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"
45 #include "odp-execute.h"
47 #include "ofp-print.h"
50 #include "poll-loop.h"
58 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
60 /* Configuration parameters. */
61 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
62 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
64 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
65 * headers to be aligned on a 4-byte boundary. */
66 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
69 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
70 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
71 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
72 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
74 struct dp_netdev_upcall {
75 struct dpif_upcall upcall; /* Queued upcall information. */
76 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
79 struct dp_netdev_queue {
80 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
81 unsigned int head, tail;
84 /* Datapath based on the network device interface from netdev.h. */
86 const struct dpif_class *class;
91 struct dp_netdev_queue queues[N_QUEUES];
92 struct hmap flow_table; /* Flow table. */
95 long long int n_hit; /* Number of flow table matches. */
96 long long int n_missed; /* Number of flow table misses. */
97 long long int n_lost; /* Number of misses not passed to client. */
100 struct dp_netdev_port *ports[MAX_PORTS];
101 struct list port_list;
105 /* A port in a netdev-based datapath. */
106 struct dp_netdev_port {
107 int port_no; /* Index into dp_netdev's 'ports'. */
108 struct list node; /* Element in dp_netdev's 'port_list'. */
109 struct netdev *netdev;
110 struct netdev_saved_flags *sf;
111 struct netdev_rx *rx;
112 char *type; /* Port type as requested by user. */
115 /* A flow in dp_netdev's 'flow_table'. */
116 struct dp_netdev_flow {
117 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
121 long long int used; /* Last used time, in monotonic msecs. */
122 long long int packet_count; /* Number of packets matched. */
123 long long int byte_count; /* Number of bytes matched. */
124 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
127 struct nlattr *actions;
131 /* Interface to netdev-based datapath. */
134 struct dp_netdev *dp;
135 unsigned int dp_serial;
138 /* All netdev-based datapaths. */
139 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
141 /* Maximum port MTU seen so far. */
142 static int max_mtu = ETH_PAYLOAD_MAX;
144 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
145 struct dp_netdev_port **portp);
146 static int get_port_by_name(struct dp_netdev *, const char *devname,
147 struct dp_netdev_port **portp);
148 static void dp_netdev_free(struct dp_netdev *);
149 static void dp_netdev_flow_flush(struct dp_netdev *);
150 static int do_add_port(struct dp_netdev *, const char *devname,
151 const char *type, uint32_t port_no);
152 static int do_del_port(struct dp_netdev *, uint32_t port_no);
153 static int dpif_netdev_open(const struct dpif_class *, const char *name,
154 bool create, struct dpif **);
155 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
156 int queue_no, const struct flow *,
157 const struct nlattr *userdata);
158 static void dp_netdev_execute_actions(struct dp_netdev *,
159 struct ofpbuf *, struct flow *,
160 const struct nlattr *actions,
163 static struct dpif_netdev *
164 dpif_netdev_cast(const struct dpif *dpif)
166 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
167 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
170 static struct dp_netdev *
171 get_dp_netdev(const struct dpif *dpif)
173 return dpif_netdev_cast(dpif)->dp;
177 dpif_netdev_enumerate(struct sset *all_dps)
179 struct shash_node *node;
181 SHASH_FOR_EACH(node, &dp_netdevs) {
182 sset_add(all_dps, node->name);
188 dpif_netdev_class_is_dummy(const struct dpif_class *class)
190 return class != &dpif_netdev_class;
194 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
196 return strcmp(type, "internal") ? type
197 : dpif_netdev_class_is_dummy(class) ? "dummy"
202 create_dpif_netdev(struct dp_netdev *dp)
204 uint16_t netflow_id = hash_string(dp->name, 0);
205 struct dpif_netdev *dpif;
209 dpif = xmalloc(sizeof *dpif);
210 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
212 dpif->dp_serial = dp->serial;
218 choose_port(struct dp_netdev *dp, const char *name)
222 if (dp->class != &dpif_netdev_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);
301 ovs_assert(dp != NULL);
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 ovs_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 netdev_saved_flags *sf;
381 struct dp_netdev_port *port;
382 struct netdev *netdev;
383 struct netdev_rx *rx;
384 const char *open_type;
388 /* XXX reject devices already in some dp_netdev. */
390 /* Open and validate network device. */
391 open_type = dpif_netdev_port_open_type(dp->class, type);
392 error = netdev_open(devname, open_type, &netdev);
396 /* XXX reject loopback devices */
397 /* XXX reject non-Ethernet devices */
399 error = netdev_rx_open(netdev, &rx);
401 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
402 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
403 devname, strerror(errno));
404 netdev_close(netdev);
408 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
411 netdev_close(netdev);
415 port = xmalloc(sizeof *port);
416 port->port_no = port_no;
417 port->netdev = netdev;
420 port->type = xstrdup(type);
422 error = netdev_get_mtu(netdev, &mtu);
423 if (!error && mtu > max_mtu) {
427 list_push_back(&dp->port_list, &port->node);
428 dp->ports[port_no] = port;
435 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
438 struct dp_netdev *dp = get_dp_netdev(dpif);
441 if (*port_nop != UINT32_MAX) {
442 if (*port_nop >= MAX_PORTS) {
444 } else if (dp->ports[*port_nop]) {
449 port_no = choose_port(dp, netdev_vport_get_dpif_port(netdev));
453 return do_add_port(dp, netdev_vport_get_dpif_port(netdev),
454 netdev_get_type(netdev), port_no);
460 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
462 struct dp_netdev *dp = get_dp_netdev(dpif);
463 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
467 is_valid_port_number(uint32_t port_no)
469 return port_no < MAX_PORTS;
473 get_port_by_number(struct dp_netdev *dp,
474 uint32_t port_no, struct dp_netdev_port **portp)
476 if (!is_valid_port_number(port_no)) {
480 *portp = dp->ports[port_no];
481 return *portp ? 0 : ENOENT;
486 get_port_by_name(struct dp_netdev *dp,
487 const char *devname, struct dp_netdev_port **portp)
489 struct dp_netdev_port *port;
491 LIST_FOR_EACH (port, node, &dp->port_list) {
492 if (!strcmp(netdev_vport_get_dpif_port(port->netdev), devname)) {
501 do_del_port(struct dp_netdev *dp, uint32_t port_no)
503 struct dp_netdev_port *port;
506 error = get_port_by_number(dp, port_no, &port);
511 list_remove(&port->node);
512 dp->ports[port->port_no] = NULL;
515 netdev_close(port->netdev);
516 netdev_restore_flags(port->sf);
517 netdev_rx_close(port->rx);
525 answer_port_query(const struct dp_netdev_port *port,
526 struct dpif_port *dpif_port)
528 dpif_port->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
529 dpif_port->type = xstrdup(port->type);
530 dpif_port->port_no = port->port_no;
534 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
535 struct dpif_port *dpif_port)
537 struct dp_netdev *dp = get_dp_netdev(dpif);
538 struct dp_netdev_port *port;
541 error = get_port_by_number(dp, port_no, &port);
542 if (!error && dpif_port) {
543 answer_port_query(port, dpif_port);
549 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
550 struct dpif_port *dpif_port)
552 struct dp_netdev *dp = get_dp_netdev(dpif);
553 struct dp_netdev_port *port;
556 error = get_port_by_name(dp, devname, &port);
557 if (!error && dpif_port) {
558 answer_port_query(port, dpif_port);
564 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
570 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
572 hmap_remove(&dp->flow_table, &flow->node);
578 dp_netdev_flow_flush(struct dp_netdev *dp)
580 struct dp_netdev_flow *flow, *next;
582 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
583 dp_netdev_free_flow(dp, flow);
588 dpif_netdev_flow_flush(struct dpif *dpif)
590 struct dp_netdev *dp = get_dp_netdev(dpif);
591 dp_netdev_flow_flush(dp);
595 struct dp_netdev_port_state {
601 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
603 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
608 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
609 struct dpif_port *dpif_port)
611 struct dp_netdev_port_state *state = state_;
612 struct dp_netdev *dp = get_dp_netdev(dpif);
615 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
616 struct dp_netdev_port *port = dp->ports[port_no];
619 state->name = xstrdup(netdev_vport_get_dpif_port(port->netdev));
620 dpif_port->name = state->name;
621 dpif_port->type = port->type;
622 dpif_port->port_no = port->port_no;
623 state->port_no = port_no + 1;
631 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
633 struct dp_netdev_port_state *state = state_;
640 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
642 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
643 if (dpif->dp_serial != dpif->dp->serial) {
644 dpif->dp_serial = dpif->dp->serial;
652 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
654 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
655 if (dpif->dp_serial != dpif->dp->serial) {
656 poll_immediate_wake();
660 static struct dp_netdev_flow *
661 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
663 struct dp_netdev_flow *flow;
665 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
666 if (flow_equal(&flow->key, key)) {
674 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
676 stats->n_packets = flow->packet_count;
677 stats->n_bytes = flow->byte_count;
678 stats->used = flow->used;
679 stats->tcp_flags = flow->tcp_flags;
683 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
686 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
687 /* This should not happen: it indicates that odp_flow_key_from_flow()
688 * and odp_flow_key_to_flow() disagree on the acceptable form of a
689 * flow. Log the problem as an error, with enough details to enable
691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
693 if (!VLOG_DROP_ERR(&rl)) {
697 odp_flow_key_format(key, key_len, &s);
698 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
705 if (flow->in_port < OFPP_MAX
706 ? flow->in_port >= MAX_PORTS
707 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
715 dpif_netdev_flow_get(const struct dpif *dpif,
716 const struct nlattr *nl_key, size_t nl_key_len,
717 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
719 struct dp_netdev *dp = get_dp_netdev(dpif);
720 struct dp_netdev_flow *flow;
724 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
729 flow = dp_netdev_lookup_flow(dp, &key);
735 get_dpif_flow_stats(flow, stats);
738 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
744 set_flow_actions(struct dp_netdev_flow *flow,
745 const struct nlattr *actions, size_t actions_len)
747 flow->actions = xrealloc(flow->actions, actions_len);
748 flow->actions_len = actions_len;
749 memcpy(flow->actions, actions, actions_len);
754 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
755 const struct nlattr *actions, size_t actions_len)
757 struct dp_netdev_flow *flow;
760 flow = xzalloc(sizeof *flow);
763 error = set_flow_actions(flow, actions, actions_len);
769 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
774 clear_stats(struct dp_netdev_flow *flow)
777 flow->packet_count = 0;
778 flow->byte_count = 0;
783 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
785 struct dp_netdev *dp = get_dp_netdev(dpif);
786 struct dp_netdev_flow *flow;
790 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
795 flow = dp_netdev_lookup_flow(dp, &key);
797 if (put->flags & DPIF_FP_CREATE) {
798 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
800 memset(put->stats, 0, sizeof *put->stats);
802 return dp_netdev_flow_add(dp, &key, put->actions,
811 if (put->flags & DPIF_FP_MODIFY) {
812 int error = set_flow_actions(flow, put->actions, put->actions_len);
815 get_dpif_flow_stats(flow, put->stats);
817 if (put->flags & DPIF_FP_ZERO_STATS) {
829 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
831 struct dp_netdev *dp = get_dp_netdev(dpif);
832 struct dp_netdev_flow *flow;
836 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
841 flow = dp_netdev_lookup_flow(dp, &key);
844 get_dpif_flow_stats(flow, del->stats);
846 dp_netdev_free_flow(dp, flow);
853 struct dp_netdev_flow_state {
856 struct nlattr *actions;
857 struct odputil_keybuf keybuf;
858 struct dpif_flow_stats stats;
862 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
864 struct dp_netdev_flow_state *state;
866 *statep = state = xmalloc(sizeof *state);
869 state->actions = NULL;
874 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
875 const struct nlattr **key, size_t *key_len,
876 const struct nlattr **actions, size_t *actions_len,
877 const struct dpif_flow_stats **stats)
879 struct dp_netdev_flow_state *state = state_;
880 struct dp_netdev *dp = get_dp_netdev(dpif);
881 struct dp_netdev_flow *flow;
882 struct hmap_node *node;
884 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
889 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
894 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
895 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
902 free(state->actions);
903 state->actions = xmemdup(flow->actions, flow->actions_len);
905 *actions = state->actions;
906 *actions_len = flow->actions_len;
910 get_dpif_flow_stats(flow, &state->stats);
911 *stats = &state->stats;
918 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
920 struct dp_netdev_flow_state *state = state_;
922 free(state->actions);
928 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
930 struct dp_netdev *dp = get_dp_netdev(dpif);
935 if (execute->packet->size < ETH_HEADER_LEN ||
936 execute->packet->size > UINT16_MAX) {
940 /* Make a deep copy of 'packet', because we might modify its data. */
941 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
942 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
943 ofpbuf_put(©, execute->packet->data, execute->packet->size);
945 flow_extract(©, 0, 0, NULL, -1, &key);
946 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
949 dp_netdev_execute_actions(dp, ©, &key,
950 execute->actions, execute->actions_len);
953 ofpbuf_uninit(©);
958 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
964 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
965 uint32_t queue_id, uint32_t *priority)
967 *priority = queue_id;
971 static struct dp_netdev_queue *
972 find_nonempty_queue(struct dpif *dpif)
974 struct dp_netdev *dp = get_dp_netdev(dpif);
977 for (i = 0; i < N_QUEUES; i++) {
978 struct dp_netdev_queue *q = &dp->queues[i];
979 if (q->head != q->tail) {
987 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
990 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
992 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
995 upcall->packet = buf;
1007 dpif_netdev_recv_wait(struct dpif *dpif)
1009 if (find_nonempty_queue(dpif)) {
1010 poll_immediate_wake();
1012 /* No messages ready to be received, and dp_wait() will ensure that we
1013 * wake up to queue new messages, so there is nothing to do. */
1018 dpif_netdev_recv_purge(struct dpif *dpif)
1020 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1021 dp_netdev_purge_queues(dpif_netdev->dp);
1025 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1027 flow->used = time_msec();
1028 flow->packet_count++;
1029 flow->byte_count += packet->size;
1030 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1034 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1035 struct ofpbuf *packet)
1037 struct dp_netdev_flow *flow;
1040 if (packet->size < ETH_HEADER_LEN) {
1043 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1044 flow = dp_netdev_lookup_flow(dp, &key);
1046 dp_netdev_flow_used(flow, packet);
1047 dp_netdev_execute_actions(dp, packet, &key,
1048 flow->actions, flow->actions_len);
1052 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1057 dpif_netdev_run(struct dpif *dpif)
1059 struct dp_netdev *dp = get_dp_netdev(dpif);
1060 struct dp_netdev_port *port;
1061 struct ofpbuf packet;
1063 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1065 LIST_FOR_EACH (port, node, &dp->port_list) {
1068 /* Reset packet contents. */
1069 ofpbuf_clear(&packet);
1070 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1072 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1074 dp_netdev_port_input(dp, port, &packet);
1075 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1076 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1077 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1078 netdev_vport_get_dpif_port(port->netdev),
1082 ofpbuf_uninit(&packet);
1086 dpif_netdev_wait(struct dpif *dpif)
1088 struct dp_netdev *dp = get_dp_netdev(dpif);
1089 struct dp_netdev_port *port;
1091 LIST_FOR_EACH (port, node, &dp->port_list) {
1093 netdev_rx_wait(port->rx);
1099 dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1101 struct dp_netdev *dp = dp_;
1102 struct dp_netdev_port *p = dp->ports[out_port];
1104 netdev_send(p->netdev, packet);
1109 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1110 int queue_no, const struct flow *flow,
1111 const struct nlattr *userdata)
1113 struct dp_netdev_queue *q = &dp->queues[queue_no];
1114 if (q->head - q->tail < MAX_QUEUE_LEN) {
1115 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1116 struct dpif_upcall *upcall = &u->upcall;
1117 struct ofpbuf *buf = &u->buf;
1120 upcall->type = queue_no;
1122 /* Allocate buffer big enough for everything. */
1123 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1125 buf_size += NLA_ALIGN(userdata->nla_len);
1127 ofpbuf_init(buf, buf_size);
1130 odp_flow_key_from_flow(buf, flow, flow->in_port);
1131 upcall->key = buf->data;
1132 upcall->key_len = buf->size;
1136 upcall->userdata = ofpbuf_put(buf, userdata,
1137 NLA_ALIGN(userdata->nla_len));
1142 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1143 * is visible in 'upcall->packet'. The ODP flow and (if present)
1144 * userdata become part of the headroom. */
1145 ofpbuf_put_zeros(buf, 2);
1146 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1147 buf->size = packet->size;
1148 upcall->packet = buf;
1158 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1159 const struct flow *key,
1160 const struct nlattr *userdata)
1162 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1166 dp_netdev_execute_actions(struct dp_netdev *dp,
1167 struct ofpbuf *packet, struct flow *key,
1168 const struct nlattr *actions,
1171 odp_execute_actions(dp, packet, key, actions, actions_len,
1172 dp_netdev_output_port, dp_netdev_action_userspace);
1175 const struct dpif_class dpif_netdev_class = {
1177 dpif_netdev_enumerate,
1178 dpif_netdev_port_open_type,
1181 dpif_netdev_destroy,
1184 dpif_netdev_get_stats,
1185 dpif_netdev_port_add,
1186 dpif_netdev_port_del,
1187 dpif_netdev_port_query_by_number,
1188 dpif_netdev_port_query_by_name,
1189 dpif_netdev_get_max_ports,
1190 NULL, /* port_get_pid */
1191 dpif_netdev_port_dump_start,
1192 dpif_netdev_port_dump_next,
1193 dpif_netdev_port_dump_done,
1194 dpif_netdev_port_poll,
1195 dpif_netdev_port_poll_wait,
1196 dpif_netdev_flow_get,
1197 dpif_netdev_flow_put,
1198 dpif_netdev_flow_del,
1199 dpif_netdev_flow_flush,
1200 dpif_netdev_flow_dump_start,
1201 dpif_netdev_flow_dump_next,
1202 dpif_netdev_flow_dump_done,
1203 dpif_netdev_execute,
1205 dpif_netdev_recv_set,
1206 dpif_netdev_queue_to_priority,
1208 dpif_netdev_recv_wait,
1209 dpif_netdev_recv_purge,
1213 dpif_dummy_register__(const char *type)
1215 struct dpif_class *class;
1217 class = xmalloc(sizeof *class);
1218 *class = dpif_netdev_class;
1219 class->type = xstrdup(type);
1220 dp_register_provider(class);
1224 dpif_dummy_register(bool override)
1231 dp_enumerate_types(&types);
1232 SSET_FOR_EACH (type, &types) {
1233 if (!dp_unregister_provider(type)) {
1234 dpif_dummy_register__(type);
1237 sset_destroy(&types);
1240 dpif_dummy_register__("dummy");