2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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"
26 #include <linux/types.h>
27 #include <linux/pkt_sched.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/sockios.h>
33 #include <sys/epoll.h>
38 #include "dpif-provider.h"
39 #include "dynamic-string.h"
42 #include "netdev-linux.h"
43 #include "netdev-vport.h"
44 #include "netlink-notifier.h"
45 #include "netlink-socket.h"
49 #include "openvswitch/datapath-compat.h"
50 #include "openvswitch/tunnel.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
61 VLOG_DEFINE_THIS_MODULE(dpif_linux);
62 enum { MAX_PORTS = USHRT_MAX };
64 /* This ethtool flag was introduced in Linux 2.6.24, so it might be
65 * missing if we have old headers. */
66 #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
68 struct dpif_linux_dp {
69 /* Generic Netlink header. */
72 /* struct ovs_header. */
76 const char *name; /* OVS_DP_ATTR_NAME. */
77 const uint32_t *upcall_pid; /* OVS_DP_UPCALL_PID. */
78 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
81 static void dpif_linux_dp_init(struct dpif_linux_dp *);
82 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
83 const struct ofpbuf *);
84 static void dpif_linux_dp_dump_start(struct nl_dump *);
85 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
86 struct dpif_linux_dp *reply,
87 struct ofpbuf **bufp);
88 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
89 struct ofpbuf **bufp);
91 struct dpif_linux_flow {
92 /* Generic Netlink header. */
95 /* struct ovs_header. */
96 unsigned int nlmsg_flags;
101 * The 'stats' member points to 64-bit data that might only be aligned on
102 * 32-bit boundaries, so get_unaligned_u64() should be used to access its
105 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
106 * the Netlink version of the command, even if actions_len is zero. */
107 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
109 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
111 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
112 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
113 const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
114 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
117 static void dpif_linux_flow_init(struct dpif_linux_flow *);
118 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
119 const struct ofpbuf *);
120 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
122 static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
123 struct dpif_linux_flow *reply,
124 struct ofpbuf **bufp);
125 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
126 struct dpif_flow_stats *);
128 /* One of the dpif channels between the kernel and userspace. */
129 struct dpif_channel {
130 struct nl_sock *sock; /* Netlink socket. */
131 long long int last_poll; /* Last time this channel was polled. */
134 static void report_loss(struct dpif *, struct dpif_channel *);
136 /* Datapath interface for the openvswitch Linux kernel module. */
141 /* Upcall messages. */
142 int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
143 struct dpif_channel *channels;
144 struct epoll_event *epoll_events;
145 int epoll_fd; /* epoll fd that includes channel socks. */
146 int n_events; /* Num events returned by epoll_wait(). */
147 int event_offset; /* Offset into 'epoll_events'. */
149 /* Change notification. */
150 struct sset changed_ports; /* Ports that have changed. */
151 struct nln_notifier *port_notifier;
155 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
157 /* Generic Netlink family numbers for OVS. */
158 static int ovs_datapath_family;
159 static int ovs_vport_family;
160 static int ovs_flow_family;
161 static int ovs_packet_family;
163 /* Generic Netlink socket. */
164 static struct nl_sock *genl_sock;
165 static struct nln *nln = NULL;
167 static int dpif_linux_init(void);
168 static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
169 static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
170 static void dpif_linux_port_changed(const void *vport, void *dpif);
171 static uint32_t dpif_linux_port_get_pid(const struct dpif *, uint32_t port_no);
173 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
175 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
176 const struct ofpbuf *);
178 static struct dpif_linux *
179 dpif_linux_cast(const struct dpif *dpif)
181 dpif_assert_class(dpif, &dpif_linux_class);
182 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
186 dpif_linux_enumerate(struct sset *all_dps)
192 error = dpif_linux_init();
197 dpif_linux_dp_dump_start(&dump);
198 while (nl_dump_next(&dump, &msg)) {
199 struct dpif_linux_dp dp;
201 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
202 sset_add(all_dps, dp.name);
205 return nl_dump_done(&dump);
209 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
210 bool create, struct dpif **dpifp)
212 struct dpif_linux_dp dp_request, dp;
217 error = dpif_linux_init();
222 /* Create or look up datapath. */
223 dpif_linux_dp_init(&dp_request);
225 dp_request.cmd = OVS_DP_CMD_NEW;
227 dp_request.upcall_pid = &upcall_pid;
229 dp_request.cmd = OVS_DP_CMD_GET;
231 dp_request.name = name;
232 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
237 open_dpif(&dp, dpifp);
243 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
245 struct dpif_linux *dpif;
247 dpif = xzalloc(sizeof *dpif);
248 dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
252 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
253 dp->dp_ifindex, dp->dp_ifindex);
255 dpif->dp_ifindex = dp->dp_ifindex;
256 sset_init(&dpif->changed_ports);
257 *dpifp = &dpif->dpif;
261 destroy_channels(struct dpif_linux *dpif)
265 if (dpif->epoll_fd < 0) {
269 for (i = 0; i < dpif->uc_array_size; i++ ) {
270 struct dpif_linux_vport vport_request;
271 struct dpif_channel *ch = &dpif->channels[i];
272 uint32_t upcall_pid = 0;
278 /* Turn off upcalls. */
279 dpif_linux_vport_init(&vport_request);
280 vport_request.cmd = OVS_VPORT_CMD_SET;
281 vport_request.dp_ifindex = dpif->dp_ifindex;
282 vport_request.port_no = i;
283 vport_request.upcall_pid = &upcall_pid;
284 dpif_linux_vport_transact(&vport_request, NULL, NULL);
286 nl_sock_destroy(ch->sock);
289 free(dpif->channels);
290 dpif->channels = NULL;
291 dpif->uc_array_size = 0;
293 free(dpif->epoll_events);
294 dpif->epoll_events = NULL;
295 dpif->n_events = dpif->event_offset = 0;
297 close(dpif->epoll_fd);
302 add_channel(struct dpif_linux *dpif, uint32_t port_no, struct nl_sock *sock)
304 struct epoll_event event;
306 if (dpif->epoll_fd < 0) {
310 /* We assume that the datapath densely chooses port numbers, which
311 * can therefore be used as an index into an array of channels. */
312 if (port_no >= dpif->uc_array_size) {
313 int new_size = port_no + 1;
316 if (new_size > 65535) {
317 VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big",
318 dpif_name(&dpif->dpif), port_no);
322 dpif->channels = xrealloc(dpif->channels,
323 new_size * sizeof *dpif->channels);
324 for (i = dpif->uc_array_size; i < new_size; i++) {
325 dpif->channels[i].sock = NULL;
328 dpif->epoll_events = xrealloc(dpif->epoll_events,
329 new_size * sizeof *dpif->epoll_events);
330 dpif->uc_array_size = new_size;
333 memset(&event, 0, sizeof event);
334 event.events = EPOLLIN;
335 event.data.u32 = port_no;
336 if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
341 dpif->channels[port_no].sock = sock;
342 dpif->channels[port_no].last_poll = LLONG_MIN;
348 del_channel(struct dpif_linux *dpif, uint32_t port_no)
350 struct dpif_channel *ch;
352 if (dpif->epoll_fd < 0 || port_no >= dpif->uc_array_size) {
356 ch = &dpif->channels[port_no];
361 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
363 nl_sock_destroy(ch->sock);
368 dpif_linux_close(struct dpif *dpif_)
370 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
372 nln_notifier_destroy(dpif->port_notifier);
373 destroy_channels(dpif);
374 sset_destroy(&dpif->changed_ports);
379 dpif_linux_destroy(struct dpif *dpif_)
381 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
382 struct dpif_linux_dp dp;
384 dpif_linux_dp_init(&dp);
385 dp.cmd = OVS_DP_CMD_DEL;
386 dp.dp_ifindex = dpif->dp_ifindex;
387 return dpif_linux_dp_transact(&dp, NULL, NULL);
391 dpif_linux_run(struct dpif *dpif_ OVS_UNUSED)
399 dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
407 dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
409 struct dpif_linux_dp dp;
413 error = dpif_linux_dp_get(dpif_, &dp, &buf);
415 stats->n_hit = dp.stats.n_hit;
416 stats->n_missed = dp.stats.n_missed;
417 stats->n_lost = dp.stats.n_lost;
418 stats->n_flows = dp.stats.n_flows;
425 get_vport_type(const struct dpif_linux_vport *vport)
427 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
429 switch (vport->type) {
430 case OVS_VPORT_TYPE_NETDEV:
433 case OVS_VPORT_TYPE_INTERNAL:
436 case OVS_VPORT_TYPE_GRE:
439 case OVS_VPORT_TYPE_GRE64:
442 case OVS_VPORT_TYPE_CAPWAP:
445 case OVS_VPORT_TYPE_VXLAN:
448 case OVS_VPORT_TYPE_UNSPEC:
449 case __OVS_VPORT_TYPE_MAX:
453 VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
454 vport->dp_ifindex, vport->name, (unsigned int) vport->type);
459 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
462 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
463 const char *name = netdev_vport_get_dpif_port(netdev);
464 const char *type = netdev_get_type(netdev);
465 struct dpif_linux_vport request, reply;
466 struct nl_sock *sock = NULL;
471 if (dpif->epoll_fd >= 0) {
472 error = nl_sock_create(NETLINK_GENERIC, &sock);
478 dpif_linux_vport_init(&request);
479 request.cmd = OVS_VPORT_CMD_NEW;
480 request.dp_ifindex = dpif->dp_ifindex;
481 request.type = netdev_vport_get_vport_type(netdev);
482 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
483 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
484 "unsupported type `%s'",
485 dpif_name(dpif_), name, type);
486 nl_sock_destroy(sock);
491 if (request.type == OVS_VPORT_TYPE_NETDEV) {
492 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
495 request.port_no = *port_nop;
496 upcall_pid = sock ? nl_sock_pid(sock) : 0;
497 request.upcall_pid = &upcall_pid;
499 error = dpif_linux_vport_transact(&request, &reply, &buf);
501 *port_nop = reply.port_no;
502 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
503 dpif_name(dpif_), reply.port_no, upcall_pid);
505 if (error == EBUSY && *port_nop != UINT32_MAX) {
506 VLOG_INFO("%s: requested port %"PRIu32" is in use",
507 dpif_name(dpif_), *port_nop);
509 nl_sock_destroy(sock);
516 error = add_channel(dpif, *port_nop, sock);
518 VLOG_INFO("%s: could not add channel for port %s",
519 dpif_name(dpif_), name);
521 /* Delete the port. */
522 dpif_linux_vport_init(&request);
523 request.cmd = OVS_VPORT_CMD_DEL;
524 request.dp_ifindex = dpif->dp_ifindex;
525 request.port_no = *port_nop;
526 dpif_linux_vport_transact(&request, NULL, NULL);
528 nl_sock_destroy(sock);
537 dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no)
539 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
540 struct dpif_linux_vport vport;
543 dpif_linux_vport_init(&vport);
544 vport.cmd = OVS_VPORT_CMD_DEL;
545 vport.dp_ifindex = dpif->dp_ifindex;
546 vport.port_no = port_no;
547 error = dpif_linux_vport_transact(&vport, NULL, NULL);
549 del_channel(dpif, port_no);
555 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
556 const char *port_name, struct dpif_port *dpif_port)
558 struct dpif_linux_vport request;
559 struct dpif_linux_vport reply;
563 dpif_linux_vport_init(&request);
564 request.cmd = OVS_VPORT_CMD_GET;
565 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
566 request.port_no = port_no;
567 request.name = port_name;
569 error = dpif_linux_vport_transact(&request, &reply, &buf);
571 if (reply.dp_ifindex != request.dp_ifindex) {
572 /* A query by name reported that 'port_name' is in some datapath
573 * other than 'dpif', but the caller wants to know about 'dpif'. */
575 } else if (dpif_port) {
576 dpif_port->name = xstrdup(reply.name);
577 dpif_port->type = xstrdup(get_vport_type(&reply));
578 dpif_port->port_no = reply.port_no;
586 dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
587 struct dpif_port *dpif_port)
589 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
593 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
594 struct dpif_port *dpif_port)
596 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
600 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
606 dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_t port_no)
608 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
610 if (dpif->epoll_fd < 0) {
613 /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s
614 * channel, since it is not heavily loaded. */
615 int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no;
616 return nl_sock_pid(dpif->channels[idx].sock);
621 dpif_linux_flow_flush(struct dpif *dpif_)
623 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
624 struct dpif_linux_flow flow;
626 dpif_linux_flow_init(&flow);
627 flow.cmd = OVS_FLOW_CMD_DEL;
628 flow.dp_ifindex = dpif->dp_ifindex;
629 return dpif_linux_flow_transact(&flow, NULL, NULL);
632 struct dpif_linux_port_state {
637 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
639 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
640 struct dpif_linux_port_state *state;
641 struct dpif_linux_vport request;
644 *statep = state = xmalloc(sizeof *state);
646 dpif_linux_vport_init(&request);
647 request.cmd = OVS_DP_CMD_GET;
648 request.dp_ifindex = dpif->dp_ifindex;
650 buf = ofpbuf_new(1024);
651 dpif_linux_vport_to_ofpbuf(&request, buf);
652 nl_dump_start(&state->dump, genl_sock, buf);
659 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
660 struct dpif_port *dpif_port)
662 struct dpif_linux_port_state *state = state_;
663 struct dpif_linux_vport vport;
667 if (!nl_dump_next(&state->dump, &buf)) {
671 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
676 dpif_port->name = CONST_CAST(char *, vport.name);
677 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
678 dpif_port->port_no = vport.port_no;
683 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
685 struct dpif_linux_port_state *state = state_;
686 int error = nl_dump_done(&state->dump);
693 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
695 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
697 if (dpif->change_error) {
698 dpif->change_error = false;
699 sset_clear(&dpif->changed_ports);
701 } else if (!sset_is_empty(&dpif->changed_ports)) {
702 *devnamep = sset_pop(&dpif->changed_ports);
710 dpif_linux_port_poll_wait(const struct dpif *dpif_)
712 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
713 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
714 poll_immediate_wake();
719 dpif_linux_flow_get__(const struct dpif *dpif_,
720 const struct nlattr *key, size_t key_len,
721 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
723 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
724 struct dpif_linux_flow request;
726 dpif_linux_flow_init(&request);
727 request.cmd = OVS_FLOW_CMD_GET;
728 request.dp_ifindex = dpif->dp_ifindex;
730 request.key_len = key_len;
731 return dpif_linux_flow_transact(&request, reply, bufp);
735 dpif_linux_flow_get(const struct dpif *dpif_,
736 const struct nlattr *key, size_t key_len,
737 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
739 struct dpif_linux_flow reply;
743 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
746 dpif_linux_flow_get_stats(&reply, stats);
749 buf->data = CONST_CAST(struct nlattr *, reply.actions);
750 buf->size = reply.actions_len;
760 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
761 struct dpif_linux_flow *request)
763 static struct nlattr dummy_action;
765 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
767 dpif_linux_flow_init(request);
768 request->cmd = (put->flags & DPIF_FP_CREATE
769 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
770 request->dp_ifindex = dpif->dp_ifindex;
771 request->key = put->key;
772 request->key_len = put->key_len;
773 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
774 request->actions = put->actions ? put->actions : &dummy_action;
775 request->actions_len = put->actions_len;
776 if (put->flags & DPIF_FP_ZERO_STATS) {
777 request->clear = true;
779 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
783 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
785 struct dpif_linux_flow request, reply;
789 dpif_linux_init_flow_put(dpif_, put, &request);
790 error = dpif_linux_flow_transact(&request,
791 put->stats ? &reply : NULL,
792 put->stats ? &buf : NULL);
793 if (!error && put->stats) {
794 dpif_linux_flow_get_stats(&reply, put->stats);
801 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
802 struct dpif_linux_flow *request)
804 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
806 dpif_linux_flow_init(request);
807 request->cmd = OVS_FLOW_CMD_DEL;
808 request->dp_ifindex = dpif->dp_ifindex;
809 request->key = del->key;
810 request->key_len = del->key_len;
814 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
816 struct dpif_linux_flow request, reply;
820 dpif_linux_init_flow_del(dpif_, del, &request);
821 error = dpif_linux_flow_transact(&request,
822 del->stats ? &reply : NULL,
823 del->stats ? &buf : NULL);
824 if (!error && del->stats) {
825 dpif_linux_flow_get_stats(&reply, del->stats);
831 struct dpif_linux_flow_state {
833 struct dpif_linux_flow flow;
834 struct dpif_flow_stats stats;
839 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
841 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
842 struct dpif_linux_flow_state *state;
843 struct dpif_linux_flow request;
846 *statep = state = xmalloc(sizeof *state);
848 dpif_linux_flow_init(&request);
849 request.cmd = OVS_DP_CMD_GET;
850 request.dp_ifindex = dpif->dp_ifindex;
852 buf = ofpbuf_new(1024);
853 dpif_linux_flow_to_ofpbuf(&request, buf);
854 nl_dump_start(&state->dump, genl_sock, buf);
863 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
864 const struct nlattr **key, size_t *key_len,
865 const struct nlattr **actions, size_t *actions_len,
866 const struct dpif_flow_stats **stats)
868 struct dpif_linux_flow_state *state = state_;
873 ofpbuf_delete(state->buf);
876 if (!nl_dump_next(&state->dump, &buf)) {
880 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
885 if (actions && !state->flow.actions) {
886 error = dpif_linux_flow_get__(dpif_, state->flow.key,
888 &state->flow, &state->buf);
889 if (error == ENOENT) {
890 VLOG_DBG("dumped flow disappeared on get");
892 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
898 *actions = state->flow.actions;
899 *actions_len = state->flow.actions_len;
902 *key = state->flow.key;
903 *key_len = state->flow.key_len;
906 dpif_linux_flow_get_stats(&state->flow, &state->stats);
907 *stats = &state->stats;
913 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
915 struct dpif_linux_flow_state *state = state_;
916 int error = nl_dump_done(&state->dump);
917 ofpbuf_delete(state->buf);
923 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
926 struct ovs_header *k_exec;
928 ofpbuf_prealloc_tailroom(buf, (64
929 + d_exec->packet->size
931 + d_exec->actions_len));
933 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
934 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
936 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
937 k_exec->dp_ifindex = dp_ifindex;
939 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
940 d_exec->packet->data, d_exec->packet->size);
941 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
942 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
943 d_exec->actions, d_exec->actions_len);
947 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
949 uint64_t request_stub[1024 / 8];
950 struct ofpbuf request;
953 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
954 dpif_linux_encode_execute(dp_ifindex, execute, &request);
955 error = nl_sock_transact(genl_sock, &request, NULL);
956 ofpbuf_uninit(&request);
962 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
964 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
966 return dpif_linux_execute__(dpif->dp_ifindex, execute);
972 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
974 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
977 struct nl_transaction txn;
979 struct ofpbuf request;
980 uint64_t request_stub[1024 / 8];
983 uint64_t reply_stub[1024 / 8];
986 struct nl_transaction *txnsp[MAX_OPS];
989 ovs_assert(n_ops <= MAX_OPS);
990 for (i = 0; i < n_ops; i++) {
991 struct op_auxdata *aux = &auxes[i];
992 struct dpif_op *op = ops[i];
993 struct dpif_flow_put *put;
994 struct dpif_flow_del *del;
995 struct dpif_execute *execute;
996 struct dpif_linux_flow flow;
998 ofpbuf_use_stub(&aux->request,
999 aux->request_stub, sizeof aux->request_stub);
1000 aux->txn.request = &aux->request;
1002 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1003 aux->txn.reply = NULL;
1006 case DPIF_OP_FLOW_PUT:
1007 put = &op->u.flow_put;
1008 dpif_linux_init_flow_put(dpif_, put, &flow);
1010 flow.nlmsg_flags |= NLM_F_ECHO;
1011 aux->txn.reply = &aux->reply;
1013 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1016 case DPIF_OP_FLOW_DEL:
1017 del = &op->u.flow_del;
1018 dpif_linux_init_flow_del(dpif_, del, &flow);
1020 flow.nlmsg_flags |= NLM_F_ECHO;
1021 aux->txn.reply = &aux->reply;
1023 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1026 case DPIF_OP_EXECUTE:
1027 execute = &op->u.execute;
1028 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1037 for (i = 0; i < n_ops; i++) {
1038 txnsp[i] = &auxes[i].txn;
1040 nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
1042 for (i = 0; i < n_ops; i++) {
1043 struct op_auxdata *aux = &auxes[i];
1044 struct nl_transaction *txn = &auxes[i].txn;
1045 struct dpif_op *op = ops[i];
1046 struct dpif_flow_put *put;
1047 struct dpif_flow_del *del;
1049 op->error = txn->error;
1052 case DPIF_OP_FLOW_PUT:
1053 put = &op->u.flow_put;
1056 struct dpif_linux_flow reply;
1058 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1061 dpif_linux_flow_get_stats(&reply, put->stats);
1066 memset(put->stats, 0, sizeof *put->stats);
1071 case DPIF_OP_FLOW_DEL:
1072 del = &op->u.flow_del;
1075 struct dpif_linux_flow reply;
1077 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1080 dpif_linux_flow_get_stats(&reply, del->stats);
1085 memset(del->stats, 0, sizeof *del->stats);
1090 case DPIF_OP_EXECUTE:
1097 ofpbuf_uninit(&aux->request);
1098 ofpbuf_uninit(&aux->reply);
1103 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1106 size_t chunk = MIN(n_ops, MAX_OPS);
1107 dpif_linux_operate__(dpif, ops, chunk);
1114 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1116 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1118 if ((dpif->epoll_fd >= 0) == enable) {
1123 destroy_channels(dpif);
1125 struct dpif_port_dump port_dump;
1126 struct dpif_port port;
1128 dpif->epoll_fd = epoll_create(10);
1129 if (dpif->epoll_fd < 0) {
1133 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
1134 struct dpif_linux_vport vport_request;
1135 struct nl_sock *sock;
1136 uint32_t upcall_pid;
1139 error = nl_sock_create(NETLINK_GENERIC, &sock);
1144 upcall_pid = nl_sock_pid(sock);
1146 dpif_linux_vport_init(&vport_request);
1147 vport_request.cmd = OVS_VPORT_CMD_SET;
1148 vport_request.dp_ifindex = dpif->dp_ifindex;
1149 vport_request.port_no = port.port_no;
1150 vport_request.upcall_pid = &upcall_pid;
1151 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1153 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1154 dpif_name(&dpif->dpif), vport_request.port_no,
1157 VLOG_WARN_RL(&error_rl,
1158 "%s: failed to set upcall pid on port: %s",
1159 dpif_name(&dpif->dpif), strerror(error));
1160 nl_sock_destroy(sock);
1162 if (error == ENODEV || error == ENOENT) {
1163 /* This device isn't there, but keep trying the others. */
1170 error = add_channel(dpif, port.port_no, sock);
1172 VLOG_INFO("%s: could not add channel for port %s",
1173 dpif_name(dpif_), port.name);
1174 nl_sock_destroy(sock);
1184 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1185 uint32_t queue_id, uint32_t *priority)
1187 if (queue_id < 0xf000) {
1188 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1196 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1199 static const struct nl_policy ovs_packet_policy[] = {
1200 /* Always present. */
1201 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1202 .min_len = ETH_HEADER_LEN },
1203 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1205 /* OVS_PACKET_CMD_ACTION only. */
1206 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
1209 struct ovs_header *ovs_header;
1210 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1211 struct nlmsghdr *nlmsg;
1212 struct genlmsghdr *genl;
1216 ofpbuf_use_const(&b, buf->data, buf->size);
1218 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1219 genl = ofpbuf_try_pull(&b, sizeof *genl);
1220 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1221 if (!nlmsg || !genl || !ovs_header
1222 || nlmsg->nlmsg_type != ovs_packet_family
1223 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1224 ARRAY_SIZE(ovs_packet_policy))) {
1228 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1229 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1235 memset(upcall, 0, sizeof *upcall);
1236 upcall->type = type;
1237 upcall->packet = buf;
1238 upcall->packet->data = CONST_CAST(struct nlattr *,
1239 nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
1240 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1241 upcall->key = CONST_CAST(struct nlattr *,
1242 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1243 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1244 upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
1245 ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
1247 *dp_ifindex = ovs_header->dp_ifindex;
1253 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1256 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1259 if (dpif->epoll_fd < 0) {
1263 if (dpif->event_offset >= dpif->n_events) {
1266 dpif->event_offset = dpif->n_events = 0;
1269 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1270 dpif->uc_array_size, 0);
1271 } while (retval < 0 && errno == EINTR);
1273 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1274 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
1275 } else if (retval > 0) {
1276 dpif->n_events = retval;
1280 while (dpif->event_offset < dpif->n_events) {
1281 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1282 struct dpif_channel *ch = &dpif->channels[idx];
1284 dpif->event_offset++;
1290 if (++read_tries > 50) {
1294 error = nl_sock_recv(ch->sock, buf, false);
1295 if (error == ENOBUFS) {
1296 /* ENOBUFS typically means that we've received so many
1297 * packets that the buffer overflowed. Try again
1298 * immediately because there's almost certainly a packet
1299 * waiting for us. */
1300 report_loss(dpif_, ch);
1304 ch->last_poll = time_msec();
1306 if (error == EAGAIN) {
1312 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1313 if (!error && dp_ifindex == dpif->dp_ifindex) {
1325 dpif_linux_recv_wait(struct dpif *dpif_)
1327 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1329 if (dpif->epoll_fd < 0) {
1333 poll_fd_wait(dpif->epoll_fd, POLLIN);
1337 dpif_linux_recv_purge(struct dpif *dpif_)
1339 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1340 struct dpif_channel *ch;
1342 if (dpif->epoll_fd < 0) {
1346 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
1348 nl_sock_drain(ch->sock);
1353 const struct dpif_class dpif_linux_class = {
1355 dpif_linux_enumerate,
1362 dpif_linux_get_stats,
1363 dpif_linux_port_add,
1364 dpif_linux_port_del,
1365 dpif_linux_port_query_by_number,
1366 dpif_linux_port_query_by_name,
1367 dpif_linux_get_max_ports,
1368 dpif_linux_port_get_pid,
1369 dpif_linux_port_dump_start,
1370 dpif_linux_port_dump_next,
1371 dpif_linux_port_dump_done,
1372 dpif_linux_port_poll,
1373 dpif_linux_port_poll_wait,
1374 dpif_linux_flow_get,
1375 dpif_linux_flow_put,
1376 dpif_linux_flow_del,
1377 dpif_linux_flow_flush,
1378 dpif_linux_flow_dump_start,
1379 dpif_linux_flow_dump_next,
1380 dpif_linux_flow_dump_done,
1383 dpif_linux_recv_set,
1384 dpif_linux_queue_to_priority,
1386 dpif_linux_recv_wait,
1387 dpif_linux_recv_purge,
1391 dpif_linux_init(void)
1393 static int error = -1;
1396 unsigned int ovs_vport_mcgroup;
1398 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1399 &ovs_datapath_family);
1401 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1402 "The Open vSwitch kernel module is probably not loaded.",
1403 OVS_DATAPATH_FAMILY);
1406 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1409 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1412 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1413 &ovs_packet_family);
1416 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1419 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1421 OVS_VPORT_MCGROUP_FALLBACK_ID);
1424 static struct dpif_linux_vport vport;
1425 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1426 dpif_linux_nln_parse, &vport);
1434 dpif_linux_is_internal_device(const char *name)
1436 struct dpif_linux_vport reply;
1440 error = dpif_linux_vport_get(name, &reply, &buf);
1443 } else if (error != ENODEV && error != ENOENT) {
1444 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1445 name, strerror(error));
1448 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1452 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1454 struct dpif_linux_vport *vport = vport_;
1455 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1459 dpif_linux_port_changed(const void *vport_, void *dpif_)
1461 const struct dpif_linux_vport *vport = vport_;
1462 struct dpif_linux *dpif = dpif_;
1465 if (vport->dp_ifindex == dpif->dp_ifindex
1466 && (vport->cmd == OVS_VPORT_CMD_NEW
1467 || vport->cmd == OVS_VPORT_CMD_DEL
1468 || vport->cmd == OVS_VPORT_CMD_SET)) {
1469 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1470 dpif->dpif.full_name, vport->name, vport->cmd);
1471 sset_add(&dpif->changed_ports, vport->name);
1474 dpif->change_error = true;
1478 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1479 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1480 * positive errno value.
1482 * 'vport' will contain pointers into 'buf', so the caller should not free
1483 * 'buf' while 'vport' is still in use. */
1485 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1486 const struct ofpbuf *buf)
1488 static const struct nl_policy ovs_vport_policy[] = {
1489 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1490 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1491 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1492 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1493 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1495 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1498 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1499 struct ovs_header *ovs_header;
1500 struct nlmsghdr *nlmsg;
1501 struct genlmsghdr *genl;
1504 dpif_linux_vport_init(vport);
1506 ofpbuf_use_const(&b, buf->data, buf->size);
1507 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1508 genl = ofpbuf_try_pull(&b, sizeof *genl);
1509 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1510 if (!nlmsg || !genl || !ovs_header
1511 || nlmsg->nlmsg_type != ovs_vport_family
1512 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1513 ARRAY_SIZE(ovs_vport_policy))) {
1517 vport->cmd = genl->cmd;
1518 vport->dp_ifindex = ovs_header->dp_ifindex;
1519 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1520 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1521 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1522 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1523 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1525 if (a[OVS_VPORT_ATTR_STATS]) {
1526 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1528 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1529 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1530 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1535 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1536 * followed by Netlink attributes corresponding to 'vport'. */
1538 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1541 struct ovs_header *ovs_header;
1543 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1544 vport->cmd, OVS_VPORT_VERSION);
1546 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1547 ovs_header->dp_ifindex = vport->dp_ifindex;
1549 if (vport->port_no != UINT32_MAX) {
1550 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1553 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1554 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1558 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1561 if (vport->upcall_pid) {
1562 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1566 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1567 vport->stats, sizeof *vport->stats);
1570 if (vport->options) {
1571 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1572 vport->options, vport->options_len);
1576 /* Clears 'vport' to "empty" values. */
1578 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1580 memset(vport, 0, sizeof *vport);
1581 vport->port_no = UINT32_MAX;
1584 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1585 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1586 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1587 * result of the command is expected to be an ovs_vport also, which is decoded
1588 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1589 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1591 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1592 struct dpif_linux_vport *reply,
1593 struct ofpbuf **bufp)
1595 struct ofpbuf *request_buf;
1598 ovs_assert((reply != NULL) == (bufp != NULL));
1600 error = dpif_linux_init();
1604 dpif_linux_vport_init(reply);
1609 request_buf = ofpbuf_new(1024);
1610 dpif_linux_vport_to_ofpbuf(request, request_buf);
1611 error = nl_sock_transact(genl_sock, request_buf, bufp);
1612 ofpbuf_delete(request_buf);
1616 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1619 dpif_linux_vport_init(reply);
1620 ofpbuf_delete(*bufp);
1627 /* Obtains information about the kernel vport named 'name' and stores it into
1628 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1629 * longer needed ('reply' will contain pointers into '*bufp'). */
1631 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1632 struct ofpbuf **bufp)
1634 struct dpif_linux_vport request;
1636 dpif_linux_vport_init(&request);
1637 request.cmd = OVS_VPORT_CMD_GET;
1638 request.name = name;
1640 return dpif_linux_vport_transact(&request, reply, bufp);
1643 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1644 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1645 * positive errno value.
1647 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1648 * while 'dp' is still in use. */
1650 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1652 static const struct nl_policy ovs_datapath_policy[] = {
1653 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1654 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1658 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1659 struct ovs_header *ovs_header;
1660 struct nlmsghdr *nlmsg;
1661 struct genlmsghdr *genl;
1664 dpif_linux_dp_init(dp);
1666 ofpbuf_use_const(&b, buf->data, buf->size);
1667 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1668 genl = ofpbuf_try_pull(&b, sizeof *genl);
1669 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1670 if (!nlmsg || !genl || !ovs_header
1671 || nlmsg->nlmsg_type != ovs_datapath_family
1672 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1673 ARRAY_SIZE(ovs_datapath_policy))) {
1677 dp->cmd = genl->cmd;
1678 dp->dp_ifindex = ovs_header->dp_ifindex;
1679 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1680 if (a[OVS_DP_ATTR_STATS]) {
1681 /* Can't use structure assignment because Netlink doesn't ensure
1682 * sufficient alignment for 64-bit members. */
1683 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1690 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1692 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1694 struct ovs_header *ovs_header;
1696 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1697 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1698 OVS_DATAPATH_VERSION);
1700 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1701 ovs_header->dp_ifindex = dp->dp_ifindex;
1704 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1707 if (dp->upcall_pid) {
1708 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1711 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1714 /* Clears 'dp' to "empty" values. */
1716 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1718 memset(dp, 0, sizeof *dp);
1722 dpif_linux_dp_dump_start(struct nl_dump *dump)
1724 struct dpif_linux_dp request;
1727 dpif_linux_dp_init(&request);
1728 request.cmd = OVS_DP_CMD_GET;
1730 buf = ofpbuf_new(1024);
1731 dpif_linux_dp_to_ofpbuf(&request, buf);
1732 nl_dump_start(dump, genl_sock, buf);
1736 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1737 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1738 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1739 * result of the command is expected to be of the same form, which is decoded
1740 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1741 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1743 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1744 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1746 struct ofpbuf *request_buf;
1749 ovs_assert((reply != NULL) == (bufp != NULL));
1751 request_buf = ofpbuf_new(1024);
1752 dpif_linux_dp_to_ofpbuf(request, request_buf);
1753 error = nl_sock_transact(genl_sock, request_buf, bufp);
1754 ofpbuf_delete(request_buf);
1758 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1761 dpif_linux_dp_init(reply);
1762 ofpbuf_delete(*bufp);
1769 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1770 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1771 * will contain pointers into '*bufp'). */
1773 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1774 struct ofpbuf **bufp)
1776 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1777 struct dpif_linux_dp request;
1779 dpif_linux_dp_init(&request);
1780 request.cmd = OVS_DP_CMD_GET;
1781 request.dp_ifindex = dpif->dp_ifindex;
1783 return dpif_linux_dp_transact(&request, reply, bufp);
1786 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1787 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1788 * positive errno value.
1790 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1791 * while 'flow' is still in use. */
1793 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1794 const struct ofpbuf *buf)
1796 static const struct nl_policy ovs_flow_policy[] = {
1797 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1798 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1799 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1801 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1802 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1803 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1806 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1807 struct ovs_header *ovs_header;
1808 struct nlmsghdr *nlmsg;
1809 struct genlmsghdr *genl;
1812 dpif_linux_flow_init(flow);
1814 ofpbuf_use_const(&b, buf->data, buf->size);
1815 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1816 genl = ofpbuf_try_pull(&b, sizeof *genl);
1817 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1818 if (!nlmsg || !genl || !ovs_header
1819 || nlmsg->nlmsg_type != ovs_flow_family
1820 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1821 ARRAY_SIZE(ovs_flow_policy))) {
1825 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1826 flow->dp_ifindex = ovs_header->dp_ifindex;
1827 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1828 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1829 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1830 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1831 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1833 if (a[OVS_FLOW_ATTR_STATS]) {
1834 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1836 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1837 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1839 if (a[OVS_FLOW_ATTR_USED]) {
1840 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1845 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1846 * followed by Netlink attributes corresponding to 'flow'. */
1848 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1851 struct ovs_header *ovs_header;
1853 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1854 NLM_F_REQUEST | flow->nlmsg_flags,
1855 flow->cmd, OVS_FLOW_VERSION);
1857 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1858 ovs_header->dp_ifindex = flow->dp_ifindex;
1860 if (flow->key_len) {
1861 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1864 if (flow->actions || flow->actions_len) {
1865 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1866 flow->actions, flow->actions_len);
1869 /* We never need to send these to the kernel. */
1870 ovs_assert(!flow->stats);
1871 ovs_assert(!flow->tcp_flags);
1872 ovs_assert(!flow->used);
1875 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1879 /* Clears 'flow' to "empty" values. */
1881 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1883 memset(flow, 0, sizeof *flow);
1886 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1887 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1888 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1889 * result of the command is expected to be a flow also, which is decoded and
1890 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1891 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1893 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1894 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1896 struct ofpbuf *request_buf;
1899 ovs_assert((reply != NULL) == (bufp != NULL));
1902 request->nlmsg_flags |= NLM_F_ECHO;
1905 request_buf = ofpbuf_new(1024);
1906 dpif_linux_flow_to_ofpbuf(request, request_buf);
1907 error = nl_sock_transact(genl_sock, request_buf, bufp);
1908 ofpbuf_delete(request_buf);
1912 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1915 dpif_linux_flow_init(reply);
1916 ofpbuf_delete(*bufp);
1924 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1925 struct dpif_flow_stats *stats)
1928 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1929 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1931 stats->n_packets = 0;
1934 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1935 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1938 /* Logs information about a packet that was recently lost in 'ch' (in
1941 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
1943 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1944 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1947 if (VLOG_DROP_WARN(&rl)) {
1952 if (ch->last_poll != LLONG_MIN) {
1953 ds_put_format(&s, " (last polled %lld ms ago)",
1954 time_msec() - ch->last_poll);
1957 VLOG_WARN("%s: lost packet on channel %td%s",
1958 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));