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 create_dpif_netdev(struct dp_netdev *dp)
195 uint16_t netflow_id = hash_string(dp->name, 0);
196 struct dpif_netdev *dpif;
200 dpif = xmalloc(sizeof *dpif);
201 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
203 dpif->dp_serial = dp->serial;
209 choose_port(struct dp_netdev *dp, const char *name)
213 if (dp->class != &dpif_netdev_class) {
217 /* If the port name begins with "br", start the number search at
218 * 100 to make writing tests easier. */
219 if (!strncmp(name, "br", 2)) {
223 /* If the port name contains a number, try to assign that port number.
224 * This can make writing unit tests easier because port numbers are
226 for (p = name; *p != '\0'; p++) {
227 if (isdigit((unsigned char) *p)) {
228 port_no = start_no + strtol(p, NULL, 10);
229 if (port_no > 0 && port_no < MAX_PORTS
230 && !dp->ports[port_no]) {
238 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
239 if (!dp->ports[port_no]) {
248 create_dp_netdev(const char *name, const struct dpif_class *class,
249 struct dp_netdev **dpp)
251 struct dp_netdev *dp;
255 dp = xzalloc(sizeof *dp);
257 dp->name = xstrdup(name);
259 for (i = 0; i < N_QUEUES; i++) {
260 dp->queues[i].head = dp->queues[i].tail = 0;
262 hmap_init(&dp->flow_table);
263 list_init(&dp->port_list);
265 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
271 shash_add(&dp_netdevs, name, dp);
278 dpif_netdev_open(const struct dpif_class *class, const char *name,
279 bool create, struct dpif **dpifp)
281 struct dp_netdev *dp;
283 dp = shash_find_data(&dp_netdevs, name);
288 int error = create_dp_netdev(name, class, &dp);
295 if (dp->class != class) {
302 *dpifp = create_dpif_netdev(dp);
307 dp_netdev_purge_queues(struct dp_netdev *dp)
311 for (i = 0; i < N_QUEUES; i++) {
312 struct dp_netdev_queue *q = &dp->queues[i];
314 while (q->tail != q->head) {
315 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
316 ofpbuf_uninit(&u->buf);
322 dp_netdev_free(struct dp_netdev *dp)
324 struct dp_netdev_port *port, *next;
326 dp_netdev_flow_flush(dp);
327 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
328 do_del_port(dp, port->port_no);
330 dp_netdev_purge_queues(dp);
331 hmap_destroy(&dp->flow_table);
337 dpif_netdev_close(struct dpif *dpif)
339 struct dp_netdev *dp = get_dp_netdev(dpif);
340 assert(dp->open_cnt > 0);
341 if (--dp->open_cnt == 0 && dp->destroyed) {
342 shash_find_and_delete(&dp_netdevs, dp->name);
349 dpif_netdev_destroy(struct dpif *dpif)
351 struct dp_netdev *dp = get_dp_netdev(dpif);
352 dp->destroyed = true;
357 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
359 struct dp_netdev *dp = get_dp_netdev(dpif);
360 stats->n_flows = hmap_count(&dp->flow_table);
361 stats->n_hit = dp->n_hit;
362 stats->n_missed = dp->n_missed;
363 stats->n_lost = dp->n_lost;
368 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
371 struct dp_netdev_port *port;
372 struct netdev *netdev;
373 const char *open_type;
377 /* XXX reject devices already in some dp_netdev. */
379 /* Open and validate network device. */
380 open_type = dpif_netdev_port_open_type(dp->class, type);
381 error = netdev_open(devname, open_type, &netdev);
385 /* XXX reject loopback devices */
386 /* XXX reject non-Ethernet devices */
388 error = netdev_listen(netdev);
390 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
391 devname, strerror(errno));
392 netdev_close(netdev);
396 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
398 netdev_close(netdev);
402 port = xmalloc(sizeof *port);
403 port->port_no = port_no;
404 port->netdev = netdev;
405 port->type = xstrdup(type);
407 error = netdev_get_mtu(netdev, &mtu);
412 list_push_back(&dp->port_list, &port->node);
413 dp->ports[port_no] = port;
420 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
423 struct dp_netdev *dp = get_dp_netdev(dpif);
426 if (*port_nop != UINT32_MAX) {
427 if (*port_nop >= MAX_PORTS) {
429 } else if (dp->ports[*port_nop]) {
434 port_no = choose_port(dp, netdev_get_name(netdev));
438 return do_add_port(dp, netdev_get_name(netdev),
439 netdev_get_type(netdev), port_no);
445 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
447 struct dp_netdev *dp = get_dp_netdev(dpif);
448 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
452 is_valid_port_number(uint32_t port_no)
454 return port_no < MAX_PORTS;
458 get_port_by_number(struct dp_netdev *dp,
459 uint32_t port_no, struct dp_netdev_port **portp)
461 if (!is_valid_port_number(port_no)) {
465 *portp = dp->ports[port_no];
466 return *portp ? 0 : ENOENT;
471 get_port_by_name(struct dp_netdev *dp,
472 const char *devname, struct dp_netdev_port **portp)
474 struct dp_netdev_port *port;
476 LIST_FOR_EACH (port, node, &dp->port_list) {
477 if (!strcmp(netdev_get_name(port->netdev), devname)) {
486 do_del_port(struct dp_netdev *dp, uint32_t port_no)
488 struct dp_netdev_port *port;
492 error = get_port_by_number(dp, port_no, &port);
497 list_remove(&port->node);
498 dp->ports[port->port_no] = NULL;
501 name = xstrdup(netdev_get_name(port->netdev));
502 netdev_close(port->netdev);
512 answer_port_query(const struct dp_netdev_port *port,
513 struct dpif_port *dpif_port)
515 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
516 dpif_port->type = xstrdup(port->type);
517 dpif_port->port_no = port->port_no;
521 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
522 struct dpif_port *dpif_port)
524 struct dp_netdev *dp = get_dp_netdev(dpif);
525 struct dp_netdev_port *port;
528 error = get_port_by_number(dp, port_no, &port);
529 if (!error && dpif_port) {
530 answer_port_query(port, dpif_port);
536 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
537 struct dpif_port *dpif_port)
539 struct dp_netdev *dp = get_dp_netdev(dpif);
540 struct dp_netdev_port *port;
543 error = get_port_by_name(dp, devname, &port);
544 if (!error && dpif_port) {
545 answer_port_query(port, dpif_port);
551 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
557 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
559 hmap_remove(&dp->flow_table, &flow->node);
565 dp_netdev_flow_flush(struct dp_netdev *dp)
567 struct dp_netdev_flow *flow, *next;
569 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
570 dp_netdev_free_flow(dp, flow);
575 dpif_netdev_flow_flush(struct dpif *dpif)
577 struct dp_netdev *dp = get_dp_netdev(dpif);
578 dp_netdev_flow_flush(dp);
582 struct dp_netdev_port_state {
588 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
590 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
595 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
596 struct dpif_port *dpif_port)
598 struct dp_netdev_port_state *state = state_;
599 struct dp_netdev *dp = get_dp_netdev(dpif);
602 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
603 struct dp_netdev_port *port = dp->ports[port_no];
606 state->name = xstrdup(netdev_get_name(port->netdev));
607 dpif_port->name = state->name;
608 dpif_port->type = port->type;
609 dpif_port->port_no = port->port_no;
610 state->port_no = port_no + 1;
618 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
620 struct dp_netdev_port_state *state = state_;
627 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
629 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
630 if (dpif->dp_serial != dpif->dp->serial) {
631 dpif->dp_serial = dpif->dp->serial;
639 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
641 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
642 if (dpif->dp_serial != dpif->dp->serial) {
643 poll_immediate_wake();
647 static struct dp_netdev_flow *
648 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
650 struct dp_netdev_flow *flow;
652 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
653 if (flow_equal(&flow->key, key)) {
661 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
663 stats->n_packets = flow->packet_count;
664 stats->n_bytes = flow->byte_count;
665 stats->used = flow->used;
666 stats->tcp_flags = flow->tcp_flags;
670 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
673 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
674 /* This should not happen: it indicates that odp_flow_key_from_flow()
675 * and odp_flow_key_to_flow() disagree on the acceptable form of a
676 * flow. Log the problem as an error, with enough details to enable
678 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
680 if (!VLOG_DROP_ERR(&rl)) {
684 odp_flow_key_format(key, key_len, &s);
685 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
692 if (flow->in_port < OFPP_MAX
693 ? flow->in_port >= MAX_PORTS
694 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
702 dpif_netdev_flow_get(const struct dpif *dpif,
703 const struct nlattr *nl_key, size_t nl_key_len,
704 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
706 struct dp_netdev *dp = get_dp_netdev(dpif);
707 struct dp_netdev_flow *flow;
711 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
716 flow = dp_netdev_lookup_flow(dp, &key);
722 get_dpif_flow_stats(flow, stats);
725 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
731 set_flow_actions(struct dp_netdev_flow *flow,
732 const struct nlattr *actions, size_t actions_len)
734 flow->actions = xrealloc(flow->actions, actions_len);
735 flow->actions_len = actions_len;
736 memcpy(flow->actions, actions, actions_len);
741 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
742 const struct nlattr *actions, size_t actions_len)
744 struct dp_netdev_flow *flow;
747 flow = xzalloc(sizeof *flow);
750 error = set_flow_actions(flow, actions, actions_len);
756 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
761 clear_stats(struct dp_netdev_flow *flow)
764 flow->packet_count = 0;
765 flow->byte_count = 0;
770 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
772 struct dp_netdev *dp = get_dp_netdev(dpif);
773 struct dp_netdev_flow *flow;
777 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
782 flow = dp_netdev_lookup_flow(dp, &key);
784 if (put->flags & DPIF_FP_CREATE) {
785 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
787 memset(put->stats, 0, sizeof *put->stats);
789 return dp_netdev_flow_add(dp, &key, put->actions,
798 if (put->flags & DPIF_FP_MODIFY) {
799 int error = set_flow_actions(flow, put->actions, put->actions_len);
802 get_dpif_flow_stats(flow, put->stats);
804 if (put->flags & DPIF_FP_ZERO_STATS) {
816 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
818 struct dp_netdev *dp = get_dp_netdev(dpif);
819 struct dp_netdev_flow *flow;
823 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
828 flow = dp_netdev_lookup_flow(dp, &key);
831 get_dpif_flow_stats(flow, del->stats);
833 dp_netdev_free_flow(dp, flow);
840 struct dp_netdev_flow_state {
843 struct nlattr *actions;
844 struct odputil_keybuf keybuf;
845 struct dpif_flow_stats stats;
849 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
851 struct dp_netdev_flow_state *state;
853 *statep = state = xmalloc(sizeof *state);
856 state->actions = NULL;
861 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
862 const struct nlattr **key, size_t *key_len,
863 const struct nlattr **actions, size_t *actions_len,
864 const struct dpif_flow_stats **stats)
866 struct dp_netdev_flow_state *state = state_;
867 struct dp_netdev *dp = get_dp_netdev(dpif);
868 struct dp_netdev_flow *flow;
869 struct hmap_node *node;
871 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
876 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
881 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
882 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
889 free(state->actions);
890 state->actions = xmemdup(flow->actions, flow->actions_len);
892 *actions = state->actions;
893 *actions_len = flow->actions_len;
897 get_dpif_flow_stats(flow, &state->stats);
898 *stats = &state->stats;
905 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
907 struct dp_netdev_flow_state *state = state_;
909 free(state->actions);
915 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
917 struct dp_netdev *dp = get_dp_netdev(dpif);
922 if (execute->packet->size < ETH_HEADER_LEN ||
923 execute->packet->size > UINT16_MAX) {
927 /* Make a deep copy of 'packet', because we might modify its data. */
928 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
929 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
930 ofpbuf_put(©, execute->packet->data, execute->packet->size);
932 flow_extract(©, 0, 0, NULL, -1, &key);
933 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
936 dp_netdev_execute_actions(dp, ©, &key,
937 execute->actions, execute->actions_len);
940 ofpbuf_uninit(©);
945 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
951 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
952 uint32_t queue_id, uint32_t *priority)
954 *priority = queue_id;
958 static struct dp_netdev_queue *
959 find_nonempty_queue(struct dpif *dpif)
961 struct dp_netdev *dp = get_dp_netdev(dpif);
964 for (i = 0; i < N_QUEUES; i++) {
965 struct dp_netdev_queue *q = &dp->queues[i];
966 if (q->head != q->tail) {
974 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
977 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
979 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
982 upcall->packet = buf;
994 dpif_netdev_recv_wait(struct dpif *dpif)
996 if (find_nonempty_queue(dpif)) {
997 poll_immediate_wake();
999 /* No messages ready to be received, and dp_wait() will ensure that we
1000 * wake up to queue new messages, so there is nothing to do. */
1005 dpif_netdev_recv_purge(struct dpif *dpif)
1007 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1008 dp_netdev_purge_queues(dpif_netdev->dp);
1012 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1014 flow->used = time_msec();
1015 flow->packet_count++;
1016 flow->byte_count += packet->size;
1017 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1021 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1022 struct ofpbuf *packet)
1024 struct dp_netdev_flow *flow;
1027 if (packet->size < ETH_HEADER_LEN) {
1030 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1031 flow = dp_netdev_lookup_flow(dp, &key);
1033 dp_netdev_flow_used(flow, packet);
1034 dp_netdev_execute_actions(dp, packet, &key,
1035 flow->actions, flow->actions_len);
1039 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1044 dpif_netdev_run(struct dpif *dpif)
1046 struct dp_netdev *dp = get_dp_netdev(dpif);
1047 struct dp_netdev_port *port;
1048 struct ofpbuf packet;
1050 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1052 LIST_FOR_EACH (port, node, &dp->port_list) {
1055 /* Reset packet contents. */
1056 ofpbuf_clear(&packet);
1057 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1059 error = netdev_recv(port->netdev, &packet);
1061 dp_netdev_port_input(dp, port, &packet);
1062 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1064 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1065 netdev_get_name(port->netdev), strerror(error));
1068 ofpbuf_uninit(&packet);
1072 dpif_netdev_wait(struct dpif *dpif)
1074 struct dp_netdev *dp = get_dp_netdev(dpif);
1075 struct dp_netdev_port *port;
1077 LIST_FOR_EACH (port, node, &dp->port_list) {
1078 netdev_recv_wait(port->netdev);
1083 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1085 struct eth_header *eh = packet->l2;
1087 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1088 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1092 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1095 struct dp_netdev_port *p = dp->ports[out_port];
1097 netdev_send(p->netdev, packet);
1102 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1103 int queue_no, const struct flow *flow, uint64_t arg)
1105 struct dp_netdev_queue *q = &dp->queues[queue_no];
1106 struct dp_netdev_upcall *u;
1107 struct dpif_upcall *upcall;
1111 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1116 u = &q->upcalls[q->head++ & QUEUE_MASK];
1119 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1120 odp_flow_key_from_flow(buf, flow, flow->in_port);
1121 key_len = buf->size;
1122 ofpbuf_pull(buf, key_len);
1123 ofpbuf_reserve(buf, 2);
1124 ofpbuf_put(buf, packet->data, packet->size);
1126 upcall = &u->upcall;
1127 upcall->type = queue_no;
1128 upcall->packet = buf;
1129 upcall->key = buf->base;
1130 upcall->key_len = key_len;
1131 upcall->userdata = arg;
1137 dp_netdev_sample(struct dp_netdev *dp,
1138 struct ofpbuf *packet, struct flow *key,
1139 const struct nlattr *action)
1141 const struct nlattr *subactions = NULL;
1142 const struct nlattr *a;
1145 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1146 int type = nl_attr_type(a);
1148 switch ((enum ovs_sample_attr) type) {
1149 case OVS_SAMPLE_ATTR_PROBABILITY:
1150 if (random_uint32() >= nl_attr_get_u32(a)) {
1155 case OVS_SAMPLE_ATTR_ACTIONS:
1159 case OVS_SAMPLE_ATTR_UNSPEC:
1160 case __OVS_SAMPLE_ATTR_MAX:
1166 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1167 nl_attr_get_size(subactions));
1171 dp_netdev_action_userspace(struct dp_netdev *dp,
1172 struct ofpbuf *packet, struct flow *key,
1173 const struct nlattr *a)
1175 const struct nlattr *userdata_attr;
1178 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1179 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1180 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1184 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1186 enum ovs_key_attr type = nl_attr_type(a);
1187 const struct ovs_key_ipv4 *ipv4_key;
1188 const struct ovs_key_ipv6 *ipv6_key;
1189 const struct ovs_key_tcp *tcp_key;
1190 const struct ovs_key_udp *udp_key;
1193 case OVS_KEY_ATTR_TUN_ID:
1194 case OVS_KEY_ATTR_PRIORITY:
1195 case OVS_KEY_ATTR_SKB_MARK:
1196 case OVS_KEY_ATTR_IPV4_TUNNEL:
1197 /* not implemented */
1200 case OVS_KEY_ATTR_ETHERNET:
1201 dp_netdev_set_dl(packet,
1202 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1205 case OVS_KEY_ATTR_IPV4:
1206 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1207 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1208 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1211 case OVS_KEY_ATTR_IPV6:
1212 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1213 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1214 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1215 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1218 case OVS_KEY_ATTR_TCP:
1219 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1220 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1223 case OVS_KEY_ATTR_UDP:
1224 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1225 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1228 case OVS_KEY_ATTR_UNSPEC:
1229 case OVS_KEY_ATTR_ENCAP:
1230 case OVS_KEY_ATTR_ETHERTYPE:
1231 case OVS_KEY_ATTR_IN_PORT:
1232 case OVS_KEY_ATTR_VLAN:
1233 case OVS_KEY_ATTR_ICMP:
1234 case OVS_KEY_ATTR_ICMPV6:
1235 case OVS_KEY_ATTR_ARP:
1236 case OVS_KEY_ATTR_ND:
1237 case __OVS_KEY_ATTR_MAX:
1244 dp_netdev_execute_actions(struct dp_netdev *dp,
1245 struct ofpbuf *packet, struct flow *key,
1246 const struct nlattr *actions,
1249 const struct nlattr *a;
1252 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1253 const struct ovs_action_push_vlan *vlan;
1254 int type = nl_attr_type(a);
1256 switch ((enum ovs_action_attr) type) {
1257 case OVS_ACTION_ATTR_OUTPUT:
1258 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1261 case OVS_ACTION_ATTR_USERSPACE:
1262 dp_netdev_action_userspace(dp, packet, key, a);
1265 case OVS_ACTION_ATTR_PUSH_VLAN:
1266 vlan = nl_attr_get(a);
1267 eth_push_vlan(packet, vlan->vlan_tci);
1270 case OVS_ACTION_ATTR_POP_VLAN:
1271 eth_pop_vlan(packet);
1274 case OVS_ACTION_ATTR_SET:
1275 execute_set_action(packet, nl_attr_get(a));
1278 case OVS_ACTION_ATTR_SAMPLE:
1279 dp_netdev_sample(dp, packet, key, a);
1282 case OVS_ACTION_ATTR_UNSPEC:
1283 case __OVS_ACTION_ATTR_MAX:
1289 const struct dpif_class dpif_netdev_class = {
1291 dpif_netdev_enumerate,
1292 dpif_netdev_port_open_type,
1295 dpif_netdev_destroy,
1298 dpif_netdev_get_stats,
1299 dpif_netdev_port_add,
1300 dpif_netdev_port_del,
1301 dpif_netdev_port_query_by_number,
1302 dpif_netdev_port_query_by_name,
1303 dpif_netdev_get_max_ports,
1304 NULL, /* port_get_pid */
1305 dpif_netdev_port_dump_start,
1306 dpif_netdev_port_dump_next,
1307 dpif_netdev_port_dump_done,
1308 dpif_netdev_port_poll,
1309 dpif_netdev_port_poll_wait,
1310 dpif_netdev_flow_get,
1311 dpif_netdev_flow_put,
1312 dpif_netdev_flow_del,
1313 dpif_netdev_flow_flush,
1314 dpif_netdev_flow_dump_start,
1315 dpif_netdev_flow_dump_next,
1316 dpif_netdev_flow_dump_done,
1317 dpif_netdev_execute,
1319 dpif_netdev_recv_set,
1320 dpif_netdev_queue_to_priority,
1322 dpif_netdev_recv_wait,
1323 dpif_netdev_recv_purge,
1327 dpif_dummy_register__(const char *type)
1329 struct dpif_class *class;
1331 class = xmalloc(sizeof *class);
1332 *class = dpif_netdev_class;
1333 class->type = xstrdup(type);
1334 dp_register_provider(class);
1338 dpif_dummy_register(bool override)
1345 dp_enumerate_types(&types);
1346 SSET_FOR_EACH (type, &types) {
1347 if (!dp_unregister_provider(type)) {
1348 dpif_dummy_register__(type);
1351 sset_destroy(&types);
1354 dpif_dummy_register__("dummy");