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 char *name = netdev_vport_get_dpif_port(netdev);
486 const char *type = netdev_get_type(netdev);
487 struct dpif_linux_vport request, reply;
488 struct nl_sock *sock = NULL;
493 if (dpif->epoll_fd >= 0) {
494 error = nl_sock_create(NETLINK_GENERIC, &sock);
500 dpif_linux_vport_init(&request);
501 request.cmd = OVS_VPORT_CMD_NEW;
502 request.dp_ifindex = dpif->dp_ifindex;
503 request.type = netdev_to_ovs_vport_type(netdev);
504 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
505 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
506 "unsupported type `%s'",
507 dpif_name(dpif_), name, type);
508 nl_sock_destroy(sock);
513 if (request.type == OVS_VPORT_TYPE_NETDEV) {
514 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
517 request.port_no = *port_nop;
518 upcall_pid = sock ? nl_sock_pid(sock) : 0;
519 request.upcall_pid = &upcall_pid;
521 error = dpif_linux_vport_transact(&request, &reply, &buf);
523 *port_nop = reply.port_no;
524 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
525 dpif_name(dpif_), reply.port_no, upcall_pid);
527 if (error == EBUSY && *port_nop != UINT32_MAX) {
528 VLOG_INFO("%s: requested port %"PRIu32" is in use",
529 dpif_name(dpif_), *port_nop);
531 nl_sock_destroy(sock);
538 error = add_channel(dpif, *port_nop, sock);
540 VLOG_INFO("%s: could not add channel for port %s",
541 dpif_name(dpif_), name);
543 /* Delete the port. */
544 dpif_linux_vport_init(&request);
545 request.cmd = OVS_VPORT_CMD_DEL;
546 request.dp_ifindex = dpif->dp_ifindex;
547 request.port_no = *port_nop;
548 dpif_linux_vport_transact(&request, NULL, NULL);
550 nl_sock_destroy(sock);
559 dpif_linux_port_del(struct dpif *dpif_, uint32_t port_no)
561 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
562 struct dpif_linux_vport vport;
565 dpif_linux_vport_init(&vport);
566 vport.cmd = OVS_VPORT_CMD_DEL;
567 vport.dp_ifindex = dpif->dp_ifindex;
568 vport.port_no = port_no;
569 error = dpif_linux_vport_transact(&vport, NULL, NULL);
571 del_channel(dpif, port_no);
577 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
578 const char *port_name, struct dpif_port *dpif_port)
580 struct dpif_linux_vport request;
581 struct dpif_linux_vport reply;
585 dpif_linux_vport_init(&request);
586 request.cmd = OVS_VPORT_CMD_GET;
587 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
588 request.port_no = port_no;
589 request.name = port_name;
591 error = dpif_linux_vport_transact(&request, &reply, &buf);
593 if (reply.dp_ifindex != request.dp_ifindex) {
594 /* A query by name reported that 'port_name' is in some datapath
595 * other than 'dpif', but the caller wants to know about 'dpif'. */
597 } else if (dpif_port) {
598 dpif_port->name = xstrdup(reply.name);
599 dpif_port->type = xstrdup(get_vport_type(&reply));
600 dpif_port->port_no = reply.port_no;
608 dpif_linux_port_query_by_number(const struct dpif *dpif, uint32_t port_no,
609 struct dpif_port *dpif_port)
611 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
615 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
616 struct dpif_port *dpif_port)
618 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
622 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
628 dpif_linux_port_get_pid(const struct dpif *dpif_, uint32_t port_no)
630 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
632 if (dpif->epoll_fd < 0) {
635 /* The UINT32_MAX "reserved" port number uses the "ovs-system"'s
636 * channel, since it is not heavily loaded. */
637 int idx = (port_no >= dpif->uc_array_size) ? 0 : port_no;
638 return nl_sock_pid(dpif->channels[idx].sock);
643 dpif_linux_flow_flush(struct dpif *dpif_)
645 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
646 struct dpif_linux_flow flow;
648 dpif_linux_flow_init(&flow);
649 flow.cmd = OVS_FLOW_CMD_DEL;
650 flow.dp_ifindex = dpif->dp_ifindex;
651 return dpif_linux_flow_transact(&flow, NULL, NULL);
654 struct dpif_linux_port_state {
659 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
661 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
662 struct dpif_linux_port_state *state;
663 struct dpif_linux_vport request;
666 *statep = state = xmalloc(sizeof *state);
668 dpif_linux_vport_init(&request);
669 request.cmd = OVS_DP_CMD_GET;
670 request.dp_ifindex = dpif->dp_ifindex;
672 buf = ofpbuf_new(1024);
673 dpif_linux_vport_to_ofpbuf(&request, buf);
674 nl_dump_start(&state->dump, genl_sock, buf);
681 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
682 struct dpif_port *dpif_port)
684 struct dpif_linux_port_state *state = state_;
685 struct dpif_linux_vport vport;
689 if (!nl_dump_next(&state->dump, &buf)) {
693 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
698 dpif_port->name = CONST_CAST(char *, vport.name);
699 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
700 dpif_port->port_no = vport.port_no;
705 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
707 struct dpif_linux_port_state *state = state_;
708 int error = nl_dump_done(&state->dump);
715 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
717 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
719 if (dpif->change_error) {
720 dpif->change_error = false;
721 sset_clear(&dpif->changed_ports);
723 } else if (!sset_is_empty(&dpif->changed_ports)) {
724 *devnamep = sset_pop(&dpif->changed_ports);
732 dpif_linux_port_poll_wait(const struct dpif *dpif_)
734 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
735 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
736 poll_immediate_wake();
741 dpif_linux_flow_get__(const struct dpif *dpif_,
742 const struct nlattr *key, size_t key_len,
743 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
745 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
746 struct dpif_linux_flow request;
748 dpif_linux_flow_init(&request);
749 request.cmd = OVS_FLOW_CMD_GET;
750 request.dp_ifindex = dpif->dp_ifindex;
752 request.key_len = key_len;
753 return dpif_linux_flow_transact(&request, reply, bufp);
757 dpif_linux_flow_get(const struct dpif *dpif_,
758 const struct nlattr *key, size_t key_len,
759 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
761 struct dpif_linux_flow reply;
765 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
768 dpif_linux_flow_get_stats(&reply, stats);
771 buf->data = CONST_CAST(struct nlattr *, reply.actions);
772 buf->size = reply.actions_len;
782 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
783 struct dpif_linux_flow *request)
785 static struct nlattr dummy_action;
787 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
789 dpif_linux_flow_init(request);
790 request->cmd = (put->flags & DPIF_FP_CREATE
791 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
792 request->dp_ifindex = dpif->dp_ifindex;
793 request->key = put->key;
794 request->key_len = put->key_len;
795 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
796 request->actions = put->actions ? put->actions : &dummy_action;
797 request->actions_len = put->actions_len;
798 if (put->flags & DPIF_FP_ZERO_STATS) {
799 request->clear = true;
801 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
805 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
807 struct dpif_linux_flow request, reply;
811 dpif_linux_init_flow_put(dpif_, put, &request);
812 error = dpif_linux_flow_transact(&request,
813 put->stats ? &reply : NULL,
814 put->stats ? &buf : NULL);
815 if (!error && put->stats) {
816 dpif_linux_flow_get_stats(&reply, put->stats);
823 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
824 struct dpif_linux_flow *request)
826 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
828 dpif_linux_flow_init(request);
829 request->cmd = OVS_FLOW_CMD_DEL;
830 request->dp_ifindex = dpif->dp_ifindex;
831 request->key = del->key;
832 request->key_len = del->key_len;
836 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
838 struct dpif_linux_flow request, reply;
842 dpif_linux_init_flow_del(dpif_, del, &request);
843 error = dpif_linux_flow_transact(&request,
844 del->stats ? &reply : NULL,
845 del->stats ? &buf : NULL);
846 if (!error && del->stats) {
847 dpif_linux_flow_get_stats(&reply, del->stats);
853 struct dpif_linux_flow_state {
855 struct dpif_linux_flow flow;
856 struct dpif_flow_stats stats;
861 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
863 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
864 struct dpif_linux_flow_state *state;
865 struct dpif_linux_flow request;
868 *statep = state = xmalloc(sizeof *state);
870 dpif_linux_flow_init(&request);
871 request.cmd = OVS_DP_CMD_GET;
872 request.dp_ifindex = dpif->dp_ifindex;
874 buf = ofpbuf_new(1024);
875 dpif_linux_flow_to_ofpbuf(&request, buf);
876 nl_dump_start(&state->dump, genl_sock, buf);
885 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
886 const struct nlattr **key, size_t *key_len,
887 const struct nlattr **actions, size_t *actions_len,
888 const struct dpif_flow_stats **stats)
890 struct dpif_linux_flow_state *state = state_;
895 ofpbuf_delete(state->buf);
898 if (!nl_dump_next(&state->dump, &buf)) {
902 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
907 if (actions && !state->flow.actions) {
908 error = dpif_linux_flow_get__(dpif_, state->flow.key,
910 &state->flow, &state->buf);
911 if (error == ENOENT) {
912 VLOG_DBG("dumped flow disappeared on get");
914 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
920 *actions = state->flow.actions;
921 *actions_len = state->flow.actions_len;
924 *key = state->flow.key;
925 *key_len = state->flow.key_len;
928 dpif_linux_flow_get_stats(&state->flow, &state->stats);
929 *stats = &state->stats;
935 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
937 struct dpif_linux_flow_state *state = state_;
938 int error = nl_dump_done(&state->dump);
939 ofpbuf_delete(state->buf);
945 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
948 struct ovs_header *k_exec;
950 ofpbuf_prealloc_tailroom(buf, (64
951 + d_exec->packet->size
953 + d_exec->actions_len));
955 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
956 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
958 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
959 k_exec->dp_ifindex = dp_ifindex;
961 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
962 d_exec->packet->data, d_exec->packet->size);
963 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
964 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
965 d_exec->actions, d_exec->actions_len);
969 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
971 uint64_t request_stub[1024 / 8];
972 struct ofpbuf request;
975 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
976 dpif_linux_encode_execute(dp_ifindex, execute, &request);
977 error = nl_sock_transact(genl_sock, &request, NULL);
978 ofpbuf_uninit(&request);
984 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
986 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
988 return dpif_linux_execute__(dpif->dp_ifindex, execute);
994 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
996 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
999 struct nl_transaction txn;
1001 struct ofpbuf request;
1002 uint64_t request_stub[1024 / 8];
1004 struct ofpbuf reply;
1005 uint64_t reply_stub[1024 / 8];
1008 struct nl_transaction *txnsp[MAX_OPS];
1011 ovs_assert(n_ops <= MAX_OPS);
1012 for (i = 0; i < n_ops; i++) {
1013 struct op_auxdata *aux = &auxes[i];
1014 struct dpif_op *op = ops[i];
1015 struct dpif_flow_put *put;
1016 struct dpif_flow_del *del;
1017 struct dpif_execute *execute;
1018 struct dpif_linux_flow flow;
1020 ofpbuf_use_stub(&aux->request,
1021 aux->request_stub, sizeof aux->request_stub);
1022 aux->txn.request = &aux->request;
1024 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1025 aux->txn.reply = NULL;
1028 case DPIF_OP_FLOW_PUT:
1029 put = &op->u.flow_put;
1030 dpif_linux_init_flow_put(dpif_, put, &flow);
1032 flow.nlmsg_flags |= NLM_F_ECHO;
1033 aux->txn.reply = &aux->reply;
1035 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1038 case DPIF_OP_FLOW_DEL:
1039 del = &op->u.flow_del;
1040 dpif_linux_init_flow_del(dpif_, del, &flow);
1042 flow.nlmsg_flags |= NLM_F_ECHO;
1043 aux->txn.reply = &aux->reply;
1045 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1048 case DPIF_OP_EXECUTE:
1049 execute = &op->u.execute;
1050 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1059 for (i = 0; i < n_ops; i++) {
1060 txnsp[i] = &auxes[i].txn;
1062 nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
1064 for (i = 0; i < n_ops; i++) {
1065 struct op_auxdata *aux = &auxes[i];
1066 struct nl_transaction *txn = &auxes[i].txn;
1067 struct dpif_op *op = ops[i];
1068 struct dpif_flow_put *put;
1069 struct dpif_flow_del *del;
1071 op->error = txn->error;
1074 case DPIF_OP_FLOW_PUT:
1075 put = &op->u.flow_put;
1078 struct dpif_linux_flow reply;
1080 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1083 dpif_linux_flow_get_stats(&reply, put->stats);
1088 memset(put->stats, 0, sizeof *put->stats);
1093 case DPIF_OP_FLOW_DEL:
1094 del = &op->u.flow_del;
1097 struct dpif_linux_flow reply;
1099 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1102 dpif_linux_flow_get_stats(&reply, del->stats);
1107 memset(del->stats, 0, sizeof *del->stats);
1112 case DPIF_OP_EXECUTE:
1119 ofpbuf_uninit(&aux->request);
1120 ofpbuf_uninit(&aux->reply);
1125 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1128 size_t chunk = MIN(n_ops, MAX_OPS);
1129 dpif_linux_operate__(dpif, ops, chunk);
1136 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1138 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1140 if ((dpif->epoll_fd >= 0) == enable) {
1145 destroy_channels(dpif);
1147 struct dpif_port_dump port_dump;
1148 struct dpif_port port;
1150 dpif->epoll_fd = epoll_create(10);
1151 if (dpif->epoll_fd < 0) {
1155 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
1156 struct dpif_linux_vport vport_request;
1157 struct nl_sock *sock;
1158 uint32_t upcall_pid;
1161 error = nl_sock_create(NETLINK_GENERIC, &sock);
1166 upcall_pid = nl_sock_pid(sock);
1168 dpif_linux_vport_init(&vport_request);
1169 vport_request.cmd = OVS_VPORT_CMD_SET;
1170 vport_request.dp_ifindex = dpif->dp_ifindex;
1171 vport_request.port_no = port.port_no;
1172 vport_request.upcall_pid = &upcall_pid;
1173 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1175 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1176 dpif_name(&dpif->dpif), vport_request.port_no,
1179 VLOG_WARN_RL(&error_rl,
1180 "%s: failed to set upcall pid on port: %s",
1181 dpif_name(&dpif->dpif), strerror(error));
1182 nl_sock_destroy(sock);
1184 if (error == ENODEV || error == ENOENT) {
1185 /* This device isn't there, but keep trying the others. */
1192 error = add_channel(dpif, port.port_no, sock);
1194 VLOG_INFO("%s: could not add channel for port %s",
1195 dpif_name(dpif_), port.name);
1196 nl_sock_destroy(sock);
1206 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1207 uint32_t queue_id, uint32_t *priority)
1209 if (queue_id < 0xf000) {
1210 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1218 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1221 static const struct nl_policy ovs_packet_policy[] = {
1222 /* Always present. */
1223 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1224 .min_len = ETH_HEADER_LEN },
1225 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1227 /* OVS_PACKET_CMD_ACTION only. */
1228 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
1231 struct ovs_header *ovs_header;
1232 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1233 struct nlmsghdr *nlmsg;
1234 struct genlmsghdr *genl;
1238 ofpbuf_use_const(&b, buf->data, buf->size);
1240 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1241 genl = ofpbuf_try_pull(&b, sizeof *genl);
1242 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1243 if (!nlmsg || !genl || !ovs_header
1244 || nlmsg->nlmsg_type != ovs_packet_family
1245 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1246 ARRAY_SIZE(ovs_packet_policy))) {
1250 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1251 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1257 memset(upcall, 0, sizeof *upcall);
1258 upcall->type = type;
1259 upcall->packet = buf;
1260 upcall->packet->data = CONST_CAST(struct nlattr *,
1261 nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
1262 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1263 upcall->key = CONST_CAST(struct nlattr *,
1264 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1265 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1266 upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
1267 ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
1269 *dp_ifindex = ovs_header->dp_ifindex;
1275 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1278 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1281 if (dpif->epoll_fd < 0) {
1285 if (dpif->event_offset >= dpif->n_events) {
1288 dpif->event_offset = dpif->n_events = 0;
1291 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1292 dpif->uc_array_size, 0);
1293 } while (retval < 0 && errno == EINTR);
1295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1296 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
1297 } else if (retval > 0) {
1298 dpif->n_events = retval;
1302 while (dpif->event_offset < dpif->n_events) {
1303 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1304 struct dpif_channel *ch = &dpif->channels[idx];
1306 dpif->event_offset++;
1312 if (++read_tries > 50) {
1316 error = nl_sock_recv(ch->sock, buf, false);
1317 if (error == ENOBUFS) {
1318 /* ENOBUFS typically means that we've received so many
1319 * packets that the buffer overflowed. Try again
1320 * immediately because there's almost certainly a packet
1321 * waiting for us. */
1322 report_loss(dpif_, ch);
1326 ch->last_poll = time_msec();
1328 if (error == EAGAIN) {
1334 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1335 if (!error && dp_ifindex == dpif->dp_ifindex) {
1347 dpif_linux_recv_wait(struct dpif *dpif_)
1349 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1351 if (dpif->epoll_fd < 0) {
1355 poll_fd_wait(dpif->epoll_fd, POLLIN);
1359 dpif_linux_recv_purge(struct dpif *dpif_)
1361 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1362 struct dpif_channel *ch;
1364 if (dpif->epoll_fd < 0) {
1368 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
1370 nl_sock_drain(ch->sock);
1375 const struct dpif_class dpif_linux_class = {
1377 dpif_linux_enumerate,
1384 dpif_linux_get_stats,
1385 dpif_linux_port_add,
1386 dpif_linux_port_del,
1387 dpif_linux_port_query_by_number,
1388 dpif_linux_port_query_by_name,
1389 dpif_linux_get_max_ports,
1390 dpif_linux_port_get_pid,
1391 dpif_linux_port_dump_start,
1392 dpif_linux_port_dump_next,
1393 dpif_linux_port_dump_done,
1394 dpif_linux_port_poll,
1395 dpif_linux_port_poll_wait,
1396 dpif_linux_flow_get,
1397 dpif_linux_flow_put,
1398 dpif_linux_flow_del,
1399 dpif_linux_flow_flush,
1400 dpif_linux_flow_dump_start,
1401 dpif_linux_flow_dump_next,
1402 dpif_linux_flow_dump_done,
1405 dpif_linux_recv_set,
1406 dpif_linux_queue_to_priority,
1408 dpif_linux_recv_wait,
1409 dpif_linux_recv_purge,
1413 dpif_linux_init(void)
1415 static int error = -1;
1418 unsigned int ovs_vport_mcgroup;
1420 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1421 &ovs_datapath_family);
1423 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1424 "The Open vSwitch kernel module is probably not loaded.",
1425 OVS_DATAPATH_FAMILY);
1428 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1431 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1434 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1435 &ovs_packet_family);
1438 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1441 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1443 OVS_VPORT_MCGROUP_FALLBACK_ID);
1446 static struct dpif_linux_vport vport;
1447 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1448 dpif_linux_nln_parse, &vport);
1456 dpif_linux_is_internal_device(const char *name)
1458 struct dpif_linux_vport reply;
1462 error = dpif_linux_vport_get(name, &reply, &buf);
1465 } else if (error != ENODEV && error != ENOENT) {
1466 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1467 name, strerror(error));
1470 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1474 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1476 struct dpif_linux_vport *vport = vport_;
1477 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1481 dpif_linux_port_changed(const void *vport_, void *dpif_)
1483 const struct dpif_linux_vport *vport = vport_;
1484 struct dpif_linux *dpif = dpif_;
1487 if (vport->dp_ifindex == dpif->dp_ifindex
1488 && (vport->cmd == OVS_VPORT_CMD_NEW
1489 || vport->cmd == OVS_VPORT_CMD_DEL
1490 || vport->cmd == OVS_VPORT_CMD_SET)) {
1491 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1492 dpif->dpif.full_name, vport->name, vport->cmd);
1493 sset_add(&dpif->changed_ports, vport->name);
1496 dpif->change_error = true;
1500 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1501 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1502 * positive errno value.
1504 * 'vport' will contain pointers into 'buf', so the caller should not free
1505 * 'buf' while 'vport' is still in use. */
1507 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1508 const struct ofpbuf *buf)
1510 static const struct nl_policy ovs_vport_policy[] = {
1511 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1512 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1513 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1514 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1515 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1517 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1520 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1521 struct ovs_header *ovs_header;
1522 struct nlmsghdr *nlmsg;
1523 struct genlmsghdr *genl;
1526 dpif_linux_vport_init(vport);
1528 ofpbuf_use_const(&b, buf->data, buf->size);
1529 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1530 genl = ofpbuf_try_pull(&b, sizeof *genl);
1531 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1532 if (!nlmsg || !genl || !ovs_header
1533 || nlmsg->nlmsg_type != ovs_vport_family
1534 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1535 ARRAY_SIZE(ovs_vport_policy))) {
1539 vport->cmd = genl->cmd;
1540 vport->dp_ifindex = ovs_header->dp_ifindex;
1541 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1542 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1543 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1544 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1545 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1547 if (a[OVS_VPORT_ATTR_STATS]) {
1548 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1550 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1551 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1552 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1557 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1558 * followed by Netlink attributes corresponding to 'vport'. */
1560 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1563 struct ovs_header *ovs_header;
1565 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1566 vport->cmd, OVS_VPORT_VERSION);
1568 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1569 ovs_header->dp_ifindex = vport->dp_ifindex;
1571 if (vport->port_no != UINT32_MAX) {
1572 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1575 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1576 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1580 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1583 if (vport->upcall_pid) {
1584 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1588 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1589 vport->stats, sizeof *vport->stats);
1592 if (vport->options) {
1593 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1594 vport->options, vport->options_len);
1598 /* Clears 'vport' to "empty" values. */
1600 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1602 memset(vport, 0, sizeof *vport);
1603 vport->port_no = UINT32_MAX;
1606 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1607 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1608 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1609 * result of the command is expected to be an ovs_vport also, which is decoded
1610 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1611 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1613 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1614 struct dpif_linux_vport *reply,
1615 struct ofpbuf **bufp)
1617 struct ofpbuf *request_buf;
1620 ovs_assert((reply != NULL) == (bufp != NULL));
1622 error = dpif_linux_init();
1626 dpif_linux_vport_init(reply);
1631 request_buf = ofpbuf_new(1024);
1632 dpif_linux_vport_to_ofpbuf(request, request_buf);
1633 error = nl_sock_transact(genl_sock, request_buf, bufp);
1634 ofpbuf_delete(request_buf);
1638 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1641 dpif_linux_vport_init(reply);
1642 ofpbuf_delete(*bufp);
1649 /* Obtains information about the kernel vport named 'name' and stores it into
1650 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1651 * longer needed ('reply' will contain pointers into '*bufp'). */
1653 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1654 struct ofpbuf **bufp)
1656 struct dpif_linux_vport request;
1658 dpif_linux_vport_init(&request);
1659 request.cmd = OVS_VPORT_CMD_GET;
1660 request.name = name;
1662 return dpif_linux_vport_transact(&request, reply, bufp);
1665 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1666 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1667 * positive errno value.
1669 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1670 * while 'dp' is still in use. */
1672 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1674 static const struct nl_policy ovs_datapath_policy[] = {
1675 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1676 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1680 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1681 struct ovs_header *ovs_header;
1682 struct nlmsghdr *nlmsg;
1683 struct genlmsghdr *genl;
1686 dpif_linux_dp_init(dp);
1688 ofpbuf_use_const(&b, buf->data, buf->size);
1689 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1690 genl = ofpbuf_try_pull(&b, sizeof *genl);
1691 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1692 if (!nlmsg || !genl || !ovs_header
1693 || nlmsg->nlmsg_type != ovs_datapath_family
1694 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1695 ARRAY_SIZE(ovs_datapath_policy))) {
1699 dp->cmd = genl->cmd;
1700 dp->dp_ifindex = ovs_header->dp_ifindex;
1701 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1702 if (a[OVS_DP_ATTR_STATS]) {
1703 /* Can't use structure assignment because Netlink doesn't ensure
1704 * sufficient alignment for 64-bit members. */
1705 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1712 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1714 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1716 struct ovs_header *ovs_header;
1718 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1719 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1720 OVS_DATAPATH_VERSION);
1722 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1723 ovs_header->dp_ifindex = dp->dp_ifindex;
1726 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1729 if (dp->upcall_pid) {
1730 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1733 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1736 /* Clears 'dp' to "empty" values. */
1738 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1740 memset(dp, 0, sizeof *dp);
1744 dpif_linux_dp_dump_start(struct nl_dump *dump)
1746 struct dpif_linux_dp request;
1749 dpif_linux_dp_init(&request);
1750 request.cmd = OVS_DP_CMD_GET;
1752 buf = ofpbuf_new(1024);
1753 dpif_linux_dp_to_ofpbuf(&request, buf);
1754 nl_dump_start(dump, genl_sock, buf);
1758 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1759 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1760 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1761 * result of the command is expected to be of the same form, which is decoded
1762 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1763 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1765 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1766 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1768 struct ofpbuf *request_buf;
1771 ovs_assert((reply != NULL) == (bufp != NULL));
1773 request_buf = ofpbuf_new(1024);
1774 dpif_linux_dp_to_ofpbuf(request, request_buf);
1775 error = nl_sock_transact(genl_sock, request_buf, bufp);
1776 ofpbuf_delete(request_buf);
1780 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1783 dpif_linux_dp_init(reply);
1784 ofpbuf_delete(*bufp);
1791 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1792 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1793 * will contain pointers into '*bufp'). */
1795 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1796 struct ofpbuf **bufp)
1798 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1799 struct dpif_linux_dp request;
1801 dpif_linux_dp_init(&request);
1802 request.cmd = OVS_DP_CMD_GET;
1803 request.dp_ifindex = dpif->dp_ifindex;
1805 return dpif_linux_dp_transact(&request, reply, bufp);
1808 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1809 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1810 * positive errno value.
1812 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1813 * while 'flow' is still in use. */
1815 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1816 const struct ofpbuf *buf)
1818 static const struct nl_policy ovs_flow_policy[] = {
1819 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1820 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1821 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1823 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1824 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1825 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1828 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1829 struct ovs_header *ovs_header;
1830 struct nlmsghdr *nlmsg;
1831 struct genlmsghdr *genl;
1834 dpif_linux_flow_init(flow);
1836 ofpbuf_use_const(&b, buf->data, buf->size);
1837 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1838 genl = ofpbuf_try_pull(&b, sizeof *genl);
1839 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1840 if (!nlmsg || !genl || !ovs_header
1841 || nlmsg->nlmsg_type != ovs_flow_family
1842 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1843 ARRAY_SIZE(ovs_flow_policy))) {
1847 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1848 flow->dp_ifindex = ovs_header->dp_ifindex;
1849 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1850 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1851 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1852 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1853 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1855 if (a[OVS_FLOW_ATTR_STATS]) {
1856 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1858 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1859 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1861 if (a[OVS_FLOW_ATTR_USED]) {
1862 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1867 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1868 * followed by Netlink attributes corresponding to 'flow'. */
1870 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1873 struct ovs_header *ovs_header;
1875 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1876 NLM_F_REQUEST | flow->nlmsg_flags,
1877 flow->cmd, OVS_FLOW_VERSION);
1879 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1880 ovs_header->dp_ifindex = flow->dp_ifindex;
1882 if (flow->key_len) {
1883 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1886 if (flow->actions || flow->actions_len) {
1887 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1888 flow->actions, flow->actions_len);
1891 /* We never need to send these to the kernel. */
1892 ovs_assert(!flow->stats);
1893 ovs_assert(!flow->tcp_flags);
1894 ovs_assert(!flow->used);
1897 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1901 /* Clears 'flow' to "empty" values. */
1903 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1905 memset(flow, 0, sizeof *flow);
1908 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1909 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1910 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1911 * result of the command is expected to be a flow also, which is decoded and
1912 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1913 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1915 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1916 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1918 struct ofpbuf *request_buf;
1921 ovs_assert((reply != NULL) == (bufp != NULL));
1924 request->nlmsg_flags |= NLM_F_ECHO;
1927 request_buf = ofpbuf_new(1024);
1928 dpif_linux_flow_to_ofpbuf(request, request_buf);
1929 error = nl_sock_transact(genl_sock, request_buf, bufp);
1930 ofpbuf_delete(request_buf);
1934 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1937 dpif_linux_flow_init(reply);
1938 ofpbuf_delete(*bufp);
1946 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1947 struct dpif_flow_stats *stats)
1950 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1951 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1953 stats->n_packets = 0;
1956 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1957 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1960 /* Logs information about a packet that was recently lost in 'ch' (in
1963 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
1965 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1966 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1969 if (VLOG_DROP_WARN(&rl)) {
1974 if (ch->last_poll != LLONG_MIN) {
1975 ds_put_format(&s, " (last polled %lld ms ago)",
1976 time_msec() - ch->last_poll);
1979 VLOG_WARN("%s: lost packet on channel %td%s",
1980 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));