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"
50 #include "poll-loop.h"
55 #include "unaligned.h"
59 VLOG_DEFINE_THIS_MODULE(dpif_linux);
60 enum { MAX_PORTS = USHRT_MAX };
62 /* This ethtool flag was introduced in Linux 2.6.24, so it might be
63 * missing if we have old headers. */
64 #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
66 struct dpif_linux_dp {
67 /* Generic Netlink header. */
70 /* struct ovs_header. */
74 const char *name; /* OVS_DP_ATTR_NAME. */
75 const uint32_t *upcall_pid; /* OVS_DP_ATTR_UPCALL_PID. */
76 uint32_t user_features; /* OVS_DP_ATTR_USER_FEATURES */
77 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
78 struct ovs_dp_megaflow_stats megaflow_stats;
79 /* OVS_DP_ATTR_MEGAFLOW_STATS.*/
82 static void dpif_linux_dp_init(struct dpif_linux_dp *);
83 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
84 const struct ofpbuf *);
85 static void dpif_linux_dp_dump_start(struct nl_dump *);
86 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
87 struct dpif_linux_dp *reply,
88 struct ofpbuf **bufp);
89 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
90 struct ofpbuf **bufp);
92 struct dpif_linux_flow {
93 /* Generic Netlink header. */
96 /* struct ovs_header. */
97 unsigned int nlmsg_flags;
102 * The 'stats' member points to 64-bit data that might only be aligned on
103 * 32-bit boundaries, so get_unaligned_u64() should be used to access its
106 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
107 * the Netlink version of the command, even if actions_len is zero. */
108 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
110 const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */
112 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
114 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
115 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
116 const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
117 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
120 static void dpif_linux_flow_init(struct dpif_linux_flow *);
121 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
122 const struct ofpbuf *);
123 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
125 static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
126 struct dpif_linux_flow *reply,
127 struct ofpbuf **bufp);
128 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
129 struct dpif_flow_stats *);
131 /* One of the dpif channels between the kernel and userspace. */
132 struct dpif_channel {
133 struct nl_sock *sock; /* Netlink socket. */
134 long long int last_poll; /* Last time this channel was polled. */
137 static void report_loss(struct dpif *, struct dpif_channel *);
139 /* Datapath interface for the openvswitch Linux kernel module. */
144 /* Upcall messages. */
145 struct ovs_mutex upcall_lock;
146 int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
147 struct dpif_channel *channels;
148 struct epoll_event *epoll_events;
149 int epoll_fd; /* epoll fd that includes channel socks. */
150 int n_events; /* Num events returned by epoll_wait(). */
151 int event_offset; /* Offset into 'epoll_events'. */
153 /* Change notification. */
154 struct nl_sock *port_notifier; /* vport multicast group subscriber. */
155 bool refresh_channels;
158 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
160 /* Generic Netlink family numbers for OVS.
162 * Initialized by dpif_linux_init(). */
163 static int ovs_datapath_family;
164 static int ovs_vport_family;
165 static int ovs_flow_family;
166 static int ovs_packet_family;
168 /* Generic Netlink multicast groups for OVS.
170 * Initialized by dpif_linux_init(). */
171 static unsigned int ovs_vport_mcgroup;
173 static int dpif_linux_init(void);
174 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
175 static uint32_t dpif_linux_port_get_pid(const struct dpif *,
177 static int dpif_linux_refresh_channels(struct dpif *);
179 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
181 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
182 const struct ofpbuf *);
184 static struct dpif_linux *
185 dpif_linux_cast(const struct dpif *dpif)
187 dpif_assert_class(dpif, &dpif_linux_class);
188 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
192 dpif_linux_enumerate(struct sset *all_dps)
198 error = dpif_linux_init();
203 dpif_linux_dp_dump_start(&dump);
204 while (nl_dump_next(&dump, &msg)) {
205 struct dpif_linux_dp dp;
207 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
208 sset_add(all_dps, dp.name);
211 return nl_dump_done(&dump);
215 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
216 bool create, struct dpif **dpifp)
218 struct dpif_linux_dp dp_request, dp;
223 error = dpif_linux_init();
228 /* Create or look up datapath. */
229 dpif_linux_dp_init(&dp_request);
231 dp_request.cmd = OVS_DP_CMD_NEW;
233 dp_request.upcall_pid = &upcall_pid;
235 /* Use OVS_DP_CMD_SET to report user features */
236 dp_request.cmd = OVS_DP_CMD_SET;
238 dp_request.name = name;
239 dp_request.user_features |= OVS_DP_F_UNALIGNED;
240 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
245 error = open_dpif(&dp, dpifp);
251 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
253 struct dpif_linux *dpif;
255 dpif = xzalloc(sizeof *dpif);
256 dpif->port_notifier = NULL;
257 ovs_mutex_init(&dpif->upcall_lock);
260 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
261 dp->dp_ifindex, dp->dp_ifindex);
263 dpif->dp_ifindex = dp->dp_ifindex;
264 *dpifp = &dpif->dpif;
270 destroy_channels(struct dpif_linux *dpif)
274 if (dpif->epoll_fd < 0) {
278 for (i = 0; i < dpif->uc_array_size; i++ ) {
279 struct dpif_linux_vport vport_request;
280 struct dpif_channel *ch = &dpif->channels[i];
281 uint32_t upcall_pid = 0;
287 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
289 /* Turn off upcalls. */
290 dpif_linux_vport_init(&vport_request);
291 vport_request.cmd = OVS_VPORT_CMD_SET;
292 vport_request.dp_ifindex = dpif->dp_ifindex;
293 vport_request.port_no = u32_to_odp(i);
294 vport_request.upcall_pid = &upcall_pid;
295 dpif_linux_vport_transact(&vport_request, NULL, NULL);
297 nl_sock_destroy(ch->sock);
300 free(dpif->channels);
301 dpif->channels = NULL;
302 dpif->uc_array_size = 0;
304 free(dpif->epoll_events);
305 dpif->epoll_events = NULL;
306 dpif->n_events = dpif->event_offset = 0;
308 /* Don't close dpif->epoll_fd since that would cause other threads that
309 * call dpif_recv_wait(dpif) to wait on an arbitrary fd or a closed fd. */
313 add_channel(struct dpif_linux *dpif, odp_port_t port_no, struct nl_sock *sock)
315 struct epoll_event event;
316 uint32_t port_idx = odp_to_u32(port_no);
318 if (dpif->epoll_fd < 0) {
322 /* We assume that the datapath densely chooses port numbers, which
323 * can therefore be used as an index into an array of channels. */
324 if (port_idx >= dpif->uc_array_size) {
325 uint32_t new_size = port_idx + 1;
328 if (new_size > MAX_PORTS) {
329 VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big",
330 dpif_name(&dpif->dpif), port_no);
334 dpif->channels = xrealloc(dpif->channels,
335 new_size * sizeof *dpif->channels);
336 for (i = dpif->uc_array_size; i < new_size; i++) {
337 dpif->channels[i].sock = NULL;
340 dpif->epoll_events = xrealloc(dpif->epoll_events,
341 new_size * sizeof *dpif->epoll_events);
342 dpif->uc_array_size = new_size;
345 memset(&event, 0, sizeof event);
346 event.events = EPOLLIN;
347 event.data.u32 = port_idx;
348 if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
353 nl_sock_destroy(dpif->channels[port_idx].sock);
354 dpif->channels[port_idx].sock = sock;
355 dpif->channels[port_idx].last_poll = LLONG_MIN;
361 del_channel(struct dpif_linux *dpif, odp_port_t port_no)
363 struct dpif_channel *ch;
364 uint32_t port_idx = odp_to_u32(port_no);
366 if (dpif->epoll_fd < 0 || port_idx >= dpif->uc_array_size) {
370 ch = &dpif->channels[port_idx];
375 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
376 dpif->event_offset = dpif->n_events = 0;
378 nl_sock_destroy(ch->sock);
383 dpif_linux_close(struct dpif *dpif_)
385 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
387 nl_sock_destroy(dpif->port_notifier);
388 destroy_channels(dpif);
389 if (dpif->epoll_fd >= 0) {
390 close(dpif->epoll_fd);
392 ovs_mutex_destroy(&dpif->upcall_lock);
397 dpif_linux_destroy(struct dpif *dpif_)
399 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
400 struct dpif_linux_dp dp;
402 dpif_linux_dp_init(&dp);
403 dp.cmd = OVS_DP_CMD_DEL;
404 dp.dp_ifindex = dpif->dp_ifindex;
405 return dpif_linux_dp_transact(&dp, NULL, NULL);
409 dpif_linux_run(struct dpif *dpif_)
411 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
412 if (dpif->refresh_channels) {
413 dpif->refresh_channels = false;
414 dpif_linux_refresh_channels(dpif_);
419 dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
421 struct dpif_linux_dp dp;
425 error = dpif_linux_dp_get(dpif_, &dp, &buf);
427 stats->n_hit = dp.stats.n_hit;
428 stats->n_missed = dp.stats.n_missed;
429 stats->n_lost = dp.stats.n_lost;
430 stats->n_flows = dp.stats.n_flows;
431 stats->n_masks = dp.megaflow_stats.n_masks;
432 stats->n_mask_hit = dp.megaflow_stats.n_mask_hit;
439 get_vport_type(const struct dpif_linux_vport *vport)
441 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
443 switch (vport->type) {
444 case OVS_VPORT_TYPE_NETDEV:
447 case OVS_VPORT_TYPE_INTERNAL:
450 case OVS_VPORT_TYPE_GRE:
453 case OVS_VPORT_TYPE_GRE64:
456 case OVS_VPORT_TYPE_VXLAN:
459 case OVS_VPORT_TYPE_LISP:
462 case OVS_VPORT_TYPE_UNSPEC:
463 case __OVS_VPORT_TYPE_MAX:
467 VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
468 vport->dp_ifindex, vport->name, (unsigned int) vport->type);
472 static enum ovs_vport_type
473 netdev_to_ovs_vport_type(const struct netdev *netdev)
475 const char *type = netdev_get_type(netdev);
477 if (!strcmp(type, "tap") || !strcmp(type, "system")) {
478 return OVS_VPORT_TYPE_NETDEV;
479 } else if (!strcmp(type, "internal")) {
480 return OVS_VPORT_TYPE_INTERNAL;
481 } else if (strstr(type, "gre64")) {
482 return OVS_VPORT_TYPE_GRE64;
483 } else if (strstr(type, "gre")) {
484 return OVS_VPORT_TYPE_GRE;
485 } else if (!strcmp(type, "vxlan")) {
486 return OVS_VPORT_TYPE_VXLAN;
487 } else if (!strcmp(type, "lisp")) {
488 return OVS_VPORT_TYPE_LISP;
490 return OVS_VPORT_TYPE_UNSPEC;
495 dpif_linux_port_add__(struct dpif *dpif_, struct netdev *netdev,
496 odp_port_t *port_nop)
498 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
499 const struct netdev_tunnel_config *tnl_cfg;
500 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
501 const char *name = netdev_vport_get_dpif_port(netdev,
502 namebuf, sizeof namebuf);
503 const char *type = netdev_get_type(netdev);
504 struct dpif_linux_vport request, reply;
505 struct nl_sock *sock = NULL;
508 uint64_t options_stub[64 / 8];
509 struct ofpbuf options;
512 if (dpif->epoll_fd >= 0) {
513 error = nl_sock_create(NETLINK_GENERIC, &sock);
519 dpif_linux_vport_init(&request);
520 request.cmd = OVS_VPORT_CMD_NEW;
521 request.dp_ifindex = dpif->dp_ifindex;
522 request.type = netdev_to_ovs_vport_type(netdev);
523 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
524 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
525 "unsupported type `%s'",
526 dpif_name(dpif_), name, type);
527 nl_sock_destroy(sock);
532 if (request.type == OVS_VPORT_TYPE_NETDEV) {
533 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
536 tnl_cfg = netdev_get_tunnel_config(netdev);
537 if (tnl_cfg && tnl_cfg->dst_port != 0) {
538 ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
539 nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
540 ntohs(tnl_cfg->dst_port));
541 request.options = options.data;
542 request.options_len = options.size;
545 request.port_no = *port_nop;
546 upcall_pid = sock ? nl_sock_pid(sock) : 0;
547 request.upcall_pid = &upcall_pid;
549 error = dpif_linux_vport_transact(&request, &reply, &buf);
551 *port_nop = reply.port_no;
552 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
553 dpif_name(dpif_), reply.port_no, upcall_pid);
555 if (error == EBUSY && *port_nop != ODPP_NONE) {
556 VLOG_INFO("%s: requested port %"PRIu32" is in use",
557 dpif_name(dpif_), *port_nop);
559 nl_sock_destroy(sock);
566 error = add_channel(dpif, *port_nop, sock);
568 VLOG_INFO("%s: could not add channel for port %s",
569 dpif_name(dpif_), name);
571 /* Delete the port. */
572 dpif_linux_vport_init(&request);
573 request.cmd = OVS_VPORT_CMD_DEL;
574 request.dp_ifindex = dpif->dp_ifindex;
575 request.port_no = *port_nop;
576 dpif_linux_vport_transact(&request, NULL, NULL);
578 nl_sock_destroy(sock);
587 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
588 odp_port_t *port_nop)
590 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
593 ovs_mutex_lock(&dpif->upcall_lock);
594 error = dpif_linux_port_add__(dpif_, netdev, port_nop);
595 ovs_mutex_unlock(&dpif->upcall_lock);
601 dpif_linux_port_del__(struct dpif *dpif_, odp_port_t port_no)
603 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
604 struct dpif_linux_vport vport;
607 dpif_linux_vport_init(&vport);
608 vport.cmd = OVS_VPORT_CMD_DEL;
609 vport.dp_ifindex = dpif->dp_ifindex;
610 vport.port_no = port_no;
611 error = dpif_linux_vport_transact(&vport, NULL, NULL);
613 del_channel(dpif, port_no);
619 dpif_linux_port_del(struct dpif *dpif_, odp_port_t port_no)
621 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
624 ovs_mutex_lock(&dpif->upcall_lock);
625 error = dpif_linux_port_del__(dpif_, port_no);
626 ovs_mutex_unlock(&dpif->upcall_lock);
632 dpif_linux_port_query__(const struct dpif *dpif, odp_port_t port_no,
633 const char *port_name, struct dpif_port *dpif_port)
635 struct dpif_linux_vport request;
636 struct dpif_linux_vport reply;
640 dpif_linux_vport_init(&request);
641 request.cmd = OVS_VPORT_CMD_GET;
642 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
643 request.port_no = port_no;
644 request.name = port_name;
646 error = dpif_linux_vport_transact(&request, &reply, &buf);
648 if (reply.dp_ifindex != request.dp_ifindex) {
649 /* A query by name reported that 'port_name' is in some datapath
650 * other than 'dpif', but the caller wants to know about 'dpif'. */
652 } else if (dpif_port) {
653 dpif_port->name = xstrdup(reply.name);
654 dpif_port->type = xstrdup(get_vport_type(&reply));
655 dpif_port->port_no = reply.port_no;
663 dpif_linux_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
664 struct dpif_port *dpif_port)
666 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
670 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
671 struct dpif_port *dpif_port)
673 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
677 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
683 dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no)
685 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
686 uint32_t port_idx = odp_to_u32(port_no);
689 ovs_mutex_lock(&dpif->upcall_lock);
690 if (dpif->epoll_fd >= 0) {
691 /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s
692 * channel, since it is not heavily loaded. */
693 uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx;
694 const struct nl_sock *sock = dpif->channels[idx].sock;
695 pid = sock ? nl_sock_pid(sock) : 0;
697 ovs_mutex_unlock(&dpif->upcall_lock);
703 dpif_linux_flow_flush(struct dpif *dpif_)
705 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
706 struct dpif_linux_flow flow;
708 dpif_linux_flow_init(&flow);
709 flow.cmd = OVS_FLOW_CMD_DEL;
710 flow.dp_ifindex = dpif->dp_ifindex;
711 return dpif_linux_flow_transact(&flow, NULL, NULL);
714 struct dpif_linux_port_state {
719 dpif_linux_port_dump_start__(const struct dpif *dpif_, struct nl_dump *dump)
721 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
722 struct dpif_linux_vport request;
725 dpif_linux_vport_init(&request);
726 request.cmd = OVS_VPORT_CMD_GET;
727 request.dp_ifindex = dpif->dp_ifindex;
729 buf = ofpbuf_new(1024);
730 dpif_linux_vport_to_ofpbuf(&request, buf);
731 nl_dump_start(dump, NETLINK_GENERIC, buf);
736 dpif_linux_port_dump_start(const struct dpif *dpif, void **statep)
738 struct dpif_linux_port_state *state;
740 *statep = state = xmalloc(sizeof *state);
741 dpif_linux_port_dump_start__(dpif, &state->dump);
747 dpif_linux_port_dump_next__(const struct dpif *dpif_, struct nl_dump *dump,
748 struct dpif_linux_vport *vport)
750 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
754 if (!nl_dump_next(dump, &buf)) {
758 error = dpif_linux_vport_from_ofpbuf(vport, &buf);
760 VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)",
761 dpif_name(&dpif->dpif), ovs_strerror(error));
767 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
768 struct dpif_port *dpif_port)
770 struct dpif_linux_port_state *state = state_;
771 struct dpif_linux_vport vport;
774 error = dpif_linux_port_dump_next__(dpif, &state->dump, &vport);
778 dpif_port->name = CONST_CAST(char *, vport.name);
779 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
780 dpif_port->port_no = vport.port_no;
785 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
787 struct dpif_linux_port_state *state = state_;
788 int error = nl_dump_done(&state->dump);
795 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
797 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
799 /* Lazily create the Netlink socket to listen for notifications. */
800 if (!dpif->port_notifier) {
801 struct nl_sock *sock;
804 error = nl_sock_create(NETLINK_GENERIC, &sock);
809 error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup);
811 nl_sock_destroy(sock);
814 dpif->port_notifier = sock;
816 /* We have no idea of the current state so report that everything
822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
823 uint64_t buf_stub[4096 / 8];
827 ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
828 error = nl_sock_recv(dpif->port_notifier, &buf, false);
830 struct dpif_linux_vport vport;
832 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
834 if (vport.dp_ifindex == dpif->dp_ifindex
835 && (vport.cmd == OVS_VPORT_CMD_NEW
836 || vport.cmd == OVS_VPORT_CMD_DEL
837 || vport.cmd == OVS_VPORT_CMD_SET)) {
838 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
839 dpif->dpif.full_name, vport.name, vport.cmd);
840 if (vport.cmd == OVS_VPORT_CMD_DEL) {
841 dpif->refresh_channels = true;
843 *devnamep = xstrdup(vport.name);
848 } else if (error != EAGAIN) {
849 VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
850 ovs_strerror(error));
851 nl_sock_drain(dpif->port_notifier);
863 dpif_linux_port_poll_wait(const struct dpif *dpif_)
865 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
867 if (dpif->port_notifier) {
868 nl_sock_wait(dpif->port_notifier, POLLIN);
870 poll_immediate_wake();
875 dpif_linux_flow_get__(const struct dpif *dpif_,
876 const struct nlattr *key, size_t key_len,
877 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
879 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
880 struct dpif_linux_flow request;
882 dpif_linux_flow_init(&request);
883 request.cmd = OVS_FLOW_CMD_GET;
884 request.dp_ifindex = dpif->dp_ifindex;
886 request.key_len = key_len;
887 return dpif_linux_flow_transact(&request, reply, bufp);
891 dpif_linux_flow_get(const struct dpif *dpif_,
892 const struct nlattr *key, size_t key_len,
893 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
895 struct dpif_linux_flow reply;
899 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
902 dpif_linux_flow_get_stats(&reply, stats);
905 buf->data = CONST_CAST(struct nlattr *, reply.actions);
906 buf->size = reply.actions_len;
916 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
917 struct dpif_linux_flow *request)
919 static const struct nlattr dummy_action;
921 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
923 dpif_linux_flow_init(request);
924 request->cmd = (put->flags & DPIF_FP_CREATE
925 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
926 request->dp_ifindex = dpif->dp_ifindex;
927 request->key = put->key;
928 request->key_len = put->key_len;
929 request->mask = put->mask;
930 request->mask_len = put->mask_len;
931 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
932 request->actions = (put->actions
934 : CONST_CAST(struct nlattr *, &dummy_action));
935 request->actions_len = put->actions_len;
936 if (put->flags & DPIF_FP_ZERO_STATS) {
937 request->clear = true;
939 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
943 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
945 struct dpif_linux_flow request, reply;
949 dpif_linux_init_flow_put(dpif_, put, &request);
950 error = dpif_linux_flow_transact(&request,
951 put->stats ? &reply : NULL,
952 put->stats ? &buf : NULL);
953 if (!error && put->stats) {
954 dpif_linux_flow_get_stats(&reply, put->stats);
961 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
962 struct dpif_linux_flow *request)
964 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
966 dpif_linux_flow_init(request);
967 request->cmd = OVS_FLOW_CMD_DEL;
968 request->dp_ifindex = dpif->dp_ifindex;
969 request->key = del->key;
970 request->key_len = del->key_len;
974 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
976 struct dpif_linux_flow request, reply;
980 dpif_linux_init_flow_del(dpif_, del, &request);
981 error = dpif_linux_flow_transact(&request,
982 del->stats ? &reply : NULL,
983 del->stats ? &buf : NULL);
984 if (!error && del->stats) {
985 dpif_linux_flow_get_stats(&reply, del->stats);
991 struct dpif_linux_flow_state {
993 struct dpif_linux_flow flow;
994 struct dpif_flow_stats stats;
999 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
1001 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1002 struct dpif_linux_flow_state *state;
1003 struct dpif_linux_flow request;
1006 *statep = state = xmalloc(sizeof *state);
1008 dpif_linux_flow_init(&request);
1009 request.cmd = OVS_FLOW_CMD_GET;
1010 request.dp_ifindex = dpif->dp_ifindex;
1012 buf = ofpbuf_new(1024);
1013 dpif_linux_flow_to_ofpbuf(&request, buf);
1014 nl_dump_start(&state->dump, NETLINK_GENERIC, buf);
1023 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
1024 const struct nlattr **key, size_t *key_len,
1025 const struct nlattr **mask, size_t *mask_len,
1026 const struct nlattr **actions, size_t *actions_len,
1027 const struct dpif_flow_stats **stats)
1029 struct dpif_linux_flow_state *state = state_;
1034 ofpbuf_delete(state->buf);
1037 if (!nl_dump_next(&state->dump, &buf)) {
1041 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
1046 if (actions && !state->flow.actions) {
1047 error = dpif_linux_flow_get__(dpif_, state->flow.key,
1048 state->flow.key_len,
1049 &state->flow, &state->buf);
1050 if (error == ENOENT) {
1051 VLOG_DBG("dumped flow disappeared on get");
1053 VLOG_WARN("error fetching dumped flow: %s",
1054 ovs_strerror(error));
1060 *actions = state->flow.actions;
1061 *actions_len = state->flow.actions_len;
1064 *key = state->flow.key;
1065 *key_len = state->flow.key_len;
1068 *mask = state->flow.mask;
1069 *mask_len = state->flow.mask ? state->flow.mask_len : 0;
1072 dpif_linux_flow_get_stats(&state->flow, &state->stats);
1073 *stats = &state->stats;
1079 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1081 struct dpif_linux_flow_state *state = state_;
1082 int error = nl_dump_done(&state->dump);
1083 ofpbuf_delete(state->buf);
1089 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
1092 struct ovs_header *k_exec;
1094 ofpbuf_prealloc_tailroom(buf, (64
1095 + d_exec->packet->size
1097 + d_exec->actions_len));
1099 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
1100 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
1102 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
1103 k_exec->dp_ifindex = dp_ifindex;
1105 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
1106 d_exec->packet->data, d_exec->packet->size);
1107 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
1108 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
1109 d_exec->actions, d_exec->actions_len);
1113 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
1115 uint64_t request_stub[1024 / 8];
1116 struct ofpbuf request;
1119 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
1120 dpif_linux_encode_execute(dp_ifindex, execute, &request);
1121 error = nl_transact(NETLINK_GENERIC, &request, NULL);
1122 ofpbuf_uninit(&request);
1128 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
1130 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1132 return dpif_linux_execute__(dpif->dp_ifindex, execute);
1138 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
1140 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1143 struct nl_transaction txn;
1145 struct ofpbuf request;
1146 uint64_t request_stub[1024 / 8];
1148 struct ofpbuf reply;
1149 uint64_t reply_stub[1024 / 8];
1152 struct nl_transaction *txnsp[MAX_OPS];
1155 ovs_assert(n_ops <= MAX_OPS);
1156 for (i = 0; i < n_ops; i++) {
1157 struct op_auxdata *aux = &auxes[i];
1158 struct dpif_op *op = ops[i];
1159 struct dpif_flow_put *put;
1160 struct dpif_flow_del *del;
1161 struct dpif_execute *execute;
1162 struct dpif_linux_flow flow;
1164 ofpbuf_use_stub(&aux->request,
1165 aux->request_stub, sizeof aux->request_stub);
1166 aux->txn.request = &aux->request;
1168 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1169 aux->txn.reply = NULL;
1172 case DPIF_OP_FLOW_PUT:
1173 put = &op->u.flow_put;
1174 dpif_linux_init_flow_put(dpif_, put, &flow);
1176 flow.nlmsg_flags |= NLM_F_ECHO;
1177 aux->txn.reply = &aux->reply;
1179 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1182 case DPIF_OP_FLOW_DEL:
1183 del = &op->u.flow_del;
1184 dpif_linux_init_flow_del(dpif_, del, &flow);
1186 flow.nlmsg_flags |= NLM_F_ECHO;
1187 aux->txn.reply = &aux->reply;
1189 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1192 case DPIF_OP_EXECUTE:
1193 execute = &op->u.execute;
1194 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1203 for (i = 0; i < n_ops; i++) {
1204 txnsp[i] = &auxes[i].txn;
1206 nl_transact_multiple(NETLINK_GENERIC, txnsp, n_ops);
1208 for (i = 0; i < n_ops; i++) {
1209 struct op_auxdata *aux = &auxes[i];
1210 struct nl_transaction *txn = &auxes[i].txn;
1211 struct dpif_op *op = ops[i];
1212 struct dpif_flow_put *put;
1213 struct dpif_flow_del *del;
1215 op->error = txn->error;
1218 case DPIF_OP_FLOW_PUT:
1219 put = &op->u.flow_put;
1222 struct dpif_linux_flow reply;
1224 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1227 dpif_linux_flow_get_stats(&reply, put->stats);
1232 memset(put->stats, 0, sizeof *put->stats);
1237 case DPIF_OP_FLOW_DEL:
1238 del = &op->u.flow_del;
1241 struct dpif_linux_flow reply;
1243 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1246 dpif_linux_flow_get_stats(&reply, del->stats);
1251 memset(del->stats, 0, sizeof *del->stats);
1256 case DPIF_OP_EXECUTE:
1263 ofpbuf_uninit(&aux->request);
1264 ofpbuf_uninit(&aux->reply);
1269 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1272 size_t chunk = MIN(n_ops, MAX_OPS);
1273 dpif_linux_operate__(dpif, ops, chunk);
1279 /* Synchronizes 'dpif->channels' with the set of vports currently in 'dpif' in
1280 * the kernel, by adding a new channel for any kernel vport that lacks one and
1281 * deleting any channels that have no backing kernel vports. */
1283 dpif_linux_refresh_channels(struct dpif *dpif_)
1285 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1286 unsigned long int *keep_channels;
1287 struct dpif_linux_vport vport;
1288 size_t keep_channels_nbits;
1289 struct nl_dump dump;
1293 /* To start with, we need an epoll fd. */
1294 if (dpif->epoll_fd < 0) {
1295 dpif->epoll_fd = epoll_create(10);
1296 if (dpif->epoll_fd < 0) {
1301 keep_channels_nbits = dpif->uc_array_size;
1302 keep_channels = bitmap_allocate(keep_channels_nbits);
1304 dpif->n_events = dpif->event_offset = 0;
1306 dpif_linux_port_dump_start__(dpif_, &dump);
1307 while (!dpif_linux_port_dump_next__(dpif_, &dump, &vport)) {
1308 uint32_t port_no = odp_to_u32(vport.port_no);
1309 struct nl_sock *sock = (port_no < dpif->uc_array_size
1310 ? dpif->channels[port_no].sock
1312 bool new_sock = !sock;
1316 error = nl_sock_create(NETLINK_GENERIC, &sock);
1323 /* Configure the vport to deliver misses to 'sock'. */
1324 if (!vport.upcall_pid || *vport.upcall_pid != nl_sock_pid(sock)) {
1325 uint32_t upcall_pid = nl_sock_pid(sock);
1326 struct dpif_linux_vport vport_request;
1328 dpif_linux_vport_init(&vport_request);
1329 vport_request.cmd = OVS_VPORT_CMD_SET;
1330 vport_request.dp_ifindex = dpif->dp_ifindex;
1331 vport_request.port_no = vport.port_no;
1332 vport_request.upcall_pid = &upcall_pid;
1333 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1335 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1336 dpif_name(&dpif->dpif), vport_request.port_no,
1339 VLOG_WARN_RL(&error_rl,
1340 "%s: failed to set upcall pid on port: %s",
1341 dpif_name(&dpif->dpif), ovs_strerror(error));
1343 if (error != ENODEV && error != ENOENT) {
1346 /* The vport isn't really there, even though the dump says
1347 * it is. Probably we just hit a race after a port
1355 error = add_channel(dpif, vport.port_no, sock);
1357 VLOG_INFO("%s: could not add channel for port %s",
1358 dpif_name(dpif_), vport.name);
1364 if (port_no < keep_channels_nbits) {
1365 bitmap_set1(keep_channels, port_no);
1370 nl_sock_destroy(sock);
1372 nl_dump_done(&dump);
1374 /* Discard any saved channels that we didn't reuse. */
1375 for (i = 0; i < keep_channels_nbits; i++) {
1376 if (!bitmap_is_set(keep_channels, i)) {
1377 nl_sock_destroy(dpif->channels[i].sock);
1378 dpif->channels[i].sock = NULL;
1381 free(keep_channels);
1387 dpif_linux_recv_set__(struct dpif *dpif_, bool enable)
1389 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1391 if ((dpif->epoll_fd >= 0) == enable) {
1393 } else if (!enable) {
1394 destroy_channels(dpif);
1397 return dpif_linux_refresh_channels(dpif_);
1402 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1404 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1407 ovs_mutex_lock(&dpif->upcall_lock);
1408 error = dpif_linux_recv_set__(dpif_, enable);
1409 ovs_mutex_unlock(&dpif->upcall_lock);
1415 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1416 uint32_t queue_id, uint32_t *priority)
1418 if (queue_id < 0xf000) {
1419 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1427 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1430 static const struct nl_policy ovs_packet_policy[] = {
1431 /* Always present. */
1432 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1433 .min_len = ETH_HEADER_LEN },
1434 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1436 /* OVS_PACKET_CMD_ACTION only. */
1437 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
1440 struct ovs_header *ovs_header;
1441 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1442 struct nlmsghdr *nlmsg;
1443 struct genlmsghdr *genl;
1447 ofpbuf_use_const(&b, buf->data, buf->size);
1449 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1450 genl = ofpbuf_try_pull(&b, sizeof *genl);
1451 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1452 if (!nlmsg || !genl || !ovs_header
1453 || nlmsg->nlmsg_type != ovs_packet_family
1454 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1455 ARRAY_SIZE(ovs_packet_policy))) {
1459 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1460 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1466 /* (Re)set ALL fields of '*upcall' on successful return. */
1467 upcall->type = type;
1468 upcall->key = CONST_CAST(struct nlattr *,
1469 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1470 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1471 upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
1473 /* Allow overwriting the netlink attribute header without reallocating. */
1474 ofpbuf_use_stub(&upcall->packet,
1475 CONST_CAST(struct nlattr *,
1476 nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1,
1477 nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) +
1478 sizeof(struct nlattr));
1479 upcall->packet.data = (char *)upcall->packet.data + sizeof(struct nlattr);
1480 upcall->packet.size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1482 *dp_ifindex = ovs_header->dp_ifindex;
1488 dpif_linux_recv__(struct dpif *dpif_, struct dpif_upcall *upcall,
1491 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1494 if (dpif->epoll_fd < 0) {
1498 if (dpif->event_offset >= dpif->n_events) {
1501 dpif->event_offset = dpif->n_events = 0;
1504 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1505 dpif->uc_array_size, 0);
1506 } while (retval < 0 && errno == EINTR);
1508 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1509 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", ovs_strerror(errno));
1510 } else if (retval > 0) {
1511 dpif->n_events = retval;
1515 while (dpif->event_offset < dpif->n_events) {
1516 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1517 struct dpif_channel *ch = &dpif->channels[idx];
1519 dpif->event_offset++;
1525 if (++read_tries > 50) {
1529 error = nl_sock_recv(ch->sock, buf, false);
1530 if (error == ENOBUFS) {
1531 /* ENOBUFS typically means that we've received so many
1532 * packets that the buffer overflowed. Try again
1533 * immediately because there's almost certainly a packet
1534 * waiting for us. */
1535 report_loss(dpif_, ch);
1539 ch->last_poll = time_msec();
1541 if (error == EAGAIN) {
1547 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1548 if (!error && dp_ifindex == dpif->dp_ifindex) {
1560 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1563 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1566 ovs_mutex_lock(&dpif->upcall_lock);
1567 error = dpif_linux_recv__(dpif_, upcall, buf);
1568 ovs_mutex_unlock(&dpif->upcall_lock);
1574 dpif_linux_recv_wait(struct dpif *dpif_)
1576 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1578 ovs_mutex_lock(&dpif->upcall_lock);
1579 if (dpif->epoll_fd >= 0) {
1580 poll_fd_wait(dpif->epoll_fd, POLLIN);
1582 ovs_mutex_unlock(&dpif->upcall_lock);
1586 dpif_linux_recv_purge(struct dpif *dpif_)
1588 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1590 ovs_mutex_lock(&dpif->upcall_lock);
1591 if (dpif->epoll_fd >= 0) {
1592 struct dpif_channel *ch;
1594 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size];
1597 nl_sock_drain(ch->sock);
1601 ovs_mutex_unlock(&dpif->upcall_lock);
1604 const struct dpif_class dpif_linux_class = {
1606 dpif_linux_enumerate,
1613 dpif_linux_get_stats,
1614 dpif_linux_port_add,
1615 dpif_linux_port_del,
1616 dpif_linux_port_query_by_number,
1617 dpif_linux_port_query_by_name,
1618 dpif_linux_get_max_ports,
1619 dpif_linux_port_get_pid,
1620 dpif_linux_port_dump_start,
1621 dpif_linux_port_dump_next,
1622 dpif_linux_port_dump_done,
1623 dpif_linux_port_poll,
1624 dpif_linux_port_poll_wait,
1625 dpif_linux_flow_get,
1626 dpif_linux_flow_put,
1627 dpif_linux_flow_del,
1628 dpif_linux_flow_flush,
1629 dpif_linux_flow_dump_start,
1630 dpif_linux_flow_dump_next,
1631 dpif_linux_flow_dump_done,
1634 dpif_linux_recv_set,
1635 dpif_linux_queue_to_priority,
1637 dpif_linux_recv_wait,
1638 dpif_linux_recv_purge,
1642 dpif_linux_init(void)
1644 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1647 if (ovsthread_once_start(&once)) {
1648 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1649 &ovs_datapath_family);
1651 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1652 "The Open vSwitch kernel module is probably not loaded.",
1653 OVS_DATAPATH_FAMILY);
1656 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1659 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1662 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1663 &ovs_packet_family);
1666 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1667 &ovs_vport_mcgroup);
1670 ovsthread_once_done(&once);
1677 dpif_linux_is_internal_device(const char *name)
1679 struct dpif_linux_vport reply;
1683 error = dpif_linux_vport_get(name, &reply, &buf);
1686 } else if (error != ENODEV && error != ENOENT) {
1687 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1688 name, ovs_strerror(error));
1691 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1694 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1695 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1696 * positive errno value.
1698 * 'vport' will contain pointers into 'buf', so the caller should not free
1699 * 'buf' while 'vport' is still in use. */
1701 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1702 const struct ofpbuf *buf)
1704 static const struct nl_policy ovs_vport_policy[] = {
1705 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1706 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1707 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1708 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1709 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1711 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1714 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1715 struct ovs_header *ovs_header;
1716 struct nlmsghdr *nlmsg;
1717 struct genlmsghdr *genl;
1720 dpif_linux_vport_init(vport);
1722 ofpbuf_use_const(&b, buf->data, buf->size);
1723 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1724 genl = ofpbuf_try_pull(&b, sizeof *genl);
1725 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1726 if (!nlmsg || !genl || !ovs_header
1727 || nlmsg->nlmsg_type != ovs_vport_family
1728 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1729 ARRAY_SIZE(ovs_vport_policy))) {
1733 vport->cmd = genl->cmd;
1734 vport->dp_ifindex = ovs_header->dp_ifindex;
1735 vport->port_no = nl_attr_get_odp_port(a[OVS_VPORT_ATTR_PORT_NO]);
1736 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1737 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1738 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1739 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1741 if (a[OVS_VPORT_ATTR_STATS]) {
1742 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1744 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1745 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1746 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1751 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1752 * followed by Netlink attributes corresponding to 'vport'. */
1754 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1757 struct ovs_header *ovs_header;
1759 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1760 vport->cmd, OVS_VPORT_VERSION);
1762 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1763 ovs_header->dp_ifindex = vport->dp_ifindex;
1765 if (vport->port_no != ODPP_NONE) {
1766 nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1769 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1770 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1774 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1777 if (vport->upcall_pid) {
1778 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1782 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1783 vport->stats, sizeof *vport->stats);
1786 if (vport->options) {
1787 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1788 vport->options, vport->options_len);
1792 /* Clears 'vport' to "empty" values. */
1794 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1796 memset(vport, 0, sizeof *vport);
1797 vport->port_no = ODPP_NONE;
1800 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1801 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1802 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1803 * result of the command is expected to be an ovs_vport also, which is decoded
1804 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1805 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1807 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1808 struct dpif_linux_vport *reply,
1809 struct ofpbuf **bufp)
1811 struct ofpbuf *request_buf;
1814 ovs_assert((reply != NULL) == (bufp != NULL));
1816 error = dpif_linux_init();
1820 dpif_linux_vport_init(reply);
1825 request_buf = ofpbuf_new(1024);
1826 dpif_linux_vport_to_ofpbuf(request, request_buf);
1827 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
1828 ofpbuf_delete(request_buf);
1832 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1835 dpif_linux_vport_init(reply);
1836 ofpbuf_delete(*bufp);
1843 /* Obtains information about the kernel vport named 'name' and stores it into
1844 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1845 * longer needed ('reply' will contain pointers into '*bufp'). */
1847 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1848 struct ofpbuf **bufp)
1850 struct dpif_linux_vport request;
1852 dpif_linux_vport_init(&request);
1853 request.cmd = OVS_VPORT_CMD_GET;
1854 request.name = name;
1856 return dpif_linux_vport_transact(&request, reply, bufp);
1859 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1860 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1861 * positive errno value.
1863 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1864 * while 'dp' is still in use. */
1866 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1868 static const struct nl_policy ovs_datapath_policy[] = {
1869 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1870 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1872 [OVS_DP_ATTR_MEGAFLOW_STATS] = {
1873 NL_POLICY_FOR(struct ovs_dp_megaflow_stats),
1877 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1878 struct ovs_header *ovs_header;
1879 struct nlmsghdr *nlmsg;
1880 struct genlmsghdr *genl;
1883 dpif_linux_dp_init(dp);
1885 ofpbuf_use_const(&b, buf->data, buf->size);
1886 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1887 genl = ofpbuf_try_pull(&b, sizeof *genl);
1888 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1889 if (!nlmsg || !genl || !ovs_header
1890 || nlmsg->nlmsg_type != ovs_datapath_family
1891 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1892 ARRAY_SIZE(ovs_datapath_policy))) {
1896 dp->cmd = genl->cmd;
1897 dp->dp_ifindex = ovs_header->dp_ifindex;
1898 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1899 if (a[OVS_DP_ATTR_STATS]) {
1900 /* Can't use structure assignment because Netlink doesn't ensure
1901 * sufficient alignment for 64-bit members. */
1902 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1906 if (a[OVS_DP_ATTR_MEGAFLOW_STATS]) {
1907 /* Can't use structure assignment because Netlink doesn't ensure
1908 * sufficient alignment for 64-bit members. */
1909 memcpy(&dp->megaflow_stats, nl_attr_get(a[OVS_DP_ATTR_MEGAFLOW_STATS]),
1910 sizeof dp->megaflow_stats);
1916 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1918 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1920 struct ovs_header *ovs_header;
1922 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1923 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1924 OVS_DATAPATH_VERSION);
1926 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1927 ovs_header->dp_ifindex = dp->dp_ifindex;
1930 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1933 if (dp->upcall_pid) {
1934 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1937 if (dp->user_features) {
1938 nl_msg_put_u32(buf, OVS_DP_ATTR_USER_FEATURES, dp->user_features);
1941 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1944 /* Clears 'dp' to "empty" values. */
1946 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1948 memset(dp, 0, sizeof *dp);
1949 dp->megaflow_stats.n_masks = UINT32_MAX;
1950 dp->megaflow_stats.n_mask_hit = UINT64_MAX;
1954 dpif_linux_dp_dump_start(struct nl_dump *dump)
1956 struct dpif_linux_dp request;
1959 dpif_linux_dp_init(&request);
1960 request.cmd = OVS_DP_CMD_GET;
1962 buf = ofpbuf_new(1024);
1963 dpif_linux_dp_to_ofpbuf(&request, buf);
1964 nl_dump_start(dump, NETLINK_GENERIC, buf);
1968 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1969 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1970 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1971 * result of the command is expected to be of the same form, which is decoded
1972 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1973 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1975 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1976 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1978 struct ofpbuf *request_buf;
1981 ovs_assert((reply != NULL) == (bufp != NULL));
1983 request_buf = ofpbuf_new(1024);
1984 dpif_linux_dp_to_ofpbuf(request, request_buf);
1985 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
1986 ofpbuf_delete(request_buf);
1989 dpif_linux_dp_init(reply);
1991 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1994 ofpbuf_delete(*bufp);
2001 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
2002 * The caller must free '*bufp' when the reply is no longer needed ('reply'
2003 * will contain pointers into '*bufp'). */
2005 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
2006 struct ofpbuf **bufp)
2008 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
2009 struct dpif_linux_dp request;
2011 dpif_linux_dp_init(&request);
2012 request.cmd = OVS_DP_CMD_GET;
2013 request.dp_ifindex = dpif->dp_ifindex;
2015 return dpif_linux_dp_transact(&request, reply, bufp);
2018 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
2019 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
2020 * positive errno value.
2022 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
2023 * while 'flow' is still in use. */
2025 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
2026 const struct ofpbuf *buf)
2028 static const struct nl_policy ovs_flow_policy[] = {
2029 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
2030 [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true },
2031 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
2032 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
2034 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
2035 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
2036 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
2039 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
2040 struct ovs_header *ovs_header;
2041 struct nlmsghdr *nlmsg;
2042 struct genlmsghdr *genl;
2045 dpif_linux_flow_init(flow);
2047 ofpbuf_use_const(&b, buf->data, buf->size);
2048 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
2049 genl = ofpbuf_try_pull(&b, sizeof *genl);
2050 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
2051 if (!nlmsg || !genl || !ovs_header
2052 || nlmsg->nlmsg_type != ovs_flow_family
2053 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
2054 ARRAY_SIZE(ovs_flow_policy))) {
2058 flow->nlmsg_flags = nlmsg->nlmsg_flags;
2059 flow->dp_ifindex = ovs_header->dp_ifindex;
2060 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
2061 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
2063 if (a[OVS_FLOW_ATTR_MASK]) {
2064 flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]);
2065 flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]);
2067 if (a[OVS_FLOW_ATTR_ACTIONS]) {
2068 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
2069 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
2071 if (a[OVS_FLOW_ATTR_STATS]) {
2072 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
2074 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
2075 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
2077 if (a[OVS_FLOW_ATTR_USED]) {
2078 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
2083 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
2084 * followed by Netlink attributes corresponding to 'flow'. */
2086 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
2089 struct ovs_header *ovs_header;
2091 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
2092 NLM_F_REQUEST | flow->nlmsg_flags,
2093 flow->cmd, OVS_FLOW_VERSION);
2095 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
2096 ovs_header->dp_ifindex = flow->dp_ifindex;
2098 if (flow->key_len) {
2099 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
2102 if (flow->mask_len) {
2103 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_MASK, flow->mask, flow->mask_len);
2106 if (flow->actions || flow->actions_len) {
2107 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
2108 flow->actions, flow->actions_len);
2111 /* We never need to send these to the kernel. */
2112 ovs_assert(!flow->stats);
2113 ovs_assert(!flow->tcp_flags);
2114 ovs_assert(!flow->used);
2117 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
2121 /* Clears 'flow' to "empty" values. */
2123 dpif_linux_flow_init(struct dpif_linux_flow *flow)
2125 memset(flow, 0, sizeof *flow);
2128 /* Executes 'request' in the kernel datapath. If the command fails, returns a
2129 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
2130 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
2131 * result of the command is expected to be a flow also, which is decoded and
2132 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
2133 * is no longer needed ('reply' will contain pointers into '*bufp'). */
2135 dpif_linux_flow_transact(struct dpif_linux_flow *request,
2136 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
2138 struct ofpbuf *request_buf;
2141 ovs_assert((reply != NULL) == (bufp != NULL));
2144 request->nlmsg_flags |= NLM_F_ECHO;
2147 request_buf = ofpbuf_new(1024);
2148 dpif_linux_flow_to_ofpbuf(request, request_buf);
2149 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
2150 ofpbuf_delete(request_buf);
2154 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
2157 dpif_linux_flow_init(reply);
2158 ofpbuf_delete(*bufp);
2166 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
2167 struct dpif_flow_stats *stats)
2170 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
2171 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
2173 stats->n_packets = 0;
2176 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
2177 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
2180 /* Logs information about a packet that was recently lost in 'ch' (in
2183 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
2185 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
2186 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
2189 if (VLOG_DROP_WARN(&rl)) {
2194 if (ch->last_poll != LLONG_MIN) {
2195 ds_put_format(&s, " (last polled %lld ms ago)",
2196 time_msec() - ch->last_poll);
2199 VLOG_WARN("%s: lost packet on channel %"PRIdPTR"%s",
2200 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));