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.
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"
45 #include "ofp-print.h"
48 #include "poll-loop.h"
56 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
58 /* Configuration parameters. */
59 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
60 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
62 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
63 * headers to be aligned on a 4-byte boundary. */
64 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
67 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
68 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
69 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
70 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
72 struct dp_netdev_upcall {
73 struct dpif_upcall upcall; /* Queued upcall information. */
74 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
77 struct dp_netdev_queue {
78 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN];
79 unsigned int head, tail;
82 /* Datapath based on the network device interface from netdev.h. */
84 const struct dpif_class *class;
89 struct dp_netdev_queue queues[N_QUEUES];
90 struct hmap flow_table; /* Flow table. */
93 long long int n_hit; /* Number of flow table matches. */
94 long long int n_missed; /* Number of flow table misses. */
95 long long int n_lost; /* Number of misses not passed to client. */
98 struct dp_netdev_port *ports[MAX_PORTS];
99 struct list port_list;
103 /* A port in a netdev-based datapath. */
104 struct dp_netdev_port {
105 int port_no; /* Index into dp_netdev's 'ports'. */
106 struct list node; /* Element in dp_netdev's 'port_list'. */
107 struct netdev *netdev;
108 char *type; /* Port type as requested by user. */
111 /* A flow in dp_netdev's 'flow_table'. */
112 struct dp_netdev_flow {
113 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
117 long long int used; /* Last used time, in monotonic msecs. */
118 long long int packet_count; /* Number of packets matched. */
119 long long int byte_count; /* Number of bytes matched. */
120 uint8_t tcp_flags; /* Bitwise-OR of seen tcp_flags values. */
123 struct nlattr *actions;
127 /* Interface to netdev-based datapath. */
130 struct dp_netdev *dp;
131 unsigned int dp_serial;
134 /* All netdev-based datapaths. */
135 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
137 /* Maximum port MTU seen so far. */
138 static int max_mtu = ETH_PAYLOAD_MAX;
140 static int get_port_by_number(struct dp_netdev *, uint32_t port_no,
141 struct dp_netdev_port **portp);
142 static int get_port_by_name(struct dp_netdev *, const char *devname,
143 struct dp_netdev_port **portp);
144 static void dp_netdev_free(struct dp_netdev *);
145 static void dp_netdev_flow_flush(struct dp_netdev *);
146 static int do_add_port(struct dp_netdev *, const char *devname,
147 const char *type, uint32_t port_no);
148 static int do_del_port(struct dp_netdev *, uint32_t port_no);
149 static int dpif_netdev_open(const struct dpif_class *, const char *name,
150 bool create, struct dpif **);
151 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
152 int queue_no, const struct flow *,
154 static void dp_netdev_execute_actions(struct dp_netdev *,
155 struct ofpbuf *, struct flow *,
156 const struct nlattr *actions,
159 static struct dpif_netdev *
160 dpif_netdev_cast(const struct dpif *dpif)
162 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
163 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
166 static struct dp_netdev *
167 get_dp_netdev(const struct dpif *dpif)
169 return dpif_netdev_cast(dpif)->dp;
173 dpif_netdev_enumerate(struct sset *all_dps)
175 struct shash_node *node;
177 SHASH_FOR_EACH(node, &dp_netdevs) {
178 sset_add(all_dps, node->name);
184 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
186 return strcmp(type, "internal") ? type
187 : class != &dpif_netdev_class ? "dummy"
192 dpif_planetlab_port_open_type(const struct dpif_class *class, const char *type)
194 return strcmp(type, "internal") ? type
195 : class != &dpif_planetlab_class ? "dummy"
200 create_dpif_netdev(struct dp_netdev *dp)
202 uint16_t netflow_id = hash_string(dp->name, 0);
203 struct dpif_netdev *dpif;
207 dpif = xmalloc(sizeof *dpif);
208 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
210 dpif->dp_serial = dp->serial;
216 choose_port(struct dp_netdev *dp, const char *name)
220 if (dp->class != &dpif_netdev_class &&
221 dp->class != &dpif_planetlab_class) {
225 /* If the port name begins with "br", start the number search at
226 * 100 to make writing tests easier. */
227 if (!strncmp(name, "br", 2)) {
231 /* If the port name contains a number, try to assign that port number.
232 * This can make writing unit tests easier because port numbers are
234 for (p = name; *p != '\0'; p++) {
235 if (isdigit((unsigned char) *p)) {
236 port_no = start_no + strtol(p, NULL, 10);
237 if (port_no > 0 && port_no < MAX_PORTS
238 && !dp->ports[port_no]) {
246 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
247 if (!dp->ports[port_no]) {
256 create_dp_netdev(const char *name, const struct dpif_class *class,
257 struct dp_netdev **dpp)
259 struct dp_netdev *dp;
263 dp = xzalloc(sizeof *dp);
265 dp->name = xstrdup(name);
267 for (i = 0; i < N_QUEUES; i++) {
268 dp->queues[i].head = dp->queues[i].tail = 0;
270 hmap_init(&dp->flow_table);
271 list_init(&dp->port_list);
273 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
279 shash_add(&dp_netdevs, name, dp);
286 dpif_netdev_open(const struct dpif_class *class, const char *name,
287 bool create, struct dpif **dpifp)
289 struct dp_netdev *dp;
291 dp = shash_find_data(&dp_netdevs, name);
296 int error = create_dp_netdev(name, class, &dp);
300 ovs_assert(dp != NULL);
303 if (dp->class != class) {
310 *dpifp = create_dpif_netdev(dp);
315 dp_netdev_purge_queues(struct dp_netdev *dp)
319 for (i = 0; i < N_QUEUES; i++) {
320 struct dp_netdev_queue *q = &dp->queues[i];
322 while (q->tail != q->head) {
323 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
324 ofpbuf_uninit(&u->buf);
330 dp_netdev_free(struct dp_netdev *dp)
332 struct dp_netdev_port *port, *next;
334 dp_netdev_flow_flush(dp);
335 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
336 do_del_port(dp, port->port_no);
338 dp_netdev_purge_queues(dp);
339 hmap_destroy(&dp->flow_table);
345 dpif_netdev_close(struct dpif *dpif)
347 struct dp_netdev *dp = get_dp_netdev(dpif);
348 ovs_assert(dp->open_cnt > 0);
349 if (--dp->open_cnt == 0 && dp->destroyed) {
350 shash_find_and_delete(&dp_netdevs, dp->name);
357 dpif_netdev_destroy(struct dpif *dpif)
359 struct dp_netdev *dp = get_dp_netdev(dpif);
360 dp->destroyed = true;
365 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
367 struct dp_netdev *dp = get_dp_netdev(dpif);
368 stats->n_flows = hmap_count(&dp->flow_table);
369 stats->n_hit = dp->n_hit;
370 stats->n_missed = dp->n_missed;
371 stats->n_lost = dp->n_lost;
376 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
379 struct dp_netdev_port *port;
380 struct netdev *netdev;
381 const char *open_type;
385 /* XXX reject devices already in some dp_netdev. */
387 /* Open and validate network device. */
388 open_type = dpif_netdev_port_open_type(dp->class, type);
389 error = netdev_open(devname, open_type, &netdev);
393 /* XXX reject loopback devices */
394 /* XXX reject non-Ethernet devices */
396 error = netdev_listen(netdev);
398 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
399 devname, strerror(errno));
400 netdev_close(netdev);
404 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
406 netdev_close(netdev);
410 port = xmalloc(sizeof *port);
411 port->port_no = port_no;
412 port->netdev = netdev;
413 port->type = xstrdup(type);
415 error = netdev_get_mtu(netdev, &mtu);
420 list_push_back(&dp->port_list, &port->node);
421 dp->ports[port_no] = port;
428 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
431 struct dp_netdev *dp = get_dp_netdev(dpif);
434 if (*port_nop != UINT32_MAX) {
435 if (*port_nop >= MAX_PORTS) {
437 } else if (dp->ports[*port_nop]) {
442 port_no = choose_port(dp, netdev_get_name(netdev));
446 return do_add_port(dp, netdev_get_name(netdev),
447 netdev_get_type(netdev), port_no);
453 dpif_netdev_port_del(struct dpif *dpif, uint32_t port_no)
455 struct dp_netdev *dp = get_dp_netdev(dpif);
456 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
460 is_valid_port_number(uint32_t port_no)
462 return port_no < MAX_PORTS;
466 get_port_by_number(struct dp_netdev *dp,
467 uint32_t port_no, struct dp_netdev_port **portp)
469 if (!is_valid_port_number(port_no)) {
473 *portp = dp->ports[port_no];
474 return *portp ? 0 : ENOENT;
479 get_port_by_name(struct dp_netdev *dp,
480 const char *devname, struct dp_netdev_port **portp)
482 struct dp_netdev_port *port;
484 LIST_FOR_EACH (port, node, &dp->port_list) {
485 if (!strcmp(netdev_get_name(port->netdev), devname)) {
494 do_del_port(struct dp_netdev *dp, uint32_t port_no)
496 struct dp_netdev_port *port;
500 error = get_port_by_number(dp, port_no, &port);
505 list_remove(&port->node);
506 dp->ports[port->port_no] = NULL;
509 name = xstrdup(netdev_get_name(port->netdev));
510 netdev_close(port->netdev);
520 answer_port_query(const struct dp_netdev_port *port,
521 struct dpif_port *dpif_port)
523 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
524 dpif_port->type = xstrdup(port->type);
525 dpif_port->port_no = port->port_no;
529 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
530 struct dpif_port *dpif_port)
532 struct dp_netdev *dp = get_dp_netdev(dpif);
533 struct dp_netdev_port *port;
536 error = get_port_by_number(dp, port_no, &port);
537 if (!error && dpif_port) {
538 answer_port_query(port, dpif_port);
544 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
545 struct dpif_port *dpif_port)
547 struct dp_netdev *dp = get_dp_netdev(dpif);
548 struct dp_netdev_port *port;
551 error = get_port_by_name(dp, devname, &port);
552 if (!error && dpif_port) {
553 answer_port_query(port, dpif_port);
559 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
565 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
567 hmap_remove(&dp->flow_table, &flow->node);
573 dp_netdev_flow_flush(struct dp_netdev *dp)
575 struct dp_netdev_flow *flow, *next;
577 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
578 dp_netdev_free_flow(dp, flow);
583 dpif_netdev_flow_flush(struct dpif *dpif)
585 struct dp_netdev *dp = get_dp_netdev(dpif);
586 dp_netdev_flow_flush(dp);
590 struct dp_netdev_port_state {
596 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
598 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
603 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
604 struct dpif_port *dpif_port)
606 struct dp_netdev_port_state *state = state_;
607 struct dp_netdev *dp = get_dp_netdev(dpif);
610 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
611 struct dp_netdev_port *port = dp->ports[port_no];
614 state->name = xstrdup(netdev_get_name(port->netdev));
615 dpif_port->name = state->name;
616 dpif_port->type = port->type;
617 dpif_port->port_no = port->port_no;
618 state->port_no = port_no + 1;
626 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
628 struct dp_netdev_port_state *state = state_;
635 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
637 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
638 if (dpif->dp_serial != dpif->dp->serial) {
639 dpif->dp_serial = dpif->dp->serial;
647 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
649 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
650 if (dpif->dp_serial != dpif->dp->serial) {
651 poll_immediate_wake();
655 static struct dp_netdev_flow *
656 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
658 struct dp_netdev_flow *flow;
660 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
661 if (flow_equal(&flow->key, key)) {
669 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
671 stats->n_packets = flow->packet_count;
672 stats->n_bytes = flow->byte_count;
673 stats->used = flow->used;
674 stats->tcp_flags = flow->tcp_flags;
678 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
681 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
682 /* This should not happen: it indicates that odp_flow_key_from_flow()
683 * and odp_flow_key_to_flow() disagree on the acceptable form of a
684 * flow. Log the problem as an error, with enough details to enable
686 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
688 if (!VLOG_DROP_ERR(&rl)) {
692 odp_flow_key_format(key, key_len, &s);
693 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
700 if (flow->in_port < OFPP_MAX
701 ? flow->in_port >= MAX_PORTS
702 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
710 dpif_netdev_flow_get(const struct dpif *dpif,
711 const struct nlattr *nl_key, size_t nl_key_len,
712 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
714 struct dp_netdev *dp = get_dp_netdev(dpif);
715 struct dp_netdev_flow *flow;
719 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
724 flow = dp_netdev_lookup_flow(dp, &key);
730 get_dpif_flow_stats(flow, stats);
733 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
739 set_flow_actions(struct dp_netdev_flow *flow,
740 const struct nlattr *actions, size_t actions_len)
742 flow->actions = xrealloc(flow->actions, actions_len);
743 flow->actions_len = actions_len;
744 memcpy(flow->actions, actions, actions_len);
749 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
750 const struct nlattr *actions, size_t actions_len)
752 struct dp_netdev_flow *flow;
755 flow = xzalloc(sizeof *flow);
758 error = set_flow_actions(flow, actions, actions_len);
764 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
769 clear_stats(struct dp_netdev_flow *flow)
772 flow->packet_count = 0;
773 flow->byte_count = 0;
778 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
780 struct dp_netdev *dp = get_dp_netdev(dpif);
781 struct dp_netdev_flow *flow;
785 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
790 flow = dp_netdev_lookup_flow(dp, &key);
792 if (put->flags & DPIF_FP_CREATE) {
793 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
795 memset(put->stats, 0, sizeof *put->stats);
797 return dp_netdev_flow_add(dp, &key, put->actions,
806 if (put->flags & DPIF_FP_MODIFY) {
807 int error = set_flow_actions(flow, put->actions, put->actions_len);
810 get_dpif_flow_stats(flow, put->stats);
812 if (put->flags & DPIF_FP_ZERO_STATS) {
824 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
826 struct dp_netdev *dp = get_dp_netdev(dpif);
827 struct dp_netdev_flow *flow;
831 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
836 flow = dp_netdev_lookup_flow(dp, &key);
839 get_dpif_flow_stats(flow, del->stats);
841 dp_netdev_free_flow(dp, flow);
848 struct dp_netdev_flow_state {
851 struct nlattr *actions;
852 struct odputil_keybuf keybuf;
853 struct dpif_flow_stats stats;
857 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
859 struct dp_netdev_flow_state *state;
861 *statep = state = xmalloc(sizeof *state);
864 state->actions = NULL;
869 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
870 const struct nlattr **key, size_t *key_len,
871 const struct nlattr **actions, size_t *actions_len,
872 const struct dpif_flow_stats **stats)
874 struct dp_netdev_flow_state *state = state_;
875 struct dp_netdev *dp = get_dp_netdev(dpif);
876 struct dp_netdev_flow *flow;
877 struct hmap_node *node;
879 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
884 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
889 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
890 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port);
897 free(state->actions);
898 state->actions = xmemdup(flow->actions, flow->actions_len);
900 *actions = state->actions;
901 *actions_len = flow->actions_len;
905 get_dpif_flow_stats(flow, &state->stats);
906 *stats = &state->stats;
913 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
915 struct dp_netdev_flow_state *state = state_;
917 free(state->actions);
923 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
925 struct dp_netdev *dp = get_dp_netdev(dpif);
930 if (execute->packet->size < ETH_HEADER_LEN ||
931 execute->packet->size > UINT16_MAX) {
935 /* Make a deep copy of 'packet', because we might modify its data. */
936 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
937 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
938 ofpbuf_put(©, execute->packet->data, execute->packet->size);
940 flow_extract(©, 0, 0, NULL, -1, &key);
941 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
944 dp_netdev_execute_actions(dp, ©, &key,
945 execute->actions, execute->actions_len);
948 ofpbuf_uninit(©);
953 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
959 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
960 uint32_t queue_id, uint32_t *priority)
962 *priority = queue_id;
966 static struct dp_netdev_queue *
967 find_nonempty_queue(struct dpif *dpif)
969 struct dp_netdev *dp = get_dp_netdev(dpif);
972 for (i = 0; i < N_QUEUES; i++) {
973 struct dp_netdev_queue *q = &dp->queues[i];
974 if (q->head != q->tail) {
982 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
985 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
987 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
990 upcall->packet = buf;
1002 dpif_netdev_recv_wait(struct dpif *dpif)
1004 if (find_nonempty_queue(dpif)) {
1005 poll_immediate_wake();
1007 /* No messages ready to be received, and dp_wait() will ensure that we
1008 * wake up to queue new messages, so there is nothing to do. */
1013 dpif_netdev_recv_purge(struct dpif *dpif)
1015 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1016 dp_netdev_purge_queues(dpif_netdev->dp);
1020 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1022 flow->used = time_msec();
1023 flow->packet_count++;
1024 flow->byte_count += packet->size;
1025 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1029 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1030 struct ofpbuf *packet)
1032 struct dp_netdev_flow *flow;
1035 if (packet->size < ETH_HEADER_LEN) {
1038 flow_extract(packet, 0, 0, NULL, port->port_no, &key);
1039 flow = dp_netdev_lookup_flow(dp, &key);
1041 dp_netdev_flow_used(flow, packet);
1042 dp_netdev_execute_actions(dp, packet, &key,
1043 flow->actions, flow->actions_len);
1047 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1052 dpif_netdev_run(struct dpif *dpif)
1054 struct dp_netdev *dp = get_dp_netdev(dpif);
1055 struct dp_netdev_port *port;
1056 struct ofpbuf packet;
1058 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1060 LIST_FOR_EACH (port, node, &dp->port_list) {
1063 /* Reset packet contents. */
1064 ofpbuf_clear(&packet);
1065 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1067 error = netdev_recv(port->netdev, &packet);
1069 dp_netdev_port_input(dp, port, &packet);
1070 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1071 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1072 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1073 netdev_get_name(port->netdev), strerror(error));
1076 ofpbuf_uninit(&packet);
1080 dpif_netdev_wait(struct dpif *dpif)
1082 struct dp_netdev *dp = get_dp_netdev(dpif);
1083 struct dp_netdev_port *port;
1085 LIST_FOR_EACH (port, node, &dp->port_list) {
1086 netdev_recv_wait(port->netdev);
1091 dp_netdev_set_dl(struct ofpbuf *packet, const struct ovs_key_ethernet *eth_key)
1093 struct eth_header *eh = packet->l2;
1095 memcpy(eh->eth_src, eth_key->eth_src, sizeof eh->eth_src);
1096 memcpy(eh->eth_dst, eth_key->eth_dst, sizeof eh->eth_dst);
1100 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1103 struct dp_netdev_port *p = dp->ports[out_port];
1105 netdev_send(p->netdev, packet);
1110 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1111 int queue_no, const struct flow *flow, uint64_t arg)
1113 struct dp_netdev_queue *q = &dp->queues[queue_no];
1114 struct dp_netdev_upcall *u;
1115 struct dpif_upcall *upcall;
1119 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1124 u = &q->upcalls[q->head++ & QUEUE_MASK];
1127 ofpbuf_init(buf, ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1128 odp_flow_key_from_flow(buf, flow, flow->in_port);
1129 key_len = buf->size;
1130 ofpbuf_pull(buf, key_len);
1131 ofpbuf_reserve(buf, 2);
1132 ofpbuf_put(buf, packet->data, packet->size);
1134 upcall = &u->upcall;
1135 upcall->type = queue_no;
1136 upcall->packet = buf;
1137 upcall->key = buf->base;
1138 upcall->key_len = key_len;
1139 upcall->userdata = arg;
1145 dp_netdev_sample(struct dp_netdev *dp,
1146 struct ofpbuf *packet, struct flow *key,
1147 const struct nlattr *action)
1149 const struct nlattr *subactions = NULL;
1150 const struct nlattr *a;
1153 NL_NESTED_FOR_EACH_UNSAFE (a, left, action) {
1154 int type = nl_attr_type(a);
1156 switch ((enum ovs_sample_attr) type) {
1157 case OVS_SAMPLE_ATTR_PROBABILITY:
1158 if (random_uint32() >= nl_attr_get_u32(a)) {
1163 case OVS_SAMPLE_ATTR_ACTIONS:
1167 case OVS_SAMPLE_ATTR_UNSPEC:
1168 case __OVS_SAMPLE_ATTR_MAX:
1174 dp_netdev_execute_actions(dp, packet, key, nl_attr_get(subactions),
1175 nl_attr_get_size(subactions));
1179 dp_netdev_action_userspace(struct dp_netdev *dp,
1180 struct ofpbuf *packet, struct flow *key,
1181 const struct nlattr *a)
1183 const struct nlattr *userdata_attr;
1186 userdata_attr = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1187 userdata = userdata_attr ? nl_attr_get_u64(userdata_attr) : 0;
1188 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1192 execute_set_action(struct ofpbuf *packet, const struct nlattr *a)
1194 enum ovs_key_attr type = nl_attr_type(a);
1195 const struct ovs_key_ipv4 *ipv4_key;
1196 const struct ovs_key_ipv6 *ipv6_key;
1197 const struct ovs_key_tcp *tcp_key;
1198 const struct ovs_key_udp *udp_key;
1201 case OVS_KEY_ATTR_TUN_ID:
1202 case OVS_KEY_ATTR_PRIORITY:
1203 case OVS_KEY_ATTR_SKB_MARK:
1204 case OVS_KEY_ATTR_TUNNEL:
1205 /* not implemented */
1208 case OVS_KEY_ATTR_ETHERNET:
1209 dp_netdev_set_dl(packet,
1210 nl_attr_get_unspec(a, sizeof(struct ovs_key_ethernet)));
1213 case OVS_KEY_ATTR_IPV4:
1214 ipv4_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv4));
1215 packet_set_ipv4(packet, ipv4_key->ipv4_src, ipv4_key->ipv4_dst,
1216 ipv4_key->ipv4_tos, ipv4_key->ipv4_ttl);
1219 case OVS_KEY_ATTR_IPV6:
1220 ipv6_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_ipv6));
1221 packet_set_ipv6(packet, ipv6_key->ipv6_proto, ipv6_key->ipv6_src,
1222 ipv6_key->ipv6_dst, ipv6_key->ipv6_tclass,
1223 ipv6_key->ipv6_label, ipv6_key->ipv6_hlimit);
1226 case OVS_KEY_ATTR_TCP:
1227 tcp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_tcp));
1228 packet_set_tcp_port(packet, tcp_key->tcp_src, tcp_key->tcp_dst);
1231 case OVS_KEY_ATTR_UDP:
1232 udp_key = nl_attr_get_unspec(a, sizeof(struct ovs_key_udp));
1233 packet_set_udp_port(packet, udp_key->udp_src, udp_key->udp_dst);
1236 case OVS_KEY_ATTR_UNSPEC:
1237 case OVS_KEY_ATTR_ENCAP:
1238 case OVS_KEY_ATTR_ETHERTYPE:
1239 case OVS_KEY_ATTR_IN_PORT:
1240 case OVS_KEY_ATTR_VLAN:
1241 case OVS_KEY_ATTR_ICMP:
1242 case OVS_KEY_ATTR_ICMPV6:
1243 case OVS_KEY_ATTR_ARP:
1244 case OVS_KEY_ATTR_ND:
1245 case __OVS_KEY_ATTR_MAX:
1252 dp_netdev_execute_actions(struct dp_netdev *dp,
1253 struct ofpbuf *packet, struct flow *key,
1254 const struct nlattr *actions,
1257 const struct nlattr *a;
1260 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1261 int type = nl_attr_type(a);
1263 switch ((enum ovs_action_attr) type) {
1264 case OVS_ACTION_ATTR_OUTPUT:
1265 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1268 case OVS_ACTION_ATTR_USERSPACE:
1269 dp_netdev_action_userspace(dp, packet, key, a);
1272 case OVS_ACTION_ATTR_PUSH_VLAN: {
1273 const struct ovs_action_push_vlan *vlan = nl_attr_get(a);
1274 eth_push_vlan(packet, vlan->vlan_tci);
1278 case OVS_ACTION_ATTR_POP_VLAN:
1279 eth_pop_vlan(packet);
1282 case OVS_ACTION_ATTR_SET:
1283 execute_set_action(packet, nl_attr_get(a));
1286 case OVS_ACTION_ATTR_SAMPLE:
1287 dp_netdev_sample(dp, packet, key, a);
1290 case OVS_ACTION_ATTR_UNSPEC:
1291 case __OVS_ACTION_ATTR_MAX:
1297 #define DPIF_NETDEV_CLASS_FUNCTIONS(PORT_OPEN_TYPE) \
1298 dpif_netdev_enumerate, \
1301 dpif_netdev_close, \
1302 dpif_netdev_destroy, \
1305 dpif_netdev_get_stats, \
1306 dpif_netdev_port_add, \
1307 dpif_netdev_port_del, \
1308 dpif_netdev_port_query_by_number, \
1309 dpif_netdev_port_query_by_name, \
1310 dpif_netdev_get_max_ports, \
1311 NULL, /* port_get_pid */ \
1312 dpif_netdev_port_dump_start, \
1313 dpif_netdev_port_dump_next, \
1314 dpif_netdev_port_dump_done, \
1315 dpif_netdev_port_poll, \
1316 dpif_netdev_port_poll_wait, \
1317 dpif_netdev_flow_get, \
1318 dpif_netdev_flow_put, \
1319 dpif_netdev_flow_del, \
1320 dpif_netdev_flow_flush, \
1321 dpif_netdev_flow_dump_start, \
1322 dpif_netdev_flow_dump_next, \
1323 dpif_netdev_flow_dump_done, \
1324 dpif_netdev_execute, \
1325 NULL, /* operate */ \
1326 dpif_netdev_recv_set, \
1327 dpif_netdev_queue_to_priority, \
1329 dpif_netdev_recv_wait, \
1330 dpif_netdev_recv_purge, \
1332 const struct dpif_class dpif_netdev_class = {
1334 DPIF_NETDEV_CLASS_FUNCTIONS(dpif_netdev_port_open_type)
1337 const struct dpif_class dpif_planetlab_class = {
1339 DPIF_NETDEV_CLASS_FUNCTIONS(dpif_planetlab_port_open_type)
1343 dpif_dummy_register__(const char *type)
1345 struct dpif_class *class;
1347 class = xmalloc(sizeof *class);
1348 *class = dpif_netdev_class;
1349 class->type = xstrdup(type);
1350 dp_register_provider(class);
1354 dpif_dummy_register(bool override)
1361 dp_enumerate_types(&types);
1362 SSET_FOR_EACH (type, &types) {
1363 if (!dp_unregister_provider(type)) {
1364 dpif_dummy_register__(type);
1367 sset_destroy(&types);
1370 dpif_dummy_register__("dummy");