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_class_is_planetlab(const struct dpif_class *class)
198 return class == &dpif_planetlab_class;
202 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
204 return strcmp(type, "internal") ? type
205 : dpif_netdev_class_is_planetlab(class) ? "pltap"
206 : dpif_netdev_class_is_dummy(class) ? "dummy"
211 create_dpif_netdev(struct dp_netdev *dp)
213 uint16_t netflow_id = hash_string(dp->name, 0);
214 struct dpif_netdev *dpif;
218 dpif = xmalloc(sizeof *dpif);
219 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
221 dpif->dp_serial = dp->serial;
226 /* Choose an unused, non-zero port number and return it on success.
227 * Return ODPP_NONE on failure. */
229 choose_port(struct dp_netdev *dp, const char *name)
233 if (dp->class != &dpif_netdev_class &&
234 dp->class != &dpif_planetlab_class) {
238 /* If the port name begins with "br", start the number search at
239 * 100 to make writing tests easier. */
240 if (!strncmp(name, "br", 2)) {
244 /* If the port name contains a number, try to assign that port number.
245 * This can make writing unit tests easier because port numbers are
247 for (p = name; *p != '\0'; p++) {
248 if (isdigit((unsigned char) *p)) {
249 port_no = start_no + strtol(p, NULL, 10);
250 if (port_no > 0 && port_no < MAX_PORTS
251 && !dp->ports[port_no]) {
252 return u32_to_odp(port_no);
259 for (port_no = 1; port_no < MAX_PORTS; port_no++) {
260 if (!dp->ports[port_no]) {
261 return u32_to_odp(port_no);
269 create_dp_netdev(const char *name, const struct dpif_class *class,
270 struct dp_netdev **dpp)
272 struct dp_netdev *dp;
276 dp = xzalloc(sizeof *dp);
278 dp->name = xstrdup(name);
280 dp->max_mtu = ETH_PAYLOAD_MAX;
281 for (i = 0; i < N_QUEUES; i++) {
282 dp->queues[i].head = dp->queues[i].tail = 0;
284 hmap_init(&dp->flow_table);
285 list_init(&dp->port_list);
287 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
293 shash_add(&dp_netdevs, name, dp);
300 dpif_netdev_open(const struct dpif_class *class, const char *name,
301 bool create, struct dpif **dpifp)
303 struct dp_netdev *dp;
305 dp = shash_find_data(&dp_netdevs, name);
310 int error = create_dp_netdev(name, class, &dp);
314 ovs_assert(dp != NULL);
317 if (dp->class != class) {
324 *dpifp = create_dpif_netdev(dp);
329 dp_netdev_purge_queues(struct dp_netdev *dp)
333 for (i = 0; i < N_QUEUES; i++) {
334 struct dp_netdev_queue *q = &dp->queues[i];
336 while (q->tail != q->head) {
337 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
338 ofpbuf_uninit(&u->buf);
344 dp_netdev_free(struct dp_netdev *dp)
346 struct dp_netdev_port *port, *next;
348 dp_netdev_flow_flush(dp);
349 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
350 do_del_port(dp, port->port_no);
352 dp_netdev_purge_queues(dp);
353 hmap_destroy(&dp->flow_table);
359 dpif_netdev_close(struct dpif *dpif)
361 struct dp_netdev *dp = get_dp_netdev(dpif);
362 ovs_assert(dp->open_cnt > 0);
363 if (--dp->open_cnt == 0 && dp->destroyed) {
364 shash_find_and_delete(&dp_netdevs, dp->name);
371 dpif_netdev_destroy(struct dpif *dpif)
373 struct dp_netdev *dp = get_dp_netdev(dpif);
374 dp->destroyed = true;
379 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
381 struct dp_netdev *dp = get_dp_netdev(dpif);
382 stats->n_flows = hmap_count(&dp->flow_table);
383 stats->n_hit = dp->n_hit;
384 stats->n_missed = dp->n_missed;
385 stats->n_lost = dp->n_lost;
390 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
393 struct netdev_saved_flags *sf;
394 struct dp_netdev_port *port;
395 struct netdev *netdev;
396 struct netdev_rx *rx;
397 const char *open_type;
401 /* XXX reject devices already in some dp_netdev. */
403 /* Open and validate network device. */
404 open_type = dpif_netdev_port_open_type(dp->class, type);
405 error = netdev_open(devname, open_type, &netdev);
409 /* XXX reject loopback devices */
410 /* XXX reject non-Ethernet devices */
412 error = netdev_rx_open(netdev, &rx);
414 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
415 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
416 devname, ovs_strerror(errno));
417 netdev_close(netdev);
421 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
424 netdev_close(netdev);
428 port = xmalloc(sizeof *port);
429 port->port_no = port_no;
430 port->netdev = netdev;
433 port->type = xstrdup(type);
435 error = netdev_get_mtu(netdev, &mtu);
436 if (!error && mtu > dp->max_mtu) {
440 list_push_back(&dp->port_list, &port->node);
441 dp->ports[odp_to_u32(port_no)] = port;
448 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
449 odp_port_t *port_nop)
451 struct dp_netdev *dp = get_dp_netdev(dpif);
452 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
453 const char *dpif_port;
456 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
457 if (*port_nop != ODPP_NONE) {
458 uint32_t port_idx = odp_to_u32(*port_nop);
459 if (port_idx >= MAX_PORTS) {
461 } else if (dp->ports[port_idx]) {
466 port_no = choose_port(dp, dpif_port);
468 if (port_no != ODPP_NONE) {
470 return do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
476 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
478 struct dp_netdev *dp = get_dp_netdev(dpif);
479 return (port_no == ODPP_LOCAL ?
480 EINVAL : do_del_port(dp, port_no));
484 is_valid_port_number(odp_port_t port_no)
486 return odp_to_u32(port_no) < MAX_PORTS;
490 get_port_by_number(struct dp_netdev *dp,
491 odp_port_t port_no, struct dp_netdev_port **portp)
493 if (!is_valid_port_number(port_no)) {
497 *portp = dp->ports[odp_to_u32(port_no)];
498 return *portp ? 0 : ENOENT;
503 get_port_by_name(struct dp_netdev *dp,
504 const char *devname, struct dp_netdev_port **portp)
506 struct dp_netdev_port *port;
508 LIST_FOR_EACH (port, node, &dp->port_list) {
509 if (!strcmp(netdev_get_name(port->netdev), devname)) {
518 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
520 struct dp_netdev_port *port;
523 error = get_port_by_number(dp, port_no, &port);
528 list_remove(&port->node);
529 dp->ports[odp_to_u32(port_no)] = NULL;
532 netdev_close(port->netdev);
533 netdev_restore_flags(port->sf);
534 netdev_rx_close(port->rx);
542 answer_port_query(const struct dp_netdev_port *port,
543 struct dpif_port *dpif_port)
545 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
546 dpif_port->type = xstrdup(port->type);
547 dpif_port->port_no = port->port_no;
551 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
552 struct dpif_port *dpif_port)
554 struct dp_netdev *dp = get_dp_netdev(dpif);
555 struct dp_netdev_port *port;
558 error = get_port_by_number(dp, port_no, &port);
559 if (!error && dpif_port) {
560 answer_port_query(port, dpif_port);
566 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
567 struct dpif_port *dpif_port)
569 struct dp_netdev *dp = get_dp_netdev(dpif);
570 struct dp_netdev_port *port;
573 error = get_port_by_name(dp, devname, &port);
574 if (!error && dpif_port) {
575 answer_port_query(port, dpif_port);
581 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
583 return u32_to_odp(MAX_PORTS);
587 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
589 hmap_remove(&dp->flow_table, &flow->node);
595 dp_netdev_flow_flush(struct dp_netdev *dp)
597 struct dp_netdev_flow *flow, *next;
599 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
600 dp_netdev_free_flow(dp, flow);
605 dpif_netdev_flow_flush(struct dpif *dpif)
607 struct dp_netdev *dp = get_dp_netdev(dpif);
608 dp_netdev_flow_flush(dp);
612 struct dp_netdev_port_state {
618 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
620 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
625 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
626 struct dpif_port *dpif_port)
628 struct dp_netdev_port_state *state = state_;
629 struct dp_netdev *dp = get_dp_netdev(dpif);
632 for (port_idx = odp_to_u32(state->port_no);
633 port_idx < MAX_PORTS; port_idx++) {
634 struct dp_netdev_port *port = dp->ports[port_idx];
637 state->name = xstrdup(netdev_get_name(port->netdev));
638 dpif_port->name = state->name;
639 dpif_port->type = port->type;
640 dpif_port->port_no = port->port_no;
641 state->port_no = u32_to_odp(port_idx + 1);
649 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
651 struct dp_netdev_port_state *state = state_;
658 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
660 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
661 if (dpif->dp_serial != dpif->dp->serial) {
662 dpif->dp_serial = dpif->dp->serial;
670 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
672 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
673 if (dpif->dp_serial != dpif->dp->serial) {
674 poll_immediate_wake();
678 static struct dp_netdev_flow *
679 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
681 struct dp_netdev_flow *flow;
683 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
684 if (flow_equal(&flow->key, key)) {
692 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
694 stats->n_packets = flow->packet_count;
695 stats->n_bytes = flow->byte_count;
696 stats->used = flow->used;
697 stats->tcp_flags = flow->tcp_flags;
701 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
706 if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
707 /* This should not happen: it indicates that odp_flow_key_from_flow()
708 * and odp_flow_key_to_flow() disagree on the acceptable form of a
709 * flow. Log the problem as an error, with enough details to enable
711 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
713 if (!VLOG_DROP_ERR(&rl)) {
717 odp_flow_key_format(key, key_len, &s);
718 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
725 in_port = flow->in_port.odp_port;
726 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
734 dpif_netdev_flow_get(const struct dpif *dpif,
735 const struct nlattr *nl_key, size_t nl_key_len,
736 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
738 struct dp_netdev *dp = get_dp_netdev(dpif);
739 struct dp_netdev_flow *flow;
743 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
748 flow = dp_netdev_lookup_flow(dp, &key);
754 get_dpif_flow_stats(flow, stats);
757 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
763 set_flow_actions(struct dp_netdev_flow *flow,
764 const struct nlattr *actions, size_t actions_len)
766 flow->actions = xrealloc(flow->actions, actions_len);
767 flow->actions_len = actions_len;
768 memcpy(flow->actions, actions, actions_len);
773 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *key,
774 const struct nlattr *actions, size_t actions_len)
776 struct dp_netdev_flow *flow;
779 flow = xzalloc(sizeof *flow);
782 error = set_flow_actions(flow, actions, actions_len);
788 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
793 clear_stats(struct dp_netdev_flow *flow)
796 flow->packet_count = 0;
797 flow->byte_count = 0;
802 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
804 struct dp_netdev *dp = get_dp_netdev(dpif);
805 struct dp_netdev_flow *flow;
809 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &key);
814 flow = dp_netdev_lookup_flow(dp, &key);
816 if (put->flags & DPIF_FP_CREATE) {
817 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
819 memset(put->stats, 0, sizeof *put->stats);
821 return dp_netdev_flow_add(dp, &key, put->actions,
830 if (put->flags & DPIF_FP_MODIFY) {
831 int error = set_flow_actions(flow, put->actions, put->actions_len);
834 get_dpif_flow_stats(flow, put->stats);
836 if (put->flags & DPIF_FP_ZERO_STATS) {
848 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
850 struct dp_netdev *dp = get_dp_netdev(dpif);
851 struct dp_netdev_flow *flow;
855 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
860 flow = dp_netdev_lookup_flow(dp, &key);
863 get_dpif_flow_stats(flow, del->stats);
865 dp_netdev_free_flow(dp, flow);
872 struct dp_netdev_flow_state {
875 struct nlattr *actions;
876 struct odputil_keybuf keybuf;
877 struct dpif_flow_stats stats;
881 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
883 struct dp_netdev_flow_state *state;
885 *statep = state = xmalloc(sizeof *state);
888 state->actions = NULL;
893 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
894 const struct nlattr **key, size_t *key_len,
895 const struct nlattr **mask, size_t *mask_len,
896 const struct nlattr **actions, size_t *actions_len,
897 const struct dpif_flow_stats **stats)
899 struct dp_netdev_flow_state *state = state_;
900 struct dp_netdev *dp = get_dp_netdev(dpif);
901 struct dp_netdev_flow *flow;
902 struct hmap_node *node;
904 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
909 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
914 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
915 odp_flow_key_from_flow(&buf, &flow->key, flow->key.in_port.odp_port);
927 free(state->actions);
928 state->actions = xmemdup(flow->actions, flow->actions_len);
930 *actions = state->actions;
931 *actions_len = flow->actions_len;
935 get_dpif_flow_stats(flow, &state->stats);
936 *stats = &state->stats;
943 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
945 struct dp_netdev_flow_state *state = state_;
947 free(state->actions);
953 dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
955 struct dp_netdev *dp = get_dp_netdev(dpif);
960 if (execute->packet->size < ETH_HEADER_LEN ||
961 execute->packet->size > UINT16_MAX) {
965 /* Make a deep copy of 'packet', because we might modify its data. */
966 ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
967 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
968 ofpbuf_put(©, execute->packet->data, execute->packet->size);
970 flow_extract(©, 0, 0, NULL, NULL, &key);
971 error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
974 dp_netdev_execute_actions(dp, ©, &key,
975 execute->actions, execute->actions_len);
978 ofpbuf_uninit(©);
983 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
989 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
990 uint32_t queue_id, uint32_t *priority)
992 *priority = queue_id;
996 static struct dp_netdev_queue *
997 find_nonempty_queue(struct dpif *dpif)
999 struct dp_netdev *dp = get_dp_netdev(dpif);
1002 for (i = 0; i < N_QUEUES; i++) {
1003 struct dp_netdev_queue *q = &dp->queues[i];
1004 if (q->head != q->tail) {
1012 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1015 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1017 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1019 *upcall = u->upcall;
1020 upcall->packet = buf;
1032 dpif_netdev_recv_wait(struct dpif *dpif)
1034 if (find_nonempty_queue(dpif)) {
1035 poll_immediate_wake();
1037 /* No messages ready to be received, and dp_wait() will ensure that we
1038 * wake up to queue new messages, so there is nothing to do. */
1043 dpif_netdev_recv_purge(struct dpif *dpif)
1045 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1046 dp_netdev_purge_queues(dpif_netdev->dp);
1050 dp_netdev_flow_used(struct dp_netdev_flow *flow, const struct ofpbuf *packet)
1052 flow->used = time_msec();
1053 flow->packet_count++;
1054 flow->byte_count += packet->size;
1055 flow->tcp_flags |= packet_get_tcp_flags(packet, &flow->key);
1059 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1060 struct ofpbuf *packet, uint32_t skb_priority,
1061 uint32_t skb_mark, const struct flow_tnl *tnl)
1063 struct dp_netdev_flow *flow;
1065 union flow_in_port in_port_;
1067 if (packet->size < ETH_HEADER_LEN) {
1070 in_port_.odp_port = port->port_no;
1071 flow_extract(packet, skb_priority, skb_mark, tnl, &in_port_, &key);
1072 flow = dp_netdev_lookup_flow(dp, &key);
1074 dp_netdev_flow_used(flow, packet);
1075 dp_netdev_execute_actions(dp, packet, &key,
1076 flow->actions, flow->actions_len);
1080 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1085 dpif_netdev_run(struct dpif *dpif)
1087 struct dp_netdev *dp = get_dp_netdev(dpif);
1088 struct dp_netdev_port *port;
1089 struct ofpbuf packet;
1091 ofpbuf_init(&packet,
1092 DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
1094 LIST_FOR_EACH (port, node, &dp->port_list) {
1097 /* Reset packet contents. */
1098 ofpbuf_clear(&packet);
1099 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1101 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1103 dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
1104 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1105 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1107 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1108 netdev_get_name(port->netdev), ovs_strerror(error));
1111 ofpbuf_uninit(&packet);
1115 dpif_netdev_wait(struct dpif *dpif)
1117 struct dp_netdev *dp = get_dp_netdev(dpif);
1118 struct dp_netdev_port *port;
1120 LIST_FOR_EACH (port, node, &dp->port_list) {
1122 netdev_rx_wait(port->rx);
1128 dp_netdev_output_port(void *dp_, struct ofpbuf *packet, uint32_t out_port)
1130 struct dp_netdev *dp = dp_;
1131 struct dp_netdev_port *p = dp->ports[out_port];
1133 netdev_send(p->netdev, packet);
1138 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1139 int queue_no, const struct flow *flow,
1140 const struct nlattr *userdata)
1142 struct dp_netdev_queue *q = &dp->queues[queue_no];
1143 if (q->head - q->tail < MAX_QUEUE_LEN) {
1144 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1145 struct dpif_upcall *upcall = &u->upcall;
1146 struct ofpbuf *buf = &u->buf;
1149 upcall->type = queue_no;
1151 /* Allocate buffer big enough for everything. */
1152 buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
1154 buf_size += NLA_ALIGN(userdata->nla_len);
1156 ofpbuf_init(buf, buf_size);
1159 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1160 upcall->key = buf->data;
1161 upcall->key_len = buf->size;
1165 upcall->userdata = ofpbuf_put(buf, userdata,
1166 NLA_ALIGN(userdata->nla_len));
1171 * We adjust 'data' and 'size' in 'buf' so that only the packet itself
1172 * is visible in 'upcall->packet'. The ODP flow and (if present)
1173 * userdata become part of the headroom. */
1174 ofpbuf_put_zeros(buf, 2);
1175 buf->data = ofpbuf_put(buf, packet->data, packet->size);
1176 buf->size = packet->size;
1177 upcall->packet = buf;
1187 dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
1188 const struct flow *key,
1189 const struct nlattr *userdata)
1191 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
1195 dp_netdev_execute_actions(struct dp_netdev *dp,
1196 struct ofpbuf *packet, struct flow *key,
1197 const struct nlattr *actions,
1200 odp_execute_actions(dp, packet, key, actions, actions_len,
1201 dp_netdev_output_port, dp_netdev_action_userspace);
1204 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1205 dpif_netdev_enumerate, \
1206 dpif_netdev_port_open_type, \
1208 dpif_netdev_close, \
1209 dpif_netdev_destroy, \
1212 dpif_netdev_get_stats, \
1213 dpif_netdev_port_add, \
1214 dpif_netdev_port_del, \
1215 dpif_netdev_port_query_by_number, \
1216 dpif_netdev_port_query_by_name, \
1217 dpif_netdev_get_max_ports, \
1218 NULL, /* port_get_pid */ \
1219 dpif_netdev_port_dump_start, \
1220 dpif_netdev_port_dump_next, \
1221 dpif_netdev_port_dump_done, \
1222 dpif_netdev_port_poll, \
1223 dpif_netdev_port_poll_wait, \
1224 dpif_netdev_flow_get, \
1225 dpif_netdev_flow_put, \
1226 dpif_netdev_flow_del, \
1227 dpif_netdev_flow_flush, \
1228 dpif_netdev_flow_dump_start, \
1229 dpif_netdev_flow_dump_next, \
1230 dpif_netdev_flow_dump_done, \
1231 dpif_netdev_execute, \
1232 NULL, /* operate */ \
1233 dpif_netdev_recv_set, \
1234 dpif_netdev_queue_to_priority, \
1236 dpif_netdev_recv_wait, \
1237 dpif_netdev_recv_purge, \
1239 const struct dpif_class dpif_netdev_class = {
1241 DPIF_NETDEV_CLASS_FUNCTIONS
1244 const struct dpif_class dpif_planetlab_class = {
1246 DPIF_NETDEV_CLASS_FUNCTIONS
1250 dpif_dummy_register__(const char *type)
1252 struct dpif_class *class;
1254 class = xmalloc(sizeof *class);
1255 *class = dpif_netdev_class;
1256 class->type = xstrdup(type);
1257 dp_register_provider(class);
1261 dpif_dummy_register(bool override)
1268 dp_enumerate_types(&types);
1269 SSET_FOR_EACH (type, &types) {
1270 if (!dp_unregister_provider(type)) {
1271 dpif_dummy_register__(type);
1274 sset_destroy(&types);
1277 dpif_dummy_register__("dummy");