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 ovs_dp_stats *stats)
302 struct dp_netdev *dp = get_dp_netdev(dpif);
303 memset(stats, 0, sizeof *stats);
304 stats->n_flows = hmap_count(&dp->flow_table);
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 *netdev;
338 /* XXX reject devices already in some dp_netdev. */
339 if (type[0] == '\0' || !strcmp(type, "system")) {
341 } else if (!strcmp(type, "internal")) {
344 VLOG_WARN("%s: unsupported port type %s", devname, type);
348 /* Open and validate network device. */
349 if (dp->class == &dpif_dummy_class) {
351 } else if (internal) {
355 error = netdev_open(devname, type, &netdev);
359 /* XXX reject loopback devices */
360 /* XXX reject non-Ethernet devices */
362 error = netdev_listen(netdev);
364 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
365 devname, strerror(errno));
366 netdev_close(netdev);
370 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, false);
372 netdev_close(netdev);
376 port = xmalloc(sizeof *port);
377 port->port_no = port_no;
378 port->netdev = netdev;
379 port->internal = internal;
381 error = netdev_get_mtu(netdev, &mtu);
386 list_push_back(&dp->port_list, &port->node);
387 dp->ports[port_no] = port;
394 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
397 struct dp_netdev *dp = get_dp_netdev(dpif);
400 for (port_no = 0; port_no < MAX_PORTS; port_no++) {
401 if (!dp->ports[port_no]) {
403 return do_add_port(dp, netdev_get_name(netdev),
404 netdev_get_type(netdev), port_no);
411 dpif_netdev_port_del(struct dpif *dpif, uint16_t port_no)
413 struct dp_netdev *dp = get_dp_netdev(dpif);
414 return port_no == OVSP_LOCAL ? EINVAL : do_del_port(dp, port_no);
418 is_valid_port_number(uint16_t port_no)
420 return port_no < MAX_PORTS;
424 get_port_by_number(struct dp_netdev *dp,
425 uint16_t port_no, struct dp_netdev_port **portp)
427 if (!is_valid_port_number(port_no)) {
431 *portp = dp->ports[port_no];
432 return *portp ? 0 : ENOENT;
437 get_port_by_name(struct dp_netdev *dp,
438 const char *devname, struct dp_netdev_port **portp)
440 struct dp_netdev_port *port;
442 LIST_FOR_EACH (port, node, &dp->port_list) {
443 if (!strcmp(netdev_get_name(port->netdev), devname)) {
452 do_del_port(struct dp_netdev *dp, uint16_t port_no)
454 struct dp_netdev_port *port;
458 error = get_port_by_number(dp, port_no, &port);
463 list_remove(&port->node);
464 dp->ports[port->port_no] = NULL;
467 name = xstrdup(netdev_get_name(port->netdev));
468 netdev_close(port->netdev);
477 answer_port_query(const struct dp_netdev_port *port,
478 struct dpif_port *dpif_port)
480 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
481 dpif_port->type = xstrdup(port->internal ? "internal" : "system");
482 dpif_port->port_no = port->port_no;
486 dpif_netdev_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
487 struct dpif_port *dpif_port)
489 struct dp_netdev *dp = get_dp_netdev(dpif);
490 struct dp_netdev_port *port;
493 error = get_port_by_number(dp, port_no, &port);
495 answer_port_query(port, dpif_port);
501 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
502 struct dpif_port *dpif_port)
504 struct dp_netdev *dp = get_dp_netdev(dpif);
505 struct dp_netdev_port *port;
508 error = get_port_by_name(dp, devname, &port);
510 answer_port_query(port, dpif_port);
516 dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
522 dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
524 hmap_remove(&dp->flow_table, &flow->node);
530 dp_netdev_flow_flush(struct dp_netdev *dp)
532 struct dp_netdev_flow *flow, *next;
534 HMAP_FOR_EACH_SAFE (flow, next, node, &dp->flow_table) {
535 dp_netdev_free_flow(dp, flow);
540 dpif_netdev_flow_flush(struct dpif *dpif)
542 struct dp_netdev *dp = get_dp_netdev(dpif);
543 dp_netdev_flow_flush(dp);
547 struct dp_netdev_port_state {
553 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
555 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
560 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
561 struct dpif_port *dpif_port)
563 struct dp_netdev_port_state *state = state_;
564 struct dp_netdev *dp = get_dp_netdev(dpif);
567 for (port_no = state->port_no; port_no < MAX_PORTS; port_no++) {
568 struct dp_netdev_port *port = dp->ports[port_no];
571 state->name = xstrdup(netdev_get_name(port->netdev));
572 dpif_port->name = state->name;
573 dpif_port->type = port->internal ? "internal" : "system";
574 dpif_port->port_no = port->port_no;
575 state->port_no = port_no + 1;
583 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
585 struct dp_netdev_port_state *state = state_;
592 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
594 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
595 if (dpif->dp_serial != dpif->dp->serial) {
596 dpif->dp_serial = dpif->dp->serial;
604 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
606 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
607 if (dpif->dp_serial != dpif->dp->serial) {
608 poll_immediate_wake();
612 static struct dp_netdev_flow *
613 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *key)
615 struct dp_netdev_flow *flow;
617 HMAP_FOR_EACH_WITH_HASH (flow, node, flow_hash(key, 0), &dp->flow_table) {
618 if (flow_equal(&flow->key, key)) {
626 get_dpif_flow_stats(struct dp_netdev_flow *flow, struct dpif_flow_stats *stats)
628 stats->n_packets = flow->packet_count;
629 stats->n_bytes = flow->byte_count;
630 stats->used = flow->used;
631 stats->tcp_flags = TCP_FLAGS(flow->tcp_ctl);
635 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
638 if (odp_flow_key_to_flow(key, key_len, flow)) {
639 /* This should not happen: it indicates that odp_flow_key_from_flow()
640 * and odp_flow_key_to_flow() disagree on the acceptable form of a
641 * flow. Log the problem as an error, with enough details to enable
643 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
645 if (!VLOG_DROP_ERR(&rl)) {
649 odp_flow_key_format(key, key_len, &s);
650 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
657 if (flow->in_port < OFPP_MAX
658 ? flow->in_port >= MAX_PORTS
659 : flow->in_port != OFPP_LOCAL && flow->in_port != OFPP_NONE) {
667 dpif_netdev_flow_get(const struct dpif *dpif,
668 const struct nlattr *nl_key, size_t nl_key_len,
669 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
671 struct dp_netdev *dp = get_dp_netdev(dpif);
672 struct dp_netdev_flow *flow;
676 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
681 flow = dp_netdev_lookup_flow(dp, &key);
687 get_dpif_flow_stats(flow, stats);
690 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
696 dpif_netdev_validate_actions(const struct nlattr *actions,
697 size_t actions_len, bool *mutates)
699 const struct nlattr *a;
703 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
704 uint16_t type = nl_attr_type(a);
705 int len = odp_action_len(type);
707 if (len != nl_attr_get_size(a)) {
712 case OVS_ACTION_ATTR_OUTPUT:
713 if (nl_attr_get_u32(a) >= MAX_PORTS) {
718 case OVS_ACTION_ATTR_USERSPACE:
721 case OVS_ACTION_ATTR_PUSH_VLAN:
723 if (nl_attr_get_be16(a) & htons(VLAN_CFI)) {
728 case OVS_ACTION_ATTR_SET_NW_TOS:
730 if (nl_attr_get_u8(a) & IP_ECN_MASK) {
735 case OVS_ACTION_ATTR_POP_VLAN:
736 case OVS_ACTION_ATTR_SET_DL_SRC:
737 case OVS_ACTION_ATTR_SET_DL_DST:
738 case OVS_ACTION_ATTR_SET_NW_SRC:
739 case OVS_ACTION_ATTR_SET_NW_DST:
740 case OVS_ACTION_ATTR_SET_TP_SRC:
741 case OVS_ACTION_ATTR_SET_TP_DST:
745 case OVS_ACTION_ATTR_SET_TUNNEL:
746 case OVS_ACTION_ATTR_SET_PRIORITY:
747 case OVS_ACTION_ATTR_POP_PRIORITY:
756 set_flow_actions(struct dp_netdev_flow *flow,
757 const struct nlattr *actions, size_t actions_len)
762 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
767 flow->actions = xrealloc(flow->actions, actions_len);
768 flow->actions_len = actions_len;
769 memcpy(flow->actions, actions, actions_len);
774 add_flow(struct dpif *dpif, const struct flow *key,
775 const struct nlattr *actions, size_t actions_len)
777 struct dp_netdev *dp = get_dp_netdev(dpif);
778 struct dp_netdev_flow *flow;
781 flow = xzalloc(sizeof *flow);
784 error = set_flow_actions(flow, actions, actions_len);
790 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
795 clear_stats(struct dp_netdev_flow *flow)
798 flow->packet_count = 0;
799 flow->byte_count = 0;
804 dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
805 const struct nlattr *nl_key, size_t nl_key_len,
806 const struct nlattr *actions, size_t actions_len,
807 struct dpif_flow_stats *stats)
809 struct dp_netdev *dp = get_dp_netdev(dpif);
810 struct dp_netdev_flow *flow;
814 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
819 flow = dp_netdev_lookup_flow(dp, &key);
821 if (flags & DPIF_FP_CREATE) {
822 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
824 memset(stats, 0, sizeof *stats);
826 return add_flow(dpif, &key, actions, actions_len);
834 if (flags & DPIF_FP_MODIFY) {
835 int error = set_flow_actions(flow, actions, actions_len);
838 get_dpif_flow_stats(flow, stats);
840 if (flags & DPIF_FP_ZERO_STATS) {
852 dpif_netdev_flow_del(struct dpif *dpif,
853 const struct nlattr *nl_key, size_t nl_key_len,
854 struct dpif_flow_stats *stats)
856 struct dp_netdev *dp = get_dp_netdev(dpif);
857 struct dp_netdev_flow *flow;
861 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
866 flow = dp_netdev_lookup_flow(dp, &key);
869 get_dpif_flow_stats(flow, stats);
871 dp_netdev_free_flow(dp, flow);
878 struct dp_netdev_flow_state {
881 struct nlattr *actions;
882 struct odputil_keybuf keybuf;
883 struct dpif_flow_stats stats;
887 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
889 struct dp_netdev_flow_state *state;
891 *statep = state = xmalloc(sizeof *state);
894 state->actions = NULL;
899 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
900 const struct nlattr **key, size_t *key_len,
901 const struct nlattr **actions, size_t *actions_len,
902 const struct dpif_flow_stats **stats)
904 struct dp_netdev_flow_state *state = state_;
905 struct dp_netdev *dp = get_dp_netdev(dpif);
906 struct dp_netdev_flow *flow;
907 struct hmap_node *node;
909 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
914 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
919 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
920 odp_flow_key_from_flow(&buf, &flow->key);
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,
954 const struct nlattr *key_attrs, size_t key_len,
955 const struct nlattr *actions, size_t actions_len,
956 const struct ofpbuf *packet)
958 struct dp_netdev *dp = get_dp_netdev(dpif);
964 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
968 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
974 /* We need a deep copy of 'packet' since we're going to modify its
976 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
977 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
978 ofpbuf_put(©, packet->data, packet->size);
980 /* We still need a shallow copy of 'packet', even though we won't
981 * modify its data, because flow_extract() modifies packet->l2, etc.
982 * We could probably get away with modifying those but it's more polite
987 flow_extract(©, 0, -1, &key);
988 error = dpif_netdev_flow_from_nlattrs(key_attrs, key_len, &key);
990 error = dp_netdev_execute_actions(dp, ©, &key,
991 actions, actions_len);
994 ofpbuf_uninit(©);
1000 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
1002 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1003 *listen_mask = dpif_netdev->listen_mask;
1008 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
1010 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1011 dpif_netdev->listen_mask = listen_mask;
1015 static struct dp_netdev_queue *
1016 find_nonempty_queue(struct dpif *dpif)
1018 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1019 struct dp_netdev *dp = get_dp_netdev(dpif);
1020 int mask = dpif_netdev->listen_mask;
1023 for (i = 0; i < N_QUEUES; i++) {
1024 struct dp_netdev_queue *q = &dp->queues[i];
1025 if (q->head != q->tail && mask & (1u << i)) {
1033 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
1035 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1037 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
1048 dpif_netdev_recv_wait(struct dpif *dpif)
1050 if (find_nonempty_queue(dpif)) {
1051 poll_immediate_wake();
1053 /* No messages ready to be received, and dp_wait() will ensure that we
1054 * wake up to queue new messages, so there is nothing to do. */
1059 dpif_netdev_recv_purge(struct dpif *dpif)
1061 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1062 dp_netdev_purge_queues(dpif_netdev->dp);
1066 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
1067 const struct ofpbuf *packet)
1069 flow->used = time_msec();
1070 flow->packet_count++;
1071 flow->byte_count += packet->size;
1072 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1073 struct tcp_header *th = packet->l4;
1074 flow->tcp_ctl |= th->tcp_ctl;
1079 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1080 struct ofpbuf *packet)
1082 struct dp_netdev_flow *flow;
1085 if (packet->size < ETH_HEADER_LEN) {
1088 if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) {
1093 flow = dp_netdev_lookup_flow(dp, &key);
1095 dp_netdev_flow_used(flow, &key, packet);
1096 dp_netdev_execute_actions(dp, packet, &key,
1097 flow->actions, flow->actions_len);
1101 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1106 dpif_netdev_run(struct dpif *dpif)
1108 struct dp_netdev *dp = get_dp_netdev(dpif);
1109 struct dp_netdev_port *port;
1110 struct ofpbuf packet;
1112 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1114 LIST_FOR_EACH (port, node, &dp->port_list) {
1117 /* Reset packet contents. */
1118 ofpbuf_clear(&packet);
1119 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1121 error = netdev_recv(port->netdev, &packet);
1123 dp_netdev_port_input(dp, port, &packet);
1124 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1125 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1126 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1127 netdev_get_name(port->netdev), strerror(error));
1130 ofpbuf_uninit(&packet);
1134 dpif_netdev_wait(struct dpif *dpif)
1136 struct dp_netdev *dp = get_dp_netdev(dpif);
1137 struct dp_netdev_port *port;
1139 LIST_FOR_EACH (port, node, &dp->port_list) {
1140 netdev_recv_wait(port->netdev);
1145 dp_netdev_pop_vlan(struct ofpbuf *packet)
1147 struct vlan_eth_header *veh = packet->l2;
1148 if (packet->size >= sizeof *veh
1149 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
1150 struct eth_header tmp;
1152 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1153 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1154 tmp.eth_type = veh->veth_next_type;
1156 ofpbuf_pull(packet, VLAN_HEADER_LEN);
1157 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
1158 memcpy(packet->data, &tmp, sizeof tmp);
1163 dp_netdev_set_dl_src(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1165 struct eth_header *eh = packet->l2;
1166 memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
1170 dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1172 struct eth_header *eh = packet->l2;
1173 memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
1177 is_ip(const struct ofpbuf *packet, const struct flow *key)
1179 return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
1183 dp_netdev_set_nw_addr(struct ofpbuf *packet, const struct flow *key,
1184 const struct nlattr *a)
1186 if (is_ip(packet, key)) {
1187 struct ip_header *nh = packet->l3;
1188 ovs_be32 ip = nl_attr_get_be32(a);
1189 uint16_t type = nl_attr_type(a);
1192 field = type == OVS_ACTION_ATTR_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
1193 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1194 struct tcp_header *th = packet->l4;
1195 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, ip);
1196 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1197 struct udp_header *uh = packet->l4;
1199 uh->udp_csum = recalc_csum32(uh->udp_csum, *field, ip);
1200 if (!uh->udp_csum) {
1201 uh->udp_csum = htons(0xffff);
1205 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, ip);
1211 dp_netdev_set_nw_tos(struct ofpbuf *packet, const struct flow *key,
1214 if (is_ip(packet, key)) {
1215 struct ip_header *nh = packet->l3;
1216 uint8_t *field = &nh->ip_tos;
1218 /* Set the DSCP bits and preserve the ECN bits. */
1219 uint8_t new = nw_tos | (nh->ip_tos & IP_ECN_MASK);
1221 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
1222 htons((uint16_t) new));
1228 dp_netdev_set_tp_port(struct ofpbuf *packet, const struct flow *key,
1229 const struct nlattr *a)
1231 if (is_ip(packet, key)) {
1232 uint16_t type = nl_attr_type(a);
1233 ovs_be16 port = nl_attr_get_be16(a);
1236 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1237 struct tcp_header *th = packet->l4;
1238 field = (type == OVS_ACTION_ATTR_SET_TP_SRC
1239 ? &th->tcp_src : &th->tcp_dst);
1240 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, port);
1242 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1243 struct udp_header *uh = packet->l4;
1244 field = (type == OVS_ACTION_ATTR_SET_TP_SRC
1245 ? &uh->udp_src : &uh->udp_dst);
1246 uh->udp_csum = recalc_csum16(uh->udp_csum, *field, port);
1255 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1258 struct dp_netdev_port *p = dp->ports[out_port];
1260 netdev_send(p->netdev, packet);
1265 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1266 int queue_no, const struct flow *flow, uint64_t arg)
1268 struct dp_netdev_queue *q = &dp->queues[queue_no];
1269 struct dpif_upcall *upcall;
1273 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1278 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1279 odp_flow_key_from_flow(buf, flow);
1280 key_len = buf->size;
1281 ofpbuf_pull(buf, key_len);
1282 ofpbuf_reserve(buf, 2);
1283 ofpbuf_put(buf, packet->data, packet->size);
1285 upcall = xzalloc(sizeof *upcall);
1286 upcall->type = queue_no;
1287 upcall->packet = buf;
1288 upcall->key = buf->base;
1289 upcall->key_len = key_len;
1290 upcall->userdata = arg;
1292 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1298 dp_netdev_execute_actions(struct dp_netdev *dp,
1299 struct ofpbuf *packet, struct flow *key,
1300 const struct nlattr *actions,
1303 const struct nlattr *a;
1306 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1307 switch (nl_attr_type(a)) {
1308 case OVS_ACTION_ATTR_OUTPUT:
1309 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1312 case OVS_ACTION_ATTR_USERSPACE:
1313 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION,
1314 key, nl_attr_get_u64(a));
1317 case OVS_ACTION_ATTR_PUSH_VLAN:
1318 eth_push_vlan(packet, nl_attr_get_be16(a));
1321 case OVS_ACTION_ATTR_POP_VLAN:
1322 dp_netdev_pop_vlan(packet);
1325 case OVS_ACTION_ATTR_SET_DL_SRC:
1326 dp_netdev_set_dl_src(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1329 case OVS_ACTION_ATTR_SET_DL_DST:
1330 dp_netdev_set_dl_dst(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1333 case OVS_ACTION_ATTR_SET_NW_SRC:
1334 case OVS_ACTION_ATTR_SET_NW_DST:
1335 dp_netdev_set_nw_addr(packet, key, a);
1338 case OVS_ACTION_ATTR_SET_NW_TOS:
1339 dp_netdev_set_nw_tos(packet, key, nl_attr_get_u8(a));
1342 case OVS_ACTION_ATTR_SET_TP_SRC:
1343 case OVS_ACTION_ATTR_SET_TP_DST:
1344 dp_netdev_set_tp_port(packet, key, a);
1351 const struct dpif_class dpif_netdev_class = {
1353 NULL, /* enumerate */
1356 dpif_netdev_destroy,
1359 dpif_netdev_get_stats,
1360 dpif_netdev_get_drop_frags,
1361 dpif_netdev_set_drop_frags,
1362 dpif_netdev_port_add,
1363 dpif_netdev_port_del,
1364 dpif_netdev_port_query_by_number,
1365 dpif_netdev_port_query_by_name,
1366 dpif_netdev_get_max_ports,
1367 dpif_netdev_port_dump_start,
1368 dpif_netdev_port_dump_next,
1369 dpif_netdev_port_dump_done,
1370 dpif_netdev_port_poll,
1371 dpif_netdev_port_poll_wait,
1372 dpif_netdev_flow_get,
1373 dpif_netdev_flow_put,
1374 dpif_netdev_flow_del,
1375 dpif_netdev_flow_flush,
1376 dpif_netdev_flow_dump_start,
1377 dpif_netdev_flow_dump_next,
1378 dpif_netdev_flow_dump_done,
1379 dpif_netdev_execute,
1380 dpif_netdev_recv_get_mask,
1381 dpif_netdev_recv_set_mask,
1382 NULL, /* get_sflow_probability */
1383 NULL, /* set_sflow_probability */
1384 NULL, /* queue_to_priority */
1386 dpif_netdev_recv_wait,
1387 dpif_netdev_recv_purge,
1391 dpif_dummy_register(void)
1393 if (!dpif_dummy_class.type) {
1394 dpif_dummy_class = dpif_netdev_class;
1395 dpif_dummy_class.type = "dummy";
1396 dp_register_provider(&dpif_dummy_class);