 const struct ovs_key_udp *udp_key;
1202 case OVS_KEY_ATTR_TUN_ID:
1203 case OVS_KEY_ATTR_PRIORITY:
1204 case OVS_KEY_ATTR_SKB_MARK:
1205 case OVS_KEY_ATTR_IPV4_TUNNEL:
1206 /* not implemented */
1209 case OVS_KEY_ATTR_ETHERNET:
1210 dp_netdev_set_dl(packet,
1211 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1214 case OVS_KEY_ATTR_IPV4:
1215 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1216 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1217 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1220 case OVS_KEY_ATTR_IPV6:
1221 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1222 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1223 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1224 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1227 case OVS_KEY_ATTR_TCP:
1228 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1229 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1232 case OVS_KEY_ATTR_UDP:
1233 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1234 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1237 case OVS_KEY_ATTR_UNSPEC:
1238 case OVS_KEY_ATTR_ENCAP:
1239 case OVS_KEY_ATTR_ETHERTYPE:
1240 case OVS_KEY_ATTR_IN_PORT:
1241 case OVS_KEY_ATTR_VLAN:
1242 case OVS_KEY_ATTR_ICMP:
1243 case OVS_KEY_ATTR_ICMPV6:
1244 case OVS_KEY_ATTR_ARP:
1245 case OVS_KEY_ATTR_ND:
1246 case __OVS_KEY_ATTR_MAX:
1253 dp_netdev_execute_actions(struct dp_netdev *dp,
1254 struct ofpbuf *packet, struct flow *key,
1255 const struct nlattr *actions,
1258 const struct nlattr *a;
1261 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1262 const struct ovs_action_push_vlan *vlan;
1263 int type = nl_attr_type(a);
1265 switch ((enum ovs_action_attr) type) {
1266 case OVS_ACTION_ATTR_OUTPUT:
1267 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1270 case OVS_ACTION_ATTR_USERSPACE:
1271 dp_netdev_action_userspace(dp, packet, key, a);
1274 case OVS_ACTION_ATTR_PUSH_VLAN:
1275 vlan = nl_attr_get(a);
1276 eth_push_vlan(packet, vlan->vlan_tci);
1279 case OVS_ACTION_ATTR_POP_VLAN:
1280 eth_pop_vlan(packet);
1283 case OVS_ACTION_ATTR_SET:
1284 execute_set_action(packet, nl_attr_get(a));
1287 case OVS_ACTION_ATTR_SAMPLE:
1288 dp_netdev_sample(dp, packet, key, a);
1291 case OVS_ACTION_ATTR_UNSPEC:
1292 case __OVS_ACTION_ATTR_MAX:
1298 #define DPIF_NETDEV_CLASS_FUNCTIONS(PORT_OPEN_TYPE) \
1299 dpif_netdev_enumerate, \
1302 dpif_netdev_close, \
1303 dpif_netdev_destroy, \
1306 dpif_netdev_get_stats, \
1307 dpif_netdev_port_add, \
1308 dpif_netdev_port_del, \
1309 dpif_netdev_port_query_by_number, \
1310 dpif_netdev_port_query_by_name, \
1311 dpif_netdev_get_max_ports, \
1312 NULL, /* port_get_pid */ \
1313 dpif_netdev_port_dump_start, \
1314 dpif_netdev_port_dump_next, \
1315 dpif_netdev_port_dump_done, \
1316 dpif_netdev_port_poll, \
1317 dpif_netdev_port_poll_wait, \
1318 dpif_netdev_flow_get, \
1319 dpif_netdev_flow_put, \
1320 dpif_netdev_flow_del, \
1321 dpif_netdev_flow_flush, \
1322 dpif_netdev_flow_dump_start, \
1323 dpif_netdev_flow_dump_next, \
1324 dpif_netdev_flow_dump_done, \
1325 dpif_netdev_execute, \
1326 NULL, /* operate */ \
1327 dpif_netdev_recv_set, \
1328 dpif_netdev_queue_to_priority, \
1330 dpif_netdev_recv_wait, \
1331 dpif_netdev_recv_purge, \
1333 const struct dpif_class dpif_netdev_class = {
1335 DPIF_NETDEV_CLASS_FUNCTIONS(dpif_netdev_port_open_type)
1338 const struct dpif_class dpif_planetlab_class = {
1340 DPIF_NETDEV_CLASS_FUNCTIONS(dpif_planetlab_port_open_type)
1344 dpif_dummy_register__(const char *type)
1346 struct dpif_class *class;
1348 class = xmalloc(sizeof *class);
1349 *class = dpif_netdev_class;
1350 class->type = xstrdup(type);
1351 dp_register_provider(class);
1355 dpif_dummy_register(bool override)
1362 dp_enumerate_types(&types);
1363 SSET_FOR_EACH (type, &types) {
1364 if (!dp_unregister_provider(type)) {
1365 dpif_dummy_register__(type);
1368 sset_destroy(&types);
1371 dpif_dummy_register__("dummy");