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_VXLAN:
445 case OVS_VPORT_TYPE_UNSPEC:
446 case __OVS_VPORT_TYPE_MAX:
450 VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
451 vport->dp_ifindex, vport->name, (unsigned int) vport->type);
455 static enum ovs_vport_type
456 netdev_to_ovs_vport_type(const struct netdev *netdev)
458 const char *type = netdev_get_type(netdev);
460 if (!strcmp(type, "tap") || !strcmp(type, "system")) {
461 return OVS_VPORT_TYPE_NETDEV;
462 } else if (!strcmp(type, "internal")) {
463 return OVS_VPORT_TYPE_INTERNAL;
464 } else if (strstr(type, "gre64")) {
465 return OVS_VPORT_TYPE_GRE64;
466 } else if (strstr(type, "gre")) {
467 return OVS_VPORT_TYPE_GRE;
468 } else if (!strcmp(type, "vxlan")) {
469 return OVS_VPORT_TYPE_VXLAN;
471 return OVS_VPORT_TYPE_UNSPEC;
476 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
479 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
480 const struct netdev_tunnel_config *tnl_cfg;
481 const char *name = netdev_vport_get_dpif_port(netdev);
482 const char *type = netdev_get_type(netdev);
483 struct dpif_linux_vport request, reply;
484 struct nl_sock *sock = NULL;
487 uint64_t options_stub[64 / 8];
488 struct ofpbuf options;
491 if (dpif->epoll_fd >= 0) {
492 error = nl_sock_create(NETLINK_GENERIC, &sock);
498 dpif_linux_vport_init(&request);
499 request.cmd = OVS_VPORT_CMD_NEW;
500 request.dp_ifindex = dpif->dp_ifindex;
501 request.type = netdev_to_ovs_vport_type(netdev);
502 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
503 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
504 "unsupported type `%s'",
505 dpif_name(dpif_), name, type);
506 nl_sock_destroy(sock);
511 if (request.type == OVS_VPORT_TYPE_NETDEV) {
512 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
515 tnl_cfg = netdev_get_tunnel_config(netdev);
516 if (tnl_cfg && tnl_cfg->dst_port != 0) {
517 ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
518 nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
519 ntohs(tnl_cfg->dst_port));
520 request.options = options.data;
521 request.options_len = options.size;
524 request.port_no = *port_nop;
525 upcall_pid = sock ? nl_sock_pid(sock) : 0;
526 request.upcall_pid = &upcall_pid;
528 error = dpif_linux_vport_transact(&request, &reply, &buf);
530 *port_nop = reply.port_no;
531 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
532 dpif_name(dpif_), reply.port_no, upcall_pid);
534 if (error == EBUSY && *port_nop != UINT32_MAX) {
535 VLOG_INFO("%s: requested port %"PRIu32" is in use",
536 dpif_name(dpif_), *port_nop);
538 nl_sock_destroy(sock);
545 error = add_channel(dpif, *port_nop, sock);
547 VLOG_INFO("%s: could not add channel for port %s",
548 dpif_name(dpif_), name);
550 /* Delete the port. */
551 dpif_linux_vport_init(&request);
552 request.cmd = OVS_VPORT_CMD_DEL;
553 request.dp_ifindex = dpif->dp_ifindex;
554 request.port_no = *port_nop;
555 dpif_linux_vport_transact(&request, NULL, NULL);
557 nl_sock_destroy(sock);
566 dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no)
568 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
569 struct dpif_linux_vport vport;
572 dpif_linux_vport_init(&vport);
573 vport.cmd = OVS_VPORT_CMD_DEL;
574 vport.dp_ifindex = dpif->dp_ifindex;
575 vport.port_no = port_no;
576 error = dpif_linux_vport_transact(&vport, NULL, NULL);
578 del_channel(dpif, port_no);
584 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
585 const char *port_name, struct dpif_port *dpif_port)
587 struct dpif_linux_vport request;
588 struct dpif_linux_vport reply;
592 dpif_linux_vport_init(&request);
593 request.cmd = OVS_VPORT_CMD_GET;
594 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
595 request.port_no = port_no;
596 request.name = port_name;
598 error = dpif_linux_vport_transact(&request, &reply, &buf);
600 if (reply.dp_ifindex != request.dp_ifindex) {
601 /* A query by name reported that 'port_name' is in some datapath
602 * other than 'dpif', but the caller wants to know about 'dpif'. */
604 } else if (dpif_port) {
605 dpif_port->name = xstrdup(reply.name);
606 dpif_port->type = xstrdup(get_vport_type(&reply));
607 dpif_port->port_no = reply.port_no;
615 dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
616 struct dpif_port *dpif_port)
618 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
622 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
623 struct dpif_port *dpif_port)
625 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
629 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
635 dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_t port_no)
637 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
639 if (dpif->epoll_fd < 0) {
642 /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s
643 * channel, since it is not heavily loaded. */
644 int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no;
645 return nl_sock_pid(dpif->channels[idx].sock);
650 dpif_linux_flow_flush(struct dpif *dpif_)
652 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
653 struct dpif_linux_flow flow;
655 dpif_linux_flow_init(&flow);
656 flow.cmd = OVS_FLOW_CMD_DEL;
657 flow.dp_ifindex = dpif->dp_ifindex;
658 return dpif_linux_flow_transact(&flow, NULL, NULL);
661 struct dpif_linux_port_state {
666 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
668 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
669 struct dpif_linux_port_state *state;
670 struct dpif_linux_vport request;
673 *statep = state = xmalloc(sizeof *state);
675 dpif_linux_vport_init(&request);
676 request.cmd = OVS_DP_CMD_GET;
677 request.dp_ifindex = dpif->dp_ifindex;
679 buf = ofpbuf_new(1024);
680 dpif_linux_vport_to_ofpbuf(&request, buf);
681 nl_dump_start(&state->dump, genl_sock, buf);
688 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
689 struct dpif_port *dpif_port)
691 struct dpif_linux_port_state *state = state_;
692 struct dpif_linux_vport vport;
696 if (!nl_dump_next(&state->dump, &buf)) {
700 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
705 dpif_port->name = CONST_CAST(char *, vport.name);
706 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
707 dpif_port->port_no = vport.port_no;
712 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
714 struct dpif_linux_port_state *state = state_;
715 int error = nl_dump_done(&state->dump);
722 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
724 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
726 if (dpif->change_error) {
727 dpif->change_error = false;
728 sset_clear(&dpif->changed_ports);
730 } else if (!sset_is_empty(&dpif->changed_ports)) {
731 *devnamep = sset_pop(&dpif->changed_ports);
739 dpif_linux_port_poll_wait(const struct dpif *dpif_)
741 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
742 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
743 poll_immediate_wake();
748 dpif_linux_flow_get__(const struct dpif *dpif_,
749 const struct nlattr *key, size_t key_len,
750 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
752 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
753 struct dpif_linux_flow request;
755 dpif_linux_flow_init(&request);
756 request.cmd = OVS_FLOW_CMD_GET;
757 request.dp_ifindex = dpif->dp_ifindex;
759 request.key_len = key_len;
760 return dpif_linux_flow_transact(&request, reply, bufp);
764 dpif_linux_flow_get(const struct dpif *dpif_,
765 const struct nlattr *key, size_t key_len,
766 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
768 struct dpif_linux_flow reply;
772 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
775 dpif_linux_flow_get_stats(&reply, stats);
778 buf->data = CONST_CAST(struct nlattr *, reply.actions);
779 buf->size = reply.actions_len;
789 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
790 struct dpif_linux_flow *request)
792 static struct nlattr dummy_action;
794 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
796 dpif_linux_flow_init(request);
797 request->cmd = (put->flags & DPIF_FP_CREATE
798 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
799 request->dp_ifindex = dpif->dp_ifindex;
800 request->key = put->key;
801 request->key_len = put->key_len;
802 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
803 request->actions = put->actions ? put->actions : &dummy_action;
804 request->actions_len = put->actions_len;
805 if (put->flags & DPIF_FP_ZERO_STATS) {
806 request->clear = true;
808 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
812 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
814 struct dpif_linux_flow request, reply;
818 dpif_linux_init_flow_put(dpif_, put, &request);
819 error = dpif_linux_flow_transact(&request,
820 put->stats ? &reply : NULL,
821 put->stats ? &buf : NULL);
822 if (!error && put->stats) {
823 dpif_linux_flow_get_stats(&reply, put->stats);
830 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
831 struct dpif_linux_flow *request)
833 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
835 dpif_linux_flow_init(request);
836 request->cmd = OVS_FLOW_CMD_DEL;
837 request->dp_ifindex = dpif->dp_ifindex;
838 request->key = del->key;
839 request->key_len = del->key_len;
843 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
845 struct dpif_linux_flow request, reply;
849 dpif_linux_init_flow_del(dpif_, del, &request);
850 error = dpif_linux_flow_transact(&request,
851 del->stats ? &reply : NULL,
852 del->stats ? &buf : NULL);
853 if (!error && del->stats) {
854 dpif_linux_flow_get_stats(&reply, del->stats);
860 struct dpif_linux_flow_state {
862 struct dpif_linux_flow flow;
863 struct dpif_flow_stats stats;
868 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
870 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
871 struct dpif_linux_flow_state *state;
872 struct dpif_linux_flow request;
875 *statep = state = xmalloc(sizeof *state);
877 dpif_linux_flow_init(&request);
878 request.cmd = OVS_DP_CMD_GET;
879 request.dp_ifindex = dpif->dp_ifindex;
881 buf = ofpbuf_new(1024);
882 dpif_linux_flow_to_ofpbuf(&request, buf);
883 nl_dump_start(&state->dump, genl_sock, buf);
892 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
893 const struct nlattr **key, size_t *key_len,
894 const struct nlattr **actions, size_t *actions_len,
895 const struct dpif_flow_stats **stats)
897 struct dpif_linux_flow_state *state = state_;
902 ofpbuf_delete(state->buf);
905 if (!nl_dump_next(&state->dump, &buf)) {
909 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
914 if (actions && !state->flow.actions) {
915 error = dpif_linux_flow_get__(dpif_, state->flow.key,
917 &state->flow, &state->buf);
918 if (error == ENOENT) {
919 VLOG_DBG("dumped flow disappeared on get");
921 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
927 *actions = state->flow.actions;
928 *actions_len = state->flow.actions_len;
931 *key = state->flow.key;
932 *key_len = state->flow.key_len;
935 dpif_linux_flow_get_stats(&state->flow, &state->stats);
936 *stats = &state->stats;
942 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
944 struct dpif_linux_flow_state *state = state_;
945 int error = nl_dump_done(&state->dump);
946 ofpbuf_delete(state->buf);
952 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
955 struct ovs_header *k_exec;
957 ofpbuf_prealloc_tailroom(buf, (64
958 + d_exec->packet->size
960 + d_exec->actions_len));
962 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
963 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
965 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
966 k_exec->dp_ifindex = dp_ifindex;
968 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
969 d_exec->packet->data, d_exec->packet->size);
970 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
971 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
972 d_exec->actions, d_exec->actions_len);
976 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
978 uint64_t request_stub[1024 / 8];
979 struct ofpbuf request;
982 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
983 dpif_linux_encode_execute(dp_ifindex, execute, &request);
984 error = nl_sock_transact(genl_sock, &request, NULL);
985 ofpbuf_uninit(&request);
991 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
993 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
995 return dpif_linux_execute__(dpif->dp_ifindex, execute);
1001 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
1003 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1006 struct nl_transaction txn;
1008 struct ofpbuf request;
1009 uint64_t request_stub[1024 / 8];
1011 struct ofpbuf reply;
1012 uint64_t reply_stub[1024 / 8];
1015 struct nl_transaction *txnsp[MAX_OPS];
1018 ovs_assert(n_ops <= MAX_OPS);
1019 for (i = 0; i < n_ops; i++) {
1020 struct op_auxdata *aux = &auxes[i];
1021 struct dpif_op *op = ops[i];
1022 struct dpif_flow_put *put;
1023 struct dpif_flow_del *del;
1024 struct dpif_execute *execute;
1025 struct dpif_linux_flow flow;
1027 ofpbuf_use_stub(&aux->request,
1028 aux->request_stub, sizeof aux->request_stub);
1029 aux->txn.request = &aux->request;
1031 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1032 aux->txn.reply = NULL;
1035 case DPIF_OP_FLOW_PUT:
1036 put = &op->u.flow_put;
1037 dpif_linux_init_flow_put(dpif_, put, &flow);
1039 flow.nlmsg_flags |= NLM_F_ECHO;
1040 aux->txn.reply = &aux->reply;
1042 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1045 case DPIF_OP_FLOW_DEL:
1046 del = &op->u.flow_del;
1047 dpif_linux_init_flow_del(dpif_, del, &flow);
1049 flow.nlmsg_flags |= NLM_F_ECHO;
1050 aux->txn.reply = &aux->reply;
1052 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1055 case DPIF_OP_EXECUTE:
1056 execute = &op->u.execute;
1057 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1066 for (i = 0; i < n_ops; i++) {
1067 txnsp[i] = &auxes[i].txn;
1069 nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
1071 for (i = 0; i < n_ops; i++) {
1072 struct op_auxdata *aux = &auxes[i];
1073 struct nl_transaction *txn = &auxes[i].txn;
1074 struct dpif_op *op = ops[i];
1075 struct dpif_flow_put *put;
1076 struct dpif_flow_del *del;
1078 op->error = txn->error;
1081 case DPIF_OP_FLOW_PUT:
1082 put = &op->u.flow_put;
1085 struct dpif_linux_flow reply;
1087 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1090 dpif_linux_flow_get_stats(&reply, put->stats);
1095 memset(put->stats, 0, sizeof *put->stats);
1100 case DPIF_OP_FLOW_DEL:
1101 del = &op->u.flow_del;
1104 struct dpif_linux_flow reply;
1106 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1109 dpif_linux_flow_get_stats(&reply, del->stats);
1114 memset(del->stats, 0, sizeof *del->stats);
1119 case DPIF_OP_EXECUTE:
1126 ofpbuf_uninit(&aux->request);
1127 ofpbuf_uninit(&aux->reply);
1132 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1135 size_t chunk = MIN(n_ops, MAX_OPS);
1136 dpif_linux_operate__(dpif, ops, chunk);
1143 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1145 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1147 if ((dpif->epoll_fd >= 0) == enable) {
1152 destroy_channels(dpif);
1154 struct dpif_port_dump port_dump;
1155 struct dpif_port port;
1157 dpif->epoll_fd = epoll_create(10);
1158 if (dpif->epoll_fd < 0) {
1162 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
1163 struct dpif_linux_vport vport_request;
1164 struct nl_sock *sock;
1165 uint32_t upcall_pid;
1168 error = nl_sock_create(NETLINK_GENERIC, &sock);
1173 upcall_pid = nl_sock_pid(sock);
1175 dpif_linux_vport_init(&vport_request);
1176 vport_request.cmd = OVS_VPORT_CMD_SET;
1177 vport_request.dp_ifindex = dpif->dp_ifindex;
1178 vport_request.port_no = port.port_no;
1179 vport_request.upcall_pid = &upcall_pid;
1180 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1182 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1183 dpif_name(&dpif->dpif), vport_request.port_no,
1186 VLOG_WARN_RL(&error_rl,
1187 "%s: failed to set upcall pid on port: %s",
1188 dpif_name(&dpif->dpif), strerror(error));
1189 nl_sock_destroy(sock);
1191 if (error == ENODEV || error == ENOENT) {
1192 /* This device isn't there, but keep trying the others. */
1199 error = add_channel(dpif, port.port_no, sock);
1201 VLOG_INFO("%s: could not add channel for port %s",
1202 dpif_name(dpif_), port.name);
1203 nl_sock_destroy(sock);
1213 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1214 uint32_t queue_id, uint32_t *priority)
1216 if (queue_id < 0xf000) {
1217 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1225 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1228 static const struct nl_policy ovs_packet_policy[] = {
1229 /* Always present. */
1230 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1231 .min_len = ETH_HEADER_LEN },
1232 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1234 /* OVS_PACKET_CMD_ACTION only. */
1235 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
1238 struct ovs_header *ovs_header;
1239 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1240 struct nlmsghdr *nlmsg;
1241 struct genlmsghdr *genl;
1245 ofpbuf_use_const(&b, buf->data, buf->size);
1247 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1248 genl = ofpbuf_try_pull(&b, sizeof *genl);
1249 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1250 if (!nlmsg || !genl || !ovs_header
1251 || nlmsg->nlmsg_type != ovs_packet_family
1252 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1253 ARRAY_SIZE(ovs_packet_policy))) {
1257 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1258 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1264 memset(upcall, 0, sizeof *upcall);
1265 upcall->type = type;
1266 upcall->packet = buf;
1267 upcall->packet->data = CONST_CAST(struct nlattr *,
1268 nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
1269 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1270 upcall->key = CONST_CAST(struct nlattr *,
1271 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1272 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1273 upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
1274 *dp_ifindex = ovs_header->dp_ifindex;
1280 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1283 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1286 if (dpif->epoll_fd < 0) {
1290 if (dpif->event_offset >= dpif->n_events) {
1293 dpif->event_offset = dpif->n_events = 0;
1296 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1297 dpif->uc_array_size, 0);
1298 } while (retval < 0 && errno == EINTR);
1300 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1301 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
1302 } else if (retval > 0) {
1303 dpif->n_events = retval;
1307 while (dpif->event_offset < dpif->n_events) {
1308 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1309 struct dpif_channel *ch = &dpif->channels[idx];
1311 dpif->event_offset++;
1317 if (++read_tries > 50) {
1321 error = nl_sock_recv(ch->sock, buf, false);
1322 if (error == ENOBUFS) {
1323 /* ENOBUFS typically means that we've received so many
1324 * packets that the buffer overflowed. Try again
1325 * immediately because there's almost certainly a packet
1326 * waiting for us. */
1327 report_loss(dpif_, ch);
1331 ch->last_poll = time_msec();
1333 if (error == EAGAIN) {
1339 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1340 if (!error && dp_ifindex == dpif->dp_ifindex) {
1352 dpif_linux_recv_wait(struct dpif *dpif_)
1354 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1356 if (dpif->epoll_fd < 0) {
1360 poll_fd_wait(dpif->epoll_fd, POLLIN);
1364 dpif_linux_recv_purge(struct dpif *dpif_)
1366 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1367 struct dpif_channel *ch;
1369 if (dpif->epoll_fd < 0) {
1373 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
1375 nl_sock_drain(ch->sock);
1380 const struct dpif_class dpif_linux_class = {
1382 dpif_linux_enumerate,
1389 dpif_linux_get_stats,
1390 dpif_linux_port_add,
1391 dpif_linux_port_del,
1392 dpif_linux_port_query_by_number,
1393 dpif_linux_port_query_by_name,
1394 dpif_linux_get_max_ports,
1395 dpif_linux_port_get_pid,
1396 dpif_linux_port_dump_start,
1397 dpif_linux_port_dump_next,
1398 dpif_linux_port_dump_done,
1399 dpif_linux_port_poll,
1400 dpif_linux_port_poll_wait,
1401 dpif_linux_flow_get,
1402 dpif_linux_flow_put,
1403 dpif_linux_flow_del,
1404 dpif_linux_flow_flush,
1405 dpif_linux_flow_dump_start,
1406 dpif_linux_flow_dump_next,
1407 dpif_linux_flow_dump_done,
1410 dpif_linux_recv_set,
1411 dpif_linux_queue_to_priority,
1413 dpif_linux_recv_wait,
1414 dpif_linux_recv_purge,
1418 dpif_linux_init(void)
1420 static int error = -1;
1423 unsigned int ovs_vport_mcgroup;
1425 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1426 &ovs_datapath_family);
1428 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1429 "The Open vSwitch kernel module is probably not loaded.",
1430 OVS_DATAPATH_FAMILY);
1433 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1436 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1439 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1440 &ovs_packet_family);
1443 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1446 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1448 OVS_VPORT_MCGROUP_FALLBACK_ID);
1451 static struct dpif_linux_vport vport;
1452 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1453 dpif_linux_nln_parse, &vport);
1461 dpif_linux_is_internal_device(const char *name)
1463 struct dpif_linux_vport reply;
1467 error = dpif_linux_vport_get(name, &reply, &buf);
1470 } else if (error != ENODEV && error != ENOENT) {
1471 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1472 name, strerror(error));
1475 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1479 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1481 struct dpif_linux_vport *vport = vport_;
1482 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1486 dpif_linux_port_changed(const void *vport_, void *dpif_)
1488 const struct dpif_linux_vport *vport = vport_;
1489 struct dpif_linux *dpif = dpif_;
1492 if (vport->dp_ifindex == dpif->dp_ifindex
1493 && (vport->cmd == OVS_VPORT_CMD_NEW
1494 || vport->cmd == OVS_VPORT_CMD_DEL
1495 || vport->cmd == OVS_VPORT_CMD_SET)) {
1496 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1497 dpif->dpif.full_name, vport->name, vport->cmd);
1498 sset_add(&dpif->changed_ports, vport->name);
1501 dpif->change_error = true;
1505 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1506 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1507 * positive errno value.
1509 * 'vport' will contain pointers into 'buf', so the caller should not free
1510 * 'buf' while 'vport' is still in use. */
1512 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1513 const struct ofpbuf *buf)
1515 static const struct nl_policy ovs_vport_policy[] = {
1516 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1517 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1518 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1519 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1520 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1522 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1525 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1526 struct ovs_header *ovs_header;
1527 struct nlmsghdr *nlmsg;
1528 struct genlmsghdr *genl;
1531 dpif_linux_vport_init(vport);
1533 ofpbuf_use_const(&b, buf->data, buf->size);
1534 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1535 genl = ofpbuf_try_pull(&b, sizeof *genl);
1536 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1537 if (!nlmsg || !genl || !ovs_header
1538 || nlmsg->nlmsg_type != ovs_vport_family
1539 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1540 ARRAY_SIZE(ovs_vport_policy))) {
1544 vport->cmd = genl->cmd;
1545 vport->dp_ifindex = ovs_header->dp_ifindex;
1546 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1547 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1548 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1549 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1550 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1552 if (a[OVS_VPORT_ATTR_STATS]) {
1553 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1555 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1556 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1557 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1562 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1563 * followed by Netlink attributes corresponding to 'vport'. */
1565 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1568 struct ovs_header *ovs_header;
1570 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1571 vport->cmd, OVS_VPORT_VERSION);
1573 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1574 ovs_header->dp_ifindex = vport->dp_ifindex;
1576 if (vport->port_no != UINT32_MAX) {
1577 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1580 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1581 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1585 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1588 if (vport->upcall_pid) {
1589 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1593 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1594 vport->stats, sizeof *vport->stats);
1597 if (vport->options) {
1598 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1599 vport->options, vport->options_len);
1603 /* Clears 'vport' to "empty" values. */
1605 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1607 memset(vport, 0, sizeof *vport);
1608 vport->port_no = UINT32_MAX;
1611 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1612 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1613 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1614 * result of the command is expected to be an ovs_vport also, which is decoded
1615 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1616 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1618 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1619 struct dpif_linux_vport *reply,
1620 struct ofpbuf **bufp)
1622 struct ofpbuf *request_buf;
1625 ovs_assert((reply != NULL) == (bufp != NULL));
1627 error = dpif_linux_init();
1631 dpif_linux_vport_init(reply);
1636 request_buf = ofpbuf_new(1024);
1637 dpif_linux_vport_to_ofpbuf(request, request_buf);
1638 error = nl_sock_transact(genl_sock, request_buf, bufp);
1639 ofpbuf_delete(request_buf);
1643 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1646 dpif_linux_vport_init(reply);
1647 ofpbuf_delete(*bufp);
1654 /* Obtains information about the kernel vport named 'name' and stores it into
1655 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1656 * longer needed ('reply' will contain pointers into '*bufp'). */
1658 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1659 struct ofpbuf **bufp)
1661 struct dpif_linux_vport request;
1663 dpif_linux_vport_init(&request);
1664 request.cmd = OVS_VPORT_CMD_GET;
1665 request.name = name;
1667 return dpif_linux_vport_transact(&request, reply, bufp);
1670 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1671 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1672 * positive errno value.
1674 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1675 * while 'dp' is still in use. */
1677 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1679 static const struct nl_policy ovs_datapath_policy[] = {
1680 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1681 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1685 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1686 struct ovs_header *ovs_header;
1687 struct nlmsghdr *nlmsg;
1688 struct genlmsghdr *genl;
1691 dpif_linux_dp_init(dp);
1693 ofpbuf_use_const(&b, buf->data, buf->size);
1694 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1695 genl = ofpbuf_try_pull(&b, sizeof *genl);
1696 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1697 if (!nlmsg || !genl || !ovs_header
1698 || nlmsg->nlmsg_type != ovs_datapath_family
1699 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1700 ARRAY_SIZE(ovs_datapath_policy))) {
1704 dp->cmd = genl->cmd;
1705 dp->dp_ifindex = ovs_header->dp_ifindex;
1706 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1707 if (a[OVS_DP_ATTR_STATS]) {
1708 /* Can't use structure assignment because Netlink doesn't ensure
1709 * sufficient alignment for 64-bit members. */
1710 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1717 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1719 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1721 struct ovs_header *ovs_header;
1723 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1724 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1725 OVS_DATAPATH_VERSION);
1727 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1728 ovs_header->dp_ifindex = dp->dp_ifindex;
1731 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1734 if (dp->upcall_pid) {
1735 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1738 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1741 /* Clears 'dp' to "empty" values. */
1743 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1745 memset(dp, 0, sizeof *dp);
1749 dpif_linux_dp_dump_start(struct nl_dump *dump)
1751 struct dpif_linux_dp request;
1754 dpif_linux_dp_init(&request);
1755 request.cmd = OVS_DP_CMD_GET;
1757 buf = ofpbuf_new(1024);
1758 dpif_linux_dp_to_ofpbuf(&request, buf);
1759 nl_dump_start(dump, genl_sock, buf);
1763 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1764 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1765 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1766 * result of the command is expected to be of the same form, which is decoded
1767 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1768 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1770 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1771 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1773 struct ofpbuf *request_buf;
1776 ovs_assert((reply != NULL) == (bufp != NULL));
1778 request_buf = ofpbuf_new(1024);
1779 dpif_linux_dp_to_ofpbuf(request, request_buf);
1780 error = nl_sock_transact(genl_sock, request_buf, bufp);
1781 ofpbuf_delete(request_buf);
1785 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1788 dpif_linux_dp_init(reply);
1789 ofpbuf_delete(*bufp);
1796 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1797 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1798 * will contain pointers into '*bufp'). */
1800 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1801 struct ofpbuf **bufp)
1803 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1804 struct dpif_linux_dp request;
1806 dpif_linux_dp_init(&request);
1807 request.cmd = OVS_DP_CMD_GET;
1808 request.dp_ifindex = dpif->dp_ifindex;
1810 return dpif_linux_dp_transact(&request, reply, bufp);
1813 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1814 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1815 * positive errno value.
1817 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1818 * while 'flow' is still in use. */
1820 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1821 const struct ofpbuf *buf)
1823 static const struct nl_policy ovs_flow_policy[] = {
1824 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1825 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1826 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1828 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1829 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1830 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1833 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1834 struct ovs_header *ovs_header;
1835 struct nlmsghdr *nlmsg;
1836 struct genlmsghdr *genl;
1839 dpif_linux_flow_init(flow);
1841 ofpbuf_use_const(&b, buf->data, buf->size);
1842 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1843 genl = ofpbuf_try_pull(&b, sizeof *genl);
1844 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1845 if (!nlmsg || !genl || !ovs_header
1846 || nlmsg->nlmsg_type != ovs_flow_family
1847 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1848 ARRAY_SIZE(ovs_flow_policy))) {
1852 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1853 flow->dp_ifindex = ovs_header->dp_ifindex;
1854 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1855 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1856 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1857 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1858 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1860 if (a[OVS_FLOW_ATTR_STATS]) {
1861 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1863 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1864 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1866 if (a[OVS_FLOW_ATTR_USED]) {
1867 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1872 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1873 * followed by Netlink attributes corresponding to 'flow'. */
1875 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1878 struct ovs_header *ovs_header;
1880 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1881 NLM_F_REQUEST | flow->nlmsg_flags,
1882 flow->cmd, OVS_FLOW_VERSION);
1884 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1885 ovs_header->dp_ifindex = flow->dp_ifindex;
1887 if (flow->key_len) {
1888 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1891 if (flow->actions || flow->actions_len) {
1892 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1893 flow->actions, flow->actions_len);
1896 /* We never need to send these to the kernel. */
1897 ovs_assert(!flow->stats);
1898 ovs_assert(!flow->tcp_flags);
1899 ovs_assert(!flow->used);
1902 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1906 /* Clears 'flow' to "empty" values. */
1908 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1910 memset(flow, 0, sizeof *flow);
1913 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1914 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1915 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1916 * result of the command is expected to be a flow also, which is decoded and
1917 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1918 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1920 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1921 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1923 struct ofpbuf *request_buf;
1926 ovs_assert((reply != NULL) == (bufp != NULL));
1929 request->nlmsg_flags |= NLM_F_ECHO;
1932 request_buf = ofpbuf_new(1024);
1933 dpif_linux_flow_to_ofpbuf(request, request_buf);
1934 error = nl_sock_transact(genl_sock, request_buf, bufp);
1935 ofpbuf_delete(request_buf);
1939 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1942 dpif_linux_flow_init(reply);
1943 ofpbuf_delete(*bufp);
1951 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1952 struct dpif_flow_stats *stats)
1955 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1956 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1958 stats->n_packets = 0;
1961 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1962 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1965 /* Logs information about a packet that was recently lost in 'ch' (in
1968 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
1970 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1971 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1974 if (VLOG_DROP_WARN(&rl)) {
1979 if (ch->last_poll != LLONG_MIN) {
1980 ds_put_format(&s, " (last polled %lld ms ago)",
1981 time_msec() - ch->last_poll);
1984 VLOG_WARN("%s: lost packet on channel %td%s",
1985 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));