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"
38 #include "netdev-vport.h"
39 #include "netlink-socket.h"
43 #include "openvswitch/tunnel.h"
45 #include "poll-loop.h"
46 #include "rtnetlink.h"
47 #include "rtnetlink-link.h"
50 #include "unaligned.h"
54 VLOG_DEFINE_THIS_MODULE(dpif_linux);
56 enum { LRU_MAX_PORTS = 1024 };
57 enum { LRU_MASK = LRU_MAX_PORTS - 1};
58 BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
60 struct dpif_linux_dp {
61 /* Generic Netlink header. */
64 /* struct odp_header. */
68 const char *name; /* ODP_DP_ATTR_NAME. */
69 struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
70 enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
71 const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
72 uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
75 static void dpif_linux_dp_init(struct dpif_linux_dp *);
76 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
77 const struct ofpbuf *);
78 static void dpif_linux_dp_dump_start(struct nl_dump *);
79 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
80 struct dpif_linux_dp *reply,
81 struct ofpbuf **bufp);
82 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
83 struct ofpbuf **bufp);
85 struct dpif_linux_flow {
86 /* Generic Netlink header. */
89 /* struct odp_header. */
90 unsigned int nlmsg_flags;
95 * The 'stats' and 'used' members point to 64-bit data that might only be
96 * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
97 * access their values.
99 * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in
100 * the Netlink version of the command, even if actions_len is zero. */
101 const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
103 const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */
105 const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */
106 const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */
107 const uint64_t *used; /* ODP_FLOW_ATTR_USED. */
108 bool clear; /* ODP_FLOW_ATTR_CLEAR. */
111 static void dpif_linux_flow_init(struct dpif_linux_flow *);
112 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
113 const struct ofpbuf *);
114 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
116 static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
117 struct dpif_linux_flow *reply,
118 struct ofpbuf **bufp);
119 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
120 struct dpif_flow_stats *);
122 /* Datapath interface for the openvswitch Linux kernel module. */
127 /* Multicast group messages. */
128 struct nl_sock *mc_sock;
129 uint32_t mcgroups[DPIF_N_UC_TYPES];
130 unsigned int listen_mask;
132 /* Change notification. */
133 struct sset changed_ports; /* Ports that have changed. */
134 struct rtnetlink_notifier port_notifier;
137 /* Queue of unused ports. */
138 unsigned long *lru_bitmap;
139 uint16_t lru_ports[LRU_MAX_PORTS];
144 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
146 /* Generic Netlink family numbers for ODP. */
147 static int odp_datapath_family;
148 static int odp_vport_family;
149 static int odp_flow_family;
150 static int odp_packet_family;
152 /* Generic Netlink socket. */
153 static struct nl_sock *genl_sock;
155 static int dpif_linux_init(void);
156 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
157 static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
160 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
162 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
163 const struct ofpbuf *);
165 static struct dpif_linux *
166 dpif_linux_cast(const struct dpif *dpif)
168 dpif_assert_class(dpif, &dpif_linux_class);
169 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
173 dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
175 if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
176 bitmap_set1(dp->lru_bitmap, port);
177 dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
182 dpif_linux_pop_port(struct dpif_linux *dp)
186 if (dp->lru_head == dp->lru_tail) {
190 port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
191 bitmap_set0(dp->lru_bitmap, port);
196 dpif_linux_enumerate(struct sset *all_dps)
202 error = dpif_linux_init();
207 dpif_linux_dp_dump_start(&dump);
208 while (nl_dump_next(&dump, &msg)) {
209 struct dpif_linux_dp dp;
211 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
212 sset_add(all_dps, dp.name);
215 return nl_dump_done(&dump);
219 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
220 bool create, struct dpif **dpifp)
222 struct dpif_linux_dp dp_request, dp;
226 error = dpif_linux_init();
231 /* Create or look up datapath. */
232 dpif_linux_dp_init(&dp_request);
233 dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
234 dp_request.name = name;
235 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
239 error = open_dpif(&dp, dpifp);
246 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
248 struct dpif_linux *dpif;
252 dpif = xmalloc(sizeof *dpif);
253 error = rtnetlink_link_notifier_register(&dpif->port_notifier,
254 dpif_linux_port_changed, dpif);
259 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
260 dp->dp_ifindex, dp->dp_ifindex);
262 dpif->mc_sock = NULL;
263 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
264 dpif->mcgroups[i] = dp->mcgroups[i];
266 dpif->listen_mask = 0;
267 dpif->dp_ifindex = dp->dp_ifindex;
268 sset_init(&dpif->changed_ports);
269 dpif->change_error = false;
270 *dpifp = &dpif->dpif;
272 dpif->lru_head = dpif->lru_tail = 0;
273 dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
274 bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL);
275 for (i = 1; i < LRU_MAX_PORTS; i++) {
276 dpif_linux_push_port(dpif, i);
286 dpif_linux_close(struct dpif *dpif_)
288 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
289 rtnetlink_link_notifier_unregister(&dpif->port_notifier);
290 sset_destroy(&dpif->changed_ports);
291 free(dpif->lru_bitmap);
296 dpif_linux_destroy(struct dpif *dpif_)
298 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
299 struct dpif_linux_dp dp;
301 dpif_linux_dp_init(&dp);
302 dp.cmd = ODP_DP_CMD_DEL;
303 dp.dp_ifindex = dpif->dp_ifindex;
304 return dpif_linux_dp_transact(&dp, NULL, NULL);
308 dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
310 struct dpif_linux_dp dp;
314 error = dpif_linux_dp_get(dpif_, &dp, &buf);
323 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
325 struct dpif_linux_dp dp;
329 error = dpif_linux_dp_get(dpif_, &dp, &buf);
331 *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
338 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
340 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
341 struct dpif_linux_dp dp;
343 dpif_linux_dp_init(&dp);
344 dp.cmd = ODP_DP_CMD_SET;
345 dp.dp_ifindex = dpif->dp_ifindex;
346 dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
347 return dpif_linux_dp_transact(&dp, NULL, NULL);
351 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
354 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
355 const char *name = netdev_get_name(netdev);
356 const char *type = netdev_get_type(netdev);
357 struct dpif_linux_vport request, reply;
358 const struct ofpbuf *options;
362 dpif_linux_vport_init(&request);
363 request.cmd = ODP_VPORT_CMD_NEW;
364 request.dp_ifindex = dpif->dp_ifindex;
365 request.type = netdev_vport_get_vport_type(netdev);
366 if (request.type == ODP_VPORT_TYPE_UNSPEC) {
367 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
368 "unsupported type `%s'",
369 dpif_name(dpif_), name, type);
374 options = netdev_vport_get_options(netdev);
375 if (options && options->size) {
376 request.options = options->data;
377 request.options_len = options->size;
380 /* Loop until we find a port that isn't used. */
382 request.port_no = dpif_linux_pop_port(dpif);
383 error = dpif_linux_vport_transact(&request, &reply, &buf);
386 *port_nop = reply.port_no;
389 } while (request.port_no != UINT32_MAX
390 && (error == EBUSY || error == EFBIG));
396 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
398 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
399 struct dpif_linux_vport vport;
402 dpif_linux_vport_init(&vport);
403 vport.cmd = ODP_VPORT_CMD_DEL;
404 vport.dp_ifindex = dpif->dp_ifindex;
405 vport.port_no = port_no;
406 error = dpif_linux_vport_transact(&vport, NULL, NULL);
409 dpif_linux_push_port(dpif, port_no);
415 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
416 const char *port_name, struct dpif_port *dpif_port)
418 struct dpif_linux_vport request;
419 struct dpif_linux_vport reply;
423 dpif_linux_vport_init(&request);
424 request.cmd = ODP_VPORT_CMD_GET;
425 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
426 request.port_no = port_no;
427 request.name = port_name;
429 error = dpif_linux_vport_transact(&request, &reply, &buf);
431 dpif_port->name = xstrdup(reply.name);
432 dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
433 dpif_port->port_no = reply.port_no;
440 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
441 struct dpif_port *dpif_port)
443 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
447 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
448 struct dpif_port *dpif_port)
450 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
454 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
456 /* If the datapath increases its range of supported ports, then it should
457 * start reporting that. */
462 dpif_linux_flow_flush(struct dpif *dpif_)
464 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
465 struct dpif_linux_flow flow;
467 dpif_linux_flow_init(&flow);
468 flow.cmd = ODP_FLOW_CMD_DEL;
469 flow.dp_ifindex = dpif->dp_ifindex;
470 return dpif_linux_flow_transact(&flow, NULL, NULL);
473 struct dpif_linux_port_state {
478 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
480 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
481 struct dpif_linux_port_state *state;
482 struct dpif_linux_vport request;
485 *statep = state = xmalloc(sizeof *state);
487 dpif_linux_vport_init(&request);
488 request.cmd = ODP_DP_CMD_GET;
489 request.dp_ifindex = dpif->dp_ifindex;
491 buf = ofpbuf_new(1024);
492 dpif_linux_vport_to_ofpbuf(&request, buf);
493 nl_dump_start(&state->dump, genl_sock, buf);
500 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
501 struct dpif_port *dpif_port)
503 struct dpif_linux_port_state *state = state_;
504 struct dpif_linux_vport vport;
508 if (!nl_dump_next(&state->dump, &buf)) {
512 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
517 dpif_port->name = (char *) vport.name;
518 dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
519 dpif_port->port_no = vport.port_no;
524 dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
526 struct dpif_linux_port_state *state = state_;
527 int error = nl_dump_done(&state->dump);
533 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
535 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
537 if (dpif->change_error) {
538 dpif->change_error = false;
539 sset_clear(&dpif->changed_ports);
541 } else if (!sset_is_empty(&dpif->changed_ports)) {
542 *devnamep = sset_pop(&dpif->changed_ports);
550 dpif_linux_port_poll_wait(const struct dpif *dpif_)
552 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
553 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
554 poll_immediate_wake();
556 rtnetlink_link_notifier_wait();
561 dpif_linux_flow_get__(const struct dpif *dpif_,
562 const struct nlattr *key, size_t key_len,
563 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
565 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
566 struct dpif_linux_flow request;
568 dpif_linux_flow_init(&request);
569 request.cmd = ODP_FLOW_CMD_GET;
570 request.dp_ifindex = dpif->dp_ifindex;
572 request.key_len = key_len;
573 return dpif_linux_flow_transact(&request, reply, bufp);
577 dpif_linux_flow_get(const struct dpif *dpif_,
578 const struct nlattr *key, size_t key_len,
579 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
581 struct dpif_linux_flow reply;
585 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
588 dpif_linux_flow_get_stats(&reply, stats);
591 buf->data = (void *) reply.actions;
592 buf->size = reply.actions_len;
602 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
603 const struct nlattr *key, size_t key_len,
604 const struct nlattr *actions, size_t actions_len,
605 struct dpif_flow_stats *stats)
607 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
608 struct dpif_linux_flow request, reply;
609 struct nlattr dummy_action;
613 dpif_linux_flow_init(&request);
614 request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
615 request.dp_ifindex = dpif->dp_ifindex;
617 request.key_len = key_len;
618 /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
619 request.actions = actions ? actions : &dummy_action;
620 request.actions_len = actions_len;
621 if (flags & DPIF_FP_ZERO_STATS) {
622 request.clear = true;
624 request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
625 error = dpif_linux_flow_transact(&request,
626 stats ? &reply : NULL,
627 stats ? &buf : NULL);
628 if (!error && stats) {
629 dpif_linux_flow_get_stats(&reply, stats);
636 dpif_linux_flow_del(struct dpif *dpif_,
637 const struct nlattr *key, size_t key_len,
638 struct dpif_flow_stats *stats)
640 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
641 struct dpif_linux_flow request, reply;
645 dpif_linux_flow_init(&request);
646 request.cmd = ODP_FLOW_CMD_DEL;
647 request.dp_ifindex = dpif->dp_ifindex;
649 request.key_len = key_len;
650 error = dpif_linux_flow_transact(&request,
651 stats ? &reply : NULL,
652 stats ? &buf : NULL);
653 if (!error && stats) {
654 dpif_linux_flow_get_stats(&reply, stats);
660 struct dpif_linux_flow_state {
662 struct dpif_linux_flow flow;
663 struct dpif_flow_stats stats;
668 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
670 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
671 struct dpif_linux_flow_state *state;
672 struct dpif_linux_flow request;
675 *statep = state = xmalloc(sizeof *state);
677 dpif_linux_flow_init(&request);
678 request.cmd = ODP_DP_CMD_GET;
679 request.dp_ifindex = dpif->dp_ifindex;
681 buf = ofpbuf_new(1024);
682 dpif_linux_flow_to_ofpbuf(&request, buf);
683 nl_dump_start(&state->dump, genl_sock, buf);
692 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
693 const struct nlattr **key, size_t *key_len,
694 const struct nlattr **actions, size_t *actions_len,
695 const struct dpif_flow_stats **stats)
697 struct dpif_linux_flow_state *state = state_;
702 ofpbuf_delete(state->buf);
705 if (!nl_dump_next(&state->dump, &buf)) {
709 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
714 if (actions && !state->flow.actions) {
715 error = dpif_linux_flow_get__(dpif_, state->flow.key,
717 &state->flow, &state->buf);
718 if (error == ENOENT) {
719 VLOG_DBG("dumped flow disappeared on get");
721 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
727 *actions = state->flow.actions;
728 *actions_len = state->flow.actions_len;
731 *key = state->flow.key;
732 *key_len = state->flow.key_len;
735 dpif_linux_flow_get_stats(&state->flow, &state->stats);
736 *stats = &state->stats;
742 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
744 struct dpif_linux_flow_state *state = state_;
745 int error = nl_dump_done(&state->dump);
746 ofpbuf_delete(state->buf);
752 dpif_linux_execute(struct dpif *dpif_,
753 const struct nlattr *actions, size_t actions_len,
754 const struct ofpbuf *packet)
756 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
757 struct odp_header *execute;
761 buf = ofpbuf_new(128 + actions_len + packet->size);
763 nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
764 ODP_PACKET_CMD_EXECUTE, 1);
766 execute = ofpbuf_put_uninit(buf, sizeof *execute);
767 execute->dp_ifindex = dpif->dp_ifindex;
769 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
770 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
772 error = nl_sock_transact(genl_sock, buf, NULL);
778 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
780 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
781 *listen_mask = dpif->listen_mask;
786 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
788 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
792 if (listen_mask == dpif->listen_mask) {
794 } else if (!listen_mask) {
795 nl_sock_destroy(dpif->mc_sock);
796 dpif->mc_sock = NULL;
797 dpif->listen_mask = 0;
799 } else if (!dpif->mc_sock) {
800 error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
806 /* Unsubscribe from old groups. */
807 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
808 if (dpif->listen_mask & (1u << i)) {
809 nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
813 /* Update listen_mask. */
814 dpif->listen_mask = listen_mask;
816 /* Subscribe to new groups. */
818 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
819 if (dpif->listen_mask & (1u << i)) {
822 retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
832 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
833 uint32_t *probability)
835 struct dpif_linux_dp dp;
839 error = dpif_linux_dp_get(dpif_, &dp, &buf);
841 *probability = dp.sampling ? *dp.sampling : 0;
848 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
850 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
851 struct dpif_linux_dp dp;
853 dpif_linux_dp_init(&dp);
854 dp.cmd = ODP_DP_CMD_SET;
855 dp.dp_ifindex = dpif->dp_ifindex;
856 dp.sampling = &probability;
857 return dpif_linux_dp_transact(&dp, NULL, NULL);
861 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
862 uint32_t queue_id, uint32_t *priority)
864 if (queue_id < 0xf000) {
865 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
873 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
876 static const struct nl_policy odp_packet_policy[] = {
877 /* Always present. */
878 [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
879 .min_len = ETH_HEADER_LEN },
880 [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
882 /* ODP_PACKET_CMD_ACTION only. */
883 [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
885 /* ODP_PACKET_CMD_SAMPLE only. */
886 [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
887 [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
890 struct odp_header *odp_header;
891 struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
892 struct nlmsghdr *nlmsg;
893 struct genlmsghdr *genl;
897 ofpbuf_use_const(&b, buf->data, buf->size);
899 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
900 genl = ofpbuf_try_pull(&b, sizeof *genl);
901 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
902 if (!nlmsg || !genl || !odp_header
903 || nlmsg->nlmsg_type != odp_packet_family
904 || !nl_policy_parse(&b, 0, odp_packet_policy, a,
905 ARRAY_SIZE(odp_packet_policy))) {
909 type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
910 : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
911 : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
917 memset(upcall, 0, sizeof *upcall);
919 upcall->packet = buf;
920 upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
921 upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
922 upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
923 upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
924 upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
925 ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
927 upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
928 ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
930 if (a[ODP_PACKET_ATTR_ACTIONS]) {
931 upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
932 upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
935 *dp_ifindex = odp_header->dp_ifindex;
941 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
943 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
948 if (!dpif->mc_sock) {
952 for (i = 0; i < 50; i++) {
955 error = nl_sock_recv(dpif->mc_sock, &buf, false);
960 error = parse_odp_packet(buf, upcall, &dp_ifindex);
962 && dp_ifindex == dpif->dp_ifindex
963 && dpif->listen_mask & (1u << upcall->type)) {
977 dpif_linux_recv_wait(struct dpif *dpif_)
979 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
981 nl_sock_wait(dpif->mc_sock, POLLIN);
986 dpif_linux_recv_purge(struct dpif *dpif_)
988 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
991 nl_sock_drain(dpif->mc_sock);
995 const struct dpif_class dpif_linux_class = {
999 dpif_linux_enumerate,
1003 dpif_linux_get_stats,
1004 dpif_linux_get_drop_frags,
1005 dpif_linux_set_drop_frags,
1006 dpif_linux_port_add,
1007 dpif_linux_port_del,
1008 dpif_linux_port_query_by_number,
1009 dpif_linux_port_query_by_name,
1010 dpif_linux_get_max_ports,
1011 dpif_linux_port_dump_start,
1012 dpif_linux_port_dump_next,
1013 dpif_linux_port_dump_done,
1014 dpif_linux_port_poll,
1015 dpif_linux_port_poll_wait,
1016 dpif_linux_flow_get,
1017 dpif_linux_flow_put,
1018 dpif_linux_flow_del,
1019 dpif_linux_flow_flush,
1020 dpif_linux_flow_dump_start,
1021 dpif_linux_flow_dump_next,
1022 dpif_linux_flow_dump_done,
1024 dpif_linux_recv_get_mask,
1025 dpif_linux_recv_set_mask,
1026 dpif_linux_get_sflow_probability,
1027 dpif_linux_set_sflow_probability,
1028 dpif_linux_queue_to_priority,
1030 dpif_linux_recv_wait,
1031 dpif_linux_recv_purge,
1035 dpif_linux_init(void)
1037 static int error = -1;
1040 error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
1041 &odp_datapath_family);
1043 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1044 "The Open vSwitch kernel module is probably not loaded.",
1045 ODP_DATAPATH_FAMILY);
1048 error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
1051 error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
1054 error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
1055 &odp_packet_family);
1058 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1066 dpif_linux_is_internal_device(const char *name)
1068 struct dpif_linux_vport reply;
1072 error = dpif_linux_vport_get(name, &reply, &buf);
1075 } else if (error != ENODEV) {
1076 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1077 name, strerror(error));
1080 return reply.type == ODP_VPORT_TYPE_INTERNAL;
1084 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1085 const void *data, size_t size)
1087 struct odp_header *execute;
1092 buf = ofpbuf_new(128 + size);
1094 nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
1095 ODP_PACKET_CMD_EXECUTE, 1);
1097 execute = ofpbuf_put_uninit(buf, sizeof *execute);
1098 execute->dp_ifindex = dp_ifindex;
1100 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size);
1102 actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS);
1103 nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no);
1104 nl_msg_end_nested(buf, actions_ofs);
1106 error = nl_sock_transact(genl_sock, buf, NULL);
1112 dpif_linux_port_changed(const struct rtnetlink_link_change *change,
1115 struct dpif_linux *dpif = dpif_;
1118 if (change->master_ifindex == dpif->dp_ifindex
1119 && (change->nlmsg_type == RTM_NEWLINK
1120 || change->nlmsg_type == RTM_DELLINK))
1122 /* Our datapath changed, either adding a new port or deleting an
1124 sset_add(&dpif->changed_ports, change->ifname);
1127 dpif->change_error = true;
1131 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1132 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1133 * positive errno value.
1135 * 'vport' will contain pointers into 'buf', so the caller should not free
1136 * 'buf' while 'vport' is still in use. */
1138 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1139 const struct ofpbuf *buf)
1141 static const struct nl_policy odp_vport_policy[] = {
1142 [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1143 [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1144 [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1145 [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1146 .min_len = sizeof(struct rtnl_link_stats64),
1147 .max_len = sizeof(struct rtnl_link_stats64),
1149 [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1150 .min_len = ETH_ADDR_LEN,
1151 .max_len = ETH_ADDR_LEN,
1153 [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
1154 [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1155 [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1156 [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
1159 struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
1160 struct odp_header *odp_header;
1161 struct nlmsghdr *nlmsg;
1162 struct genlmsghdr *genl;
1165 dpif_linux_vport_init(vport);
1167 ofpbuf_use_const(&b, buf->data, buf->size);
1168 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1169 genl = ofpbuf_try_pull(&b, sizeof *genl);
1170 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1171 if (!nlmsg || !genl || !odp_header
1172 || nlmsg->nlmsg_type != odp_vport_family
1173 || !nl_policy_parse(&b, 0, odp_vport_policy, a,
1174 ARRAY_SIZE(odp_vport_policy))) {
1178 vport->cmd = genl->cmd;
1179 vport->dp_ifindex = odp_header->dp_ifindex;
1180 vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1181 vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1182 vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
1183 if (a[ODP_VPORT_ATTR_STATS]) {
1184 vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
1186 if (a[ODP_VPORT_ATTR_ADDRESS]) {
1187 vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
1189 if (a[ODP_VPORT_ATTR_MTU]) {
1190 vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
1192 vport->mtu = INT_MAX;
1194 if (a[ODP_VPORT_ATTR_OPTIONS]) {
1195 vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
1196 vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]);
1198 if (a[ODP_VPORT_ATTR_IFINDEX]) {
1199 vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
1201 if (a[ODP_VPORT_ATTR_IFLINK]) {
1202 vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]);
1207 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1208 * followed by Netlink attributes corresponding to 'vport'. */
1210 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1213 struct odp_header *odp_header;
1215 nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1218 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1219 odp_header->dp_ifindex = vport->dp_ifindex;
1221 if (vport->port_no != UINT32_MAX) {
1222 nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1225 if (vport->type != ODP_VPORT_TYPE_UNSPEC) {
1226 nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type);
1230 nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
1234 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
1235 vport->stats, sizeof *vport->stats);
1238 if (vport->address) {
1239 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS,
1240 vport->address, ETH_ADDR_LEN);
1243 if (vport->mtu && vport->mtu != INT_MAX) {
1244 nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
1247 if (vport->options) {
1248 nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS,
1249 vport->options, vport->options_len);
1252 if (vport->ifindex) {
1253 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex);
1256 if (vport->iflink) {
1257 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
1261 /* Clears 'vport' to "empty" values. */
1263 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1265 memset(vport, 0, sizeof *vport);
1266 vport->port_no = UINT32_MAX;
1269 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1270 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1271 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1272 * result of the command is expected to be an odp_vport also, which is decoded
1273 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1274 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1276 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1277 struct dpif_linux_vport *reply,
1278 struct ofpbuf **bufp)
1280 struct ofpbuf *request_buf;
1283 assert((reply != NULL) == (bufp != NULL));
1285 request_buf = ofpbuf_new(1024);
1286 dpif_linux_vport_to_ofpbuf(request, request_buf);
1287 error = nl_sock_transact(genl_sock, request_buf, bufp);
1288 ofpbuf_delete(request_buf);
1292 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1295 dpif_linux_vport_init(reply);
1296 ofpbuf_delete(*bufp);
1303 /* Obtains information about the kernel vport named 'name' and stores it into
1304 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1305 * longer needed ('reply' will contain pointers into '*bufp'). */
1307 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1308 struct ofpbuf **bufp)
1310 struct dpif_linux_vport request;
1312 dpif_linux_vport_init(&request);
1313 request.cmd = ODP_VPORT_CMD_GET;
1314 request.name = name;
1316 return dpif_linux_vport_transact(&request, reply, bufp);
1319 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1320 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1321 * positive errno value.
1323 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1324 * while 'dp' is still in use. */
1326 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1328 static const struct nl_policy odp_datapath_policy[] = {
1329 [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1330 [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1331 .min_len = sizeof(struct odp_stats),
1332 .max_len = sizeof(struct odp_stats),
1334 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1335 [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1336 [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
1339 struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
1340 struct odp_header *odp_header;
1341 struct nlmsghdr *nlmsg;
1342 struct genlmsghdr *genl;
1345 dpif_linux_dp_init(dp);
1347 ofpbuf_use_const(&b, buf->data, buf->size);
1348 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1349 genl = ofpbuf_try_pull(&b, sizeof *genl);
1350 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1351 if (!nlmsg || !genl || !odp_header
1352 || nlmsg->nlmsg_type != odp_datapath_family
1353 || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
1354 ARRAY_SIZE(odp_datapath_policy))) {
1358 dp->cmd = genl->cmd;
1359 dp->dp_ifindex = odp_header->dp_ifindex;
1360 dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
1361 if (a[ODP_DP_ATTR_STATS]) {
1362 /* Can't use structure assignment because Netlink doesn't ensure
1363 * sufficient alignment for 64-bit members. */
1364 memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
1367 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1368 dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1370 if (a[ODP_DP_ATTR_SAMPLING]) {
1371 dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
1374 if (a[ODP_DP_ATTR_MCGROUPS]) {
1375 static const struct nl_policy odp_mcgroup_policy[] = {
1376 [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
1377 [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
1378 [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
1381 struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
1383 if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
1384 mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
1388 if (mcgroups[ODP_PACKET_CMD_MISS]) {
1389 dp->mcgroups[DPIF_UC_MISS]
1390 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
1392 if (mcgroups[ODP_PACKET_CMD_ACTION]) {
1393 dp->mcgroups[DPIF_UC_ACTION]
1394 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
1396 if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
1397 dp->mcgroups[DPIF_UC_SAMPLE]
1398 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
1405 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1407 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1409 struct odp_header *odp_header;
1411 nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
1412 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1414 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1415 odp_header->dp_ifindex = dp->dp_ifindex;
1418 nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
1421 /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
1423 if (dp->ipv4_frags) {
1424 nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1428 nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
1432 /* Clears 'dp' to "empty" values. */
1434 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1436 memset(dp, 0, sizeof *dp);
1440 dpif_linux_dp_dump_start(struct nl_dump *dump)
1442 struct dpif_linux_dp request;
1445 dpif_linux_dp_init(&request);
1446 request.cmd = ODP_DP_CMD_GET;
1448 buf = ofpbuf_new(1024);
1449 dpif_linux_dp_to_ofpbuf(&request, buf);
1450 nl_dump_start(dump, genl_sock, buf);
1454 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1455 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1456 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1457 * result of the command is expected to be of the same form, which is decoded
1458 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1459 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1461 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1462 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1464 struct ofpbuf *request_buf;
1467 assert((reply != NULL) == (bufp != NULL));
1469 request_buf = ofpbuf_new(1024);
1470 dpif_linux_dp_to_ofpbuf(request, request_buf);
1471 error = nl_sock_transact(genl_sock, request_buf, bufp);
1472 ofpbuf_delete(request_buf);
1476 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1479 dpif_linux_dp_init(reply);
1480 ofpbuf_delete(*bufp);
1487 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1488 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1489 * will contain pointers into '*bufp'). */
1491 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1492 struct ofpbuf **bufp)
1494 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1495 struct dpif_linux_dp request;
1497 dpif_linux_dp_init(&request);
1498 request.cmd = ODP_DP_CMD_GET;
1499 request.dp_ifindex = dpif->dp_ifindex;
1501 return dpif_linux_dp_transact(&request, reply, bufp);
1504 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1505 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1506 * positive errno value.
1508 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1509 * while 'flow' is still in use. */
1511 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1512 const struct ofpbuf *buf)
1514 static const struct nl_policy odp_flow_policy[] = {
1515 [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1516 [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1517 [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1518 .min_len = sizeof(struct odp_flow_stats),
1519 .max_len = sizeof(struct odp_flow_stats),
1521 [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1522 [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1523 /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
1526 struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
1527 struct odp_header *odp_header;
1528 struct nlmsghdr *nlmsg;
1529 struct genlmsghdr *genl;
1532 dpif_linux_flow_init(flow);
1534 ofpbuf_use_const(&b, buf->data, buf->size);
1535 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1536 genl = ofpbuf_try_pull(&b, sizeof *genl);
1537 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1538 if (!nlmsg || !genl || !odp_header
1539 || nlmsg->nlmsg_type != odp_flow_family
1540 || !nl_policy_parse(&b, 0, odp_flow_policy, a,
1541 ARRAY_SIZE(odp_flow_policy))) {
1545 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1546 flow->dp_ifindex = odp_header->dp_ifindex;
1547 flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
1548 flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
1549 if (a[ODP_FLOW_ATTR_ACTIONS]) {
1550 flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
1551 flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
1553 if (a[ODP_FLOW_ATTR_STATS]) {
1554 flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
1556 if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
1557 flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
1559 if (a[ODP_FLOW_ATTR_USED]) {
1560 flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
1565 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1566 * followed by Netlink attributes corresponding to 'flow'. */
1568 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1571 struct odp_header *odp_header;
1573 nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
1574 NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1577 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1578 odp_header->dp_ifindex = flow->dp_ifindex;
1580 if (flow->key_len) {
1581 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
1584 if (flow->actions || flow->actions_len) {
1585 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
1586 flow->actions, flow->actions_len);
1589 /* We never need to send these to the kernel. */
1590 assert(!flow->stats);
1591 assert(!flow->tcp_flags);
1592 assert(!flow->used);
1595 nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
1599 /* Clears 'flow' to "empty" values. */
1601 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1603 memset(flow, 0, sizeof *flow);
1606 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1607 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1608 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1609 * result of the command is expected to be a flow also, which is decoded and
1610 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1611 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1613 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1614 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1616 struct ofpbuf *request_buf;
1619 assert((reply != NULL) == (bufp != NULL));
1621 request_buf = ofpbuf_new(1024);
1622 dpif_linux_flow_to_ofpbuf(request, request_buf);
1623 error = nl_sock_transact(genl_sock, request_buf, bufp);
1624 ofpbuf_delete(request_buf);
1628 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1631 dpif_linux_flow_init(reply);
1632 ofpbuf_delete(*bufp);
1640 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1641 struct dpif_flow_stats *stats)
1644 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1645 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1647 stats->n_packets = 0;
1650 stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1651 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;