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"
51 #include "poll-loop.h"
56 #include "unaligned.h"
60 VLOG_DEFINE_THIS_MODULE(dpif_linux);
61 enum { MAX_PORTS = USHRT_MAX };
63 /* This ethtool flag was introduced in Linux 2.6.24, so it might be
64 * missing if we have old headers. */
65 #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
67 struct dpif_linux_dp {
68 /* Generic Netlink header. */
71 /* struct ovs_header. */
75 const char *name; /* OVS_DP_ATTR_NAME. */
76 const uint32_t *upcall_pid; /* OVS_DP_UPCALL_PID. */
77 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
80 static void dpif_linux_dp_init(struct dpif_linux_dp *);
81 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
82 const struct ofpbuf *);
83 static void dpif_linux_dp_dump_start(struct nl_dump *);
84 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
85 struct dpif_linux_dp *reply,
86 struct ofpbuf **bufp);
87 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
88 struct ofpbuf **bufp);
90 struct dpif_linux_flow {
91 /* Generic Netlink header. */
94 /* struct ovs_header. */
95 unsigned int nlmsg_flags;
100 * The 'stats' member points to 64-bit data that might only be aligned on
101 * 32-bit boundaries, so get_unaligned_u64() should be used to access its
104 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
105 * the Netlink version of the command, even if actions_len is zero. */
106 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
108 const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */
110 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
112 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
113 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
114 const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
115 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
118 static void dpif_linux_flow_init(struct dpif_linux_flow *);
119 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
120 const struct ofpbuf *);
121 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
123 static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
124 struct dpif_linux_flow *reply,
125 struct ofpbuf **bufp);
126 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
127 struct dpif_flow_stats *);
129 /* One of the dpif channels between the kernel and userspace. */
130 struct dpif_channel {
131 struct nl_sock *sock; /* Netlink socket. */
132 long long int last_poll; /* Last time this channel was polled. */
135 static void report_loss(struct dpif *, struct dpif_channel *);
137 /* Datapath interface for the openvswitch Linux kernel module. */
142 /* Upcall messages. */
143 int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
144 struct dpif_channel *channels;
145 struct epoll_event *epoll_events;
146 int epoll_fd; /* epoll fd that includes channel socks. */
147 int n_events; /* Num events returned by epoll_wait(). */
148 int event_offset; /* Offset into 'epoll_events'. */
150 /* Change notification. */
151 struct sset changed_ports; /* Ports that have changed. */
152 struct nln_notifier *port_notifier;
156 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
158 /* Generic Netlink family numbers for OVS. */
159 static int ovs_datapath_family;
160 static int ovs_vport_family;
161 static int ovs_flow_family;
162 static int ovs_packet_family;
164 /* Generic Netlink socket. */
165 static struct nl_sock *genl_sock;
166 static struct nln *nln = NULL;
168 static int dpif_linux_init(void);
169 static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
170 static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
171 static void dpif_linux_port_changed(const void *vport, void *dpif);
172 static uint32_t dpif_linux_port_get_pid(const struct dpif *,
175 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
177 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
178 const struct ofpbuf *);
180 static struct dpif_linux *
181 dpif_linux_cast(const struct dpif *dpif)
183 dpif_assert_class(dpif, &dpif_linux_class);
184 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
188 dpif_linux_enumerate(struct sset *all_dps)
194 error = dpif_linux_init();
199 dpif_linux_dp_dump_start(&dump);
200 while (nl_dump_next(&dump, &msg)) {
201 struct dpif_linux_dp dp;
203 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
204 sset_add(all_dps, dp.name);
207 return nl_dump_done(&dump);
211 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
212 bool create, struct dpif **dpifp)
214 struct dpif_linux_dp dp_request, dp;
219 error = dpif_linux_init();
224 /* Create or look up datapath. */
225 dpif_linux_dp_init(&dp_request);
227 dp_request.cmd = OVS_DP_CMD_NEW;
229 dp_request.upcall_pid = &upcall_pid;
231 dp_request.cmd = OVS_DP_CMD_GET;
233 dp_request.name = name;
234 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
239 open_dpif(&dp, dpifp);
245 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
247 struct dpif_linux *dpif;
249 dpif = xzalloc(sizeof *dpif);
250 dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
254 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
255 dp->dp_ifindex, dp->dp_ifindex);
257 dpif->dp_ifindex = dp->dp_ifindex;
258 sset_init(&dpif->changed_ports);
259 *dpifp = &dpif->dpif;
263 destroy_channels(struct dpif_linux *dpif)
267 if (dpif->epoll_fd < 0) {
271 for (i = 0; i < dpif->uc_array_size; i++ ) {
272 struct dpif_linux_vport vport_request;
273 struct dpif_channel *ch = &dpif->channels[i];
274 uint32_t upcall_pid = 0;
280 /* Turn off upcalls. */
281 dpif_linux_vport_init(&vport_request);
282 vport_request.cmd = OVS_VPORT_CMD_SET;
283 vport_request.dp_ifindex = dpif->dp_ifindex;
284 vport_request.port_no = u32_to_odp(i);
285 vport_request.upcall_pid = &upcall_pid;
286 dpif_linux_vport_transact(&vport_request, NULL, NULL);
288 nl_sock_destroy(ch->sock);
291 free(dpif->channels);
292 dpif->channels = NULL;
293 dpif->uc_array_size = 0;
295 free(dpif->epoll_events);
296 dpif->epoll_events = NULL;
297 dpif->n_events = dpif->event_offset = 0;
299 close(dpif->epoll_fd);
304 add_channel(struct dpif_linux *dpif, odp_port_t port_no, struct nl_sock *sock)
306 struct epoll_event event;
307 uint32_t port_idx = odp_to_u32(port_no);
309 if (dpif->epoll_fd < 0) {
313 /* We assume that the datapath densely chooses port numbers, which
314 * can therefore be used as an index into an array of channels. */
315 if (port_idx >= dpif->uc_array_size) {
316 uint32_t new_size = port_idx + 1;
319 if (new_size > MAX_PORTS) {
320 VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big",
321 dpif_name(&dpif->dpif), port_no);
325 dpif->channels = xrealloc(dpif->channels,
326 new_size * sizeof *dpif->channels);
327 for (i = dpif->uc_array_size; i < new_size; i++) {
328 dpif->channels[i].sock = NULL;
331 dpif->epoll_events = xrealloc(dpif->epoll_events,
332 new_size * sizeof *dpif->epoll_events);
333 dpif->uc_array_size = new_size;
336 memset(&event, 0, sizeof event);
337 event.events = EPOLLIN;
338 event.data.u32 = port_idx;
339 if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
344 nl_sock_destroy(dpif->channels[port_idx].sock);
345 dpif->channels[port_idx].sock = sock;
346 dpif->channels[port_idx].last_poll = LLONG_MIN;
352 del_channel(struct dpif_linux *dpif, odp_port_t port_no)
354 struct dpif_channel *ch;
355 uint32_t port_idx = odp_to_u32(port_no);
357 if (dpif->epoll_fd < 0 || port_idx >= dpif->uc_array_size) {
361 ch = &dpif->channels[port_idx];
366 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
367 dpif->event_offset = dpif->n_events = 0;
369 nl_sock_destroy(ch->sock);
374 dpif_linux_close(struct dpif *dpif_)
376 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
378 nln_notifier_destroy(dpif->port_notifier);
379 destroy_channels(dpif);
380 sset_destroy(&dpif->changed_ports);
385 dpif_linux_destroy(struct dpif *dpif_)
387 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
388 struct dpif_linux_dp dp;
390 dpif_linux_dp_init(&dp);
391 dp.cmd = OVS_DP_CMD_DEL;
392 dp.dp_ifindex = dpif->dp_ifindex;
393 return dpif_linux_dp_transact(&dp, NULL, NULL);
397 dpif_linux_run(struct dpif *dpif_ OVS_UNUSED)
405 dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
413 dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
415 struct dpif_linux_dp dp;
419 error = dpif_linux_dp_get(dpif_, &dp, &buf);
421 stats->n_hit = dp.stats.n_hit;
422 stats->n_missed = dp.stats.n_missed;
423 stats->n_lost = dp.stats.n_lost;
424 stats->n_flows = dp.stats.n_flows;
431 get_vport_type(const struct dpif_linux_vport *vport)
433 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
435 switch (vport->type) {
436 case OVS_VPORT_TYPE_NETDEV:
439 case OVS_VPORT_TYPE_INTERNAL:
442 case OVS_VPORT_TYPE_GRE:
445 case OVS_VPORT_TYPE_GRE64:
448 case OVS_VPORT_TYPE_VXLAN:
451 case OVS_VPORT_TYPE_LISP:
454 case OVS_VPORT_TYPE_UNSPEC:
455 case __OVS_VPORT_TYPE_MAX:
459 VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
460 vport->dp_ifindex, vport->name, (unsigned int) vport->type);
464 static enum ovs_vport_type
465 netdev_to_ovs_vport_type(const struct netdev *netdev)
467 const char *type = netdev_get_type(netdev);
469 if (!strcmp(type, "tap") || !strcmp(type, "system")) {
470 return OVS_VPORT_TYPE_NETDEV;
471 } else if (!strcmp(type, "internal")) {
472 return OVS_VPORT_TYPE_INTERNAL;
473 } else if (strstr(type, "gre64")) {
474 return OVS_VPORT_TYPE_GRE64;
475 } else if (strstr(type, "gre")) {
476 return OVS_VPORT_TYPE_GRE;
477 } else if (!strcmp(type, "vxlan")) {
478 return OVS_VPORT_TYPE_VXLAN;
479 } else if (!strcmp(type, "lisp")) {
480 return OVS_VPORT_TYPE_LISP;
482 return OVS_VPORT_TYPE_UNSPEC;
487 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
488 odp_port_t *port_nop)
490 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
491 const struct netdev_tunnel_config *tnl_cfg;
492 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
493 const char *name = netdev_vport_get_dpif_port(netdev,
494 namebuf, sizeof namebuf);
495 const char *type = netdev_get_type(netdev);
496 struct dpif_linux_vport request, reply;
497 struct nl_sock *sock = NULL;
500 uint64_t options_stub[64 / 8];
501 struct ofpbuf options;
504 if (dpif->epoll_fd >= 0) {
505 error = nl_sock_create(NETLINK_GENERIC, &sock);
511 dpif_linux_vport_init(&request);
512 request.cmd = OVS_VPORT_CMD_NEW;
513 request.dp_ifindex = dpif->dp_ifindex;
514 request.type = netdev_to_ovs_vport_type(netdev);
515 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
516 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
517 "unsupported type `%s'",
518 dpif_name(dpif_), name, type);
519 nl_sock_destroy(sock);
524 if (request.type == OVS_VPORT_TYPE_NETDEV) {
525 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
528 tnl_cfg = netdev_get_tunnel_config(netdev);
529 if (tnl_cfg && tnl_cfg->dst_port != 0) {
530 ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
531 nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
532 ntohs(tnl_cfg->dst_port));
533 request.options = options.data;
534 request.options_len = options.size;
537 request.port_no = *port_nop;
538 upcall_pid = sock ? nl_sock_pid(sock) : 0;
539 request.upcall_pid = &upcall_pid;
541 error = dpif_linux_vport_transact(&request, &reply, &buf);
543 *port_nop = reply.port_no;
544 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
545 dpif_name(dpif_), reply.port_no, upcall_pid);
547 if (error == EBUSY && *port_nop != ODPP_NONE) {
548 VLOG_INFO("%s: requested port %"PRIu32" is in use",
549 dpif_name(dpif_), *port_nop);
551 nl_sock_destroy(sock);
558 error = add_channel(dpif, *port_nop, sock);
560 VLOG_INFO("%s: could not add channel for port %s",
561 dpif_name(dpif_), name);
563 /* Delete the port. */
564 dpif_linux_vport_init(&request);
565 request.cmd = OVS_VPORT_CMD_DEL;
566 request.dp_ifindex = dpif->dp_ifindex;
567 request.port_no = *port_nop;
568 dpif_linux_vport_transact(&request, NULL, NULL);
570 nl_sock_destroy(sock);
579 dpif_linux_port_del(struct dpif *dpif_, odp_port_t port_no)
581 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
582 struct dpif_linux_vport vport;
585 dpif_linux_vport_init(&vport);
586 vport.cmd = OVS_VPORT_CMD_DEL;
587 vport.dp_ifindex = dpif->dp_ifindex;
588 vport.port_no = port_no;
589 error = dpif_linux_vport_transact(&vport, NULL, NULL);
591 del_channel(dpif, port_no);
597 dpif_linux_port_query__(const struct dpif *dpif, odp_port_t port_no,
598 const char *port_name, struct dpif_port *dpif_port)
600 struct dpif_linux_vport request;
601 struct dpif_linux_vport reply;
605 dpif_linux_vport_init(&request);
606 request.cmd = OVS_VPORT_CMD_GET;
607 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
608 request.port_no = port_no;
609 request.name = port_name;
611 error = dpif_linux_vport_transact(&request, &reply, &buf);
613 if (reply.dp_ifindex != request.dp_ifindex) {
614 /* A query by name reported that 'port_name' is in some datapath
615 * other than 'dpif', but the caller wants to know about 'dpif'. */
617 } else if (dpif_port) {
618 dpif_port->name = xstrdup(reply.name);
619 dpif_port->type = xstrdup(get_vport_type(&reply));
620 dpif_port->port_no = reply.port_no;
628 dpif_linux_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
629 struct dpif_port *dpif_port)
631 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
635 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
636 struct dpif_port *dpif_port)
638 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
642 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
644 return u32_to_odp(MAX_PORTS);
648 dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no)
650 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
651 uint32_t port_idx = odp_to_u32(port_no);
653 if (dpif->epoll_fd < 0) {
656 /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s
657 * channel, since it is not heavily loaded. */
658 uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx;
659 return nl_sock_pid(dpif->channels[idx].sock);
664 dpif_linux_flow_flush(struct dpif *dpif_)
666 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
667 struct dpif_linux_flow flow;
669 dpif_linux_flow_init(&flow);
670 flow.cmd = OVS_FLOW_CMD_DEL;
671 flow.dp_ifindex = dpif->dp_ifindex;
672 return dpif_linux_flow_transact(&flow, NULL, NULL);
675 struct dpif_linux_port_state {
680 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
682 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
683 struct dpif_linux_port_state *state;
684 struct dpif_linux_vport request;
687 *statep = state = xmalloc(sizeof *state);
689 dpif_linux_vport_init(&request);
690 request.cmd = OVS_DP_CMD_GET;
691 request.dp_ifindex = dpif->dp_ifindex;
693 buf = ofpbuf_new(1024);
694 dpif_linux_vport_to_ofpbuf(&request, buf);
695 nl_dump_start(&state->dump, genl_sock, buf);
702 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
703 struct dpif_port *dpif_port)
705 struct dpif_linux_port_state *state = state_;
706 struct dpif_linux_vport vport;
710 if (!nl_dump_next(&state->dump, &buf)) {
714 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
719 dpif_port->name = CONST_CAST(char *, vport.name);
720 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
721 dpif_port->port_no = vport.port_no;
726 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
728 struct dpif_linux_port_state *state = state_;
729 int error = nl_dump_done(&state->dump);
736 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
738 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
740 if (dpif->change_error) {
741 dpif->change_error = false;
742 sset_clear(&dpif->changed_ports);
744 } else if (!sset_is_empty(&dpif->changed_ports)) {
745 *devnamep = sset_pop(&dpif->changed_ports);
753 dpif_linux_port_poll_wait(const struct dpif *dpif_)
755 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
756 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
757 poll_immediate_wake();
762 dpif_linux_flow_get__(const struct dpif *dpif_,
763 const struct nlattr *key, size_t key_len,
764 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
766 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
767 struct dpif_linux_flow request;
769 dpif_linux_flow_init(&request);
770 request.cmd = OVS_FLOW_CMD_GET;
771 request.dp_ifindex = dpif->dp_ifindex;
773 request.key_len = key_len;
774 return dpif_linux_flow_transact(&request, reply, bufp);
778 dpif_linux_flow_get(const struct dpif *dpif_,
779 const struct nlattr *key, size_t key_len,
780 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
782 struct dpif_linux_flow reply;
786 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
789 dpif_linux_flow_get_stats(&reply, stats);
792 buf->data = CONST_CAST(struct nlattr *, reply.actions);
793 buf->size = reply.actions_len;
803 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
804 struct dpif_linux_flow *request)
806 static const struct nlattr dummy_action;
808 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
810 dpif_linux_flow_init(request);
811 request->cmd = (put->flags & DPIF_FP_CREATE
812 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
813 request->dp_ifindex = dpif->dp_ifindex;
814 request->key = put->key;
815 request->key_len = put->key_len;
816 request->mask = put->mask;
817 request->mask_len = put->mask_len;
818 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
819 request->actions = (put->actions
821 : CONST_CAST(struct nlattr *, &dummy_action));
822 request->actions_len = put->actions_len;
823 if (put->flags & DPIF_FP_ZERO_STATS) {
824 request->clear = true;
826 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
830 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
832 struct dpif_linux_flow request, reply;
836 dpif_linux_init_flow_put(dpif_, put, &request);
837 error = dpif_linux_flow_transact(&request,
838 put->stats ? &reply : NULL,
839 put->stats ? &buf : NULL);
840 if (!error && put->stats) {
841 dpif_linux_flow_get_stats(&reply, put->stats);
848 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
849 struct dpif_linux_flow *request)
851 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
853 dpif_linux_flow_init(request);
854 request->cmd = OVS_FLOW_CMD_DEL;
855 request->dp_ifindex = dpif->dp_ifindex;
856 request->key = del->key;
857 request->key_len = del->key_len;
861 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
863 struct dpif_linux_flow request, reply;
867 dpif_linux_init_flow_del(dpif_, del, &request);
868 error = dpif_linux_flow_transact(&request,
869 del->stats ? &reply : NULL,
870 del->stats ? &buf : NULL);
871 if (!error && del->stats) {
872 dpif_linux_flow_get_stats(&reply, del->stats);
878 struct dpif_linux_flow_state {
880 struct dpif_linux_flow flow;
881 struct dpif_flow_stats stats;
886 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
888 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
889 struct dpif_linux_flow_state *state;
890 struct dpif_linux_flow request;
893 *statep = state = xmalloc(sizeof *state);
895 dpif_linux_flow_init(&request);
896 request.cmd = OVS_DP_CMD_GET;
897 request.dp_ifindex = dpif->dp_ifindex;
899 buf = ofpbuf_new(1024);
900 dpif_linux_flow_to_ofpbuf(&request, buf);
901 nl_dump_start(&state->dump, genl_sock, buf);
910 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
911 const struct nlattr **key, size_t *key_len,
912 const struct nlattr **mask, size_t *mask_len,
913 const struct nlattr **actions, size_t *actions_len,
914 const struct dpif_flow_stats **stats)
916 struct dpif_linux_flow_state *state = state_;
921 ofpbuf_delete(state->buf);
924 if (!nl_dump_next(&state->dump, &buf)) {
928 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
933 if (actions && !state->flow.actions) {
934 error = dpif_linux_flow_get__(dpif_, state->flow.key,
936 &state->flow, &state->buf);
937 if (error == ENOENT) {
938 VLOG_DBG("dumped flow disappeared on get");
940 VLOG_WARN("error fetching dumped flow: %s",
941 ovs_strerror(error));
947 *actions = state->flow.actions;
948 *actions_len = state->flow.actions_len;
951 *key = state->flow.key;
952 *key_len = state->flow.key_len;
955 *mask = state->flow.mask;
956 *mask_len = state->flow.mask ? state->flow.mask_len : 0;
959 dpif_linux_flow_get_stats(&state->flow, &state->stats);
960 *stats = &state->stats;
966 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
968 struct dpif_linux_flow_state *state = state_;
969 int error = nl_dump_done(&state->dump);
970 ofpbuf_delete(state->buf);
976 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
979 struct ovs_header *k_exec;
981 ofpbuf_prealloc_tailroom(buf, (64
982 + d_exec->packet->size
984 + d_exec->actions_len));
986 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
987 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
989 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
990 k_exec->dp_ifindex = dp_ifindex;
992 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
993 d_exec->packet->data, d_exec->packet->size);
994 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
995 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
996 d_exec->actions, d_exec->actions_len);
1000 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
1002 uint64_t request_stub[1024 / 8];
1003 struct ofpbuf request;
1006 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
1007 dpif_linux_encode_execute(dp_ifindex, execute, &request);
1008 error = nl_sock_transact(genl_sock, &request, NULL);
1009 ofpbuf_uninit(&request);
1015 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
1017 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1019 return dpif_linux_execute__(dpif->dp_ifindex, execute);
1025 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
1027 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1030 struct nl_transaction txn;
1032 struct ofpbuf request;
1033 uint64_t request_stub[1024 / 8];
1035 struct ofpbuf reply;
1036 uint64_t reply_stub[1024 / 8];
1039 struct nl_transaction *txnsp[MAX_OPS];
1042 ovs_assert(n_ops <= MAX_OPS);
1043 for (i = 0; i < n_ops; i++) {
1044 struct op_auxdata *aux = &auxes[i];
1045 struct dpif_op *op = ops[i];
1046 struct dpif_flow_put *put;
1047 struct dpif_flow_del *del;
1048 struct dpif_execute *execute;
1049 struct dpif_linux_flow flow;
1051 ofpbuf_use_stub(&aux->request,
1052 aux->request_stub, sizeof aux->request_stub);
1053 aux->txn.request = &aux->request;
1055 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1056 aux->txn.reply = NULL;
1059 case DPIF_OP_FLOW_PUT:
1060 put = &op->u.flow_put;
1061 dpif_linux_init_flow_put(dpif_, put, &flow);
1063 flow.nlmsg_flags |= NLM_F_ECHO;
1064 aux->txn.reply = &aux->reply;
1066 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1069 case DPIF_OP_FLOW_DEL:
1070 del = &op->u.flow_del;
1071 dpif_linux_init_flow_del(dpif_, del, &flow);
1073 flow.nlmsg_flags |= NLM_F_ECHO;
1074 aux->txn.reply = &aux->reply;
1076 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1079 case DPIF_OP_EXECUTE:
1080 execute = &op->u.execute;
1081 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1090 for (i = 0; i < n_ops; i++) {
1091 txnsp[i] = &auxes[i].txn;
1093 nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
1095 for (i = 0; i < n_ops; i++) {
1096 struct op_auxdata *aux = &auxes[i];
1097 struct nl_transaction *txn = &auxes[i].txn;
1098 struct dpif_op *op = ops[i];
1099 struct dpif_flow_put *put;
1100 struct dpif_flow_del *del;
1102 op->error = txn->error;
1105 case DPIF_OP_FLOW_PUT:
1106 put = &op->u.flow_put;
1109 struct dpif_linux_flow reply;
1111 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1114 dpif_linux_flow_get_stats(&reply, put->stats);
1119 memset(put->stats, 0, sizeof *put->stats);
1124 case DPIF_OP_FLOW_DEL:
1125 del = &op->u.flow_del;
1128 struct dpif_linux_flow reply;
1130 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1133 dpif_linux_flow_get_stats(&reply, del->stats);
1138 memset(del->stats, 0, sizeof *del->stats);
1143 case DPIF_OP_EXECUTE:
1150 ofpbuf_uninit(&aux->request);
1151 ofpbuf_uninit(&aux->reply);
1156 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1159 size_t chunk = MIN(n_ops, MAX_OPS);
1160 dpif_linux_operate__(dpif, ops, chunk);
1167 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1169 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1171 if ((dpif->epoll_fd >= 0) == enable) {
1176 destroy_channels(dpif);
1178 struct dpif_port_dump port_dump;
1179 struct dpif_port port;
1181 dpif->epoll_fd = epoll_create(10);
1182 if (dpif->epoll_fd < 0) {
1186 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
1187 struct dpif_linux_vport vport_request;
1188 struct nl_sock *sock;
1189 uint32_t upcall_pid;
1192 error = nl_sock_create(NETLINK_GENERIC, &sock);
1197 upcall_pid = nl_sock_pid(sock);
1199 dpif_linux_vport_init(&vport_request);
1200 vport_request.cmd = OVS_VPORT_CMD_SET;
1201 vport_request.dp_ifindex = dpif->dp_ifindex;
1202 vport_request.port_no = port.port_no;
1203 vport_request.upcall_pid = &upcall_pid;
1204 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1206 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1207 dpif_name(&dpif->dpif), vport_request.port_no,
1210 VLOG_WARN_RL(&error_rl,
1211 "%s: failed to set upcall pid on port: %s",
1212 dpif_name(&dpif->dpif), ovs_strerror(error));
1213 nl_sock_destroy(sock);
1215 if (error == ENODEV || error == ENOENT) {
1216 /* This device isn't there, but keep trying the others. */
1223 error = add_channel(dpif, port.port_no, sock);
1225 VLOG_INFO("%s: could not add channel for port %s",
1226 dpif_name(dpif_), port.name);
1227 nl_sock_destroy(sock);
1237 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1238 uint32_t queue_id, uint32_t *priority)
1240 if (queue_id < 0xf000) {
1241 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1249 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1252 static const struct nl_policy ovs_packet_policy[] = {
1253 /* Always present. */
1254 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1255 .min_len = ETH_HEADER_LEN },
1256 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1258 /* OVS_PACKET_CMD_ACTION only. */
1259 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
1262 struct ovs_header *ovs_header;
1263 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1264 struct nlmsghdr *nlmsg;
1265 struct genlmsghdr *genl;
1269 ofpbuf_use_const(&b, buf->data, buf->size);
1271 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1272 genl = ofpbuf_try_pull(&b, sizeof *genl);
1273 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1274 if (!nlmsg || !genl || !ovs_header
1275 || nlmsg->nlmsg_type != ovs_packet_family
1276 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1277 ARRAY_SIZE(ovs_packet_policy))) {
1281 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1282 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1288 memset(upcall, 0, sizeof *upcall);
1289 upcall->type = type;
1290 upcall->packet = buf;
1291 upcall->packet->data = CONST_CAST(struct nlattr *,
1292 nl_attr_get(a[OVS_PACKET_ATTR_PACKET]));
1293 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1294 upcall->key = CONST_CAST(struct nlattr *,
1295 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1296 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1297 upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
1298 *dp_ifindex = ovs_header->dp_ifindex;
1304 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1307 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1310 if (dpif->epoll_fd < 0) {
1314 if (dpif->event_offset >= dpif->n_events) {
1317 dpif->event_offset = dpif->n_events = 0;
1320 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1321 dpif->uc_array_size, 0);
1322 } while (retval < 0 && errno == EINTR);
1324 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1325 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", ovs_strerror(errno));
1326 } else if (retval > 0) {
1327 dpif->n_events = retval;
1331 while (dpif->event_offset < dpif->n_events) {
1332 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1333 struct dpif_channel *ch = &dpif->channels[idx];
1335 dpif->event_offset++;
1341 if (++read_tries > 50) {
1345 error = nl_sock_recv(ch->sock, buf, false);
1346 if (error == ENOBUFS) {
1347 /* ENOBUFS typically means that we've received so many
1348 * packets that the buffer overflowed. Try again
1349 * immediately because there's almost certainly a packet
1350 * waiting for us. */
1351 report_loss(dpif_, ch);
1355 ch->last_poll = time_msec();
1357 if (error == EAGAIN) {
1363 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1364 if (!error && dp_ifindex == dpif->dp_ifindex) {
1376 dpif_linux_recv_wait(struct dpif *dpif_)
1378 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1380 if (dpif->epoll_fd < 0) {
1384 poll_fd_wait(dpif->epoll_fd, POLLIN);
1388 dpif_linux_recv_purge(struct dpif *dpif_)
1390 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1391 struct dpif_channel *ch;
1393 if (dpif->epoll_fd < 0) {
1397 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size]; ch++) {
1399 nl_sock_drain(ch->sock);
1404 const struct dpif_class dpif_linux_class = {
1406 dpif_linux_enumerate,
1413 dpif_linux_get_stats,
1414 dpif_linux_port_add,
1415 dpif_linux_port_del,
1416 dpif_linux_port_query_by_number,
1417 dpif_linux_port_query_by_name,
1418 dpif_linux_get_max_ports,
1419 dpif_linux_port_get_pid,
1420 dpif_linux_port_dump_start,
1421 dpif_linux_port_dump_next,
1422 dpif_linux_port_dump_done,
1423 dpif_linux_port_poll,
1424 dpif_linux_port_poll_wait,
1425 dpif_linux_flow_get,
1426 dpif_linux_flow_put,
1427 dpif_linux_flow_del,
1428 dpif_linux_flow_flush,
1429 dpif_linux_flow_dump_start,
1430 dpif_linux_flow_dump_next,
1431 dpif_linux_flow_dump_done,
1434 dpif_linux_recv_set,
1435 dpif_linux_queue_to_priority,
1437 dpif_linux_recv_wait,
1438 dpif_linux_recv_purge,
1442 dpif_linux_init(void)
1444 static int error = -1;
1447 unsigned int ovs_vport_mcgroup;
1449 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1450 &ovs_datapath_family);
1452 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1453 "The Open vSwitch kernel module is probably not loaded.",
1454 OVS_DATAPATH_FAMILY);
1457 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1460 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1463 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1464 &ovs_packet_family);
1467 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1470 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1472 OVS_VPORT_MCGROUP_FALLBACK_ID);
1475 static struct dpif_linux_vport vport;
1476 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1477 dpif_linux_nln_parse, &vport);
1485 dpif_linux_is_internal_device(const char *name)
1487 struct dpif_linux_vport reply;
1491 error = dpif_linux_vport_get(name, &reply, &buf);
1494 } else if (error != ENODEV && error != ENOENT) {
1495 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1496 name, ovs_strerror(error));
1499 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1503 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1505 struct dpif_linux_vport *vport = vport_;
1506 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1510 dpif_linux_port_changed(const void *vport_, void *dpif_)
1512 const struct dpif_linux_vport *vport = vport_;
1513 struct dpif_linux *dpif = dpif_;
1516 if (vport->dp_ifindex == dpif->dp_ifindex
1517 && (vport->cmd == OVS_VPORT_CMD_NEW
1518 || vport->cmd == OVS_VPORT_CMD_DEL
1519 || vport->cmd == OVS_VPORT_CMD_SET)) {
1520 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1521 dpif->dpif.full_name, vport->name, vport->cmd);
1522 sset_add(&dpif->changed_ports, vport->name);
1525 dpif->change_error = true;
1529 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1530 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1531 * positive errno value.
1533 * 'vport' will contain pointers into 'buf', so the caller should not free
1534 * 'buf' while 'vport' is still in use. */
1536 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1537 const struct ofpbuf *buf)
1539 static const struct nl_policy ovs_vport_policy[] = {
1540 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1541 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1542 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1543 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1544 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1546 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1549 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1550 struct ovs_header *ovs_header;
1551 struct nlmsghdr *nlmsg;
1552 struct genlmsghdr *genl;
1555 dpif_linux_vport_init(vport);
1557 ofpbuf_use_const(&b, buf->data, buf->size);
1558 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1559 genl = ofpbuf_try_pull(&b, sizeof *genl);
1560 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1561 if (!nlmsg || !genl || !ovs_header
1562 || nlmsg->nlmsg_type != ovs_vport_family
1563 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1564 ARRAY_SIZE(ovs_vport_policy))) {
1568 vport->cmd = genl->cmd;
1569 vport->dp_ifindex = ovs_header->dp_ifindex;
1570 vport->port_no = nl_attr_get_odp_port(a[OVS_VPORT_ATTR_PORT_NO]);
1571 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1572 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1573 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1574 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1576 if (a[OVS_VPORT_ATTR_STATS]) {
1577 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1579 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1580 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1581 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1586 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1587 * followed by Netlink attributes corresponding to 'vport'. */
1589 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1592 struct ovs_header *ovs_header;
1594 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1595 vport->cmd, OVS_VPORT_VERSION);
1597 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1598 ovs_header->dp_ifindex = vport->dp_ifindex;
1600 if (vport->port_no != ODPP_NONE) {
1601 nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1604 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1605 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1609 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1612 if (vport->upcall_pid) {
1613 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1617 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1618 vport->stats, sizeof *vport->stats);
1621 if (vport->options) {
1622 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1623 vport->options, vport->options_len);
1627 /* Clears 'vport' to "empty" values. */
1629 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1631 memset(vport, 0, sizeof *vport);
1632 vport->port_no = ODPP_NONE;
1635 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1636 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1637 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1638 * result of the command is expected to be an ovs_vport also, which is decoded
1639 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1640 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1642 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1643 struct dpif_linux_vport *reply,
1644 struct ofpbuf **bufp)
1646 struct ofpbuf *request_buf;
1649 ovs_assert((reply != NULL) == (bufp != NULL));
1651 error = dpif_linux_init();
1655 dpif_linux_vport_init(reply);
1660 request_buf = ofpbuf_new(1024);
1661 dpif_linux_vport_to_ofpbuf(request, request_buf);
1662 error = nl_sock_transact(genl_sock, request_buf, bufp);
1663 ofpbuf_delete(request_buf);
1667 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1670 dpif_linux_vport_init(reply);
1671 ofpbuf_delete(*bufp);
1678 /* Obtains information about the kernel vport named 'name' and stores it into
1679 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1680 * longer needed ('reply' will contain pointers into '*bufp'). */
1682 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1683 struct ofpbuf **bufp)
1685 struct dpif_linux_vport request;
1687 dpif_linux_vport_init(&request);
1688 request.cmd = OVS_VPORT_CMD_GET;
1689 request.name = name;
1691 return dpif_linux_vport_transact(&request, reply, bufp);
1694 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1695 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1696 * positive errno value.
1698 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1699 * while 'dp' is still in use. */
1701 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1703 static const struct nl_policy ovs_datapath_policy[] = {
1704 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1705 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1709 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1710 struct ovs_header *ovs_header;
1711 struct nlmsghdr *nlmsg;
1712 struct genlmsghdr *genl;
1715 dpif_linux_dp_init(dp);
1717 ofpbuf_use_const(&b, buf->data, buf->size);
1718 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1719 genl = ofpbuf_try_pull(&b, sizeof *genl);
1720 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1721 if (!nlmsg || !genl || !ovs_header
1722 || nlmsg->nlmsg_type != ovs_datapath_family
1723 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1724 ARRAY_SIZE(ovs_datapath_policy))) {
1728 dp->cmd = genl->cmd;
1729 dp->dp_ifindex = ovs_header->dp_ifindex;
1730 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1731 if (a[OVS_DP_ATTR_STATS]) {
1732 /* Can't use structure assignment because Netlink doesn't ensure
1733 * sufficient alignment for 64-bit members. */
1734 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1741 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1743 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1745 struct ovs_header *ovs_header;
1747 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1748 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1749 OVS_DATAPATH_VERSION);
1751 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1752 ovs_header->dp_ifindex = dp->dp_ifindex;
1755 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1758 if (dp->upcall_pid) {
1759 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1762 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1765 /* Clears 'dp' to "empty" values. */
1767 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1769 memset(dp, 0, sizeof *dp);
1773 dpif_linux_dp_dump_start(struct nl_dump *dump)
1775 struct dpif_linux_dp request;
1778 dpif_linux_dp_init(&request);
1779 request.cmd = OVS_DP_CMD_GET;
1781 buf = ofpbuf_new(1024);
1782 dpif_linux_dp_to_ofpbuf(&request, buf);
1783 nl_dump_start(dump, genl_sock, buf);
1787 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1788 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1789 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1790 * result of the command is expected to be of the same form, which is decoded
1791 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1792 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1794 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1795 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1797 struct ofpbuf *request_buf;
1800 ovs_assert((reply != NULL) == (bufp != NULL));
1802 request_buf = ofpbuf_new(1024);
1803 dpif_linux_dp_to_ofpbuf(request, request_buf);
1804 error = nl_sock_transact(genl_sock, request_buf, bufp);
1805 ofpbuf_delete(request_buf);
1809 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1812 dpif_linux_dp_init(reply);
1813 ofpbuf_delete(*bufp);
1820 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1821 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1822 * will contain pointers into '*bufp'). */
1824 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1825 struct ofpbuf **bufp)
1827 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1828 struct dpif_linux_dp request;
1830 dpif_linux_dp_init(&request);
1831 request.cmd = OVS_DP_CMD_GET;
1832 request.dp_ifindex = dpif->dp_ifindex;
1834 return dpif_linux_dp_transact(&request, reply, bufp);
1837 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1838 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1839 * positive errno value.
1841 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1842 * while 'flow' is still in use. */
1844 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1845 const struct ofpbuf *buf)
1847 static const struct nl_policy ovs_flow_policy[] = {
1848 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1849 [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true },
1850 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1851 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
1853 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1854 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1855 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1858 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1859 struct ovs_header *ovs_header;
1860 struct nlmsghdr *nlmsg;
1861 struct genlmsghdr *genl;
1864 dpif_linux_flow_init(flow);
1866 ofpbuf_use_const(&b, buf->data, buf->size);
1867 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1868 genl = ofpbuf_try_pull(&b, sizeof *genl);
1869 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1870 if (!nlmsg || !genl || !ovs_header
1871 || nlmsg->nlmsg_type != ovs_flow_family
1872 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1873 ARRAY_SIZE(ovs_flow_policy))) {
1877 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1878 flow->dp_ifindex = ovs_header->dp_ifindex;
1879 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1880 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1882 if (a[OVS_FLOW_ATTR_MASK]) {
1883 flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]);
1884 flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]);
1886 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1887 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1888 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1890 if (a[OVS_FLOW_ATTR_STATS]) {
1891 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1893 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1894 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1896 if (a[OVS_FLOW_ATTR_USED]) {
1897 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1902 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1903 * followed by Netlink attributes corresponding to 'flow'. */
1905 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1908 struct ovs_header *ovs_header;
1910 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1911 NLM_F_REQUEST | flow->nlmsg_flags,
1912 flow->cmd, OVS_FLOW_VERSION);
1914 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1915 ovs_header->dp_ifindex = flow->dp_ifindex;
1917 if (flow->key_len) {
1918 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1921 if (flow->mask_len) {
1922 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_MASK, flow->mask, flow->mask_len);
1925 if (flow->actions || flow->actions_len) {
1926 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1927 flow->actions, flow->actions_len);
1930 /* We never need to send these to the kernel. */
1931 ovs_assert(!flow->stats);
1932 ovs_assert(!flow->tcp_flags);
1933 ovs_assert(!flow->used);
1936 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1940 /* Clears 'flow' to "empty" values. */
1942 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1944 memset(flow, 0, sizeof *flow);
1947 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1948 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1949 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1950 * result of the command is expected to be a flow also, which is decoded and
1951 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1952 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1954 dpif_linux_flow_transact(struct dpif_linux_flow *request,
1955 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1957 struct ofpbuf *request_buf;
1960 ovs_assert((reply != NULL) == (bufp != NULL));
1963 request->nlmsg_flags |= NLM_F_ECHO;
1966 request_buf = ofpbuf_new(1024);
1967 dpif_linux_flow_to_ofpbuf(request, request_buf);
1968 error = nl_sock_transact(genl_sock, request_buf, bufp);
1969 ofpbuf_delete(request_buf);
1973 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1976 dpif_linux_flow_init(reply);
1977 ofpbuf_delete(*bufp);
1985 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1986 struct dpif_flow_stats *stats)
1989 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1990 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1992 stats->n_packets = 0;
1995 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
1996 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
1999 /* Logs information about a packet that was recently lost in 'ch' (in
2002 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
2004 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
2005 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
2008 if (VLOG_DROP_WARN(&rl)) {
2013 if (ch->last_poll != LLONG_MIN) {
2014 ds_put_format(&s, " (last polled %lld ms ago)",
2015 time_msec() - ch->last_poll);
2018 VLOG_WARN("%s: lost packet on channel %td%s",
2019 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));