2 * Copyright (c) 2008, 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.
19 #include "dpif-linux.h"
27 #include <linux/types.h>
28 #include <linux/pkt_sched.h>
29 #include <linux/rtnetlink.h>
30 #include <linux/sockios.h>
36 #include "dpif-provider.h"
37 #include "dynamic-string.h"
40 #include "netdev-linux.h"
41 #include "netdev-vport.h"
42 #include "netlink-socket.h"
46 #include "openvswitch/tunnel.h"
48 #include "poll-loop.h"
49 #include "rtnetlink.h"
50 #include "rtnetlink-link.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(dpif_linux);
59 enum { LRU_MAX_PORTS = 1024 };
60 enum { LRU_MASK = LRU_MAX_PORTS - 1};
61 BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
63 struct dpif_linux_dp {
64 /* Generic Netlink header. */
67 /* struct ovs_header. */
71 const char *name; /* OVS_DP_ATTR_NAME. */
72 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
73 enum ovs_frag_handling ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */
74 const uint32_t *sampling; /* OVS_DP_ATTR_SAMPLING. */
75 uint32_t mcgroups[DPIF_N_UC_TYPES]; /* OVS_DP_ATTR_MCGROUPS. */
78 static void dpif_linux_dp_init(struct dpif_linux_dp *);
79 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
80 const struct ofpbuf *);
81 static void dpif_linux_dp_dump_start(struct nl_dump *);
82 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
83 struct dpif_linux_dp *reply,
84 struct ofpbuf **bufp);
85 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
86 struct ofpbuf **bufp);
88 struct dpif_linux_flow {
89 /* Generic Netlink header. */
92 /* struct ovs_header. */
93 unsigned int nlmsg_flags;
98 * The 'stats' and 'used' members point to 64-bit data that might only be
99 * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
100 * access their values.
102 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
103 * the Netlink version of the command, even if actions_len is zero. */
104 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
106 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
108 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
109 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
110 const uint64_t *used; /* OVS_FLOW_ATTR_USED. */
111 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
114 static void dpif_linux_flow_init(struct dpif_linux_flow *);
115 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
116 const struct ofpbuf *);
117 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
119 static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
120 struct dpif_linux_flow *reply,
121 struct ofpbuf **bufp);
122 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
123 struct dpif_flow_stats *);
125 /* Datapath interface for the openvswitch Linux kernel module. */
130 /* Multicast group messages. */
131 struct nl_sock *mc_sock;
132 uint32_t mcgroups[DPIF_N_UC_TYPES];
133 unsigned int listen_mask;
135 /* Change notification. */
136 struct sset changed_ports; /* Ports that have changed. */
137 struct rtnetlink_notifier port_notifier;
140 /* Queue of unused ports. */
141 unsigned long *lru_bitmap;
142 uint16_t lru_ports[LRU_MAX_PORTS];
147 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
149 /* Generic Netlink family numbers for OVS. */
150 static int ovs_datapath_family;
151 static int ovs_vport_family;
152 static int ovs_flow_family;
153 static int ovs_packet_family;
155 /* Generic Netlink socket. */
156 static struct nl_sock *genl_sock;
158 static int dpif_linux_init(void);
159 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
160 static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
163 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
165 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
166 const struct ofpbuf *);
168 static struct dpif_linux *
169 dpif_linux_cast(const struct dpif *dpif)
171 dpif_assert_class(dpif, &dpif_linux_class);
172 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
176 dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
178 if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
179 bitmap_set1(dp->lru_bitmap, port);
180 dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
185 dpif_linux_pop_port(struct dpif_linux *dp)
189 if (dp->lru_head == dp->lru_tail) {
193 port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
194 bitmap_set0(dp->lru_bitmap, port);
199 dpif_linux_enumerate(struct sset *all_dps)
205 error = dpif_linux_init();
210 dpif_linux_dp_dump_start(&dump);
211 while (nl_dump_next(&dump, &msg)) {
212 struct dpif_linux_dp dp;
214 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
215 sset_add(all_dps, dp.name);
218 return nl_dump_done(&dump);
222 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
223 bool create, struct dpif **dpifp)
225 struct dpif_linux_dp dp_request, dp;
229 error = dpif_linux_init();
234 /* Create or look up datapath. */
235 dpif_linux_dp_init(&dp_request);
236 dp_request.cmd = create ? OVS_DP_CMD_NEW : OVS_DP_CMD_GET;
237 dp_request.name = name;
238 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
242 error = open_dpif(&dp, dpifp);
249 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
251 struct dpif_linux *dpif;
255 dpif = xmalloc(sizeof *dpif);
256 error = rtnetlink_link_notifier_register(&dpif->port_notifier,
257 dpif_linux_port_changed, dpif);
262 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
263 dp->dp_ifindex, dp->dp_ifindex);
265 dpif->mc_sock = NULL;
266 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
267 dpif->mcgroups[i] = dp->mcgroups[i];
269 dpif->listen_mask = 0;
270 dpif->dp_ifindex = dp->dp_ifindex;
271 sset_init(&dpif->changed_ports);
272 dpif->change_error = false;
273 *dpifp = &dpif->dpif;
275 dpif->lru_head = dpif->lru_tail = 0;
276 dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
277 bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL);
278 for (i = 1; i < LRU_MAX_PORTS; i++) {
279 dpif_linux_push_port(dpif, i);
289 dpif_linux_close(struct dpif *dpif_)
291 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
292 nl_sock_destroy(dpif->mc_sock);
293 rtnetlink_link_notifier_unregister(&dpif->port_notifier);
294 sset_destroy(&dpif->changed_ports);
295 free(dpif->lru_bitmap);
300 dpif_linux_destroy(struct dpif *dpif_)
302 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
303 struct dpif_linux_dp dp;
305 dpif_linux_dp_init(&dp);
306 dp.cmd = OVS_DP_CMD_DEL;
307 dp.dp_ifindex = dpif->dp_ifindex;
308 return dpif_linux_dp_transact(&dp, NULL, NULL);
312 dpif_linux_run(struct dpif *dpif OVS_UNUSED)
314 rtnetlink_link_notifier_run();
318 dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
320 rtnetlink_link_notifier_wait();
324 dpif_linux_get_stats(const struct dpif *dpif_, struct ovs_dp_stats *stats)
326 struct dpif_linux_dp dp;
330 error = dpif_linux_dp_get(dpif_, &dp, &buf);
339 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
341 struct dpif_linux_dp dp;
345 error = dpif_linux_dp_get(dpif_, &dp, &buf);
347 *drop_fragsp = dp.ipv4_frags == OVS_DP_FRAG_DROP;
354 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
356 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
357 struct dpif_linux_dp dp;
359 dpif_linux_dp_init(&dp);
360 dp.cmd = OVS_DP_CMD_SET;
361 dp.dp_ifindex = dpif->dp_ifindex;
362 dp.ipv4_frags = drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO;
363 return dpif_linux_dp_transact(&dp, NULL, NULL);
367 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
370 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
371 const char *name = netdev_get_name(netdev);
372 const char *type = netdev_get_type(netdev);
373 struct dpif_linux_vport request, reply;
374 const struct ofpbuf *options;
378 dpif_linux_vport_init(&request);
379 request.cmd = OVS_VPORT_CMD_NEW;
380 request.dp_ifindex = dpif->dp_ifindex;
381 request.type = netdev_vport_get_vport_type(netdev);
382 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
383 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
384 "unsupported type `%s'",
385 dpif_name(dpif_), name, type);
390 options = netdev_vport_get_options(netdev);
391 if (options && options->size) {
392 request.options = options->data;
393 request.options_len = options->size;
396 /* Loop until we find a port that isn't used. */
398 request.port_no = dpif_linux_pop_port(dpif);
399 error = dpif_linux_vport_transact(&request, &reply, &buf);
402 *port_nop = reply.port_no;
405 } while (request.port_no != UINT32_MAX
406 && (error == EBUSY || error == EFBIG));
412 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
414 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
415 struct dpif_linux_vport vport;
418 dpif_linux_vport_init(&vport);
419 vport.cmd = OVS_VPORT_CMD_DEL;
420 vport.dp_ifindex = dpif->dp_ifindex;
421 vport.port_no = port_no;
422 error = dpif_linux_vport_transact(&vport, NULL, NULL);
425 dpif_linux_push_port(dpif, port_no);
431 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
432 const char *port_name, struct dpif_port *dpif_port)
434 struct dpif_linux_vport request;
435 struct dpif_linux_vport reply;
439 dpif_linux_vport_init(&request);
440 request.cmd = OVS_VPORT_CMD_GET;
441 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
442 request.port_no = port_no;
443 request.name = port_name;
445 error = dpif_linux_vport_transact(&request, &reply, &buf);
447 dpif_port->name = xstrdup(reply.name);
448 dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
449 dpif_port->port_no = reply.port_no;
451 netdev_stats_from_rtnl_link_stats64(&dpif_port->stats,
454 memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
462 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
463 struct dpif_port *dpif_port)
465 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
469 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
470 struct dpif_port *dpif_port)
472 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
476 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
478 /* If the datapath increases its range of supported ports, then it should
479 * start reporting that. */
484 dpif_linux_flow_flush(struct dpif *dpif_)
486 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
487 struct dpif_linux_flow flow;
489 dpif_linux_flow_init(&flow);
490 flow.cmd = OVS_FLOW_CMD_DEL;
491 flow.dp_ifindex = dpif->dp_ifindex;
492 return dpif_linux_flow_transact(&flow, NULL, NULL);
495 struct dpif_linux_port_state {
497 unsigned long *port_bitmap; /* Ports in the datapath. */
498 bool complete; /* Dump completed without error. */
502 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
504 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
505 struct dpif_linux_port_state *state;
506 struct dpif_linux_vport request;
509 *statep = state = xmalloc(sizeof *state);
510 state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
511 state->complete = false;
513 dpif_linux_vport_init(&request);
514 request.cmd = OVS_DP_CMD_GET;
515 request.dp_ifindex = dpif->dp_ifindex;
517 buf = ofpbuf_new(1024);
518 dpif_linux_vport_to_ofpbuf(&request, buf);
519 nl_dump_start(&state->dump, genl_sock, buf);
526 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
527 struct dpif_port *dpif_port)
529 struct dpif_linux_port_state *state = state_;
530 struct dpif_linux_vport vport;
534 if (!nl_dump_next(&state->dump, &buf)) {
535 state->complete = true;
539 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
544 if (vport.port_no < LRU_MAX_PORTS) {
545 bitmap_set1(state->port_bitmap, vport.port_no);
548 dpif_port->name = (char *) vport.name;
549 dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
550 dpif_port->port_no = vport.port_no;
552 netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, vport.stats);
554 memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
560 dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
562 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
563 struct dpif_linux_port_state *state = state_;
564 int error = nl_dump_done(&state->dump);
566 if (state->complete) {
569 for (i = 0; i < LRU_MAX_PORTS; i++) {
570 if (!bitmap_is_set(state->port_bitmap, i)) {
571 dpif_linux_push_port(dpif, i);
576 free(state->port_bitmap);
582 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
584 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
586 if (dpif->change_error) {
587 dpif->change_error = false;
588 sset_clear(&dpif->changed_ports);
590 } else if (!sset_is_empty(&dpif->changed_ports)) {
591 *devnamep = sset_pop(&dpif->changed_ports);
599 dpif_linux_port_poll_wait(const struct dpif *dpif_)
601 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
602 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
603 poll_immediate_wake();
608 dpif_linux_flow_get__(const struct dpif *dpif_,
609 const struct nlattr *key, size_t key_len,
610 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
612 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
613 struct dpif_linux_flow request;
615 dpif_linux_flow_init(&request);
616 request.cmd = OVS_FLOW_CMD_GET;
617 request.dp_ifindex = dpif->dp_ifindex;
619 request.key_len = key_len;
620 return dpif_linux_flow_transact(&request, reply, bufp);
624 dpif_linux_flow_get(const struct dpif *dpif_,
625 const struct nlattr *key, size_t key_len,
626 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
628 struct dpif_linux_flow reply;
632 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
635 dpif_linux_flow_get_stats(&reply, stats);
638 buf->data = (void *) reply.actions;
639 buf->size = reply.actions_len;
649 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
650 const struct nlattr *key, size_t key_len,
651 const struct nlattr *actions, size_t actions_len,
652 struct dpif_flow_stats *stats)
654 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
655 struct dpif_linux_flow request, reply;
656 struct nlattr dummy_action;
660 dpif_linux_flow_init(&request);
661 request.cmd = flags & DPIF_FP_CREATE ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET;
662 request.dp_ifindex = dpif->dp_ifindex;
664 request.key_len = key_len;
665 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
666 request.actions = actions ? actions : &dummy_action;
667 request.actions_len = actions_len;
668 if (flags & DPIF_FP_ZERO_STATS) {
669 request.clear = true;
671 request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
672 error = dpif_linux_flow_transact(&request,
673 stats ? &reply : NULL,
674 stats ? &buf : NULL);
675 if (!error && stats) {
676 dpif_linux_flow_get_stats(&reply, stats);
683 dpif_linux_flow_del(struct dpif *dpif_,
684 const struct nlattr *key, size_t key_len,
685 struct dpif_flow_stats *stats)
687 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
688 struct dpif_linux_flow request, reply;
692 dpif_linux_flow_init(&request);
693 request.cmd = OVS_FLOW_CMD_DEL;
694 request.dp_ifindex = dpif->dp_ifindex;
696 request.key_len = key_len;
697 error = dpif_linux_flow_transact(&request,
698 stats ? &reply : NULL,
699 stats ? &buf : NULL);
700 if (!error && stats) {
701 dpif_linux_flow_get_stats(&reply, stats);
707 struct dpif_linux_flow_state {
709 struct dpif_linux_flow flow;
710 struct dpif_flow_stats stats;
715 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
717 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
718 struct dpif_linux_flow_state *state;
719 struct dpif_linux_flow request;
722 *statep = state = xmalloc(sizeof *state);
724 dpif_linux_flow_init(&request);
725 request.cmd = OVS_DP_CMD_GET;
726 request.dp_ifindex = dpif->dp_ifindex;
728 buf = ofpbuf_new(1024);
729 dpif_linux_flow_to_ofpbuf(&request, buf);
730 nl_dump_start(&state->dump, genl_sock, buf);
739 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
740 const struct nlattr **key, size_t *key_len,
741 const struct nlattr **actions, size_t *actions_len,
742 const struct dpif_flow_stats **stats)
744 struct dpif_linux_flow_state *state = state_;
749 ofpbuf_delete(state->buf);
752 if (!nl_dump_next(&state->dump, &buf)) {
756 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
761 if (actions && !state->flow.actions) {
762 error = dpif_linux_flow_get__(dpif_, state->flow.key,
764 &state->flow, &state->buf);
765 if (error == ENOENT) {
766 VLOG_DBG("dumped flow disappeared on get");
768 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
774 *actions = state->flow.actions;
775 *actions_len = state->flow.actions_len;
778 *key = state->flow.key;
779 *key_len = state->flow.key_len;
782 dpif_linux_flow_get_stats(&state->flow, &state->stats);
783 *stats = &state->stats;
789 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
791 struct dpif_linux_flow_state *state = state_;
792 int error = nl_dump_done(&state->dump);
793 ofpbuf_delete(state->buf);
799 dpif_linux_execute__(int dp_ifindex,
800 const struct nlattr *key, size_t key_len,
801 const struct nlattr *actions, size_t actions_len,
802 const struct ofpbuf *packet)
804 struct ovs_header *execute;
808 buf = ofpbuf_new(128 + actions_len + packet->size);
810 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
811 OVS_PACKET_CMD_EXECUTE, 1);
813 execute = ofpbuf_put_uninit(buf, sizeof *execute);
814 execute->dp_ifindex = dp_ifindex;
816 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, packet->data, packet->size);
817 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len);
818 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len);
820 error = nl_sock_transact(genl_sock, buf, NULL);
826 dpif_linux_execute(struct dpif *dpif_,
827 const struct nlattr *key, size_t key_len,
828 const struct nlattr *actions, size_t actions_len,
829 const struct ofpbuf *packet)
831 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
833 return dpif_linux_execute__(dpif->dp_ifindex, key, key_len,
834 actions, actions_len, packet);
838 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
840 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
841 *listen_mask = dpif->listen_mask;
846 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
848 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
852 if (listen_mask == dpif->listen_mask) {
854 } else if (!listen_mask) {
855 nl_sock_destroy(dpif->mc_sock);
856 dpif->mc_sock = NULL;
857 dpif->listen_mask = 0;
859 } else if (!dpif->mc_sock) {
860 error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
866 /* Unsubscribe from old groups. */
867 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
868 if (dpif->listen_mask & (1u << i)) {
869 nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
873 /* Update listen_mask. */
874 dpif->listen_mask = listen_mask;
876 /* Subscribe to new groups. */
878 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
879 if (dpif->listen_mask & (1u << i)) {
882 retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
892 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
893 uint32_t *probability)
895 struct dpif_linux_dp dp;
899 error = dpif_linux_dp_get(dpif_, &dp, &buf);
901 *probability = dp.sampling ? *dp.sampling : 0;
908 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
910 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
911 struct dpif_linux_dp dp;
913 dpif_linux_dp_init(&dp);
914 dp.cmd = OVS_DP_CMD_SET;
915 dp.dp_ifindex = dpif->dp_ifindex;
916 dp.sampling = &probability;
917 return dpif_linux_dp_transact(&dp, NULL, NULL);
921 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
922 uint32_t queue_id, uint32_t *priority)
924 if (queue_id < 0xf000) {
925 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
933 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
936 static const struct nl_policy ovs_packet_policy[] = {
937 /* Always present. */
938 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
939 .min_len = ETH_HEADER_LEN },
940 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
942 /* OVS_PACKET_CMD_ACTION only. */
943 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
945 /* OVS_PACKET_CMD_SAMPLE only. */
946 [OVS_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
947 [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
950 struct ovs_header *ovs_header;
951 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
952 struct nlmsghdr *nlmsg;
953 struct genlmsghdr *genl;
957 ofpbuf_use_const(&b, buf->data, buf->size);
959 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
960 genl = ofpbuf_try_pull(&b, sizeof *genl);
961 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
962 if (!nlmsg || !genl || !ovs_header
963 || nlmsg->nlmsg_type != ovs_packet_family
964 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
965 ARRAY_SIZE(ovs_packet_policy))) {
969 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
970 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
971 : genl->cmd == OVS_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
977 memset(upcall, 0, sizeof *upcall);
979 upcall->packet = buf;
980 upcall->packet->data = (void *) nl_attr_get(a[OVS_PACKET_ATTR_PACKET]);
981 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
982 upcall->key = (void *) nl_attr_get(a[OVS_PACKET_ATTR_KEY]);
983 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
984 upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
985 ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
987 upcall->sample_pool = (a[OVS_PACKET_ATTR_SAMPLE_POOL]
988 ? nl_attr_get_u32(a[OVS_PACKET_ATTR_SAMPLE_POOL])
990 if (a[OVS_PACKET_ATTR_ACTIONS]) {
991 upcall->actions = (void *) nl_attr_get(a[OVS_PACKET_ATTR_ACTIONS]);
992 upcall->actions_len = nl_attr_get_size(a[OVS_PACKET_ATTR_ACTIONS]);
995 *dp_ifindex = ovs_header->dp_ifindex;
1001 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
1003 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1008 if (!dpif->mc_sock) {
1012 for (i = 0; i < 50; i++) {
1015 error = nl_sock_recv(dpif->mc_sock, &buf, false);
1020 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1022 && dp_ifindex == dpif->dp_ifindex
1023 && dpif->listen_mask & (1u << upcall->type)) {
1037 dpif_linux_recv_wait(struct dpif *dpif_)
1039 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1040 if (dpif->mc_sock) {
1041 nl_sock_wait(dpif->mc_sock, POLLIN);
1046 dpif_linux_recv_purge(struct dpif *dpif_)
1048 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1050 if (dpif->mc_sock) {
1051 nl_sock_drain(dpif->mc_sock);
1055 const struct dpif_class dpif_linux_class = {
1057 dpif_linux_enumerate,
1063 dpif_linux_get_stats,
1064 dpif_linux_get_drop_frags,
1065 dpif_linux_set_drop_frags,
1066 dpif_linux_port_add,
1067 dpif_linux_port_del,
1068 dpif_linux_port_query_by_number,
1069 dpif_linux_port_query_by_name,
1070 dpif_linux_get_max_ports,
1071 dpif_linux_port_dump_start,
1072 dpif_linux_port_dump_next,
1073 dpif_linux_port_dump_done,
1074 dpif_linux_port_poll,
1075 dpif_linux_port_poll_wait,
1076 dpif_linux_flow_get,
1077 dpif_linux_flow_put,
1078 dpif_linux_flow_del,
1079 dpif_linux_flow_flush,
1080 dpif_linux_flow_dump_start,
1081 dpif_linux_flow_dump_next,
1082 dpif_linux_flow_dump_done,
1084 dpif_linux_recv_get_mask,
1085 dpif_linux_recv_set_mask,
1086 dpif_linux_get_sflow_probability,
1087 dpif_linux_set_sflow_probability,
1088 dpif_linux_queue_to_priority,
1090 dpif_linux_recv_wait,
1091 dpif_linux_recv_purge,
1095 dpif_linux_init(void)
1097 static int error = -1;
1100 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1101 &ovs_datapath_family);
1103 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1104 "The Open vSwitch kernel module is probably not loaded.",
1105 OVS_DATAPATH_FAMILY);
1108 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1111 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1114 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1115 &ovs_packet_family);
1118 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1126 dpif_linux_is_internal_device(const char *name)
1128 struct dpif_linux_vport reply;
1132 error = dpif_linux_vport_get(name, &reply, &buf);
1135 } else if (error != ENODEV && error != ENOENT) {
1136 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1137 name, strerror(error));
1140 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1144 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1145 const void *data, size_t size)
1147 struct ofpbuf actions, key, packet;
1148 struct odputil_keybuf keybuf;
1152 ofpbuf_use_const(&packet, data, size);
1153 flow_extract(&packet, htonll(0), 0, &flow);
1155 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1156 odp_flow_key_from_flow(&key, &flow);
1158 ofpbuf_use_stack(&actions, &action, sizeof action);
1159 nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
1161 return dpif_linux_execute__(dp_ifindex, key.data, key.size,
1162 actions.data, actions.size, &packet);
1166 dpif_linux_port_changed(const struct rtnetlink_link_change *change,
1169 struct dpif_linux *dpif = dpif_;
1172 if (change->master_ifindex == dpif->dp_ifindex
1173 && (change->nlmsg_type == RTM_NEWLINK
1174 || change->nlmsg_type == RTM_DELLINK))
1176 /* Our datapath changed, either adding a new port or deleting an
1178 sset_add(&dpif->changed_ports, change->ifname);
1181 dpif->change_error = true;
1185 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1186 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1187 * positive errno value.
1189 * 'vport' will contain pointers into 'buf', so the caller should not free
1190 * 'buf' while 'vport' is still in use. */
1192 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1193 const struct ofpbuf *buf)
1195 static const struct nl_policy ovs_vport_policy[] = {
1196 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1197 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1198 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1199 [OVS_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1200 .min_len = sizeof(struct rtnl_link_stats64),
1201 .max_len = sizeof(struct rtnl_link_stats64),
1203 [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1204 .min_len = ETH_ADDR_LEN,
1205 .max_len = ETH_ADDR_LEN,
1207 [OVS_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
1208 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1209 [OVS_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1210 [OVS_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
1213 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1214 struct ovs_header *ovs_header;
1215 struct nlmsghdr *nlmsg;
1216 struct genlmsghdr *genl;
1219 dpif_linux_vport_init(vport);
1221 ofpbuf_use_const(&b, buf->data, buf->size);
1222 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1223 genl = ofpbuf_try_pull(&b, sizeof *genl);
1224 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1225 if (!nlmsg || !genl || !ovs_header
1226 || nlmsg->nlmsg_type != ovs_vport_family
1227 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1228 ARRAY_SIZE(ovs_vport_policy))) {
1232 vport->cmd = genl->cmd;
1233 vport->dp_ifindex = ovs_header->dp_ifindex;
1234 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1235 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1236 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1237 if (a[OVS_VPORT_ATTR_STATS]) {
1238 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1240 if (a[OVS_VPORT_ATTR_ADDRESS]) {
1241 vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
1243 if (a[OVS_VPORT_ATTR_MTU]) {
1244 vport->mtu = nl_attr_get_u32(a[OVS_VPORT_ATTR_MTU]);
1246 vport->mtu = INT_MAX;
1248 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1249 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1250 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1252 if (a[OVS_VPORT_ATTR_IFINDEX]) {
1253 vport->ifindex = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFINDEX]);
1255 if (a[OVS_VPORT_ATTR_IFLINK]) {
1256 vport->iflink = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFLINK]);
1261 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1262 * followed by Netlink attributes corresponding to 'vport'. */
1264 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1267 struct ovs_header *ovs_header;
1269 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1272 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1273 ovs_header->dp_ifindex = vport->dp_ifindex;
1275 if (vport->port_no != UINT32_MAX) {
1276 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1279 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1280 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1284 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1288 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1289 vport->stats, sizeof *vport->stats);
1292 if (vport->address) {
1293 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS,
1294 vport->address, ETH_ADDR_LEN);
1297 if (vport->mtu && vport->mtu != INT_MAX) {
1298 nl_msg_put_u32(buf, OVS_VPORT_ATTR_MTU, vport->mtu);
1301 if (vport->options) {
1302 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1303 vport->options, vport->options_len);
1306 if (vport->ifindex) {
1307 nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFINDEX, vport->ifindex);
1310 if (vport->iflink) {
1311 nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFLINK, vport->iflink);
1315 /* Clears 'vport' to "empty" values. */
1317 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1319 memset(vport, 0, sizeof *vport);
1320 vport->port_no = UINT32_MAX;
1323 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1324 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1325 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1326 * result of the command is expected to be an ovs_vport also, which is decoded
1327 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1328 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1330 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1331 struct dpif_linux_vport *reply,
1332 struct ofpbuf **bufp)
1334 struct ofpbuf *request_buf;
1337 assert((reply != NULL) == (bufp != NULL));
1339 error = dpif_linux_init();
1343 dpif_linux_vport_init(reply);
1348 request_buf = ofpbuf_new(1024);
1349 dpif_linux_vport_to_ofpbuf(request, request_buf);
1350 error = nl_sock_transact(genl_sock, request_buf, bufp);
1351 ofpbuf_delete(request_buf);
1355 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1358 dpif_linux_vport_init(reply);
1359 ofpbuf_delete(*bufp);
1366 /* Obtains information about the kernel vport named 'name' and stores it into
1367 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1368 * longer needed ('reply' will contain pointers into '*bufp'). */
1370 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1371 struct ofpbuf **bufp)
1373 struct dpif_linux_vport request;
1375 dpif_linux_vport_init(&request);
1376 request.cmd = OVS_VPORT_CMD_GET;
1377 request.name = name;
1379 return dpif_linux_vport_transact(&request, reply, bufp);
1382 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1383 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1384 * positive errno value.
1386 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1387 * while 'dp' is still in use. */
1389 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1391 static const struct nl_policy ovs_datapath_policy[] = {
1392 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1393 [OVS_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1394 .min_len = sizeof(struct ovs_dp_stats),
1395 .max_len = sizeof(struct ovs_dp_stats),
1397 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1398 [OVS_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1399 [OVS_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
1402 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1403 struct ovs_header *ovs_header;
1404 struct nlmsghdr *nlmsg;
1405 struct genlmsghdr *genl;
1408 dpif_linux_dp_init(dp);
1410 ofpbuf_use_const(&b, buf->data, buf->size);
1411 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1412 genl = ofpbuf_try_pull(&b, sizeof *genl);
1413 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1414 if (!nlmsg || !genl || !ovs_header
1415 || nlmsg->nlmsg_type != ovs_datapath_family
1416 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1417 ARRAY_SIZE(ovs_datapath_policy))) {
1421 dp->cmd = genl->cmd;
1422 dp->dp_ifindex = ovs_header->dp_ifindex;
1423 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1424 if (a[OVS_DP_ATTR_STATS]) {
1425 /* Can't use structure assignment because Netlink doesn't ensure
1426 * sufficient alignment for 64-bit members. */
1427 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1430 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1431 dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1433 if (a[OVS_DP_ATTR_SAMPLING]) {
1434 dp->sampling = nl_attr_get(a[OVS_DP_ATTR_SAMPLING]);
1437 if (a[OVS_DP_ATTR_MCGROUPS]) {
1438 static const struct nl_policy ovs_mcgroup_policy[] = {
1439 [OVS_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
1440 [OVS_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
1441 [OVS_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
1444 struct nlattr *mcgroups[ARRAY_SIZE(ovs_mcgroup_policy)];
1446 if (!nl_parse_nested(a[OVS_DP_ATTR_MCGROUPS], ovs_mcgroup_policy,
1447 mcgroups, ARRAY_SIZE(ovs_mcgroup_policy))) {
1451 if (mcgroups[OVS_PACKET_CMD_MISS]) {
1452 dp->mcgroups[DPIF_UC_MISS]
1453 = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_MISS]);
1455 if (mcgroups[OVS_PACKET_CMD_ACTION]) {
1456 dp->mcgroups[DPIF_UC_ACTION]
1457 = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_ACTION]);
1459 if (mcgroups[OVS_PACKET_CMD_SAMPLE]) {
1460 dp->mcgroups[DPIF_UC_SAMPLE]
1461 = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_SAMPLE]);
1468 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1470 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1472 struct ovs_header *ovs_header;
1474 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1475 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1477 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1478 ovs_header->dp_ifindex = dp->dp_ifindex;
1481 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1484 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1486 if (dp->ipv4_frags) {
1487 nl_msg_put_u32(buf, OVS_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1491 nl_msg_put_u32(buf, OVS_DP_ATTR_SAMPLING, *dp->sampling);
1495 /* Clears 'dp' to "empty" values. */
1497 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1499 memset(dp, 0, sizeof *dp);
1503 dpif_linux_dp_dump_start(struct nl_dump *dump)
1505 struct dpif_linux_dp request;
1508 dpif_linux_dp_init(&request);
1509 request.cmd = OVS_DP_CMD_GET;
1511 buf = ofpbuf_new(1024);
1512 dpif_linux_dp_to_ofpbuf(&request, buf);
1513 nl_dump_start(dump, genl_sock, buf);
1517 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1518 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1519 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1520 * result of the command is expected to be of the same form, which is decoded
1521 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1522 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1524 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1525 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1527 struct ofpbuf *request_buf;
1530 assert((reply != NULL) == (bufp != NULL));
1532 request_buf = ofpbuf_new(1024);
1533 dpif_linux_dp_to_ofpbuf(request, request_buf);
1534 error = nl_sock_transact(genl_sock, request_buf, bufp);
1535 ofpbuf_delete(request_buf);
1539 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1542 dpif_linux_dp_init(reply);
1543 ofpbuf_delete(*bufp);
1550 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1551 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1552 * will contain pointers into '*bufp'). */
1554 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1555 struct ofpbuf **bufp)
1557 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1558 struct dpif_linux_dp request;
1560 dpif_linux_dp_init(&request);
1561 request.cmd = OVS_DP_CMD_GET;
1562 request.dp_ifindex = dpif->dp_ifindex;
1564 return dpif_linux_dp_transact(&request, reply, bufp);
1567 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1568 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1569 * positive errno value.
1571 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1572 * while 'flow' is still in use. */
1574 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1575 const struct ofpbuf *buf)
1577 static const struct nl_policy ovs_flow_policy[] = {
1578 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1579 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1580 [OVS_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1581 .min_len = sizeof(struct ovs_flow_stats),
1582 .max_len = sizeof(struct ovs_flow_stats),
1584 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1585 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1586 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1589 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1590 struct ovs_header *ovs_header;
1591 struct nlmsghdr *nlmsg;
1592 struct genlmsghdr *genl;
1595 dpif_linux_flow_init(flow);
1597 ofpbuf_use_const(&b, buf->data, buf->size);
1598 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1599 genl = ofpbuf_try_pull(&b, sizeof *genl);
1600 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1601 if (!nlmsg || !genl || !ovs_header
1602 || nlmsg->nlmsg_type != ovs_flow_family
1603 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1604 ARRAY_SIZE(ovs_flow_policy))) {
1608 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1609 flow->dp_ifindex = ovs_header->dp_ifindex;
1610 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1611 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1612 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1613 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1614 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1616 if (a[OVS_FLOW_ATTR_STATS]) {
1617 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1619 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1620 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1622 if (a[OVS_FLOW_ATTR_USED]) {
1623 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1628 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1629 * followed by Netlink attributes corresponding to 'flow'. */
1631 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1634 struct ovs_header *ovs_header;
1636 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1637 NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1640 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1641 ovs_header->dp_ifindex = flow->dp_ifindex;
1643 if (flow->key_len) {
1644 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1647 if (flow->actions || flow->actions_len) {
1648 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1649 flow->actions, flow->actions_len);
1652 /* We never need to send these to the kernel. */
1653 assert(!flow->stats);
1654 assert(!flow->tcp_flags);
1655 assert(!flow->used);
1658 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1662 /* Clears 'flow' to "empty" values. */
1664 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1666 memset(flow, 0, sizeof *flow);
1669 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1670 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1671 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1672 * result of the command is expected to be a flow also, which is decoded and
1673 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1674 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1676 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1677 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1679 struct ofpbuf *request_buf;
1682 assert((reply != NULL) == (bufp != NULL));
1684 request_buf = ofpbuf_new(1024);
1685 dpif_linux_flow_to_ofpbuf(request, request_buf);
1686 error = nl_sock_transact(genl_sock, request_buf, bufp);
1687 ofpbuf_delete(request_buf);
1691 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1694 dpif_linux_flow_init(reply);
1695 ofpbuf_delete(*bufp);
1703 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1704 struct dpif_flow_stats *stats)
1707 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1708 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1710 stats->n_packets = 0;
1713 stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1714 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;