2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
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"
27 #include <linux/types.h>
28 #include <linux/pkt_sched.h>
29 #include <linux/rtnetlink.h>
30 #include <linux/sockios.h>
36 #include "dpif-provider.h"
37 #include "dynamic-string.h"
40 #include "netdev-linux.h"
41 #include "netdev-vport.h"
42 #include "netlink-notifier.h"
43 #include "netlink-socket.h"
47 #include "openvswitch/datapath-compat.h"
48 #include "openvswitch/tunnel.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(dpif_linux);
60 enum { LRU_MAX_PORTS = 1024 };
61 enum { LRU_MASK = LRU_MAX_PORTS - 1};
62 BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
64 enum { N_UPCALL_SOCKS = 16 };
65 BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
67 /* This ethtool flag was introduced in Linux 2.6.24, so it might be
68 * missing if we have old headers. */
69 #define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
71 struct dpif_linux_dp {
72 /* Generic Netlink header. */
75 /* struct ovs_header. */
79 const char *name; /* OVS_DP_ATTR_NAME. */
80 uint32_t upcall_pid; /* OVS_DP_UPCALL_PID. */
81 struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
82 enum ovs_frag_handling ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */
83 const uint32_t *sampling; /* OVS_DP_ATTR_SAMPLING. */
86 static void dpif_linux_dp_init(struct dpif_linux_dp *);
87 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
88 const struct ofpbuf *);
89 static void dpif_linux_dp_dump_start(struct nl_dump *);
90 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
91 struct dpif_linux_dp *reply,
92 struct ofpbuf **bufp);
93 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
94 struct ofpbuf **bufp);
96 struct dpif_linux_flow {
97 /* Generic Netlink header. */
100 /* struct ovs_header. */
101 unsigned int nlmsg_flags;
106 * The 'stats' and 'used' members point to 64-bit data that might only be
107 * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
108 * access their values.
110 * If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
111 * the Netlink version of the command, even if actions_len is zero. */
112 const struct nlattr *key; /* OVS_FLOW_ATTR_KEY. */
114 const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
116 uint32_t upcall_pid; /* OVS_FLOW_ATTR_UPCALL_PID. */
117 const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
118 const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
119 const uint64_t *used; /* OVS_FLOW_ATTR_USED. */
120 bool clear; /* OVS_FLOW_ATTR_CLEAR. */
123 static void dpif_linux_flow_init(struct dpif_linux_flow *);
124 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
125 const struct ofpbuf *);
126 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
128 static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
129 struct dpif_linux_flow *reply,
130 struct ofpbuf **bufp);
131 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
132 struct dpif_flow_stats *);
134 /* Datapath interface for the openvswitch Linux kernel module. */
139 /* Upcall messages. */
140 struct nl_sock *upcall_socks[N_UPCALL_SOCKS];
141 int last_read_upcall;
142 unsigned int listen_mask;
144 /* Change notification. */
145 struct sset changed_ports; /* Ports that have changed. */
146 struct nln_notifier *port_notifier;
149 /* Queue of unused ports. */
150 unsigned long *lru_bitmap;
151 uint16_t lru_ports[LRU_MAX_PORTS];
156 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
158 /* Generic Netlink family numbers for OVS. */
159 static int ovs_datapath_family;
160 static int ovs_vport_family;
161 static int ovs_flow_family;
162 static int ovs_packet_family;
164 /* Generic Netlink socket. */
165 static struct nl_sock *genl_sock;
166 static struct nln *nln = NULL;
168 static int dpif_linux_init(void);
169 static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
170 static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
171 static void dpif_linux_port_changed(const void *vport, void *dpif);
172 static uint32_t get_upcall_pid_port(struct dpif_linux *, uint32_t port);
173 static uint32_t get_upcall_pid_flow(struct dpif_linux *,
174 const struct nlattr *key, size_t key_len);
176 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
178 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
179 const struct ofpbuf *);
181 static struct dpif_linux *
182 dpif_linux_cast(const struct dpif *dpif)
184 dpif_assert_class(dpif, &dpif_linux_class);
185 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
189 dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
191 if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
192 bitmap_set1(dp->lru_bitmap, port);
193 dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
198 dpif_linux_pop_port(struct dpif_linux *dp)
202 if (dp->lru_head == dp->lru_tail) {
206 port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
207 bitmap_set0(dp->lru_bitmap, port);
212 dpif_linux_enumerate(struct sset *all_dps)
218 error = dpif_linux_init();
223 dpif_linux_dp_dump_start(&dump);
224 while (nl_dump_next(&dump, &msg)) {
225 struct dpif_linux_dp dp;
227 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
228 sset_add(all_dps, dp.name);
231 return nl_dump_done(&dump);
235 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
236 bool create, struct dpif **dpifp)
238 struct dpif_linux_dp dp_request, dp;
242 error = dpif_linux_init();
247 /* Create or look up datapath. */
248 dpif_linux_dp_init(&dp_request);
249 dp_request.cmd = create ? OVS_DP_CMD_NEW : OVS_DP_CMD_GET;
250 dp_request.name = name;
251 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
256 open_dpif(&dp, dpifp);
262 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
264 struct dpif_linux *dpif;
267 dpif = xzalloc(sizeof *dpif);
268 dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
271 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
272 dp->dp_ifindex, dp->dp_ifindex);
274 dpif->dp_ifindex = dp->dp_ifindex;
275 sset_init(&dpif->changed_ports);
276 *dpifp = &dpif->dpif;
278 dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
279 bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL);
280 for (i = 1; i < LRU_MAX_PORTS; i++) {
281 dpif_linux_push_port(dpif, i);
286 destroy_upcall_socks(struct dpif_linux *dpif)
290 for (i = 0; i < N_UPCALL_SOCKS; i++) {
291 nl_sock_destroy(dpif->upcall_socks[i]);
292 dpif->upcall_socks[i] = NULL;
297 dpif_linux_close(struct dpif *dpif_)
299 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
301 nln_notifier_destroy(dpif->port_notifier);
302 destroy_upcall_socks(dpif);
303 sset_destroy(&dpif->changed_ports);
304 free(dpif->lru_bitmap);
309 dpif_linux_destroy(struct dpif *dpif_)
311 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
312 struct dpif_linux_dp dp;
314 dpif_linux_dp_init(&dp);
315 dp.cmd = OVS_DP_CMD_DEL;
316 dp.dp_ifindex = dpif->dp_ifindex;
317 return dpif_linux_dp_transact(&dp, NULL, NULL);
321 dpif_linux_run(struct dpif *dpif OVS_UNUSED)
329 dpif_linux_wait(struct dpif *dpif OVS_UNUSED)
337 dpif_linux_get_stats(const struct dpif *dpif_, struct ovs_dp_stats *stats)
339 struct dpif_linux_dp dp;
343 error = dpif_linux_dp_get(dpif_, &dp, &buf);
352 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
354 struct dpif_linux_dp dp;
358 error = dpif_linux_dp_get(dpif_, &dp, &buf);
360 *drop_fragsp = dp.ipv4_frags == OVS_DP_FRAG_DROP;
367 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
369 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
370 struct dpif_linux_dp dp;
372 dpif_linux_dp_init(&dp);
373 dp.cmd = OVS_DP_CMD_SET;
374 dp.dp_ifindex = dpif->dp_ifindex;
375 dp.ipv4_frags = drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO;
376 return dpif_linux_dp_transact(&dp, NULL, NULL);
380 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
383 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
384 const char *name = netdev_get_name(netdev);
385 const char *type = netdev_get_type(netdev);
386 struct dpif_linux_vport request, reply;
387 const struct ofpbuf *options;
391 dpif_linux_vport_init(&request);
392 request.cmd = OVS_VPORT_CMD_NEW;
393 request.dp_ifindex = dpif->dp_ifindex;
394 request.type = netdev_vport_get_vport_type(netdev);
395 if (request.type == OVS_VPORT_TYPE_UNSPEC) {
396 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
397 "unsupported type `%s'",
398 dpif_name(dpif_), name, type);
403 options = netdev_vport_get_options(netdev);
404 if (options && options->size) {
405 request.options = options->data;
406 request.options_len = options->size;
409 if (request.type == OVS_VPORT_TYPE_NETDEV) {
410 netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
413 /* Loop until we find a port that isn't used. */
415 request.port_no = dpif_linux_pop_port(dpif);
416 request.upcall_pid = get_upcall_pid_port(dpif, request.port_no);
417 error = dpif_linux_vport_transact(&request, &reply, &buf);
420 *port_nop = reply.port_no;
421 VLOG_DBG("%s: assigning port %"PRIu32" to netlink "
423 dpif_name(dpif_), request.port_no,
427 } while (request.port_no != UINT32_MAX
428 && (error == EBUSY || error == EFBIG));
434 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
436 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
437 struct dpif_linux_vport vport;
440 dpif_linux_vport_init(&vport);
441 vport.cmd = OVS_VPORT_CMD_DEL;
442 vport.dp_ifindex = dpif->dp_ifindex;
443 vport.port_no = port_no;
444 error = dpif_linux_vport_transact(&vport, NULL, NULL);
447 dpif_linux_push_port(dpif, port_no);
453 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
454 const char *port_name, struct dpif_port *dpif_port)
456 struct dpif_linux_vport request;
457 struct dpif_linux_vport reply;
461 dpif_linux_vport_init(&request);
462 request.cmd = OVS_VPORT_CMD_GET;
463 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
464 request.port_no = port_no;
465 request.name = port_name;
467 error = dpif_linux_vport_transact(&request, &reply, &buf);
469 dpif_port->name = xstrdup(reply.name);
470 dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
471 dpif_port->port_no = reply.port_no;
478 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
479 struct dpif_port *dpif_port)
481 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
485 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
486 struct dpif_port *dpif_port)
488 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
492 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
494 /* If the datapath increases its range of supported ports, then it should
495 * start reporting that. */
500 dpif_linux_flow_flush(struct dpif *dpif_)
502 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
503 struct dpif_linux_flow flow;
505 dpif_linux_flow_init(&flow);
506 flow.cmd = OVS_FLOW_CMD_DEL;
507 flow.dp_ifindex = dpif->dp_ifindex;
508 return dpif_linux_flow_transact(&flow, NULL, NULL);
511 struct dpif_linux_port_state {
513 unsigned long *port_bitmap; /* Ports in the datapath. */
514 bool complete; /* Dump completed without error. */
518 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
520 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
521 struct dpif_linux_port_state *state;
522 struct dpif_linux_vport request;
525 *statep = state = xmalloc(sizeof *state);
526 state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
527 state->complete = false;
529 dpif_linux_vport_init(&request);
530 request.cmd = OVS_DP_CMD_GET;
531 request.dp_ifindex = dpif->dp_ifindex;
533 buf = ofpbuf_new(1024);
534 dpif_linux_vport_to_ofpbuf(&request, buf);
535 nl_dump_start(&state->dump, genl_sock, buf);
542 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
543 struct dpif_port *dpif_port)
545 struct dpif_linux_port_state *state = state_;
546 struct dpif_linux_vport vport;
550 if (!nl_dump_next(&state->dump, &buf)) {
551 state->complete = true;
555 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
560 if (vport.port_no < LRU_MAX_PORTS) {
561 bitmap_set1(state->port_bitmap, vport.port_no);
564 dpif_port->name = (char *) vport.name;
565 dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
566 dpif_port->port_no = vport.port_no;
571 dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
573 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
574 struct dpif_linux_port_state *state = state_;
575 int error = nl_dump_done(&state->dump);
577 if (state->complete) {
580 for (i = 0; i < LRU_MAX_PORTS; i++) {
581 if (!bitmap_is_set(state->port_bitmap, i)) {
582 dpif_linux_push_port(dpif, i);
587 free(state->port_bitmap);
593 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
595 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
597 if (dpif->change_error) {
598 dpif->change_error = false;
599 sset_clear(&dpif->changed_ports);
601 } else if (!sset_is_empty(&dpif->changed_ports)) {
602 *devnamep = sset_pop(&dpif->changed_ports);
610 dpif_linux_port_poll_wait(const struct dpif *dpif_)
612 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
613 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
614 poll_immediate_wake();
619 dpif_linux_flow_get__(const struct dpif *dpif_,
620 const struct nlattr *key, size_t key_len,
621 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
623 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
624 struct dpif_linux_flow request;
626 dpif_linux_flow_init(&request);
627 request.cmd = OVS_FLOW_CMD_GET;
628 request.dp_ifindex = dpif->dp_ifindex;
630 request.key_len = key_len;
631 return dpif_linux_flow_transact(&request, reply, bufp);
635 dpif_linux_flow_get(const struct dpif *dpif_,
636 const struct nlattr *key, size_t key_len,
637 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
639 struct dpif_linux_flow reply;
643 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
646 dpif_linux_flow_get_stats(&reply, stats);
649 buf->data = (void *) reply.actions;
650 buf->size = reply.actions_len;
660 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
661 const struct nlattr *key, size_t key_len,
662 const struct nlattr *actions, size_t actions_len,
663 struct dpif_flow_stats *stats)
665 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
666 struct dpif_linux_flow request, reply;
667 struct nlattr dummy_action;
671 dpif_linux_flow_init(&request);
672 request.cmd = flags & DPIF_FP_CREATE ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET;
673 request.dp_ifindex = dpif->dp_ifindex;
675 request.key_len = key_len;
676 /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
677 request.actions = actions ? actions : &dummy_action;
678 request.actions_len = actions_len;
679 request.upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
680 if (flags & DPIF_FP_ZERO_STATS) {
681 request.clear = true;
683 request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
684 error = dpif_linux_flow_transact(&request,
685 stats ? &reply : NULL,
686 stats ? &buf : NULL);
687 if (!error && stats) {
688 dpif_linux_flow_get_stats(&reply, stats);
695 dpif_linux_flow_del(struct dpif *dpif_,
696 const struct nlattr *key, size_t key_len,
697 struct dpif_flow_stats *stats)
699 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
700 struct dpif_linux_flow request, reply;
704 dpif_linux_flow_init(&request);
705 request.cmd = OVS_FLOW_CMD_DEL;
706 request.dp_ifindex = dpif->dp_ifindex;
708 request.key_len = key_len;
709 error = dpif_linux_flow_transact(&request,
710 stats ? &reply : NULL,
711 stats ? &buf : NULL);
712 if (!error && stats) {
713 dpif_linux_flow_get_stats(&reply, stats);
719 struct dpif_linux_flow_state {
721 struct dpif_linux_flow flow;
722 struct dpif_flow_stats stats;
727 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
729 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
730 struct dpif_linux_flow_state *state;
731 struct dpif_linux_flow request;
734 *statep = state = xmalloc(sizeof *state);
736 dpif_linux_flow_init(&request);
737 request.cmd = OVS_DP_CMD_GET;
738 request.dp_ifindex = dpif->dp_ifindex;
740 buf = ofpbuf_new(1024);
741 dpif_linux_flow_to_ofpbuf(&request, buf);
742 nl_dump_start(&state->dump, genl_sock, buf);
751 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
752 const struct nlattr **key, size_t *key_len,
753 const struct nlattr **actions, size_t *actions_len,
754 const struct dpif_flow_stats **stats)
756 struct dpif_linux_flow_state *state = state_;
761 ofpbuf_delete(state->buf);
764 if (!nl_dump_next(&state->dump, &buf)) {
768 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
773 if (actions && !state->flow.actions) {
774 error = dpif_linux_flow_get__(dpif_, state->flow.key,
776 &state->flow, &state->buf);
777 if (error == ENOENT) {
778 VLOG_DBG("dumped flow disappeared on get");
780 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
786 *actions = state->flow.actions;
787 *actions_len = state->flow.actions_len;
790 *key = state->flow.key;
791 *key_len = state->flow.key_len;
794 dpif_linux_flow_get_stats(&state->flow, &state->stats);
795 *stats = &state->stats;
801 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
803 struct dpif_linux_flow_state *state = state_;
804 int error = nl_dump_done(&state->dump);
805 ofpbuf_delete(state->buf);
811 dpif_linux_execute__(int dp_ifindex, uint32_t upcall_pid,
812 const struct nlattr *key, size_t key_len,
813 const struct nlattr *actions, size_t actions_len,
814 const struct ofpbuf *packet)
816 struct ovs_header *execute;
820 buf = ofpbuf_new(128 + actions_len + packet->size);
822 nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
823 OVS_PACKET_CMD_EXECUTE, 1);
825 execute = ofpbuf_put_uninit(buf, sizeof *execute);
826 execute->dp_ifindex = dp_ifindex;
828 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, packet->data, packet->size);
829 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len);
830 nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len);
831 nl_msg_put_u32(buf, OVS_PACKET_ATTR_UPCALL_PID, upcall_pid);
833 error = nl_sock_transact(genl_sock, buf, NULL);
839 dpif_linux_execute(struct dpif *dpif_,
840 const struct nlattr *key, size_t key_len,
841 const struct nlattr *actions, size_t actions_len,
842 const struct ofpbuf *packet)
844 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
845 uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
847 return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len,
848 actions, actions_len, packet);
852 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
854 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
855 *listen_mask = dpif->listen_mask;
860 get_upcall_pid_port__(struct dpif_linux *dpif, uint32_t port)
862 int idx = port & (N_UPCALL_SOCKS - 1);
863 return nl_sock_pid(dpif->upcall_socks[idx]);
867 get_upcall_pid_port(struct dpif_linux *dpif, uint32_t port)
869 if (!(dpif->listen_mask & (1u << DPIF_UC_MISS))) {
873 return get_upcall_pid_port__(dpif, port);
877 get_upcall_pid_flow(struct dpif_linux *dpif,
878 const struct nlattr *key, size_t key_len)
880 const struct nlattr *nla;
883 if (!(dpif->listen_mask &
884 ((1u << DPIF_UC_ACTION) | (1u << DPIF_UC_SAMPLE)))) {
888 nla = nl_attr_find__(key, key_len, OVS_KEY_ATTR_IN_PORT);
890 port = nl_attr_get_u32(nla);
892 port = random_uint32();
895 return get_upcall_pid_port__(dpif, port);
899 set_upcall_pids(struct dpif_linux *dpif)
901 struct dpif_port port;
902 struct dpif_port_dump port_dump;
903 struct dpif_flow_dump flow_dump;
904 const struct nlattr *key;
908 DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
909 struct dpif_linux_vport vport_request;
911 dpif_linux_vport_init(&vport_request);
912 vport_request.cmd = OVS_VPORT_CMD_SET;
913 vport_request.dp_ifindex = dpif->dp_ifindex;
914 vport_request.port_no = port.port_no;
915 vport_request.upcall_pid = get_upcall_pid_port(dpif,
916 vport_request.port_no);
917 error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
919 VLOG_DBG("%s: assigning port %"PRIu32" to netlink "
921 dpif_name(&dpif->dpif), vport_request.port_no,
922 vport_request.upcall_pid);
924 VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
925 dpif_name(&dpif->dpif), strerror(error));
929 dpif_flow_dump_start(&flow_dump, &dpif->dpif);
930 while (dpif_flow_dump_next(&flow_dump, &key, &key_len,
932 struct dpif_linux_flow flow_request;
934 dpif_linux_flow_init(&flow_request);
935 flow_request.cmd = OVS_FLOW_CMD_SET;
936 flow_request.dp_ifindex = dpif->dp_ifindex;
937 flow_request.key = key;
938 flow_request.key_len = key_len;
939 flow_request.upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
940 error = dpif_linux_flow_transact(&flow_request, NULL, NULL);
942 VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on flow: %s",
943 dpif_name(&dpif->dpif), strerror(error));
946 dpif_flow_dump_done(&flow_dump);
950 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
952 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
954 if (listen_mask == dpif->listen_mask) {
959 destroy_upcall_socks(dpif);
960 } else if (!dpif->listen_mask) {
964 for (i = 0; i < N_UPCALL_SOCKS; i++) {
965 error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]);
967 destroy_upcall_socks(dpif);
973 dpif->listen_mask = listen_mask;
974 set_upcall_pids(dpif);
980 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
981 uint32_t *probability)
983 struct dpif_linux_dp dp;
987 error = dpif_linux_dp_get(dpif_, &dp, &buf);
989 *probability = dp.sampling ? *dp.sampling : 0;
996 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
998 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
999 struct dpif_linux_dp dp;
1001 dpif_linux_dp_init(&dp);
1002 dp.cmd = OVS_DP_CMD_SET;
1003 dp.dp_ifindex = dpif->dp_ifindex;
1004 dp.sampling = &probability;
1005 return dpif_linux_dp_transact(&dp, NULL, NULL);
1009 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1010 uint32_t queue_id, uint32_t *priority)
1012 if (queue_id < 0xf000) {
1013 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
1021 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
1024 static const struct nl_policy ovs_packet_policy[] = {
1025 /* Always present. */
1026 [OVS_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
1027 .min_len = ETH_HEADER_LEN },
1028 [OVS_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
1030 /* OVS_PACKET_CMD_ACTION only. */
1031 [OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
1033 /* OVS_PACKET_CMD_SAMPLE only. */
1034 [OVS_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
1035 [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1038 struct ovs_header *ovs_header;
1039 struct nlattr *a[ARRAY_SIZE(ovs_packet_policy)];
1040 struct nlmsghdr *nlmsg;
1041 struct genlmsghdr *genl;
1045 ofpbuf_use_const(&b, buf->data, buf->size);
1047 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1048 genl = ofpbuf_try_pull(&b, sizeof *genl);
1049 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1050 if (!nlmsg || !genl || !ovs_header
1051 || nlmsg->nlmsg_type != ovs_packet_family
1052 || !nl_policy_parse(&b, 0, ovs_packet_policy, a,
1053 ARRAY_SIZE(ovs_packet_policy))) {
1057 type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
1058 : genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
1059 : genl->cmd == OVS_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
1065 memset(upcall, 0, sizeof *upcall);
1066 upcall->type = type;
1067 upcall->packet = buf;
1068 upcall->packet->data = (void *) nl_attr_get(a[OVS_PACKET_ATTR_PACKET]);
1069 upcall->packet->size = nl_attr_get_size(a[OVS_PACKET_ATTR_PACKET]);
1070 upcall->key = (void *) nl_attr_get(a[OVS_PACKET_ATTR_KEY]);
1071 upcall->key_len = nl_attr_get_size(a[OVS_PACKET_ATTR_KEY]);
1072 upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
1073 ? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
1075 upcall->sample_pool = (a[OVS_PACKET_ATTR_SAMPLE_POOL]
1076 ? nl_attr_get_u32(a[OVS_PACKET_ATTR_SAMPLE_POOL])
1078 if (a[OVS_PACKET_ATTR_ACTIONS]) {
1079 upcall->actions = (void *) nl_attr_get(a[OVS_PACKET_ATTR_ACTIONS]);
1080 upcall->actions_len = nl_attr_get_size(a[OVS_PACKET_ATTR_ACTIONS]);
1083 *dp_ifindex = ovs_header->dp_ifindex;
1089 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
1091 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1095 if (!dpif->listen_mask) {
1099 for (i = 0; i < N_UPCALL_SOCKS; i++) {
1100 struct nl_sock *upcall_sock;
1101 dpif->last_read_upcall = (dpif->last_read_upcall + 1) &
1102 (N_UPCALL_SOCKS - 1);
1103 upcall_sock = dpif->upcall_socks[dpif->last_read_upcall];
1105 if (nl_sock_woke(upcall_sock)) {
1112 if (++read_tries > 50) {
1116 error = nl_sock_recv(upcall_sock, &buf, false);
1117 if (error == EAGAIN) {
1123 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1125 && dp_ifindex == dpif->dp_ifindex
1126 && dpif->listen_mask & (1u << upcall->type)) {
1142 dpif_linux_recv_wait(struct dpif *dpif_)
1144 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1147 if (!dpif->listen_mask) {
1151 for (i = 0; i < N_UPCALL_SOCKS; i++) {
1152 nl_sock_wait(dpif->upcall_socks[i], POLLIN);
1157 dpif_linux_recv_purge(struct dpif *dpif_)
1159 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1162 if (!dpif->listen_mask) {
1166 for (i = 0; i < N_UPCALL_SOCKS; i++) {
1167 nl_sock_drain(dpif->upcall_socks[i]);
1171 const struct dpif_class dpif_linux_class = {
1173 dpif_linux_enumerate,
1179 dpif_linux_get_stats,
1180 dpif_linux_get_drop_frags,
1181 dpif_linux_set_drop_frags,
1182 dpif_linux_port_add,
1183 dpif_linux_port_del,
1184 dpif_linux_port_query_by_number,
1185 dpif_linux_port_query_by_name,
1186 dpif_linux_get_max_ports,
1187 dpif_linux_port_dump_start,
1188 dpif_linux_port_dump_next,
1189 dpif_linux_port_dump_done,
1190 dpif_linux_port_poll,
1191 dpif_linux_port_poll_wait,
1192 dpif_linux_flow_get,
1193 dpif_linux_flow_put,
1194 dpif_linux_flow_del,
1195 dpif_linux_flow_flush,
1196 dpif_linux_flow_dump_start,
1197 dpif_linux_flow_dump_next,
1198 dpif_linux_flow_dump_done,
1200 dpif_linux_recv_get_mask,
1201 dpif_linux_recv_set_mask,
1202 dpif_linux_get_sflow_probability,
1203 dpif_linux_set_sflow_probability,
1204 dpif_linux_queue_to_priority,
1206 dpif_linux_recv_wait,
1207 dpif_linux_recv_purge,
1211 dpif_linux_init(void)
1213 static int error = -1;
1216 unsigned int ovs_vport_mcgroup;
1218 error = nl_lookup_genl_family(OVS_DATAPATH_FAMILY,
1219 &ovs_datapath_family);
1221 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1222 "The Open vSwitch kernel module is probably not loaded.",
1223 OVS_DATAPATH_FAMILY);
1226 error = nl_lookup_genl_family(OVS_VPORT_FAMILY, &ovs_vport_family);
1229 error = nl_lookup_genl_family(OVS_FLOW_FAMILY, &ovs_flow_family);
1232 error = nl_lookup_genl_family(OVS_PACKET_FAMILY,
1233 &ovs_packet_family);
1236 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1239 error = nl_lookup_genl_mcgroup(OVS_VPORT_FAMILY, OVS_VPORT_MCGROUP,
1241 OVS_VPORT_MCGROUP_FALLBACK_ID);
1244 static struct dpif_linux_vport vport;
1245 nln = nln_create(NETLINK_GENERIC, ovs_vport_mcgroup,
1246 dpif_linux_nln_parse, &vport);
1254 dpif_linux_is_internal_device(const char *name)
1256 struct dpif_linux_vport reply;
1260 error = dpif_linux_vport_get(name, &reply, &buf);
1263 } else if (error != ENODEV && error != ENOENT) {
1264 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1265 name, strerror(error));
1268 return reply.type == OVS_VPORT_TYPE_INTERNAL;
1272 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1273 const void *data, size_t size)
1275 struct ofpbuf actions, key, packet;
1276 struct odputil_keybuf keybuf;
1280 ofpbuf_use_const(&packet, data, size);
1281 flow_extract(&packet, htonll(0), 0, &flow);
1283 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1284 odp_flow_key_from_flow(&key, &flow);
1286 ofpbuf_use_stack(&actions, &action, sizeof action);
1287 nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
1289 return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size,
1290 actions.data, actions.size, &packet);
1294 dpif_linux_nln_parse(struct ofpbuf *buf, void *vport_)
1296 struct dpif_linux_vport *vport = vport_;
1297 return dpif_linux_vport_from_ofpbuf(vport, buf) == 0;
1301 dpif_linux_port_changed(const void *vport_, void *dpif_)
1303 const struct dpif_linux_vport *vport = vport_;
1304 struct dpif_linux *dpif = dpif_;
1307 if (vport->dp_ifindex == dpif->dp_ifindex
1308 && (vport->cmd == OVS_VPORT_CMD_NEW
1309 || vport->cmd == OVS_VPORT_CMD_DEL
1310 || vport->cmd == OVS_VPORT_CMD_SET)) {
1311 VLOG_DBG("port_changed: dpif:%s vport:%s cmd:%"PRIu8,
1312 dpif->dpif.full_name, vport->name, vport->cmd);
1313 sset_add(&dpif->changed_ports, vport->name);
1316 dpif->change_error = true;
1320 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1321 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1322 * positive errno value.
1324 * 'vport' will contain pointers into 'buf', so the caller should not free
1325 * 'buf' while 'vport' is still in use. */
1327 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1328 const struct ofpbuf *buf)
1330 static const struct nl_policy ovs_vport_policy[] = {
1331 [OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1332 [OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1333 [OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1334 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1335 [OVS_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1336 .min_len = sizeof(struct ovs_vport_stats),
1337 .max_len = sizeof(struct ovs_vport_stats),
1339 [OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1340 .min_len = ETH_ADDR_LEN,
1341 .max_len = ETH_ADDR_LEN,
1343 [OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1344 [OVS_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1347 struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
1348 struct ovs_header *ovs_header;
1349 struct nlmsghdr *nlmsg;
1350 struct genlmsghdr *genl;
1353 dpif_linux_vport_init(vport);
1355 ofpbuf_use_const(&b, buf->data, buf->size);
1356 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1357 genl = ofpbuf_try_pull(&b, sizeof *genl);
1358 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1359 if (!nlmsg || !genl || !ovs_header
1360 || nlmsg->nlmsg_type != ovs_vport_family
1361 || !nl_policy_parse(&b, 0, ovs_vport_policy, a,
1362 ARRAY_SIZE(ovs_vport_policy))) {
1366 vport->cmd = genl->cmd;
1367 vport->dp_ifindex = ovs_header->dp_ifindex;
1368 vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1369 vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1370 vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
1371 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1372 vport->upcall_pid = nl_attr_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1374 if (a[OVS_VPORT_ATTR_STATS]) {
1375 vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
1377 if (a[OVS_VPORT_ATTR_ADDRESS]) {
1378 vport->address = nl_attr_get(a[OVS_VPORT_ATTR_ADDRESS]);
1380 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1381 vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
1382 vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
1384 if (a[OVS_VPORT_ATTR_IFINDEX]) {
1385 vport->ifindex = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFINDEX]);
1390 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1391 * followed by Netlink attributes corresponding to 'vport'. */
1393 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1396 struct ovs_header *ovs_header;
1398 nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1401 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1402 ovs_header->dp_ifindex = vport->dp_ifindex;
1404 if (vport->port_no != UINT32_MAX) {
1405 nl_msg_put_u32(buf, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1408 if (vport->type != OVS_VPORT_TYPE_UNSPEC) {
1409 nl_msg_put_u32(buf, OVS_VPORT_ATTR_TYPE, vport->type);
1413 nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
1416 nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1419 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
1420 vport->stats, sizeof *vport->stats);
1423 if (vport->address) {
1424 nl_msg_put_unspec(buf, OVS_VPORT_ATTR_ADDRESS,
1425 vport->address, ETH_ADDR_LEN);
1428 if (vport->options) {
1429 nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
1430 vport->options, vport->options_len);
1433 if (vport->ifindex) {
1434 nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFINDEX, vport->ifindex);
1438 /* Clears 'vport' to "empty" values. */
1440 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1442 memset(vport, 0, sizeof *vport);
1443 vport->port_no = UINT32_MAX;
1446 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1447 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1448 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1449 * result of the command is expected to be an ovs_vport also, which is decoded
1450 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1451 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1453 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1454 struct dpif_linux_vport *reply,
1455 struct ofpbuf **bufp)
1457 struct ofpbuf *request_buf;
1460 assert((reply != NULL) == (bufp != NULL));
1462 error = dpif_linux_init();
1466 dpif_linux_vport_init(reply);
1471 request_buf = ofpbuf_new(1024);
1472 dpif_linux_vport_to_ofpbuf(request, request_buf);
1473 error = nl_sock_transact(genl_sock, request_buf, bufp);
1474 ofpbuf_delete(request_buf);
1478 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1481 dpif_linux_vport_init(reply);
1482 ofpbuf_delete(*bufp);
1489 /* Obtains information about the kernel vport named 'name' and stores it into
1490 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1491 * longer needed ('reply' will contain pointers into '*bufp'). */
1493 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1494 struct ofpbuf **bufp)
1496 struct dpif_linux_vport request;
1498 dpif_linux_vport_init(&request);
1499 request.cmd = OVS_VPORT_CMD_GET;
1500 request.name = name;
1502 return dpif_linux_vport_transact(&request, reply, bufp);
1505 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1506 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1507 * positive errno value.
1509 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1510 * while 'dp' is still in use. */
1512 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1514 static const struct nl_policy ovs_datapath_policy[] = {
1515 [OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1516 [OVS_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1517 .min_len = sizeof(struct ovs_dp_stats),
1518 .max_len = sizeof(struct ovs_dp_stats),
1520 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1521 [OVS_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1524 struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
1525 struct ovs_header *ovs_header;
1526 struct nlmsghdr *nlmsg;
1527 struct genlmsghdr *genl;
1530 dpif_linux_dp_init(dp);
1532 ofpbuf_use_const(&b, buf->data, buf->size);
1533 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1534 genl = ofpbuf_try_pull(&b, sizeof *genl);
1535 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1536 if (!nlmsg || !genl || !ovs_header
1537 || nlmsg->nlmsg_type != ovs_datapath_family
1538 || !nl_policy_parse(&b, 0, ovs_datapath_policy, a,
1539 ARRAY_SIZE(ovs_datapath_policy))) {
1543 dp->cmd = genl->cmd;
1544 dp->dp_ifindex = ovs_header->dp_ifindex;
1545 dp->name = nl_attr_get_string(a[OVS_DP_ATTR_NAME]);
1546 if (a[OVS_DP_ATTR_STATS]) {
1547 /* Can't use structure assignment because Netlink doesn't ensure
1548 * sufficient alignment for 64-bit members. */
1549 memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
1552 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1553 dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1555 if (a[OVS_DP_ATTR_SAMPLING]) {
1556 dp->sampling = nl_attr_get(a[OVS_DP_ATTR_SAMPLING]);
1562 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1564 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1566 struct ovs_header *ovs_header;
1568 nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
1569 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1571 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1572 ovs_header->dp_ifindex = dp->dp_ifindex;
1575 nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
1578 nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, dp->upcall_pid);
1580 /* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
1582 if (dp->ipv4_frags) {
1583 nl_msg_put_u32(buf, OVS_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1587 nl_msg_put_u32(buf, OVS_DP_ATTR_SAMPLING, *dp->sampling);
1591 /* Clears 'dp' to "empty" values. */
1593 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1595 memset(dp, 0, sizeof *dp);
1599 dpif_linux_dp_dump_start(struct nl_dump *dump)
1601 struct dpif_linux_dp request;
1604 dpif_linux_dp_init(&request);
1605 request.cmd = OVS_DP_CMD_GET;
1607 buf = ofpbuf_new(1024);
1608 dpif_linux_dp_to_ofpbuf(&request, buf);
1609 nl_dump_start(dump, genl_sock, buf);
1613 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1614 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1615 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1616 * result of the command is expected to be of the same form, which is decoded
1617 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1618 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1620 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1621 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1623 struct ofpbuf *request_buf;
1626 assert((reply != NULL) == (bufp != NULL));
1628 request_buf = ofpbuf_new(1024);
1629 dpif_linux_dp_to_ofpbuf(request, request_buf);
1630 error = nl_sock_transact(genl_sock, request_buf, bufp);
1631 ofpbuf_delete(request_buf);
1635 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1638 dpif_linux_dp_init(reply);
1639 ofpbuf_delete(*bufp);
1646 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1647 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1648 * will contain pointers into '*bufp'). */
1650 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1651 struct ofpbuf **bufp)
1653 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1654 struct dpif_linux_dp request;
1656 dpif_linux_dp_init(&request);
1657 request.cmd = OVS_DP_CMD_GET;
1658 request.dp_ifindex = dpif->dp_ifindex;
1660 return dpif_linux_dp_transact(&request, reply, bufp);
1663 /* Parses the contents of 'buf', which contains a "struct ovs_header" followed
1664 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1665 * positive errno value.
1667 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1668 * while 'flow' is still in use. */
1670 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1671 const struct ofpbuf *buf)
1673 static const struct nl_policy ovs_flow_policy[] = {
1674 [OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1675 [OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1676 [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
1677 [OVS_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1678 .min_len = sizeof(struct ovs_flow_stats),
1679 .max_len = sizeof(struct ovs_flow_stats),
1681 [OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1682 [OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1683 /* The kernel never uses OVS_FLOW_ATTR_CLEAR. */
1686 struct nlattr *a[ARRAY_SIZE(ovs_flow_policy)];
1687 struct ovs_header *ovs_header;
1688 struct nlmsghdr *nlmsg;
1689 struct genlmsghdr *genl;
1692 dpif_linux_flow_init(flow);
1694 ofpbuf_use_const(&b, buf->data, buf->size);
1695 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1696 genl = ofpbuf_try_pull(&b, sizeof *genl);
1697 ovs_header = ofpbuf_try_pull(&b, sizeof *ovs_header);
1698 if (!nlmsg || !genl || !ovs_header
1699 || nlmsg->nlmsg_type != ovs_flow_family
1700 || !nl_policy_parse(&b, 0, ovs_flow_policy, a,
1701 ARRAY_SIZE(ovs_flow_policy))) {
1705 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1706 flow->dp_ifindex = ovs_header->dp_ifindex;
1707 flow->key = nl_attr_get(a[OVS_FLOW_ATTR_KEY]);
1708 flow->key_len = nl_attr_get_size(a[OVS_FLOW_ATTR_KEY]);
1709 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1710 flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
1711 flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
1713 if (a[OVS_FLOW_ATTR_UPCALL_PID]) {
1714 flow->upcall_pid = nl_attr_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
1716 if (a[OVS_FLOW_ATTR_STATS]) {
1717 flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
1719 if (a[OVS_FLOW_ATTR_TCP_FLAGS]) {
1720 flow->tcp_flags = nl_attr_get(a[OVS_FLOW_ATTR_TCP_FLAGS]);
1722 if (a[OVS_FLOW_ATTR_USED]) {
1723 flow->used = nl_attr_get(a[OVS_FLOW_ATTR_USED]);
1728 /* Appends to 'buf' (which must initially be empty) a "struct ovs_header"
1729 * followed by Netlink attributes corresponding to 'flow'. */
1731 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1734 struct ovs_header *ovs_header;
1736 nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
1737 NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1740 ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
1741 ovs_header->dp_ifindex = flow->dp_ifindex;
1743 if (flow->key_len) {
1744 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_KEY, flow->key, flow->key_len);
1747 if (flow->actions || flow->actions_len) {
1748 nl_msg_put_unspec(buf, OVS_FLOW_ATTR_ACTIONS,
1749 flow->actions, flow->actions_len);
1752 nl_msg_put_u32(buf, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
1754 /* We never need to send these to the kernel. */
1755 assert(!flow->stats);
1756 assert(!flow->tcp_flags);
1757 assert(!flow->used);
1760 nl_msg_put_flag(buf, OVS_FLOW_ATTR_CLEAR);
1764 /* Clears 'flow' to "empty" values. */
1766 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1768 memset(flow, 0, sizeof *flow);
1771 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1772 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1773 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1774 * result of the command is expected to be a flow also, which is decoded and
1775 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1776 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1778 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1779 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1781 struct ofpbuf *request_buf;
1784 assert((reply != NULL) == (bufp != NULL));
1786 request_buf = ofpbuf_new(1024);
1787 dpif_linux_flow_to_ofpbuf(request, request_buf);
1788 error = nl_sock_transact(genl_sock, request_buf, bufp);
1789 ofpbuf_delete(request_buf);
1793 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1796 dpif_linux_flow_init(reply);
1797 ofpbuf_delete(*bufp);
1805 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1806 struct dpif_flow_stats *stats)
1809 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1810 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1812 stats->n_packets = 0;
1815 stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1816 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;