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);
458 static enum ovs_vport_type
459 netdev_to_ovs_vport_type(const struct netdev *netdev)
461 const char *type = netdev_get_type(netdev);
463 if (!strcmp(type, "tap") || !strcmp(type, "system")) {
464 return OVS_VPORT_TYPE_NETDEV;
465 } else if (!strcmp(type, "internal")) {
466 return OVS_VPORT_TYPE_INTERNAL;
467 } else if (strstr(type, "gre64")) {
468 return OVS_VPORT_TYPE_GRE64;
469 } else if (strstr(type, "gre")) {
470 return OVS_VPORT_TYPE_GRE;
471 } else if (!strcmp(type, "capwap")) {
472 return OVS_VPORT_TYPE_CAPWAP;
473 } else if (!strcmp(type, "vxlan")) {
474 return OVS_VPORT_TYPE_VXLAN;
476 return OVS_VPORT_TYPE_UNSPEC;
481 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
484 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
485 const struct netdev_tunnel_config *tnl_cfg;
486 const char *name = netdev_vport_get_dpif_port(netdev);
487 const char *type = netdev_get_type(netdev);
488 struct dpif_linux_vport request, reply;
489 struct nl_sock *sock = NULL;
492 uint64_t options_stub[64 / 8];
493 struct ofpbuf options;
496 if (dpif->epoll_fd >= 0) {
497 error = nl_sock_create(NETLINK_GENERIC, &sock);
503 dpif_linux_vport_init(&request);
504 request.cmd = OVS_VPORT_CMD_NEW;
505 request.dp_ifindex = dpif->dp_ifindex;
506 request.type = netdev_to_ovs_vport_type(netdev);
507 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
508 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
509 "unsupported type `%s'",
510 dpif_name(dpif_), name, type);
511 nl_sock_destroy(sock);
516 if (request.type == OVS_VPORT_TYPE_NETDEV) {
517 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
520 tnl_cfg = netdev_get_tunnel_config(netdev);
521 if (tnl_cfg && tnl_cfg->dst_port != 0) {
522 ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
523 nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
524 ntohs(tnl_cfg->dst_port));
525 request.options = options.data;
526 request.options_len = options.size;
529 request.port_no = *port_nop;
530 upcall_pid = sock ? nl_sock_pid(sock) : 0;
531 request.upcall_pid = &upcall_pid;
533 error = dpif_linux_vport_transact(&request, &reply, &buf);
535 *port_nop = reply.port_no;
536 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
537 dpif_name(dpif_), reply.port_no, upcall_pid);
539 if (error == EBUSY && *port_nop != UINT32_MAX) {
540 VLOG_INFO("%s: requested port %"PRIu32" is in use",
541 dpif_name(dpif_), *port_nop);
543 nl_sock_destroy(sock);
550 error = add_channel(dpif, *port_nop, sock);
552 VLOG_INFO("%s: could not add channel for port %s",
553 dpif_name(dpif_), name);
555 /* Delete the port. */
556 dpif_linux_vport_init(&request);
557 request.cmd = OVS_VPORT_CMD_DEL;
558 request.dp_ifindex = dpif->dp_ifindex;
559 request.port_no = *port_nop;
560 dpif_linux_vport_transact(&request, NULL, NULL);
562 nl_sock_destroy(sock);
571 dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no)
573 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
574 struct dpif_linux_vport vport;
577 dpif_linux_vport_init(&vport);
578 vport.cmd = OVS_VPORT_CMD_DEL;
579 vport.dp_ifindex = dpif->dp_ifindex;
580 vport.port_no = port_no;
581 error = dpif_linux_vport_transact(&vport, NULL, NULL);
583 del_channel(dpif, port_no);
589 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
590 const char *port_name, struct dpif_port *dpif_port)
592 struct dpif_linux_vport request;
593 struct dpif_linux_vport reply;
597 dpif_linux_vport_init(&request);
598 request.cmd = OVS_VPORT_CMD_GET;
599 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
600 request.port_no = port_no;
601 request.name = port_name;
603 error = dpif_linux_vport_transact(&request, &reply, &buf);
605 if (reply.dp_ifindex != request.dp_ifindex) {
606 /* A query by name reported that 'port_name' is in some datapath
607 * other than 'dpif', but the caller wants to know about 'dpif'. */
609 } else if (dpif_port) {
610 dpif_port->name = xstrdup(reply.name);
611 dpif_port->type = xstrdup(get_vport_type(&reply));
612 dpif_port->port_no = reply.port_no;
620 dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
621 struct dpif_port *dpif_port)
623 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
627 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
628 struct dpif_port *dpif_port)
630 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
634 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
640 dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_t port_no)
642 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
644 if (dpif->epoll_fd < 0) {
647 /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s
648 * channel, since it is not heavily loaded. */
649 int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no;
650 return nl_sock_pid(dpif->channels[idx].sock);
655 dpif_linux_flow_flush(struct dpif *dpif_)
657 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
658 struct dpif_linux_flow flow;
660 dpif_linux_flow_init(&flow);
661 flow.cmd = OVS_FLOW_CMD_DEL;
662 flow.dp_ifindex = dpif->dp_ifindex;
663 return dpif_linux_flow_transact(&flow, NULL, NULL);
666 struct dpif_linux_port_state {
671 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
673 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
674 struct dpif_linux_port_state *state;
675 struct dpif_linux_vport request;
678 *statep = state = xmalloc(sizeof *state);
680 dpif_linux_vport_init(&request);
681 request.cmd = OVS_DP_CMD_GET;
682 request.dp_ifindex = dpif->dp_ifindex;
684 buf = ofpbuf_new(1024);
685 dpif_linux_vport_to_ofpbuf(&request, buf);
686 nl_dump_start(&state->dump, genl_sock, buf);
693 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
694 struct dpif_port *dpif_port)
696 struct dpif_linux_port_state *state = state_;
697 struct dpif_linux_vport vport;
701 if (!nl_dump_next(&state->dump, &buf)) {
705 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
710 dpif_port->name = CONST_CAST(char *, vport.name);
711 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
712 dpif_port->port_no = vport.port_no;
717 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
719 struct dpif_linux_port_state *state = state_;
720 int error = nl_dump_done(&state->dump);
727 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
729 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
731 if (dpif->change_error) {
732 dpif->change_error = false;
733 sset_clear(&dpif->changed_ports);
735 } else if (!sset_is_empty(&dpif->changed_ports)) {
736 *devnamep = sset_pop(&dpif->changed_ports);
744 dpif_linux_port_poll_wait(const struct dpif *dpif_)
746 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
747 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
748 poll_immediate_wake();
753 dpif_linux_flow_get__(const struct dpif *dpif_,
754 const struct nlattr *key, size_t key_len,
755 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
757 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
758 struct dpif_linux_flow request;
760 dpif_linux_flow_init(&request);
761 request.cmd = OVS_FLOW_CMD_GET;
762 request.dp_ifindex = dpif->dp_ifindex;
764 request.key_len = key_len;
765 return dpif_linux_flow_transact(&request, reply, bufp);
769 dpif_linux_flow_get(const struct dpif *dpif_,
770 const struct nlattr *key, size_t key_len,
771 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
773 struct dpif_linux_flow reply;
777 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
780 dpif_linux_flow_get_stats(&reply, stats);
783 buf->data = CONST_CAST(struct nlattr *, reply.actions);
784 buf->size = reply.actions_len;
794 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
795 struct dpif_linux_flow *request)
797 static struct nlattr dummy_action;
799 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
801 dpif_linux_flow_init(request);
802 request->cmd = (put->flags & DPIF_FP_CREATE
803 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
804 request->dp_ifindex = dpif->dp_ifindex;
805 request->key = put->key;
806 request->key_len = put->key_len;
807 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
808 request->actions = put->actions ? put->actions : &dummy_action;
809 request->actions_len = put->actions_len;
810 if (put->flags & DPIF_FP_ZERO_STATS) {
811 request->clear = true;
813 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
817 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
819 struct dpif_linux_flow request, reply;
823 dpif_linux_init_flow_put(dpif_, put, &request);
824 error = dpif_linux_flow_transact(&request,
825 put->stats ? &reply : NULL,
826 put->stats ? &buf : NULL);
827 if (!error && put->stats) {
828 dpif_linux_flow_get_stats(&reply, put->stats);
835 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
836 struct dpif_linux_flow *request)
838 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
840 dpif_linux_flow_init(request);
841 request->cmd = OVS_FLOW_CMD_DEL;
842 request->dp_ifindex = dpif->dp_ifindex;
843 request->key = del->key;
844 request->key_len = del->key_len;
848 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
850 struct dpif_linux_flow request, reply;
854 dpif_linux_init_flow_del(dpif_, del, &request);
855 error = dpif_linux_flow_transact(&request,
856 del->stats ? &reply : NULL,
857 del->stats ? &buf : NULL);
858 if (!error && del->stats) {
859 dpif_linux_flow_get_stats(&reply, del->stats);
865 struct dpif_linux_flow_state {
867 struct dpif_linux_flow flow;
868 struct dpif_flow_stats stats;
873 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
875 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
876 struct dpif_linux_flow_state *state;
877 struct dpif_linux_flow request;
880 *statep = state = xmalloc(sizeof *state);
882 dpif_linux_flow_init(&request);
883 request.cmd = OVS_DP_CMD_GET;
884 request.dp_ifindex = dpif->dp_ifindex;
886 buf = ofpbuf_new(1024);
887 dpif_linux_flow_to_ofpbuf(&request, buf);
888 nl_dump_start(&state->dump, genl_sock, buf);
897 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
898 const struct nlattr **key, size_t *key_len,
899 const struct nlattr **actions, size_t *actions_len,
900 const struct dpif_flow_stats **stats)
902 struct dpif_linux_flow_state *state = state_;
907 ofpbuf_delete(state->buf);
910 if (!nl_dump_next(&state->dump, &buf)) {
914 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
919 if (actions && !state->flow.actions) {
920 error = dpif_linux_flow_get__(dpif_, state->flow.key,
922 &state->flow, &state->buf);
923 if (error == ENOENT) {
924 VLOG_DBG("dumped flow disappeared on get");
926 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
932 *actions = state->flow.actions;
933 *actions_len = state->flow.actions_len;
936 *key = state->flow.key;
937 *key_len = state->flow.key_len;
940 dpif_linux_flow_get_stats(&state->flow, &state->stats);
941 *stats = &state->stats;
947 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
949 struct dpif_linux_flow_state *state = state_;
950 int error = nl_dump_done(&state->dump);
951 ofpbuf_delete(state->buf);
957 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
960 struct ovs_header *k_exec;
962 ofpbuf_prealloc_tailroom(buf, (64
963 + d_exec->packet->size
965 + d_exec->actions_len));
967 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
968 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
970 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
971 k_exec->dp_ifindex = dp_ifindex;
973 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
974 d_exec->packet->data, d_exec->packet->size);
975 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
976 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
977 d_exec->actions, d_exec->actions_len);
981 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
983 uint64_t request_stub[1024 / 8];
984 struct ofpbuf request;
987 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
988 dpif_linux_encode_execute(dp_ifindex, execute, &request);
989 error = nl_sock_transact(genl_sock, &request, NULL);
990 ofpbuf_uninit(&request);
996 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
998 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1000 return dpif_linux_execute__(dpif->dp_ifindex, execute);
1006 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
1008 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1011 struct nl_transaction txn;
1013 struct ofpbuf request;
1014 uint64_t request_stub[1024 / 8];
1016 struct ofpbuf reply;
1017 uint64_t reply_stub[1024 / 8];
1020 struct nl_transaction *txnsp[MAX_OPS];
1023 ovs_assert(n_ops <= MAX_OPS);
1024 for (i = 0; i < n_ops; i++) {
1025 struct op_auxdata *aux = &auxes[i];
1026 struct dpif_op *op = ops[i];
1027 struct dpif_flow_put *put;
1028 struct dpif_flow_del *del;
1029 struct dpif_execute *execute;
1030 struct dpif_linux_flow flow;
1032 ofpbuf_use_stub(&aux->request,
1033 aux->request_stub, sizeof aux->request_stub);
1034 aux->txn.request = &aux->request;
1036 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1037 aux->txn.reply = NULL;
1040 case DPIF_OP_FLOW_PUT:
1041 put = &op->u.flow_put;
1042 dpif_linux_init_flow_put(dpif_, put, &flow);
1044 flow.nlmsg_flags |= NLM_F_ECHO;
1045 aux->txn.reply = &aux->reply;
1047 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1050 case DPIF_OP_FLOW_DEL:
1051 del = &op->u.flow_del;
1052 dpif_linux_init_flow_del(dpif_, del, &flow);
1054 flow.nlmsg_flags |= NLM_F_ECHO;
1055 aux->txn.reply = &aux->reply;
1057 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1060 case DPIF_OP_EXECUTE:
1061 execute = &op->u.execute;
1062 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1071 for (i = 0; i < n_ops; i++) {
1072 txnsp[i] = &auxes[i].txn;
1074 nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
1076 for (i = 0; i < n_ops; i++) {
1077 struct op_auxdata *aux = &auxes[i];
1078 struct nl_transaction *txn = &auxes[i].txn;
1079 struct dpif_op *op = ops[i];
1080 struct dpif_flow_put *put;
1081 struct dpif_flow_del *del;
1083 op->error = txn->error;
1086 case DPIF_OP_FLOW_PUT:
1087 put = &op->u.flow_put;
1090 struct dpif_linux_flow reply;
1092 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1095 dpif_linux_flow_get_stats(&reply, put->stats);
1100 memset(put->stats, 0, sizeof *put->stats);
1105 case DPIF_OP_FLOW_DEL:
1106 del = &op->u.flow_del;
1109 struct dpif_linux_flow reply;
1111 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1114 dpif_linux_flow_get_stats(&reply, del->stats);
1119 memset(del->stats, 0, sizeof *del->stats);
1124 case DPIF_OP_EXECUTE:
1131 ofpbuf_uninit(&aux->request);
1132 ofpbuf_uninit(&aux->reply);
1137 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1140 size_t chunk = MIN(n_ops, MAX_OPS);
1141 dpif_linux_operate__(dpif, ops, chunk);
1148 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1150 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1152 if ((dpif->epoll_fd >= 0) == enable) {
1157 destroy_channels(dpif);
1159 struct dpif_port_dump port_dump;
1160 struct dpif_port port;
1162 dpif->epoll_fd = epoll_create(10);
1163 if (dpif->epoll_fd < 0) {
1167 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
1168 struct dpif_linux_vport vport_request;
1169 struct nl_sock *sock;
1170 uint32_t upcall_pid;
1173 error = nl_sock_create(NETLINK_GENERIC, &sock);
1178 upcall_pid = nl_sock_pid(sock);
1180 dpif_linux_vport_init(&vport_request);
1181 vport_request.cmd = OVS_VPORT_CMD_SET;
1182 vport_request.dp_ifindex = dpif->dp_ifindex;
1183 vport_request.port_no = port.port_no;
1184 vport_request.upcall_pid = &upcall_pid;
1185 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1187 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1188 dpif_name(&dpif->dpif), vport_request.port_no,
1191 VLOG_WARN_RL(&error_rl,
1192 "%s: failed to set upcall pid on port: %s",
1193 dpif_name(&dpif->dpif), strerror(error));
1194 nl_sock_destroy(sock);
1196 if (error == ENODEV || error == ENOENT) {
1197 /* This device isn't there, but keep trying the others. */
1204 error = add_channel(dpif, port.port_no, sock);
1206 VLOG_INFO("%s: could not add channel for port %s",
1207 dpif_name(dpif_), port.name);
1208 nl_sock_destroy(sock);
1218 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1219 uint32_t queue_id, uint32_t *priority)
1221 if (queue_id < 0xf000) {
1222 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1230 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1233 static const struct nl_policy ovs_packet_policy[] = {
1234 /* Always present. */
1235 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1236 .min_len = ETH_HEADER_LEN },
1237 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1239 /* OVS_PACKET_CMD_ACTION only. */
1240 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
1243 struct ovs_header *ovs_header;
1244 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1245 struct nlmsghdr *nlmsg;
1246 struct genlmsghdr *genl;
1250 ofpbuf_use_const(&b, buf->data, buf->size);
1252 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1253 genl = ofpbuf_try_pull(&b, sizeof *genl);
1254 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1255 if (!nlmsg || !genl || !ovs_header
1256 || nlmsg->nlmsg_type != ovs_packet_family
1257 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1258 ARRAY_SIZE(ovs_packet_policy))) {
1262 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1263 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1269 memset(upcall, 0, sizeof *upcall);
1270 upcall->type = type;
1271 upcall->packet = buf;
1272 upcall->packet->data = CONST_CAST(struct nlattr *,
1273 nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
1274 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1275 upcall->key = CONST_CAST(struct nlattr *,
1276 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1277 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1278 upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
1279 *dp_ifindex = ovs_header->dp_ifindex;
1285 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1288 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1291 if (dpif->epoll_fd < 0) {
1295 if (dpif->event_offset >= dpif->n_events) {
1298 dpif->event_offset = dpif->n_events = 0;
1301 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1302 dpif->uc_array_size, 0);
1303 } while (retval < 0 && errno == EINTR);
1305 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1306 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
1307 } else if (retval > 0) {
1308 dpif->n_events = retval;
1312 while (dpif->event_offset < dpif->n_events) {
1313 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1314 struct dpif_channel *ch = &dpif->channels[idx];
1316 dpif->event_offset++;
1322 if (++read_tries > 50) {
1326 error = nl_sock_recv(ch->sock, buf, false);
1327 if (error == ENOBUFS) {
1328 /* ENOBUFS typically means that we've received so many
1329 * packets that the buffer overflowed. Try again
1330 * immediately because there's almost certainly a packet
1331 * waiting for us. */
1332 report_loss(dpif_, ch);
1336 ch->last_poll = time_msec();
1338 if (error == EAGAIN) {
1344 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1345 if (!error && dp_ifindex == dpif->dp_ifindex) {
1357 dpif_linux_recv_wait(struct dpif *dpif_)
1359 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1361 if (dpif->epoll_fd < 0) {
1365 poll_fd_wait(dpif->epoll_fd, POLLIN);
1369 dpif_linux_recv_purge(struct dpif *dpif_)
1371 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1372 struct dpif_channel *ch;
1374 if (dpif->epoll_fd < 0) {
1378 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
1380 nl_sock_drain(ch->sock);
1385 const struct dpif_class dpif_linux_class = {
1387 dpif_linux_enumerate,
1394 dpif_linux_get_stats,
1395 dpif_linux_port_add,
1396 dpif_linux_port_del,
1397 dpif_linux_port_query_by_number,
1398 dpif_linux_port_query_by_name,
1399 dpif_linux_get_max_ports,
1400 dpif_linux_port_get_pid,
1401 dpif_linux_port_dump_start,
1402 dpif_linux_port_dump_next,
1403 dpif_linux_port_dump_done,
1404 dpif_linux_port_poll,
1405 dpif_linux_port_poll_wait,
1406 dpif_linux_flow_get,
1407 dpif_linux_flow_put,
1408 dpif_linux_flow_del,
1409 dpif_linux_flow_flush,
1410 dpif_linux_flow_dump_start,
1411 dpif_linux_flow_dump_next,
1412 dpif_linux_flow_dump_done,
1415 dpif_linux_recv_set,
1416 dpif_linux_queue_to_priority,
1418 dpif_linux_recv_wait,
1419 dpif_linux_recv_purge,
1423 dpif_linux_init(void)
1425 static int error = -1;
1428 unsigned int ovs_vport_mcgroup;
1430 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1431 &ovs_datapath_family);
1433 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1434 "The Open vSwitch kernel module is probably not loaded.",
1435 OVS_DATAPATH_FAMILY);
1438 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1441 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1444 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1445 &ovs_packet_family);
1448 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1451 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1453 OVS_VPORT_MCGROUP_FALLBACK_ID);
1456 static struct dpif_linux_vport vport;
1457 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1458 dpif_linux_nln_parse, &vport);
1466 dpif_linux_is_internal_device(const char *name)
1468 struct dpif_linux_vport reply;
1472 error = dpif_linux_vport_get(name, &reply, &buf);
1475 } else if (error != ENODEV && error != ENOENT) {
1476 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1477 name, strerror(error));
1480 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1484 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1486 struct dpif_linux_vport *vport = vport_;
1487 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1491 dpif_linux_port_changed(const void *vport_, void *dpif_)
1493 const struct dpif_linux_vport *vport = vport_;
1494 struct dpif_linux *dpif = dpif_;
1497 if (vport->dp_ifindex == dpif->dp_ifindex
1498 && (vport->cmd == OVS_VPORT_CMD_NEW
1499 || vport->cmd == OVS_VPORT_CMD_DEL
1500 || vport->cmd == OVS_VPORT_CMD_SET)) {
1501 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1502 dpif->dpif.full_name, vport->name, vport->cmd);
1503 sset_add(&dpif->changed_ports, vport->name);
1506 dpif->change_error = true;
1510 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1511 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1512 * positive errno value.
1514 * 'vport' will contain pointers into 'buf', so the caller should not free
1515 * 'buf' while 'vport' is still in use. */
1517 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1518 const struct ofpbuf *buf)
1520 static const struct nl_policy ovs_vport_policy[] = {
1521 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1522 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1523 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1524 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1525 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1527 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1530 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1531 struct ovs_header *ovs_header;
1532 struct nlmsghdr *nlmsg;
1533 struct genlmsghdr *genl;
1536 dpif_linux_vport_init(vport);
1538 ofpbuf_use_const(&b, buf->data, buf->size);
1539 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1540 genl = ofpbuf_try_pull(&b, sizeof *genl);
1541 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1542 if (!nlmsg || !genl || !ovs_header
1543 || nlmsg->nlmsg_type != ovs_vport_family
1544 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1545 ARRAY_SIZE(ovs_vport_policy))) {
1549 vport->cmd = genl->cmd;
1550 vport->dp_ifindex = ovs_header->dp_ifindex;
1551 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1552 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1553 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1554 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1555 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1557 if (a[OVS_VPORT_ATTR_STATS]) {
1558 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1560 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1561 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1562 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1567 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1568 * followed by Netlink attributes corresponding to 'vport'. */
1570 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1573 struct ovs_header *ovs_header;
1575 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1576 vport->cmd, OVS_VPORT_VERSION);
1578 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1579 ovs_header->dp_ifindex = vport->dp_ifindex;
1581 if (vport->port_no != UINT32_MAX) {
1582 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1585 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1586 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1590 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1593 if (vport->upcall_pid) {
1594 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1598 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1599 vport->stats, sizeof *vport->stats);
1602 if (vport->options) {
1603 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1604 vport->options, vport->options_len);
1608 /* Clears 'vport' to "empty" values. */
1610 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1612 memset(vport, 0, sizeof *vport);
1613 vport->port_no = UINT32_MAX;
1616 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1617 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1618 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1619 * result of the command is expected to be an ovs_vport also, which is decoded
1620 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1621 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1623 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1624 struct dpif_linux_vport *reply,
1625 struct ofpbuf **bufp)
1627 struct ofpbuf *request_buf;
1630 ovs_assert((reply != NULL) == (bufp != NULL));
1632 error = dpif_linux_init();
1636 dpif_linux_vport_init(reply);
1641 request_buf = ofpbuf_new(1024);
1642 dpif_linux_vport_to_ofpbuf(request, request_buf);
1643 error = nl_sock_transact(genl_sock, request_buf, bufp);
1644 ofpbuf_delete(request_buf);
1648 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1651 dpif_linux_vport_init(reply);
1652 ofpbuf_delete(*bufp);
1659 /* Obtains information about the kernel vport named 'name' and stores it into
1660 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1661 * longer needed ('reply' will contain pointers into '*bufp'). */
1663 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1664 struct ofpbuf **bufp)
1666 struct dpif_linux_vport request;
1668 dpif_linux_vport_init(&request);
1669 request.cmd = OVS_VPORT_CMD_GET;
1670 request.name = name;
1672 return dpif_linux_vport_transact(&request, reply, bufp);
1675 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1676 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1677 * positive errno value.
1679 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1680 * while 'dp' is still in use. */
1682 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1684 static const struct nl_policy ovs_datapath_policy[] = {
1685 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1686 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1690 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1691 struct ovs_header *ovs_header;
1692 struct nlmsghdr *nlmsg;
1693 struct genlmsghdr *genl;
1696 dpif_linux_dp_init(dp);
1698 ofpbuf_use_const(&b, buf->data, buf->size);
1699 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1700 genl = ofpbuf_try_pull(&b, sizeof *genl);
1701 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1702 if (!nlmsg || !genl || !ovs_header
1703 || nlmsg->nlmsg_type != ovs_datapath_family
1704 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1705 ARRAY_SIZE(ovs_datapath_policy))) {
1709 dp->cmd = genl->cmd;
1710 dp->dp_ifindex = ovs_header->dp_ifindex;
1711 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1712 if (a[OVS_DP_ATTR_STATS]) {
1713 /* Can't use structure assignment because Netlink doesn't ensure
1714 * sufficient alignment for 64-bit members. */
1715 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1722 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1724 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1726 struct ovs_header *ovs_header;
1728 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1729 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1730 OVS_DATAPATH_VERSION);
1732 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1733 ovs_header->dp_ifindex = dp->dp_ifindex;
1736 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1739 if (dp->upcall_pid) {
1740 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1743 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1746 /* Clears 'dp' to "empty" values. */
1748 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1750 memset(dp, 0, sizeof *dp);
1754 dpif_linux_dp_dump_start(struct nl_dump *dump)
1756 struct dpif_linux_dp request;
1759 dpif_linux_dp_init(&request);
1760 request.cmd = OVS_DP_CMD_GET;
1762 buf = ofpbuf_new(1024);
1763 dpif_linux_dp_to_ofpbuf(&request, buf);
1764 nl_dump_start(dump, genl_sock, buf);
1768 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1769 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1770 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1771 * result of the command is expected to be of the same form, which is decoded
1772 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1773 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1775 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1776 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1778 struct ofpbuf *request_buf;
1781 ovs_assert((reply != NULL) == (bufp != NULL));
1783 request_buf = ofpbuf_new(1024);
1784 dpif_linux_dp_to_ofpbuf(request, request_buf);
1785 error = nl_sock_transact(genl_sock, request_buf, bufp);
1786 ofpbuf_delete(request_buf);
1790 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1793 dpif_linux_dp_init(reply);
1794 ofpbuf_delete(*bufp);
1801 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1802 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1803 * will contain pointers into '*bufp'). */
1805 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1806 struct ofpbuf **bufp)
1808 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1809 struct dpif_linux_dp request;
1811 dpif_linux_dp_init(&request);
1812 request.cmd = OVS_DP_CMD_GET;
1813 request.dp_ifindex = dpif->dp_ifindex;
1815 return dpif_linux_dp_transact(&request, reply, bufp);
1818 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1819 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1820 * positive errno value.
1822 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1823 * while 'flow' is still in use. */
1825 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1826 const struct ofpbuf *buf)
1828 static const struct nl_policy ovs_flow_policy[] = {
1829 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1830 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1831 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1833 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1834 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1835 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1838 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1839 struct ovs_header *ovs_header;
1840 struct nlmsghdr *nlmsg;
1841 struct genlmsghdr *genl;
1844 dpif_linux_flow_init(flow);
1846 ofpbuf_use_const(&b, buf->data, buf->size);
1847 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1848 genl = ofpbuf_try_pull(&b, sizeof *genl);
1849 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1850 if (!nlmsg || !genl || !ovs_header
1851 || nlmsg->nlmsg_type != ovs_flow_family
1852 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1853 ARRAY_SIZE(ovs_flow_policy))) {
1857 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1858 flow->dp_ifindex = ovs_header->dp_ifindex;
1859 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1860 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1861 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1862 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1863 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1865 if (a[OVS_FLOW_ATTR_STATS]) {
1866 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1868 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1869 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1871 if (a[OVS_FLOW_ATTR_USED]) {
1872 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1877 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1878 * followed by Netlink attributes corresponding to 'flow'. */
1880 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1883 struct ovs_header *ovs_header;
1885 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1886 NLM_F_REQUEST | flow->nlmsg_flags,
1887 flow->cmd, OVS_FLOW_VERSION);
1889 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1890 ovs_header->dp_ifindex = flow->dp_ifindex;
1892 if (flow->key_len) {
1893 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1896 if (flow->actions || flow->actions_len) {
1897 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1898 flow->actions, flow->actions_len);
1901 /* We never need to send these to the kernel. */
1902 ovs_assert(!flow->stats);
1903 ovs_assert(!flow->tcp_flags);
1904 ovs_assert(!flow->used);
1907 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1911 /* Clears 'flow' to "empty" values. */
1913 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1915 memset(flow, 0, sizeof *flow);
1918 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1919 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1920 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1921 * result of the command is expected to be a flow also, which is decoded and
1922 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1923 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1925 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1926 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1928 struct ofpbuf *request_buf;
1931 ovs_assert((reply != NULL) == (bufp != NULL));
1934 request->nlmsg_flags |= NLM_F_ECHO;
1937 request_buf = ofpbuf_new(1024);
1938 dpif_linux_flow_to_ofpbuf(request, request_buf);
1939 error = nl_sock_transact(genl_sock, request_buf, bufp);
1940 ofpbuf_delete(request_buf);
1944 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1947 dpif_linux_flow_init(reply);
1948 ofpbuf_delete(*bufp);
1956 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1957 struct dpif_flow_stats *stats)
1960 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1961 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1963 stats->n_packets = 0;
1966 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1967 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1970 /* Logs information about a packet that was recently lost in 'ch' (in
1973 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
1975 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1976 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1979 if (VLOG_DROP_WARN(&rl)) {
1984 if (ch->last_poll != LLONG_MIN) {
1985 ds_put_format(&s, " (last polled %lld ms ago)",
1986 time_msec() - ch->last_poll);
1989 VLOG_WARN("%s: lost packet on channel %td%s",
1990 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));