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", ODPP_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 odp_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 netdev_get_mtu(netdev, &mtu);
382 if (mtu != INT_MAX && mtu > max_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 == ODPP_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));
661 dpif_netdev_flow_get(const struct dpif *dpif,
662 const struct nlattr *nl_key, size_t nl_key_len,
663 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
665 struct dp_netdev *dp = get_dp_netdev(dpif);
666 struct dp_netdev_flow *flow;
670 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
675 flow = dp_netdev_lookup_flow(dp, &key);
681 get_dpif_flow_stats(flow, stats);
684 *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
690 dpif_netdev_validate_actions(const struct nlattr *actions,
691 size_t actions_len, bool *mutates)
693 const struct nlattr *a;
697 NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
698 uint16_t type = nl_attr_type(a);
699 int len = odp_action_len(type);
701 if (len != nl_attr_get_size(a)) {
706 case ODP_ACTION_ATTR_OUTPUT:
707 if (nl_attr_get_u32(a) >= MAX_PORTS) {
712 case ODP_ACTION_ATTR_USERSPACE:
715 case ODP_ACTION_ATTR_SET_DL_TCI:
717 if (nl_attr_get_be16(a) & htons(VLAN_CFI)) {
722 case ODP_ACTION_ATTR_SET_NW_TOS:
724 if (nl_attr_get_u8(a) & IP_ECN_MASK) {
729 case ODP_ACTION_ATTR_STRIP_VLAN:
730 case ODP_ACTION_ATTR_SET_DL_SRC:
731 case ODP_ACTION_ATTR_SET_DL_DST:
732 case ODP_ACTION_ATTR_SET_NW_SRC:
733 case ODP_ACTION_ATTR_SET_NW_DST:
734 case ODP_ACTION_ATTR_SET_TP_SRC:
735 case ODP_ACTION_ATTR_SET_TP_DST:
739 case ODP_ACTION_ATTR_SET_TUNNEL:
740 case ODP_ACTION_ATTR_SET_PRIORITY:
741 case ODP_ACTION_ATTR_POP_PRIORITY:
750 set_flow_actions(struct dp_netdev_flow *flow,
751 const struct nlattr *actions, size_t actions_len)
756 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
761 flow->actions = xrealloc(flow->actions, actions_len);
762 flow->actions_len = actions_len;
763 memcpy(flow->actions, actions, actions_len);
768 add_flow(struct dpif *dpif, const struct flow *key,
769 const struct nlattr *actions, size_t actions_len)
771 struct dp_netdev *dp = get_dp_netdev(dpif);
772 struct dp_netdev_flow *flow;
775 flow = xzalloc(sizeof *flow);
778 error = set_flow_actions(flow, actions, actions_len);
784 hmap_insert(&dp->flow_table, &flow->node, flow_hash(&flow->key, 0));
789 clear_stats(struct dp_netdev_flow *flow)
792 flow->packet_count = 0;
793 flow->byte_count = 0;
798 dpif_netdev_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
799 const struct nlattr *nl_key, size_t nl_key_len,
800 const struct nlattr *actions, size_t actions_len,
801 struct dpif_flow_stats *stats)
803 struct dp_netdev *dp = get_dp_netdev(dpif);
804 struct dp_netdev_flow *flow;
808 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
813 flow = dp_netdev_lookup_flow(dp, &key);
815 if (flags & DPIF_FP_CREATE) {
816 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
818 memset(stats, 0, sizeof *stats);
820 return add_flow(dpif, &key, actions, actions_len);
828 if (flags & DPIF_FP_MODIFY) {
829 int error = set_flow_actions(flow, actions, actions_len);
832 get_dpif_flow_stats(flow, stats);
834 if (flags & DPIF_FP_ZERO_STATS) {
846 dpif_netdev_flow_del(struct dpif *dpif,
847 const struct nlattr *nl_key, size_t nl_key_len,
848 struct dpif_flow_stats *stats)
850 struct dp_netdev *dp = get_dp_netdev(dpif);
851 struct dp_netdev_flow *flow;
855 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
860 flow = dp_netdev_lookup_flow(dp, &key);
863 get_dpif_flow_stats(flow, 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 **actions, size_t *actions_len,
896 const struct dpif_flow_stats **stats)
898 struct dp_netdev_flow_state *state = state_;
899 struct dp_netdev *dp = get_dp_netdev(dpif);
900 struct dp_netdev_flow *flow;
901 struct hmap_node *node;
903 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
908 flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
913 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
914 odp_flow_key_from_flow(&buf, &flow->key);
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 *key_attrs, size_t key_len,
949 const struct nlattr *actions, size_t actions_len,
950 const struct ofpbuf *packet)
952 struct dp_netdev *dp = get_dp_netdev(dpif);
958 if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
962 error = dpif_netdev_validate_actions(actions, actions_len, &mutates);
968 /* We need a deep copy of 'packet' since we're going to modify its
970 ofpbuf_init(©, DP_NETDEV_HEADROOM + packet->size);
971 ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
972 ofpbuf_put(©, packet->data, packet->size);
974 /* We still need a shallow copy of 'packet', even though we won't
975 * modify its data, because flow_extract() modifies packet->l2, etc.
976 * We could probably get away with modifying those but it's more polite
981 flow_extract(©, 0, -1, &key);
982 dpif_netdev_flow_from_nlattrs(key_attrs, key_len, &key);
984 error = dp_netdev_execute_actions(dp, ©, &key, actions, actions_len);
986 ofpbuf_uninit(©);
992 dpif_netdev_recv_get_mask(const struct dpif *dpif, int *listen_mask)
994 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
995 *listen_mask = dpif_netdev->listen_mask;
1000 dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
1002 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1003 dpif_netdev->listen_mask = listen_mask;
1007 static struct dp_netdev_queue *
1008 find_nonempty_queue(struct dpif *dpif)
1010 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1011 struct dp_netdev *dp = get_dp_netdev(dpif);
1012 int mask = dpif_netdev->listen_mask;
1015 for (i = 0; i < N_QUEUES; i++) {
1016 struct dp_netdev_queue *q = &dp->queues[i];
1017 if (q->head != q->tail && mask & (1u << i)) {
1025 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall)
1027 struct dp_netdev_queue *q = find_nonempty_queue(dpif);
1029 struct dpif_upcall *u = q->upcalls[q->tail++ & QUEUE_MASK];
1040 dpif_netdev_recv_wait(struct dpif *dpif)
1042 if (find_nonempty_queue(dpif)) {
1043 poll_immediate_wake();
1045 /* No messages ready to be received, and dp_wait() will ensure that we
1046 * wake up to queue new messages, so there is nothing to do. */
1051 dpif_netdev_recv_purge(struct dpif *dpif)
1053 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1054 dp_netdev_purge_queues(dpif_netdev->dp);
1058 dp_netdev_flow_used(struct dp_netdev_flow *flow, struct flow *key,
1059 const struct ofpbuf *packet)
1061 flow->used = time_msec();
1062 flow->packet_count++;
1063 flow->byte_count += packet->size;
1064 if (key->dl_type == htons(ETH_TYPE_IP) && key->nw_proto == IPPROTO_TCP) {
1065 struct tcp_header *th = packet->l4;
1066 flow->tcp_ctl |= th->tcp_ctl;
1071 dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
1072 struct ofpbuf *packet)
1074 struct dp_netdev_flow *flow;
1077 if (packet->size < ETH_HEADER_LEN) {
1080 if (flow_extract(packet, 0, port->port_no, &key) && dp->drop_frags) {
1085 flow = dp_netdev_lookup_flow(dp, &key);
1087 dp_netdev_flow_used(flow, &key, packet);
1088 dp_netdev_execute_actions(dp, packet, &key,
1089 flow->actions, flow->actions_len);
1093 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, 0);
1098 dpif_netdev_run(struct dpif *dpif)
1100 struct dp_netdev *dp = get_dp_netdev(dpif);
1101 struct dp_netdev_port *port;
1102 struct ofpbuf packet;
1104 ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
1106 LIST_FOR_EACH (port, node, &dp->port_list) {
1109 /* Reset packet contents. */
1110 ofpbuf_clear(&packet);
1111 ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
1113 error = netdev_recv(port->netdev, &packet);
1115 dp_netdev_port_input(dp, port, &packet);
1116 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1118 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1119 netdev_get_name(port->netdev), strerror(error));
1122 ofpbuf_uninit(&packet);
1126 dpif_netdev_wait(struct dpif *dpif)
1128 struct dp_netdev *dp = get_dp_netdev(dpif);
1129 struct dp_netdev_port *port;
1131 LIST_FOR_EACH (port, node, &dp->port_list) {
1132 netdev_recv_wait(port->netdev);
1137 dp_netdev_strip_vlan(struct ofpbuf *packet)
1139 struct vlan_eth_header *veh = packet->l2;
1140 if (packet->size >= sizeof *veh
1141 && veh->veth_type == htons(ETH_TYPE_VLAN)) {
1142 struct eth_header tmp;
1144 memcpy(tmp.eth_dst, veh->veth_dst, ETH_ADDR_LEN);
1145 memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
1146 tmp.eth_type = veh->veth_next_type;
1148 ofpbuf_pull(packet, VLAN_HEADER_LEN);
1149 packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
1150 memcpy(packet->data, &tmp, sizeof tmp);
1155 dp_netdev_set_dl_src(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1157 struct eth_header *eh = packet->l2;
1158 memcpy(eh->eth_src, dl_addr, sizeof eh->eth_src);
1162 dp_netdev_set_dl_dst(struct ofpbuf *packet, const uint8_t dl_addr[ETH_ADDR_LEN])
1164 struct eth_header *eh = packet->l2;
1165 memcpy(eh->eth_dst, dl_addr, sizeof eh->eth_dst);
1169 is_ip(const struct ofpbuf *packet, const struct flow *key)
1171 return key->dl_type == htons(ETH_TYPE_IP) && packet->l4;
1175 dp_netdev_set_nw_addr(struct ofpbuf *packet, const struct flow *key,
1176 const struct nlattr *a)
1178 if (is_ip(packet, key)) {
1179 struct ip_header *nh = packet->l3;
1180 ovs_be32 ip = nl_attr_get_be32(a);
1181 uint16_t type = nl_attr_type(a);
1184 field = type == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->ip_src : &nh->ip_dst;
1185 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1186 struct tcp_header *th = packet->l4;
1187 th->tcp_csum = recalc_csum32(th->tcp_csum, *field, ip);
1188 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1189 struct udp_header *uh = packet->l4;
1191 uh->udp_csum = recalc_csum32(uh->udp_csum, *field, ip);
1192 if (!uh->udp_csum) {
1193 uh->udp_csum = htons(0xffff);
1197 nh->ip_csum = recalc_csum32(nh->ip_csum, *field, ip);
1203 dp_netdev_set_nw_tos(struct ofpbuf *packet, const struct flow *key,
1206 if (is_ip(packet, key)) {
1207 struct ip_header *nh = packet->l3;
1208 uint8_t *field = &nh->ip_tos;
1210 /* Set the DSCP bits and preserve the ECN bits. */
1211 uint8_t new = nw_tos | (nh->ip_tos & IP_ECN_MASK);
1213 nh->ip_csum = recalc_csum16(nh->ip_csum, htons((uint16_t)*field),
1214 htons((uint16_t) new));
1220 dp_netdev_set_tp_port(struct ofpbuf *packet, const struct flow *key,
1221 const struct nlattr *a)
1223 if (is_ip(packet, key)) {
1224 uint16_t type = nl_attr_type(a);
1225 ovs_be16 port = nl_attr_get_be16(a);
1228 if (key->nw_proto == IPPROTO_TCP && packet->l7) {
1229 struct tcp_header *th = packet->l4;
1230 field = (type == ODP_ACTION_ATTR_SET_TP_SRC
1231 ? &th->tcp_src : &th->tcp_dst);
1232 th->tcp_csum = recalc_csum16(th->tcp_csum, *field, port);
1234 } else if (key->nw_proto == IPPROTO_UDP && packet->l7) {
1235 struct udp_header *uh = packet->l4;
1236 field = (type == ODP_ACTION_ATTR_SET_TP_SRC
1237 ? &uh->udp_src : &uh->udp_dst);
1238 uh->udp_csum = recalc_csum16(uh->udp_csum, *field, port);
1247 dp_netdev_output_port(struct dp_netdev *dp, struct ofpbuf *packet,
1250 struct dp_netdev_port *p = dp->ports[out_port];
1252 netdev_send(p->netdev, packet);
1257 dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
1258 int queue_no, const struct flow *flow, uint64_t arg)
1260 struct dp_netdev_queue *q = &dp->queues[queue_no];
1261 struct dpif_upcall *upcall;
1265 if (q->head - q->tail >= MAX_QUEUE_LEN) {
1270 buf = ofpbuf_new(ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size);
1271 odp_flow_key_from_flow(buf, flow);
1272 key_len = buf->size;
1273 ofpbuf_pull(buf, key_len);
1274 ofpbuf_reserve(buf, 2);
1275 ofpbuf_put(buf, packet->data, packet->size);
1277 upcall = xzalloc(sizeof *upcall);
1278 upcall->type = queue_no;
1279 upcall->packet = buf;
1280 upcall->key = buf->base;
1281 upcall->key_len = key_len;
1282 upcall->userdata = arg;
1284 q->upcalls[q->head++ & QUEUE_MASK] = upcall;
1290 dp_netdev_execute_actions(struct dp_netdev *dp,
1291 struct ofpbuf *packet, struct flow *key,
1292 const struct nlattr *actions,
1295 const struct nlattr *a;
1298 NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
1299 switch (nl_attr_type(a)) {
1300 case ODP_ACTION_ATTR_OUTPUT:
1301 dp_netdev_output_port(dp, packet, nl_attr_get_u32(a));
1304 case ODP_ACTION_ATTR_USERSPACE:
1305 dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION,
1306 key, nl_attr_get_u64(a));
1309 case ODP_ACTION_ATTR_SET_DL_TCI:
1310 eth_set_vlan_tci(packet, nl_attr_get_be16(a));
1313 case ODP_ACTION_ATTR_STRIP_VLAN:
1314 dp_netdev_strip_vlan(packet);
1317 case ODP_ACTION_ATTR_SET_DL_SRC:
1318 dp_netdev_set_dl_src(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1321 case ODP_ACTION_ATTR_SET_DL_DST:
1322 dp_netdev_set_dl_dst(packet, nl_attr_get_unspec(a, ETH_ADDR_LEN));
1325 case ODP_ACTION_ATTR_SET_NW_SRC:
1326 case ODP_ACTION_ATTR_SET_NW_DST:
1327 dp_netdev_set_nw_addr(packet, key, a);
1330 case ODP_ACTION_ATTR_SET_NW_TOS:
1331 dp_netdev_set_nw_tos(packet, key, nl_attr_get_u8(a));
1334 case ODP_ACTION_ATTR_SET_TP_SRC:
1335 case ODP_ACTION_ATTR_SET_TP_DST:
1336 dp_netdev_set_tp_port(packet, key, a);
1343 const struct dpif_class dpif_netdev_class = {
1345 NULL, /* enumerate */
1348 dpif_netdev_destroy,
1351 dpif_netdev_get_stats,
1352 dpif_netdev_get_drop_frags,
1353 dpif_netdev_set_drop_frags,
1354 dpif_netdev_port_add,
1355 dpif_netdev_port_del,
1356 dpif_netdev_port_query_by_number,
1357 dpif_netdev_port_query_by_name,
1358 dpif_netdev_get_max_ports,
1359 dpif_netdev_port_dump_start,
1360 dpif_netdev_port_dump_next,
1361 dpif_netdev_port_dump_done,
1362 dpif_netdev_port_poll,
1363 dpif_netdev_port_poll_wait,
1364 dpif_netdev_flow_get,
1365 dpif_netdev_flow_put,
1366 dpif_netdev_flow_del,
1367 dpif_netdev_flow_flush,
1368 dpif_netdev_flow_dump_start,
1369 dpif_netdev_flow_dump_next,
1370 dpif_netdev_flow_dump_done,
1371 dpif_netdev_execute,
1372 dpif_netdev_recv_get_mask,
1373 dpif_netdev_recv_set_mask,
1374 NULL, /* get_sflow_probability */
1375 NULL, /* set_sflow_probability */
1376 NULL, /* queue_to_priority */
1378 dpif_netdev_recv_wait,
1379 dpif_netdev_recv_purge,
1383 dpif_dummy_register(void)
1385 if (!dpif_dummy_class.type) {
1386 dpif_dummy_class = dpif_netdev_class;
1387 dpif_dummy_class.type = "dummy";
1388 dp_register_provider(&dpif_dummy_class);