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;
90 int max_mtu; /* Maximum MTU of any port added so far. */
92 struct dp_netdev_queue queues[N_QUEUES];
93 struct hmap flow_table; /* Flow table. */
96 long long int n_hit; /* Number of flow table matches. */
97 long long int n_missed; /* Number of flow table misses. */
98 long long int n_lost; /* Number of misses not passed to client. */
101 struct dp_netdev_port *ports[MAX_PORTS];
102 struct list port_list;
106 /* A port in a netdev-based datapath. */
107 struct dp_netdev_port {
108 odp_port_t port_no; /* Index into dp_netdev's 'ports'. */
109 struct list node; /* Element in dp_netdev's 'port_list'. */
110 struct netdev *netdev;
111 struct netdev_saved_flags *sf;
112 struct netdev_rx *rx;
113 char *type; /* Port type as requested by user. */
116 /* A flow in dp_netdev's 'flow_table'. */
117 struct dp_netdev_flow {
118 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
122 long long int used; /* Last used time, in monotonic msecs. */
123 long long int packet_count; /* Number of packets matched. */
124 long long int byte_count; /* Number of bytes matched. */
125 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
128 struct nlattr *actions;
132 /* Interface to netdev-based datapath. */
135 struct dp_netdev *dp;
136 unsigned int dp_serial;
139 /* All netdev-based datapaths. */
140 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
142 static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
143 struct dp_netdev_port **portp);
144 static int get_port_by_name(struct dp_netdev *, const char *devname,
145 struct dp_netdev_port **portp);
146 static void dp_netdev_free(struct dp_netdev *);
147 static void dp_netdev_flow_flush(struct dp_netdev *);
148 static int do_add_port(struct dp_netdev *, const char *devname,
149 const char *type, odp_port_t port_no);
150 static int do_del_port(struct dp_netdev *, odp_port_t port_no);
151 static int dpif_netdev_open(const struct dpif_class *, const char *name,
152 bool create, struct dpif **);
153 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
154 int queue_no, const struct flow *,
155 const struct nlattr *userdata);
156 static void dp_netdev_execute_actions(struct dp_netdev *,
157 struct ofpbuf *, struct flow *,
158 const struct nlattr *actions,
160 static void dp_netdev_port_input(struct dp_netdev *dp,
161 struct dp_netdev_port *port,
162 struct ofpbuf *packet, uint32_t skb_priority,
163 uint32_t skb_mark, const struct flow_tnl *tnl);
165 static struct dpif_netdev *
166 dpif_netdev_cast(const struct dpif *dpif)
168 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
169 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
172 static struct dp_netdev *
173 get_dp_netdev(const struct dpif *dpif)
175 return dpif_netdev_cast(dpif)->dp;
179 dpif_netdev_enumerate(struct sset *all_dps)
181 struct shash_node *node;
183 SHASH_FOR_EACH(node, &dp_netdevs) {
184 sset_add(all_dps, node->name);
190 dpif_netdev_class_is_dummy(const struct dpif_class *class)
192 return class != &dpif_netdev_class;
196 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
198 return strcmp(type, "internal") ? type
199 : dpif_netdev_class_is_dummy(class) ? "dummy"
204 create_dpif_netdev(struct dp_netdev *dp)
206 uint16_t netflow_id = hash_string(dp->name, 0);
207 struct dpif_netdev *dpif;
211 dpif = xmalloc(sizeof *dpif);
212 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
214 dpif->dp_serial = dp->serial;
219 /* Choose an unused, non-zero port number and return it on success.
220 * Return ODPP_NONE on failure. */
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]) {
244 return u32_to_odp(port_no);
251 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
252 if (!dp->ports[port_no]) {
253 return u32_to_odp(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 dp->max_mtu = ETH_PAYLOAD_MAX;
273 for (i = 0; i < N_QUEUES; i++) {
274 dp->queues[i].head = dp->queues[i].tail = 0;
276 hmap_init(&dp->flow_table);
277 list_init(&dp->port_list);
279 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
285 shash_add(&dp_netdevs, name, dp);
292 dpif_netdev_open(const struct dpif_class *class, const char *name,
293 bool create, struct dpif **dpifp)
295 struct dp_netdev *dp;
297 dp = shash_find_data(&dp_netdevs, name);
302 int error = create_dp_netdev(name, class, &dp);
306 ovs_assert(dp != NULL);
309 if (dp->class != class) {
316 *dpifp = create_dpif_netdev(dp);
321 dp_netdev_purge_queues(struct dp_netdev *dp)
325 for (i = 0; i < N_QUEUES; i++) {
326 struct dp_netdev_queue *q = &dp->queues[i];
328 while (q->tail != q->head) {
329 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
330 ofpbuf_uninit(&u->buf);
336 dp_netdev_free(struct dp_netdev *dp)
338 struct dp_netdev_port *port, *next;
340 dp_netdev_flow_flush(dp);
341 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
342 do_del_port(dp, port->port_no);
344 dp_netdev_purge_queues(dp);
345 hmap_destroy(&dp->flow_table);
351 dpif_netdev_close(struct dpif *dpif)
353 struct dp_netdev *dp = get_dp_netdev(dpif);
354 ovs_assert(dp->open_cnt > 0);
355 if (--dp->open_cnt == 0 && dp->destroyed) {
356 shash_find_and_delete(&dp_netdevs, dp->name);
363 dpif_netdev_destroy(struct dpif *dpif)
365 struct dp_netdev *dp = get_dp_netdev(dpif);
366 dp->destroyed = true;
371 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
373 struct dp_netdev *dp = get_dp_netdev(dpif);
374 stats->n_flows = hmap_count(&dp->flow_table);
375 stats->n_hit = dp->n_hit;
376 stats->n_missed = dp->n_missed;
377 stats->n_lost = dp->n_lost;
382 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
385 struct netdev_saved_flags *sf;
386 struct dp_netdev_port *port;
387 struct netdev *netdev;
388 struct netdev_rx *rx;
389 const char *open_type;
393 /* XXX reject devices already in some dp_netdev. */
395 /* Open and validate network device. */
396 open_type = dpif_netdev_port_open_type(dp->class, type);
397 error = netdev_open(devname, open_type, &netdev);
401 /* XXX reject loopback devices */
402 /* XXX reject non-Ethernet devices */
404 error = netdev_rx_open(netdev, &rx);
406 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
407 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
408 devname, ovs_strerror(errno));
409 netdev_close(netdev);
413 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
416 netdev_close(netdev);
420 port = xmalloc(sizeof *port);
421 port->port_no = port_no;
422 port->netdev = netdev;
425 port->type = xstrdup(type);
427 error = netdev_get_mtu(netdev, &mtu);
428 if (!error && mtu > dp->max_mtu) {
432 list_push_back(&dp->port_list, &port->node);
433 dp->ports[odp_to_u32(port_no)] = port;
440 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
441 odp_port_t *port_nop)
443 struct dp_netdev *dp = get_dp_netdev(dpif);
444 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
445 const char *dpif_port;
448 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
449 if (*port_nop != ODPP_NONE) {
450 uint32_t port_idx = odp_to_u32(*port_nop);
451 if (port_idx >= MAX_PORTS) {
453 } else if (dp->ports[port_idx]) {
458 port_no = choose_port(dp, dpif_port);
460 if (port_no != ODPP_NONE) {
462 return do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
468 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
470 struct dp_netdev *dp = get_dp_netdev(dpif);
471 return (port_no == ODPP_LOCAL ?
472 EINVAL : do_del_port(dp, port_no));
476 is_valid_port_number(odp_port_t port_no)
478 return odp_to_u32(port_no) < MAX_PORTS;
482 get_port_by_number(struct dp_netdev *dp,
483 odp_port_t port_no, struct dp_netdev_port **portp)
485 if (!is_valid_port_number(port_no)) {
489 *portp = dp->ports[odp_to_u32(port_no)];
490 return *portp ? 0 : ENOENT;
495 get_port_by_name(struct dp_netdev *dp,
496 const char *devname, struct dp_netdev_port **portp)
498 struct dp_netdev_port *port;
500 LIST_FOR_EACH (port, node, &dp->port_list) {
501 if (!strcmp(netdev_get_name(port->netdev), devname)) {
510 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
512 struct dp_netdev_port *port;
515 error = get_port_by_number(dp, port_no, &port);
520 list_remove(&port->node);
521 dp->ports[odp_to_u32(port_no)] = NULL;
524 netdev_close(port->netdev);
525 netdev_restore_flags(port->sf);
526 netdev_rx_close(port->rx);
534 answer_port_query(const struct dp_netdev_port *port,
535 struct dpif_port *dpif_port)
537 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
538 dpif_port->type = xstrdup(port->type);
539 dpif_port->port_no = port->port_no;
543 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
544 struct dpif_port *dpif_port)
546 struct dp_netdev *dp = get_dp_netdev(dpif);
547 struct dp_netdev_port *port;
550 error = get_port_by_number(dp, port_no, &port);
551 if (!error && dpif_port) {
552 answer_port_query(port, dpif_port);
558 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
559 struct dpif_port *dpif_port)
561 struct dp_netdev *dp = get_dp_netdev(dpif);
562 struct dp_netdev_port *port;
565 error = get_port_by_name(dp, devname, &port);
566 if (!error && dpif_port) {
567 answer_port_query(port, dpif_port);
573 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
575 return u32_to_odp(MAX_PORTS);
579 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
581 hmap_remove(&dp->flow_table, &flow->node);
587 dp_netdev_flow_flush(struct dp_netdev *dp)
589 struct dp_netdev_flow *flow, *next;
591 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
592 dp_netdev_free_flow(dp, flow);
597 dpif_netdev_flow_flush(struct dpif *dpif)
599 struct dp_netdev *dp = get_dp_netdev(dpif);
600 dp_netdev_flow_flush(dp);
604 struct dp_netdev_port_state {
610 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
612 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
617 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
618 struct dpif_port *dpif_port)
620 struct dp_netdev_port_state *state = state_;
621 struct dp_netdev *dp = get_dp_netdev(dpif);
624 for (port_idx = odp_to_u32(state->port_no);
625 port_idx < MAX_PORTS; port_idx++) {
626 struct dp_netdev_port *port = dp->ports[port_idx];
629 state->name = xstrdup(netdev_get_name(port->netdev));
630 dpif_port->name = state->name;
631 dpif_port->type = port->type;
632 dpif_port->port_no = port->port_no;
633 state->port_no = u32_to_odp(port_idx + 1);
641 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
643 struct dp_netdev_port_state *state = state_;
650 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
652 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
653 if (dpif->dp_serial != dpif->dp->serial) {
654 dpif->dp_serial = dpif->dp->serial;
662 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
664 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
665 if (dpif->dp_serial != dpif->dp->serial) {
666 poll_immediate_wake();
670 static struct dp_netdev_flow *
671 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
673 struct dp_netdev_flow *flow;
675 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
676 if (flow_equal(&flow->key, key)) {
684 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
686 stats->n_packets = flow->packet_count;
687 stats->n_bytes = flow->byte_count;
688 stats->used = flow->used;
689 stats->tcp_flags = flow->tcp_flags;
693 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
698 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
699 /* This should not happen: it indicates that odp_flow_key_from_flow()
700 * and odp_flow_key_to_flow() disagree on the acceptable form of a
701 * flow. Log the problem as an error, with enough details to enable
703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
705 if (!VLOG_DROP_ERR(&rl)) {
709 odp_flow_key_format(key, key_len, &s);
710 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
717 in_port = flow->in_port.odp_port;
718 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
726 dpif_netdev_flow_get(const struct dpif *dpif,
727 const struct nlattr *nl_key, size_t nl_key_len,
728 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
730 struct dp_netdev *dp = get_dp_netdev(dpif);
731 struct dp_netdev_flow *flow;
735 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
740 flow = dp_netdev_lookup_flow(dp, &key);
746 get_dpif_flow_stats(flow, stats);
749 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
755 set_flow_actions(struct dp_netdev_flow *flow,
756 const struct nlattr *actions, size_t actions_len)
758 flow->actions = xrealloc(flow->actions, actions_len);
759 flow->actions_len = actions_len;
760 memcpy(flow->actions, actions, actions_len);
765 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
766 const struct nlattr *actions, size_t actions_len)
768 struct dp_netdev_flow *flow;
771 flow = xzalloc(sizeof *flow);
774 error = set_flow_actions(flow, actions, actions_len);
780 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
785 clear_stats(struct dp_netdev_flow *flow)
788 flow->packet_count = 0;
789 flow->byte_count = 0;
794 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
796 struct dp_netdev *dp = get_dp_netdev(dpif);
797 struct dp_netdev_flow *flow;
801 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
806 flow = dp_netdev_lookup_flow(dp, &key);
808 if (put->flags & DPIF_FP_CREATE) {
809 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
811 memset(put->stats, 0, sizeof *put->stats);
813 return dp_netdev_flow_add(dp, &key, put->actions,
822 if (put->flags & DPIF_FP_MODIFY) {
823 int error = set_flow_actions(flow, put->actions, put->actions_len);
826 get_dpif_flow_stats(flow, put->stats);
828 if (put->flags & DPIF_FP_ZERO_STATS) {
840 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
842 struct dp_netdev *dp = get_dp_netdev(dpif);
843 struct dp_netdev_flow *flow;
847 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
852 flow = dp_netdev_lookup_flow(dp, &key);
855 get_dpif_flow_stats(flow, del->stats);
857 dp_netdev_free_flow(dp, flow);
864 struct dp_netdev_flow_state {
867 struct nlattr *actions;
868 struct odputil_keybuf keybuf;
869 struct dpif_flow_stats stats;
873 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
875 struct dp_netdev_flow_state *state;
877 *statep = state = xmalloc(sizeof *state);
880 state->actions = NULL;
885 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
886 const struct nlattr **key, size_t *key_len,
887 const struct nlattr **mask, size_t *mask_len,
888 const struct nlattr **actions, size_t *actions_len,
889 const struct dpif_flow_stats **stats)
891 struct dp_netdev_flow_state *state = state_;
892 struct dp_netdev *dp = get_dp_netdev(dpif);
893 struct dp_netdev_flow *flow;
894 struct hmap_node *node;
896 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
901 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
906 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
907 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port.odp_port);
919 free(state->actions);
920 state->actions = xmemdup(flow->actions, flow->actions_len);
922 *actions = state->actions;
923 *actions_len = flow->actions_len;
927 get_dpif_flow_stats(flow, &state->stats);
928 *stats = &state->stats;
935 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
937 struct dp_netdev_flow_state *state = state_;
939 free(state->actions);
945 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
947 struct dp_netdev *dp = get_dp_netdev(dpif);
952 if (execute->packet->size < ETH_HEADER_LEN ||
953 execute->packet->size > UINT16_MAX) {
957 /* Make a deep copy of 'packet', because we might modify its data. */
958 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
959 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
960 ofpbuf_put(©, execute->packet->data, execute->packet->size);
962 flow_extract(©, 0, 0, NULL, NULL, &key);
963 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
966 dp_netdev_execute_actions(dp, ©, &key,
967 execute->actions, execute->actions_len);
970 ofpbuf_uninit(©);
975 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
981 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
982 uint32_t queue_id, uint32_t *priority)
984 *priority = queue_id;
988 static struct dp_netdev_queue *
989 find_nonempty_queue(struct dpif *dpif)
991 struct dp_netdev *dp = get_dp_netdev(dpif);
994 for (i = 0; i < N_QUEUES; i++) {
995 struct dp_netdev_queue *q = &dp->queues[i];
996 if (q->head != q->tail) {
1004 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1007 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1009 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1011 *upcall = u->upcall;
1012 upcall->packet = buf;
1024 dpif_netdev_recv_wait(struct dpif *dpif)
1026 if (find_nonempty_queue(dpif)) {
1027 poll_immediate_wake();
1029 /* No messages ready to be received, and dp_wait() will ensure that we
1030 * wake up to queue new messages, so there is nothing to do. */
1035 dpif_netdev_recv_purge(struct dpif *dpif)
1037 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1038 dp_netdev_purge_queues(dpif_netdev->dp);
1042 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1044 flow->used = time_msec();
1045 flow->packet_count++;
1046 flow->byte_count += packet->size;
1047 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1051 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1052 struct ofpbuf *packet, uint32_t skb_priority,
1053 uint32_t skb_mark, const struct flow_tnl *tnl)
1055 struct dp_netdev_flow *flow;
1057 union flow_in_port in_port_;
1059 if (packet->size < ETH_HEADER_LEN) {
1062 in_port_.odp_port = port->port_no;
1063 flow_extract(packet, skb_priority, skb_mark, tnl, &in_port_, &key);
1064 flow = dp_netdev_lookup_flow(dp, &key);
1066 dp_netdev_flow_used(flow, packet);
1067 dp_netdev_execute_actions(dp, packet, &key,
1068 flow->actions, flow->actions_len);
1072 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1077 dpif_netdev_run(struct dpif *dpif)
1079 struct dp_netdev *dp = get_dp_netdev(dpif);
1080 struct dp_netdev_port *port;
1081 struct ofpbuf packet;
1083 ofpbuf_init(&packet,
1084 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1086 LIST_FOR_EACH (port, node, &dp->port_list) {
1089 /* Reset packet contents. */
1090 ofpbuf_clear(&packet);
1091 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1093 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1095 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1096 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1097 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1099 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1100 netdev_get_name(port->netdev), ovs_strerror(error));
1103 ofpbuf_uninit(&packet);
1107 dpif_netdev_wait(struct dpif *dpif)
1109 struct dp_netdev *dp = get_dp_netdev(dpif);
1110 struct dp_netdev_port *port;
1112 LIST_FOR_EACH (port, node, &dp->port_list) {
1114 netdev_rx_wait(port->rx);
1120 dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1122 struct dp_netdev *dp = dp_;
1123 struct dp_netdev_port *p = dp->ports[out_port];
1125 netdev_send(p->netdev, packet);
1130 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1131 int queue_no, const struct flow *flow,
1132 const struct nlattr *userdata)
1134 struct dp_netdev_queue *q = &dp->queues[queue_no];
1135 if (q->head - q->tail < MAX_QUEUE_LEN) {
1136 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1137 struct dpif_upcall *upcall = &u->upcall;
1138 struct ofpbuf *buf = &u->buf;
1141 upcall->type = queue_no;
1143 /* Allocate buffer big enough for everything. */
1144 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1146 buf_size += NLA_ALIGN(userdata->nla_len);
1148 ofpbuf_init(buf, buf_size);
1151 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1152 upcall->key = buf->data;
1153 upcall->key_len = buf->size;
1157 upcall->userdata = ofpbuf_put(buf, userdata,
1158 NLA_ALIGN(userdata->nla_len));
1163 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1164 * is visible in 'upcall->packet'. The ODP flow and (if present)
1165 * userdata become part of the headroom. */
1166 ofpbuf_put_zeros(buf, 2);
1167 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1168 buf->size = packet->size;
1169 upcall->packet = buf;
1179 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1180 const struct flow *key,
1181 const struct nlattr *userdata)
1183 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1187 dp_netdev_execute_actions(struct dp_netdev *dp,
1188 struct ofpbuf *packet, struct flow *key,
1189 const struct nlattr *actions,
1192 odp_execute_actions(dp, packet, key, actions, actions_len,
1193 dp_netdev_output_port, dp_netdev_action_userspace);
1196 const struct dpif_class dpif_netdev_class = {
1198 dpif_netdev_enumerate,
1199 dpif_netdev_port_open_type,
1202 dpif_netdev_destroy,
1205 dpif_netdev_get_stats,
1206 dpif_netdev_port_add,
1207 dpif_netdev_port_del,
1208 dpif_netdev_port_query_by_number,
1209 dpif_netdev_port_query_by_name,
1210 dpif_netdev_get_max_ports,
1211 NULL, /* port_get_pid */
1212 dpif_netdev_port_dump_start,
1213 dpif_netdev_port_dump_next,
1214 dpif_netdev_port_dump_done,
1215 dpif_netdev_port_poll,
1216 dpif_netdev_port_poll_wait,
1217 dpif_netdev_flow_get,
1218 dpif_netdev_flow_put,
1219 dpif_netdev_flow_del,
1220 dpif_netdev_flow_flush,
1221 dpif_netdev_flow_dump_start,
1222 dpif_netdev_flow_dump_next,
1223 dpif_netdev_flow_dump_done,
1224 dpif_netdev_execute,
1226 dpif_netdev_recv_set,
1227 dpif_netdev_queue_to_priority,
1229 dpif_netdev_recv_wait,
1230 dpif_netdev_recv_purge,
1234 dpif_dummy_register__(const char *type)
1236 struct dpif_class *class;
1238 class = xmalloc(sizeof *class);
1239 *class = dpif_netdev_class;
1240 class->type = xstrdup(type);
1241 dp_register_provider(class);
1245 dpif_dummy_register(bool override)
1252 dp_enumerate_types(&types);
1253 SSET_FOR_EACH (type, &types) {
1254 if (!dp_unregister_provider(type)) {
1255 dpif_dummy_register__(type);
1258 sset_destroy(&types);
1261 dpif_dummy_register__("dummy");