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 {
475 unsigned long *port_bitmap; /* Ports in the datapath. */
476 bool complete; /* Dump completed without error. */
480 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
482 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
483 struct dpif_linux_port_state *state;
484 struct dpif_linux_vport request;
487 *statep = state = xmalloc(sizeof *state);
488 state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
489 state->complete = false;
491 dpif_linux_vport_init(&request);
492 request.cmd = ODP_DP_CMD_GET;
493 request.dp_ifindex = dpif->dp_ifindex;
495 buf = ofpbuf_new(1024);
496 dpif_linux_vport_to_ofpbuf(&request, buf);
497 nl_dump_start(&state->dump, genl_sock, buf);
504 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
505 struct dpif_port *dpif_port)
507 struct dpif_linux_port_state *state = state_;
508 struct dpif_linux_vport vport;
512 if (!nl_dump_next(&state->dump, &buf)) {
513 state->complete = true;
517 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
522 if (vport.port_no < LRU_MAX_PORTS) {
523 bitmap_set1(state->port_bitmap, vport.port_no);
526 dpif_port->name = (char *) vport.name;
527 dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
528 dpif_port->port_no = vport.port_no;
533 dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
535 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
536 struct dpif_linux_port_state *state = state_;
537 int error = nl_dump_done(&state->dump);
539 if (state->complete) {
542 for (i = 0; i < LRU_MAX_PORTS; i++) {
543 if (!bitmap_is_set(state->port_bitmap, i)) {
544 dpif_linux_push_port(dpif, i);
549 free(state->port_bitmap);
555 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
557 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
559 if (dpif->change_error) {
560 dpif->change_error = false;
561 sset_clear(&dpif->changed_ports);
563 } else if (!sset_is_empty(&dpif->changed_ports)) {
564 *devnamep = sset_pop(&dpif->changed_ports);
572 dpif_linux_port_poll_wait(const struct dpif *dpif_)
574 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
575 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
576 poll_immediate_wake();
578 rtnetlink_link_notifier_wait();
583 dpif_linux_flow_get__(const struct dpif *dpif_,
584 const struct nlattr *key, size_t key_len,
585 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
587 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
588 struct dpif_linux_flow request;
590 dpif_linux_flow_init(&request);
591 request.cmd = ODP_FLOW_CMD_GET;
592 request.dp_ifindex = dpif->dp_ifindex;
594 request.key_len = key_len;
595 return dpif_linux_flow_transact(&request, reply, bufp);
599 dpif_linux_flow_get(const struct dpif *dpif_,
600 const struct nlattr *key, size_t key_len,
601 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
603 struct dpif_linux_flow reply;
607 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
610 dpif_linux_flow_get_stats(&reply, stats);
613 buf->data = (void *) reply.actions;
614 buf->size = reply.actions_len;
624 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
625 const struct nlattr *key, size_t key_len,
626 const struct nlattr *actions, size_t actions_len,
627 struct dpif_flow_stats *stats)
629 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
630 struct dpif_linux_flow request, reply;
631 struct nlattr dummy_action;
635 dpif_linux_flow_init(&request);
636 request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
637 request.dp_ifindex = dpif->dp_ifindex;
639 request.key_len = key_len;
640 /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
641 request.actions = actions ? actions : &dummy_action;
642 request.actions_len = actions_len;
643 if (flags & DPIF_FP_ZERO_STATS) {
644 request.clear = true;
646 request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
647 error = dpif_linux_flow_transact(&request,
648 stats ? &reply : NULL,
649 stats ? &buf : NULL);
650 if (!error && stats) {
651 dpif_linux_flow_get_stats(&reply, stats);
658 dpif_linux_flow_del(struct dpif *dpif_,
659 const struct nlattr *key, size_t key_len,
660 struct dpif_flow_stats *stats)
662 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
663 struct dpif_linux_flow request, reply;
667 dpif_linux_flow_init(&request);
668 request.cmd = ODP_FLOW_CMD_DEL;
669 request.dp_ifindex = dpif->dp_ifindex;
671 request.key_len = key_len;
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);
682 struct dpif_linux_flow_state {
684 struct dpif_linux_flow flow;
685 struct dpif_flow_stats stats;
690 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
692 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
693 struct dpif_linux_flow_state *state;
694 struct dpif_linux_flow request;
697 *statep = state = xmalloc(sizeof *state);
699 dpif_linux_flow_init(&request);
700 request.cmd = ODP_DP_CMD_GET;
701 request.dp_ifindex = dpif->dp_ifindex;
703 buf = ofpbuf_new(1024);
704 dpif_linux_flow_to_ofpbuf(&request, buf);
705 nl_dump_start(&state->dump, genl_sock, buf);
714 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
715 const struct nlattr **key, size_t *key_len,
716 const struct nlattr **actions, size_t *actions_len,
717 const struct dpif_flow_stats **stats)
719 struct dpif_linux_flow_state *state = state_;
724 ofpbuf_delete(state->buf);
727 if (!nl_dump_next(&state->dump, &buf)) {
731 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
736 if (actions && !state->flow.actions) {
737 error = dpif_linux_flow_get__(dpif_, state->flow.key,
739 &state->flow, &state->buf);
740 if (error == ENOENT) {
741 VLOG_DBG("dumped flow disappeared on get");
743 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
749 *actions = state->flow.actions;
750 *actions_len = state->flow.actions_len;
753 *key = state->flow.key;
754 *key_len = state->flow.key_len;
757 dpif_linux_flow_get_stats(&state->flow, &state->stats);
758 *stats = &state->stats;
764 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
766 struct dpif_linux_flow_state *state = state_;
767 int error = nl_dump_done(&state->dump);
768 ofpbuf_delete(state->buf);
774 dpif_linux_execute(struct dpif *dpif_,
775 const struct nlattr *actions, size_t actions_len,
776 const struct ofpbuf *packet)
778 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
779 struct odp_header *execute;
783 buf = ofpbuf_new(128 + actions_len + packet->size);
785 nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
786 ODP_PACKET_CMD_EXECUTE, 1);
788 execute = ofpbuf_put_uninit(buf, sizeof *execute);
789 execute->dp_ifindex = dpif->dp_ifindex;
791 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
792 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
794 error = nl_sock_transact(genl_sock, buf, NULL);
800 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
802 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
803 *listen_mask = dpif->listen_mask;
808 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
810 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
814 if (listen_mask == dpif->listen_mask) {
816 } else if (!listen_mask) {
817 nl_sock_destroy(dpif->mc_sock);
818 dpif->mc_sock = NULL;
819 dpif->listen_mask = 0;
821 } else if (!dpif->mc_sock) {
822 error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
828 /* Unsubscribe from old groups. */
829 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
830 if (dpif->listen_mask & (1u << i)) {
831 nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
835 /* Update listen_mask. */
836 dpif->listen_mask = listen_mask;
838 /* Subscribe to new groups. */
840 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
841 if (dpif->listen_mask & (1u << i)) {
844 retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
854 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
855 uint32_t *probability)
857 struct dpif_linux_dp dp;
861 error = dpif_linux_dp_get(dpif_, &dp, &buf);
863 *probability = dp.sampling ? *dp.sampling : 0;
870 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
872 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
873 struct dpif_linux_dp dp;
875 dpif_linux_dp_init(&dp);
876 dp.cmd = ODP_DP_CMD_SET;
877 dp.dp_ifindex = dpif->dp_ifindex;
878 dp.sampling = &probability;
879 return dpif_linux_dp_transact(&dp, NULL, NULL);
883 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
884 uint32_t queue_id, uint32_t *priority)
886 if (queue_id < 0xf000) {
887 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
895 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
898 static const struct nl_policy odp_packet_policy[] = {
899 /* Always present. */
900 [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
901 .min_len = ETH_HEADER_LEN },
902 [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
904 /* ODP_PACKET_CMD_ACTION only. */
905 [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
907 /* ODP_PACKET_CMD_SAMPLE only. */
908 [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
909 [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
912 struct odp_header *odp_header;
913 struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
914 struct nlmsghdr *nlmsg;
915 struct genlmsghdr *genl;
919 ofpbuf_use_const(&b, buf->data, buf->size);
921 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
922 genl = ofpbuf_try_pull(&b, sizeof *genl);
923 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
924 if (!nlmsg || !genl || !odp_header
925 || nlmsg->nlmsg_type != odp_packet_family
926 || !nl_policy_parse(&b, 0, odp_packet_policy, a,
927 ARRAY_SIZE(odp_packet_policy))) {
931 type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
932 : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
933 : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
939 memset(upcall, 0, sizeof *upcall);
941 upcall->packet = buf;
942 upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
943 upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
944 upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
945 upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
946 upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
947 ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
949 upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
950 ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
952 if (a[ODP_PACKET_ATTR_ACTIONS]) {
953 upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
954 upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
957 *dp_ifindex = odp_header->dp_ifindex;
963 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
965 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
970 if (!dpif->mc_sock) {
974 for (i = 0; i < 50; i++) {
977 error = nl_sock_recv(dpif->mc_sock, &buf, false);
982 error = parse_odp_packet(buf, upcall, &dp_ifindex);
984 && dp_ifindex == dpif->dp_ifindex
985 && dpif->listen_mask & (1u << upcall->type)) {
999 dpif_linux_recv_wait(struct dpif *dpif_)
1001 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1002 if (dpif->mc_sock) {
1003 nl_sock_wait(dpif->mc_sock, POLLIN);
1008 dpif_linux_recv_purge(struct dpif *dpif_)
1010 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1012 if (dpif->mc_sock) {
1013 nl_sock_drain(dpif->mc_sock);
1017 const struct dpif_class dpif_linux_class = {
1021 dpif_linux_enumerate,
1025 dpif_linux_get_stats,
1026 dpif_linux_get_drop_frags,
1027 dpif_linux_set_drop_frags,
1028 dpif_linux_port_add,
1029 dpif_linux_port_del,
1030 dpif_linux_port_query_by_number,
1031 dpif_linux_port_query_by_name,
1032 dpif_linux_get_max_ports,
1033 dpif_linux_port_dump_start,
1034 dpif_linux_port_dump_next,
1035 dpif_linux_port_dump_done,
1036 dpif_linux_port_poll,
1037 dpif_linux_port_poll_wait,
1038 dpif_linux_flow_get,
1039 dpif_linux_flow_put,
1040 dpif_linux_flow_del,
1041 dpif_linux_flow_flush,
1042 dpif_linux_flow_dump_start,
1043 dpif_linux_flow_dump_next,
1044 dpif_linux_flow_dump_done,
1046 dpif_linux_recv_get_mask,
1047 dpif_linux_recv_set_mask,
1048 dpif_linux_get_sflow_probability,
1049 dpif_linux_set_sflow_probability,
1050 dpif_linux_queue_to_priority,
1052 dpif_linux_recv_wait,
1053 dpif_linux_recv_purge,
1057 dpif_linux_init(void)
1059 static int error = -1;
1062 error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
1063 &odp_datapath_family);
1065 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1066 "The Open vSwitch kernel module is probably not loaded.",
1067 ODP_DATAPATH_FAMILY);
1070 error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
1073 error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
1076 error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
1077 &odp_packet_family);
1080 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1088 dpif_linux_is_internal_device(const char *name)
1090 struct dpif_linux_vport reply;
1094 error = dpif_linux_vport_get(name, &reply, &buf);
1097 } else if (error != ENODEV && error != ENOENT) {
1098 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1099 name, strerror(error));
1102 return reply.type == ODP_VPORT_TYPE_INTERNAL;
1106 dpif_linux_port_changed(const struct rtnetlink_link_change *change,
1109 struct dpif_linux *dpif = dpif_;
1112 if (change->master_ifindex == dpif->dp_ifindex
1113 && (change->nlmsg_type == RTM_NEWLINK
1114 || change->nlmsg_type == RTM_DELLINK))
1116 /* Our datapath changed, either adding a new port or deleting an
1118 sset_add(&dpif->changed_ports, change->ifname);
1121 dpif->change_error = true;
1125 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1126 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1127 * positive errno value.
1129 * 'vport' will contain pointers into 'buf', so the caller should not free
1130 * 'buf' while 'vport' is still in use. */
1132 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1133 const struct ofpbuf *buf)
1135 static const struct nl_policy odp_vport_policy[] = {
1136 [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1137 [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1138 [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1139 [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1140 .min_len = sizeof(struct rtnl_link_stats64),
1141 .max_len = sizeof(struct rtnl_link_stats64),
1143 [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1144 .min_len = ETH_ADDR_LEN,
1145 .max_len = ETH_ADDR_LEN,
1147 [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
1148 [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1149 [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1150 [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
1153 struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
1154 struct odp_header *odp_header;
1155 struct nlmsghdr *nlmsg;
1156 struct genlmsghdr *genl;
1159 dpif_linux_vport_init(vport);
1161 ofpbuf_use_const(&b, buf->data, buf->size);
1162 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1163 genl = ofpbuf_try_pull(&b, sizeof *genl);
1164 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1165 if (!nlmsg || !genl || !odp_header
1166 || nlmsg->nlmsg_type != odp_vport_family
1167 || !nl_policy_parse(&b, 0, odp_vport_policy, a,
1168 ARRAY_SIZE(odp_vport_policy))) {
1172 vport->cmd = genl->cmd;
1173 vport->dp_ifindex = odp_header->dp_ifindex;
1174 vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1175 vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1176 vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
1177 if (a[ODP_VPORT_ATTR_STATS]) {
1178 vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
1180 if (a[ODP_VPORT_ATTR_ADDRESS]) {
1181 vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
1183 if (a[ODP_VPORT_ATTR_MTU]) {
1184 vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
1186 vport->mtu = INT_MAX;
1188 if (a[ODP_VPORT_ATTR_OPTIONS]) {
1189 vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
1190 vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]);
1192 if (a[ODP_VPORT_ATTR_IFINDEX]) {
1193 vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
1195 if (a[ODP_VPORT_ATTR_IFLINK]) {
1196 vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]);
1201 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1202 * followed by Netlink attributes corresponding to 'vport'. */
1204 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1207 struct odp_header *odp_header;
1209 nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1212 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1213 odp_header->dp_ifindex = vport->dp_ifindex;
1215 if (vport->port_no != UINT32_MAX) {
1216 nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1219 if (vport->type != ODP_VPORT_TYPE_UNSPEC) {
1220 nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type);
1224 nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
1228 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
1229 vport->stats, sizeof *vport->stats);
1232 if (vport->address) {
1233 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS,
1234 vport->address, ETH_ADDR_LEN);
1237 if (vport->mtu && vport->mtu != INT_MAX) {
1238 nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
1241 if (vport->options) {
1242 nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS,
1243 vport->options, vport->options_len);
1246 if (vport->ifindex) {
1247 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex);
1250 if (vport->iflink) {
1251 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
1255 /* Clears 'vport' to "empty" values. */
1257 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1259 memset(vport, 0, sizeof *vport);
1260 vport->port_no = UINT32_MAX;
1263 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1264 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1265 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1266 * result of the command is expected to be an odp_vport also, which is decoded
1267 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1268 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1270 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1271 struct dpif_linux_vport *reply,
1272 struct ofpbuf **bufp)
1274 struct ofpbuf *request_buf;
1277 assert((reply != NULL) == (bufp != NULL));
1279 error = dpif_linux_init();
1283 dpif_linux_vport_init(reply);
1288 request_buf = ofpbuf_new(1024);
1289 dpif_linux_vport_to_ofpbuf(request, request_buf);
1290 error = nl_sock_transact(genl_sock, request_buf, bufp);
1291 ofpbuf_delete(request_buf);
1295 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1298 dpif_linux_vport_init(reply);
1299 ofpbuf_delete(*bufp);
1306 /* Obtains information about the kernel vport named 'name' and stores it into
1307 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1308 * longer needed ('reply' will contain pointers into '*bufp'). */
1310 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1311 struct ofpbuf **bufp)
1313 struct dpif_linux_vport request;
1315 dpif_linux_vport_init(&request);
1316 request.cmd = ODP_VPORT_CMD_GET;
1317 request.name = name;
1319 return dpif_linux_vport_transact(&request, reply, bufp);
1322 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1323 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1324 * positive errno value.
1326 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1327 * while 'dp' is still in use. */
1329 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1331 static const struct nl_policy odp_datapath_policy[] = {
1332 [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1333 [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1334 .min_len = sizeof(struct odp_stats),
1335 .max_len = sizeof(struct odp_stats),
1337 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1338 [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1339 [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
1342 struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
1343 struct odp_header *odp_header;
1344 struct nlmsghdr *nlmsg;
1345 struct genlmsghdr *genl;
1348 dpif_linux_dp_init(dp);
1350 ofpbuf_use_const(&b, buf->data, buf->size);
1351 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1352 genl = ofpbuf_try_pull(&b, sizeof *genl);
1353 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1354 if (!nlmsg || !genl || !odp_header
1355 || nlmsg->nlmsg_type != odp_datapath_family
1356 || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
1357 ARRAY_SIZE(odp_datapath_policy))) {
1361 dp->cmd = genl->cmd;
1362 dp->dp_ifindex = odp_header->dp_ifindex;
1363 dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
1364 if (a[ODP_DP_ATTR_STATS]) {
1365 /* Can't use structure assignment because Netlink doesn't ensure
1366 * sufficient alignment for 64-bit members. */
1367 memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
1370 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1371 dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1373 if (a[ODP_DP_ATTR_SAMPLING]) {
1374 dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
1377 if (a[ODP_DP_ATTR_MCGROUPS]) {
1378 static const struct nl_policy odp_mcgroup_policy[] = {
1379 [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
1380 [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
1381 [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
1384 struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
1386 if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
1387 mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
1391 if (mcgroups[ODP_PACKET_CMD_MISS]) {
1392 dp->mcgroups[DPIF_UC_MISS]
1393 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
1395 if (mcgroups[ODP_PACKET_CMD_ACTION]) {
1396 dp->mcgroups[DPIF_UC_ACTION]
1397 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
1399 if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
1400 dp->mcgroups[DPIF_UC_SAMPLE]
1401 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
1408 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1410 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1412 struct odp_header *odp_header;
1414 nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
1415 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1417 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1418 odp_header->dp_ifindex = dp->dp_ifindex;
1421 nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
1424 /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
1426 if (dp->ipv4_frags) {
1427 nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1431 nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
1435 /* Clears 'dp' to "empty" values. */
1437 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1439 memset(dp, 0, sizeof *dp);
1443 dpif_linux_dp_dump_start(struct nl_dump *dump)
1445 struct dpif_linux_dp request;
1448 dpif_linux_dp_init(&request);
1449 request.cmd = ODP_DP_CMD_GET;
1451 buf = ofpbuf_new(1024);
1452 dpif_linux_dp_to_ofpbuf(&request, buf);
1453 nl_dump_start(dump, genl_sock, buf);
1457 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1458 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1459 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1460 * result of the command is expected to be of the same form, which is decoded
1461 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1462 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1464 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1465 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1467 struct ofpbuf *request_buf;
1470 assert((reply != NULL) == (bufp != NULL));
1472 request_buf = ofpbuf_new(1024);
1473 dpif_linux_dp_to_ofpbuf(request, request_buf);
1474 error = nl_sock_transact(genl_sock, request_buf, bufp);
1475 ofpbuf_delete(request_buf);
1479 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1482 dpif_linux_dp_init(reply);
1483 ofpbuf_delete(*bufp);
1490 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1491 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1492 * will contain pointers into '*bufp'). */
1494 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1495 struct ofpbuf **bufp)
1497 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1498 struct dpif_linux_dp request;
1500 dpif_linux_dp_init(&request);
1501 request.cmd = ODP_DP_CMD_GET;
1502 request.dp_ifindex = dpif->dp_ifindex;
1504 return dpif_linux_dp_transact(&request, reply, bufp);
1507 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1508 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1509 * positive errno value.
1511 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1512 * while 'flow' is still in use. */
1514 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1515 const struct ofpbuf *buf)
1517 static const struct nl_policy odp_flow_policy[] = {
1518 [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1519 [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1520 [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1521 .min_len = sizeof(struct odp_flow_stats),
1522 .max_len = sizeof(struct odp_flow_stats),
1524 [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1525 [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1526 /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
1529 struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
1530 struct odp_header *odp_header;
1531 struct nlmsghdr *nlmsg;
1532 struct genlmsghdr *genl;
1535 dpif_linux_flow_init(flow);
1537 ofpbuf_use_const(&b, buf->data, buf->size);
1538 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1539 genl = ofpbuf_try_pull(&b, sizeof *genl);
1540 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1541 if (!nlmsg || !genl || !odp_header
1542 || nlmsg->nlmsg_type != odp_flow_family
1543 || !nl_policy_parse(&b, 0, odp_flow_policy, a,
1544 ARRAY_SIZE(odp_flow_policy))) {
1548 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1549 flow->dp_ifindex = odp_header->dp_ifindex;
1550 flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
1551 flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
1552 if (a[ODP_FLOW_ATTR_ACTIONS]) {
1553 flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
1554 flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
1556 if (a[ODP_FLOW_ATTR_STATS]) {
1557 flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
1559 if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
1560 flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
1562 if (a[ODP_FLOW_ATTR_USED]) {
1563 flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
1568 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1569 * followed by Netlink attributes corresponding to 'flow'. */
1571 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1574 struct odp_header *odp_header;
1576 nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
1577 NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1580 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1581 odp_header->dp_ifindex = flow->dp_ifindex;
1583 if (flow->key_len) {
1584 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
1587 if (flow->actions || flow->actions_len) {
1588 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
1589 flow->actions, flow->actions_len);
1592 /* We never need to send these to the kernel. */
1593 assert(!flow->stats);
1594 assert(!flow->tcp_flags);
1595 assert(!flow->used);
1598 nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
1602 /* Clears 'flow' to "empty" values. */
1604 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1606 memset(flow, 0, sizeof *flow);
1609 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1610 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1611 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1612 * result of the command is expected to be a flow also, which is decoded and
1613 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1614 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1616 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1617 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1619 struct ofpbuf *request_buf;
1622 assert((reply != NULL) == (bufp != NULL));
1624 request_buf = ofpbuf_new(1024);
1625 dpif_linux_flow_to_ofpbuf(request, request_buf);
1626 error = nl_sock_transact(genl_sock, request_buf, bufp);
1627 ofpbuf_delete(request_buf);
1631 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1634 dpif_linux_flow_init(reply);
1635 ofpbuf_delete(*bufp);
1643 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1644 struct dpif_flow_stats *stats)
1647 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1648 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1650 stats->n_packets = 0;
1653 stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1654 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;