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,
162 static void dp_netdev_port_input(struct dp_netdev *dp,
163 struct dp_netdev_port *port,
164 struct ofpbuf *packet, uint32_t skb_priority,
165 uint32_t skb_mark, const struct flow_tnl *tnl);
167 static struct dpif_netdev *
168 dpif_netdev_cast(const struct dpif *dpif)
170 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
171 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
174 static struct dp_netdev *
175 get_dp_netdev(const struct dpif *dpif)
177 return dpif_netdev_cast(dpif)->dp;
181 dpif_netdev_enumerate(struct sset *all_dps)
183 struct shash_node *node;
185 SHASH_FOR_EACH(node, &dp_netdevs) {
186 sset_add(all_dps, node->name);
192 dpif_netdev_class_is_dummy(const struct dpif_class *class)
194 return class != &dpif_netdev_class;
198 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
200 return strcmp(type, "internal") ? type
201 : dpif_netdev_class_is_dummy(class) ? "dummy"
206 create_dpif_netdev(struct dp_netdev *dp)
208 uint16_t netflow_id = hash_string(dp->name, 0);
209 struct dpif_netdev *dpif;
213 dpif = xmalloc(sizeof *dpif);
214 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
216 dpif->dp_serial = dp->serial;
222 choose_port(struct dp_netdev *dp, const char *name)
226 if (dp->class != &dpif_netdev_class) {
230 /* If the port name begins with "br", start the number search at
231 * 100 to make writing tests easier. */
232 if (!strncmp(name, "br", 2)) {
236 /* If the port name contains a number, try to assign that port number.
237 * This can make writing unit tests easier because port numbers are
239 for (p = name; *p != '\0'; p++) {
240 if (isdigit((unsigned char) *p)) {
241 port_no = start_no + strtol(p, NULL, 10);
242 if (port_no > 0 && port_no < MAX_PORTS
243 && !dp->ports[port_no]) {
251 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
252 if (!dp->ports[port_no]) {
261 create_dp_netdev(const char *name, const struct dpif_class *class,
262 struct dp_netdev **dpp)
264 struct dp_netdev *dp;
268 dp = xzalloc(sizeof *dp);
270 dp->name = xstrdup(name);
272 for (i = 0; i < N_QUEUES; i++) {
273 dp->queues[i].head = dp->queues[i].tail = 0;
275 hmap_init(&dp->flow_table);
276 list_init(&dp->port_list);
278 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
284 shash_add(&dp_netdevs, name, dp);
291 dpif_netdev_open(const struct dpif_class *class, const char *name,
292 bool create, struct dpif **dpifp)
294 struct dp_netdev *dp;
296 dp = shash_find_data(&dp_netdevs, name);
301 int error = create_dp_netdev(name, class, &dp);
305 ovs_assert(dp != NULL);
308 if (dp->class != class) {
315 *dpifp = create_dpif_netdev(dp);
320 dp_netdev_purge_queues(struct dp_netdev *dp)
324 for (i = 0; i < N_QUEUES; i++) {
325 struct dp_netdev_queue *q = &dp->queues[i];
327 while (q->tail != q->head) {
328 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
329 ofpbuf_uninit(&u->buf);
335 dp_netdev_free(struct dp_netdev *dp)
337 struct dp_netdev_port *port, *next;
339 dp_netdev_flow_flush(dp);
340 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
341 do_del_port(dp, port->port_no);
343 dp_netdev_purge_queues(dp);
344 hmap_destroy(&dp->flow_table);
350 dpif_netdev_close(struct dpif *dpif)
352 struct dp_netdev *dp = get_dp_netdev(dpif);
353 ovs_assert(dp->open_cnt > 0);
354 if (--dp->open_cnt == 0 && dp->destroyed) {
355 shash_find_and_delete(&dp_netdevs, dp->name);
362 dpif_netdev_destroy(struct dpif *dpif)
364 struct dp_netdev *dp = get_dp_netdev(dpif);
365 dp->destroyed = true;
370 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
372 struct dp_netdev *dp = get_dp_netdev(dpif);
373 stats->n_flows = hmap_count(&dp->flow_table);
374 stats->n_hit = dp->n_hit;
375 stats->n_missed = dp->n_missed;
376 stats->n_lost = dp->n_lost;
381 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
384 struct netdev_saved_flags *sf;
385 struct dp_netdev_port *port;
386 struct netdev *netdev;
387 struct netdev_rx *rx;
388 const char *open_type;
392 /* XXX reject devices already in some dp_netdev. */
394 /* Open and validate network device. */
395 open_type = dpif_netdev_port_open_type(dp->class, type);
396 error = netdev_open(devname, open_type, &netdev);
400 /* XXX reject loopback devices */
401 /* XXX reject non-Ethernet devices */
403 error = netdev_rx_open(netdev, &rx);
405 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
406 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
407 devname, strerror(errno));
408 netdev_close(netdev);
412 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
415 netdev_close(netdev);
419 port = xmalloc(sizeof *port);
420 port->port_no = port_no;
421 port->netdev = netdev;
424 port->type = xstrdup(type);
426 error = netdev_get_mtu(netdev, &mtu);
427 if (!error && mtu > max_mtu) {
431 list_push_back(&dp->port_list, &port->node);
432 dp->ports[port_no] = port;
439 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
442 struct dp_netdev *dp = get_dp_netdev(dpif);
445 if (*port_nop != UINT32_MAX) {
446 if (*port_nop >= MAX_PORTS) {
448 } else if (dp->ports[*port_nop]) {
453 port_no = choose_port(dp, netdev_vport_get_dpif_port(netdev));
457 return do_add_port(dp, netdev_vport_get_dpif_port(netdev),
458 netdev_get_type(netdev), port_no);
464 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
466 struct dp_netdev *dp = get_dp_netdev(dpif);
467 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
471 is_valid_port_number(uint32_t port_no)
473 return port_no < MAX_PORTS;
477 get_port_by_number(struct dp_netdev *dp,
478 uint32_t port_no, struct dp_netdev_port **portp)
480 if (!is_valid_port_number(port_no)) {
484 *portp = dp->ports[port_no];
485 return *portp ? 0 : ENOENT;
490 get_port_by_name(struct dp_netdev *dp,
491 const char *devname, struct dp_netdev_port **portp)
493 struct dp_netdev_port *port;
495 LIST_FOR_EACH (port, node, &dp->port_list) {
496 if (!strcmp(netdev_get_name(port->netdev), devname)) {
505 do_del_port(struct dp_netdev *dp, uint32_t port_no)
507 struct dp_netdev_port *port;
510 error = get_port_by_number(dp, port_no, &port);
515 list_remove(&port->node);
516 dp->ports[port->port_no] = NULL;
519 netdev_close(port->netdev);
520 netdev_restore_flags(port->sf);
521 netdev_rx_close(port->rx);
529 answer_port_query(const struct dp_netdev_port *port,
530 struct dpif_port *dpif_port)
532 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
533 dpif_port->type = xstrdup(port->type);
534 dpif_port->port_no = port->port_no;
538 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
539 struct dpif_port *dpif_port)
541 struct dp_netdev *dp = get_dp_netdev(dpif);
542 struct dp_netdev_port *port;
545 error = get_port_by_number(dp, port_no, &port);
546 if (!error && dpif_port) {
547 answer_port_query(port, dpif_port);
553 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
554 struct dpif_port *dpif_port)
556 struct dp_netdev *dp = get_dp_netdev(dpif);
557 struct dp_netdev_port *port;
560 error = get_port_by_name(dp, devname, &port);
561 if (!error && dpif_port) {
562 answer_port_query(port, dpif_port);
568 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
574 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
576 hmap_remove(&dp->flow_table, &flow->node);
582 dp_netdev_flow_flush(struct dp_netdev *dp)
584 struct dp_netdev_flow *flow, *next;
586 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
587 dp_netdev_free_flow(dp, flow);
592 dpif_netdev_flow_flush(struct dpif *dpif)
594 struct dp_netdev *dp = get_dp_netdev(dpif);
595 dp_netdev_flow_flush(dp);
599 struct dp_netdev_port_state {
605 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
607 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
612 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
613 struct dpif_port *dpif_port)
615 struct dp_netdev_port_state *state = state_;
616 struct dp_netdev *dp = get_dp_netdev(dpif);
619 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
620 struct dp_netdev_port *port = dp->ports[port_no];
623 state->name = xstrdup(netdev_get_name(port->netdev));
624 dpif_port->name = state->name;
625 dpif_port->type = port->type;
626 dpif_port->port_no = port->port_no;
627 state->port_no = port_no + 1;
635 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
637 struct dp_netdev_port_state *state = state_;
644 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
646 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
647 if (dpif->dp_serial != dpif->dp->serial) {
648 dpif->dp_serial = dpif->dp->serial;
656 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
658 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
659 if (dpif->dp_serial != dpif->dp->serial) {
660 poll_immediate_wake();
664 static struct dp_netdev_flow *
665 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
667 struct dp_netdev_flow *flow;
669 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
670 if (flow_equal(&flow->key, key)) {
678 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
680 stats->n_packets = flow->packet_count;
681 stats->n_bytes = flow->byte_count;
682 stats->used = flow->used;
683 stats->tcp_flags = flow->tcp_flags;
687 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
690 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
691 /* This should not happen: it indicates that odp_flow_key_from_flow()
692 * and odp_flow_key_to_flow() disagree on the acceptable form of a
693 * flow. Log the problem as an error, with enough details to enable
695 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
697 if (!VLOG_DROP_ERR(&rl)) {
701 odp_flow_key_format(key, key_len, &s);
702 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
709 if (flow->in_port < OFPP_MAX
710 ? flow->in_port >= MAX_PORTS
711 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
719 dpif_netdev_flow_get(const struct dpif *dpif,
720 const struct nlattr *nl_key, size_t nl_key_len,
721 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
723 struct dp_netdev *dp = get_dp_netdev(dpif);
724 struct dp_netdev_flow *flow;
728 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
733 flow = dp_netdev_lookup_flow(dp, &key);
739 get_dpif_flow_stats(flow, stats);
742 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
748 set_flow_actions(struct dp_netdev_flow *flow,
749 const struct nlattr *actions, size_t actions_len)
751 flow->actions = xrealloc(flow->actions, actions_len);
752 flow->actions_len = actions_len;
753 memcpy(flow->actions, actions, actions_len);
758 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
759 const struct nlattr *actions, size_t actions_len)
761 struct dp_netdev_flow *flow;
764 flow = xzalloc(sizeof *flow);
767 error = set_flow_actions(flow, actions, actions_len);
773 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
778 clear_stats(struct dp_netdev_flow *flow)
781 flow->packet_count = 0;
782 flow->byte_count = 0;
787 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
789 struct dp_netdev *dp = get_dp_netdev(dpif);
790 struct dp_netdev_flow *flow;
794 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
799 flow = dp_netdev_lookup_flow(dp, &key);
801 if (put->flags & DPIF_FP_CREATE) {
802 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
804 memset(put->stats, 0, sizeof *put->stats);
806 return dp_netdev_flow_add(dp, &key, put->actions,
815 if (put->flags & DPIF_FP_MODIFY) {
816 int error = set_flow_actions(flow, put->actions, put->actions_len);
819 get_dpif_flow_stats(flow, put->stats);
821 if (put->flags & DPIF_FP_ZERO_STATS) {
833 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
835 struct dp_netdev *dp = get_dp_netdev(dpif);
836 struct dp_netdev_flow *flow;
840 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
845 flow = dp_netdev_lookup_flow(dp, &key);
848 get_dpif_flow_stats(flow, del->stats);
850 dp_netdev_free_flow(dp, flow);
857 struct dp_netdev_flow_state {
860 struct nlattr *actions;
861 struct odputil_keybuf keybuf;
862 struct dpif_flow_stats stats;
866 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
868 struct dp_netdev_flow_state *state;
870 *statep = state = xmalloc(sizeof *state);
873 state->actions = NULL;
878 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
879 const struct nlattr **key, size_t *key_len,
880 const struct nlattr **actions, size_t *actions_len,
881 const struct dpif_flow_stats **stats)
883 struct dp_netdev_flow_state *state = state_;
884 struct dp_netdev *dp = get_dp_netdev(dpif);
885 struct dp_netdev_flow *flow;
886 struct hmap_node *node;
888 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
893 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
898 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
899 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
906 free(state->actions);
907 state->actions = xmemdup(flow->actions, flow->actions_len);
909 *actions = state->actions;
910 *actions_len = flow->actions_len;
914 get_dpif_flow_stats(flow, &state->stats);
915 *stats = &state->stats;
922 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
924 struct dp_netdev_flow_state *state = state_;
926 free(state->actions);
932 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
934 struct dp_netdev *dp = get_dp_netdev(dpif);
939 if (execute->packet->size < ETH_HEADER_LEN ||
940 execute->packet->size > UINT16_MAX) {
944 /* Make a deep copy of 'packet', because we might modify its data. */
945 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
946 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
947 ofpbuf_put(©, execute->packet->data, execute->packet->size);
949 flow_extract(©, 0, 0, NULL, -1, &key);
950 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
953 dp_netdev_execute_actions(dp, ©, &key,
954 execute->actions, execute->actions_len);
957 ofpbuf_uninit(©);
962 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
968 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
969 uint32_t queue_id, uint32_t *priority)
971 *priority = queue_id;
975 static struct dp_netdev_queue *
976 find_nonempty_queue(struct dpif *dpif)
978 struct dp_netdev *dp = get_dp_netdev(dpif);
981 for (i = 0; i < N_QUEUES; i++) {
982 struct dp_netdev_queue *q = &dp->queues[i];
983 if (q->head != q->tail) {
991 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
994 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
996 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
999 upcall->packet = buf;
1011 dpif_netdev_recv_wait(struct dpif *dpif)
1013 if (find_nonempty_queue(dpif)) {
1014 poll_immediate_wake();
1016 /* No messages ready to be received, and dp_wait() will ensure that we
1017 * wake up to queue new messages, so there is nothing to do. */
1022 dpif_netdev_recv_purge(struct dpif *dpif)
1024 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1025 dp_netdev_purge_queues(dpif_netdev->dp);
1029 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1031 flow->used = time_msec();
1032 flow->packet_count++;
1033 flow->byte_count += packet->size;
1034 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1038 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1039 struct ofpbuf *packet, uint32_t skb_priority,
1040 uint32_t skb_mark, const struct flow_tnl *tnl)
1042 struct dp_netdev_flow *flow;
1045 if (packet->size < ETH_HEADER_LEN) {
1048 flow_extract(packet, skb_priority, skb_mark, tnl, port->port_no, &key);
1049 flow = dp_netdev_lookup_flow(dp, &key);
1051 dp_netdev_flow_used(flow, packet);
1052 dp_netdev_execute_actions(dp, packet, &key,
1053 flow->actions, flow->actions_len);
1057 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1062 dpif_netdev_run(struct dpif *dpif)
1064 struct dp_netdev *dp = get_dp_netdev(dpif);
1065 struct dp_netdev_port *port;
1066 struct ofpbuf packet;
1068 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1070 LIST_FOR_EACH (port, node, &dp->port_list) {
1073 /* Reset packet contents. */
1074 ofpbuf_clear(&packet);
1075 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1077 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1079 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1080 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1081 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1082 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1083 netdev_get_name(port->netdev), strerror(error));
1086 ofpbuf_uninit(&packet);
1090 dpif_netdev_wait(struct dpif *dpif)
1092 struct dp_netdev *dp = get_dp_netdev(dpif);
1093 struct dp_netdev_port *port;
1095 LIST_FOR_EACH (port, node, &dp->port_list) {
1097 netdev_rx_wait(port->rx);
1103 dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1105 struct dp_netdev *dp = dp_;
1106 struct dp_netdev_port *p = dp->ports[out_port];
1108 netdev_send(p->netdev, packet);
1113 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1114 int queue_no, const struct flow *flow,
1115 const struct nlattr *userdata)
1117 struct dp_netdev_queue *q = &dp->queues[queue_no];
1118 if (q->head - q->tail < MAX_QUEUE_LEN) {
1119 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1120 struct dpif_upcall *upcall = &u->upcall;
1121 struct ofpbuf *buf = &u->buf;
1124 upcall->type = queue_no;
1126 /* Allocate buffer big enough for everything. */
1127 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1129 buf_size += NLA_ALIGN(userdata->nla_len);
1131 ofpbuf_init(buf, buf_size);
1134 odp_flow_key_from_flow(buf, flow, flow->in_port);
1135 upcall->key = buf->data;
1136 upcall->key_len = buf->size;
1140 upcall->userdata = ofpbuf_put(buf, userdata,
1141 NLA_ALIGN(userdata->nla_len));
1146 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1147 * is visible in 'upcall->packet'. The ODP flow and (if present)
1148 * userdata become part of the headroom. */
1149 ofpbuf_put_zeros(buf, 2);
1150 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1151 buf->size = packet->size;
1152 upcall->packet = buf;
1162 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1163 const struct flow *key,
1164 const struct nlattr *userdata)
1166 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1170 dp_netdev_execute_actions(struct dp_netdev *dp,
1171 struct ofpbuf *packet, struct flow *key,
1172 const struct nlattr *actions,
1175 odp_execute_actions(dp, packet, key, actions, actions_len,
1176 dp_netdev_output_port, dp_netdev_action_userspace);
1179 const struct dpif_class dpif_netdev_class = {
1181 dpif_netdev_enumerate,
1182 dpif_netdev_port_open_type,
1185 dpif_netdev_destroy,
1188 dpif_netdev_get_stats,
1189 dpif_netdev_port_add,
1190 dpif_netdev_port_del,
1191 dpif_netdev_port_query_by_number,
1192 dpif_netdev_port_query_by_name,
1193 dpif_netdev_get_max_ports,
1194 NULL, /* port_get_pid */
1195 dpif_netdev_port_dump_start,
1196 dpif_netdev_port_dump_next,
1197 dpif_netdev_port_dump_done,
1198 dpif_netdev_port_poll,
1199 dpif_netdev_port_poll_wait,
1200 dpif_netdev_flow_get,
1201 dpif_netdev_flow_put,
1202 dpif_netdev_flow_del,
1203 dpif_netdev_flow_flush,
1204 dpif_netdev_flow_dump_start,
1205 dpif_netdev_flow_dump_next,
1206 dpif_netdev_flow_dump_done,
1207 dpif_netdev_execute,
1209 dpif_netdev_recv_set,
1210 dpif_netdev_queue_to_priority,
1212 dpif_netdev_recv_wait,
1213 dpif_netdev_recv_purge,
1217 dpif_dummy_register__(const char *type)
1219 struct dpif_class *class;
1221 class = xmalloc(sizeof *class);
1222 *class = dpif_netdev_class;
1223 class->type = xstrdup(type);
1224 dp_register_provider(class);
1228 dpif_dummy_register(bool override)
1235 dp_enumerate_types(&types);
1236 SSET_FOR_EACH (type, &types) {
1237 if (!dp_unregister_provider(type)) {
1238 dpif_dummy_register__(type);
1241 sset_destroy(&types);
1244 dpif_dummy_register__("dummy");