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;
126 unsigned int dp_serial;
129 /* All netdev-based datapaths. */
130 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
132 /* Maximum port MTU seen so far. */
133 static int max_mtu = ETH_PAYLOAD_MAX;
135 static int get_port_by_number(struct dp_netdev *, uint16_t port_no,
136 struct dp_netdev_port **portp);
137 static int get_port_by_name(struct dp_netdev *, const char *devname,
138 struct dp_netdev_port **portp);
139 static void dp_netdev_free(struct dp_netdev *);
140 static void dp_netdev_flow_flush(struct dp_netdev *);
141 static int do_add_port(struct dp_netdev *, const char *devname,
142 const char *type, uint16_t port_no);
143 static int do_del_port(struct dp_netdev *, uint16_t port_no);
144 static int dpif_netdev_open(const struct dpif_class *, const char *name,
145 bool create, struct dpif **);
146 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
147 int queue_no, const struct flow *,
149 static void dp_netdev_execute_actions(struct dp_netdev *,
150 struct ofpbuf *, struct flow *,
151 const struct nlattr *actions,
154 static struct dpif_class dpif_dummy_class;
156 static struct dpif_netdev *
157 dpif_netdev_cast(const struct dpif *dpif)
159 assert(dpif->dpif_class->open == dpif_netdev_open);
160 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
163 static struct dp_netdev *
164 get_dp_netdev(const struct dpif *dpif)
166 return dpif_netdev_cast(dpif)->dp;
170 create_dpif_netdev(struct dp_netdev *dp)
172 uint16_t netflow_id = hash_string(dp->name, 0);
173 struct dpif_netdev *dpif;
177 dpif = xmalloc(sizeof *dpif);
178 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
180 dpif->dp_serial = dp->serial;
186 create_dp_netdev(const char *name, const struct dpif_class *class,
187 struct dp_netdev **dpp)
189 struct dp_netdev *dp;
193 dp = xzalloc(sizeof *dp);
195 dp->name = xstrdup(name);
197 for (i = 0; i < N_QUEUES; i++) {
198 dp->queues[i].head = dp->queues[i].tail = 0;
200 hmap_init(&dp->flow_table);
201 list_init(&dp->port_list);
202 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
208 shash_add(&dp_netdevs, name, dp);
215 dpif_netdev_open(const struct dpif_class *class, const char *name,
216 bool create, struct dpif **dpifp)
218 struct dp_netdev *dp;
220 dp = shash_find_data(&dp_netdevs, name);
225 int error = create_dp_netdev(name, class, &dp);
232 if (dp->class != class) {
239 *dpifp = create_dpif_netdev(dp);
244 dp_netdev_purge_queues(struct dp_netdev *dp)
248 for (i = 0; i < N_QUEUES; i++) {
249 struct dp_netdev_queue *q = &dp->queues[i];
251 while (q->tail != q->head) {
252 struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK];
254 ofpbuf_delete(upcall->packet);
261 dp_netdev_free(struct dp_netdev *dp)
263 struct dp_netdev_port *port, *next;
265 dp_netdev_flow_flush(dp);
266 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
267 do_del_port(dp, port->port_no);
269 dp_netdev_purge_queues(dp);
270 hmap_destroy(&dp->flow_table);
276 dpif_netdev_close(struct dpif *dpif)
278 struct dp_netdev *dp = get_dp_netdev(dpif);
279 assert(dp->open_cnt > 0);
280 if (--dp->open_cnt == 0 && dp->destroyed) {
281 shash_find_and_delete(&dp_netdevs, dp->name);
288 dpif_netdev_destroy(struct dpif *dpif)
290 struct dp_netdev *dp = get_dp_netdev(dpif);
291 dp->destroyed = true;
296 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
298 struct dp_netdev *dp = get_dp_netdev(dpif);
299 stats->n_flows = hmap_count(&dp->flow_table);
300 stats->n_hit = dp->n_hit;
301 stats->n_missed = dp->n_missed;
302 stats->n_lost = dp->n_lost;
307 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
310 struct dp_netdev_port *port;
311 struct netdev *netdev;
316 /* XXX reject devices already in some dp_netdev. */
317 if (type[0] == '\0' || !strcmp(type, "system") || !strcmp(type, "dummy")) {
319 } else if (!strcmp(type, "internal")) {
322 VLOG_WARN("%s: unsupported port type %s", devname, type);
326 /* Open and validate network device. */
327 if (dp->class == &dpif_dummy_class) {
329 } else if (internal) {
333 error = netdev_open(devname, type, &netdev);
337 /* XXX reject loopback devices */
338 /* XXX reject non-Ethernet devices */
340 error = netdev_listen(netdev);
342 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
343 devname, strerror(errno));
344 netdev_close(netdev);
348 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
350 netdev_close(netdev);
354 port = xmalloc(sizeof *port);
355 port->port_no = port_no;
356 port->netdev = netdev;
357 port->internal = internal;
359 error = netdev_get_mtu(netdev, &mtu);
364 list_push_back(&dp->port_list, &port->node);
365 dp->ports[port_no] = port;
372 choose_port(struct dpif *dpif, struct netdev *netdev)
374 struct dp_netdev *dp = get_dp_netdev(dpif);
377 if (dpif->dpif_class == &dpif_dummy_class) {
378 /* If the port name contains a number, try to assign that port number.
379 * This can make writing unit tests easier because port numbers are
383 for (p = netdev_get_name(netdev); *p != '\0'; p++) {
384 if (isdigit((unsigned char) *p)) {
385 port_no = strtol(p, NULL, 10);
386 if (port_no > 0 && port_no < MAX_PORTS
387 && !dp->ports[port_no]) {
395 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
396 if (!dp->ports[port_no]) {
405 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
408 struct dp_netdev *dp = get_dp_netdev(dpif);
411 port_no = choose_port(dpif, netdev);
414 return do_add_port(dp, netdev_get_name(netdev),
415 netdev_get_type(netdev), port_no);
421 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
423 struct dp_netdev *dp = get_dp_netdev(dpif);
424 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
428 is_valid_port_number(uint16_t port_no)
430 return port_no < MAX_PORTS;
434 get_port_by_number(struct dp_netdev *dp,
435 uint16_t port_no, struct dp_netdev_port **portp)
437 if (!is_valid_port_number(port_no)) {
441 *portp = dp->ports[port_no];
442 return *portp ? 0 : ENOENT;
447 get_port_by_name(struct dp_netdev *dp,
448 const char *devname, struct dp_netdev_port **portp)
450 struct dp_netdev_port *port;
452 LIST_FOR_EACH (port, node, &dp->port_list) {
453 if (!strcmp(netdev_get_name(port->netdev), devname)) {
462 do_del_port(struct dp_netdev *dp, uint16_t port_no)
464 struct dp_netdev_port *port;
468 error = get_port_by_number(dp, port_no, &port);
473 list_remove(&port->node);
474 dp->ports[port->port_no] = NULL;
477 name = xstrdup(netdev_get_name(port->netdev));
478 netdev_close(port->netdev);
487 answer_port_query(const struct dp_netdev_port *port,
488 struct dpif_port *dpif_port)
490 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
491 dpif_port->type = xstrdup(port->internal ? "internal"
492 : netdev_get_type(port->netdev));
493 dpif_port->port_no = port->port_no;
497 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
498 struct dpif_port *dpif_port)
500 struct dp_netdev *dp = get_dp_netdev(dpif);
501 struct dp_netdev_port *port;
504 error = get_port_by_number(dp, port_no, &port);
506 answer_port_query(port, dpif_port);
512 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
513 struct dpif_port *dpif_port)
515 struct dp_netdev *dp = get_dp_netdev(dpif);
516 struct dp_netdev_port *port;
519 error = get_port_by_name(dp, devname, &port);
521 answer_port_query(port, dpif_port);
527 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
533 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
535 hmap_remove(&dp->flow_table, &flow->node);
541 dp_netdev_flow_flush(struct dp_netdev *dp)
543 struct dp_netdev_flow *flow, *next;
545 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
546 dp_netdev_free_flow(dp, flow);
551 dpif_netdev_flow_flush(struct dpif *dpif)
553 struct dp_netdev *dp = get_dp_netdev(dpif);
554 dp_netdev_flow_flush(dp);
558 struct dp_netdev_port_state {
564 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
566 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
571 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
572 struct dpif_port *dpif_port)
574 struct dp_netdev_port_state *state = state_;
575 struct dp_netdev *dp = get_dp_netdev(dpif);
578 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
579 struct dp_netdev_port *port = dp->ports[port_no];
582 state->name = xstrdup(netdev_get_name(port->netdev));
583 dpif_port->name = state->name;
584 dpif_port->type = (char *) (port->internal ? "internal"
585 : netdev_get_type(port->netdev));
586 dpif_port->port_no = port->port_no;
587 state->port_no = port_no + 1;
595 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
597 struct dp_netdev_port_state *state = state_;
604 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
606 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
607 if (dpif->dp_serial != dpif->dp->serial) {
608 dpif->dp_serial = dpif->dp->serial;
616 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
618 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
619 if (dpif->dp_serial != dpif->dp->serial) {
620 poll_immediate_wake();
624 static struct dp_netdev_flow *
625 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
627 struct dp_netdev_flow *flow;
629 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
630 if (flow_equal(&flow->key, key)) {
638 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
640 stats->n_packets = flow->packet_count;
641 stats->n_bytes = flow->byte_count;
642 stats->used = flow->used;
643 stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl);
647 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
650 if (odp_flow_key_to_flow(key, key_len, flow)) {
651 /* This should not happen: it indicates that odp_flow_key_from_flow()
652 * and odp_flow_key_to_flow() disagree on the acceptable form of a
653 * flow. Log the problem as an error, with enough details to enable
655 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
657 if (!VLOG_DROP_ERR(&rl)) {
661 odp_flow_key_format(key, key_len, &s);
662 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
669 if (flow->in_port < OFPP_MAX
670 ? flow->in_port >= MAX_PORTS
671 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
679 dpif_netdev_flow_get(const struct dpif *dpif,
680 const struct nlattr *nl_key, size_t nl_key_len,
681 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
683 struct dp_netdev *dp = get_dp_netdev(dpif);
684 struct dp_netdev_flow *flow;
688 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
693 flow = dp_netdev_lookup_flow(dp, &key);
699 get_dpif_flow_stats(flow, stats);
702 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
708 set_flow_actions(struct dp_netdev_flow *flow,
709 const struct nlattr *actions, size_t actions_len)
711 flow->actions = xrealloc(flow->actions, actions_len);
712 flow->actions_len = actions_len;
713 memcpy(flow->actions, actions, actions_len);
718 add_flow(struct dpif *dpif, const struct flow *key,
719 const struct nlattr *actions, size_t actions_len)
721 struct dp_netdev *dp = get_dp_netdev(dpif);
722 struct dp_netdev_flow *flow;
725 flow = xzalloc(sizeof *flow);
728 error = set_flow_actions(flow, actions, actions_len);
734 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
739 clear_stats(struct dp_netdev_flow *flow)
742 flow->packet_count = 0;
743 flow->byte_count = 0;
748 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
750 struct dp_netdev *dp = get_dp_netdev(dpif);
751 struct dp_netdev_flow *flow;
755 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
760 flow = dp_netdev_lookup_flow(dp, &key);
762 if (put->flags & DPIF_FP_CREATE) {
763 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
765 memset(put->stats, 0, sizeof *put->stats);
767 return add_flow(dpif, &key, put->actions, put->actions_len);
775 if (put->flags & DPIF_FP_MODIFY) {
776 int error = set_flow_actions(flow, put->actions, put->actions_len);
779 get_dpif_flow_stats(flow, put->stats);
781 if (put->flags & DPIF_FP_ZERO_STATS) {
793 dpif_netdev_flow_del(struct dpif *dpif,
794 const struct nlattr *nl_key, size_t nl_key_len,
795 struct dpif_flow_stats *stats)
797 struct dp_netdev *dp = get_dp_netdev(dpif);
798 struct dp_netdev_flow *flow;
802 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
807 flow = dp_netdev_lookup_flow(dp, &key);
810 get_dpif_flow_stats(flow, stats);
812 dp_netdev_free_flow(dp, flow);
819 struct dp_netdev_flow_state {
822 struct nlattr *actions;
823 struct odputil_keybuf keybuf;
824 struct dpif_flow_stats stats;
828 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
830 struct dp_netdev_flow_state *state;
832 *statep = state = xmalloc(sizeof *state);
835 state->actions = NULL;
840 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
841 const struct nlattr **key, size_t *key_len,
842 const struct nlattr **actions, size_t *actions_len,
843 const struct dpif_flow_stats **stats)
845 struct dp_netdev_flow_state *state = state_;
846 struct dp_netdev *dp = get_dp_netdev(dpif);
847 struct dp_netdev_flow *flow;
848 struct hmap_node *node;
850 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
855 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
860 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
861 odp_flow_key_from_flow(&buf, &flow->key);
868 free(state->actions);
869 state->actions = xmemdup(flow->actions, flow->actions_len);
871 *actions = state->actions;
872 *actions_len = flow->actions_len;
876 get_dpif_flow_stats(flow, &state->stats);
877 *stats = &state->stats;
884 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
886 struct dp_netdev_flow_state *state = state_;
888 free(state->actions);
894 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
896 struct dp_netdev *dp = get_dp_netdev(dpif);
901 if (execute->packet->size < ETH_HEADER_LEN ||
902 execute->packet->size > UINT16_MAX) {
906 /* Make a deep copy of 'packet', because we might modify its data. */
907 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
908 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
909 ofpbuf_put(©, execute->packet->data, execute->packet->size);
911 flow_extract(©, 0, 0, -1, &key);
912 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
915 dp_netdev_execute_actions(dp, ©, &key,
916 execute->actions, execute->actions_len);
919 ofpbuf_uninit(©);
924 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
930 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
931 uint32_t queue_id, uint32_t *priority)
933 *priority = queue_id;
937 static struct dp_netdev_queue *
938 find_nonempty_queue(struct dpif *dpif)
940 struct dp_netdev *dp = get_dp_netdev(dpif);
943 for (i = 0; i < N_QUEUES; i++) {
944 struct dp_netdev_queue *q = &dp->queues[i];
945 if (q->head != q->tail) {
953 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
955 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
957 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
968 dpif_netdev_recv_wait(struct dpif *dpif)
970 if (find_nonempty_queue(dpif)) {
971 poll_immediate_wake();
973 /* No messages ready to be received, and dp_wait() will ensure that we
974 * wake up to queue new messages, so there is nothing to do. */
979 dpif_netdev_recv_purge(struct dpif *dpif)
981 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
982 dp_netdev_purge_queues(dpif_netdev->dp);
986 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
987 const struct ofpbuf *packet)
989 flow->used = time_msec();
990 flow->packet_count++;
991 flow->byte_count += packet->size;
992 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
993 struct tcp_header *th = packet->l4;
994 flow->tcp_ctl |= th->tcp_ctl;
999 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1000 struct ofpbuf *packet)
1002 struct dp_netdev_flow *flow;
1005 if (packet->size < ETH_HEADER_LEN) {
1008 flow_extract(packet, 0, 0, port->port_no, &key);
1009 flow = dp_netdev_lookup_flow(dp, &key);
1011 dp_netdev_flow_used(flow, &key, packet);
1012 dp_netdev_execute_actions(dp, packet, &key,
1013 flow->actions, flow->actions_len);
1017 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1022 dpif_netdev_run(struct dpif *dpif)
1024 struct dp_netdev *dp = get_dp_netdev(dpif);
1025 struct dp_netdev_port *port;
1026 struct ofpbuf packet;
1028 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1030 LIST_FOR_EACH (port, node, &dp->port_list) {
1033 /* Reset packet contents. */
1034 ofpbuf_clear(&packet);
1035 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1037 error = netdev_recv(port->netdev, &packet);
1039 dp_netdev_port_input(dp, port, &packet);
1040 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1041 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1042 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1043 netdev_get_name(port->netdev), strerror(error));
1046 ofpbuf_uninit(&packet);
1050 dpif_netdev_wait(struct dpif *dpif)
1052 struct dp_netdev *dp = get_dp_netdev(dpif);
1053 struct dp_netdev_port *port;
1055 LIST_FOR_EACH (port, node, &dp->port_list) {
1056 netdev_recv_wait(port->netdev);
1061 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1063 struct eth_header *eh = packet->l2;
1065 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1066 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1070 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1073 struct dp_netdev_port *p = dp->ports[out_port];
1075 netdev_send(p->netdev, packet);
1080 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1081 int queue_no, const struct flow *flow, uint64_t arg)
1083 struct dp_netdev_queue *q = &dp->queues[queue_no];
1084 struct dpif_upcall *upcall;
1088 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1093 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1094 odp_flow_key_from_flow(buf, flow);
1095 key_len = buf->size;
1096 ofpbuf_pull(buf, key_len);
1097 ofpbuf_reserve(buf, 2);
1098 ofpbuf_put(buf, packet->data, packet->size);
1100 upcall = xzalloc(sizeof *upcall);
1101 upcall->type = queue_no;
1102 upcall->packet = buf;
1103 upcall->key = buf->base;
1104 upcall->key_len = key_len;
1105 upcall->userdata = arg;
1107 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1113 dp_netdev_sample(struct dp_netdev *dp,
1114 struct ofpbuf *packet, struct flow *key,
1115 const struct nlattr *action)
1117 const struct nlattr *subactions = NULL;
1118 const struct nlattr *a;
1121 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1122 int type = nl_attr_type(a);
1124 switch ((enum ovs_sample_attr) type) {
1125 case OVS_SAMPLE_ATTR_PROBABILITY:
1126 if (random_uint32() >= nl_attr_get_u32(a)) {
1131 case OVS_SAMPLE_ATTR_ACTIONS:
1135 case OVS_SAMPLE_ATTR_UNSPEC:
1136 case __OVS_SAMPLE_ATTR_MAX:
1142 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1143 nl_attr_get_size(subactions));
1147 dp_netdev_action_userspace(struct dp_netdev *dp,
1148 struct ofpbuf *packet, struct flow *key,
1149 const struct nlattr *a)
1151 const struct nlattr *userdata_attr;
1154 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1155 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1156 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1160 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1162 enum ovs_key_attr type = nl_attr_type(a);
1163 const struct ovs_key_ipv4 *ipv4_key;
1164 const struct ovs_key_tcp *tcp_key;
1165 const struct ovs_key_udp *udp_key;
1168 case OVS_KEY_ATTR_TUN_ID:
1169 case OVS_KEY_ATTR_PRIORITY:
1170 case OVS_KEY_ATTR_IPV6:
1171 /* not implemented */
1174 case OVS_KEY_ATTR_ETHERNET:
1175 dp_netdev_set_dl(packet,
1176 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1179 case OVS_KEY_ATTR_IPV4:
1180 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1181 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1182 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1185 case OVS_KEY_ATTR_TCP:
1186 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1187 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1190 case OVS_KEY_ATTR_UDP:
1191 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1192 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1195 case OVS_KEY_ATTR_UNSPEC:
1196 case OVS_KEY_ATTR_ENCAP:
1197 case OVS_KEY_ATTR_ETHERTYPE:
1198 case OVS_KEY_ATTR_IN_PORT:
1199 case OVS_KEY_ATTR_VLAN:
1200 case OVS_KEY_ATTR_ICMP:
1201 case OVS_KEY_ATTR_ICMPV6:
1202 case OVS_KEY_ATTR_ARP:
1203 case OVS_KEY_ATTR_ND:
1204 case __OVS_KEY_ATTR_MAX:
1211 dp_netdev_execute_actions(struct dp_netdev *dp,
1212 struct ofpbuf *packet, struct flow *key,
1213 const struct nlattr *actions,
1216 const struct nlattr *a;
1219 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1220 const struct ovs_action_push_vlan *vlan;
1221 int type = nl_attr_type(a);
1223 switch ((enum ovs_action_attr) type) {
1224 case OVS_ACTION_ATTR_OUTPUT:
1225 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1228 case OVS_ACTION_ATTR_USERSPACE:
1229 dp_netdev_action_userspace(dp, packet, key, a);
1232 case OVS_ACTION_ATTR_PUSH_VLAN:
1233 vlan = nl_attr_get(a);
1234 eth_push_vlan(packet, vlan->vlan_tci);
1237 case OVS_ACTION_ATTR_POP_VLAN:
1238 eth_pop_vlan(packet);
1241 case OVS_ACTION_ATTR_SET:
1242 execute_set_action(packet, nl_attr_get(a));
1245 case OVS_ACTION_ATTR_SAMPLE:
1246 dp_netdev_sample(dp, packet, key, a);
1249 case OVS_ACTION_ATTR_UNSPEC:
1250 case __OVS_ACTION_ATTR_MAX:
1256 const struct dpif_class dpif_netdev_class = {
1258 NULL, /* enumerate */
1261 dpif_netdev_destroy,
1264 dpif_netdev_get_stats,
1265 dpif_netdev_port_add,
1266 dpif_netdev_port_del,
1267 dpif_netdev_port_query_by_number,
1268 dpif_netdev_port_query_by_name,
1269 dpif_netdev_get_max_ports,
1270 NULL, /* port_get_pid */
1271 dpif_netdev_port_dump_start,
1272 dpif_netdev_port_dump_next,
1273 dpif_netdev_port_dump_done,
1274 dpif_netdev_port_poll,
1275 dpif_netdev_port_poll_wait,
1276 dpif_netdev_flow_get,
1277 dpif_netdev_flow_put,
1278 dpif_netdev_flow_del,
1279 dpif_netdev_flow_flush,
1280 dpif_netdev_flow_dump_start,
1281 dpif_netdev_flow_dump_next,
1282 dpif_netdev_flow_dump_done,
1283 dpif_netdev_execute,
1285 dpif_netdev_recv_set,
1286 dpif_netdev_queue_to_priority,
1288 dpif_netdev_recv_wait,
1289 dpif_netdev_recv_purge,
1293 dpif_dummy_register(void)
1295 if (!dpif_dummy_class.type) {
1296 dpif_dummy_class = dpif_netdev_class;
1297 dpif_dummy_class.type = "dummy";
1298 dp_register_provider(&dpif_dummy_class);