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 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
77 struct ovs_dp_megaflow_stats megaflow_stats;
78 /* OVS_DP_ATTR_MEGAFLOW_STATS.*/
81 static void dpif_linux_dp_init(struct dpif_linux_dp *);
82 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
83 const struct ofpbuf *);
84 static void dpif_linux_dp_dump_start(struct nl_dump *);
85 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
86 struct dpif_linux_dp *reply,
87 struct ofpbuf **bufp);
88 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
89 struct ofpbuf **bufp);
91 struct dpif_linux_flow {
92 /* Generic Netlink header. */
95 /* struct ovs_header. */
96 unsigned int nlmsg_flags;
101 * The 'stats' member points to 64-bit data that might only be aligned on
102 * 32-bit boundaries, so get_unaligned_u64() should be used to access its
105 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
106 * the Netlink version of the command, even if actions_len is zero. */
107 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
109 const struct nlattr *mask; /* OVS_FLOW_ATTR_MASK. */
111 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
113 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
114 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
115 const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
116 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
119 static void dpif_linux_flow_init(struct dpif_linux_flow *);
120 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
121 const struct ofpbuf *);
122 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
124 static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
125 struct dpif_linux_flow *reply,
126 struct ofpbuf **bufp);
127 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
128 struct dpif_flow_stats *);
130 /* One of the dpif channels between the kernel and userspace. */
131 struct dpif_channel {
132 struct nl_sock *sock; /* Netlink socket. */
133 long long int last_poll; /* Last time this channel was polled. */
136 static void report_loss(struct dpif *, struct dpif_channel *);
138 /* Datapath interface for the openvswitch Linux kernel module. */
143 /* Upcall messages. */
144 struct ovs_mutex upcall_lock;
145 int uc_array_size; /* Size of 'channels' and 'epoll_events'. */
146 struct dpif_channel *channels;
147 struct epoll_event *epoll_events;
148 int epoll_fd; /* epoll fd that includes channel socks. */
149 int n_events; /* Num events returned by epoll_wait(). */
150 int event_offset; /* Offset into 'epoll_events'. */
152 /* Change notification. */
153 struct nl_sock *port_notifier; /* vport multicast group subscriber. */
154 bool refresh_channels;
157 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
159 /* Generic Netlink family numbers for OVS.
161 * Initialized by dpif_linux_init(). */
162 static int ovs_datapath_family;
163 static int ovs_vport_family;
164 static int ovs_flow_family;
165 static int ovs_packet_family;
167 /* Generic Netlink multicast groups for OVS.
169 * Initialized by dpif_linux_init(). */
170 static unsigned int ovs_vport_mcgroup;
172 static int dpif_linux_init(void);
173 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
174 static uint32_t dpif_linux_port_get_pid(const struct dpif *,
176 static int dpif_linux_refresh_channels(struct dpif *);
178 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
180 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
181 const struct ofpbuf *);
183 static struct dpif_linux *
184 dpif_linux_cast(const struct dpif *dpif)
186 dpif_assert_class(dpif, &dpif_linux_class);
187 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
191 dpif_linux_enumerate(struct sset *all_dps)
197 error = dpif_linux_init();
202 dpif_linux_dp_dump_start(&dump);
203 while (nl_dump_next(&dump, &msg)) {
204 struct dpif_linux_dp dp;
206 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
207 sset_add(all_dps, dp.name);
210 return nl_dump_done(&dump);
214 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
215 bool create, struct dpif **dpifp)
217 struct dpif_linux_dp dp_request, dp;
222 error = dpif_linux_init();
227 /* Create or look up datapath. */
228 dpif_linux_dp_init(&dp_request);
230 dp_request.cmd = OVS_DP_CMD_NEW;
232 dp_request.upcall_pid = &upcall_pid;
234 dp_request.cmd = OVS_DP_CMD_GET;
236 dp_request.name = name;
237 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
242 error = open_dpif(&dp, dpifp);
248 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
250 struct dpif_linux *dpif;
252 dpif = xzalloc(sizeof *dpif);
253 dpif->port_notifier = NULL;
254 ovs_mutex_init(&dpif->upcall_lock);
257 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
258 dp->dp_ifindex, dp->dp_ifindex);
260 dpif->dp_ifindex = dp->dp_ifindex;
261 *dpifp = &dpif->dpif;
267 destroy_channels(struct dpif_linux *dpif)
271 if (dpif->epoll_fd < 0) {
275 for (i = 0; i < dpif->uc_array_size; i++ ) {
276 struct dpif_linux_vport vport_request;
277 struct dpif_channel *ch = &dpif->channels[i];
278 uint32_t upcall_pid = 0;
284 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
286 /* Turn off upcalls. */
287 dpif_linux_vport_init(&vport_request);
288 vport_request.cmd = OVS_VPORT_CMD_SET;
289 vport_request.dp_ifindex = dpif->dp_ifindex;
290 vport_request.port_no = u32_to_odp(i);
291 vport_request.upcall_pid = &upcall_pid;
292 dpif_linux_vport_transact(&vport_request, NULL, NULL);
294 nl_sock_destroy(ch->sock);
297 free(dpif->channels);
298 dpif->channels = NULL;
299 dpif->uc_array_size = 0;
301 free(dpif->epoll_events);
302 dpif->epoll_events = NULL;
303 dpif->n_events = dpif->event_offset = 0;
305 /* Don't close dpif->epoll_fd since that would cause other threads that
306 * call dpif_recv_wait(dpif) to wait on an arbitrary fd or a closed fd. */
310 add_channel(struct dpif_linux *dpif, odp_port_t port_no, struct nl_sock *sock)
312 struct epoll_event event;
313 uint32_t port_idx = odp_to_u32(port_no);
315 if (dpif->epoll_fd < 0) {
319 /* We assume that the datapath densely chooses port numbers, which
320 * can therefore be used as an index into an array of channels. */
321 if (port_idx >= dpif->uc_array_size) {
322 uint32_t new_size = port_idx + 1;
325 if (new_size > MAX_PORTS) {
326 VLOG_WARN_RL(&error_rl, "%s: datapath port %"PRIu32" too big",
327 dpif_name(&dpif->dpif), port_no);
331 dpif->channels = xrealloc(dpif->channels,
332 new_size * sizeof *dpif->channels);
333 for (i = dpif->uc_array_size; i < new_size; i++) {
334 dpif->channels[i].sock = NULL;
337 dpif->epoll_events = xrealloc(dpif->epoll_events,
338 new_size * sizeof *dpif->epoll_events);
339 dpif->uc_array_size = new_size;
342 memset(&event, 0, sizeof event);
343 event.events = EPOLLIN;
344 event.data.u32 = port_idx;
345 if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD, nl_sock_fd(sock),
350 nl_sock_destroy(dpif->channels[port_idx].sock);
351 dpif->channels[port_idx].sock = sock;
352 dpif->channels[port_idx].last_poll = LLONG_MIN;
358 del_channel(struct dpif_linux *dpif, odp_port_t port_no)
360 struct dpif_channel *ch;
361 uint32_t port_idx = odp_to_u32(port_no);
363 if (dpif->epoll_fd < 0 || port_idx >= dpif->uc_array_size) {
367 ch = &dpif->channels[port_idx];
372 epoll_ctl(dpif->epoll_fd, EPOLL_CTL_DEL, nl_sock_fd(ch->sock), NULL);
373 dpif->event_offset = dpif->n_events = 0;
375 nl_sock_destroy(ch->sock);
380 dpif_linux_close(struct dpif *dpif_)
382 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
384 nl_sock_destroy(dpif->port_notifier);
385 destroy_channels(dpif);
386 if (dpif->epoll_fd >= 0) {
387 close(dpif->epoll_fd);
389 ovs_mutex_destroy(&dpif->upcall_lock);
394 dpif_linux_destroy(struct dpif *dpif_)
396 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
397 struct dpif_linux_dp dp;
399 dpif_linux_dp_init(&dp);
400 dp.cmd = OVS_DP_CMD_DEL;
401 dp.dp_ifindex = dpif->dp_ifindex;
402 return dpif_linux_dp_transact(&dp, NULL, NULL);
406 dpif_linux_run(struct dpif *dpif_)
408 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
409 if (dpif->refresh_channels) {
410 dpif->refresh_channels = false;
411 dpif_linux_refresh_channels(dpif_);
416 dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
418 struct dpif_linux_dp dp;
422 error = dpif_linux_dp_get(dpif_, &dp, &buf);
424 stats->n_hit = dp.stats.n_hit;
425 stats->n_missed = dp.stats.n_missed;
426 stats->n_lost = dp.stats.n_lost;
427 stats->n_flows = dp.stats.n_flows;
428 stats->n_masks = dp.megaflow_stats.n_masks;
429 stats->n_mask_hit = dp.megaflow_stats.n_mask_hit;
436 get_vport_type(const struct dpif_linux_vport *vport)
438 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
440 switch (vport->type) {
441 case OVS_VPORT_TYPE_NETDEV:
444 case OVS_VPORT_TYPE_INTERNAL:
447 case OVS_VPORT_TYPE_GRE:
450 case OVS_VPORT_TYPE_GRE64:
453 case OVS_VPORT_TYPE_VXLAN:
456 case OVS_VPORT_TYPE_LISP:
459 case OVS_VPORT_TYPE_UNSPEC:
460 case __OVS_VPORT_TYPE_MAX:
464 VLOG_WARN_RL(&rl, "dp%d: port `%s' has unsupported type %u",
465 vport->dp_ifindex, vport->name, (unsigned int) vport->type);
469 static enum ovs_vport_type
470 netdev_to_ovs_vport_type(const struct netdev *netdev)
472 const char *type = netdev_get_type(netdev);
474 if (!strcmp(type, "tap") || !strcmp(type, "system")) {
475 return OVS_VPORT_TYPE_NETDEV;
476 } else if (!strcmp(type, "internal")) {
477 return OVS_VPORT_TYPE_INTERNAL;
478 } else if (strstr(type, "gre64")) {
479 return OVS_VPORT_TYPE_GRE64;
480 } else if (strstr(type, "gre")) {
481 return OVS_VPORT_TYPE_GRE;
482 } else if (!strcmp(type, "vxlan")) {
483 return OVS_VPORT_TYPE_VXLAN;
484 } else if (!strcmp(type, "lisp")) {
485 return OVS_VPORT_TYPE_LISP;
487 return OVS_VPORT_TYPE_UNSPEC;
492 dpif_linux_port_add__(struct dpif *dpif_, struct netdev *netdev,
493 odp_port_t *port_nop)
495 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
496 const struct netdev_tunnel_config *tnl_cfg;
497 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
498 const char *name = netdev_vport_get_dpif_port(netdev,
499 namebuf, sizeof namebuf);
500 const char *type = netdev_get_type(netdev);
501 struct dpif_linux_vport request, reply;
502 struct nl_sock *sock = NULL;
505 uint64_t options_stub[64 / 8];
506 struct ofpbuf options;
509 if (dpif->epoll_fd >= 0) {
510 error = nl_sock_create(NETLINK_GENERIC, &sock);
516 dpif_linux_vport_init(&request);
517 request.cmd = OVS_VPORT_CMD_NEW;
518 request.dp_ifindex = dpif->dp_ifindex;
519 request.type = netdev_to_ovs_vport_type(netdev);
520 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
521 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
522 "unsupported type `%s'",
523 dpif_name(dpif_), name, type);
524 nl_sock_destroy(sock);
529 if (request.type == OVS_VPORT_TYPE_NETDEV) {
530 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
533 tnl_cfg = netdev_get_tunnel_config(netdev);
534 if (tnl_cfg && tnl_cfg->dst_port != 0) {
535 ofpbuf_use_stack(&options, options_stub, sizeof options_stub);
536 nl_msg_put_u16(&options, OVS_TUNNEL_ATTR_DST_PORT,
537 ntohs(tnl_cfg->dst_port));
538 request.options = options.data;
539 request.options_len = options.size;
542 request.port_no = *port_nop;
543 upcall_pid = sock ? nl_sock_pid(sock) : 0;
544 request.upcall_pid = &upcall_pid;
546 error = dpif_linux_vport_transact(&request, &reply, &buf);
548 *port_nop = reply.port_no;
549 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
550 dpif_name(dpif_), reply.port_no, upcall_pid);
552 if (error == EBUSY && *port_nop != ODPP_NONE) {
553 VLOG_INFO("%s: requested port %"PRIu32" is in use",
554 dpif_name(dpif_), *port_nop);
556 nl_sock_destroy(sock);
563 error = add_channel(dpif, *port_nop, sock);
565 VLOG_INFO("%s: could not add channel for port %s",
566 dpif_name(dpif_), name);
568 /* Delete the port. */
569 dpif_linux_vport_init(&request);
570 request.cmd = OVS_VPORT_CMD_DEL;
571 request.dp_ifindex = dpif->dp_ifindex;
572 request.port_no = *port_nop;
573 dpif_linux_vport_transact(&request, NULL, NULL);
575 nl_sock_destroy(sock);
584 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
585 odp_port_t *port_nop)
587 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
590 ovs_mutex_lock(&dpif->upcall_lock);
591 error = dpif_linux_port_add__(dpif_, netdev, port_nop);
592 ovs_mutex_unlock(&dpif->upcall_lock);
598 dpif_linux_port_del__(struct dpif *dpif_, odp_port_t port_no)
600 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
601 struct dpif_linux_vport vport;
604 dpif_linux_vport_init(&vport);
605 vport.cmd = OVS_VPORT_CMD_DEL;
606 vport.dp_ifindex = dpif->dp_ifindex;
607 vport.port_no = port_no;
608 error = dpif_linux_vport_transact(&vport, NULL, NULL);
610 del_channel(dpif, port_no);
616 dpif_linux_port_del(struct dpif *dpif_, odp_port_t port_no)
618 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
621 ovs_mutex_lock(&dpif->upcall_lock);
622 error = dpif_linux_port_del__(dpif_, port_no);
623 ovs_mutex_unlock(&dpif->upcall_lock);
629 dpif_linux_port_query__(const struct dpif *dpif, odp_port_t port_no,
630 const char *port_name, struct dpif_port *dpif_port)
632 struct dpif_linux_vport request;
633 struct dpif_linux_vport reply;
637 dpif_linux_vport_init(&request);
638 request.cmd = OVS_VPORT_CMD_GET;
639 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
640 request.port_no = port_no;
641 request.name = port_name;
643 error = dpif_linux_vport_transact(&request, &reply, &buf);
645 if (reply.dp_ifindex != request.dp_ifindex) {
646 /* A query by name reported that 'port_name' is in some datapath
647 * other than 'dpif', but the caller wants to know about 'dpif'. */
649 } else if (dpif_port) {
650 dpif_port->name = xstrdup(reply.name);
651 dpif_port->type = xstrdup(get_vport_type(&reply));
652 dpif_port->port_no = reply.port_no;
660 dpif_linux_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
661 struct dpif_port *dpif_port)
663 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
667 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
668 struct dpif_port *dpif_port)
670 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
674 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
680 dpif_linux_port_get_pid(const struct dpif *dpif_, odp_port_t port_no)
682 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
683 uint32_t port_idx = odp_to_u32(port_no);
686 ovs_mutex_lock(&dpif->upcall_lock);
687 if (dpif->epoll_fd >= 0) {
688 /* The ODPP_NONE "reserved" port number uses the "ovs-system"'s
689 * channel, since it is not heavily loaded. */
690 uint32_t idx = port_idx >= dpif->uc_array_size ? 0 : port_idx;
691 const struct nl_sock *sock = dpif->channels[idx].sock;
692 pid = sock ? nl_sock_pid(sock) : 0;
694 ovs_mutex_unlock(&dpif->upcall_lock);
700 dpif_linux_flow_flush(struct dpif *dpif_)
702 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
703 struct dpif_linux_flow flow;
705 dpif_linux_flow_init(&flow);
706 flow.cmd = OVS_FLOW_CMD_DEL;
707 flow.dp_ifindex = dpif->dp_ifindex;
708 return dpif_linux_flow_transact(&flow, NULL, NULL);
711 struct dpif_linux_port_state {
716 dpif_linux_port_dump_start__(const struct dpif *dpif_, struct nl_dump *dump)
718 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
719 struct dpif_linux_vport request;
722 dpif_linux_vport_init(&request);
723 request.cmd = OVS_VPORT_CMD_GET;
724 request.dp_ifindex = dpif->dp_ifindex;
726 buf = ofpbuf_new(1024);
727 dpif_linux_vport_to_ofpbuf(&request, buf);
728 nl_dump_start(dump, NETLINK_GENERIC, buf);
733 dpif_linux_port_dump_start(const struct dpif *dpif, void **statep)
735 struct dpif_linux_port_state *state;
737 *statep = state = xmalloc(sizeof *state);
738 dpif_linux_port_dump_start__(dpif, &state->dump);
744 dpif_linux_port_dump_next__(const struct dpif *dpif_, struct nl_dump *dump,
745 struct dpif_linux_vport *vport)
747 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
751 if (!nl_dump_next(dump, &buf)) {
755 error = dpif_linux_vport_from_ofpbuf(vport, &buf);
757 VLOG_WARN_RL(&error_rl, "%s: failed to parse vport record (%s)",
758 dpif_name(&dpif->dpif), ovs_strerror(error));
764 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
765 struct dpif_port *dpif_port)
767 struct dpif_linux_port_state *state = state_;
768 struct dpif_linux_vport vport;
771 error = dpif_linux_port_dump_next__(dpif, &state->dump, &vport);
775 dpif_port->name = CONST_CAST(char *, vport.name);
776 dpif_port->type = CONST_CAST(char *, get_vport_type(&vport));
777 dpif_port->port_no = vport.port_no;
782 dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
784 struct dpif_linux_port_state *state = state_;
785 int error = nl_dump_done(&state->dump);
792 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
794 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
796 /* Lazily create the Netlink socket to listen for notifications. */
797 if (!dpif->port_notifier) {
798 struct nl_sock *sock;
801 error = nl_sock_create(NETLINK_GENERIC, &sock);
806 error = nl_sock_join_mcgroup(sock, ovs_vport_mcgroup);
808 nl_sock_destroy(sock);
811 dpif->port_notifier = sock;
813 /* We have no idea of the current state so report that everything
819 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
820 uint64_t buf_stub[4096 / 8];
824 ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
825 error = nl_sock_recv(dpif->port_notifier, &buf, false);
827 struct dpif_linux_vport vport;
829 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
831 if (vport.dp_ifindex == dpif->dp_ifindex
832 && (vport.cmd == OVS_VPORT_CMD_NEW
833 || vport.cmd == OVS_VPORT_CMD_DEL
834 || vport.cmd == OVS_VPORT_CMD_SET)) {
835 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
836 dpif->dpif.full_name, vport.name, vport.cmd);
837 if (vport.cmd == OVS_VPORT_CMD_DEL) {
838 dpif->refresh_channels = true;
840 *devnamep = xstrdup(vport.name);
845 } else if (error != EAGAIN) {
846 VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)",
847 ovs_strerror(error));
848 nl_sock_drain(dpif->port_notifier);
860 dpif_linux_port_poll_wait(const struct dpif *dpif_)
862 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
864 if (dpif->port_notifier) {
865 nl_sock_wait(dpif->port_notifier, POLLIN);
867 poll_immediate_wake();
872 dpif_linux_flow_get__(const struct dpif *dpif_,
873 const struct nlattr *key, size_t key_len,
874 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
876 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
877 struct dpif_linux_flow request;
879 dpif_linux_flow_init(&request);
880 request.cmd = OVS_FLOW_CMD_GET;
881 request.dp_ifindex = dpif->dp_ifindex;
883 request.key_len = key_len;
884 return dpif_linux_flow_transact(&request, reply, bufp);
888 dpif_linux_flow_get(const struct dpif *dpif_,
889 const struct nlattr *key, size_t key_len,
890 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
892 struct dpif_linux_flow reply;
896 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
899 dpif_linux_flow_get_stats(&reply, stats);
902 buf->data = CONST_CAST(struct nlattr *, reply.actions);
903 buf->size = reply.actions_len;
913 dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
914 struct dpif_linux_flow *request)
916 static const struct nlattr dummy_action;
918 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
920 dpif_linux_flow_init(request);
921 request->cmd = (put->flags & DPIF_FP_CREATE
922 ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
923 request->dp_ifindex = dpif->dp_ifindex;
924 request->key = put->key;
925 request->key_len = put->key_len;
926 request->mask = put->mask;
927 request->mask_len = put->mask_len;
928 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
929 request->actions = (put->actions
931 : CONST_CAST(struct nlattr *, &dummy_action));
932 request->actions_len = put->actions_len;
933 if (put->flags & DPIF_FP_ZERO_STATS) {
934 request->clear = true;
936 request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
940 dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
942 struct dpif_linux_flow request, reply;
946 dpif_linux_init_flow_put(dpif_, put, &request);
947 error = dpif_linux_flow_transact(&request,
948 put->stats ? &reply : NULL,
949 put->stats ? &buf : NULL);
950 if (!error && put->stats) {
951 dpif_linux_flow_get_stats(&reply, put->stats);
958 dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
959 struct dpif_linux_flow *request)
961 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
963 dpif_linux_flow_init(request);
964 request->cmd = OVS_FLOW_CMD_DEL;
965 request->dp_ifindex = dpif->dp_ifindex;
966 request->key = del->key;
967 request->key_len = del->key_len;
971 dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
973 struct dpif_linux_flow request, reply;
977 dpif_linux_init_flow_del(dpif_, del, &request);
978 error = dpif_linux_flow_transact(&request,
979 del->stats ? &reply : NULL,
980 del->stats ? &buf : NULL);
981 if (!error && del->stats) {
982 dpif_linux_flow_get_stats(&reply, del->stats);
988 struct dpif_linux_flow_state {
990 struct dpif_linux_flow flow;
991 struct dpif_flow_stats stats;
996 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
998 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
999 struct dpif_linux_flow_state *state;
1000 struct dpif_linux_flow request;
1003 *statep = state = xmalloc(sizeof *state);
1005 dpif_linux_flow_init(&request);
1006 request.cmd = OVS_FLOW_CMD_GET;
1007 request.dp_ifindex = dpif->dp_ifindex;
1009 buf = ofpbuf_new(1024);
1010 dpif_linux_flow_to_ofpbuf(&request, buf);
1011 nl_dump_start(&state->dump, NETLINK_GENERIC, buf);
1020 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
1021 const struct nlattr **key, size_t *key_len,
1022 const struct nlattr **mask, size_t *mask_len,
1023 const struct nlattr **actions, size_t *actions_len,
1024 const struct dpif_flow_stats **stats)
1026 struct dpif_linux_flow_state *state = state_;
1031 ofpbuf_delete(state->buf);
1034 if (!nl_dump_next(&state->dump, &buf)) {
1038 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
1043 if (actions && !state->flow.actions) {
1044 error = dpif_linux_flow_get__(dpif_, state->flow.key,
1045 state->flow.key_len,
1046 &state->flow, &state->buf);
1047 if (error == ENOENT) {
1048 VLOG_DBG("dumped flow disappeared on get");
1050 VLOG_WARN("error fetching dumped flow: %s",
1051 ovs_strerror(error));
1057 *actions = state->flow.actions;
1058 *actions_len = state->flow.actions_len;
1061 *key = state->flow.key;
1062 *key_len = state->flow.key_len;
1065 *mask = state->flow.mask;
1066 *mask_len = state->flow.mask ? state->flow.mask_len : 0;
1069 dpif_linux_flow_get_stats(&state->flow, &state->stats);
1070 *stats = &state->stats;
1076 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1078 struct dpif_linux_flow_state *state = state_;
1079 int error = nl_dump_done(&state->dump);
1080 ofpbuf_delete(state->buf);
1086 dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
1089 struct ovs_header *k_exec;
1091 ofpbuf_prealloc_tailroom(buf, (64
1092 + d_exec->packet->size
1094 + d_exec->actions_len));
1096 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
1097 OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
1099 k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
1100 k_exec->dp_ifindex = dp_ifindex;
1102 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
1103 d_exec->packet->data, d_exec->packet->size);
1104 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
1105 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
1106 d_exec->actions, d_exec->actions_len);
1110 dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
1112 uint64_t request_stub[1024 / 8];
1113 struct ofpbuf request;
1116 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
1117 dpif_linux_encode_execute(dp_ifindex, execute, &request);
1118 error = nl_transact(NETLINK_GENERIC, &request, NULL);
1119 ofpbuf_uninit(&request);
1125 dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
1127 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1129 return dpif_linux_execute__(dpif->dp_ifindex, execute);
1135 dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
1137 const struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1140 struct nl_transaction txn;
1142 struct ofpbuf request;
1143 uint64_t request_stub[1024 / 8];
1145 struct ofpbuf reply;
1146 uint64_t reply_stub[1024 / 8];
1149 struct nl_transaction *txnsp[MAX_OPS];
1152 ovs_assert(n_ops <= MAX_OPS);
1153 for (i = 0; i < n_ops; i++) {
1154 struct op_auxdata *aux = &auxes[i];
1155 struct dpif_op *op = ops[i];
1156 struct dpif_flow_put *put;
1157 struct dpif_flow_del *del;
1158 struct dpif_execute *execute;
1159 struct dpif_linux_flow flow;
1161 ofpbuf_use_stub(&aux->request,
1162 aux->request_stub, sizeof aux->request_stub);
1163 aux->txn.request = &aux->request;
1165 ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
1166 aux->txn.reply = NULL;
1169 case DPIF_OP_FLOW_PUT:
1170 put = &op->u.flow_put;
1171 dpif_linux_init_flow_put(dpif_, put, &flow);
1173 flow.nlmsg_flags |= NLM_F_ECHO;
1174 aux->txn.reply = &aux->reply;
1176 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1179 case DPIF_OP_FLOW_DEL:
1180 del = &op->u.flow_del;
1181 dpif_linux_init_flow_del(dpif_, del, &flow);
1183 flow.nlmsg_flags |= NLM_F_ECHO;
1184 aux->txn.reply = &aux->reply;
1186 dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
1189 case DPIF_OP_EXECUTE:
1190 execute = &op->u.execute;
1191 dpif_linux_encode_execute(dpif->dp_ifindex, execute,
1200 for (i = 0; i < n_ops; i++) {
1201 txnsp[i] = &auxes[i].txn;
1203 nl_transact_multiple(NETLINK_GENERIC, txnsp, n_ops);
1205 for (i = 0; i < n_ops; i++) {
1206 struct op_auxdata *aux = &auxes[i];
1207 struct nl_transaction *txn = &auxes[i].txn;
1208 struct dpif_op *op = ops[i];
1209 struct dpif_flow_put *put;
1210 struct dpif_flow_del *del;
1212 op->error = txn->error;
1215 case DPIF_OP_FLOW_PUT:
1216 put = &op->u.flow_put;
1219 struct dpif_linux_flow reply;
1221 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1224 dpif_linux_flow_get_stats(&reply, put->stats);
1229 memset(put->stats, 0, sizeof *put->stats);
1234 case DPIF_OP_FLOW_DEL:
1235 del = &op->u.flow_del;
1238 struct dpif_linux_flow reply;
1240 op->error = dpif_linux_flow_from_ofpbuf(&reply,
1243 dpif_linux_flow_get_stats(&reply, del->stats);
1248 memset(del->stats, 0, sizeof *del->stats);
1253 case DPIF_OP_EXECUTE:
1260 ofpbuf_uninit(&aux->request);
1261 ofpbuf_uninit(&aux->reply);
1266 dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1269 size_t chunk = MIN(n_ops, MAX_OPS);
1270 dpif_linux_operate__(dpif, ops, chunk);
1276 /* Synchronizes 'dpif->channels' with the set of vports currently in 'dpif' in
1277 * the kernel, by adding a new channel for any kernel vport that lacks one and
1278 * deleting any channels that have no backing kernel vports. */
1280 dpif_linux_refresh_channels(struct dpif *dpif_)
1282 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1283 unsigned long int *keep_channels;
1284 struct dpif_linux_vport vport;
1285 size_t keep_channels_nbits;
1286 struct nl_dump dump;
1290 /* To start with, we need an epoll fd. */
1291 if (dpif->epoll_fd < 0) {
1292 dpif->epoll_fd = epoll_create(10);
1293 if (dpif->epoll_fd < 0) {
1298 keep_channels_nbits = dpif->uc_array_size;
1299 keep_channels = bitmap_allocate(keep_channels_nbits);
1301 dpif->n_events = dpif->event_offset = 0;
1303 dpif_linux_port_dump_start__(dpif_, &dump);
1304 while (!dpif_linux_port_dump_next__(dpif_, &dump, &vport)) {
1305 uint32_t port_no = odp_to_u32(vport.port_no);
1306 struct nl_sock *sock = (port_no < dpif->uc_array_size
1307 ? dpif->channels[port_no].sock
1309 bool new_sock = !sock;
1313 error = nl_sock_create(NETLINK_GENERIC, &sock);
1320 /* Configure the vport to deliver misses to 'sock'. */
1321 if (!vport.upcall_pid || *vport.upcall_pid != nl_sock_pid(sock)) {
1322 uint32_t upcall_pid = nl_sock_pid(sock);
1323 struct dpif_linux_vport vport_request;
1325 dpif_linux_vport_init(&vport_request);
1326 vport_request.cmd = OVS_VPORT_CMD_SET;
1327 vport_request.dp_ifindex = dpif->dp_ifindex;
1328 vport_request.port_no = vport.port_no;
1329 vport_request.upcall_pid = &upcall_pid;
1330 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
1332 VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
1333 dpif_name(&dpif->dpif), vport_request.port_no,
1336 VLOG_WARN_RL(&error_rl,
1337 "%s: failed to set upcall pid on port: %s",
1338 dpif_name(&dpif->dpif), ovs_strerror(error));
1340 if (error != ENODEV && error != ENOENT) {
1343 /* The vport isn't really there, even though the dump says
1344 * it is. Probably we just hit a race after a port
1352 error = add_channel(dpif, vport.port_no, sock);
1354 VLOG_INFO("%s: could not add channel for port %s",
1355 dpif_name(dpif_), vport.name);
1361 if (port_no < keep_channels_nbits) {
1362 bitmap_set1(keep_channels, port_no);
1367 nl_sock_destroy(sock);
1369 nl_dump_done(&dump);
1371 /* Discard any saved channels that we didn't reuse. */
1372 for (i = 0; i < keep_channels_nbits; i++) {
1373 if (!bitmap_is_set(keep_channels, i)) {
1374 nl_sock_destroy(dpif->channels[i].sock);
1375 dpif->channels[i].sock = NULL;
1378 free(keep_channels);
1384 dpif_linux_recv_set__(struct dpif *dpif_, bool enable)
1386 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1388 if ((dpif->epoll_fd >= 0) == enable) {
1390 } else if (!enable) {
1391 destroy_channels(dpif);
1394 return dpif_linux_refresh_channels(dpif_);
1399 dpif_linux_recv_set(struct dpif *dpif_, bool enable)
1401 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1404 ovs_mutex_lock(&dpif->upcall_lock);
1405 error = dpif_linux_recv_set__(dpif_, enable);
1406 ovs_mutex_unlock(&dpif->upcall_lock);
1412 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1413 uint32_t queue_id, uint32_t *priority)
1415 if (queue_id < 0xf000) {
1416 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1424 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1427 static const struct nl_policy ovs_packet_policy[] = {
1428 /* Always present. */
1429 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1430 .min_len = ETH_HEADER_LEN },
1431 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1433 /* OVS_PACKET_CMD_ACTION only. */
1434 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_UNSPEC, .optional = true },
1437 struct ovs_header *ovs_header;
1438 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1439 struct nlmsghdr *nlmsg;
1440 struct genlmsghdr *genl;
1444 ofpbuf_use_const(&b, buf->data, buf->size);
1446 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1447 genl = ofpbuf_try_pull(&b, sizeof *genl);
1448 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1449 if (!nlmsg || !genl || !ovs_header
1450 || nlmsg->nlmsg_type != ovs_packet_family
1451 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1452 ARRAY_SIZE(ovs_packet_policy))) {
1456 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1457 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1463 /* (Re)set ALL fields of '*upcall' on successful return. */
1464 upcall->type = type;
1465 upcall->key = CONST_CAST(struct nlattr *,
1466 nl_attr_get(a[OVS_PACKET_ATTR_KEY]));
1467 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1468 upcall->userdata = a[OVS_PACKET_ATTR_USERDATA];
1470 /* Allow overwriting the netlink attribute header without reallocating. */
1471 ofpbuf_use_stub(&upcall->packet,
1472 CONST_CAST(struct nlattr *,
1473 nl_attr_get(a[OVS_PACKET_ATTR_PACKET])) - 1,
1474 nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]) +
1475 sizeof(struct nlattr));
1476 upcall->packet.data = (char *)upcall->packet.data + sizeof(struct nlattr);
1477 upcall->packet.size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1479 *dp_ifindex = ovs_header->dp_ifindex;
1485 dpif_linux_recv__(struct dpif *dpif_, struct dpif_upcall *upcall,
1488 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1491 if (dpif->epoll_fd < 0) {
1495 if (dpif->event_offset >= dpif->n_events) {
1498 dpif->event_offset = dpif->n_events = 0;
1501 retval = epoll_wait(dpif->epoll_fd, dpif->epoll_events,
1502 dpif->uc_array_size, 0);
1503 } while (retval < 0 && errno == EINTR);
1505 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1506 VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", ovs_strerror(errno));
1507 } else if (retval > 0) {
1508 dpif->n_events = retval;
1512 while (dpif->event_offset < dpif->n_events) {
1513 int idx = dpif->epoll_events[dpif->event_offset].data.u32;
1514 struct dpif_channel *ch = &dpif->channels[idx];
1516 dpif->event_offset++;
1522 if (++read_tries > 50) {
1526 error = nl_sock_recv(ch->sock, buf, false);
1527 if (error == ENOBUFS) {
1528 /* ENOBUFS typically means that we've received so many
1529 * packets that the buffer overflowed. Try again
1530 * immediately because there's almost certainly a packet
1531 * waiting for us. */
1532 report_loss(dpif_, ch);
1536 ch->last_poll = time_msec();
1538 if (error == EAGAIN) {
1544 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1545 if (!error && dp_ifindex == dpif->dp_ifindex) {
1557 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
1560 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1563 ovs_mutex_lock(&dpif->upcall_lock);
1564 error = dpif_linux_recv__(dpif_, upcall, buf);
1565 ovs_mutex_unlock(&dpif->upcall_lock);
1571 dpif_linux_recv_wait(struct dpif *dpif_)
1573 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1575 ovs_mutex_lock(&dpif->upcall_lock);
1576 if (dpif->epoll_fd >= 0) {
1577 poll_fd_wait(dpif->epoll_fd, POLLIN);
1579 ovs_mutex_unlock(&dpif->upcall_lock);
1583 dpif_linux_recv_purge(struct dpif *dpif_)
1585 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1587 ovs_mutex_lock(&dpif->upcall_lock);
1588 if (dpif->epoll_fd >= 0) {
1589 struct dpif_channel *ch;
1591 for (ch = dpif->channels; ch < &dpif->channels[dpif->uc_array_size];
1594 nl_sock_drain(ch->sock);
1598 ovs_mutex_unlock(&dpif->upcall_lock);
1601 const struct dpif_class dpif_linux_class = {
1603 dpif_linux_enumerate,
1610 dpif_linux_get_stats,
1611 dpif_linux_port_add,
1612 dpif_linux_port_del,
1613 dpif_linux_port_query_by_number,
1614 dpif_linux_port_query_by_name,
1615 dpif_linux_get_max_ports,
1616 dpif_linux_port_get_pid,
1617 dpif_linux_port_dump_start,
1618 dpif_linux_port_dump_next,
1619 dpif_linux_port_dump_done,
1620 dpif_linux_port_poll,
1621 dpif_linux_port_poll_wait,
1622 dpif_linux_flow_get,
1623 dpif_linux_flow_put,
1624 dpif_linux_flow_del,
1625 dpif_linux_flow_flush,
1626 dpif_linux_flow_dump_start,
1627 dpif_linux_flow_dump_next,
1628 dpif_linux_flow_dump_done,
1631 dpif_linux_recv_set,
1632 dpif_linux_queue_to_priority,
1634 dpif_linux_recv_wait,
1635 dpif_linux_recv_purge,
1639 dpif_linux_init(void)
1641 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1644 if (ovsthread_once_start(&once)) {
1645 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1646 &ovs_datapath_family);
1648 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1649 "The Open vSwitch kernel module is probably not loaded.",
1650 OVS_DATAPATH_FAMILY);
1653 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1656 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1659 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1660 &ovs_packet_family);
1663 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1664 &ovs_vport_mcgroup);
1667 ovsthread_once_done(&once);
1674 dpif_linux_is_internal_device(const char *name)
1676 struct dpif_linux_vport reply;
1680 error = dpif_linux_vport_get(name, &reply, &buf);
1683 } else if (error != ENODEV && error != ENOENT) {
1684 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1685 name, ovs_strerror(error));
1688 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1691 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1692 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1693 * positive errno value.
1695 * 'vport' will contain pointers into 'buf', so the caller should not free
1696 * 'buf' while 'vport' is still in use. */
1698 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1699 const struct ofpbuf *buf)
1701 static const struct nl_policy ovs_vport_policy[] = {
1702 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1703 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1704 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1705 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1706 [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
1708 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1711 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1712 struct ovs_header *ovs_header;
1713 struct nlmsghdr *nlmsg;
1714 struct genlmsghdr *genl;
1717 dpif_linux_vport_init(vport);
1719 ofpbuf_use_const(&b, buf->data, buf->size);
1720 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1721 genl = ofpbuf_try_pull(&b, sizeof *genl);
1722 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1723 if (!nlmsg || !genl || !ovs_header
1724 || nlmsg->nlmsg_type != ovs_vport_family
1725 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1726 ARRAY_SIZE(ovs_vport_policy))) {
1730 vport->cmd = genl->cmd;
1731 vport->dp_ifindex = ovs_header->dp_ifindex;
1732 vport->port_no = nl_attr_get_odp_port(a[OVS_VPORT_ATTR_PORT_NO]);
1733 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1734 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1735 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1736 vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
1738 if (a[OVS_VPORT_ATTR_STATS]) {
1739 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1741 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1742 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1743 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1748 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1749 * followed by Netlink attributes corresponding to 'vport'. */
1751 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1754 struct ovs_header *ovs_header;
1756 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1757 vport->cmd, OVS_VPORT_VERSION);
1759 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1760 ovs_header->dp_ifindex = vport->dp_ifindex;
1762 if (vport->port_no != ODPP_NONE) {
1763 nl_msg_put_odp_port(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1766 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1767 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1771 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1774 if (vport->upcall_pid) {
1775 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
1779 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1780 vport->stats, sizeof *vport->stats);
1783 if (vport->options) {
1784 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1785 vport->options, vport->options_len);
1789 /* Clears 'vport' to "empty" values. */
1791 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1793 memset(vport, 0, sizeof *vport);
1794 vport->port_no = ODPP_NONE;
1797 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1798 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1799 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1800 * result of the command is expected to be an ovs_vport also, which is decoded
1801 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1802 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1804 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1805 struct dpif_linux_vport *reply,
1806 struct ofpbuf **bufp)
1808 struct ofpbuf *request_buf;
1811 ovs_assert((reply != NULL) == (bufp != NULL));
1813 error = dpif_linux_init();
1817 dpif_linux_vport_init(reply);
1822 request_buf = ofpbuf_new(1024);
1823 dpif_linux_vport_to_ofpbuf(request, request_buf);
1824 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
1825 ofpbuf_delete(request_buf);
1829 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1832 dpif_linux_vport_init(reply);
1833 ofpbuf_delete(*bufp);
1840 /* Obtains information about the kernel vport named 'name' and stores it into
1841 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1842 * longer needed ('reply' will contain pointers into '*bufp'). */
1844 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1845 struct ofpbuf **bufp)
1847 struct dpif_linux_vport request;
1849 dpif_linux_vport_init(&request);
1850 request.cmd = OVS_VPORT_CMD_GET;
1851 request.name = name;
1853 return dpif_linux_vport_transact(&request, reply, bufp);
1856 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1857 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1858 * positive errno value.
1860 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1861 * while 'dp' is still in use. */
1863 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1865 static const struct nl_policy ovs_datapath_policy[] = {
1866 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1867 [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
1869 [OVS_DP_ATTR_MEGAFLOW_STATS] = {
1870 NL_POLICY_FOR(struct ovs_dp_megaflow_stats),
1874 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1875 struct ovs_header *ovs_header;
1876 struct nlmsghdr *nlmsg;
1877 struct genlmsghdr *genl;
1880 dpif_linux_dp_init(dp);
1882 ofpbuf_use_const(&b, buf->data, buf->size);
1883 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1884 genl = ofpbuf_try_pull(&b, sizeof *genl);
1885 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1886 if (!nlmsg || !genl || !ovs_header
1887 || nlmsg->nlmsg_type != ovs_datapath_family
1888 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1889 ARRAY_SIZE(ovs_datapath_policy))) {
1893 dp->cmd = genl->cmd;
1894 dp->dp_ifindex = ovs_header->dp_ifindex;
1895 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1896 if (a[OVS_DP_ATTR_STATS]) {
1897 /* Can't use structure assignment because Netlink doesn't ensure
1898 * sufficient alignment for 64-bit members. */
1899 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1903 if (a[OVS_DP_ATTR_MEGAFLOW_STATS]) {
1904 /* Can't use structure assignment because Netlink doesn't ensure
1905 * sufficient alignment for 64-bit members. */
1906 memcpy(&dp->megaflow_stats, nl_attr_get(a[OVS_DP_ATTR_MEGAFLOW_STATS]),
1907 sizeof dp->megaflow_stats);
1913 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1915 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1917 struct ovs_header *ovs_header;
1919 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1920 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
1921 OVS_DATAPATH_VERSION);
1923 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1924 ovs_header->dp_ifindex = dp->dp_ifindex;
1927 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1930 if (dp->upcall_pid) {
1931 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
1934 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1937 /* Clears 'dp' to "empty" values. */
1939 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1941 memset(dp, 0, sizeof *dp);
1942 dp->megaflow_stats.n_masks = UINT32_MAX;
1943 dp->megaflow_stats.n_mask_hit = UINT64_MAX;
1947 dpif_linux_dp_dump_start(struct nl_dump *dump)
1949 struct dpif_linux_dp request;
1952 dpif_linux_dp_init(&request);
1953 request.cmd = OVS_DP_CMD_GET;
1955 buf = ofpbuf_new(1024);
1956 dpif_linux_dp_to_ofpbuf(&request, buf);
1957 nl_dump_start(dump, NETLINK_GENERIC, buf);
1961 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1962 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1963 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1964 * result of the command is expected to be of the same form, which is decoded
1965 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1966 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1968 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1969 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1971 struct ofpbuf *request_buf;
1974 ovs_assert((reply != NULL) == (bufp != NULL));
1976 request_buf = ofpbuf_new(1024);
1977 dpif_linux_dp_to_ofpbuf(request, request_buf);
1978 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
1979 ofpbuf_delete(request_buf);
1982 dpif_linux_dp_init(reply);
1984 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1987 ofpbuf_delete(*bufp);
1994 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1995 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1996 * will contain pointers into '*bufp'). */
1998 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1999 struct ofpbuf **bufp)
2001 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
2002 struct dpif_linux_dp request;
2004 dpif_linux_dp_init(&request);
2005 request.cmd = OVS_DP_CMD_GET;
2006 request.dp_ifindex = dpif->dp_ifindex;
2008 return dpif_linux_dp_transact(&request, reply, bufp);
2011 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
2012 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
2013 * positive errno value.
2015 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
2016 * while 'flow' is still in use. */
2018 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
2019 const struct ofpbuf *buf)
2021 static const struct nl_policy ovs_flow_policy[] = {
2022 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
2023 [OVS_FLOW_ATTR_MASK] = { .type = NL_A_NESTED, .optional = true },
2024 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
2025 [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
2027 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
2028 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
2029 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
2032 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
2033 struct ovs_header *ovs_header;
2034 struct nlmsghdr *nlmsg;
2035 struct genlmsghdr *genl;
2038 dpif_linux_flow_init(flow);
2040 ofpbuf_use_const(&b, buf->data, buf->size);
2041 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
2042 genl = ofpbuf_try_pull(&b, sizeof *genl);
2043 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
2044 if (!nlmsg || !genl || !ovs_header
2045 || nlmsg->nlmsg_type != ovs_flow_family
2046 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
2047 ARRAY_SIZE(ovs_flow_policy))) {
2051 flow->nlmsg_flags = nlmsg->nlmsg_flags;
2052 flow->dp_ifindex = ovs_header->dp_ifindex;
2053 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
2054 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
2056 if (a[OVS_FLOW_ATTR_MASK]) {
2057 flow->mask = nl_attr_get(a[OVS_FLOW_ATTR_MASK]);
2058 flow->mask_len = nl_attr_get_size(a[OVS_FLOW_ATTR_MASK]);
2060 if (a[OVS_FLOW_ATTR_ACTIONS]) {
2061 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
2062 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
2064 if (a[OVS_FLOW_ATTR_STATS]) {
2065 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
2067 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
2068 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
2070 if (a[OVS_FLOW_ATTR_USED]) {
2071 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
2076 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
2077 * followed by Netlink attributes corresponding to 'flow'. */
2079 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
2082 struct ovs_header *ovs_header;
2084 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
2085 NLM_F_REQUEST | flow->nlmsg_flags,
2086 flow->cmd, OVS_FLOW_VERSION);
2088 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
2089 ovs_header->dp_ifindex = flow->dp_ifindex;
2091 if (flow->key_len) {
2092 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
2095 if (flow->mask_len) {
2096 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_MASK, flow->mask, flow->mask_len);
2099 if (flow->actions || flow->actions_len) {
2100 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
2101 flow->actions, flow->actions_len);
2104 /* We never need to send these to the kernel. */
2105 ovs_assert(!flow->stats);
2106 ovs_assert(!flow->tcp_flags);
2107 ovs_assert(!flow->used);
2110 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
2114 /* Clears 'flow' to "empty" values. */
2116 dpif_linux_flow_init(struct dpif_linux_flow *flow)
2118 memset(flow, 0, sizeof *flow);
2121 /* Executes 'request' in the kernel datapath. If the command fails, returns a
2122 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
2123 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
2124 * result of the command is expected to be a flow also, which is decoded and
2125 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
2126 * is no longer needed ('reply' will contain pointers into '*bufp'). */
2128 dpif_linux_flow_transact(struct dpif_linux_flow *request,
2129 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
2131 struct ofpbuf *request_buf;
2134 ovs_assert((reply != NULL) == (bufp != NULL));
2137 request->nlmsg_flags |= NLM_F_ECHO;
2140 request_buf = ofpbuf_new(1024);
2141 dpif_linux_flow_to_ofpbuf(request, request_buf);
2142 error = nl_transact(NETLINK_GENERIC, request_buf, bufp);
2143 ofpbuf_delete(request_buf);
2147 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
2150 dpif_linux_flow_init(reply);
2151 ofpbuf_delete(*bufp);
2159 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
2160 struct dpif_flow_stats *stats)
2163 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
2164 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
2166 stats->n_packets = 0;
2169 stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
2170 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
2173 /* Logs information about a packet that was recently lost in 'ch' (in
2176 report_loss(struct dpif *dpif_, struct dpif_channel *ch)
2178 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
2179 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
2182 if (VLOG_DROP_WARN(&rl)) {
2187 if (ch->last_poll != LLONG_MIN) {
2188 ds_put_format(&s, " (last polled %lld ms ago)",
2189 time_msec() - ch->last_poll);
2192 VLOG_WARN("%s: lost packet on channel %"PRIdPTR"%s",
2193 dpif_name(dpif_), ch - dpif->channels, ds_cstr(&s));