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_class_is_planetlab(const struct dpif_class *class)
195 return class == &dpif_planetlab_class;
199 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
201 return strcmp(type, "internal") ? type
202 : dpif_netdev_class_is_planetlab(class) ? "pltap"
203 : dpif_netdev_class_is_dummy(class) ? "dummy"
208 create_dpif_netdev(struct dp_netdev *dp)
210 uint16_t netflow_id = hash_string(dp->name, 0);
211 struct dpif_netdev *dpif;
215 dpif = xmalloc(sizeof *dpif);
216 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
218 dpif->dp_serial = dp->serial;
224 choose_port(struct dp_netdev *dp, const char *name)
228 if (dp->class != &dpif_netdev_class &&
229 dp->class != &dpif_planetlab_class) {
233 /* If the port name begins with "br", start the number search at
234 * 100 to make writing tests easier. */
235 if (!strncmp(name, "br", 2)) {
239 /* If the port name contains a number, try to assign that port number.
240 * This can make writing unit tests easier because port numbers are
242 for (p = name; *p != '\0'; p++) {
243 if (isdigit((unsigned char) *p)) {
244 port_no = start_no + strtol(p, NULL, 10);
245 if (port_no > 0 && port_no < MAX_PORTS
246 && !dp->ports[port_no]) {
254 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
255 if (!dp->ports[port_no]) {
264 create_dp_netdev(const char *name, const struct dpif_class *class,
265 struct dp_netdev **dpp)
267 struct dp_netdev *dp;
271 dp = xzalloc(sizeof *dp);
273 dp->name = xstrdup(name);
275 for (i = 0; i < N_QUEUES; i++) {
276 dp->queues[i].head = dp->queues[i].tail = 0;
278 hmap_init(&dp->flow_table);
279 list_init(&dp->port_list);
281 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
287 shash_add(&dp_netdevs, name, dp);
294 dpif_netdev_open(const struct dpif_class *class, const char *name,
295 bool create, struct dpif **dpifp)
297 struct dp_netdev *dp;
299 dp = shash_find_data(&dp_netdevs, name);
304 int error = create_dp_netdev(name, class, &dp);
308 ovs_assert(dp != NULL);
311 if (dp->class != class) {
318 *dpifp = create_dpif_netdev(dp);
323 dp_netdev_purge_queues(struct dp_netdev *dp)
327 for (i = 0; i < N_QUEUES; i++) {
328 struct dp_netdev_queue *q = &dp->queues[i];
330 while (q->tail != q->head) {
331 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
332 ofpbuf_uninit(&u->buf);
338 dp_netdev_free(struct dp_netdev *dp)
340 struct dp_netdev_port *port, *next;
342 dp_netdev_flow_flush(dp);
343 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
344 do_del_port(dp, port->port_no);
346 dp_netdev_purge_queues(dp);
347 hmap_destroy(&dp->flow_table);
353 dpif_netdev_close(struct dpif *dpif)
355 struct dp_netdev *dp = get_dp_netdev(dpif);
356 ovs_assert(dp->open_cnt > 0);
357 if (--dp->open_cnt == 0 && dp->destroyed) {
358 shash_find_and_delete(&dp_netdevs, dp->name);
365 dpif_netdev_destroy(struct dpif *dpif)
367 struct dp_netdev *dp = get_dp_netdev(dpif);
368 dp->destroyed = true;
373 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
375 struct dp_netdev *dp = get_dp_netdev(dpif);
376 stats->n_flows = hmap_count(&dp->flow_table);
377 stats->n_hit = dp->n_hit;
378 stats->n_missed = dp->n_missed;
379 stats->n_lost = dp->n_lost;
384 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
387 struct netdev_saved_flags *sf;
388 struct dp_netdev_port *port;
389 struct netdev *netdev;
390 struct netdev_rx *rx;
391 const char *open_type;
395 /* XXX reject devices already in some dp_netdev. */
397 /* Open and validate network device. */
398 open_type = dpif_netdev_port_open_type(dp->class, type);
399 error = netdev_open(devname, open_type, &netdev);
403 /* XXX reject loopback devices */
404 /* XXX reject non-Ethernet devices */
406 error = netdev_rx_open(netdev, &rx);
408 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
409 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
410 devname, strerror(errno));
411 netdev_close(netdev);
415 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
418 netdev_close(netdev);
422 port = xmalloc(sizeof *port);
423 port->port_no = port_no;
424 port->netdev = netdev;
427 port->type = xstrdup(type);
429 error = netdev_get_mtu(netdev, &mtu);
430 if (!error && mtu > max_mtu) {
434 list_push_back(&dp->port_list, &port->node);
435 dp->ports[port_no] = port;
442 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
445 struct dp_netdev *dp = get_dp_netdev(dpif);
448 if (*port_nop != UINT32_MAX) {
449 if (*port_nop >= MAX_PORTS) {
451 } else if (dp->ports[*port_nop]) {
456 port_no = choose_port(dp, netdev_vport_get_dpif_port(netdev));
460 return do_add_port(dp, netdev_vport_get_dpif_port(netdev),
461 netdev_get_type(netdev), port_no);
467 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
469 struct dp_netdev *dp = get_dp_netdev(dpif);
470 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
474 is_valid_port_number(uint32_t port_no)
476 return port_no < MAX_PORTS;
480 get_port_by_number(struct dp_netdev *dp,
481 uint32_t port_no, struct dp_netdev_port **portp)
483 if (!is_valid_port_number(port_no)) {
487 *portp = dp->ports[port_no];
488 return *portp ? 0 : ENOENT;
493 get_port_by_name(struct dp_netdev *dp,
494 const char *devname, struct dp_netdev_port **portp)
496 struct dp_netdev_port *port;
498 LIST_FOR_EACH (port, node, &dp->port_list) {
499 if (!strcmp(netdev_vport_get_dpif_port(port->netdev), devname)) {
508 do_del_port(struct dp_netdev *dp, uint32_t port_no)
510 struct dp_netdev_port *port;
513 error = get_port_by_number(dp, port_no, &port);
518 list_remove(&port->node);
519 dp->ports[port->port_no] = NULL;
522 netdev_close(port->netdev);
523 netdev_restore_flags(port->sf);
524 netdev_rx_close(port->rx);
532 answer_port_query(const struct dp_netdev_port *port,
533 struct dpif_port *dpif_port)
535 dpif_port->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
536 dpif_port->type = xstrdup(port->type);
537 dpif_port->port_no = port->port_no;
541 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
542 struct dpif_port *dpif_port)
544 struct dp_netdev *dp = get_dp_netdev(dpif);
545 struct dp_netdev_port *port;
548 error = get_port_by_number(dp, port_no, &port);
549 if (!error && dpif_port) {
550 answer_port_query(port, dpif_port);
556 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
557 struct dpif_port *dpif_port)
559 struct dp_netdev *dp = get_dp_netdev(dpif);
560 struct dp_netdev_port *port;
563 error = get_port_by_name(dp, devname, &port);
564 if (!error && dpif_port) {
565 answer_port_query(port, dpif_port);
571 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
577 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
579 hmap_remove(&dp->flow_table, &flow->node);
585 dp_netdev_flow_flush(struct dp_netdev *dp)
587 struct dp_netdev_flow *flow, *next;
589 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
590 dp_netdev_free_flow(dp, flow);
595 dpif_netdev_flow_flush(struct dpif *dpif)
597 struct dp_netdev *dp = get_dp_netdev(dpif);
598 dp_netdev_flow_flush(dp);
602 struct dp_netdev_port_state {
608 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
610 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
615 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
616 struct dpif_port *dpif_port)
618 struct dp_netdev_port_state *state = state_;
619 struct dp_netdev *dp = get_dp_netdev(dpif);
622 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
623 struct dp_netdev_port *port = dp->ports[port_no];
626 state->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
627 dpif_port->name = state->name;
628 dpif_port->type = port->type;
629 dpif_port->port_no = port->port_no;
630 state->port_no = port_no + 1;
638 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
640 struct dp_netdev_port_state *state = state_;
647 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
649 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
650 if (dpif->dp_serial != dpif->dp->serial) {
651 dpif->dp_serial = dpif->dp->serial;
659 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
661 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
662 if (dpif->dp_serial != dpif->dp->serial) {
663 poll_immediate_wake();
667 static struct dp_netdev_flow *
668 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
670 struct dp_netdev_flow *flow;
672 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
673 if (flow_equal(&flow->key, key)) {
681 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
683 stats->n_packets = flow->packet_count;
684 stats->n_bytes = flow->byte_count;
685 stats->used = flow->used;
686 stats->tcp_flags = flow->tcp_flags;
690 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
693 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
694 /* This should not happen: it indicates that odp_flow_key_from_flow()
695 * and odp_flow_key_to_flow() disagree on the acceptable form of a
696 * flow. Log the problem as an error, with enough details to enable
698 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
700 if (!VLOG_DROP_ERR(&rl)) {
704 odp_flow_key_format(key, key_len, &s);
705 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
712 if (flow->in_port < OFPP_MAX
713 ? flow->in_port >= MAX_PORTS
714 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
722 dpif_netdev_flow_get(const struct dpif *dpif,
723 const struct nlattr *nl_key, size_t nl_key_len,
724 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
726 struct dp_netdev *dp = get_dp_netdev(dpif);
727 struct dp_netdev_flow *flow;
731 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
736 flow = dp_netdev_lookup_flow(dp, &key);
742 get_dpif_flow_stats(flow, stats);
745 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
751 set_flow_actions(struct dp_netdev_flow *flow,
752 const struct nlattr *actions, size_t actions_len)
754 flow->actions = xrealloc(flow->actions, actions_len);
755 flow->actions_len = actions_len;
756 memcpy(flow->actions, actions, actions_len);
761 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
762 const struct nlattr *actions, size_t actions_len)
764 struct dp_netdev_flow *flow;
767 flow = xzalloc(sizeof *flow);
770 error = set_flow_actions(flow, actions, actions_len);
776 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
781 clear_stats(struct dp_netdev_flow *flow)
784 flow->packet_count = 0;
785 flow->byte_count = 0;
790 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
792 struct dp_netdev *dp = get_dp_netdev(dpif);
793 struct dp_netdev_flow *flow;
797 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
802 flow = dp_netdev_lookup_flow(dp, &key);
804 if (put->flags & DPIF_FP_CREATE) {
805 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
807 memset(put->stats, 0, sizeof *put->stats);
809 return dp_netdev_flow_add(dp, &key, put->actions,
818 if (put->flags & DPIF_FP_MODIFY) {
819 int error = set_flow_actions(flow, put->actions, put->actions_len);
822 get_dpif_flow_stats(flow, put->stats);
824 if (put->flags & DPIF_FP_ZERO_STATS) {
836 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
838 struct dp_netdev *dp = get_dp_netdev(dpif);
839 struct dp_netdev_flow *flow;
843 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
848 flow = dp_netdev_lookup_flow(dp, &key);
851 get_dpif_flow_stats(flow, del->stats);
853 dp_netdev_free_flow(dp, flow);
860 struct dp_netdev_flow_state {
863 struct nlattr *actions;
864 struct odputil_keybuf keybuf;
865 struct dpif_flow_stats stats;
869 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
871 struct dp_netdev_flow_state *state;
873 *statep = state = xmalloc(sizeof *state);
876 state->actions = NULL;
881 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
882 const struct nlattr **key, size_t *key_len,
883 const struct nlattr **actions, size_t *actions_len,
884 const struct dpif_flow_stats **stats)
886 struct dp_netdev_flow_state *state = state_;
887 struct dp_netdev *dp = get_dp_netdev(dpif);
888 struct dp_netdev_flow *flow;
889 struct hmap_node *node;
891 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
896 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
901 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
902 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
909 free(state->actions);
910 state->actions = xmemdup(flow->actions, flow->actions_len);
912 *actions = state->actions;
913 *actions_len = flow->actions_len;
917 get_dpif_flow_stats(flow, &state->stats);
918 *stats = &state->stats;
925 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
927 struct dp_netdev_flow_state *state = state_;
929 free(state->actions);
935 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
937 struct dp_netdev *dp = get_dp_netdev(dpif);
942 if (execute->packet->size < ETH_HEADER_LEN ||
943 execute->packet->size > UINT16_MAX) {
947 /* Make a deep copy of 'packet', because we might modify its data. */
948 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
949 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
950 ofpbuf_put(©, execute->packet->data, execute->packet->size);
952 flow_extract(©, 0, 0, NULL, -1, &key);
953 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
956 dp_netdev_execute_actions(dp, ©, &key,
957 execute->actions, execute->actions_len);
960 ofpbuf_uninit(©);
965 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
971 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
972 uint32_t queue_id, uint32_t *priority)
974 *priority = queue_id;
978 static struct dp_netdev_queue *
979 find_nonempty_queue(struct dpif *dpif)
981 struct dp_netdev *dp = get_dp_netdev(dpif);
984 for (i = 0; i < N_QUEUES; i++) {
985 struct dp_netdev_queue *q = &dp->queues[i];
986 if (q->head != q->tail) {
994 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
997 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
999 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1001 *upcall = u->upcall;
1002 upcall->packet = buf;
1014 dpif_netdev_recv_wait(struct dpif *dpif)
1016 if (find_nonempty_queue(dpif)) {
1017 poll_immediate_wake();
1019 /* No messages ready to be received, and dp_wait() will ensure that we
1020 * wake up to queue new messages, so there is nothing to do. */
1025 dpif_netdev_recv_purge(struct dpif *dpif)
1027 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1028 dp_netdev_purge_queues(dpif_netdev->dp);
1032 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1034 flow->used = time_msec();
1035 flow->packet_count++;
1036 flow->byte_count += packet->size;
1037 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1041 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1042 struct ofpbuf *packet)
1044 struct dp_netdev_flow *flow;
1047 if (packet->size < ETH_HEADER_LEN) {
1050 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1051 flow = dp_netdev_lookup_flow(dp, &key);
1053 dp_netdev_flow_used(flow, packet);
1054 dp_netdev_execute_actions(dp, packet, &key,
1055 flow->actions, flow->actions_len);
1059 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1064 dpif_netdev_run(struct dpif *dpif)
1066 struct dp_netdev *dp = get_dp_netdev(dpif);
1067 struct dp_netdev_port *port;
1068 struct ofpbuf packet;
1070 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1072 LIST_FOR_EACH (port, node, &dp->port_list) {
1075 /* Reset packet contents. */
1076 ofpbuf_clear(&packet);
1077 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1079 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1081 dp_netdev_port_input(dp, port, &packet);
1082 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1083 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1084 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1085 netdev_vport_get_dpif_port(port->netdev),
1089 ofpbuf_uninit(&packet);
1093 dpif_netdev_wait(struct dpif *dpif)
1095 struct dp_netdev *dp = get_dp_netdev(dpif);
1096 struct dp_netdev_port *port;
1098 LIST_FOR_EACH (port, node, &dp->port_list) {
1100 netdev_rx_wait(port->rx);
1106 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1108 struct eth_header *eh = packet->l2;
1110 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1111 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1115 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1118 struct dp_netdev_port *p = dp->ports[out_port];
1120 netdev_send(p->netdev, packet);
1125 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1126 int queue_no, const struct flow *flow,
1127 const struct nlattr *userdata)
1129 struct dp_netdev_queue *q = &dp->queues[queue_no];
1130 if (q->head - q->tail < MAX_QUEUE_LEN) {
1131 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1132 struct dpif_upcall *upcall = &u->upcall;
1133 struct ofpbuf *buf = &u->buf;
1136 upcall->type = queue_no;
1138 /* Allocate buffer big enough for everything. */
1139 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1141 buf_size += NLA_ALIGN(userdata->nla_len);
1143 ofpbuf_init(buf, buf_size);
1146 odp_flow_key_from_flow(buf, flow, flow->in_port);
1147 upcall->key = buf->data;
1148 upcall->key_len = buf->size;
1152 upcall->userdata = ofpbuf_put(buf, userdata,
1153 NLA_ALIGN(userdata->nla_len));
1158 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1159 * is visible in 'upcall->packet'. The ODP flow and (if present)
1160 * userdata become part of the headroom. */
1161 ofpbuf_put_zeros(buf, 2);
1162 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1163 buf->size = packet->size;
1164 upcall->packet = buf;
1174 dp_netdev_sample(struct dp_netdev *dp,
1175 struct ofpbuf *packet, struct flow *key,
1176 const struct nlattr *action)
1178 const struct nlattr *subactions = NULL;
1179 const struct nlattr *a;
1182 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1183 int type = nl_attr_type(a);
1185 switch ((enum ovs_sample_attr) type) {
1186 case OVS_SAMPLE_ATTR_PROBABILITY:
1187 if (random_uint32() >= nl_attr_get_u32(a)) {
1192 case OVS_SAMPLE_ATTR_ACTIONS:
1196 case OVS_SAMPLE_ATTR_UNSPEC:
1197 case __OVS_SAMPLE_ATTR_MAX:
1203 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1204 nl_attr_get_size(subactions));
1208 dp_netdev_action_userspace(struct dp_netdev *dp,
1209 struct ofpbuf *packet, struct flow *key,
1210 const struct nlattr *a)
1212 const struct nlattr *userdata;
1214 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1215 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1219 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1221 enum ovs_key_attr type = nl_attr_type(a);
1222 const struct ovs_key_ipv4 *ipv4_key;
1223 const struct ovs_key_ipv6 *ipv6_key;
1224 const struct ovs_key_tcp *tcp_key;
1225 const struct ovs_key_udp *udp_key;
1228 case OVS_KEY_ATTR_PRIORITY:
1229 case OVS_KEY_ATTR_SKB_MARK:
1230 case OVS_KEY_ATTR_TUNNEL:
1231 /* not implemented */
1234 case OVS_KEY_ATTR_ETHERNET:
1235 dp_netdev_set_dl(packet,
1236 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1239 case OVS_KEY_ATTR_IPV4:
1240 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1241 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1242 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1245 case OVS_KEY_ATTR_IPV6:
1246 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1247 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1248 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1249 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1252 case OVS_KEY_ATTR_TCP:
1253 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1254 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1257 case OVS_KEY_ATTR_UDP:
1258 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1259 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1262 case OVS_KEY_ATTR_MPLS:
1263 set_mpls_lse(packet, nl_attr_get_be32(a));
1266 case OVS_KEY_ATTR_UNSPEC:
1267 case OVS_KEY_ATTR_ENCAP:
1268 case OVS_KEY_ATTR_ETHERTYPE:
1269 case OVS_KEY_ATTR_IN_PORT:
1270 case OVS_KEY_ATTR_VLAN:
1271 case OVS_KEY_ATTR_ICMP:
1272 case OVS_KEY_ATTR_ICMPV6:
1273 case OVS_KEY_ATTR_ARP:
1274 case OVS_KEY_ATTR_ND:
1275 case __OVS_KEY_ATTR_MAX:
1282 dp_netdev_execute_actions(struct dp_netdev *dp,
1283 struct ofpbuf *packet, struct flow *key,
1284 const struct nlattr *actions,
1287 const struct nlattr *a;
1290 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1291 int type = nl_attr_type(a);
1293 switch ((enum ovs_action_attr) type) {
1294 case OVS_ACTION_ATTR_OUTPUT:
1295 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1298 case OVS_ACTION_ATTR_USERSPACE:
1299 dp_netdev_action_userspace(dp, packet, key, a);
1302 case OVS_ACTION_ATTR_PUSH_VLAN: {
1303 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1304 eth_push_vlan(packet, vlan->vlan_tci);
1308 case OVS_ACTION_ATTR_POP_VLAN:
1309 eth_pop_vlan(packet);
1312 case OVS_ACTION_ATTR_PUSH_MPLS: {
1313 const struct ovs_action_push_mpls *mpls = nl_attr_get(a);
1314 push_mpls(packet, mpls->mpls_ethertype, mpls->mpls_lse);
1318 case OVS_ACTION_ATTR_POP_MPLS:
1319 pop_mpls(packet, nl_attr_get_be16(a));
1322 case OVS_ACTION_ATTR_SET:
1323 execute_set_action(packet, nl_attr_get(a));
1326 case OVS_ACTION_ATTR_SAMPLE:
1327 dp_netdev_sample(dp, packet, key, a);
1330 case OVS_ACTION_ATTR_UNSPEC:
1331 case __OVS_ACTION_ATTR_MAX:
1337 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1338 dpif_netdev_enumerate, \
1339 dpif_netdev_port_open_type, \
1341 dpif_netdev_close, \
1342 dpif_netdev_destroy, \
1345 dpif_netdev_get_stats, \
1346 dpif_netdev_port_add, \
1347 dpif_netdev_port_del, \
1348 dpif_netdev_port_query_by_number, \
1349 dpif_netdev_port_query_by_name, \
1350 dpif_netdev_get_max_ports, \
1351 NULL, /* port_get_pid */ \
1352 dpif_netdev_port_dump_start, \
1353 dpif_netdev_port_dump_next, \
1354 dpif_netdev_port_dump_done, \
1355 dpif_netdev_port_poll, \
1356 dpif_netdev_port_poll_wait, \
1357 dpif_netdev_flow_get, \
1358 dpif_netdev_flow_put, \
1359 dpif_netdev_flow_del, \
1360 dpif_netdev_flow_flush, \
1361 dpif_netdev_flow_dump_start, \
1362 dpif_netdev_flow_dump_next, \
1363 dpif_netdev_flow_dump_done, \
1364 dpif_netdev_execute, \
1365 NULL, /* operate */ \
1366 dpif_netdev_recv_set, \
1367 dpif_netdev_queue_to_priority, \
1369 dpif_netdev_recv_wait, \
1370 dpif_netdev_recv_purge, \
1372 const struct dpif_class dpif_netdev_class = {
1374 DPIF_NETDEV_CLASS_FUNCTIONS
1377 const struct dpif_class dpif_planetlab_class = {
1379 DPIF_NETDEV_CLASS_FUNCTIONS
1383 dpif_dummy_register__(const char *type)
1385 struct dpif_class *class;
1387 class = xmalloc(sizeof *class);
1388 *class = dpif_netdev_class;
1389 class->type = xstrdup(type);
1390 dp_register_provider(class);
1394 dpif_dummy_register(bool override)
1401 dp_enumerate_types(&types);
1402 SSET_FOR_EACH (type, &types) {
1403 if (!dp_unregister_provider(type)) {
1404 dpif_dummy_register__(type);
1407 sset_destroy(&types);
1410 dpif_dummy_register__("dummy");