2 * Copyright (c) 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 "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_revalidate);
72 COVERAGE_DEFINE(facet_unexpected);
73 COVERAGE_DEFINE(facet_create);
74 COVERAGE_DEFINE(facet_remove);
75 COVERAGE_DEFINE(subfacet_create);
76 COVERAGE_DEFINE(subfacet_destroy);
77 COVERAGE_DEFINE(subfacet_install_fail);
78 COVERAGE_DEFINE(packet_in_overflow);
80 /* Number of implemented OpenFlow tables. */
81 enum { N_TABLES = 255 };
82 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
83 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
93 * - Do include packets and bytes from facets that have been deleted or
94 * whose own statistics have been folded into the rule.
96 * - Do include packets and bytes sent "by hand" that were accounted to
97 * the rule without any facet being involved (this is a rare corner
98 * case in rule_execute()).
100 * - Do not include packet or bytes that can be obtained from any facet's
101 * packet_count or byte_count member or that can be obtained from the
102 * datapath by, e.g., dpif_flow_get() for any subfacet.
104 struct ovs_mutex stats_mutex;
105 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
106 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
109 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
110 static struct rule_dpif *rule_dpif_cast(const struct rule *);
113 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 void *aux; /* Key supplied by ofproto's client. */
116 char *name; /* Identifier for log messages. */
119 struct list ports; /* Contains "struct ofport"s. */
120 enum port_vlan_mode vlan_mode; /* VLAN mode */
121 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
125 struct bond *bond; /* Nonnull iff more than one port. */
126 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
129 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
132 static void bundle_remove(struct ofport *);
133 static void bundle_update(struct ofbundle *);
134 static void bundle_destroy(struct ofbundle *);
135 static void bundle_del_port(struct ofport_dpif *);
136 static void bundle_run(struct ofbundle *);
137 static void bundle_wait(struct ofbundle *);
139 static void stp_run(struct ofproto_dpif *ofproto);
140 static void stp_wait(struct ofproto_dpif *ofproto);
141 static int set_stp_port(struct ofport *,
142 const struct ofproto_port_stp_settings *);
144 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
145 enum slow_path_reason,
146 uint64_t *stub, size_t stub_size,
147 const struct nlattr **actionsp,
148 size_t *actions_lenp);
150 /* A subfacet (see "struct subfacet" below) has three possible installation
153 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
154 * case just after the subfacet is created, just before the subfacet is
155 * destroyed, or if the datapath returns an error when we try to install a
158 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
160 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
161 * ofproto_dpif is installed in the datapath.
164 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
165 SF_FAST_PATH, /* Full actions are installed. */
166 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
169 /* A dpif flow and actions associated with a facet.
171 * See also the large comment on struct facet. */
174 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
175 struct list list_node; /* In struct facet's 'facets' list. */
176 struct facet *facet; /* Owning facet. */
177 struct dpif_backer *backer; /* Owning backer. */
182 long long int used; /* Time last used; time created if not used. */
183 long long int created; /* Time created. */
185 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
186 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
188 enum subfacet_path path; /* Installed in datapath? */
191 #define SUBFACET_DESTROY_MAX_BATCH 50
193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
194 static struct subfacet *subfacet_find(struct dpif_backer *,
195 const struct nlattr *key, size_t key_len,
197 static void subfacet_destroy(struct subfacet *);
198 static void subfacet_destroy__(struct subfacet *);
199 static void subfacet_destroy_batch(struct dpif_backer *,
200 struct subfacet **, int n);
201 static void subfacet_reset_dp_stats(struct subfacet *,
202 struct dpif_flow_stats *);
203 static void subfacet_update_stats(struct subfacet *,
204 const struct dpif_flow_stats *);
205 static int subfacet_install(struct subfacet *,
206 const struct ofpbuf *odp_actions,
207 struct dpif_flow_stats *);
208 static void subfacet_uninstall(struct subfacet *);
210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
212 * A facet associates a "struct flow", which represents the Open vSwitch
213 * userspace idea of an exact-match flow, with one or more subfacets.
214 * While the facet is created based on an exact-match flow, it is stored
215 * within the ofproto based on the wildcards that could be expressed
216 * based on the flow table and other configuration. (See the 'wc'
217 * description in "struct xlate_out" for more details.)
219 * Each subfacet tracks the datapath's idea of the flow equivalent to
220 * the facet. When the kernel module (or other dpif implementation) and
221 * Open vSwitch userspace agree on the definition of a flow key, there
222 * is exactly one subfacet per facet. If the dpif implementation
223 * supports more-specific flow matching than userspace, however, a facet
224 * can have more than one subfacet. Examples include the dpif
225 * implementation not supporting the same wildcards as userspace or some
226 * distinction in flow that userspace simply doesn't understand.
228 * Flow expiration works in terms of subfacets, so a facet must have at
229 * least one subfacet or it will never expire, leaking memory. */
232 struct ofproto_dpif *ofproto;
235 struct list subfacets;
236 long long int used; /* Time last used; time created if not used. */
239 struct flow flow; /* Flow of the creating subfacet. */
240 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
244 * - Do include packets and bytes sent "by hand", e.g. with
247 * - Do include packets and bytes that were obtained from the datapath
248 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
249 * DPIF_FP_ZERO_STATS).
251 * - Do not include packets or bytes that can be obtained from the
252 * datapath for any existing subfacet.
254 uint64_t packet_count; /* Number of packets received. */
255 uint64_t byte_count; /* Number of bytes received. */
257 /* Resubmit statistics. */
258 uint64_t prev_packet_count; /* Number of packets from last stats push. */
259 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
260 long long int prev_used; /* Used time from last stats push. */
263 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
264 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
265 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
267 struct xlate_out xout;
269 /* Storage for a single subfacet, to reduce malloc() time and space
270 * overhead. (A facet always has at least one subfacet and in the common
271 * case has exactly one subfacet. However, 'one_subfacet' may not
272 * always be valid, since it could have been removed after newer
273 * subfacets were pushed onto the 'subfacets' list.) */
274 struct subfacet one_subfacet;
276 long long int learn_rl; /* Rate limiter for facet_learn(). */
279 static struct facet *facet_create(const struct flow_miss *);
280 static void facet_remove(struct facet *);
281 static void facet_free(struct facet *);
283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
285 const struct flow *);
286 static bool facet_revalidate(struct facet *);
287 static bool facet_check_consistency(struct facet *);
289 static void facet_flush_stats(struct facet *);
291 static void facet_reset_counters(struct facet *);
292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
293 struct dpif_flow_stats *, bool may_learn);
294 static void facet_push_stats(struct facet *, bool may_learn);
295 static void facet_learn(struct facet *);
296 static void facet_account(struct facet *);
297 static void push_all_stats(void);
299 static bool facet_is_controller_flow(struct facet *);
302 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
306 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
307 struct list bundle_node; /* In struct ofbundle's "ports" list. */
308 struct cfm *cfm; /* Connectivity Fault Management, if any. */
309 struct bfd *bfd; /* BFD, if any. */
310 bool may_enable; /* May be enabled in bonds. */
311 bool is_tunnel; /* This port is a tunnel. */
312 long long int carrier_seq; /* Carrier status changes. */
313 struct ofport_dpif *peer; /* Peer if patch port. */
316 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
317 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
318 long long int stp_state_entered;
320 /* Queue to DSCP mapping. */
321 struct ofproto_port_queue *qdscp;
324 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
326 * This is deprecated. It is only for compatibility with broken device
327 * drivers in old versions of Linux that do not properly support VLANs when
328 * VLAN devices are not used. When broken device drivers are no longer in
329 * widespread use, we will delete these interfaces. */
330 ofp_port_t realdev_ofp_port;
334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
336 * This is deprecated. It is only for compatibility with broken device drivers
337 * in old versions of Linux that do not properly support VLANs when VLAN
338 * devices are not used. When broken device drivers are no longer in
339 * widespread use, we will delete these interfaces. */
340 struct vlan_splinter {
341 struct hmap_node realdev_vid_node;
342 struct hmap_node vlandev_node;
343 ofp_port_t realdev_ofp_port;
344 ofp_port_t vlandev_ofp_port;
348 static void vsp_remove(struct ofport_dpif *);
349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
357 static struct ofport_dpif *
358 ofport_dpif_cast(const struct ofport *ofport)
360 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
363 static void port_run(struct ofport_dpif *);
364 static void port_run_fast(struct ofport_dpif *);
365 static void port_wait(struct ofport_dpif *);
366 static int set_bfd(struct ofport *, const struct smap *);
367 static int set_cfm(struct ofport *, const struct cfm_settings *);
368 static void ofport_update_peer(struct ofport_dpif *);
369 static void run_fast_rl(void);
370 static int run_fast(struct ofproto *);
372 struct dpif_completion {
373 struct list list_node;
374 struct ofoperation *op;
377 /* Reasons that we might need to revalidate every facet, and corresponding
380 * A value of 0 means that there is no need to revalidate.
382 * It would be nice to have some cleaner way to integrate with coverage
383 * counters, but with only a few reasons I guess this is good enough for
385 enum revalidate_reason {
386 REV_RECONFIGURE = 1, /* Switch configuration changed. */
387 REV_STP, /* Spanning tree protocol port status change. */
388 REV_BOND, /* Bonding changed. */
389 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
390 REV_FLOW_TABLE, /* Flow table changed. */
391 REV_MAC_LEARNING, /* Mac learning changed. */
392 REV_INCONSISTENCY /* Facet self-check failed. */
394 COVERAGE_DEFINE(rev_reconfigure);
395 COVERAGE_DEFINE(rev_stp);
396 COVERAGE_DEFINE(rev_bond);
397 COVERAGE_DEFINE(rev_port_toggled);
398 COVERAGE_DEFINE(rev_flow_table);
399 COVERAGE_DEFINE(rev_mac_learning);
400 COVERAGE_DEFINE(rev_inconsistency);
402 struct avg_subfacet_rates {
403 double add_rate; /* Moving average of new flows created per minute. */
404 double del_rate; /* Moving average of flows deleted per minute. */
407 /* All datapaths of a given type share a single dpif backer instance. */
413 struct timer next_expiration;
415 struct ovs_rwlock odp_to_ofport_lock;
416 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
418 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
420 /* Facet revalidation flags applying to facets which use this backer. */
421 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
423 struct hmap drop_keys; /* Set of dropped odp keys. */
424 bool recv_set_enable; /* Enables or disables receiving packets. */
426 struct hmap subfacets;
427 struct governor *governor;
429 /* Subfacet statistics.
431 * These keep track of the total number of subfacets added and deleted and
432 * flow life span. They are useful for computing the flow rates stats
433 * exposed via "ovs-appctl dpif/show". The goal is to learn about
434 * traffic patterns in ways that we can use later to improve Open vSwitch
435 * performance in new situations. */
436 long long int created; /* Time when it is created. */
437 unsigned max_n_subfacet; /* Maximum number of flows */
438 unsigned avg_n_subfacet; /* Average number of flows. */
439 long long int avg_subfacet_life; /* Average life span of subfacets. */
441 /* Number of upcall handling threads. */
442 unsigned int n_handler_threads;
445 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
446 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
448 static void drop_key_clear(struct dpif_backer *);
450 struct ofproto_dpif {
451 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
453 struct dpif_backer *backer;
455 /* Special OpenFlow rules. */
456 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
457 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
458 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
461 struct netflow *netflow;
462 struct dpif_sflow *sflow;
463 struct dpif_ipfix *ipfix;
464 struct hmap bundles; /* Contains "struct ofbundle"s. */
465 struct mac_learning *ml;
466 bool has_bonded_bundles;
467 struct mbridge *mbridge;
470 struct classifier facets; /* Contains 'struct facet's. */
471 long long int consistency_rl;
473 struct netdev_stats stats; /* To account packets generated and consumed in
478 long long int stp_last_tick;
480 /* VLAN splinters. */
481 struct ovs_mutex vsp_mutex;
482 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
483 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
486 struct sset ports; /* Set of standard port names. */
487 struct sset ghost_ports; /* Ports with no datapath port. */
488 struct sset port_poll_set; /* Queued names for port_poll() reply. */
489 int port_poll_errno; /* Last errno for port_poll() reply. */
491 /* Per ofproto's dpif stats. */
496 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
499 /* By default, flows in the datapath are wildcarded (megaflows). They
500 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
501 static bool enable_megaflows = true;
503 /* All existing ofproto_dpif instances, indexed by ->up.name. */
504 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
506 static void ofproto_dpif_unixctl_init(void);
508 static inline struct ofproto_dpif *
509 ofproto_dpif_cast(const struct ofproto *ofproto)
511 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
512 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
515 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
516 ofp_port_t ofp_port);
517 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
518 const struct ofpbuf *packet, struct ds *);
521 static void handle_upcalls(struct dpif_backer *);
523 /* Flow expiration. */
524 static int expire(struct dpif_backer *);
527 static void send_netflow_active_timeouts(struct ofproto_dpif *);
530 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
532 /* Global variables. */
533 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
535 /* Initial mappings of port to bridge mappings. */
536 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
538 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
541 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
542 struct ofputil_flow_mod *fm)
544 ofproto_flow_mod(&ofproto->up, fm);
547 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
548 * Takes ownership of 'pin' and pin->packet. */
550 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
551 struct ofputil_packet_in *pin)
553 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
554 COVERAGE_INC(packet_in_overflow);
555 free(CONST_CAST(void *, pin->packet));
560 /* Factory functions. */
563 init(const struct shash *iface_hints)
565 struct shash_node *node;
567 /* Make a local copy, since we don't own 'iface_hints' elements. */
568 SHASH_FOR_EACH(node, iface_hints) {
569 const struct iface_hint *orig_hint = node->data;
570 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
572 new_hint->br_name = xstrdup(orig_hint->br_name);
573 new_hint->br_type = xstrdup(orig_hint->br_type);
574 new_hint->ofp_port = orig_hint->ofp_port;
576 shash_add(&init_ofp_ports, node->name, new_hint);
581 enumerate_types(struct sset *types)
583 dp_enumerate_types(types);
587 enumerate_names(const char *type, struct sset *names)
589 struct ofproto_dpif *ofproto;
592 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
593 if (strcmp(type, ofproto->up.type)) {
596 sset_add(names, ofproto->up.name);
603 del(const char *type, const char *name)
608 error = dpif_open(name, type, &dpif);
610 error = dpif_delete(dpif);
617 port_open_type(const char *datapath_type, const char *port_type)
619 return dpif_port_open_type(datapath_type, port_type);
622 /* Type functions. */
624 static void process_dpif_port_changes(struct dpif_backer *);
625 static void process_dpif_all_ports_changed(struct dpif_backer *);
626 static void process_dpif_port_change(struct dpif_backer *,
627 const char *devname);
628 static void process_dpif_port_error(struct dpif_backer *, int error);
630 static struct ofproto_dpif *
631 lookup_ofproto_dpif_by_port_name(const char *name)
633 struct ofproto_dpif *ofproto;
635 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
636 if (sset_contains(&ofproto->ports, name)) {
645 type_run(const char *type)
647 static long long int push_timer = LLONG_MIN;
648 struct dpif_backer *backer;
650 backer = shash_find_data(&all_dpif_backers, type);
652 /* This is not necessarily a problem, since backers are only
653 * created on demand. */
657 dpif_run(backer->dpif);
659 /* The most natural place to push facet statistics is when they're pulled
660 * from the datapath. However, when there are many flows in the datapath,
661 * this expensive operation can occur so frequently, that it reduces our
662 * ability to quickly set up flows. To reduce the cost, we push statistics
664 if (time_msec() > push_timer) {
665 push_timer = time_msec() + 2000;
669 /* If vswitchd started with other_config:flow_restore_wait set as "true",
670 * and the configuration has now changed to "false", enable receiving
671 * packets from the datapath. */
672 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
675 backer->recv_set_enable = true;
677 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
679 udpif_recv_set(backer->udpif, 0, false);
680 VLOG_ERR("Failed to enable receiving packets in dpif.");
683 udpif_recv_set(backer->udpif, n_handler_threads,
684 backer->recv_set_enable);
685 dpif_flow_flush(backer->dpif);
686 backer->need_revalidate = REV_RECONFIGURE;
689 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
690 * to reset the handler threads. */
691 if (backer->n_handler_threads != n_handler_threads) {
692 udpif_recv_set(backer->udpif, n_handler_threads,
693 backer->recv_set_enable);
694 backer->n_handler_threads = n_handler_threads;
697 if (backer->need_revalidate) {
698 struct ofproto_dpif *ofproto;
699 struct simap_node *node;
700 struct simap tmp_backers;
702 /* Handle tunnel garbage collection. */
703 simap_init(&tmp_backers);
704 simap_swap(&backer->tnl_backers, &tmp_backers);
706 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
707 struct ofport_dpif *iter;
709 if (backer != ofproto->backer) {
713 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
714 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
717 if (!iter->is_tunnel) {
721 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
722 namebuf, sizeof namebuf);
723 node = simap_find(&tmp_backers, dp_port);
725 simap_put(&backer->tnl_backers, dp_port, node->data);
726 simap_delete(&tmp_backers, node);
727 node = simap_find(&backer->tnl_backers, dp_port);
729 node = simap_find(&backer->tnl_backers, dp_port);
731 odp_port_t odp_port = ODPP_NONE;
733 if (!dpif_port_add(backer->dpif, iter->up.netdev,
735 simap_put(&backer->tnl_backers, dp_port,
736 odp_to_u32(odp_port));
737 node = simap_find(&backer->tnl_backers, dp_port);
742 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
743 if (tnl_port_reconfigure(iter, iter->up.netdev,
745 backer->need_revalidate = REV_RECONFIGURE;
750 SIMAP_FOR_EACH (node, &tmp_backers) {
751 dpif_port_del(backer->dpif, u32_to_odp(node->data));
753 simap_destroy(&tmp_backers);
755 switch (backer->need_revalidate) {
756 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
757 case REV_STP: COVERAGE_INC(rev_stp); break;
758 case REV_BOND: COVERAGE_INC(rev_bond); break;
759 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
760 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
761 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
762 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
764 backer->need_revalidate = 0;
766 /* Clear the drop_keys in case we should now be accepting some
767 * formerly dropped flows. */
768 drop_key_clear(backer);
770 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
771 struct facet *facet, *next;
772 struct ofport_dpif *ofport;
773 struct cls_cursor cursor;
774 struct ofbundle *bundle;
776 if (ofproto->backer != backer) {
780 ovs_rwlock_wrlock(&xlate_rwlock);
781 xlate_ofproto_set(ofproto, ofproto->up.name,
782 ofproto->backer->dpif, ofproto->miss_rule,
783 ofproto->no_packet_in_rule, ofproto->ml,
784 ofproto->stp, ofproto->mbridge,
785 ofproto->sflow, ofproto->ipfix,
786 ofproto->up.frag_handling,
787 ofproto->up.forward_bpdu,
788 connmgr_has_in_band(ofproto->up.connmgr),
789 ofproto->netflow != NULL);
791 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
792 xlate_bundle_set(ofproto, bundle, bundle->name,
793 bundle->vlan_mode, bundle->vlan,
794 bundle->trunks, bundle->use_priority_tags,
795 bundle->bond, bundle->lacp,
799 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
800 int stp_port = ofport->stp_port
801 ? stp_port_no(ofport->stp_port)
803 xlate_ofport_set(ofproto, ofport->bundle, ofport,
804 ofport->up.ofp_port, ofport->odp_port,
805 ofport->up.netdev, ofport->cfm,
806 ofport->bfd, ofport->peer, stp_port,
807 ofport->qdscp, ofport->n_qdscp,
808 ofport->up.pp.config, ofport->is_tunnel,
811 ovs_rwlock_unlock(&xlate_rwlock);
813 /* Only ofproto-dpif cares about the facet classifier so we just
814 * lock cls_cursor_init() to appease the thread safety analysis. */
815 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
816 cls_cursor_init(&cursor, &ofproto->facets, NULL);
817 ovs_rwlock_unlock(&ofproto->facets.rwlock);
818 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
819 facet_revalidate(facet);
824 udpif_revalidate(backer->udpif);
827 if (!backer->recv_set_enable) {
828 /* Wake up before a max of 1000ms. */
829 timer_set_duration(&backer->next_expiration, 1000);
830 } else if (timer_expired(&backer->next_expiration)) {
831 int delay = expire(backer);
832 timer_set_duration(&backer->next_expiration, delay);
835 process_dpif_port_changes(backer);
837 if (backer->governor) {
840 governor_run(backer->governor);
842 /* If the governor has shrunk to its minimum size and the number of
843 * subfacets has dwindled, then drop the governor entirely.
845 * For hysteresis, the number of subfacets to drop the governor is
846 * smaller than the number needed to trigger its creation. */
847 n_subfacets = hmap_count(&backer->subfacets);
848 if (n_subfacets * 4 < flow_eviction_threshold
849 && governor_is_idle(backer->governor)) {
850 governor_destroy(backer->governor);
851 backer->governor = NULL;
858 /* Check for and handle port changes in 'backer''s dpif. */
860 process_dpif_port_changes(struct dpif_backer *backer)
866 error = dpif_port_poll(backer->dpif, &devname);
872 process_dpif_all_ports_changed(backer);
876 process_dpif_port_change(backer, devname);
881 process_dpif_port_error(backer, error);
888 process_dpif_all_ports_changed(struct dpif_backer *backer)
890 struct ofproto_dpif *ofproto;
891 struct dpif_port dpif_port;
892 struct dpif_port_dump dump;
893 struct sset devnames;
896 sset_init(&devnames);
897 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
898 if (ofproto->backer == backer) {
899 struct ofport *ofport;
901 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
902 sset_add(&devnames, netdev_get_name(ofport->netdev));
906 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
907 sset_add(&devnames, dpif_port.name);
910 SSET_FOR_EACH (devname, &devnames) {
911 process_dpif_port_change(backer, devname);
913 sset_destroy(&devnames);
917 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
919 struct ofproto_dpif *ofproto;
920 struct dpif_port port;
922 /* Don't report on the datapath's device. */
923 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
927 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
928 &all_ofproto_dpifs) {
929 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
934 ofproto = lookup_ofproto_dpif_by_port_name(devname);
935 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
936 /* The port was removed. If we know the datapath,
937 * report it through poll_set(). If we don't, it may be
938 * notifying us of a removal we initiated, so ignore it.
939 * If there's a pending ENOBUFS, let it stand, since
940 * everything will be reevaluated. */
941 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
942 sset_add(&ofproto->port_poll_set, devname);
943 ofproto->port_poll_errno = 0;
945 } else if (!ofproto) {
946 /* The port was added, but we don't know with which
947 * ofproto we should associate it. Delete it. */
948 dpif_port_del(backer->dpif, port.port_no);
950 struct ofport_dpif *ofport;
952 ofport = ofport_dpif_cast(shash_find_data(
953 &ofproto->up.port_by_name, devname));
955 && ofport->odp_port != port.port_no
956 && !odp_port_to_ofport(backer, port.port_no))
958 /* 'ofport''s datapath port number has changed from
959 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
960 * structures to match. */
961 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
962 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
963 ofport->odp_port = port.port_no;
964 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
965 hash_odp_port(port.port_no));
966 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
967 backer->need_revalidate = REV_RECONFIGURE;
970 dpif_port_destroy(&port);
973 /* Propagate 'error' to all ofprotos based on 'backer'. */
975 process_dpif_port_error(struct dpif_backer *backer, int error)
977 struct ofproto_dpif *ofproto;
979 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
980 if (ofproto->backer == backer) {
981 sset_clear(&ofproto->port_poll_set);
982 ofproto->port_poll_errno = error;
988 dpif_backer_run_fast(struct dpif_backer *backer)
990 handle_upcalls(backer);
996 type_run_fast(const char *type)
998 struct dpif_backer *backer;
1000 backer = shash_find_data(&all_dpif_backers, type);
1002 /* This is not necessarily a problem, since backers are only
1003 * created on demand. */
1007 return dpif_backer_run_fast(backer);
1013 static long long int port_rl = LLONG_MIN;
1015 if (time_msec() >= port_rl) {
1016 struct ofproto_dpif *ofproto;
1018 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1019 run_fast(&ofproto->up);
1021 port_rl = time_msec() + 200;
1026 type_wait(const char *type)
1028 struct dpif_backer *backer;
1030 backer = shash_find_data(&all_dpif_backers, type);
1032 /* This is not necessarily a problem, since backers are only
1033 * created on demand. */
1037 if (backer->governor) {
1038 governor_wait(backer->governor);
1041 timer_wait(&backer->next_expiration);
1042 dpif_wait(backer->dpif);
1043 udpif_wait(backer->udpif);
1046 /* Basic life-cycle. */
1048 static int add_internal_flows(struct ofproto_dpif *);
1050 static struct ofproto *
1053 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1054 return &ofproto->up;
1058 dealloc(struct ofproto *ofproto_)
1060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1065 close_dpif_backer(struct dpif_backer *backer)
1067 ovs_assert(backer->refcount > 0);
1069 if (--backer->refcount) {
1073 drop_key_clear(backer);
1074 hmap_destroy(&backer->drop_keys);
1076 udpif_destroy(backer->udpif);
1078 simap_destroy(&backer->tnl_backers);
1079 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1080 hmap_destroy(&backer->odp_to_ofport_map);
1081 shash_find_and_delete(&all_dpif_backers, backer->type);
1083 dpif_close(backer->dpif);
1085 ovs_assert(hmap_is_empty(&backer->subfacets));
1086 hmap_destroy(&backer->subfacets);
1087 governor_destroy(backer->governor);
1092 /* Datapath port slated for removal from datapath. */
1093 struct odp_garbage {
1094 struct list list_node;
1095 odp_port_t odp_port;
1099 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1101 struct dpif_backer *backer;
1102 struct dpif_port_dump port_dump;
1103 struct dpif_port port;
1104 struct shash_node *node;
1105 struct list garbage_list;
1106 struct odp_garbage *garbage, *next;
1112 backer = shash_find_data(&all_dpif_backers, type);
1119 backer_name = xasprintf("ovs-%s", type);
1121 /* Remove any existing datapaths, since we assume we're the only
1122 * userspace controlling the datapath. */
1124 dp_enumerate_names(type, &names);
1125 SSET_FOR_EACH(name, &names) {
1126 struct dpif *old_dpif;
1128 /* Don't remove our backer if it exists. */
1129 if (!strcmp(name, backer_name)) {
1133 if (dpif_open(name, type, &old_dpif)) {
1134 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1136 dpif_delete(old_dpif);
1137 dpif_close(old_dpif);
1140 sset_destroy(&names);
1142 backer = xmalloc(sizeof *backer);
1144 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1147 VLOG_ERR("failed to open datapath of type %s: %s", type,
1148 ovs_strerror(error));
1152 backer->udpif = udpif_create(backer, backer->dpif);
1154 backer->type = xstrdup(type);
1155 backer->governor = NULL;
1156 backer->refcount = 1;
1157 hmap_init(&backer->odp_to_ofport_map);
1158 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1159 hmap_init(&backer->drop_keys);
1160 hmap_init(&backer->subfacets);
1161 timer_set_duration(&backer->next_expiration, 1000);
1162 backer->need_revalidate = 0;
1163 simap_init(&backer->tnl_backers);
1164 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1167 if (backer->recv_set_enable) {
1168 dpif_flow_flush(backer->dpif);
1171 /* Loop through the ports already on the datapath and remove any
1172 * that we don't need anymore. */
1173 list_init(&garbage_list);
1174 dpif_port_dump_start(&port_dump, backer->dpif);
1175 while (dpif_port_dump_next(&port_dump, &port)) {
1176 node = shash_find(&init_ofp_ports, port.name);
1177 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1178 garbage = xmalloc(sizeof *garbage);
1179 garbage->odp_port = port.port_no;
1180 list_push_front(&garbage_list, &garbage->list_node);
1183 dpif_port_dump_done(&port_dump);
1185 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1186 dpif_port_del(backer->dpif, garbage->odp_port);
1187 list_remove(&garbage->list_node);
1191 shash_add(&all_dpif_backers, type, backer);
1193 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1195 VLOG_ERR("failed to listen on datapath of type %s: %s",
1196 type, ovs_strerror(error));
1197 close_dpif_backer(backer);
1200 udpif_recv_set(backer->udpif, n_handler_threads,
1201 backer->recv_set_enable);
1202 backer->n_handler_threads = n_handler_threads;
1204 backer->max_n_subfacet = 0;
1205 backer->created = time_msec();
1206 backer->avg_n_subfacet = 0;
1207 backer->avg_subfacet_life = 0;
1213 construct(struct ofproto *ofproto_)
1215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1216 struct shash_node *node, *next;
1220 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1225 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1226 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1228 ofproto->netflow = NULL;
1229 ofproto->sflow = NULL;
1230 ofproto->ipfix = NULL;
1231 ofproto->stp = NULL;
1232 hmap_init(&ofproto->bundles);
1233 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1234 ofproto->mbridge = mbridge_create();
1235 ofproto->has_bonded_bundles = false;
1236 ovs_mutex_init(&ofproto->vsp_mutex);
1238 classifier_init(&ofproto->facets);
1239 ofproto->consistency_rl = LLONG_MIN;
1241 guarded_list_init(&ofproto->pins);
1243 ofproto_dpif_unixctl_init();
1245 hmap_init(&ofproto->vlandev_map);
1246 hmap_init(&ofproto->realdev_vid_map);
1248 sset_init(&ofproto->ports);
1249 sset_init(&ofproto->ghost_ports);
1250 sset_init(&ofproto->port_poll_set);
1251 ofproto->port_poll_errno = 0;
1253 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1254 struct iface_hint *iface_hint = node->data;
1256 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1257 /* Check if the datapath already has this port. */
1258 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1259 sset_add(&ofproto->ports, node->name);
1262 free(iface_hint->br_name);
1263 free(iface_hint->br_type);
1265 shash_delete(&init_ofp_ports, node);
1269 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1270 hash_string(ofproto->up.name, 0));
1271 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1273 ofproto_init_tables(ofproto_, N_TABLES);
1274 error = add_internal_flows(ofproto);
1275 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1278 ofproto->n_missed = 0;
1284 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1285 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1287 struct ofputil_flow_mod fm;
1290 match_init_catchall(&fm.match);
1292 match_set_reg(&fm.match, 0, id);
1293 fm.new_cookie = htonll(0);
1294 fm.cookie = htonll(0);
1295 fm.cookie_mask = htonll(0);
1296 fm.modify_cookie = false;
1297 fm.table_id = TBL_INTERNAL;
1298 fm.command = OFPFC_ADD;
1299 fm.idle_timeout = 0;
1300 fm.hard_timeout = 0;
1304 fm.ofpacts = ofpacts->data;
1305 fm.ofpacts_len = ofpacts->size;
1307 error = ofproto_flow_mod(&ofproto->up, &fm);
1309 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1310 id, ofperr_to_string(error));
1314 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1316 rule_dpif_unref(*rulep);
1325 add_internal_flows(struct ofproto_dpif *ofproto)
1327 struct ofpact_controller *controller;
1328 uint64_t ofpacts_stub[128 / 8];
1329 struct ofpbuf ofpacts;
1333 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1336 controller = ofpact_put_CONTROLLER(&ofpacts);
1337 controller->max_len = UINT16_MAX;
1338 controller->controller_id = 0;
1339 controller->reason = OFPR_NO_MATCH;
1340 ofpact_pad(&ofpacts);
1342 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1347 ofpbuf_clear(&ofpacts);
1348 error = add_internal_flow(ofproto, id++, &ofpacts,
1349 &ofproto->no_packet_in_rule);
1354 error = add_internal_flow(ofproto, id++, &ofpacts,
1355 &ofproto->drop_frags_rule);
1360 destruct(struct ofproto *ofproto_)
1362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1363 struct rule_dpif *rule, *next_rule;
1364 struct ofputil_packet_in *pin, *next_pin;
1365 struct facet *facet, *next_facet;
1366 struct cls_cursor cursor;
1367 struct oftable *table;
1370 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1371 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1372 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1373 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1374 facet_remove(facet);
1377 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1378 ovs_rwlock_wrlock(&xlate_rwlock);
1379 xlate_remove_ofproto(ofproto);
1380 ovs_rwlock_unlock(&xlate_rwlock);
1382 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1383 * use-after-free error. */
1384 udpif_revalidate(ofproto->backer->udpif);
1386 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1388 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1389 struct cls_cursor cursor;
1391 ovs_rwlock_rdlock(&table->cls.rwlock);
1392 cls_cursor_init(&cursor, &table->cls, NULL);
1393 ovs_rwlock_unlock(&table->cls.rwlock);
1394 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1395 ofproto_rule_delete(&ofproto->up, &rule->up);
1399 guarded_list_pop_all(&ofproto->pins, &pins);
1400 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1401 list_remove(&pin->list_node);
1402 free(CONST_CAST(void *, pin->packet));
1405 guarded_list_destroy(&ofproto->pins);
1407 mbridge_unref(ofproto->mbridge);
1409 netflow_destroy(ofproto->netflow);
1410 dpif_sflow_unref(ofproto->sflow);
1411 hmap_destroy(&ofproto->bundles);
1412 mac_learning_unref(ofproto->ml);
1414 classifier_destroy(&ofproto->facets);
1416 hmap_destroy(&ofproto->vlandev_map);
1417 hmap_destroy(&ofproto->realdev_vid_map);
1419 sset_destroy(&ofproto->ports);
1420 sset_destroy(&ofproto->ghost_ports);
1421 sset_destroy(&ofproto->port_poll_set);
1423 ovs_mutex_destroy(&ofproto->vsp_mutex);
1425 close_dpif_backer(ofproto->backer);
1429 run_fast(struct ofproto *ofproto_)
1431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1432 struct ofputil_packet_in *pin, *next_pin;
1433 struct ofport_dpif *ofport;
1436 /* Do not perform any periodic activity required by 'ofproto' while
1437 * waiting for flow restore to complete. */
1438 if (ofproto_get_flow_restore_wait()) {
1442 guarded_list_pop_all(&ofproto->pins, &pins);
1443 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1444 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1445 list_remove(&pin->list_node);
1446 free(CONST_CAST(void *, pin->packet));
1450 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1451 port_run_fast(ofport);
1458 run(struct ofproto *ofproto_)
1460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1461 struct ofport_dpif *ofport;
1462 struct ofbundle *bundle;
1465 if (mbridge_need_revalidate(ofproto->mbridge)) {
1466 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1467 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1468 mac_learning_flush(ofproto->ml);
1469 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1472 /* Do not perform any periodic activity below required by 'ofproto' while
1473 * waiting for flow restore to complete. */
1474 if (ofproto_get_flow_restore_wait()) {
1478 error = run_fast(ofproto_);
1483 if (ofproto->netflow) {
1484 if (netflow_run(ofproto->netflow)) {
1485 send_netflow_active_timeouts(ofproto);
1488 if (ofproto->sflow) {
1489 dpif_sflow_run(ofproto->sflow);
1491 if (ofproto->ipfix) {
1492 dpif_ipfix_run(ofproto->ipfix);
1495 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1498 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1503 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1504 if (mac_learning_run(ofproto->ml)) {
1505 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1507 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1509 /* Check the consistency of a random facet, to aid debugging. */
1510 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1511 if (time_msec() >= ofproto->consistency_rl
1512 && !classifier_is_empty(&ofproto->facets)
1513 && !ofproto->backer->need_revalidate) {
1514 struct cls_table *table;
1515 struct cls_rule *cr;
1516 struct facet *facet;
1518 ofproto->consistency_rl = time_msec() + 250;
1520 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1521 struct cls_table, hmap_node);
1522 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1524 facet = CONTAINER_OF(cr, struct facet, cr);
1526 if (!facet_check_consistency(facet)) {
1527 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1530 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1536 wait(struct ofproto *ofproto_)
1538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1539 struct ofport_dpif *ofport;
1540 struct ofbundle *bundle;
1542 if (ofproto_get_flow_restore_wait()) {
1546 if (ofproto->sflow) {
1547 dpif_sflow_wait(ofproto->sflow);
1549 if (ofproto->ipfix) {
1550 dpif_ipfix_wait(ofproto->ipfix);
1552 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1555 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1556 bundle_wait(bundle);
1558 if (ofproto->netflow) {
1559 netflow_wait(ofproto->netflow);
1561 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1562 mac_learning_wait(ofproto->ml);
1563 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1565 if (ofproto->backer->need_revalidate) {
1566 /* Shouldn't happen, but if it does just go around again. */
1567 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1568 poll_immediate_wake();
1573 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1575 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1576 struct cls_cursor cursor;
1577 size_t n_subfacets = 0;
1578 struct facet *facet;
1580 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1581 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1582 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1584 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1585 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1586 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1587 n_subfacets += list_size(&facet->subfacets);
1589 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1590 simap_increase(usage, "subfacets", n_subfacets);
1594 flush(struct ofproto *ofproto_)
1596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597 struct subfacet *subfacet, *next_subfacet;
1598 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1602 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1603 &ofproto->backer->subfacets) {
1604 if (subfacet->facet->ofproto != ofproto) {
1608 if (subfacet->path != SF_NOT_INSTALLED) {
1609 batch[n_batch++] = subfacet;
1610 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1611 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1615 subfacet_destroy(subfacet);
1620 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1625 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1626 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1628 *arp_match_ip = true;
1629 *actions = (OFPUTIL_A_OUTPUT |
1630 OFPUTIL_A_SET_VLAN_VID |
1631 OFPUTIL_A_SET_VLAN_PCP |
1632 OFPUTIL_A_STRIP_VLAN |
1633 OFPUTIL_A_SET_DL_SRC |
1634 OFPUTIL_A_SET_DL_DST |
1635 OFPUTIL_A_SET_NW_SRC |
1636 OFPUTIL_A_SET_NW_DST |
1637 OFPUTIL_A_SET_NW_TOS |
1638 OFPUTIL_A_SET_TP_SRC |
1639 OFPUTIL_A_SET_TP_DST |
1644 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1647 struct dpif_dp_stats s;
1648 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1651 strcpy(ots->name, "classifier");
1653 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1654 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1655 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1656 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1658 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1659 ots->lookup_count = htonll(n_lookup);
1660 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1663 static struct ofport *
1666 struct ofport_dpif *port = xmalloc(sizeof *port);
1671 port_dealloc(struct ofport *port_)
1673 struct ofport_dpif *port = ofport_dpif_cast(port_);
1678 port_construct(struct ofport *port_)
1680 struct ofport_dpif *port = ofport_dpif_cast(port_);
1681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1682 const struct netdev *netdev = port->up.netdev;
1683 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1684 struct dpif_port dpif_port;
1687 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1688 port->bundle = NULL;
1691 port->may_enable = true;
1692 port->stp_port = NULL;
1693 port->stp_state = STP_DISABLED;
1694 port->is_tunnel = false;
1698 port->realdev_ofp_port = 0;
1699 port->vlandev_vid = 0;
1700 port->carrier_seq = netdev_get_carrier_resets(netdev);
1702 if (netdev_vport_is_patch(netdev)) {
1703 /* By bailing out here, we don't submit the port to the sFlow module
1704 * to be considered for counter polling export. This is correct
1705 * because the patch port represents an interface that sFlow considers
1706 * to be "internal" to the switch as a whole, and therefore not an
1707 * candidate for counter polling. */
1708 port->odp_port = ODPP_NONE;
1709 ofport_update_peer(port);
1713 error = dpif_port_query_by_name(ofproto->backer->dpif,
1714 netdev_vport_get_dpif_port(netdev, namebuf,
1721 port->odp_port = dpif_port.port_no;
1723 if (netdev_get_tunnel_config(netdev)) {
1724 tnl_port_add(port, port->up.netdev, port->odp_port);
1725 port->is_tunnel = true;
1727 /* Sanity-check that a mapping doesn't already exist. This
1728 * shouldn't happen for non-tunnel ports. */
1729 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1730 VLOG_ERR("port %s already has an OpenFlow port number",
1732 dpif_port_destroy(&dpif_port);
1736 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1737 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1738 hash_odp_port(port->odp_port));
1739 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1741 dpif_port_destroy(&dpif_port);
1743 if (ofproto->sflow) {
1744 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1751 port_destruct(struct ofport *port_)
1753 struct ofport_dpif *port = ofport_dpif_cast(port_);
1754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1755 const char *devname = netdev_get_name(port->up.netdev);
1756 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1757 const char *dp_port_name;
1759 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1760 ovs_rwlock_wrlock(&xlate_rwlock);
1761 xlate_ofport_remove(port);
1762 ovs_rwlock_unlock(&xlate_rwlock);
1764 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1766 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1767 /* The underlying device is still there, so delete it. This
1768 * happens when the ofproto is being destroyed, since the caller
1769 * assumes that removal of attached ports will happen as part of
1771 if (!port->is_tunnel) {
1772 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1777 port->peer->peer = NULL;
1781 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1782 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1783 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1784 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1788 sset_find_and_delete(&ofproto->ports, devname);
1789 sset_find_and_delete(&ofproto->ghost_ports, devname);
1790 bundle_remove(port_);
1791 set_cfm(port_, NULL);
1792 set_bfd(port_, NULL);
1793 if (ofproto->sflow) {
1794 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1801 port_modified(struct ofport *port_)
1803 struct ofport_dpif *port = ofport_dpif_cast(port_);
1805 if (port->bundle && port->bundle->bond) {
1806 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1810 cfm_set_netdev(port->cfm, port->up.netdev);
1814 bfd_set_netdev(port->bfd, port->up.netdev);
1817 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1819 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1823 ofport_update_peer(port);
1827 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1829 struct ofport_dpif *port = ofport_dpif_cast(port_);
1830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1831 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1833 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1834 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1835 OFPUTIL_PC_NO_PACKET_IN)) {
1836 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1838 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1839 bundle_update(port->bundle);
1845 set_sflow(struct ofproto *ofproto_,
1846 const struct ofproto_sflow_options *sflow_options)
1848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1849 struct dpif_sflow *ds = ofproto->sflow;
1851 if (sflow_options) {
1853 struct ofport_dpif *ofport;
1855 ds = ofproto->sflow = dpif_sflow_create();
1856 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1857 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861 dpif_sflow_set_options(ds, sflow_options);
1864 dpif_sflow_unref(ds);
1865 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1866 ofproto->sflow = NULL;
1874 struct ofproto *ofproto_,
1875 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1876 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1877 size_t n_flow_exporters_options)
1879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1880 struct dpif_ipfix *di = ofproto->ipfix;
1881 bool has_options = bridge_exporter_options || flow_exporters_options;
1883 if (has_options && !di) {
1884 di = ofproto->ipfix = dpif_ipfix_create();
1888 /* Call set_options in any case to cleanly flush the flow
1889 * caches in the last exporters that are to be destroyed. */
1890 dpif_ipfix_set_options(
1891 di, bridge_exporter_options, flow_exporters_options,
1892 n_flow_exporters_options);
1895 dpif_ipfix_unref(di);
1896 ofproto->ipfix = NULL;
1904 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1906 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1913 struct ofproto_dpif *ofproto;
1915 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1916 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1917 ofport->cfm = cfm_create(ofport->up.netdev);
1920 if (cfm_configure(ofport->cfm, s)) {
1926 cfm_unref(ofport->cfm);
1932 get_cfm_status(const struct ofport *ofport_,
1933 struct ofproto_cfm_status *status)
1935 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1938 status->faults = cfm_get_fault(ofport->cfm);
1939 status->remote_opstate = cfm_get_opup(ofport->cfm);
1940 status->health = cfm_get_health(ofport->cfm);
1941 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1949 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1952 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1957 cfg, ofport->up.netdev);
1958 if (ofport->bfd != old) {
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1968 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1971 bfd_get_status(ofport->bfd, smap);
1978 /* Spanning Tree. */
1981 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1983 struct ofproto_dpif *ofproto = ofproto_;
1984 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1985 struct ofport_dpif *ofport;
1987 ofport = stp_port_get_aux(sp);
1989 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1990 ofproto->up.name, port_num);
1992 struct eth_header *eth = pkt->l2;
1994 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1995 if (eth_addr_is_zero(eth->eth_src)) {
1996 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1997 "with unknown MAC", ofproto->up.name, port_num);
1999 send_packet(ofport, pkt);
2005 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2007 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2011 /* Only revalidate flows if the configuration changed. */
2012 if (!s != !ofproto->stp) {
2013 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2017 if (!ofproto->stp) {
2018 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2019 send_bpdu_cb, ofproto);
2020 ofproto->stp_last_tick = time_msec();
2023 stp_set_bridge_id(ofproto->stp, s->system_id);
2024 stp_set_bridge_priority(ofproto->stp, s->priority);
2025 stp_set_hello_time(ofproto->stp, s->hello_time);
2026 stp_set_max_age(ofproto->stp, s->max_age);
2027 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2029 struct ofport *ofport;
2031 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2032 set_stp_port(ofport, NULL);
2035 stp_unref(ofproto->stp);
2036 ofproto->stp = NULL;
2043 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2049 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2050 s->designated_root = stp_get_designated_root(ofproto->stp);
2051 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2060 update_stp_port_state(struct ofport_dpif *ofport)
2062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2063 enum stp_state state;
2065 /* Figure out new state. */
2066 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2070 if (ofport->stp_state != state) {
2071 enum ofputil_port_state of_state;
2074 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2075 netdev_get_name(ofport->up.netdev),
2076 stp_state_name(ofport->stp_state),
2077 stp_state_name(state));
2078 if (stp_learn_in_state(ofport->stp_state)
2079 != stp_learn_in_state(state)) {
2080 /* xxx Learning action flows should also be flushed. */
2081 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2082 mac_learning_flush(ofproto->ml);
2083 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2085 fwd_change = stp_forward_in_state(ofport->stp_state)
2086 != stp_forward_in_state(state);
2088 ofproto->backer->need_revalidate = REV_STP;
2089 ofport->stp_state = state;
2090 ofport->stp_state_entered = time_msec();
2092 if (fwd_change && ofport->bundle) {
2093 bundle_update(ofport->bundle);
2096 /* Update the STP state bits in the OpenFlow port description. */
2097 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2098 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2099 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2100 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2101 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2103 ofproto_port_set_state(&ofport->up, of_state);
2107 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2108 * caller is responsible for assigning STP port numbers and ensuring
2109 * there are no duplicates. */
2111 set_stp_port(struct ofport *ofport_,
2112 const struct ofproto_port_stp_settings *s)
2114 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2116 struct stp_port *sp = ofport->stp_port;
2118 if (!s || !s->enable) {
2120 ofport->stp_port = NULL;
2121 stp_port_disable(sp);
2122 update_stp_port_state(ofport);
2125 } else if (sp && stp_port_no(sp) != s->port_num
2126 && ofport == stp_port_get_aux(sp)) {
2127 /* The port-id changed, so disable the old one if it's not
2128 * already in use by another port. */
2129 stp_port_disable(sp);
2132 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2133 stp_port_enable(sp);
2135 stp_port_set_aux(sp, ofport);
2136 stp_port_set_priority(sp, s->priority);
2137 stp_port_set_path_cost(sp, s->path_cost);
2139 update_stp_port_state(ofport);
2145 get_stp_port_status(struct ofport *ofport_,
2146 struct ofproto_port_stp_status *s)
2148 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2150 struct stp_port *sp = ofport->stp_port;
2152 if (!ofproto->stp || !sp) {
2158 s->port_id = stp_port_get_id(sp);
2159 s->state = stp_port_get_state(sp);
2160 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2161 s->role = stp_port_get_role(sp);
2162 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2168 stp_run(struct ofproto_dpif *ofproto)
2171 long long int now = time_msec();
2172 long long int elapsed = now - ofproto->stp_last_tick;
2173 struct stp_port *sp;
2176 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2177 ofproto->stp_last_tick = now;
2179 while (stp_get_changed_port(ofproto->stp, &sp)) {
2180 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2183 update_stp_port_state(ofport);
2187 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2188 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2189 mac_learning_flush(ofproto->ml);
2190 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2196 stp_wait(struct ofproto_dpif *ofproto)
2199 poll_timer_wait(1000);
2204 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2207 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2210 if (ofport->n_qdscp != n_qdscp
2211 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2212 n_qdscp * sizeof *qdscp))) {
2213 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2214 free(ofport->qdscp);
2215 ofport->qdscp = n_qdscp
2216 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2218 ofport->n_qdscp = n_qdscp;
2226 /* Expires all MAC learning entries associated with 'bundle' and forces its
2227 * ofproto to revalidate every flow.
2229 * Normally MAC learning entries are removed only from the ofproto associated
2230 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2231 * are removed from every ofproto. When patch ports and SLB bonds are in use
2232 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2233 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2234 * with the host from which it migrated. */
2236 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2238 struct ofproto_dpif *ofproto = bundle->ofproto;
2239 struct mac_learning *ml = ofproto->ml;
2240 struct mac_entry *mac, *next_mac;
2242 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2243 ovs_rwlock_wrlock(&ml->rwlock);
2244 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2245 if (mac->port.p == bundle) {
2247 struct ofproto_dpif *o;
2249 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2251 struct mac_entry *e;
2253 ovs_rwlock_wrlock(&o->ml->rwlock);
2254 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2256 mac_learning_expire(o->ml, e);
2258 ovs_rwlock_unlock(&o->ml->rwlock);
2263 mac_learning_expire(ml, mac);
2266 ovs_rwlock_unlock(&ml->rwlock);
2269 static struct ofbundle *
2270 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2272 struct ofbundle *bundle;
2274 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2275 &ofproto->bundles) {
2276 if (bundle->aux == aux) {
2284 bundle_update(struct ofbundle *bundle)
2286 struct ofport_dpif *port;
2288 bundle->floodable = true;
2289 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2290 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2291 || !stp_forward_in_state(port->stp_state)) {
2292 bundle->floodable = false;
2299 bundle_del_port(struct ofport_dpif *port)
2301 struct ofbundle *bundle = port->bundle;
2303 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2305 list_remove(&port->bundle_node);
2306 port->bundle = NULL;
2309 lacp_slave_unregister(bundle->lacp, port);
2312 bond_slave_unregister(bundle->bond, port);
2315 bundle_update(bundle);
2319 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2320 struct lacp_slave_settings *lacp)
2322 struct ofport_dpif *port;
2324 port = get_ofp_port(bundle->ofproto, ofp_port);
2329 if (port->bundle != bundle) {
2330 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2332 bundle_remove(&port->up);
2335 port->bundle = bundle;
2336 list_push_back(&bundle->ports, &port->bundle_node);
2337 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2338 || !stp_forward_in_state(port->stp_state)) {
2339 bundle->floodable = false;
2343 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344 lacp_slave_register(bundle->lacp, port, lacp);
2351 bundle_destroy(struct ofbundle *bundle)
2353 struct ofproto_dpif *ofproto;
2354 struct ofport_dpif *port, *next_port;
2360 ofproto = bundle->ofproto;
2361 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2363 ovs_rwlock_wrlock(&xlate_rwlock);
2364 xlate_bundle_remove(bundle);
2365 ovs_rwlock_unlock(&xlate_rwlock);
2367 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2368 bundle_del_port(port);
2371 bundle_flush_macs(bundle, true);
2372 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2374 free(bundle->trunks);
2375 lacp_unref(bundle->lacp);
2376 bond_unref(bundle->bond);
2381 bundle_set(struct ofproto *ofproto_, void *aux,
2382 const struct ofproto_bundle_settings *s)
2384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2385 bool need_flush = false;
2386 struct ofport_dpif *port;
2387 struct ofbundle *bundle;
2388 unsigned long *trunks;
2394 bundle_destroy(bundle_lookup(ofproto, aux));
2398 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2399 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2401 bundle = bundle_lookup(ofproto, aux);
2403 bundle = xmalloc(sizeof *bundle);
2405 bundle->ofproto = ofproto;
2406 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2407 hash_pointer(aux, 0));
2409 bundle->name = NULL;
2411 list_init(&bundle->ports);
2412 bundle->vlan_mode = PORT_VLAN_TRUNK;
2414 bundle->trunks = NULL;
2415 bundle->use_priority_tags = s->use_priority_tags;
2416 bundle->lacp = NULL;
2417 bundle->bond = NULL;
2419 bundle->floodable = true;
2420 mbridge_register_bundle(ofproto->mbridge, bundle);
2423 if (!bundle->name || strcmp(s->name, bundle->name)) {
2425 bundle->name = xstrdup(s->name);
2430 if (!bundle->lacp) {
2431 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432 bundle->lacp = lacp_create();
2434 lacp_configure(bundle->lacp, s->lacp);
2436 lacp_unref(bundle->lacp);
2437 bundle->lacp = NULL;
2440 /* Update set of ports. */
2442 for (i = 0; i < s->n_slaves; i++) {
2443 if (!bundle_add_port(bundle, s->slaves[i],
2444 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2448 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2449 struct ofport_dpif *next_port;
2451 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2452 for (i = 0; i < s->n_slaves; i++) {
2453 if (s->slaves[i] == port->up.ofp_port) {
2458 bundle_del_port(port);
2462 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2464 if (list_is_empty(&bundle->ports)) {
2465 bundle_destroy(bundle);
2469 /* Set VLAN tagging mode */
2470 if (s->vlan_mode != bundle->vlan_mode
2471 || s->use_priority_tags != bundle->use_priority_tags) {
2472 bundle->vlan_mode = s->vlan_mode;
2473 bundle->use_priority_tags = s->use_priority_tags;
2478 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2479 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2481 if (vlan != bundle->vlan) {
2482 bundle->vlan = vlan;
2486 /* Get trunked VLANs. */
2487 switch (s->vlan_mode) {
2488 case PORT_VLAN_ACCESS:
2492 case PORT_VLAN_TRUNK:
2493 trunks = CONST_CAST(unsigned long *, s->trunks);
2496 case PORT_VLAN_NATIVE_UNTAGGED:
2497 case PORT_VLAN_NATIVE_TAGGED:
2498 if (vlan != 0 && (!s->trunks
2499 || !bitmap_is_set(s->trunks, vlan)
2500 || bitmap_is_set(s->trunks, 0))) {
2501 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2503 trunks = bitmap_clone(s->trunks, 4096);
2505 trunks = bitmap_allocate1(4096);
2507 bitmap_set1(trunks, vlan);
2508 bitmap_set0(trunks, 0);
2510 trunks = CONST_CAST(unsigned long *, s->trunks);
2517 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2518 free(bundle->trunks);
2519 if (trunks == s->trunks) {
2520 bundle->trunks = vlan_bitmap_clone(trunks);
2522 bundle->trunks = trunks;
2527 if (trunks != s->trunks) {
2532 if (!list_is_short(&bundle->ports)) {
2533 bundle->ofproto->has_bonded_bundles = true;
2535 if (bond_reconfigure(bundle->bond, s->bond)) {
2536 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2539 bundle->bond = bond_create(s->bond);
2540 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2543 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2544 bond_slave_register(bundle->bond, port, port->up.netdev);
2547 bond_unref(bundle->bond);
2548 bundle->bond = NULL;
2551 /* If we changed something that would affect MAC learning, un-learn
2552 * everything on this port and force flow revalidation. */
2554 bundle_flush_macs(bundle, false);
2561 bundle_remove(struct ofport *port_)
2563 struct ofport_dpif *port = ofport_dpif_cast(port_);
2564 struct ofbundle *bundle = port->bundle;
2567 bundle_del_port(port);
2568 if (list_is_empty(&bundle->ports)) {
2569 bundle_destroy(bundle);
2570 } else if (list_is_short(&bundle->ports)) {
2571 bond_unref(bundle->bond);
2572 bundle->bond = NULL;
2578 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2580 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2581 struct ofport_dpif *port = port_;
2582 uint8_t ea[ETH_ADDR_LEN];
2585 error = netdev_get_etheraddr(port->up.netdev, ea);
2587 struct ofpbuf packet;
2590 ofpbuf_init(&packet, 0);
2591 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2593 memcpy(packet_pdu, pdu, pdu_size);
2595 send_packet(port, &packet);
2596 ofpbuf_uninit(&packet);
2598 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2599 "%s (%s)", port->bundle->name,
2600 netdev_get_name(port->up.netdev), ovs_strerror(error));
2605 bundle_send_learning_packets(struct ofbundle *bundle)
2607 struct ofproto_dpif *ofproto = bundle->ofproto;
2608 struct ofpbuf *learning_packet;
2609 int error, n_packets, n_errors;
2610 struct mac_entry *e;
2611 struct list packets;
2613 list_init(&packets);
2614 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2615 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2616 if (e->port.p != bundle) {
2619 learning_packet = bond_compose_learning_packet(bundle->bond,
2622 learning_packet->private_p = port_void;
2623 list_push_back(&packets, &learning_packet->list_node);
2626 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2628 error = n_packets = n_errors = 0;
2629 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2632 ret = send_packet(learning_packet->private_p, learning_packet);
2639 ofpbuf_list_delete(&packets);
2642 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2643 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2644 "packets, last error was: %s",
2645 bundle->name, n_errors, n_packets, ovs_strerror(error));
2647 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2648 bundle->name, n_packets);
2653 bundle_run(struct ofbundle *bundle)
2656 lacp_run(bundle->lacp, send_pdu_cb);
2659 struct ofport_dpif *port;
2661 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2662 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2665 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2666 bundle->ofproto->backer->need_revalidate = REV_BOND;
2669 if (bond_should_send_learning_packets(bundle->bond)) {
2670 bundle_send_learning_packets(bundle);
2676 bundle_wait(struct ofbundle *bundle)
2679 lacp_wait(bundle->lacp);
2682 bond_wait(bundle->bond);
2689 mirror_set__(struct ofproto *ofproto_, void *aux,
2690 const struct ofproto_mirror_settings *s)
2692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2693 struct ofbundle **srcs, **dsts;
2698 mirror_destroy(ofproto->mbridge, aux);
2702 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2703 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2705 for (i = 0; i < s->n_srcs; i++) {
2706 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2709 for (i = 0; i < s->n_dsts; i++) {
2710 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2713 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2714 s->n_dsts, s->src_vlans,
2715 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2722 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2723 uint64_t *packets, uint64_t *bytes)
2726 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2731 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2734 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2735 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2736 mac_learning_flush(ofproto->ml);
2738 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2743 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2746 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2747 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2751 forward_bpdu_changed(struct ofproto *ofproto_)
2753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2758 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2762 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2763 mac_learning_set_idle_time(ofproto->ml, idle_time);
2764 mac_learning_set_max_entries(ofproto->ml, max_entries);
2765 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2770 static struct ofport_dpif *
2771 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2773 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2774 return ofport ? ofport_dpif_cast(ofport) : NULL;
2777 static struct ofport_dpif *
2778 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2780 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2781 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2785 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2786 struct ofproto_port *ofproto_port,
2787 struct dpif_port *dpif_port)
2789 ofproto_port->name = dpif_port->name;
2790 ofproto_port->type = dpif_port->type;
2791 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2795 ofport_update_peer(struct ofport_dpif *ofport)
2797 const struct ofproto_dpif *ofproto;
2798 struct dpif_backer *backer;
2801 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2805 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2806 backer->need_revalidate = REV_RECONFIGURE;
2809 ofport->peer->peer = NULL;
2810 ofport->peer = NULL;
2813 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2818 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2819 struct ofport *peer_ofport;
2820 struct ofport_dpif *peer;
2823 if (ofproto->backer != backer) {
2827 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2832 peer = ofport_dpif_cast(peer_ofport);
2833 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2834 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2836 ofport->peer = peer;
2837 ofport->peer->peer = ofport;
2847 port_run_fast(struct ofport_dpif *ofport)
2849 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2850 struct ofpbuf packet;
2852 ofpbuf_init(&packet, 0);
2853 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2854 send_packet(ofport, &packet);
2855 ofpbuf_uninit(&packet);
2858 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2859 struct ofpbuf packet;
2861 ofpbuf_init(&packet, 0);
2862 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2863 send_packet(ofport, &packet);
2864 ofpbuf_uninit(&packet);
2869 port_run(struct ofport_dpif *ofport)
2871 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2872 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2873 bool enable = netdev_get_carrier(ofport->up.netdev);
2874 bool cfm_enable = false;
2875 bool bfd_enable = false;
2877 ofport->carrier_seq = carrier_seq;
2879 port_run_fast(ofport);
2882 int cfm_opup = cfm_get_opup(ofport->cfm);
2884 cfm_run(ofport->cfm);
2885 cfm_enable = !cfm_get_fault(ofport->cfm);
2887 if (cfm_opup >= 0) {
2888 cfm_enable = cfm_enable && cfm_opup;
2893 bfd_run(ofport->bfd);
2894 bfd_enable = bfd_forwarding(ofport->bfd);
2897 if (ofport->bfd || ofport->cfm) {
2898 enable = enable && (cfm_enable || bfd_enable);
2901 if (ofport->bundle) {
2902 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2903 if (carrier_changed) {
2904 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2908 if (ofport->may_enable != enable) {
2909 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2910 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2913 ofport->may_enable = enable;
2917 port_wait(struct ofport_dpif *ofport)
2920 cfm_wait(ofport->cfm);
2924 bfd_wait(ofport->bfd);
2929 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2930 struct ofproto_port *ofproto_port)
2932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2933 struct dpif_port dpif_port;
2936 if (sset_contains(&ofproto->ghost_ports, devname)) {
2937 const char *type = netdev_get_type_from_name(devname);
2939 /* We may be called before ofproto->up.port_by_name is populated with
2940 * the appropriate ofport. For this reason, we must get the name and
2941 * type from the netdev layer directly. */
2943 const struct ofport *ofport;
2945 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2946 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2947 ofproto_port->name = xstrdup(devname);
2948 ofproto_port->type = xstrdup(type);
2954 if (!sset_contains(&ofproto->ports, devname)) {
2957 error = dpif_port_query_by_name(ofproto->backer->dpif,
2958 devname, &dpif_port);
2960 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2966 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2969 const char *devname = netdev_get_name(netdev);
2970 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2971 const char *dp_port_name;
2973 if (netdev_vport_is_patch(netdev)) {
2974 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2978 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2979 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2980 odp_port_t port_no = ODPP_NONE;
2983 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2987 if (netdev_get_tunnel_config(netdev)) {
2988 simap_put(&ofproto->backer->tnl_backers,
2989 dp_port_name, odp_to_u32(port_no));
2993 if (netdev_get_tunnel_config(netdev)) {
2994 sset_add(&ofproto->ghost_ports, devname);
2996 sset_add(&ofproto->ports, devname);
3002 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3005 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3012 sset_find_and_delete(&ofproto->ghost_ports,
3013 netdev_get_name(ofport->up.netdev));
3014 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3015 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3016 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3018 /* The caller is going to close ofport->up.netdev. If this is a
3019 * bonded port, then the bond is using that netdev, so remove it
3020 * from the bond. The client will need to reconfigure everything
3021 * after deleting ports, so then the slave will get re-added. */
3022 bundle_remove(&ofport->up);
3029 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3031 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3036 error = netdev_get_stats(ofport->up.netdev, stats);
3038 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3041 /* ofproto->stats.tx_packets represents packets that we created
3042 * internally and sent to some port (e.g. packets sent with
3043 * send_packet()). Account for them as if they had come from
3044 * OFPP_LOCAL and got forwarded. */
3046 if (stats->rx_packets != UINT64_MAX) {
3047 stats->rx_packets += ofproto->stats.tx_packets;
3050 if (stats->rx_bytes != UINT64_MAX) {
3051 stats->rx_bytes += ofproto->stats.tx_bytes;
3054 /* ofproto->stats.rx_packets represents packets that were received on
3055 * some port and we processed internally and dropped (e.g. STP).
3056 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3058 if (stats->tx_packets != UINT64_MAX) {
3059 stats->tx_packets += ofproto->stats.rx_packets;
3062 if (stats->tx_bytes != UINT64_MAX) {
3063 stats->tx_bytes += ofproto->stats.rx_bytes;
3070 struct port_dump_state {
3075 struct ofproto_port port;
3080 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3082 *statep = xzalloc(sizeof(struct port_dump_state));
3087 port_dump_next(const struct ofproto *ofproto_, void *state_,
3088 struct ofproto_port *port)
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3091 struct port_dump_state *state = state_;
3092 const struct sset *sset;
3093 struct sset_node *node;
3095 if (state->has_port) {
3096 ofproto_port_destroy(&state->port);
3097 state->has_port = false;
3099 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3100 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3103 error = port_query_by_name(ofproto_, node->name, &state->port);
3105 *port = state->port;
3106 state->has_port = true;
3108 } else if (error != ENODEV) {
3113 if (!state->ghost) {
3114 state->ghost = true;
3117 return port_dump_next(ofproto_, state_, port);
3124 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3126 struct port_dump_state *state = state_;
3128 if (state->has_port) {
3129 ofproto_port_destroy(&state->port);
3136 port_poll(const struct ofproto *ofproto_, char **devnamep)
3138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3140 if (ofproto->port_poll_errno) {
3141 int error = ofproto->port_poll_errno;
3142 ofproto->port_poll_errno = 0;
3146 if (sset_is_empty(&ofproto->port_poll_set)) {
3150 *devnamep = sset_pop(&ofproto->port_poll_set);
3155 port_poll_wait(const struct ofproto *ofproto_)
3157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3158 dpif_port_poll_wait(ofproto->backer->dpif);
3162 port_is_lacp_current(const struct ofport *ofport_)
3164 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3165 return (ofport->bundle && ofport->bundle->lacp
3166 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3170 /* Upcall handling. */
3172 struct flow_miss_op {
3173 struct dpif_op dpif_op;
3175 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3176 struct xlate_out xout;
3177 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3179 struct ofpbuf mask; /* Flow mask for "put" ops. */
3180 struct odputil_keybuf maskbuf;
3182 /* If this is a "put" op, then a pointer to the subfacet that should
3183 * be marked as uninstalled if the operation fails. */
3184 struct subfacet *subfacet;
3187 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3188 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3189 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3190 * return value of true). However, for short flows the cost of bookkeeping is
3191 * much higher than the benefits, so when the datapath holds a large number of
3192 * flows we impose some heuristics to decide which flows are likely to be worth
3195 flow_miss_should_make_facet(struct flow_miss *miss)
3197 struct dpif_backer *backer = miss->ofproto->backer;
3200 switch (flow_miss_model) {
3201 case OFPROTO_HANDLE_MISS_AUTO:
3203 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3205 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3209 if (!backer->governor) {
3212 n_subfacets = hmap_count(&backer->subfacets);
3213 if (n_subfacets * 2 <= flow_eviction_threshold) {
3217 backer->governor = governor_create();
3220 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3221 return governor_should_install_flow(backer->governor, hash,
3222 miss->stats.n_packets);
3225 /* Handles 'miss', which matches 'facet'. May add any required datapath
3226 * operations to 'ops', incrementing '*n_ops' for each new op.
3228 * All of the packets in 'miss' are considered to have arrived at time
3229 * 'miss->stats.used'. This is really important only for new facets: if we
3230 * just called time_msec() here, then the new subfacet or its packets could
3231 * look (occasionally) as though it was used some time after the facet was
3232 * used. That can make a one-packet flow look like it has a nonzero duration,
3233 * which looks odd in e.g. NetFlow statistics. */
3235 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3236 struct flow_miss_op *ops, size_t *n_ops)
3238 enum subfacet_path want_path;
3239 struct subfacet *subfacet;
3241 facet->packet_count += miss->stats.n_packets;
3242 facet->prev_packet_count += miss->stats.n_packets;
3243 facet->byte_count += miss->stats.n_bytes;
3244 facet->prev_byte_count += miss->stats.n_bytes;
3246 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3248 /* Don't install the flow if it's the result of the "userspace"
3249 * action for an already installed facet. This can occur when a
3250 * datapath flow with wildcards has a "userspace" action and flows
3251 * sent to userspace result in a different subfacet, which will then
3252 * be rejected as overlapping by the datapath. */
3253 if (miss->upcall_type == DPIF_UC_ACTION
3254 && !list_is_empty(&facet->subfacets)) {
3258 subfacet = subfacet_create(facet, miss);
3259 if (subfacet->path != want_path) {
3260 struct flow_miss_op *op = &ops[(*n_ops)++];
3261 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3263 subfacet->path = want_path;
3265 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3266 if (enable_megaflows) {
3267 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3268 &miss->flow, UINT32_MAX);
3271 op->xout_garbage = false;
3272 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3273 op->subfacet = subfacet;
3274 put->flags = DPIF_FP_CREATE;
3275 put->key = miss->key;
3276 put->key_len = miss->key_len;
3277 put->mask = op->mask.data;
3278 put->mask_len = op->mask.size;
3280 if (want_path == SF_FAST_PATH) {
3281 put->actions = facet->xout.odp_actions.data;
3282 put->actions_len = facet->xout.odp_actions.size;
3284 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3285 op->slow_stub, sizeof op->slow_stub,
3286 &put->actions, &put->actions_len);
3292 /* Handles flow miss 'miss'. May add any required datapath operations
3293 * to 'ops', incrementing '*n_ops' for each new op. */
3295 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3298 struct facet *facet;
3300 miss->ofproto->n_missed += miss->stats.n_packets;
3302 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3304 /* There does not exist a bijection between 'struct flow' and datapath
3305 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3306 * assumption used throughout the facet and subfacet handling code.
3307 * Since we have to handle these misses in userspace anyway, we simply
3308 * skip facet creation, avoiding the problem altogether. */
3309 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3310 || !flow_miss_should_make_facet(miss)) {
3314 facet = facet_create(miss);
3316 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3319 static struct drop_key *
3320 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3323 struct drop_key *drop_key;
3325 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3326 &backer->drop_keys) {
3327 if (drop_key->key_len == key_len
3328 && !memcmp(drop_key->key, key, key_len)) {
3336 drop_key_clear(struct dpif_backer *backer)
3338 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3339 struct drop_key *drop_key, *next;
3341 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3344 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3346 if (error && !VLOG_DROP_WARN(&rl)) {
3347 struct ds ds = DS_EMPTY_INITIALIZER;
3348 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3349 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3350 ovs_strerror(error), ds_cstr(&ds));
3354 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3355 drop_key_destroy(drop_key);
3358 udpif_drop_key_clear(backer->udpif);
3362 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3364 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3365 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3366 struct flow_miss *miss;
3369 /* Process each element in the to-do list, constructing the set of
3370 * operations to batch. */
3372 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3373 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3375 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3377 /* Execute batch. */
3378 for (i = 0; i < n_ops; i++) {
3379 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3381 dpif_operate(backer->dpif, dpif_ops, n_ops);
3383 for (i = 0; i < n_ops; i++) {
3384 if (dpif_ops[i]->error != 0
3385 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3386 && flow_miss_ops[i].subfacet) {
3387 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3389 COVERAGE_INC(subfacet_install_fail);
3391 /* Zero-out subfacet counters when installation failed, but
3392 * datapath reported hits. This should not happen and
3393 * indicates a bug, since if the datapath flow exists, we
3394 * should not be attempting to create a new subfacet. A
3395 * buggy datapath could trigger this, so just zero out the
3396 * counters and log an error. */
3397 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3398 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3399 "datapath reported hits");
3400 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3403 subfacet->path = SF_NOT_INSTALLED;
3409 handle_upcalls(struct dpif_backer *backer)
3411 struct flow_miss_batch *fmb;
3414 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3415 struct drop_key *drop_key = drop_key_next(backer->udpif);
3420 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3421 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3422 hash_bytes(drop_key->key, drop_key->key_len, 0));
3423 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3424 drop_key->key, drop_key->key_len,
3425 NULL, 0, NULL, 0, NULL);
3427 drop_key_destroy(drop_key);
3431 fmb = flow_miss_batch_next(backer->udpif);
3433 handle_flow_misses(backer, fmb);
3434 flow_miss_batch_destroy(fmb);
3438 /* Flow expiration. */
3440 static int subfacet_max_idle(const struct dpif_backer *);
3441 static void update_stats(struct dpif_backer *);
3442 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3443 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3445 /* This function is called periodically by run(). Its job is to collect
3446 * updates for the flows that have been installed into the datapath, most
3447 * importantly when they last were used, and then use that information to
3448 * expire flows that have not been used recently.
3450 * Returns the number of milliseconds after which it should be called again. */
3452 expire(struct dpif_backer *backer)
3454 struct ofproto_dpif *ofproto;
3458 /* Periodically clear out the drop keys in an effort to keep them
3459 * relatively few. */
3460 drop_key_clear(backer);
3462 /* Update stats for each flow in the backer. */
3463 update_stats(backer);
3465 n_subfacets = hmap_count(&backer->subfacets);
3467 struct subfacet *subfacet;
3468 long long int total, now;
3472 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3473 total += now - subfacet->created;
3475 backer->avg_subfacet_life += total / n_subfacets;
3477 backer->avg_subfacet_life /= 2;
3479 backer->avg_n_subfacet += n_subfacets;
3480 backer->avg_n_subfacet /= 2;
3482 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3484 max_idle = subfacet_max_idle(backer);
3485 expire_subfacets(backer, max_idle);
3487 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3488 struct rule *rule, *next_rule;
3490 if (ofproto->backer != backer) {
3494 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3496 ovs_mutex_lock(&ofproto_mutex);
3497 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3498 &ofproto->up.expirable) {
3499 rule_expire(rule_dpif_cast(rule));
3501 ovs_mutex_unlock(&ofproto_mutex);
3503 /* All outstanding data in existing flows has been accounted, so it's a
3504 * good time to do bond rebalancing. */
3505 if (ofproto->has_bonded_bundles) {
3506 struct ofbundle *bundle;
3508 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3510 bond_rebalance(bundle->bond);
3516 return MIN(max_idle, 1000);
3519 /* Updates flow table statistics given that the datapath just reported 'stats'
3520 * as 'subfacet''s statistics. */
3522 update_subfacet_stats(struct subfacet *subfacet,
3523 const struct dpif_flow_stats *stats)
3525 struct facet *facet = subfacet->facet;
3526 struct dpif_flow_stats diff;
3528 diff.tcp_flags = stats->tcp_flags;
3529 diff.used = stats->used;
3531 if (stats->n_packets >= subfacet->dp_packet_count) {
3532 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3534 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3538 if (stats->n_bytes >= subfacet->dp_byte_count) {
3539 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3541 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3545 facet->ofproto->n_hit += diff.n_packets;
3546 subfacet->dp_packet_count = stats->n_packets;
3547 subfacet->dp_byte_count = stats->n_bytes;
3548 subfacet_update_stats(subfacet, &diff);
3550 if (facet->accounted_bytes < facet->byte_count) {
3552 facet_account(facet);
3553 facet->accounted_bytes = facet->byte_count;
3557 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3558 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3560 delete_unexpected_flow(struct dpif_backer *backer,
3561 const struct nlattr *key, size_t key_len)
3563 if (!VLOG_DROP_WARN(&rl)) {
3567 odp_flow_key_format(key, key_len, &s);
3568 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3572 COVERAGE_INC(facet_unexpected);
3573 dpif_flow_del(backer->dpif, key, key_len, NULL);
3576 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3578 * This function also pushes statistics updates to rules which each facet
3579 * resubmits into. Generally these statistics will be accurate. However, if a
3580 * facet changes the rule it resubmits into at some time in between
3581 * update_stats() runs, it is possible that statistics accrued to the
3582 * old rule will be incorrectly attributed to the new rule. This could be
3583 * avoided by calling update_stats() whenever rules are created or
3584 * deleted. However, the performance impact of making so many calls to the
3585 * datapath do not justify the benefit of having perfectly accurate statistics.
3587 * In addition, this function maintains per ofproto flow hit counts. The patch
3588 * port is not treated specially. e.g. A packet ingress from br0 patched into
3589 * br1 will increase the hit count of br0 by 1, however, does not affect
3590 * the hit or miss counts of br1.
3593 update_stats(struct dpif_backer *backer)
3595 const struct dpif_flow_stats *stats;
3596 struct dpif_flow_dump dump;
3597 const struct nlattr *key, *mask;
3598 size_t key_len, mask_len;
3600 dpif_flow_dump_start(&dump, backer->dpif);
3601 while (dpif_flow_dump_next(&dump, &key, &key_len,
3602 &mask, &mask_len, NULL, NULL, &stats)) {
3603 struct subfacet *subfacet;
3606 key_hash = odp_flow_key_hash(key, key_len);
3607 subfacet = subfacet_find(backer, key, key_len, key_hash);
3608 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3610 update_subfacet_stats(subfacet, stats);
3614 /* Stats are updated per-packet. */
3617 case SF_NOT_INSTALLED:
3619 delete_unexpected_flow(backer, key, key_len);
3624 dpif_flow_dump_done(&dump);
3627 /* Calculates and returns the number of milliseconds of idle time after which
3628 * subfacets should expire from the datapath. When a subfacet expires, we fold
3629 * its statistics into its facet, and when a facet's last subfacet expires, we
3630 * fold its statistic into its rule. */
3632 subfacet_max_idle(const struct dpif_backer *backer)
3635 * Idle time histogram.
3637 * Most of the time a switch has a relatively small number of subfacets.
3638 * When this is the case we might as well keep statistics for all of them
3639 * in userspace and to cache them in the kernel datapath for performance as
3642 * As the number of subfacets increases, the memory required to maintain
3643 * statistics about them in userspace and in the kernel becomes
3644 * significant. However, with a large number of subfacets it is likely
3645 * that only a few of them are "heavy hitters" that consume a large amount
3646 * of bandwidth. At this point, only heavy hitters are worth caching in
3647 * the kernel and maintaining in userspaces; other subfacets we can
3650 * The technique used to compute the idle time is to build a histogram with
3651 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3652 * that is installed in the kernel gets dropped in the appropriate bucket.
3653 * After the histogram has been built, we compute the cutoff so that only
3654 * the most-recently-used 1% of subfacets (but at least
3655 * flow_eviction_threshold flows) are kept cached. At least
3656 * the most-recently-used bucket of subfacets is kept, so actually an
3657 * arbitrary number of subfacets can be kept in any given expiration run
3658 * (though the next run will delete most of those unless they receive
3661 * This requires a second pass through the subfacets, in addition to the
3662 * pass made by update_stats(), because the former function never looks at
3663 * uninstallable subfacets.
3665 enum { BUCKET_WIDTH = 100 };
3666 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3667 int buckets[N_BUCKETS] = { 0 };
3668 int total, subtotal, bucket;
3669 struct subfacet *subfacet;
3673 total = hmap_count(&backer->subfacets);
3674 if (total <= flow_eviction_threshold) {
3675 return N_BUCKETS * BUCKET_WIDTH;
3678 /* Build histogram. */
3680 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3681 long long int idle = now - subfacet->used;
3682 int bucket = (idle <= 0 ? 0
3683 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3684 : (unsigned int) idle / BUCKET_WIDTH);
3688 /* Find the first bucket whose flows should be expired. */
3689 subtotal = bucket = 0;
3691 subtotal += buckets[bucket++];
3692 } while (bucket < N_BUCKETS &&
3693 subtotal < MAX(flow_eviction_threshold, total / 100));
3695 if (VLOG_IS_DBG_ENABLED()) {
3699 ds_put_cstr(&s, "keep");
3700 for (i = 0; i < N_BUCKETS; i++) {
3702 ds_put_cstr(&s, ", drop");
3705 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3708 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3712 return bucket * BUCKET_WIDTH;
3716 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3718 /* Cutoff time for most flows. */
3719 long long int normal_cutoff = time_msec() - dp_max_idle;
3721 /* We really want to keep flows for special protocols around, so use a more
3722 * conservative cutoff. */
3723 long long int special_cutoff = time_msec() - 10000;
3725 struct subfacet *subfacet, *next_subfacet;
3726 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3730 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3731 &backer->subfacets) {
3732 long long int cutoff;
3734 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3738 if (subfacet->used < cutoff) {
3739 if (subfacet->path != SF_NOT_INSTALLED) {
3740 batch[n_batch++] = subfacet;
3741 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3742 subfacet_destroy_batch(backer, batch, n_batch);
3746 subfacet_destroy(subfacet);
3752 subfacet_destroy_batch(backer, batch, n_batch);
3756 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3757 * then delete it entirely. */
3759 rule_expire(struct rule_dpif *rule)
3760 OVS_REQUIRES(ofproto_mutex)
3762 uint16_t idle_timeout, hard_timeout;
3763 long long int now = time_msec();
3766 ovs_assert(!rule->up.pending);
3768 /* Has 'rule' expired? */
3769 ovs_mutex_lock(&rule->up.mutex);
3770 hard_timeout = rule->up.hard_timeout;
3771 idle_timeout = rule->up.idle_timeout;
3772 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3773 reason = OFPRR_HARD_TIMEOUT;
3774 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3775 reason = OFPRR_IDLE_TIMEOUT;
3779 ovs_mutex_unlock(&rule->up.mutex);
3782 COVERAGE_INC(ofproto_dpif_expired);
3783 ofproto_rule_expire(&rule->up, reason);
3789 /* Creates and returns a new facet based on 'miss'.
3791 * The caller must already have determined that no facet with an identical
3792 * 'miss->flow' exists in 'miss->ofproto'.
3794 * 'rule' and 'xout' must have been created based on 'miss'.
3796 * 'facet'' statistics are initialized based on 'stats'.
3798 * The facet will initially have no subfacets. The caller should create (at
3799 * least) one subfacet with subfacet_create(). */
3800 static struct facet *
3801 facet_create(const struct flow_miss *miss)
3803 struct ofproto_dpif *ofproto = miss->ofproto;
3804 struct facet *facet;
3807 COVERAGE_INC(facet_create);
3808 facet = xzalloc(sizeof *facet);
3809 facet->ofproto = miss->ofproto;
3810 facet->used = miss->stats.used;
3811 facet->flow = miss->flow;
3812 facet->learn_rl = time_msec() + 500;
3814 list_init(&facet->subfacets);
3815 netflow_flow_init(&facet->nf_flow);
3816 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3818 xlate_out_copy(&facet->xout, &miss->xout);
3820 match_init(&match, &facet->flow, &facet->xout.wc);
3821 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3822 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3823 classifier_insert(&ofproto->facets, &facet->cr);
3824 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3826 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3831 facet_free(struct facet *facet)
3834 xlate_out_uninit(&facet->xout);
3839 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3840 * 'packet', which arrived on 'in_port'. */
3842 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3843 const struct nlattr *odp_actions, size_t actions_len,
3844 struct ofpbuf *packet)
3846 struct odputil_keybuf keybuf;
3850 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3851 odp_flow_key_from_flow(&key, flow,
3852 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3854 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3855 odp_actions, actions_len, packet);
3859 /* Remove 'facet' from its ofproto and free up the associated memory:
3861 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3862 * rule's statistics, via subfacet_uninstall().
3864 * - Removes 'facet' from its rule and from ofproto->facets.
3867 facet_remove(struct facet *facet)
3869 struct subfacet *subfacet, *next_subfacet;
3871 COVERAGE_INC(facet_remove);
3872 ovs_assert(!list_is_empty(&facet->subfacets));
3874 /* First uninstall all of the subfacets to get final statistics. */
3875 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3876 subfacet_uninstall(subfacet);
3879 /* Flush the final stats to the rule.
3881 * This might require us to have at least one subfacet around so that we
3882 * can use its actions for accounting in facet_account(), which is why we
3883 * have uninstalled but not yet destroyed the subfacets. */
3884 facet_flush_stats(facet);
3886 /* Now we're really all done so destroy everything. */
3887 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3888 &facet->subfacets) {
3889 subfacet_destroy__(subfacet);
3891 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3892 classifier_remove(&facet->ofproto->facets, &facet->cr);
3893 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3894 cls_rule_destroy(&facet->cr);
3898 /* Feed information from 'facet' back into the learning table to keep it in
3899 * sync with what is actually flowing through the datapath. */
3901 facet_learn(struct facet *facet)
3903 long long int now = time_msec();
3905 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3909 facet->learn_rl = now + 500;
3911 if (!facet->xout.has_learn
3912 && !facet->xout.has_normal
3913 && (!facet->xout.has_fin_timeout
3914 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3918 facet_push_stats(facet, true);
3922 facet_account(struct facet *facet)
3924 const struct nlattr *a;
3929 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3932 n_bytes = facet->byte_count - facet->accounted_bytes;
3934 /* This loop feeds byte counters to bond_account() for rebalancing to use
3935 * as a basis. We also need to track the actual VLAN on which the packet
3936 * is going to be sent to ensure that it matches the one passed to
3937 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3940 * We use the actions from an arbitrary subfacet because they should all
3941 * be equally valid for our purpose. */
3942 vlan_tci = facet->flow.vlan_tci;
3943 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3944 facet->xout.odp_actions.size) {
3945 const struct ovs_action_push_vlan *vlan;
3946 struct ofport_dpif *port;
3948 switch (nl_attr_type(a)) {
3949 case OVS_ACTION_ATTR_OUTPUT:
3950 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3951 if (port && port->bundle && port->bundle->bond) {
3952 bond_account(port->bundle->bond, &facet->flow,
3953 vlan_tci_to_vid(vlan_tci), n_bytes);
3957 case OVS_ACTION_ATTR_POP_VLAN:
3958 vlan_tci = htons(0);
3961 case OVS_ACTION_ATTR_PUSH_VLAN:
3962 vlan = nl_attr_get(a);
3963 vlan_tci = vlan->vlan_tci;
3969 /* Returns true if the only action for 'facet' is to send to the controller.
3970 * (We don't report NetFlow expiration messages for such facets because they
3971 * are just part of the control logic for the network, not real traffic). */
3973 facet_is_controller_flow(struct facet *facet)
3976 struct ofproto_dpif *ofproto = facet->ofproto;
3977 const struct ofpact *ofpacts;
3978 struct rule_actions *actions;
3979 struct rule_dpif *rule;
3983 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3984 actions = rule_dpif_get_actions(rule);
3985 rule_dpif_unref(rule);
3987 ofpacts_len = actions->ofpacts_len;
3988 ofpacts = actions->ofpacts;
3989 is_controller = ofpacts_len > 0
3990 && ofpacts->type == OFPACT_CONTROLLER
3991 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3992 rule_actions_unref(actions);
3994 return is_controller;
3999 /* Folds all of 'facet''s statistics into its rule. Also updates the
4000 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4001 * 'facet''s statistics in the datapath should have been zeroed and folded into
4002 * its packet and byte counts before this function is called. */
4004 facet_flush_stats(struct facet *facet)
4006 struct ofproto_dpif *ofproto = facet->ofproto;
4007 struct subfacet *subfacet;
4009 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4010 ovs_assert(!subfacet->dp_byte_count);
4011 ovs_assert(!subfacet->dp_packet_count);
4014 facet_push_stats(facet, false);
4015 if (facet->accounted_bytes < facet->byte_count) {
4016 facet_account(facet);
4017 facet->accounted_bytes = facet->byte_count;
4020 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4021 struct ofexpired expired;
4022 expired.flow = facet->flow;
4023 expired.packet_count = facet->packet_count;
4024 expired.byte_count = facet->byte_count;
4025 expired.used = facet->used;
4026 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4029 /* Reset counters to prevent double counting if 'facet' ever gets
4031 facet_reset_counters(facet);
4033 netflow_flow_clear(&facet->nf_flow);
4034 facet->tcp_flags = 0;
4037 /* Searches 'ofproto''s table of facets for one which would be responsible for
4038 * 'flow'. Returns it if found, otherwise a null pointer.
4040 * The returned facet might need revalidation; use facet_lookup_valid()
4041 * instead if that is important. */
4042 static struct facet *
4043 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4045 struct cls_rule *cr;
4047 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4048 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4049 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4050 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4053 /* Searches 'ofproto''s table of facets for one capable that covers
4054 * 'flow'. Returns it if found, otherwise a null pointer.
4056 * The returned facet is guaranteed to be valid. */
4057 static struct facet *
4058 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4060 struct facet *facet;
4062 facet = facet_find(ofproto, flow);
4064 && ofproto->backer->need_revalidate
4065 && !facet_revalidate(facet)) {
4073 facet_check_consistency(struct facet *facet)
4075 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4077 struct xlate_out xout;
4078 struct xlate_in xin;
4080 struct rule_dpif *rule;
4083 /* Check the datapath actions for consistency. */
4084 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4085 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4086 xlate_actions(&xin, &xout);
4087 rule_dpif_unref(rule);
4089 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4090 && facet->xout.slow == xout.slow;
4091 if (!ok && !VLOG_DROP_WARN(&rl)) {
4092 struct ds s = DS_EMPTY_INITIALIZER;
4094 flow_format(&s, &facet->flow);
4095 ds_put_cstr(&s, ": inconsistency in facet");
4097 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4098 ds_put_cstr(&s, " (actions were: ");
4099 format_odp_actions(&s, facet->xout.odp_actions.data,
4100 facet->xout.odp_actions.size);
4101 ds_put_cstr(&s, ") (correct actions: ");
4102 format_odp_actions(&s, xout.odp_actions.data,
4103 xout.odp_actions.size);
4104 ds_put_char(&s, ')');
4107 if (facet->xout.slow != xout.slow) {
4108 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4113 xlate_out_uninit(&xout);
4118 /* Re-searches the classifier for 'facet':
4120 * - If the rule found is different from 'facet''s current rule, moves
4121 * 'facet' to the new rule and recompiles its actions.
4123 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4124 * where it is and recompiles its actions anyway.
4126 * - If any of 'facet''s subfacets correspond to a new flow according to
4127 * xlate_receive(), 'facet' is removed.
4129 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4131 facet_revalidate(struct facet *facet)
4133 struct ofproto_dpif *ofproto = facet->ofproto;
4134 struct rule_dpif *new_rule;
4135 struct subfacet *subfacet;
4136 struct flow_wildcards wc;
4137 struct xlate_out xout;
4138 struct xlate_in xin;
4140 COVERAGE_INC(facet_revalidate);
4142 /* Check that child subfacets still correspond to this facet. Tunnel
4143 * configuration changes could cause a subfacet's OpenFlow in_port to
4145 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4146 struct ofproto_dpif *recv_ofproto;
4147 struct flow recv_flow;
4150 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4151 subfacet->key_len, &recv_flow, NULL,
4152 &recv_ofproto, NULL);
4154 || recv_ofproto != ofproto
4155 || facet != facet_find(ofproto, &recv_flow)) {
4156 facet_remove(facet);
4161 flow_wildcards_init_catchall(&wc);
4162 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4164 /* Calculate new datapath actions.
4166 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4167 * emit a NetFlow expiration and, if so, we need to have the old state
4168 * around to properly compose it. */
4169 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4170 xlate_actions(&xin, &xout);
4171 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4173 /* A facet's slow path reason should only change under dramatic
4174 * circumstances. Rather than try to update everything, it's simpler to
4175 * remove the facet and start over.
4177 * More importantly, if a facet's wildcards change, it will be relatively
4178 * difficult to figure out if its subfacets still belong to it, and if not
4179 * which facet they may belong to. Again, to avoid the complexity, we
4180 * simply give up instead. */
4181 if (facet->xout.slow != xout.slow
4182 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4183 facet_remove(facet);
4184 xlate_out_uninit(&xout);
4185 rule_dpif_unref(new_rule);
4189 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4190 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4191 if (subfacet->path == SF_FAST_PATH) {
4192 struct dpif_flow_stats stats;
4194 subfacet_install(subfacet, &xout.odp_actions, &stats);
4195 subfacet_update_stats(subfacet, &stats);
4199 facet_flush_stats(facet);
4201 ofpbuf_clear(&facet->xout.odp_actions);
4202 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4203 xout.odp_actions.size);
4206 /* Update 'facet' now that we've taken care of all the old state. */
4207 facet->xout.slow = xout.slow;
4208 facet->xout.has_learn = xout.has_learn;
4209 facet->xout.has_normal = xout.has_normal;
4210 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4211 facet->xout.nf_output_iface = xout.nf_output_iface;
4212 facet->xout.mirrors = xout.mirrors;
4213 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4215 ovs_mutex_lock(&new_rule->up.mutex);
4216 facet->used = MAX(facet->used, new_rule->up.created);
4217 ovs_mutex_unlock(&new_rule->up.mutex);
4219 xlate_out_uninit(&xout);
4220 rule_dpif_unref(new_rule);
4225 facet_reset_counters(struct facet *facet)
4227 facet->packet_count = 0;
4228 facet->byte_count = 0;
4229 facet->prev_packet_count = 0;
4230 facet->prev_byte_count = 0;
4231 facet->accounted_bytes = 0;
4235 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4236 struct dpif_flow_stats *stats, bool may_learn)
4238 struct ofport_dpif *in_port;
4239 struct rule_dpif *rule;
4240 struct xlate_in xin;
4242 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4243 if (in_port && in_port->is_tunnel) {
4244 netdev_vport_inc_rx(in_port->up.netdev, stats);
4247 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4248 rule_dpif_credit_stats(rule, stats);
4249 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4250 xin.resubmit_stats = stats;
4251 xin.may_learn = may_learn;
4252 xlate_actions_for_side_effects(&xin);
4253 rule_dpif_unref(rule);
4257 facet_push_stats(struct facet *facet, bool may_learn)
4259 struct dpif_flow_stats stats;
4261 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4262 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4263 ovs_assert(facet->used >= facet->prev_used);
4265 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4266 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4267 stats.used = facet->used;
4268 stats.tcp_flags = facet->tcp_flags;
4270 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4271 facet->prev_packet_count = facet->packet_count;
4272 facet->prev_byte_count = facet->byte_count;
4273 facet->prev_used = facet->used;
4275 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4277 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4278 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4279 stats.n_packets, stats.n_bytes);
4280 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4285 push_all_stats__(bool run_fast)
4287 static long long int rl = LLONG_MIN;
4288 struct ofproto_dpif *ofproto;
4290 if (time_msec() < rl) {
4294 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4295 struct cls_cursor cursor;
4296 struct facet *facet;
4298 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4299 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4300 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4301 facet_push_stats(facet, false);
4306 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4309 rl = time_msec() + 100;
4313 push_all_stats(void)
4315 push_all_stats__(true);
4319 rule_dpif_credit_stats(struct rule_dpif *rule,
4320 const struct dpif_flow_stats *stats)
4322 ovs_mutex_lock(&rule->stats_mutex);
4323 rule->packet_count += stats->n_packets;
4324 rule->byte_count += stats->n_bytes;
4325 rule->up.used = MAX(rule->up.used, stats->used);
4326 ovs_mutex_unlock(&rule->stats_mutex);
4330 rule_dpif_fail_open(const struct rule_dpif *rule)
4332 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4336 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4337 OVS_REQUIRES(rule->up.mutex)
4339 return rule->up.flow_cookie;
4343 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4344 uint16_t hard_timeout)
4346 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4349 /* Returns 'rule''s actions. The caller owns a reference on the returned
4350 * actions and must eventually release it (with rule_actions_unref()) to avoid
4352 struct rule_actions *
4353 rule_dpif_get_actions(const struct rule_dpif *rule)
4355 return rule_get_actions(&rule->up);
4360 static struct subfacet *
4361 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4362 size_t key_len, uint32_t key_hash)
4364 struct subfacet *subfacet;
4366 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4367 &backer->subfacets) {
4368 if (subfacet->key_len == key_len
4369 && !memcmp(key, subfacet->key, key_len)) {
4377 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4378 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4379 * existing subfacet if there is one, otherwise creates and returns a
4381 static struct subfacet *
4382 subfacet_create(struct facet *facet, struct flow_miss *miss)
4384 struct dpif_backer *backer = miss->ofproto->backer;
4385 const struct nlattr *key = miss->key;
4386 size_t key_len = miss->key_len;
4388 struct subfacet *subfacet;
4390 key_hash = odp_flow_key_hash(key, key_len);
4392 if (list_is_empty(&facet->subfacets)) {
4393 subfacet = &facet->one_subfacet;
4395 subfacet = subfacet_find(backer, key, key_len, key_hash);
4397 if (subfacet->facet == facet) {
4401 /* This shouldn't happen. */
4402 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4403 subfacet_destroy(subfacet);
4406 subfacet = xmalloc(sizeof *subfacet);
4409 COVERAGE_INC(subfacet_create);
4410 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4411 list_push_back(&facet->subfacets, &subfacet->list_node);
4412 subfacet->facet = facet;
4413 subfacet->key = xmemdup(key, key_len);
4414 subfacet->key_len = key_len;
4415 subfacet->used = miss->stats.used;
4416 subfacet->created = subfacet->used;
4417 subfacet->dp_packet_count = 0;
4418 subfacet->dp_byte_count = 0;
4419 subfacet->path = SF_NOT_INSTALLED;
4420 subfacet->backer = backer;
4425 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4426 * its facet within 'ofproto', and frees it. */
4428 subfacet_destroy__(struct subfacet *subfacet)
4430 struct facet *facet = subfacet->facet;
4432 COVERAGE_INC(subfacet_destroy);
4433 subfacet_uninstall(subfacet);
4434 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4435 list_remove(&subfacet->list_node);
4436 free(subfacet->key);
4437 if (subfacet != &facet->one_subfacet) {
4442 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4443 * last remaining subfacet in its facet destroys the facet too. */
4445 subfacet_destroy(struct subfacet *subfacet)
4447 struct facet *facet = subfacet->facet;
4449 if (list_is_singleton(&facet->subfacets)) {
4450 /* facet_remove() needs at least one subfacet (it will remove it). */
4451 facet_remove(facet);
4453 subfacet_destroy__(subfacet);
4458 subfacet_destroy_batch(struct dpif_backer *backer,
4459 struct subfacet **subfacets, int n)
4461 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4462 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4463 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4466 for (i = 0; i < n; i++) {
4467 ops[i].type = DPIF_OP_FLOW_DEL;
4468 ops[i].u.flow_del.key = subfacets[i]->key;
4469 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4470 ops[i].u.flow_del.stats = &stats[i];
4474 dpif_operate(backer->dpif, opsp, n);
4475 for (i = 0; i < n; i++) {
4476 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4477 subfacets[i]->path = SF_NOT_INSTALLED;
4478 subfacet_destroy(subfacets[i]);
4483 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4484 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4485 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4486 * since 'subfacet' was last updated.
4488 * Returns 0 if successful, otherwise a positive errno value. */
4490 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4491 struct dpif_flow_stats *stats)
4493 struct facet *facet = subfacet->facet;
4494 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4495 const struct nlattr *actions = odp_actions->data;
4496 size_t actions_len = odp_actions->size;
4497 struct odputil_keybuf maskbuf;
4500 uint64_t slow_path_stub[128 / 8];
4501 enum dpif_flow_put_flags flags;
4504 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4507 flags |= DPIF_FP_ZERO_STATS;
4510 if (path == SF_SLOW_PATH) {
4511 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4512 slow_path_stub, sizeof slow_path_stub,
4513 &actions, &actions_len);
4516 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4517 if (enable_megaflows) {
4518 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4519 &facet->flow, UINT32_MAX);
4522 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4523 subfacet->key_len, mask.data, mask.size,
4524 actions, actions_len, stats);
4527 subfacet_reset_dp_stats(subfacet, stats);
4531 COVERAGE_INC(subfacet_install_fail);
4533 subfacet->path = path;
4538 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4540 subfacet_uninstall(struct subfacet *subfacet)
4542 if (subfacet->path != SF_NOT_INSTALLED) {
4543 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4544 struct dpif_flow_stats stats;
4547 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4548 subfacet->key_len, &stats);
4549 subfacet_reset_dp_stats(subfacet, &stats);
4551 subfacet_update_stats(subfacet, &stats);
4553 subfacet->path = SF_NOT_INSTALLED;
4555 ovs_assert(subfacet->dp_packet_count == 0);
4556 ovs_assert(subfacet->dp_byte_count == 0);
4560 /* Resets 'subfacet''s datapath statistics counters. This should be called
4561 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4562 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4563 * was reset in the datapath. 'stats' will be modified to include only
4564 * statistics new since 'subfacet' was last updated. */
4566 subfacet_reset_dp_stats(struct subfacet *subfacet,
4567 struct dpif_flow_stats *stats)
4570 && subfacet->dp_packet_count <= stats->n_packets
4571 && subfacet->dp_byte_count <= stats->n_bytes) {
4572 stats->n_packets -= subfacet->dp_packet_count;
4573 stats->n_bytes -= subfacet->dp_byte_count;
4576 subfacet->dp_packet_count = 0;
4577 subfacet->dp_byte_count = 0;
4580 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4582 * Because of the meaning of a subfacet's counters, it only makes sense to do
4583 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4584 * represents a packet that was sent by hand or if it represents statistics
4585 * that have been cleared out of the datapath. */
4587 subfacet_update_stats(struct subfacet *subfacet,
4588 const struct dpif_flow_stats *stats)
4590 if (stats->n_packets || stats->used > subfacet->used) {
4591 struct facet *facet = subfacet->facet;
4593 subfacet->used = MAX(subfacet->used, stats->used);
4594 facet->used = MAX(facet->used, stats->used);
4595 facet->packet_count += stats->n_packets;
4596 facet->byte_count += stats->n_bytes;
4597 facet->tcp_flags |= stats->tcp_flags;
4603 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4604 * the fields that were relevant as part of the lookup. */
4606 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4607 struct flow_wildcards *wc, struct rule_dpif **rule)
4609 struct ofport_dpif *port;
4611 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4614 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4616 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4617 flow->in_port.ofp_port);
4620 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4621 ofproto->no_packet_in_rule, rule);
4625 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4626 const struct flow *flow, struct flow_wildcards *wc,
4627 uint8_t table_id, struct rule_dpif **rule)
4629 const struct cls_rule *cls_rule;
4630 struct classifier *cls;
4634 if (table_id >= N_TABLES) {
4639 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4640 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4643 cls = &ofproto->up.tables[table_id].cls;
4644 ovs_rwlock_rdlock(&cls->rwlock);
4645 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4646 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4647 /* We must pretend that transport ports are unavailable. */
4648 struct flow ofpc_normal_flow = *flow;
4649 ofpc_normal_flow.tp_src = htons(0);
4650 ofpc_normal_flow.tp_dst = htons(0);
4651 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4652 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4653 cls_rule = &ofproto->drop_frags_rule->up.cr;
4655 flow_wildcards_init_exact(wc);
4658 cls_rule = classifier_lookup(cls, flow, wc);
4661 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4662 rule_dpif_ref(*rule);
4663 ovs_rwlock_unlock(&cls->rwlock);
4665 return *rule != NULL;
4668 /* Given a port configuration (specified as zero if there's no port), chooses
4669 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4670 * flow table miss. */
4672 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4673 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4675 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4676 rule_dpif_ref(*rule);
4680 rule_dpif_ref(struct rule_dpif *rule)
4683 ofproto_rule_ref(&rule->up);
4688 rule_dpif_unref(struct rule_dpif *rule)
4691 ofproto_rule_unref(&rule->up);
4696 complete_operation(struct rule_dpif *rule)
4697 OVS_REQUIRES(ofproto_mutex)
4699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4701 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4702 ofoperation_complete(rule->up.pending, 0);
4705 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4707 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4710 static struct rule *
4713 struct rule_dpif *rule = xmalloc(sizeof *rule);
4718 rule_dealloc(struct rule *rule_)
4720 struct rule_dpif *rule = rule_dpif_cast(rule_);
4725 rule_construct(struct rule *rule_)
4727 struct rule_dpif *rule = rule_dpif_cast(rule_);
4728 ovs_mutex_init(&rule->stats_mutex);
4729 ovs_mutex_lock(&rule->stats_mutex);
4730 rule->packet_count = 0;
4731 rule->byte_count = 0;
4732 ovs_mutex_unlock(&rule->stats_mutex);
4737 rule_insert(struct rule *rule_)
4738 OVS_REQUIRES(ofproto_mutex)
4740 struct rule_dpif *rule = rule_dpif_cast(rule_);
4741 complete_operation(rule);
4745 rule_delete(struct rule *rule_)
4746 OVS_REQUIRES(ofproto_mutex)
4748 struct rule_dpif *rule = rule_dpif_cast(rule_);
4749 complete_operation(rule);
4753 rule_destruct(struct rule *rule_)
4755 struct rule_dpif *rule = rule_dpif_cast(rule_);
4756 ovs_mutex_destroy(&rule->stats_mutex);
4760 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4762 struct rule_dpif *rule = rule_dpif_cast(rule_);
4764 /* push_all_stats() can handle flow misses which, when using the learn
4765 * action, can cause rules to be added and deleted. This can corrupt our
4766 * caller's datastructures which assume that rule_get_stats() doesn't have
4767 * an impact on the flow table. To be safe, we disable miss handling. */
4768 push_all_stats__(false);
4770 /* Start from historical data for 'rule' itself that are no longer tracked
4771 * in facets. This counts, for example, facets that have expired. */
4772 ovs_mutex_lock(&rule->stats_mutex);
4773 *packets = rule->packet_count;
4774 *bytes = rule->byte_count;
4775 ovs_mutex_unlock(&rule->stats_mutex);
4779 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4780 struct ofpbuf *packet)
4782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4783 struct dpif_flow_stats stats;
4784 struct xlate_out xout;
4785 struct xlate_in xin;
4787 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4788 rule_dpif_credit_stats(rule, &stats);
4790 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4791 xin.resubmit_stats = &stats;
4792 xlate_actions(&xin, &xout);
4794 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4795 xout.odp_actions.size, packet);
4797 xlate_out_uninit(&xout);
4801 rule_execute(struct rule *rule, const struct flow *flow,
4802 struct ofpbuf *packet)
4804 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4805 ofpbuf_delete(packet);
4810 rule_modify_actions(struct rule *rule_, bool reset_counters)
4811 OVS_REQUIRES(ofproto_mutex)
4813 struct rule_dpif *rule = rule_dpif_cast(rule_);
4815 if (reset_counters) {
4816 ovs_mutex_lock(&rule->stats_mutex);
4817 rule->packet_count = 0;
4818 rule->byte_count = 0;
4819 ovs_mutex_unlock(&rule->stats_mutex);
4822 complete_operation(rule);
4825 /* Sends 'packet' out 'ofport'.
4826 * May modify 'packet'.
4827 * Returns 0 if successful, otherwise a positive errno value. */
4829 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4831 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4832 uint64_t odp_actions_stub[1024 / 8];
4833 struct ofpbuf key, odp_actions;
4834 struct dpif_flow_stats stats;
4835 struct odputil_keybuf keybuf;
4836 struct ofpact_output output;
4837 struct xlate_out xout;
4838 struct xlate_in xin;
4840 union flow_in_port in_port_;
4843 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4844 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4846 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4847 in_port_.ofp_port = OFPP_NONE;
4848 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4849 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4851 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4853 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4854 output.port = ofport->up.ofp_port;
4857 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4858 xin.ofpacts_len = sizeof output;
4859 xin.ofpacts = &output.ofpact;
4860 xin.resubmit_stats = &stats;
4861 xlate_actions(&xin, &xout);
4863 error = dpif_execute(ofproto->backer->dpif,
4865 xout.odp_actions.data, xout.odp_actions.size,
4867 xlate_out_uninit(&xout);
4870 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4871 ofproto->up.name, netdev_get_name(ofport->up.netdev),
4872 ovs_strerror(error));
4875 ofproto->stats.tx_packets++;
4876 ofproto->stats.tx_bytes += packet->size;
4880 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4881 * The action will state 'slow' as the reason that the action is in the slow
4882 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4883 * dump-flows" output to see why a flow is in the slow path.)
4885 * The 'stub_size' bytes in 'stub' will be used to store the action.
4886 * 'stub_size' must be large enough for the action.
4888 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4891 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4892 enum slow_path_reason slow,
4893 uint64_t *stub, size_t stub_size,
4894 const struct nlattr **actionsp, size_t *actions_lenp)
4896 union user_action_cookie cookie;
4899 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4900 cookie.slow_path.unused = 0;
4901 cookie.slow_path.reason = slow;
4903 ofpbuf_use_stack(&buf, stub, stub_size);
4904 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4905 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4907 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4909 odp_port_t odp_port;
4912 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4913 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4914 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4916 *actionsp = buf.data;
4917 *actions_lenp = buf.size;
4921 set_frag_handling(struct ofproto *ofproto_,
4922 enum ofp_config_flags frag_handling)
4924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4925 if (frag_handling != OFPC_FRAG_REASM) {
4926 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4934 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4935 const struct flow *flow,
4936 const struct ofpact *ofpacts, size_t ofpacts_len)
4938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4939 struct odputil_keybuf keybuf;
4940 struct dpif_flow_stats stats;
4941 struct xlate_out xout;
4942 struct xlate_in xin;
4946 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4947 odp_flow_key_from_flow(&key, flow,
4948 ofp_port_to_odp_port(ofproto,
4949 flow->in_port.ofp_port));
4951 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4953 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
4954 xin.resubmit_stats = &stats;
4955 xin.ofpacts_len = ofpacts_len;
4956 xin.ofpacts = ofpacts;
4958 xlate_actions(&xin, &xout);
4959 dpif_execute(ofproto->backer->dpif, key.data, key.size,
4960 xout.odp_actions.data, xout.odp_actions.size, packet);
4961 xlate_out_uninit(&xout);
4969 set_netflow(struct ofproto *ofproto_,
4970 const struct netflow_options *netflow_options)
4972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4974 if (netflow_options) {
4975 if (!ofproto->netflow) {
4976 ofproto->netflow = netflow_create();
4977 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4979 return netflow_set_options(ofproto->netflow, netflow_options);
4980 } else if (ofproto->netflow) {
4981 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4982 netflow_destroy(ofproto->netflow);
4983 ofproto->netflow = NULL;
4990 get_netflow_ids(const struct ofproto *ofproto_,
4991 uint8_t *engine_type, uint8_t *engine_id)
4993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4995 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4999 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5001 if (!facet_is_controller_flow(facet) &&
5002 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5003 struct subfacet *subfacet;
5004 struct ofexpired expired;
5006 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5007 if (subfacet->path == SF_FAST_PATH) {
5008 struct dpif_flow_stats stats;
5010 subfacet_install(subfacet, &facet->xout.odp_actions,
5012 subfacet_update_stats(subfacet, &stats);
5016 expired.flow = facet->flow;
5017 expired.packet_count = facet->packet_count;
5018 expired.byte_count = facet->byte_count;
5019 expired.used = facet->used;
5020 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5025 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5027 struct cls_cursor cursor;
5028 struct facet *facet;
5030 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5031 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5032 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5033 send_active_timeout(ofproto, facet);
5035 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5038 static struct ofproto_dpif *
5039 ofproto_dpif_lookup(const char *name)
5041 struct ofproto_dpif *ofproto;
5043 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5044 hash_string(name, 0), &all_ofproto_dpifs) {
5045 if (!strcmp(ofproto->up.name, name)) {
5053 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5054 const char *argv[], void *aux OVS_UNUSED)
5056 struct ofproto_dpif *ofproto;
5059 ofproto = ofproto_dpif_lookup(argv[1]);
5061 unixctl_command_reply_error(conn, "no such bridge");
5064 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5065 mac_learning_flush(ofproto->ml);
5066 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5068 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5069 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5070 mac_learning_flush(ofproto->ml);
5071 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5075 unixctl_command_reply(conn, "table successfully flushed");
5078 static struct ofport_dpif *
5079 ofbundle_get_a_port(const struct ofbundle *bundle)
5081 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5086 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5087 const char *argv[], void *aux OVS_UNUSED)
5089 struct ds ds = DS_EMPTY_INITIALIZER;
5090 const struct ofproto_dpif *ofproto;
5091 const struct mac_entry *e;
5093 ofproto = ofproto_dpif_lookup(argv[1]);
5095 unixctl_command_reply_error(conn, "no such bridge");
5099 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5100 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5101 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5102 struct ofbundle *bundle = e->port.p;
5103 char name[OFP_MAX_PORT_NAME_LEN];
5105 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5107 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5108 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5109 mac_entry_age(ofproto->ml, e));
5111 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5112 unixctl_command_reply(conn, ds_cstr(&ds));
5117 struct xlate_out xout;
5118 struct xlate_in xin;
5124 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5126 struct rule_actions *actions;
5129 ds_put_char_multiple(result, '\t', level);
5131 ds_put_cstr(result, "No match\n");
5135 ovs_mutex_lock(&rule->up.mutex);
5136 cookie = rule->up.flow_cookie;
5137 ovs_mutex_unlock(&rule->up.mutex);
5139 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5140 rule ? rule->up.table_id : 0, ntohll(cookie));
5141 cls_rule_format(&rule->up.cr, result);
5142 ds_put_char(result, '\n');
5144 actions = rule_dpif_get_actions(rule);
5146 ds_put_char_multiple(result, '\t', level);
5147 ds_put_cstr(result, "OpenFlow ");
5148 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5149 ds_put_char(result, '\n');
5151 rule_actions_unref(actions);
5155 trace_format_flow(struct ds *result, int level, const char *title,
5156 struct trace_ctx *trace)
5158 ds_put_char_multiple(result, '\t', level);
5159 ds_put_format(result, "%s: ", title);
5160 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5161 ds_put_cstr(result, "unchanged");
5163 flow_format(result, &trace->xin.flow);
5164 trace->flow = trace->xin.flow;
5166 ds_put_char(result, '\n');
5170 trace_format_regs(struct ds *result, int level, const char *title,
5171 struct trace_ctx *trace)
5175 ds_put_char_multiple(result, '\t', level);
5176 ds_put_format(result, "%s:", title);
5177 for (i = 0; i < FLOW_N_REGS; i++) {
5178 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5180 ds_put_char(result, '\n');
5184 trace_format_odp(struct ds *result, int level, const char *title,
5185 struct trace_ctx *trace)
5187 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5189 ds_put_char_multiple(result, '\t', level);
5190 ds_put_format(result, "%s: ", title);
5191 format_odp_actions(result, odp_actions->data, odp_actions->size);
5192 ds_put_char(result, '\n');
5196 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5198 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5199 struct ds *result = trace->result;
5201 ds_put_char(result, '\n');
5202 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5203 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5204 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5205 trace_format_rule(result, recurse + 1, rule);
5209 trace_report(struct xlate_in *xin, const char *s, int recurse)
5211 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5212 struct ds *result = trace->result;
5214 ds_put_char_multiple(result, '\t', recurse);
5215 ds_put_cstr(result, s);
5216 ds_put_char(result, '\n');
5220 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5221 void *aux OVS_UNUSED)
5223 const struct dpif_backer *backer = NULL;
5224 struct ofproto_dpif *ofproto;
5225 struct ofpbuf odp_key, odp_mask;
5226 struct ofpbuf *packet;
5229 struct simap port_names;
5235 ofpbuf_init(&odp_key, 0);
5236 ofpbuf_init(&odp_mask, 0);
5237 simap_init(&port_names);
5239 /* Handle "-generate" or a hex string as the last argument. */
5240 if (!strcmp(argv[argc - 1], "-generate")) {
5241 packet = ofpbuf_new(0);
5244 const char *error = eth_from_hex(argv[argc - 1], &packet);
5247 } else if (argc == 4) {
5248 /* The 3-argument form must end in "-generate' or a hex string. */
5249 unixctl_command_reply_error(conn, error);
5254 /* odp_flow can have its in_port specified as a name instead of port no.
5255 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5256 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5257 * we need to create a simap of name to port no. */
5259 const char *dp_type;
5260 if (!strncmp(argv[1], "ovs-", 4)) {
5261 dp_type = argv[1] + 4;
5265 backer = shash_find_data(&all_dpif_backers, dp_type);
5267 struct shash_node *node;
5268 if (shash_count(&all_dpif_backers) == 1) {
5269 node = shash_first(&all_dpif_backers);
5270 backer = node->data;
5273 if (backer && backer->dpif) {
5274 struct dpif_port dpif_port;
5275 struct dpif_port_dump port_dump;
5276 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5277 simap_put(&port_names, dpif_port.name,
5278 odp_to_u32(dpif_port.port_no));
5282 /* Parse the flow and determine whether a datapath or
5283 * bridge is specified. If function odp_flow_key_from_string()
5284 * returns 0, the flow is a odp_flow. If function
5285 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5286 if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) {
5288 unixctl_command_reply_error(conn, "Cannot find the datapath");
5292 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5293 NULL, &ofproto, NULL)) {
5294 unixctl_command_reply_error(conn, "Invalid datapath flow");
5297 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5298 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5300 unixctl_command_reply_error(conn, "Must specify bridge name");
5304 ofproto = ofproto_dpif_lookup(argv[1]);
5306 unixctl_command_reply_error(conn, "Unknown bridge name");
5310 unixctl_command_reply_error(conn, "Bad flow syntax");
5314 /* Generate a packet, if requested. */
5316 if (!packet->size) {
5317 flow_compose(packet, &flow);
5319 union flow_in_port in_port_;
5321 in_port_ = flow.in_port;
5322 ds_put_cstr(&result, "Packet: ");
5323 s = ofp_packet_to_string(packet->data, packet->size);
5324 ds_put_cstr(&result, s);
5327 /* Use the metadata from the flow and the packet argument
5328 * to reconstruct the flow. */
5329 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5334 ofproto_trace(ofproto, &flow, packet, &result);
5335 unixctl_command_reply(conn, ds_cstr(&result));
5338 ds_destroy(&result);
5339 ofpbuf_delete(packet);
5340 ofpbuf_uninit(&odp_key);
5341 ofpbuf_uninit(&odp_mask);
5342 simap_destroy(&port_names);
5346 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5347 const struct ofpbuf *packet, struct ds *ds)
5349 struct rule_dpif *rule;
5350 struct flow_wildcards wc;
5352 ds_put_cstr(ds, "Flow: ");
5353 flow_format(ds, flow);
5354 ds_put_char(ds, '\n');
5356 flow_wildcards_init_catchall(&wc);
5357 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5359 trace_format_rule(ds, 0, rule);
5360 if (rule == ofproto->miss_rule) {
5361 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5362 } else if (rule == ofproto->no_packet_in_rule) {
5363 ds_put_cstr(ds, "\nNo match, packets dropped because "
5364 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5365 } else if (rule == ofproto->drop_frags_rule) {
5366 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5367 "and the fragment handling mode is \"drop\".\n");
5371 uint64_t odp_actions_stub[1024 / 8];
5372 struct ofpbuf odp_actions;
5373 struct trace_ctx trace;
5377 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5380 ofpbuf_use_stub(&odp_actions,
5381 odp_actions_stub, sizeof odp_actions_stub);
5382 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5383 trace.xin.resubmit_hook = trace_resubmit;
5384 trace.xin.report_hook = trace_report;
5386 xlate_actions(&trace.xin, &trace.xout);
5387 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5389 ds_put_char(ds, '\n');
5390 trace_format_flow(ds, 0, "Final flow", &trace);
5392 match_init(&match, flow, &trace.xout.wc);
5393 ds_put_cstr(ds, "Relevant fields: ");
5394 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5395 ds_put_char(ds, '\n');
5397 ds_put_cstr(ds, "Datapath actions: ");
5398 format_odp_actions(ds, trace.xout.odp_actions.data,
5399 trace.xout.odp_actions.size);
5401 if (trace.xout.slow) {
5402 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5403 "slow path because it:");
5404 switch (trace.xout.slow) {
5406 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5409 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5412 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5415 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5417 case SLOW_CONTROLLER:
5418 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5419 "to the OpenFlow controller.");
5426 xlate_out_uninit(&trace.xout);
5429 rule_dpif_unref(rule);
5432 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5433 * 'reply' describing the results. */
5435 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5437 struct cls_cursor cursor;
5438 struct facet *facet;
5442 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5443 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5444 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5445 if (!facet_check_consistency(facet)) {
5449 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5451 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5455 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5456 ofproto->up.name, errors);
5458 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5463 ofproto_dpif_self_check(struct unixctl_conn *conn,
5464 int argc, const char *argv[], void *aux OVS_UNUSED)
5466 struct ds reply = DS_EMPTY_INITIALIZER;
5467 struct ofproto_dpif *ofproto;
5470 ofproto = ofproto_dpif_lookup(argv[1]);
5472 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5473 "ofproto/list for help)");
5476 ofproto_dpif_self_check__(ofproto, &reply);
5478 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5479 ofproto_dpif_self_check__(ofproto, &reply);
5483 unixctl_command_reply(conn, ds_cstr(&reply));
5487 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5488 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5489 * to destroy 'ofproto_shash' and free the returned value. */
5490 static const struct shash_node **
5491 get_ofprotos(struct shash *ofproto_shash)
5493 const struct ofproto_dpif *ofproto;
5495 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5496 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5497 shash_add_nocopy(ofproto_shash, name, ofproto);
5500 return shash_sort(ofproto_shash);
5504 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5505 const char *argv[] OVS_UNUSED,
5506 void *aux OVS_UNUSED)
5508 struct ds ds = DS_EMPTY_INITIALIZER;
5509 struct shash ofproto_shash;
5510 const struct shash_node **sorted_ofprotos;
5513 shash_init(&ofproto_shash);
5514 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5515 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5516 const struct shash_node *node = sorted_ofprotos[i];
5517 ds_put_format(&ds, "%s\n", node->name);
5520 shash_destroy(&ofproto_shash);
5521 free(sorted_ofprotos);
5523 unixctl_command_reply(conn, ds_cstr(&ds));
5528 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5530 const struct shash_node **ofprotos;
5531 struct ofproto_dpif *ofproto;
5532 struct shash ofproto_shash;
5533 uint64_t n_hit, n_missed;
5536 n_hit = n_missed = 0;
5537 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5538 if (ofproto->backer == backer) {
5539 n_missed += ofproto->n_missed;
5540 n_hit += ofproto->n_hit;
5544 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5545 dpif_name(backer->dpif), n_hit, n_missed);
5546 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5547 " life span: %lldms\n", hmap_count(&backer->subfacets),
5548 backer->avg_n_subfacet, backer->max_n_subfacet,
5549 backer->avg_subfacet_life);
5551 shash_init(&ofproto_shash);
5552 ofprotos = get_ofprotos(&ofproto_shash);
5553 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5554 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5555 const struct shash_node **ports;
5558 if (ofproto->backer != backer) {
5562 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5563 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5565 ports = shash_sort(&ofproto->up.port_by_name);
5566 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5567 const struct shash_node *node = ports[j];
5568 struct ofport *ofport = node->data;
5570 odp_port_t odp_port;
5572 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5575 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5576 if (odp_port != ODPP_NONE) {
5577 ds_put_format(ds, "%"PRIu32":", odp_port);
5579 ds_put_cstr(ds, "none:");
5582 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5585 if (!netdev_get_config(ofport->netdev, &config)) {
5586 const struct smap_node **nodes;
5589 nodes = smap_sort(&config);
5590 for (i = 0; i < smap_count(&config); i++) {
5591 const struct smap_node *node = nodes[i];
5592 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5593 node->key, node->value);
5597 smap_destroy(&config);
5599 ds_put_char(ds, ')');
5600 ds_put_char(ds, '\n');
5604 shash_destroy(&ofproto_shash);
5609 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5610 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5612 struct ds ds = DS_EMPTY_INITIALIZER;
5613 const struct shash_node **backers;
5616 backers = shash_sort(&all_dpif_backers);
5617 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5618 dpif_show_backer(backers[i]->data, &ds);
5622 unixctl_command_reply(conn, ds_cstr(&ds));
5626 /* Dump the megaflow (facet) cache. This is useful to check the
5627 * correctness of flow wildcarding, since the same mechanism is used for
5628 * both xlate caching and kernel wildcarding.
5630 * It's important to note that in the output the flow description uses
5631 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5633 * This command is only needed for advanced debugging, so it's not
5634 * documented in the man page. */
5636 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5637 int argc OVS_UNUSED, const char *argv[],
5638 void *aux OVS_UNUSED)
5640 struct ds ds = DS_EMPTY_INITIALIZER;
5641 const struct ofproto_dpif *ofproto;
5642 long long int now = time_msec();
5643 struct cls_cursor cursor;
5644 struct facet *facet;
5646 ofproto = ofproto_dpif_lookup(argv[1]);
5648 unixctl_command_reply_error(conn, "no such bridge");
5652 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5653 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5654 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5655 cls_rule_format(&facet->cr, &ds);
5656 ds_put_cstr(&ds, ", ");
5657 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5658 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5659 ds_put_cstr(&ds, "Datapath actions: ");
5660 if (facet->xout.slow) {
5661 uint64_t slow_path_stub[128 / 8];
5662 const struct nlattr *actions;
5665 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5666 slow_path_stub, sizeof slow_path_stub,
5667 &actions, &actions_len);
5668 format_odp_actions(&ds, actions, actions_len);
5670 format_odp_actions(&ds, facet->xout.odp_actions.data,
5671 facet->xout.odp_actions.size);
5673 ds_put_cstr(&ds, "\n");
5675 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5677 ds_chomp(&ds, '\n');
5678 unixctl_command_reply(conn, ds_cstr(&ds));
5682 /* Disable using the megaflows.
5684 * This command is only needed for advanced debugging, so it's not
5685 * documented in the man page. */
5687 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5688 int argc OVS_UNUSED,
5689 const char *argv[] OVS_UNUSED,
5690 void *aux OVS_UNUSED)
5692 struct ofproto_dpif *ofproto;
5694 enable_megaflows = false;
5696 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5697 flush(&ofproto->up);
5700 unixctl_command_reply(conn, "megaflows disabled");
5703 /* Re-enable using megaflows.
5705 * This command is only needed for advanced debugging, so it's not
5706 * documented in the man page. */
5708 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5709 int argc OVS_UNUSED,
5710 const char *argv[] OVS_UNUSED,
5711 void *aux OVS_UNUSED)
5713 struct ofproto_dpif *ofproto;
5715 enable_megaflows = true;
5717 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5718 flush(&ofproto->up);
5721 unixctl_command_reply(conn, "megaflows enabled");
5725 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5726 int argc OVS_UNUSED, const char *argv[],
5727 void *aux OVS_UNUSED)
5729 struct ds ds = DS_EMPTY_INITIALIZER;
5730 const struct ofproto_dpif *ofproto;
5731 struct subfacet *subfacet;
5733 ofproto = ofproto_dpif_lookup(argv[1]);
5735 unixctl_command_reply_error(conn, "no such bridge");
5739 update_stats(ofproto->backer);
5741 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5742 struct facet *facet = subfacet->facet;
5743 struct odputil_keybuf maskbuf;
5746 if (facet->ofproto != ofproto) {
5750 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5751 if (enable_megaflows) {
5752 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5753 &facet->flow, UINT32_MAX);
5756 odp_flow_format(subfacet->key, subfacet->key_len,
5757 mask.data, mask.size, NULL, &ds, false);
5759 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5760 subfacet->dp_packet_count, subfacet->dp_byte_count);
5761 if (subfacet->used) {
5762 ds_put_format(&ds, "%.3fs",
5763 (time_msec() - subfacet->used) / 1000.0);
5765 ds_put_format(&ds, "never");
5767 if (subfacet->facet->tcp_flags) {
5768 ds_put_cstr(&ds, ", flags:");
5769 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5772 ds_put_cstr(&ds, ", actions:");
5773 if (facet->xout.slow) {
5774 uint64_t slow_path_stub[128 / 8];
5775 const struct nlattr *actions;
5778 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5779 slow_path_stub, sizeof slow_path_stub,
5780 &actions, &actions_len);
5781 format_odp_actions(&ds, actions, actions_len);
5783 format_odp_actions(&ds, facet->xout.odp_actions.data,
5784 facet->xout.odp_actions.size);
5786 ds_put_char(&ds, '\n');
5789 unixctl_command_reply(conn, ds_cstr(&ds));
5794 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5795 int argc OVS_UNUSED, const char *argv[],
5796 void *aux OVS_UNUSED)
5798 struct ds ds = DS_EMPTY_INITIALIZER;
5799 struct ofproto_dpif *ofproto;
5801 ofproto = ofproto_dpif_lookup(argv[1]);
5803 unixctl_command_reply_error(conn, "no such bridge");
5807 flush(&ofproto->up);
5809 unixctl_command_reply(conn, ds_cstr(&ds));
5814 ofproto_dpif_unixctl_init(void)
5816 static bool registered;
5822 unixctl_command_register(
5824 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5825 1, 3, ofproto_unixctl_trace, NULL);
5826 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5827 ofproto_unixctl_fdb_flush, NULL);
5828 unixctl_command_register("fdb/show", "bridge", 1, 1,
5829 ofproto_unixctl_fdb_show, NULL);
5830 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5831 ofproto_dpif_self_check, NULL);
5832 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5833 ofproto_unixctl_dpif_dump_dps, NULL);
5834 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5836 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5837 ofproto_unixctl_dpif_dump_flows, NULL);
5838 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5839 ofproto_unixctl_dpif_del_flows, NULL);
5840 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5841 ofproto_unixctl_dpif_dump_megaflows, NULL);
5842 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5843 ofproto_unixctl_dpif_disable_megaflows, NULL);
5844 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5845 ofproto_unixctl_dpif_enable_megaflows, NULL);
5848 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5850 * This is deprecated. It is only for compatibility with broken device drivers
5851 * in old versions of Linux that do not properly support VLANs when VLAN
5852 * devices are not used. When broken device drivers are no longer in
5853 * widespread use, we will delete these interfaces. */
5856 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5858 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5859 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5861 if (realdev_ofp_port == ofport->realdev_ofp_port
5862 && vid == ofport->vlandev_vid) {
5866 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5868 if (ofport->realdev_ofp_port) {
5871 if (realdev_ofp_port && ofport->bundle) {
5872 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5873 * themselves be part of a bundle. */
5874 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5877 ofport->realdev_ofp_port = realdev_ofp_port;
5878 ofport->vlandev_vid = vid;
5880 if (realdev_ofp_port) {
5881 vsp_add(ofport, realdev_ofp_port, vid);
5888 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5890 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5894 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5895 OVS_EXCLUDED(ofproto->vsp_mutex)
5899 ovs_mutex_lock(&ofproto->vsp_mutex);
5900 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5901 ovs_mutex_unlock(&ofproto->vsp_mutex);
5906 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5907 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5908 OVS_REQUIRES(ofproto->vsp_mutex)
5910 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5911 int vid = vlan_tci_to_vid(vlan_tci);
5912 const struct vlan_splinter *vsp;
5914 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5915 hash_realdev_vid(realdev_ofp_port, vid),
5916 &ofproto->realdev_vid_map) {
5917 if (vsp->realdev_ofp_port == realdev_ofp_port
5918 && vsp->vid == vid) {
5919 return vsp->vlandev_ofp_port;
5923 return realdev_ofp_port;
5926 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5927 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5928 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5929 * 'vlan_tci' 9, it would return the port number of eth0.9.
5931 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5932 * function just returns its 'realdev_ofp_port' argument. */
5934 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5935 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5936 OVS_EXCLUDED(ofproto->vsp_mutex)
5940 ovs_mutex_lock(&ofproto->vsp_mutex);
5941 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5942 ovs_mutex_unlock(&ofproto->vsp_mutex);
5946 static struct vlan_splinter *
5947 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5949 struct vlan_splinter *vsp;
5951 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5952 hash_ofp_port(vlandev_ofp_port),
5953 &ofproto->vlandev_map) {
5954 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5962 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5963 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5964 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
5965 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5966 * eth0 and store 9 in '*vid'.
5968 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5969 * VLAN device. Unless VLAN splinters are enabled, this is what this function
5972 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5973 ofp_port_t vlandev_ofp_port, int *vid)
5974 OVS_REQUIRES(ofproto->vsp_mutex)
5976 if (!hmap_is_empty(&ofproto->vlandev_map)) {
5977 const struct vlan_splinter *vsp;
5979 vsp = vlandev_find(ofproto, vlandev_ofp_port);
5984 return vsp->realdev_ofp_port;
5990 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5991 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
5992 * 'flow->in_port' to the "real" device backing the VLAN device, sets
5993 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
5994 * always the case unless VLAN splinters are enabled), returns false without
5995 * making any changes. */
5997 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
5998 OVS_EXCLUDED(ofproto->vsp_mutex)
6003 ovs_mutex_lock(&ofproto->vsp_mutex);
6004 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6005 ovs_mutex_unlock(&ofproto->vsp_mutex);
6010 /* Cause the flow to be processed as if it came in on the real device with
6011 * the VLAN device's VLAN ID. */
6012 flow->in_port.ofp_port = realdev;
6013 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6018 vsp_remove(struct ofport_dpif *port)
6020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6021 struct vlan_splinter *vsp;
6023 ovs_mutex_lock(&ofproto->vsp_mutex);
6024 vsp = vlandev_find(ofproto, port->up.ofp_port);
6026 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6027 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6030 port->realdev_ofp_port = 0;
6032 VLOG_ERR("missing vlan device record");
6034 ovs_mutex_unlock(&ofproto->vsp_mutex);
6038 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6042 ovs_mutex_lock(&ofproto->vsp_mutex);
6043 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6044 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6045 == realdev_ofp_port)) {
6046 struct vlan_splinter *vsp;
6048 vsp = xmalloc(sizeof *vsp);
6049 vsp->realdev_ofp_port = realdev_ofp_port;
6050 vsp->vlandev_ofp_port = port->up.ofp_port;
6053 port->realdev_ofp_port = realdev_ofp_port;
6055 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6056 hash_ofp_port(port->up.ofp_port));
6057 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6058 hash_realdev_vid(realdev_ofp_port, vid));
6060 VLOG_ERR("duplicate vlan device record");
6062 ovs_mutex_unlock(&ofproto->vsp_mutex);
6066 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6068 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6069 return ofport ? ofport->odp_port : ODPP_NONE;
6072 struct ofport_dpif *
6073 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6075 struct ofport_dpif *port;
6077 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6078 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6079 &backer->odp_to_ofport_map) {
6080 if (port->odp_port == odp_port) {
6081 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6086 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6091 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6093 struct ofport_dpif *port;
6095 port = odp_port_to_ofport(ofproto->backer, odp_port);
6096 if (port && &ofproto->up == port->up.ofproto) {
6097 return port->up.ofp_port;
6103 const struct ofproto_class ofproto_dpif_class = {
6138 port_is_lacp_current,
6139 NULL, /* rule_choose_table */
6148 rule_modify_actions,
6162 get_stp_port_status,
6169 is_mirror_output_bundle,
6170 forward_bpdu_changed,
6171 set_mac_table_config,
6173 NULL, /* meter_get_features */
6174 NULL, /* meter_set */
6175 NULL, /* meter_get */
6176 NULL, /* meter_del */
6177 NULL, /* group_alloc */
6178 NULL, /* group_construct */
6179 NULL, /* group_destruct */
6180 NULL, /* group_dealloc */
6181 NULL, /* group_modify */
6182 NULL, /* group_get_stats */