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. */
95 struct dp_netdev_port *ports[MAX_PORTS];
96 struct list port_list;
100 /* A port in a netdev-based datapath. */
101 struct dp_netdev_port {
102 int port_no; /* Index into dp_netdev's 'ports'. */
103 struct list node; /* Element in dp_netdev's 'port_list'. */
104 struct netdev *netdev;
105 bool internal; /* Internal port? */
108 /* A flow in dp_netdev's 'flow_table'. */
109 struct dp_netdev_flow {
110 struct hmap_node node; /* Element in dp_netdev's 'flow_table'. */
114 long long int used; /* Last used time, in monotonic msecs. */
115 long long int packet_count; /* Number of packets matched. */
116 long long int byte_count; /* Number of bytes matched. */
117 uint16_t tcp_ctl; /* Bitwise-OR of seen tcp_ctl values. */
120 struct nlattr *actions;
124 /* Interface to netdev-based datapath. */
127 struct dp_netdev *dp;
129 unsigned int dp_serial;
132 /* All netdev-based datapaths. */
133 static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
135 /* Maximum port MTU seen so far. */
136 static int max_mtu = ETH_PAYLOAD_MAX;
138 static int get_port_by_number(struct dp_netdev *, uint16_t port_no,
139 struct dp_netdev_port **portp);
140 static int get_port_by_name(struct dp_netdev *, const char *devname,
141 struct dp_netdev_port **portp);
142 static void dp_netdev_free(struct dp_netdev *);
143 static void dp_netdev_flow_flush(struct dp_netdev *);
144 static int do_add_port(struct dp_netdev *, const char *devname,
145 const char *type, uint16_t port_no);
146 static int do_del_port(struct dp_netdev *, uint16_t port_no);
147 static int dpif_netdev_open(const struct dpif_class *, const char *name,
148 bool create, struct dpif **);
149 static int dp_netdev_output_control(struct dp_netdev *, const struct ofpbuf *,
150 int queue_no, const struct flow *,
152 static int dp_netdev_execute_actions(struct dp_netdev *,
153 struct ofpbuf *, struct flow *,
154 const struct nlattr *actions,
157 static struct dpif_class dpif_dummy_class;
159 static struct dpif_netdev *
160 dpif_netdev_cast(const struct dpif *dpif)
162 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 create_dpif_netdev(struct dp_netdev *dp)
175 uint16_t netflow_id = hash_string(dp->name, 0);
176 struct dpif_netdev *dpif;
180 dpif = xmalloc(sizeof *dpif);
181 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
183 dpif->listen_mask = 0;
184 dpif->dp_serial = dp->serial;
190 create_dp_netdev(const char *name, const struct dpif_class *class,
191 struct dp_netdev **dpp)
193 struct dp_netdev *dp;
197 dp = xzalloc(sizeof *dp);
199 dp->name = xstrdup(name);
201 dp->drop_frags = false;
202 for (i = 0; i < N_QUEUES; i++) {
203 dp->queues[i].head = dp->queues[i].tail = 0;
205 hmap_init(&dp->flow_table);
206 list_init(&dp->port_list);
207 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
213 shash_add(&dp_netdevs, name, dp);
220 dpif_netdev_open(const struct dpif_class *class, const char *name,
221 bool create, struct dpif **dpifp)
223 struct dp_netdev *dp;
225 dp = shash_find_data(&dp_netdevs, name);
230 int error = create_dp_netdev(name, class, &dp);
237 if (dp->class != class) {
244 *dpifp = create_dpif_netdev(dp);
249 dp_netdev_purge_queues(struct dp_netdev *dp)
253 for (i = 0; i < N_QUEUES; i++) {
254 struct dp_netdev_queue *q = &dp->queues[i];
256 while (q->tail != q->head) {
257 struct dpif_upcall *upcall = q->upcalls[q->tail++ & QUEUE_MASK];
259 ofpbuf_delete(upcall->packet);
266 dp_netdev_free(struct dp_netdev *dp)
268 dp_netdev_flow_flush(dp);
269 while (dp->n_ports > 0) {
270 struct dp_netdev_port *port = CONTAINER_OF(
271 dp->port_list.next, struct dp_netdev_port, node);
272 do_del_port(dp, port->port_no);
274 dp_netdev_purge_queues(dp);
275 hmap_destroy(&dp->flow_table);
281 dpif_netdev_close(struct dpif *dpif)
283 struct dp_netdev *dp = get_dp_netdev(dpif);
284 assert(dp->open_cnt > 0);
285 if (--dp->open_cnt == 0 && dp->destroyed) {
286 shash_find_and_delete(&dp_netdevs, dp->name);
293 dpif_netdev_destroy(struct dpif *dpif)
295 struct dp_netdev *dp = get_dp_netdev(dpif);
296 dp->destroyed = true;
301 dpif_netdev_get_stats(const struct dpif *dpif, struct odp_stats *stats)
303 struct dp_netdev *dp = get_dp_netdev(dpif);
304 memset(stats, 0, sizeof *stats);
305 stats->n_frags = dp->n_frags;
306 stats->n_hit = dp->n_hit;
307 stats->n_missed = dp->n_missed;
308 stats->n_lost = dp->n_lost;
313 dpif_netdev_get_drop_frags(const struct dpif *dpif, bool *drop_fragsp)
315 struct dp_netdev *dp = get_dp_netdev(dpif);
316 *drop_fragsp = dp->drop_frags;
321 dpif_netdev_set_drop_frags(struct dpif *dpif, bool drop_frags)
323 struct dp_netdev *dp = get_dp_netdev(dpif);
324 dp->drop_frags = drop_frags;
329 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
332 struct dp_netdev_port *port;
333 struct netdev_options netdev_options;
334 struct netdev *netdev;
339 /* XXX reject devices already in some dp_netdev. */
340 if (type[0] == '\0' || !strcmp(type, "system")) {
342 } else if (!strcmp(type, "internal")) {
345 VLOG_WARN("%s: unsupported port type %s", devname, type);
349 /* Open and validate network device. */
350 memset(&netdev_options, 0, sizeof netdev_options);
351 netdev_options.name = devname;
352 netdev_options.ethertype = NETDEV_ETH_TYPE_ANY;
353 if (dp->class == &dpif_dummy_class) {
354 netdev_options.type = "dummy";
355 } else if (internal) {
356 netdev_options.type = "tap";
359 error = netdev_open(&netdev_options, &netdev);
363 /* XXX reject loopback devices */
364 /* XXX reject non-Ethernet devices */
366 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
368 netdev_close(netdev);
372 port = xmalloc(sizeof *port);
373 port->port_no = port_no;
374 port->netdev = netdev;
375 port->internal = internal;
377 netdev_get_mtu(netdev, &mtu);
378 if (mtu != INT_MAX && mtu > max_mtu) {
382 list_push_back(&dp->port_list, &port->node);
383 dp->ports[port_no] = port;
391 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
394 struct dp_netdev *dp = get_dp_netdev(dpif);
397 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
398 if (!dp->ports[port_no]) {
400 return do_add_port(dp, netdev_get_name(netdev),
401 netdev_get_type(netdev), port_no);
408 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
410 struct dp_netdev *dp = get_dp_netdev(dpif);
411 return port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
415 is_valid_port_number(uint16_t port_no)
417 return port_no < MAX_PORTS;
421 get_port_by_number(struct dp_netdev *dp,
422 uint16_t port_no, struct dp_netdev_port **portp)
424 if (!is_valid_port_number(port_no)) {
428 *portp = dp->ports[port_no];
429 return *portp ? 0 : ENOENT;
434 get_port_by_name(struct dp_netdev *dp,
435 const char *devname, struct dp_netdev_port **portp)
437 struct dp_netdev_port *port;
439 LIST_FOR_EACH (port, node, &dp->port_list) {
440 if (!strcmp(netdev_get_name(port->netdev), devname)) {
449 do_del_port(struct dp_netdev *dp, uint16_t port_no)
451 struct dp_netdev_port *port;
455 error = get_port_by_number(dp, port_no, &port);
460 list_remove(&port->node);
461 dp->ports[port->port_no] = NULL;
465 name = xstrdup(netdev_get_name(port->netdev));
466 netdev_close(port->netdev);
475 answer_port_query(const struct dp_netdev_port *port,
476 struct dpif_port *dpif_port)
478 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
479 dpif_port->type = xstrdup(port->internal ? "internal" : "system");
480 dpif_port->port_no = port->port_no;
484 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
485 struct dpif_port *dpif_port)
487 struct dp_netdev *dp = get_dp_netdev(dpif);
488 struct dp_netdev_port *port;
491 error = get_port_by_number(dp, port_no, &port);
493 answer_port_query(port, dpif_port);
499 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
500 struct dpif_port *dpif_port)
502 struct dp_netdev *dp = get_dp_netdev(dpif);
503 struct dp_netdev_port *port;
506 error = get_port_by_name(dp, devname, &port);
508 answer_port_query(port, dpif_port);
514 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
520 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
522 hmap_remove(&dp->flow_table, &flow->node);
528 dp_netdev_flow_flush(struct dp_netdev *dp)
530 struct dp_netdev_flow *flow, *next;
532 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
533 dp_netdev_free_flow(dp, flow);
538 dpif_netdev_flow_flush(struct dpif *dpif)
540 struct dp_netdev *dp = get_dp_netdev(dpif);
541 dp_netdev_flow_flush(dp);
545 struct dp_netdev_port_state {
551 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
553 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
558 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
559 struct dpif_port *dpif_port)
561 struct dp_netdev_port_state *state = state_;
562 struct dp_netdev *dp = get_dp_netdev(dpif);
565 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
566 struct dp_netdev_port *port = dp->ports[port_no];
569 state->name = xstrdup(netdev_get_name(port->netdev));
570 dpif_port->name = state->name;
571 dpif_port->type = port->internal ? "internal" : "system";
572 dpif_port->port_no = port->port_no;
573 state->port_no = port_no + 1;
581 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
583 struct dp_netdev_port_state *state = state_;
590 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
592 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
593 if (dpif->dp_serial != dpif->dp->serial) {
594 dpif->dp_serial = dpif->dp->serial;
602 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
604 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
605 if (dpif->dp_serial != dpif->dp->serial) {
606 poll_immediate_wake();
610 static struct dp_netdev_flow *
611 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
613 struct dp_netdev_flow *flow;
615 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
616 if (flow_equal(&flow->key, key)) {
624 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
626 stats->n_packets = flow->packet_count;
627 stats->n_bytes = flow->byte_count;
628 stats->used = flow->used;
629 stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl);
633 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
636 if (odp_flow_key_to_flow(key, key_len, flow)) {
637 /* This should not happen: it indicates that odp_flow_key_from_flow()
638 * and odp_flow_key_to_flow() disagree on the acceptable form of a
639 * flow. Log the problem as an error, with enough details to enable
641 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
643 if (!VLOG_DROP_ERR(&rl)) {
647 odp_flow_key_format(key, key_len, &s);
648 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
659 dpif_netdev_flow_get(const struct dpif *dpif,
660 const struct nlattr *nl_key, size_t nl_key_len,
661 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
663 struct dp_netdev *dp = get_dp_netdev(dpif);
664 struct dp_netdev_flow *flow;
668 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
673 flow = dp_netdev_lookup_flow(dp, &key);
679 get_dpif_flow_stats(flow, stats);
682 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
688 dpif_netdev_validate_actions(const struct nlattr *actions,
689 size_t actions_len, bool *mutates)
691 const struct nlattr *a;
695 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
696 uint16_t type = nl_attr_type(a);
697 int len = odp_action_len(type);
699 if (len != nl_attr_get_size(a)) {
704 case ODP_ACTION_ATTR_OUTPUT:
705 if (nl_attr_get_u32(a) >= MAX_PORTS) {
710 case ODP_ACTION_ATTR_CONTROLLER:
711 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
714 case ODP_ACTION_ATTR_SET_DL_TCI:
716 if (nl_attr_get_be16(a) & htons(VLAN_CFI)) {
721 case ODP_ACTION_ATTR_SET_NW_TOS:
723 if (nl_attr_get_u8(a) & IP_ECN_MASK) {
728 case ODP_ACTION_ATTR_STRIP_VLAN:
729 case ODP_ACTION_ATTR_SET_DL_SRC:
730 case ODP_ACTION_ATTR_SET_DL_DST:
731 case ODP_ACTION_ATTR_SET_NW_SRC:
732 case ODP_ACTION_ATTR_SET_NW_DST:
733 case ODP_ACTION_ATTR_SET_TP_SRC:
734 case ODP_ACTION_ATTR_SET_TP_DST:
738 case ODP_ACTION_ATTR_SET_TUNNEL:
739 case ODP_ACTION_ATTR_SET_PRIORITY:
740 case ODP_ACTION_ATTR_POP_PRIORITY:
749 set_flow_actions(struct dp_netdev_flow *flow,
750 const struct nlattr *actions, size_t actions_len)
755 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
760 flow->actions = xrealloc(flow->actions, actions_len);
761 flow->actions_len = actions_len;
762 memcpy(flow->actions, actions, actions_len);
767 add_flow(struct dpif *dpif, const struct flow *key,
768 const struct nlattr *actions, size_t actions_len)
770 struct dp_netdev *dp = get_dp_netdev(dpif);
771 struct dp_netdev_flow *flow;
774 flow = xzalloc(sizeof *flow);
777 error = set_flow_actions(flow, actions, actions_len);
783 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
788 clear_stats(struct dp_netdev_flow *flow)
791 flow->packet_count = 0;
792 flow->byte_count = 0;
797 dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
798 const struct nlattr *nl_key, size_t nl_key_len,
799 const struct nlattr *actions, size_t actions_len,
800 struct dpif_flow_stats *stats)
802 struct dp_netdev *dp = get_dp_netdev(dpif);
803 struct dp_netdev_flow *flow;
807 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
812 flow = dp_netdev_lookup_flow(dp, &key);
814 if (flags & DPIF_FP_CREATE) {
815 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
817 memset(stats, 0, sizeof *stats);
819 return add_flow(dpif, &key, actions, actions_len);
827 if (flags & DPIF_FP_MODIFY) {
828 int error = set_flow_actions(flow, actions, actions_len);
831 get_dpif_flow_stats(flow, stats);
833 if (flags & DPIF_FP_ZERO_STATS) {
845 dpif_netdev_flow_del(struct dpif *dpif,
846 const struct nlattr *nl_key, size_t nl_key_len,
847 struct dpif_flow_stats *stats)
849 struct dp_netdev *dp = get_dp_netdev(dpif);
850 struct dp_netdev_flow *flow;
854 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
859 flow = dp_netdev_lookup_flow(dp, &key);
862 get_dpif_flow_stats(flow, stats);
864 dp_netdev_free_flow(dp, flow);
871 struct dp_netdev_flow_state {
874 struct nlattr *actions;
875 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
876 struct dpif_flow_stats stats;
880 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
882 struct dp_netdev_flow_state *state;
884 *statep = state = xmalloc(sizeof *state);
887 state->actions = NULL;
892 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
893 const struct nlattr **key, size_t *key_len,
894 const struct nlattr **actions, size_t *actions_len,
895 const struct dpif_flow_stats **stats)
897 struct dp_netdev_flow_state *state = state_;
898 struct dp_netdev *dp = get_dp_netdev(dpif);
899 struct dp_netdev_flow *flow;
900 struct hmap_node *node;
902 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
907 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
912 ofpbuf_use_stack(&buf, state->keybuf, sizeof state->keybuf);
913 odp_flow_key_from_flow(&buf, &flow->key);
914 assert(buf.base == state->keybuf);
921 free(state->actions);
922 state->actions = xmemdup(flow->actions, flow->actions_len);
924 *actions = state->actions;
925 *actions_len = flow->actions_len;
929 get_dpif_flow_stats(flow, &state->stats);
930 *stats = &state->stats;
937 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
939 struct dp_netdev_flow_state *state = state_;
941 free(state->actions);
947 dpif_netdev_execute(struct dpif *dpif,
948 const struct nlattr *actions, size_t actions_len,
949 const struct ofpbuf *packet)
951 struct dp_netdev *dp = get_dp_netdev(dpif);
957 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
961 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
967 /* We need a deep copy of 'packet' since we're going to modify its
969 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
970 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
971 ofpbuf_put(©, packet->data, packet->size);
973 /* We still need a shallow copy of 'packet', even though we won't
974 * modify its data, because flow_extract() modifies packet->l2, etc.
975 * We could probably get away with modifying those but it's more polite
979 flow_extract(©, 0, -1, &key);
980 error = dp_netdev_execute_actions(dp, ©, &key, actions, actions_len);
982 ofpbuf_uninit(©);
988 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
990 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
991 *listen_mask = dpif_netdev->listen_mask;
996 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
998 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
999 dpif_netdev->listen_mask = listen_mask;
1003 static struct dp_netdev_queue *
1004 find_nonempty_queue(struct dpif *dpif)
1006 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1007 struct dp_netdev *dp = get_dp_netdev(dpif);
1008 int mask = dpif_netdev->listen_mask;
1011 for (i = 0; i < N_QUEUES; i++) {
1012 struct dp_netdev_queue *q = &dp->queues[i];
1013 if (q->head != q->tail && mask & (1u << i)) {
1021 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
1023 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1025 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
1036 dpif_netdev_recv_wait(struct dpif *dpif)
1038 if (find_nonempty_queue(dpif)) {
1039 poll_immediate_wake();
1041 /* No messages ready to be received, and dp_wait() will ensure that we
1042 * wake up to queue new messages, so there is nothing to do. */
1047 dpif_netdev_recv_purge(struct dpif *dpif)
1049 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1050 dp_netdev_purge_queues(dpif_netdev->dp);
1054 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
1055 const struct ofpbuf *packet)
1057 flow->used = time_msec();
1058 flow->packet_count++;
1059 flow->byte_count += packet->size;
1060 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1061 struct tcp_header *th = packet->l4;
1062 flow->tcp_ctl |= th->tcp_ctl;
1067 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1068 struct ofpbuf *packet)
1070 struct dp_netdev_flow *flow;
1073 if (packet->size < ETH_HEADER_LEN) {
1076 if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) {
1081 flow = dp_netdev_lookup_flow(dp, &key);
1083 dp_netdev_flow_used(flow, &key, packet);
1084 dp_netdev_execute_actions(dp, packet, &key,
1085 flow->actions, flow->actions_len);
1089 dp_netdev_output_control(dp, packet, DPIF_UC_MISS, &key, 0);
1096 struct shash_node *node;
1097 struct ofpbuf packet;
1099 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1100 SHASH_FOR_EACH (node, &dp_netdevs) {
1101 struct dp_netdev *dp = node->data;
1102 struct dp_netdev_port *port;
1104 LIST_FOR_EACH (port, node, &dp->port_list) {
1107 /* Reset packet contents. */
1108 ofpbuf_clear(&packet);
1109 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1111 error = netdev_recv(port->netdev, &packet);
1113 dp_netdev_port_input(dp, port, &packet);
1114 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1115 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1116 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1117 netdev_get_name(port->netdev), strerror(error));
1121 ofpbuf_uninit(&packet);
1125 dp_netdev_wait(void)
1127 struct shash_node *node;
1129 SHASH_FOR_EACH (node, &dp_netdevs) {
1130 struct dp_netdev *dp = node->data;
1131 struct dp_netdev_port *port;
1133 LIST_FOR_EACH (port, node, &dp->port_list) {
1134 netdev_recv_wait(port->netdev);
1140 /* Modify the TCI field of 'packet'. If a VLAN tag is present, its TCI field
1141 * is replaced by 'tci'. If a VLAN tag is not present, one is added with the
1142 * TCI field set to 'tci'.
1145 dp_netdev_set_dl_tci(struct ofpbuf *packet, uint16_t tci)
1147 struct vlan_eth_header *veh;
1148 struct eth_header *eh;
1151 if (packet->size >= sizeof(struct vlan_eth_header)
1152 && eh->eth_type == htons(ETH_TYPE_VLAN)) {
1154 veh->veth_tci = tci;
1156 /* Insert new 802.1Q header. */
1157 struct vlan_eth_header tmp;
1158 memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
1159 memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
1160 tmp.veth_type = htons(ETH_TYPE_VLAN);
1162 tmp.veth_next_type = eh->eth_type;
1164 veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN);
1165 memcpy(veh, &tmp, sizeof tmp);
1166 packet->l2 = (char*)packet->l2 - VLAN_HEADER_LEN;
1171 dp_netdev_strip_vlan(struct ofpbuf *packet)
1173 struct vlan_eth_header *veh = packet->l2;
1174 if (packet->size >= sizeof *veh
1175 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
1176 struct eth_header tmp;
1178 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1179 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1180 tmp.eth_type = veh->veth_next_type;
1182 ofpbuf_pull(packet, VLAN_HEADER_LEN);
1183 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
1184 memcpy(packet->data, &tmp, sizeof tmp);
1189 dp_netdev_set_dl_src(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1191 struct eth_header *eh = packet->l2;
1192 memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
1196 dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1198 struct eth_header *eh = packet->l2;
1199 memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
1203 is_ip(const struct ofpbuf *packet, const struct flow *key)
1205 return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
1209 dp_netdev_set_nw_addr(struct ofpbuf *packet, const struct flow *key,
1210 const struct nlattr *a)
1212 if (is_ip(packet, key)) {
1213 struct ip_header *nh = packet->l3;
1214 ovs_be32 ip = nl_attr_get_be32(a);
1215 uint16_t type = nl_attr_type(a);
1218 field = type == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
1219 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1220 struct tcp_header *th = packet->l4;
1221 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, ip);
1222 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1223 struct udp_header *uh = packet->l4;
1225 uh->udp_csum = recalc_csum32(uh->udp_csum, *field, ip);
1226 if (!uh->udp_csum) {
1227 uh->udp_csum = 0xffff;
1231 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, ip);
1237 dp_netdev_set_nw_tos(struct ofpbuf *packet, const struct flow *key,
1240 if (is_ip(packet, key)) {
1241 struct ip_header *nh = packet->l3;
1242 uint8_t *field = &nh->ip_tos;
1244 /* Set the DSCP bits and preserve the ECN bits. */
1245 uint8_t new = nw_tos | (nh->ip_tos & IP_ECN_MASK);
1247 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
1248 htons((uint16_t) new));
1254 dp_netdev_set_tp_port(struct ofpbuf *packet, const struct flow *key,
1255 const struct nlattr *a)
1257 if (is_ip(packet, key)) {
1258 uint16_t type = nl_attr_type(a);
1259 ovs_be16 port = nl_attr_get_be16(a);
1262 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1263 struct tcp_header *th = packet->l4;
1264 field = (type == ODP_ACTION_ATTR_SET_TP_SRC
1265 ? &th->tcp_src : &th->tcp_dst);
1266 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, port);
1268 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1269 struct udp_header *uh = packet->l4;
1270 field = (type == ODP_ACTION_ATTR_SET_TP_SRC
1271 ? &uh->udp_src : &uh->udp_dst);
1272 uh->udp_csum = recalc_csum16(uh->udp_csum, *field, port);
1281 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1284 struct dp_netdev_port *p = dp->ports[out_port];
1286 netdev_send(p->netdev, packet);
1291 dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet,
1292 int queue_no, const struct flow *flow, uint64_t arg)
1294 struct dp_netdev_queue *q = &dp->queues[queue_no];
1295 struct dpif_upcall *upcall;
1299 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1304 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1305 odp_flow_key_from_flow(buf, flow);
1306 key_len = buf->size;
1307 ofpbuf_pull(buf, key_len);
1308 ofpbuf_reserve(buf, 2);
1309 ofpbuf_put(buf, packet->data, packet->size);
1311 upcall = xzalloc(sizeof *upcall);
1312 upcall->type = queue_no;
1313 upcall->packet = buf;
1314 upcall->key = buf->base;
1315 upcall->key_len = key_len;
1316 upcall->userdata = arg;
1318 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1323 /* Returns true if 'packet' is an invalid Ethernet+IPv4 ARP packet: one with
1324 * screwy or truncated header fields or one whose inner and outer Ethernet
1325 * address differ. */
1327 dp_netdev_is_spoofed_arp(struct ofpbuf *packet, const struct flow *key)
1329 struct arp_eth_header *arp;
1330 struct eth_header *eth;
1333 if (key->dl_type != htons(ETH_TYPE_ARP)) {
1337 l3_size = (char *) ofpbuf_end(packet) - (char *) packet->l3;
1338 if (l3_size < sizeof(struct arp_eth_header)) {
1344 return (arp->ar_hrd != htons(ARP_HRD_ETHERNET)
1345 || arp->ar_pro != htons(ARP_PRO_IP)
1346 || arp->ar_hln != ETH_HEADER_LEN
1348 || !eth_addr_equals(arp->ar_sha, eth->eth_src));
1352 dp_netdev_execute_actions(struct dp_netdev *dp,
1353 struct ofpbuf *packet, struct flow *key,
1354 const struct nlattr *actions,
1357 const struct nlattr *a;
1360 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1361 switch (nl_attr_type(a)) {
1362 case ODP_ACTION_ATTR_OUTPUT:
1363 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1366 case ODP_ACTION_ATTR_CONTROLLER:
1367 dp_netdev_output_control(dp, packet, DPIF_UC_ACTION,
1368 key, nl_attr_get_u64(a));
1371 case ODP_ACTION_ATTR_SET_DL_TCI:
1372 dp_netdev_set_dl_tci(packet, nl_attr_get_be16(a));
1375 case ODP_ACTION_ATTR_STRIP_VLAN:
1376 dp_netdev_strip_vlan(packet);
1379 case ODP_ACTION_ATTR_SET_DL_SRC:
1380 dp_netdev_set_dl_src(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1383 case ODP_ACTION_ATTR_SET_DL_DST:
1384 dp_netdev_set_dl_dst(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1387 case ODP_ACTION_ATTR_SET_NW_SRC:
1388 case ODP_ACTION_ATTR_SET_NW_DST:
1389 dp_netdev_set_nw_addr(packet, key, a);
1392 case ODP_ACTION_ATTR_SET_NW_TOS:
1393 dp_netdev_set_nw_tos(packet, key, nl_attr_get_u8(a));
1396 case ODP_ACTION_ATTR_SET_TP_SRC:
1397 case ODP_ACTION_ATTR_SET_TP_DST:
1398 dp_netdev_set_tp_port(packet, key, a);
1401 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
1402 if (dp_netdev_is_spoofed_arp(packet, key)) {
1410 const struct dpif_class dpif_netdev_class = {
1414 NULL, /* enumerate */
1417 dpif_netdev_destroy,
1418 dpif_netdev_get_stats,
1419 dpif_netdev_get_drop_frags,
1420 dpif_netdev_set_drop_frags,
1421 dpif_netdev_port_add,
1422 dpif_netdev_port_del,
1423 dpif_netdev_port_query_by_number,
1424 dpif_netdev_port_query_by_name,
1425 dpif_netdev_get_max_ports,
1426 dpif_netdev_port_dump_start,
1427 dpif_netdev_port_dump_next,
1428 dpif_netdev_port_dump_done,
1429 dpif_netdev_port_poll,
1430 dpif_netdev_port_poll_wait,
1431 dpif_netdev_flow_get,
1432 dpif_netdev_flow_put,
1433 dpif_netdev_flow_del,
1434 dpif_netdev_flow_flush,
1435 dpif_netdev_flow_dump_start,
1436 dpif_netdev_flow_dump_next,
1437 dpif_netdev_flow_dump_done,
1438 dpif_netdev_execute,
1439 dpif_netdev_recv_get_mask,
1440 dpif_netdev_recv_set_mask,
1441 NULL, /* get_sflow_probability */
1442 NULL, /* set_sflow_probability */
1443 NULL, /* queue_to_priority */
1445 dpif_netdev_recv_wait,
1446 dpif_netdev_recv_purge,
1450 dpif_dummy_register(void)
1452 if (!dpif_dummy_class.type) {
1453 dpif_dummy_class = dpif_netdev_class;
1454 dpif_dummy_class.type = "dummy";
1455 dp_register_provider(&dpif_dummy_class);