2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
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.
25 #include <netinet/in.h>
26 #include <sys/socket.h>
31 #include <sys/ioctl.h>
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
46 #include "ofp-print.h"
49 #include "poll-loop.h"
55 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
57 /* Configuration parameters. */
58 enum { MAX_PORTS = 256 }; /* Maximum number of ports. */
59 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
61 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
62 * headers to be aligned on a 4-byte boundary. */
63 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
66 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
67 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
68 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
69 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
71 struct dp_netdev_queue {
72 struct dpif_upcall *upcalls[MAX_QUEUE_LEN];
73 unsigned int head, tail;
76 /* Datapath based on the network device interface from netdev.h. */
78 const struct dpif_class *class;
83 bool drop_frags; /* Drop all IP fragments, if true. */
84 struct dp_netdev_queue queues[N_QUEUES];
85 struct hmap flow_table; /* Flow table. */
88 long long int n_frags; /* Number of dropped IP fragments. */
89 long long int n_hit; /* Number of flow table matches. */
90 long long int n_missed; /* Number of flow table misses. */
91 long long int n_lost; /* Number of misses not passed to client. */
94 struct dp_netdev_port *ports[MAX_PORTS];
95 struct list port_list;
99 /* A port in a netdev-based datapath. */
100 struct dp_netdev_port {
101 int port_no; /* Index into dp_netdev's 'ports'. */
102 struct list node; /* Element in dp_netdev's 'port_list'. */
103 struct netdev *netdev;
104 bool internal; /* Internal port? */
107 /* A flow in dp_netdev's 'flow_table'. */
108 struct dp_netdev_flow {
109 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
113 long long int used; /* Last used time, in monotonic msecs. */
114 long long int packet_count; /* Number of packets matched. */
115 long long int byte_count; /* Number of bytes matched. */
116 ovs_be16 tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */
119 struct nlattr *actions;
123 /* Interface to netdev-based datapath. */
126 struct dp_netdev *dp;
128 unsigned int dp_serial;
131 /* All netdev-based datapaths. */
132 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
134 /* Maximum port MTU seen so far. */
135 static int max_mtu = ETH_PAYLOAD_MAX;
137 static int get_port_by_number(struct dp_netdev *, uint16_t port_no,
138 struct dp_netdev_port **portp);
139 static int get_port_by_name(struct dp_netdev *, const char *devname,
140 struct dp_netdev_port **portp);
141 static void dp_netdev_free(struct dp_netdev *);
142 static void dp_netdev_flow_flush(struct dp_netdev *);
143 static int do_add_port(struct dp_netdev *, const char *devname,
144 const char *type, uint16_t port_no);
145 static int do_del_port(struct dp_netdev *, uint16_t port_no);
146 static int dpif_netdev_open(const struct dpif_class *, const char *name,
147 bool create, struct dpif **);
148 static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
149 int queue_no, const struct flow *,
151 static int dp_netdev_execute_actions(struct dp_netdev *,
152 struct ofpbuf *, struct flow *,
153 const struct nlattr *actions,
156 static struct dpif_class dpif_dummy_class;
158 static struct dpif_netdev *
159 dpif_netdev_cast(const struct dpif *dpif)
161 assert(dpif->dpif_class->open == dpif_netdev_open);
162 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
165 static struct dp_netdev *
166 get_dp_netdev(const struct dpif *dpif)
168 return dpif_netdev_cast(dpif)->dp;
172 create_dpif_netdev(struct dp_netdev *dp)
174 uint16_t netflow_id = hash_string(dp->name, 0);
175 struct dpif_netdev *dpif;
179 dpif = xmalloc(sizeof *dpif);
180 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
182 dpif->listen_mask = 0;
183 dpif->dp_serial = dp->serial;
189 create_dp_netdev(const char *name, const struct dpif_class *class,
190 struct dp_netdev **dpp)
192 struct dp_netdev *dp;
196 dp = xzalloc(sizeof *dp);
198 dp->name = xstrdup(name);
200 dp->drop_frags = false;
201 for (i = 0; i < N_QUEUES; i++) {
202 dp->queues[i].head = dp->queues[i].tail = 0;
204 hmap_init(&dp->flow_table);
205 list_init(&dp->port_list);
206 error = do_add_port(dp, name, "internal", OVSP_LOCAL);
212 shash_add(&dp_netdevs, name, dp);
219 dpif_netdev_open(const struct dpif_class *class, const char *name,
220 bool create, struct dpif **dpifp)
222 struct dp_netdev *dp;
224 dp = shash_find_data(&dp_netdevs, name);
229 int error = create_dp_netdev(name, class, &dp);
236 if (dp->class != class) {
243 *dpifp = create_dpif_netdev(dp);
248 dp_netdev_purge_queues(struct dp_netdev *dp)
252 for (i = 0; i < N_QUEUES; i++) {
253 struct dp_netdev_queue *q = &dp->queues[i];
255 while (q->tail != q->head) {
256 struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK];
258 ofpbuf_delete(upcall->packet);
265 dp_netdev_free(struct dp_netdev *dp)
267 struct dp_netdev_port *port, *next;
269 dp_netdev_flow_flush(dp);
270 LIST_FOR_EACH_SAFE (port, next, node, &dp->port_list) {
271 do_del_port(dp, port->port_no);
273 dp_netdev_purge_queues(dp);
274 hmap_destroy(&dp->flow_table);
280 dpif_netdev_close(struct dpif *dpif)
282 struct dp_netdev *dp = get_dp_netdev(dpif);
283 assert(dp->open_cnt > 0);
284 if (--dp->open_cnt == 0 && dp->destroyed) {
285 shash_find_and_delete(&dp_netdevs, dp->name);
292 dpif_netdev_destroy(struct dpif *dpif)
294 struct dp_netdev *dp = get_dp_netdev(dpif);
295 dp->destroyed = true;
300 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
302 struct dp_netdev *dp = get_dp_netdev(dpif);
303 stats->n_flows = hmap_count(&dp->flow_table);
304 stats->n_frags = dp->n_frags;
305 stats->n_hit = dp->n_hit;
306 stats->n_missed = dp->n_missed;
307 stats->n_lost = dp->n_lost;
312 dpif_netdev_get_drop_frags(const struct dpif *dpif, bool *drop_fragsp)
314 struct dp_netdev *dp = get_dp_netdev(dpif);
315 *drop_fragsp = dp->drop_frags;
320 dpif_netdev_set_drop_frags(struct dpif *dpif, bool drop_frags)
322 struct dp_netdev *dp = get_dp_netdev(dpif);
323 dp->drop_frags = drop_frags;
328 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
331 struct dp_netdev_port *port;
332 struct netdev *netdev;
337 /* XXX reject devices already in some dp_netdev. */
338 if (type[0] == '\0' || !strcmp(type, "system") || !strcmp(type, "dummy")) {
340 } else if (!strcmp(type, "internal")) {
343 VLOG_WARN("%s: unsupported port type %s", devname, type);
347 /* Open and validate network device. */
348 if (dp->class == &dpif_dummy_class) {
350 } else if (internal) {
354 error = netdev_open(devname, type, &netdev);
358 /* XXX reject loopback devices */
359 /* XXX reject non-Ethernet devices */
361 error = netdev_listen(netdev);
363 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
364 devname, strerror(errno));
365 netdev_close(netdev);
369 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
371 netdev_close(netdev);
375 port = xmalloc(sizeof *port);
376 port->port_no = port_no;
377 port->netdev = netdev;
378 port->internal = internal;
380 error = netdev_get_mtu(netdev, &mtu);
385 list_push_back(&dp->port_list, &port->node);
386 dp->ports[port_no] = port;
393 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
396 struct dp_netdev *dp = get_dp_netdev(dpif);
399 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
400 if (!dp->ports[port_no]) {
402 return do_add_port(dp, netdev_get_name(netdev),
403 netdev_get_type(netdev), port_no);
410 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
412 struct dp_netdev *dp = get_dp_netdev(dpif);
413 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
417 is_valid_port_number(uint16_t port_no)
419 return port_no < MAX_PORTS;
423 get_port_by_number(struct dp_netdev *dp,
424 uint16_t port_no, struct dp_netdev_port **portp)
426 if (!is_valid_port_number(port_no)) {
430 *portp = dp->ports[port_no];
431 return *portp ? 0 : ENOENT;
436 get_port_by_name(struct dp_netdev *dp,
437 const char *devname, struct dp_netdev_port **portp)
439 struct dp_netdev_port *port;
441 LIST_FOR_EACH (port, node, &dp->port_list) {
442 if (!strcmp(netdev_get_name(port->netdev), devname)) {
451 do_del_port(struct dp_netdev *dp, uint16_t port_no)
453 struct dp_netdev_port *port;
457 error = get_port_by_number(dp, port_no, &port);
462 list_remove(&port->node);
463 dp->ports[port->port_no] = NULL;
466 name = xstrdup(netdev_get_name(port->netdev));
467 netdev_close(port->netdev);
476 answer_port_query(const struct dp_netdev_port *port,
477 struct dpif_port *dpif_port)
479 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
480 dpif_port->type = xstrdup(port->internal ? "internal" : "system");
481 dpif_port->port_no = port->port_no;
485 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
486 struct dpif_port *dpif_port)
488 struct dp_netdev *dp = get_dp_netdev(dpif);
489 struct dp_netdev_port *port;
492 error = get_port_by_number(dp, port_no, &port);
494 answer_port_query(port, dpif_port);
500 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
501 struct dpif_port *dpif_port)
503 struct dp_netdev *dp = get_dp_netdev(dpif);
504 struct dp_netdev_port *port;
507 error = get_port_by_name(dp, devname, &port);
509 answer_port_query(port, dpif_port);
515 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
521 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
523 hmap_remove(&dp->flow_table, &flow->node);
529 dp_netdev_flow_flush(struct dp_netdev *dp)
531 struct dp_netdev_flow *flow, *next;
533 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
534 dp_netdev_free_flow(dp, flow);
539 dpif_netdev_flow_flush(struct dpif *dpif)
541 struct dp_netdev *dp = get_dp_netdev(dpif);
542 dp_netdev_flow_flush(dp);
546 struct dp_netdev_port_state {
552 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
554 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
559 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
560 struct dpif_port *dpif_port)
562 struct dp_netdev_port_state *state = state_;
563 struct dp_netdev *dp = get_dp_netdev(dpif);
566 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
567 struct dp_netdev_port *port = dp->ports[port_no];
570 state->name = xstrdup(netdev_get_name(port->netdev));
571 dpif_port->name = state->name;
572 dpif_port->type = port->internal ? "internal" : "system";
573 dpif_port->port_no = port->port_no;
574 state->port_no = port_no + 1;
582 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
584 struct dp_netdev_port_state *state = state_;
591 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
593 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
594 if (dpif->dp_serial != dpif->dp->serial) {
595 dpif->dp_serial = dpif->dp->serial;
603 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
605 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
606 if (dpif->dp_serial != dpif->dp->serial) {
607 poll_immediate_wake();
611 static struct dp_netdev_flow *
612 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
614 struct dp_netdev_flow *flow;
616 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
617 if (flow_equal(&flow->key, key)) {
625 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
627 stats->n_packets = flow->packet_count;
628 stats->n_bytes = flow->byte_count;
629 stats->used = flow->used;
630 stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl);
634 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
637 if (odp_flow_key_to_flow(key, key_len, flow)) {
638 /* This should not happen: it indicates that odp_flow_key_from_flow()
639 * and odp_flow_key_to_flow() disagree on the acceptable form of a
640 * flow. Log the problem as an error, with enough details to enable
642 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
644 if (!VLOG_DROP_ERR(&rl)) {
648 odp_flow_key_format(key, key_len, &s);
649 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
656 if (flow->in_port < OFPP_MAX
657 ? flow->in_port >= MAX_PORTS
658 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
666 dpif_netdev_flow_get(const struct dpif *dpif,
667 const struct nlattr *nl_key, size_t nl_key_len,
668 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
670 struct dp_netdev *dp = get_dp_netdev(dpif);
671 struct dp_netdev_flow *flow;
675 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
680 flow = dp_netdev_lookup_flow(dp, &key);
686 get_dpif_flow_stats(flow, stats);
689 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
695 dpif_netdev_validate_actions(const struct nlattr *actions,
696 size_t actions_len, bool *mutates)
698 const struct nlattr *a;
702 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
703 uint16_t type = nl_attr_type(a);
704 int len = odp_action_len(type);
706 if (len != nl_attr_get_size(a)) {
711 case OVS_ACTION_ATTR_OUTPUT:
712 if (nl_attr_get_u32(a) >= MAX_PORTS) {
717 case OVS_ACTION_ATTR_USERSPACE:
720 case OVS_ACTION_ATTR_PUSH_VLAN:
722 if (nl_attr_get_be16(a) & htons(VLAN_CFI)) {
727 case OVS_ACTION_ATTR_SET_NW_TOS:
729 if (nl_attr_get_u8(a) & IP_ECN_MASK) {
734 case OVS_ACTION_ATTR_POP_VLAN:
735 case OVS_ACTION_ATTR_SET_DL_SRC:
736 case OVS_ACTION_ATTR_SET_DL_DST:
737 case OVS_ACTION_ATTR_SET_NW_SRC:
738 case OVS_ACTION_ATTR_SET_NW_DST:
739 case OVS_ACTION_ATTR_SET_TP_SRC:
740 case OVS_ACTION_ATTR_SET_TP_DST:
744 case OVS_ACTION_ATTR_SET_TUNNEL:
745 case OVS_ACTION_ATTR_SET_PRIORITY:
746 case OVS_ACTION_ATTR_POP_PRIORITY:
755 set_flow_actions(struct dp_netdev_flow *flow,
756 const struct nlattr *actions, size_t actions_len)
761 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
766 flow->actions = xrealloc(flow->actions, actions_len);
767 flow->actions_len = actions_len;
768 memcpy(flow->actions, actions, actions_len);
773 add_flow(struct dpif *dpif, const struct flow *key,
774 const struct nlattr *actions, size_t actions_len)
776 struct dp_netdev *dp = get_dp_netdev(dpif);
777 struct dp_netdev_flow *flow;
780 flow = xzalloc(sizeof *flow);
783 error = set_flow_actions(flow, actions, actions_len);
789 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
794 clear_stats(struct dp_netdev_flow *flow)
797 flow->packet_count = 0;
798 flow->byte_count = 0;
803 dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
804 const struct nlattr *nl_key, size_t nl_key_len,
805 const struct nlattr *actions, size_t actions_len,
806 struct dpif_flow_stats *stats)
808 struct dp_netdev *dp = get_dp_netdev(dpif);
809 struct dp_netdev_flow *flow;
813 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
818 flow = dp_netdev_lookup_flow(dp, &key);
820 if (flags & DPIF_FP_CREATE) {
821 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
823 memset(stats, 0, sizeof *stats);
825 return add_flow(dpif, &key, actions, actions_len);
833 if (flags & DPIF_FP_MODIFY) {
834 int error = set_flow_actions(flow, actions, actions_len);
837 get_dpif_flow_stats(flow, stats);
839 if (flags & DPIF_FP_ZERO_STATS) {
851 dpif_netdev_flow_del(struct dpif *dpif,
852 const struct nlattr *nl_key, size_t nl_key_len,
853 struct dpif_flow_stats *stats)
855 struct dp_netdev *dp = get_dp_netdev(dpif);
856 struct dp_netdev_flow *flow;
860 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
865 flow = dp_netdev_lookup_flow(dp, &key);
868 get_dpif_flow_stats(flow, stats);
870 dp_netdev_free_flow(dp, flow);
877 struct dp_netdev_flow_state {
880 struct nlattr *actions;
881 struct odputil_keybuf keybuf;
882 struct dpif_flow_stats stats;
886 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
888 struct dp_netdev_flow_state *state;
890 *statep = state = xmalloc(sizeof *state);
893 state->actions = NULL;
898 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
899 const struct nlattr **key, size_t *key_len,
900 const struct nlattr **actions, size_t *actions_len,
901 const struct dpif_flow_stats **stats)
903 struct dp_netdev_flow_state *state = state_;
904 struct dp_netdev *dp = get_dp_netdev(dpif);
905 struct dp_netdev_flow *flow;
906 struct hmap_node *node;
908 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
913 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
918 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
919 odp_flow_key_from_flow(&buf, &flow->key);
926 free(state->actions);
927 state->actions = xmemdup(flow->actions, flow->actions_len);
929 *actions = state->actions;
930 *actions_len = flow->actions_len;
934 get_dpif_flow_stats(flow, &state->stats);
935 *stats = &state->stats;
942 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
944 struct dp_netdev_flow_state *state = state_;
946 free(state->actions);
952 dpif_netdev_execute(struct dpif *dpif,
953 const struct nlattr *key_attrs, size_t key_len,
954 const struct nlattr *actions, size_t actions_len,
955 const struct ofpbuf *packet)
957 struct dp_netdev *dp = get_dp_netdev(dpif);
963 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
967 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
973 /* We need a deep copy of 'packet' since we're going to modify its
975 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
976 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
977 ofpbuf_put(©, packet->data, packet->size);
979 /* We still need a shallow copy of 'packet', even though we won't
980 * modify its data, because flow_extract() modifies packet->l2, etc.
981 * We could probably get away with modifying those but it's more polite
986 flow_extract(©, 0, -1, &key);
987 error = dpif_netdev_flow_from_nlattrs(key_attrs, key_len, &key);
989 error = dp_netdev_execute_actions(dp, ©, &key,
990 actions, actions_len);
993 ofpbuf_uninit(©);
999 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
1001 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1002 *listen_mask = dpif_netdev->listen_mask;
1007 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
1009 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1010 dpif_netdev->listen_mask = listen_mask;
1014 static struct dp_netdev_queue *
1015 find_nonempty_queue(struct dpif *dpif)
1017 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1018 struct dp_netdev *dp = get_dp_netdev(dpif);
1019 int mask = dpif_netdev->listen_mask;
1022 for (i = 0; i < N_QUEUES; i++) {
1023 struct dp_netdev_queue *q = &dp->queues[i];
1024 if (q->head != q->tail && mask & (1u << i)) {
1032 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
1034 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1036 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
1047 dpif_netdev_recv_wait(struct dpif *dpif)
1049 if (find_nonempty_queue(dpif)) {
1050 poll_immediate_wake();
1052 /* No messages ready to be received, and dp_wait() will ensure that we
1053 * wake up to queue new messages, so there is nothing to do. */
1058 dpif_netdev_recv_purge(struct dpif *dpif)
1060 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1061 dp_netdev_purge_queues(dpif_netdev->dp);
1065 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
1066 const struct ofpbuf *packet)
1068 flow->used = time_msec();
1069 flow->packet_count++;
1070 flow->byte_count += packet->size;
1071 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1072 struct tcp_header *th = packet->l4;
1073 flow->tcp_ctl |= th->tcp_ctl;
1078 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1079 struct ofpbuf *packet)
1081 struct dp_netdev_flow *flow;
1084 if (packet->size < ETH_HEADER_LEN) {
1087 if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) {
1092 flow = dp_netdev_lookup_flow(dp, &key);
1094 dp_netdev_flow_used(flow, &key, packet);
1095 dp_netdev_execute_actions(dp, packet, &key,
1096 flow->actions, flow->actions_len);
1100 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1105 dpif_netdev_run(struct dpif *dpif)
1107 struct dp_netdev *dp = get_dp_netdev(dpif);
1108 struct dp_netdev_port *port;
1109 struct ofpbuf packet;
1111 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1113 LIST_FOR_EACH (port, node, &dp->port_list) {
1116 /* Reset packet contents. */
1117 ofpbuf_clear(&packet);
1118 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1120 error = netdev_recv(port->netdev, &packet);
1122 dp_netdev_port_input(dp, port, &packet);
1123 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1124 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1125 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1126 netdev_get_name(port->netdev), strerror(error));
1129 ofpbuf_uninit(&packet);
1133 dpif_netdev_wait(struct dpif *dpif)
1135 struct dp_netdev *dp = get_dp_netdev(dpif);
1136 struct dp_netdev_port *port;
1138 LIST_FOR_EACH (port, node, &dp->port_list) {
1139 netdev_recv_wait(port->netdev);
1144 dp_netdev_pop_vlan(struct ofpbuf *packet)
1146 struct vlan_eth_header *veh = packet->l2;
1147 if (packet->size >= sizeof *veh
1148 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
1149 struct eth_header tmp;
1151 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1152 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1153 tmp.eth_type = veh->veth_next_type;
1155 ofpbuf_pull(packet, VLAN_HEADER_LEN);
1156 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
1157 memcpy(packet->data, &tmp, sizeof tmp);
1162 dp_netdev_set_dl_src(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1164 struct eth_header *eh = packet->l2;
1165 memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
1169 dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1171 struct eth_header *eh = packet->l2;
1172 memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
1176 is_ip(const struct ofpbuf *packet, const struct flow *key)
1178 return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
1182 dp_netdev_set_nw_addr(struct ofpbuf *packet, const struct flow *key,
1183 const struct nlattr *a)
1185 if (is_ip(packet, key)) {
1186 struct ip_header *nh = packet->l3;
1187 ovs_be32 ip = nl_attr_get_be32(a);
1188 uint16_t type = nl_attr_type(a);
1191 field = type == OVS_ACTION_ATTR_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
1192 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1193 struct tcp_header *th = packet->l4;
1194 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, ip);
1195 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1196 struct udp_header *uh = packet->l4;
1198 uh->udp_csum = recalc_csum32(uh->udp_csum, *field, ip);
1199 if (!uh->udp_csum) {
1200 uh->udp_csum = htons(0xffff);
1204 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, ip);
1210 dp_netdev_set_nw_tos(struct ofpbuf *packet, const struct flow *key,
1213 if (is_ip(packet, key)) {
1214 struct ip_header *nh = packet->l3;
1215 uint8_t *field = &nh->ip_tos;
1217 /* Set the DSCP bits and preserve the ECN bits. */
1218 uint8_t new = nw_tos | (nh->ip_tos & IP_ECN_MASK);
1220 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
1221 htons((uint16_t) new));
1227 dp_netdev_set_tp_port(struct ofpbuf *packet, const struct flow *key,
1228 const struct nlattr *a)
1230 if (is_ip(packet, key)) {
1231 uint16_t type = nl_attr_type(a);
1232 ovs_be16 port = nl_attr_get_be16(a);
1235 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1236 struct tcp_header *th = packet->l4;
1237 field = (type == OVS_ACTION_ATTR_SET_TP_SRC
1238 ? &th->tcp_src : &th->tcp_dst);
1239 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, port);
1241 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1242 struct udp_header *uh = packet->l4;
1243 field = (type == OVS_ACTION_ATTR_SET_TP_SRC
1244 ? &uh->udp_src : &uh->udp_dst);
1245 uh->udp_csum = recalc_csum16(uh->udp_csum, *field, port);
1254 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1257 struct dp_netdev_port *p = dp->ports[out_port];
1259 netdev_send(p->netdev, packet);
1264 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1265 int queue_no, const struct flow *flow, uint64_t arg)
1267 struct dp_netdev_queue *q = &dp->queues[queue_no];
1268 struct dpif_upcall *upcall;
1272 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1277 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1278 odp_flow_key_from_flow(buf, flow);
1279 key_len = buf->size;
1280 ofpbuf_pull(buf, key_len);
1281 ofpbuf_reserve(buf, 2);
1282 ofpbuf_put(buf, packet->data, packet->size);
1284 upcall = xzalloc(sizeof *upcall);
1285 upcall->type = queue_no;
1286 upcall->packet = buf;
1287 upcall->key = buf->base;
1288 upcall->key_len = key_len;
1289 upcall->userdata = arg;
1291 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1297 dp_netdev_execute_actions(struct dp_netdev *dp,
1298 struct ofpbuf *packet, struct flow *key,
1299 const struct nlattr *actions,
1302 const struct nlattr *a;
1305 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1306 switch (nl_attr_type(a)) {
1307 case OVS_ACTION_ATTR_OUTPUT:
1308 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1311 case OVS_ACTION_ATTR_USERSPACE:
1312 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION,
1313 key, nl_attr_get_u64(a));
1316 case OVS_ACTION_ATTR_PUSH_VLAN:
1317 eth_push_vlan(packet, nl_attr_get_be16(a));
1320 case OVS_ACTION_ATTR_POP_VLAN:
1321 dp_netdev_pop_vlan(packet);
1324 case OVS_ACTION_ATTR_SET_DL_SRC:
1325 dp_netdev_set_dl_src(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1328 case OVS_ACTION_ATTR_SET_DL_DST:
1329 dp_netdev_set_dl_dst(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1332 case OVS_ACTION_ATTR_SET_NW_SRC:
1333 case OVS_ACTION_ATTR_SET_NW_DST:
1334 dp_netdev_set_nw_addr(packet, key, a);
1337 case OVS_ACTION_ATTR_SET_NW_TOS:
1338 dp_netdev_set_nw_tos(packet, key, nl_attr_get_u8(a));
1341 case OVS_ACTION_ATTR_SET_TP_SRC:
1342 case OVS_ACTION_ATTR_SET_TP_DST:
1343 dp_netdev_set_tp_port(packet, key, a);
1350 const struct dpif_class dpif_netdev_class = {
1352 NULL, /* enumerate */
1355 dpif_netdev_destroy,
1358 dpif_netdev_get_stats,
1359 dpif_netdev_get_drop_frags,
1360 dpif_netdev_set_drop_frags,
1361 dpif_netdev_port_add,
1362 dpif_netdev_port_del,
1363 dpif_netdev_port_query_by_number,
1364 dpif_netdev_port_query_by_name,
1365 dpif_netdev_get_max_ports,
1366 dpif_netdev_port_dump_start,
1367 dpif_netdev_port_dump_next,
1368 dpif_netdev_port_dump_done,
1369 dpif_netdev_port_poll,
1370 dpif_netdev_port_poll_wait,
1371 dpif_netdev_flow_get,
1372 dpif_netdev_flow_put,
1373 dpif_netdev_flow_del,
1374 dpif_netdev_flow_flush,
1375 dpif_netdev_flow_dump_start,
1376 dpif_netdev_flow_dump_next,
1377 dpif_netdev_flow_dump_done,
1378 dpif_netdev_execute,
1379 dpif_netdev_recv_get_mask,
1380 dpif_netdev_recv_set_mask,
1381 NULL, /* queue_to_priority */
1383 dpif_netdev_recv_wait,
1384 dpif_netdev_recv_purge,
1388 dpif_dummy_register(void)
1390 if (!dpif_dummy_class.type) {
1391 dpif_dummy_class = dpif_netdev_class;
1392 dpif_dummy_class.type = "dummy";
1393 dp_register_provider(&dpif_dummy_class);