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-monitor.h"
56 #include "ofproto-dpif-sflow.h"
57 #include "ofproto-dpif-upcall.h"
58 #include "ofproto-dpif-xlate.h"
59 #include "poll-loop.h"
64 #include "unaligned.h"
66 #include "vlan-bitmap.h"
69 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
71 COVERAGE_DEFINE(ofproto_dpif_expired);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(facet_unexpected);
74 COVERAGE_DEFINE(facet_create);
75 COVERAGE_DEFINE(facet_remove);
76 COVERAGE_DEFINE(subfacet_create);
77 COVERAGE_DEFINE(subfacet_destroy);
78 COVERAGE_DEFINE(subfacet_install_fail);
79 COVERAGE_DEFINE(packet_in_overflow);
81 /* Number of implemented OpenFlow tables. */
82 enum { N_TABLES = 255 };
83 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
84 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
94 * - Do include packets and bytes from facets that have been deleted or
95 * whose own statistics have been folded into the rule.
97 * - Do include packets and bytes sent "by hand" that were accounted to
98 * the rule without any facet being involved (this is a rare corner
99 * case in rule_execute()).
101 * - Do not include packet or bytes that can be obtained from any facet's
102 * packet_count or byte_count member or that can be obtained from the
103 * datapath by, e.g., dpif_flow_get() for any subfacet.
105 struct ovs_mutex stats_mutex;
106 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
107 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
110 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
118 * - Do include packets and bytes from facets that have been deleted or
119 * whose own statistics have been folded into the rule.
121 * - Do include packets and bytes sent "by hand" that were accounted to
122 * the rule without any facet being involved (this is a rare corner
123 * case in rule_execute()).
125 * - Do not include packet or bytes that can be obtained from any facet's
126 * packet_count or byte_count member or that can be obtained from the
127 * datapath by, e.g., dpif_flow_get() for any subfacet.
129 struct ovs_mutex stats_mutex;
130 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
131 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
132 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
136 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
137 struct ofproto_dpif *ofproto; /* Owning ofproto. */
138 void *aux; /* Key supplied by ofproto's client. */
139 char *name; /* Identifier for log messages. */
142 struct list ports; /* Contains "struct ofport"s. */
143 enum port_vlan_mode vlan_mode; /* VLAN mode */
144 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
145 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
146 * NULL if all VLANs are trunked. */
147 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
148 struct bond *bond; /* Nonnull iff more than one port. */
149 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
152 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
155 static void bundle_remove(struct ofport *);
156 static void bundle_update(struct ofbundle *);
157 static void bundle_destroy(struct ofbundle *);
158 static void bundle_del_port(struct ofport_dpif *);
159 static void bundle_run(struct ofbundle *);
160 static void bundle_wait(struct ofbundle *);
162 static void stp_run(struct ofproto_dpif *ofproto);
163 static void stp_wait(struct ofproto_dpif *ofproto);
164 static int set_stp_port(struct ofport *,
165 const struct ofproto_port_stp_settings *);
167 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
168 enum slow_path_reason,
169 uint64_t *stub, size_t stub_size,
170 const struct nlattr **actionsp,
171 size_t *actions_lenp);
173 /* A subfacet (see "struct subfacet" below) has three possible installation
176 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
177 * case just after the subfacet is created, just before the subfacet is
178 * destroyed, or if the datapath returns an error when we try to install a
181 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
183 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
184 * ofproto_dpif is installed in the datapath.
187 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
188 SF_FAST_PATH, /* Full actions are installed. */
189 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
192 /* A dpif flow and actions associated with a facet.
194 * See also the large comment on struct facet. */
197 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
198 struct list list_node; /* In struct facet's 'facets' list. */
199 struct facet *facet; /* Owning facet. */
200 struct dpif_backer *backer; /* Owning backer. */
205 long long int used; /* Time last used; time created if not used. */
206 long long int created; /* Time created. */
208 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
209 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
211 enum subfacet_path path; /* Installed in datapath? */
214 #define SUBFACET_DESTROY_MAX_BATCH 50
216 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *,
218 static struct subfacet *subfacet_find(struct dpif_backer *,
219 const struct nlattr *key, size_t key_len,
221 static void subfacet_destroy(struct subfacet *);
222 static void subfacet_destroy__(struct subfacet *);
223 static void subfacet_destroy_batch(struct dpif_backer *,
224 struct subfacet **, int n);
225 static void subfacet_reset_dp_stats(struct subfacet *,
226 struct dpif_flow_stats *);
227 static void subfacet_update_stats(struct subfacet *,
228 const struct dpif_flow_stats *);
229 static int subfacet_install(struct subfacet *,
230 const struct ofpbuf *odp_actions,
231 struct dpif_flow_stats *);
232 static void subfacet_uninstall(struct subfacet *);
234 /* A unique, non-overlapping instantiation of an OpenFlow flow.
236 * A facet associates a "struct flow", which represents the Open vSwitch
237 * userspace idea of an exact-match flow, with one or more subfacets.
238 * While the facet is created based on an exact-match flow, it is stored
239 * within the ofproto based on the wildcards that could be expressed
240 * based on the flow table and other configuration. (See the 'wc'
241 * description in "struct xlate_out" for more details.)
243 * Each subfacet tracks the datapath's idea of the flow equivalent to
244 * the facet. When the kernel module (or other dpif implementation) and
245 * Open vSwitch userspace agree on the definition of a flow key, there
246 * is exactly one subfacet per facet. If the dpif implementation
247 * supports more-specific flow matching than userspace, however, a facet
248 * can have more than one subfacet. Examples include the dpif
249 * implementation not supporting the same wildcards as userspace or some
250 * distinction in flow that userspace simply doesn't understand.
252 * Flow expiration works in terms of subfacets, so a facet must have at
253 * least one subfacet or it will never expire, leaking memory. */
256 struct ofproto_dpif *ofproto;
259 struct list subfacets;
260 long long int used; /* Time last used; time created if not used. */
263 struct flow flow; /* Flow of the creating subfacet. */
264 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
268 * - Do include packets and bytes sent "by hand", e.g. with
271 * - Do include packets and bytes that were obtained from the datapath
272 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
273 * DPIF_FP_ZERO_STATS).
275 * - Do not include packets or bytes that can be obtained from the
276 * datapath for any existing subfacet.
278 uint64_t packet_count; /* Number of packets received. */
279 uint64_t byte_count; /* Number of bytes received. */
281 /* Resubmit statistics. */
282 uint64_t prev_packet_count; /* Number of packets from last stats push. */
283 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
284 long long int prev_used; /* Used time from last stats push. */
287 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
288 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
289 uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */
291 struct xlate_out xout;
293 /* Storage for a single subfacet, to reduce malloc() time and space
294 * overhead. (A facet always has at least one subfacet and in the common
295 * case has exactly one subfacet. However, 'one_subfacet' may not
296 * always be valid, since it could have been removed after newer
297 * subfacets were pushed onto the 'subfacets' list.) */
298 struct subfacet one_subfacet;
300 long long int learn_rl; /* Rate limiter for facet_learn(). */
303 static struct facet *facet_create(const struct flow_miss *);
304 static void facet_remove(struct facet *);
305 static void facet_free(struct facet *);
307 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
308 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
309 const struct flow *);
310 static bool facet_revalidate(struct facet *);
311 static bool facet_check_consistency(struct facet *);
313 static void facet_flush_stats(struct facet *);
315 static void facet_reset_counters(struct facet *);
316 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
317 struct dpif_flow_stats *, bool may_learn);
318 static void facet_push_stats(struct facet *, bool may_learn);
319 static void facet_learn(struct facet *);
320 static void facet_account(struct facet *);
321 static void push_all_stats(void);
323 static bool facet_is_controller_flow(struct facet *);
326 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
330 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
331 struct list bundle_node; /* In struct ofbundle's "ports" list. */
332 struct cfm *cfm; /* Connectivity Fault Management, if any. */
333 struct bfd *bfd; /* BFD, if any. */
334 bool may_enable; /* May be enabled in bonds. */
335 bool is_tunnel; /* This port is a tunnel. */
336 bool is_layer3; /* This is a layer 3 port. */
337 long long int carrier_seq; /* Carrier status changes. */
338 struct ofport_dpif *peer; /* Peer if patch port. */
341 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
342 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
343 long long int stp_state_entered;
345 /* Queue to DSCP mapping. */
346 struct ofproto_port_queue *qdscp;
349 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
351 * This is deprecated. It is only for compatibility with broken device
352 * drivers in old versions of Linux that do not properly support VLANs when
353 * VLAN devices are not used. When broken device drivers are no longer in
354 * widespread use, we will delete these interfaces. */
355 ofp_port_t realdev_ofp_port;
359 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
361 * This is deprecated. It is only for compatibility with broken device drivers
362 * in old versions of Linux that do not properly support VLANs when VLAN
363 * devices are not used. When broken device drivers are no longer in
364 * widespread use, we will delete these interfaces. */
365 struct vlan_splinter {
366 struct hmap_node realdev_vid_node;
367 struct hmap_node vlandev_node;
368 ofp_port_t realdev_ofp_port;
369 ofp_port_t vlandev_ofp_port;
373 static void vsp_remove(struct ofport_dpif *);
374 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
376 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
379 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
382 static struct ofport_dpif *
383 ofport_dpif_cast(const struct ofport *ofport)
385 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
388 static void port_run(struct ofport_dpif *);
389 static int set_bfd(struct ofport *, const struct smap *);
390 static int set_cfm(struct ofport *, const struct cfm_settings *);
391 static void ofport_update_peer(struct ofport_dpif *);
392 static void run_fast_rl(void);
393 static int run_fast(struct ofproto *);
395 struct dpif_completion {
396 struct list list_node;
397 struct ofoperation *op;
400 /* Reasons that we might need to revalidate every facet, and corresponding
403 * A value of 0 means that there is no need to revalidate.
405 * It would be nice to have some cleaner way to integrate with coverage
406 * counters, but with only a few reasons I guess this is good enough for
408 enum revalidate_reason {
409 REV_RECONFIGURE = 1, /* Switch configuration changed. */
410 REV_STP, /* Spanning tree protocol port status change. */
411 REV_BOND, /* Bonding changed. */
412 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
413 REV_FLOW_TABLE, /* Flow table changed. */
414 REV_MAC_LEARNING, /* Mac learning changed. */
415 REV_INCONSISTENCY /* Facet self-check failed. */
417 COVERAGE_DEFINE(rev_reconfigure);
418 COVERAGE_DEFINE(rev_stp);
419 COVERAGE_DEFINE(rev_bond);
420 COVERAGE_DEFINE(rev_port_toggled);
421 COVERAGE_DEFINE(rev_flow_table);
422 COVERAGE_DEFINE(rev_mac_learning);
423 COVERAGE_DEFINE(rev_inconsistency);
425 struct avg_subfacet_rates {
426 double add_rate; /* Moving average of new flows created per minute. */
427 double del_rate; /* Moving average of flows deleted per minute. */
430 /* All datapaths of a given type share a single dpif backer instance. */
436 struct timer next_expiration;
438 struct ovs_rwlock odp_to_ofport_lock;
439 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
441 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
443 /* Facet revalidation flags applying to facets which use this backer. */
444 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
446 struct hmap drop_keys; /* Set of dropped odp keys. */
447 bool recv_set_enable; /* Enables or disables receiving packets. */
449 struct hmap subfacets;
450 struct governor *governor;
452 /* Subfacet statistics.
454 * These keep track of the total number of subfacets added and deleted and
455 * flow life span. They are useful for computing the flow rates stats
456 * exposed via "ovs-appctl dpif/show". The goal is to learn about
457 * traffic patterns in ways that we can use later to improve Open vSwitch
458 * performance in new situations. */
459 long long int created; /* Time when it is created. */
460 unsigned max_n_subfacet; /* Maximum number of flows */
461 unsigned avg_n_subfacet; /* Average number of flows. */
462 long long int avg_subfacet_life; /* Average life span of subfacets. */
464 /* Number of upcall handling threads. */
465 unsigned int n_handler_threads;
468 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
469 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
471 static void drop_key_clear(struct dpif_backer *);
473 struct ofproto_dpif {
474 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
476 struct dpif_backer *backer;
478 /* Special OpenFlow rules. */
479 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
480 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
481 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
484 struct netflow *netflow;
485 struct dpif_sflow *sflow;
486 struct dpif_ipfix *ipfix;
487 struct hmap bundles; /* Contains "struct ofbundle"s. */
488 struct mac_learning *ml;
489 bool has_bonded_bundles;
490 struct mbridge *mbridge;
493 struct classifier facets; /* Contains 'struct facet's. */
494 long long int consistency_rl;
496 struct ovs_mutex stats_mutex;
497 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
498 * consumed in userspace. */
502 long long int stp_last_tick;
504 /* VLAN splinters. */
505 struct ovs_mutex vsp_mutex;
506 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
507 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
510 struct sset ports; /* Set of standard port names. */
511 struct sset ghost_ports; /* Ports with no datapath port. */
512 struct sset port_poll_set; /* Queued names for port_poll() reply. */
513 int port_poll_errno; /* Last errno for port_poll() reply. */
515 /* Per ofproto's dpif stats. */
520 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
523 /* By default, flows in the datapath are wildcarded (megaflows). They
524 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
525 static bool enable_megaflows = true;
527 /* All existing ofproto_dpif instances, indexed by ->up.name. */
528 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
530 static void ofproto_dpif_unixctl_init(void);
532 static inline struct ofproto_dpif *
533 ofproto_dpif_cast(const struct ofproto *ofproto)
535 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
536 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
539 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
540 ofp_port_t ofp_port);
541 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
542 const struct ofpbuf *packet,
543 const struct ofpact[], size_t ofpacts_len,
547 static void handle_upcalls(struct dpif_backer *);
549 /* Flow expiration. */
550 static int expire(struct dpif_backer *);
553 static void send_netflow_active_timeouts(struct ofproto_dpif *);
555 /* Global variables. */
556 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
558 /* Initial mappings of port to bridge mappings. */
559 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
561 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
564 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
565 struct ofputil_flow_mod *fm)
567 ofproto_flow_mod(&ofproto->up, fm);
570 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
571 * Takes ownership of 'pin' and pin->packet. */
573 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
574 struct ofproto_packet_in *pin)
576 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
577 COVERAGE_INC(packet_in_overflow);
578 free(CONST_CAST(void *, pin->up.packet));
583 /* Factory functions. */
586 init(const struct shash *iface_hints)
588 struct shash_node *node;
590 /* Make a local copy, since we don't own 'iface_hints' elements. */
591 SHASH_FOR_EACH(node, iface_hints) {
592 const struct iface_hint *orig_hint = node->data;
593 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
595 new_hint->br_name = xstrdup(orig_hint->br_name);
596 new_hint->br_type = xstrdup(orig_hint->br_type);
597 new_hint->ofp_port = orig_hint->ofp_port;
599 shash_add(&init_ofp_ports, node->name, new_hint);
604 enumerate_types(struct sset *types)
606 dp_enumerate_types(types);
610 enumerate_names(const char *type, struct sset *names)
612 struct ofproto_dpif *ofproto;
615 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
616 if (strcmp(type, ofproto->up.type)) {
619 sset_add(names, ofproto->up.name);
626 del(const char *type, const char *name)
631 error = dpif_open(name, type, &dpif);
633 error = dpif_delete(dpif);
640 port_open_type(const char *datapath_type, const char *port_type)
642 return dpif_port_open_type(datapath_type, port_type);
645 /* Type functions. */
647 static void process_dpif_port_changes(struct dpif_backer *);
648 static void process_dpif_all_ports_changed(struct dpif_backer *);
649 static void process_dpif_port_change(struct dpif_backer *,
650 const char *devname);
651 static void process_dpif_port_error(struct dpif_backer *, int error);
653 static struct ofproto_dpif *
654 lookup_ofproto_dpif_by_port_name(const char *name)
656 struct ofproto_dpif *ofproto;
658 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
659 if (sset_contains(&ofproto->ports, name)) {
668 type_run(const char *type)
670 static long long int push_timer = LLONG_MIN;
671 struct dpif_backer *backer;
673 backer = shash_find_data(&all_dpif_backers, type);
675 /* This is not necessarily a problem, since backers are only
676 * created on demand. */
680 dpif_run(backer->dpif);
682 /* The most natural place to push facet statistics is when they're pulled
683 * from the datapath. However, when there are many flows in the datapath,
684 * this expensive operation can occur so frequently, that it reduces our
685 * ability to quickly set up flows. To reduce the cost, we push statistics
687 if (time_msec() > push_timer) {
688 push_timer = time_msec() + 2000;
692 /* If vswitchd started with other_config:flow_restore_wait set as "true",
693 * and the configuration has now changed to "false", enable receiving
694 * packets from the datapath. */
695 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
698 backer->recv_set_enable = true;
700 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
702 udpif_recv_set(backer->udpif, 0, false);
703 VLOG_ERR("Failed to enable receiving packets in dpif.");
706 udpif_recv_set(backer->udpif, n_handler_threads,
707 backer->recv_set_enable);
708 dpif_flow_flush(backer->dpif);
709 backer->need_revalidate = REV_RECONFIGURE;
712 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
713 * to reset the handler threads. */
714 if (backer->n_handler_threads != n_handler_threads) {
715 udpif_recv_set(backer->udpif, n_handler_threads,
716 backer->recv_set_enable);
717 backer->n_handler_threads = n_handler_threads;
720 if (backer->need_revalidate) {
721 struct ofproto_dpif *ofproto;
722 struct simap_node *node;
723 struct simap tmp_backers;
725 /* Handle tunnel garbage collection. */
726 simap_init(&tmp_backers);
727 simap_swap(&backer->tnl_backers, &tmp_backers);
729 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
730 struct ofport_dpif *iter;
732 if (backer != ofproto->backer) {
736 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
737 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
740 if (!iter->is_tunnel) {
744 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
745 namebuf, sizeof namebuf);
746 node = simap_find(&tmp_backers, dp_port);
748 simap_put(&backer->tnl_backers, dp_port, node->data);
749 simap_delete(&tmp_backers, node);
750 node = simap_find(&backer->tnl_backers, dp_port);
752 node = simap_find(&backer->tnl_backers, dp_port);
754 odp_port_t odp_port = ODPP_NONE;
756 if (!dpif_port_add(backer->dpif, iter->up.netdev,
758 simap_put(&backer->tnl_backers, dp_port,
759 odp_to_u32(odp_port));
760 node = simap_find(&backer->tnl_backers, dp_port);
765 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
766 if (tnl_port_reconfigure(iter, iter->up.netdev,
768 backer->need_revalidate = REV_RECONFIGURE;
773 SIMAP_FOR_EACH (node, &tmp_backers) {
774 dpif_port_del(backer->dpif, u32_to_odp(node->data));
776 simap_destroy(&tmp_backers);
778 switch (backer->need_revalidate) {
779 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
780 case REV_STP: COVERAGE_INC(rev_stp); break;
781 case REV_BOND: COVERAGE_INC(rev_bond); break;
782 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
783 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
784 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
785 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
787 backer->need_revalidate = 0;
789 /* Clear the drop_keys in case we should now be accepting some
790 * formerly dropped flows. */
791 drop_key_clear(backer);
793 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
794 struct facet *facet, *next;
795 struct ofport_dpif *ofport;
796 struct cls_cursor cursor;
797 struct ofbundle *bundle;
799 if (ofproto->backer != backer) {
803 ovs_rwlock_wrlock(&xlate_rwlock);
804 xlate_ofproto_set(ofproto, ofproto->up.name,
805 ofproto->backer->dpif, ofproto->miss_rule,
806 ofproto->no_packet_in_rule, ofproto->ml,
807 ofproto->stp, ofproto->mbridge,
808 ofproto->sflow, ofproto->ipfix,
809 ofproto->up.frag_handling,
810 ofproto->up.forward_bpdu,
811 connmgr_has_in_band(ofproto->up.connmgr),
812 ofproto->netflow != NULL);
814 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
815 xlate_bundle_set(ofproto, bundle, bundle->name,
816 bundle->vlan_mode, bundle->vlan,
817 bundle->trunks, bundle->use_priority_tags,
818 bundle->bond, bundle->lacp,
822 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
823 int stp_port = ofport->stp_port
824 ? stp_port_no(ofport->stp_port)
826 xlate_ofport_set(ofproto, ofport->bundle, ofport,
827 ofport->up.ofp_port, ofport->odp_port,
828 ofport->up.netdev, ofport->cfm,
829 ofport->bfd, ofport->peer, stp_port,
830 ofport->qdscp, ofport->n_qdscp,
831 ofport->up.pp.config, ofport->up.pp.state,
832 ofport->is_tunnel, ofport->may_enable);
834 ovs_rwlock_unlock(&xlate_rwlock);
836 /* Only ofproto-dpif cares about the facet classifier so we just
837 * lock cls_cursor_init() to appease the thread safety analysis. */
838 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
839 cls_cursor_init(&cursor, &ofproto->facets, NULL);
840 ovs_rwlock_unlock(&ofproto->facets.rwlock);
841 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
842 facet_revalidate(facet);
847 udpif_revalidate(backer->udpif);
850 if (!backer->recv_set_enable) {
851 /* Wake up before a max of 1000ms. */
852 timer_set_duration(&backer->next_expiration, 1000);
853 } else if (timer_expired(&backer->next_expiration)) {
854 int delay = expire(backer);
855 timer_set_duration(&backer->next_expiration, delay);
858 process_dpif_port_changes(backer);
860 if (backer->governor) {
863 governor_run(backer->governor);
865 /* If the governor has shrunk to its minimum size and the number of
866 * subfacets has dwindled, then drop the governor entirely.
868 * For hysteresis, the number of subfacets to drop the governor is
869 * smaller than the number needed to trigger its creation. */
870 n_subfacets = hmap_count(&backer->subfacets);
871 if (n_subfacets * 4 < flow_eviction_threshold
872 && governor_is_idle(backer->governor)) {
873 governor_destroy(backer->governor);
874 backer->governor = NULL;
881 /* Check for and handle port changes in 'backer''s dpif. */
883 process_dpif_port_changes(struct dpif_backer *backer)
889 error = dpif_port_poll(backer->dpif, &devname);
895 process_dpif_all_ports_changed(backer);
899 process_dpif_port_change(backer, devname);
904 process_dpif_port_error(backer, error);
911 process_dpif_all_ports_changed(struct dpif_backer *backer)
913 struct ofproto_dpif *ofproto;
914 struct dpif_port dpif_port;
915 struct dpif_port_dump dump;
916 struct sset devnames;
919 sset_init(&devnames);
920 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
921 if (ofproto->backer == backer) {
922 struct ofport *ofport;
924 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
925 sset_add(&devnames, netdev_get_name(ofport->netdev));
929 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
930 sset_add(&devnames, dpif_port.name);
933 SSET_FOR_EACH (devname, &devnames) {
934 process_dpif_port_change(backer, devname);
936 sset_destroy(&devnames);
940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
942 struct ofproto_dpif *ofproto;
943 struct dpif_port port;
945 /* Don't report on the datapath's device. */
946 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
950 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
951 &all_ofproto_dpifs) {
952 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
957 ofproto = lookup_ofproto_dpif_by_port_name(devname);
958 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
959 /* The port was removed. If we know the datapath,
960 * report it through poll_set(). If we don't, it may be
961 * notifying us of a removal we initiated, so ignore it.
962 * If there's a pending ENOBUFS, let it stand, since
963 * everything will be reevaluated. */
964 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
965 sset_add(&ofproto->port_poll_set, devname);
966 ofproto->port_poll_errno = 0;
968 } else if (!ofproto) {
969 /* The port was added, but we don't know with which
970 * ofproto we should associate it. Delete it. */
971 dpif_port_del(backer->dpif, port.port_no);
973 struct ofport_dpif *ofport;
975 ofport = ofport_dpif_cast(shash_find_data(
976 &ofproto->up.port_by_name, devname));
978 && ofport->odp_port != port.port_no
979 && !odp_port_to_ofport(backer, port.port_no))
981 /* 'ofport''s datapath port number has changed from
982 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
983 * structures to match. */
984 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
985 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
986 ofport->odp_port = port.port_no;
987 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
988 hash_odp_port(port.port_no));
989 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
990 backer->need_revalidate = REV_RECONFIGURE;
993 dpif_port_destroy(&port);
996 /* Propagate 'error' to all ofprotos based on 'backer'. */
998 process_dpif_port_error(struct dpif_backer *backer, int error)
1000 struct ofproto_dpif *ofproto;
1002 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003 if (ofproto->backer == backer) {
1004 sset_clear(&ofproto->port_poll_set);
1005 ofproto->port_poll_errno = error;
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1013 handle_upcalls(backer);
1019 type_run_fast(const char *type)
1021 struct dpif_backer *backer;
1023 backer = shash_find_data(&all_dpif_backers, type);
1025 /* This is not necessarily a problem, since backers are only
1026 * created on demand. */
1030 return dpif_backer_run_fast(backer);
1036 static long long int port_rl = LLONG_MIN;
1038 if (time_msec() >= port_rl) {
1039 struct ofproto_dpif *ofproto;
1041 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1042 run_fast(&ofproto->up);
1044 port_rl = time_msec() + 200;
1049 type_wait(const char *type)
1051 struct dpif_backer *backer;
1053 backer = shash_find_data(&all_dpif_backers, type);
1055 /* This is not necessarily a problem, since backers are only
1056 * created on demand. */
1060 if (backer->governor) {
1061 governor_wait(backer->governor);
1064 timer_wait(&backer->next_expiration);
1065 dpif_wait(backer->dpif);
1066 udpif_wait(backer->udpif);
1069 /* Basic life-cycle. */
1071 static int add_internal_flows(struct ofproto_dpif *);
1073 static struct ofproto *
1076 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1077 return &ofproto->up;
1081 dealloc(struct ofproto *ofproto_)
1083 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1088 close_dpif_backer(struct dpif_backer *backer)
1090 ovs_assert(backer->refcount > 0);
1092 if (--backer->refcount) {
1096 drop_key_clear(backer);
1097 hmap_destroy(&backer->drop_keys);
1099 udpif_destroy(backer->udpif);
1101 simap_destroy(&backer->tnl_backers);
1102 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1103 hmap_destroy(&backer->odp_to_ofport_map);
1104 shash_find_and_delete(&all_dpif_backers, backer->type);
1106 dpif_close(backer->dpif);
1108 ovs_assert(hmap_is_empty(&backer->subfacets));
1109 hmap_destroy(&backer->subfacets);
1110 governor_destroy(backer->governor);
1115 /* Datapath port slated for removal from datapath. */
1116 struct odp_garbage {
1117 struct list list_node;
1118 odp_port_t odp_port;
1122 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1124 struct dpif_backer *backer;
1125 struct dpif_port_dump port_dump;
1126 struct dpif_port port;
1127 struct shash_node *node;
1128 struct list garbage_list;
1129 struct odp_garbage *garbage, *next;
1135 backer = shash_find_data(&all_dpif_backers, type);
1142 backer_name = xasprintf("ovs-%s", type);
1144 /* Remove any existing datapaths, since we assume we're the only
1145 * userspace controlling the datapath. */
1147 dp_enumerate_names(type, &names);
1148 SSET_FOR_EACH(name, &names) {
1149 struct dpif *old_dpif;
1151 /* Don't remove our backer if it exists. */
1152 if (!strcmp(name, backer_name)) {
1156 if (dpif_open(name, type, &old_dpif)) {
1157 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1159 dpif_delete(old_dpif);
1160 dpif_close(old_dpif);
1163 sset_destroy(&names);
1165 backer = xmalloc(sizeof *backer);
1167 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1170 VLOG_ERR("failed to open datapath of type %s: %s", type,
1171 ovs_strerror(error));
1175 backer->udpif = udpif_create(backer, backer->dpif);
1177 backer->type = xstrdup(type);
1178 backer->governor = NULL;
1179 backer->refcount = 1;
1180 hmap_init(&backer->odp_to_ofport_map);
1181 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1182 hmap_init(&backer->drop_keys);
1183 hmap_init(&backer->subfacets);
1184 timer_set_duration(&backer->next_expiration, 1000);
1185 backer->need_revalidate = 0;
1186 simap_init(&backer->tnl_backers);
1187 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1190 if (backer->recv_set_enable) {
1191 dpif_flow_flush(backer->dpif);
1194 /* Loop through the ports already on the datapath and remove any
1195 * that we don't need anymore. */
1196 list_init(&garbage_list);
1197 dpif_port_dump_start(&port_dump, backer->dpif);
1198 while (dpif_port_dump_next(&port_dump, &port)) {
1199 node = shash_find(&init_ofp_ports, port.name);
1200 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1201 garbage = xmalloc(sizeof *garbage);
1202 garbage->odp_port = port.port_no;
1203 list_push_front(&garbage_list, &garbage->list_node);
1206 dpif_port_dump_done(&port_dump);
1208 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1209 dpif_port_del(backer->dpif, garbage->odp_port);
1210 list_remove(&garbage->list_node);
1214 shash_add(&all_dpif_backers, type, backer);
1216 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1218 VLOG_ERR("failed to listen on datapath of type %s: %s",
1219 type, ovs_strerror(error));
1220 close_dpif_backer(backer);
1223 udpif_recv_set(backer->udpif, n_handler_threads,
1224 backer->recv_set_enable);
1225 backer->n_handler_threads = n_handler_threads;
1227 backer->max_n_subfacet = 0;
1228 backer->created = time_msec();
1229 backer->avg_n_subfacet = 0;
1230 backer->avg_subfacet_life = 0;
1236 construct(struct ofproto *ofproto_)
1238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1239 struct shash_node *node, *next;
1242 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1247 ofproto->netflow = NULL;
1248 ofproto->sflow = NULL;
1249 ofproto->ipfix = NULL;
1250 ofproto->stp = NULL;
1251 hmap_init(&ofproto->bundles);
1252 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1253 ofproto->mbridge = mbridge_create();
1254 ofproto->has_bonded_bundles = false;
1255 ovs_mutex_init(&ofproto->stats_mutex);
1256 ovs_mutex_init(&ofproto->vsp_mutex);
1258 classifier_init(&ofproto->facets, NULL);
1259 ofproto->consistency_rl = LLONG_MIN;
1261 guarded_list_init(&ofproto->pins);
1263 ofproto_dpif_unixctl_init();
1265 hmap_init(&ofproto->vlandev_map);
1266 hmap_init(&ofproto->realdev_vid_map);
1268 sset_init(&ofproto->ports);
1269 sset_init(&ofproto->ghost_ports);
1270 sset_init(&ofproto->port_poll_set);
1271 ofproto->port_poll_errno = 0;
1273 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1274 struct iface_hint *iface_hint = node->data;
1276 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1277 /* Check if the datapath already has this port. */
1278 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1279 sset_add(&ofproto->ports, node->name);
1282 free(iface_hint->br_name);
1283 free(iface_hint->br_type);
1285 shash_delete(&init_ofp_ports, node);
1289 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1290 hash_string(ofproto->up.name, 0));
1291 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1293 ofproto_init_tables(ofproto_, N_TABLES);
1294 error = add_internal_flows(ofproto);
1295 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1298 ofproto->n_missed = 0;
1304 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1305 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1307 struct ofputil_flow_mod fm;
1310 match_init_catchall(&fm.match);
1312 match_set_reg(&fm.match, 0, id);
1313 fm.new_cookie = htonll(0);
1314 fm.cookie = htonll(0);
1315 fm.cookie_mask = htonll(0);
1316 fm.modify_cookie = false;
1317 fm.table_id = TBL_INTERNAL;
1318 fm.command = OFPFC_ADD;
1319 fm.idle_timeout = 0;
1320 fm.hard_timeout = 0;
1324 fm.ofpacts = ofpacts->data;
1325 fm.ofpacts_len = ofpacts->size;
1327 error = ofproto_flow_mod(&ofproto->up, &fm);
1329 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1330 id, ofperr_to_string(error));
1334 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1336 rule_dpif_unref(*rulep);
1345 add_internal_flows(struct ofproto_dpif *ofproto)
1347 struct ofpact_controller *controller;
1348 uint64_t ofpacts_stub[128 / 8];
1349 struct ofpbuf ofpacts;
1353 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1356 controller = ofpact_put_CONTROLLER(&ofpacts);
1357 controller->max_len = UINT16_MAX;
1358 controller->controller_id = 0;
1359 controller->reason = OFPR_NO_MATCH;
1360 ofpact_pad(&ofpacts);
1362 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1367 ofpbuf_clear(&ofpacts);
1368 error = add_internal_flow(ofproto, id++, &ofpacts,
1369 &ofproto->no_packet_in_rule);
1374 error = add_internal_flow(ofproto, id++, &ofpacts,
1375 &ofproto->drop_frags_rule);
1380 destruct(struct ofproto *ofproto_)
1382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1383 struct rule_dpif *rule, *next_rule;
1384 struct ofproto_packet_in *pin, *next_pin;
1385 struct facet *facet, *next_facet;
1386 struct cls_cursor cursor;
1387 struct oftable *table;
1390 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1391 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1392 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1393 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1394 facet_remove(facet);
1397 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1398 ovs_rwlock_wrlock(&xlate_rwlock);
1399 xlate_remove_ofproto(ofproto);
1400 ovs_rwlock_unlock(&xlate_rwlock);
1402 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1403 * use-after-free error. */
1404 udpif_revalidate(ofproto->backer->udpif);
1406 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1408 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1409 struct cls_cursor cursor;
1411 ovs_rwlock_rdlock(&table->cls.rwlock);
1412 cls_cursor_init(&cursor, &table->cls, NULL);
1413 ovs_rwlock_unlock(&table->cls.rwlock);
1414 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1415 ofproto_rule_delete(&ofproto->up, &rule->up);
1419 guarded_list_pop_all(&ofproto->pins, &pins);
1420 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1421 list_remove(&pin->list_node);
1422 free(CONST_CAST(void *, pin->up.packet));
1425 guarded_list_destroy(&ofproto->pins);
1427 mbridge_unref(ofproto->mbridge);
1429 netflow_destroy(ofproto->netflow);
1430 dpif_sflow_unref(ofproto->sflow);
1431 hmap_destroy(&ofproto->bundles);
1432 mac_learning_unref(ofproto->ml);
1434 classifier_destroy(&ofproto->facets);
1436 hmap_destroy(&ofproto->vlandev_map);
1437 hmap_destroy(&ofproto->realdev_vid_map);
1439 sset_destroy(&ofproto->ports);
1440 sset_destroy(&ofproto->ghost_ports);
1441 sset_destroy(&ofproto->port_poll_set);
1443 ovs_mutex_destroy(&ofproto->stats_mutex);
1444 ovs_mutex_destroy(&ofproto->vsp_mutex);
1446 close_dpif_backer(ofproto->backer);
1450 run_fast(struct ofproto *ofproto_)
1452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1453 struct ofproto_packet_in *pin, *next_pin;
1456 /* Do not perform any periodic activity required by 'ofproto' while
1457 * waiting for flow restore to complete. */
1458 if (ofproto_get_flow_restore_wait()) {
1462 guarded_list_pop_all(&ofproto->pins, &pins);
1463 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1464 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1465 list_remove(&pin->list_node);
1466 free(CONST_CAST(void *, pin->up.packet));
1474 run(struct ofproto *ofproto_)
1476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1477 struct ofport_dpif *ofport;
1478 struct ofbundle *bundle;
1481 if (mbridge_need_revalidate(ofproto->mbridge)) {
1482 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1483 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1484 mac_learning_flush(ofproto->ml);
1485 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1488 /* Do not perform any periodic activity below required by 'ofproto' while
1489 * waiting for flow restore to complete. */
1490 if (ofproto_get_flow_restore_wait()) {
1494 error = run_fast(ofproto_);
1499 if (ofproto->netflow) {
1500 if (netflow_run(ofproto->netflow)) {
1501 send_netflow_active_timeouts(ofproto);
1504 if (ofproto->sflow) {
1505 dpif_sflow_run(ofproto->sflow);
1507 if (ofproto->ipfix) {
1508 dpif_ipfix_run(ofproto->ipfix);
1511 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1514 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1519 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1520 if (mac_learning_run(ofproto->ml)) {
1521 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1523 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1525 /* Check the consistency of a random facet, to aid debugging. */
1526 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1527 if (time_msec() >= ofproto->consistency_rl
1528 && !classifier_is_empty(&ofproto->facets)
1529 && !ofproto->backer->need_revalidate) {
1530 struct cls_subtable *table;
1531 struct cls_rule *cr;
1532 struct facet *facet;
1534 ofproto->consistency_rl = time_msec() + 250;
1536 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1537 struct cls_subtable, hmap_node);
1538 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1540 facet = CONTAINER_OF(cr, struct facet, cr);
1542 if (!facet_check_consistency(facet)) {
1543 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1546 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1552 wait(struct ofproto *ofproto_)
1554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555 struct ofbundle *bundle;
1557 if (ofproto_get_flow_restore_wait()) {
1561 if (ofproto->sflow) {
1562 dpif_sflow_wait(ofproto->sflow);
1564 if (ofproto->ipfix) {
1565 dpif_ipfix_wait(ofproto->ipfix);
1567 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1568 bundle_wait(bundle);
1570 if (ofproto->netflow) {
1571 netflow_wait(ofproto->netflow);
1573 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1574 mac_learning_wait(ofproto->ml);
1575 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1577 if (ofproto->backer->need_revalidate) {
1578 /* Shouldn't happen, but if it does just go around again. */
1579 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1580 poll_immediate_wake();
1585 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1587 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1588 struct cls_cursor cursor;
1589 size_t n_subfacets = 0;
1590 struct facet *facet;
1592 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1593 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1594 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1596 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1597 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1598 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1599 n_subfacets += list_size(&facet->subfacets);
1601 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1602 simap_increase(usage, "subfacets", n_subfacets);
1606 flush(struct ofproto *ofproto_)
1608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1609 struct subfacet *subfacet, *next_subfacet;
1610 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1614 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1615 &ofproto->backer->subfacets) {
1616 if (subfacet->facet->ofproto != ofproto) {
1620 if (subfacet->path != SF_NOT_INSTALLED) {
1621 batch[n_batch++] = subfacet;
1622 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1623 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1627 subfacet_destroy(subfacet);
1632 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1637 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1638 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1640 *arp_match_ip = true;
1641 *actions = (OFPUTIL_A_OUTPUT |
1642 OFPUTIL_A_SET_VLAN_VID |
1643 OFPUTIL_A_SET_VLAN_PCP |
1644 OFPUTIL_A_STRIP_VLAN |
1645 OFPUTIL_A_SET_DL_SRC |
1646 OFPUTIL_A_SET_DL_DST |
1647 OFPUTIL_A_SET_NW_SRC |
1648 OFPUTIL_A_SET_NW_DST |
1649 OFPUTIL_A_SET_NW_TOS |
1650 OFPUTIL_A_SET_TP_SRC |
1651 OFPUTIL_A_SET_TP_DST |
1656 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659 struct dpif_dp_stats s;
1660 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1663 strcpy(ots->name, "classifier");
1665 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1666 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1667 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1668 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1670 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1671 ots->lookup_count = htonll(n_lookup);
1672 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1675 static struct ofport *
1678 struct ofport_dpif *port = xmalloc(sizeof *port);
1683 port_dealloc(struct ofport *port_)
1685 struct ofport_dpif *port = ofport_dpif_cast(port_);
1690 port_construct(struct ofport *port_)
1692 struct ofport_dpif *port = ofport_dpif_cast(port_);
1693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1694 const struct netdev *netdev = port->up.netdev;
1695 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1696 struct dpif_port dpif_port;
1699 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1700 port->bundle = NULL;
1703 port->may_enable = true;
1704 port->stp_port = NULL;
1705 port->stp_state = STP_DISABLED;
1706 port->is_tunnel = false;
1710 port->realdev_ofp_port = 0;
1711 port->vlandev_vid = 0;
1712 port->carrier_seq = netdev_get_carrier_resets(netdev);
1713 port->is_layer3 = netdev_vport_is_layer3(netdev);
1715 if (netdev_vport_is_patch(netdev)) {
1716 /* By bailing out here, we don't submit the port to the sFlow module
1717 * to be considered for counter polling export. This is correct
1718 * because the patch port represents an interface that sFlow considers
1719 * to be "internal" to the switch as a whole, and therefore not an
1720 * candidate for counter polling. */
1721 port->odp_port = ODPP_NONE;
1722 ofport_update_peer(port);
1726 error = dpif_port_query_by_name(ofproto->backer->dpif,
1727 netdev_vport_get_dpif_port(netdev, namebuf,
1734 port->odp_port = dpif_port.port_no;
1736 if (netdev_get_tunnel_config(netdev)) {
1737 tnl_port_add(port, port->up.netdev, port->odp_port);
1738 port->is_tunnel = true;
1740 /* Sanity-check that a mapping doesn't already exist. This
1741 * shouldn't happen for non-tunnel ports. */
1742 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1743 VLOG_ERR("port %s already has an OpenFlow port number",
1745 dpif_port_destroy(&dpif_port);
1749 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1750 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1751 hash_odp_port(port->odp_port));
1752 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1754 dpif_port_destroy(&dpif_port);
1756 if (ofproto->sflow) {
1757 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1764 port_destruct(struct ofport *port_)
1766 struct ofport_dpif *port = ofport_dpif_cast(port_);
1767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1768 const char *devname = netdev_get_name(port->up.netdev);
1769 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1770 const char *dp_port_name;
1772 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1773 ovs_rwlock_wrlock(&xlate_rwlock);
1774 xlate_ofport_remove(port);
1775 ovs_rwlock_unlock(&xlate_rwlock);
1777 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1779 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1780 /* The underlying device is still there, so delete it. This
1781 * happens when the ofproto is being destroyed, since the caller
1782 * assumes that removal of attached ports will happen as part of
1784 if (!port->is_tunnel) {
1785 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1790 port->peer->peer = NULL;
1794 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1795 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1796 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1797 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1801 sset_find_and_delete(&ofproto->ports, devname);
1802 sset_find_and_delete(&ofproto->ghost_ports, devname);
1803 bundle_remove(port_);
1804 set_cfm(port_, NULL);
1805 set_bfd(port_, NULL);
1806 if (ofproto->sflow) {
1807 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1814 port_modified(struct ofport *port_)
1816 struct ofport_dpif *port = ofport_dpif_cast(port_);
1818 if (port->bundle && port->bundle->bond) {
1819 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1823 cfm_set_netdev(port->cfm, port->up.netdev);
1827 bfd_set_netdev(port->bfd, port->up.netdev);
1830 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1831 port->up.pp.hw_addr);
1833 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1835 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1839 ofport_update_peer(port);
1843 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1845 struct ofport_dpif *port = ofport_dpif_cast(port_);
1846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1847 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1849 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1850 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1851 OFPUTIL_PC_NO_PACKET_IN)) {
1852 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1854 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1855 bundle_update(port->bundle);
1861 set_sflow(struct ofproto *ofproto_,
1862 const struct ofproto_sflow_options *sflow_options)
1864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1865 struct dpif_sflow *ds = ofproto->sflow;
1867 if (sflow_options) {
1869 struct ofport_dpif *ofport;
1871 ds = ofproto->sflow = dpif_sflow_create();
1872 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1873 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1875 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1877 dpif_sflow_set_options(ds, sflow_options);
1880 dpif_sflow_unref(ds);
1881 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 ofproto->sflow = NULL;
1890 struct ofproto *ofproto_,
1891 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1892 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1893 size_t n_flow_exporters_options)
1895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1896 struct dpif_ipfix *di = ofproto->ipfix;
1897 bool has_options = bridge_exporter_options || flow_exporters_options;
1899 if (has_options && !di) {
1900 di = ofproto->ipfix = dpif_ipfix_create();
1904 /* Call set_options in any case to cleanly flush the flow
1905 * caches in the last exporters that are to be destroyed. */
1906 dpif_ipfix_set_options(
1907 di, bridge_exporter_options, flow_exporters_options,
1908 n_flow_exporters_options);
1911 dpif_ipfix_unref(di);
1912 ofproto->ipfix = NULL;
1920 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1922 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1927 struct ofproto_dpif *ofproto;
1929 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1930 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1931 ofport->cfm = cfm_create(ofport->up.netdev);
1934 if (cfm_configure(ofport->cfm, s)) {
1941 cfm_unref(ofport->cfm);
1944 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1945 ofport->up.pp.hw_addr);
1950 get_cfm_status(const struct ofport *ofport_,
1951 struct ofproto_cfm_status *status)
1953 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 status->faults = cfm_get_fault(ofport->cfm);
1957 status->flap_count = cfm_get_flap_count(ofport->cfm);
1958 status->remote_opstate = cfm_get_opup(ofport->cfm);
1959 status->health = cfm_get_health(ofport->cfm);
1960 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1968 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1971 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1975 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1976 cfg, ofport->up.netdev);
1977 if (ofport->bfd != old) {
1978 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1980 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1981 ofport->up.pp.hw_addr);
1986 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1988 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1991 bfd_get_status(ofport->bfd, smap);
1998 /* Spanning Tree. */
2001 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2003 struct ofproto_dpif *ofproto = ofproto_;
2004 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2005 struct ofport_dpif *ofport;
2007 ofport = stp_port_get_aux(sp);
2009 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2010 ofproto->up.name, port_num);
2012 struct eth_header *eth = pkt->l2;
2014 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2015 if (eth_addr_is_zero(eth->eth_src)) {
2016 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2017 "with unknown MAC", ofproto->up.name, port_num);
2019 ofproto_dpif_send_packet(ofport, pkt);
2025 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2027 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2031 /* Only revalidate flows if the configuration changed. */
2032 if (!s != !ofproto->stp) {
2033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2037 if (!ofproto->stp) {
2038 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2039 send_bpdu_cb, ofproto);
2040 ofproto->stp_last_tick = time_msec();
2043 stp_set_bridge_id(ofproto->stp, s->system_id);
2044 stp_set_bridge_priority(ofproto->stp, s->priority);
2045 stp_set_hello_time(ofproto->stp, s->hello_time);
2046 stp_set_max_age(ofproto->stp, s->max_age);
2047 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2049 struct ofport *ofport;
2051 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2052 set_stp_port(ofport, NULL);
2055 stp_unref(ofproto->stp);
2056 ofproto->stp = NULL;
2063 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2069 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2070 s->designated_root = stp_get_designated_root(ofproto->stp);
2071 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2080 update_stp_port_state(struct ofport_dpif *ofport)
2082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2083 enum stp_state state;
2085 /* Figure out new state. */
2086 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2090 if (ofport->stp_state != state) {
2091 enum ofputil_port_state of_state;
2094 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2095 netdev_get_name(ofport->up.netdev),
2096 stp_state_name(ofport->stp_state),
2097 stp_state_name(state));
2098 if (stp_learn_in_state(ofport->stp_state)
2099 != stp_learn_in_state(state)) {
2100 /* xxx Learning action flows should also be flushed. */
2101 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2102 mac_learning_flush(ofproto->ml);
2103 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2105 fwd_change = stp_forward_in_state(ofport->stp_state)
2106 != stp_forward_in_state(state);
2108 ofproto->backer->need_revalidate = REV_STP;
2109 ofport->stp_state = state;
2110 ofport->stp_state_entered = time_msec();
2112 if (fwd_change && ofport->bundle) {
2113 bundle_update(ofport->bundle);
2116 /* Update the STP state bits in the OpenFlow port description. */
2117 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2118 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2119 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2120 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2121 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2123 ofproto_port_set_state(&ofport->up, of_state);
2127 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2128 * caller is responsible for assigning STP port numbers and ensuring
2129 * there are no duplicates. */
2131 set_stp_port(struct ofport *ofport_,
2132 const struct ofproto_port_stp_settings *s)
2134 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2135 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2136 struct stp_port *sp = ofport->stp_port;
2138 if (!s || !s->enable) {
2140 ofport->stp_port = NULL;
2141 stp_port_disable(sp);
2142 update_stp_port_state(ofport);
2145 } else if (sp && stp_port_no(sp) != s->port_num
2146 && ofport == stp_port_get_aux(sp)) {
2147 /* The port-id changed, so disable the old one if it's not
2148 * already in use by another port. */
2149 stp_port_disable(sp);
2152 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2153 stp_port_enable(sp);
2155 stp_port_set_aux(sp, ofport);
2156 stp_port_set_priority(sp, s->priority);
2157 stp_port_set_path_cost(sp, s->path_cost);
2159 update_stp_port_state(ofport);
2165 get_stp_port_status(struct ofport *ofport_,
2166 struct ofproto_port_stp_status *s)
2168 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2170 struct stp_port *sp = ofport->stp_port;
2172 if (!ofproto->stp || !sp) {
2178 s->port_id = stp_port_get_id(sp);
2179 s->state = stp_port_get_state(sp);
2180 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2181 s->role = stp_port_get_role(sp);
2182 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2188 stp_run(struct ofproto_dpif *ofproto)
2191 long long int now = time_msec();
2192 long long int elapsed = now - ofproto->stp_last_tick;
2193 struct stp_port *sp;
2196 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2197 ofproto->stp_last_tick = now;
2199 while (stp_get_changed_port(ofproto->stp, &sp)) {
2200 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2203 update_stp_port_state(ofport);
2207 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2208 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2209 mac_learning_flush(ofproto->ml);
2210 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2216 stp_wait(struct ofproto_dpif *ofproto)
2219 poll_timer_wait(1000);
2224 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2227 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2230 if (ofport->n_qdscp != n_qdscp
2231 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2232 n_qdscp * sizeof *qdscp))) {
2233 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2234 free(ofport->qdscp);
2235 ofport->qdscp = n_qdscp
2236 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2238 ofport->n_qdscp = n_qdscp;
2246 /* Expires all MAC learning entries associated with 'bundle' and forces its
2247 * ofproto to revalidate every flow.
2249 * Normally MAC learning entries are removed only from the ofproto associated
2250 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2251 * are removed from every ofproto. When patch ports and SLB bonds are in use
2252 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2253 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2254 * with the host from which it migrated. */
2256 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2258 struct ofproto_dpif *ofproto = bundle->ofproto;
2259 struct mac_learning *ml = ofproto->ml;
2260 struct mac_entry *mac, *next_mac;
2262 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2263 ovs_rwlock_wrlock(&ml->rwlock);
2264 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2265 if (mac->port.p == bundle) {
2267 struct ofproto_dpif *o;
2269 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2271 struct mac_entry *e;
2273 ovs_rwlock_wrlock(&o->ml->rwlock);
2274 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2276 mac_learning_expire(o->ml, e);
2278 ovs_rwlock_unlock(&o->ml->rwlock);
2283 mac_learning_expire(ml, mac);
2286 ovs_rwlock_unlock(&ml->rwlock);
2289 static struct ofbundle *
2290 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2292 struct ofbundle *bundle;
2294 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2295 &ofproto->bundles) {
2296 if (bundle->aux == aux) {
2304 bundle_update(struct ofbundle *bundle)
2306 struct ofport_dpif *port;
2308 bundle->floodable = true;
2309 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2310 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2312 || !stp_forward_in_state(port->stp_state)) {
2313 bundle->floodable = false;
2320 bundle_del_port(struct ofport_dpif *port)
2322 struct ofbundle *bundle = port->bundle;
2324 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2326 list_remove(&port->bundle_node);
2327 port->bundle = NULL;
2330 lacp_slave_unregister(bundle->lacp, port);
2333 bond_slave_unregister(bundle->bond, port);
2336 bundle_update(bundle);
2340 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2341 struct lacp_slave_settings *lacp)
2343 struct ofport_dpif *port;
2345 port = get_ofp_port(bundle->ofproto, ofp_port);
2350 if (port->bundle != bundle) {
2351 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2353 bundle_remove(&port->up);
2356 port->bundle = bundle;
2357 list_push_back(&bundle->ports, &port->bundle_node);
2358 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2360 || !stp_forward_in_state(port->stp_state)) {
2361 bundle->floodable = false;
2365 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2366 lacp_slave_register(bundle->lacp, port, lacp);
2373 bundle_destroy(struct ofbundle *bundle)
2375 struct ofproto_dpif *ofproto;
2376 struct ofport_dpif *port, *next_port;
2382 ofproto = bundle->ofproto;
2383 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2385 ovs_rwlock_wrlock(&xlate_rwlock);
2386 xlate_bundle_remove(bundle);
2387 ovs_rwlock_unlock(&xlate_rwlock);
2389 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2390 bundle_del_port(port);
2393 bundle_flush_macs(bundle, true);
2394 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2396 free(bundle->trunks);
2397 lacp_unref(bundle->lacp);
2398 bond_unref(bundle->bond);
2403 bundle_set(struct ofproto *ofproto_, void *aux,
2404 const struct ofproto_bundle_settings *s)
2406 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2407 bool need_flush = false;
2408 struct ofport_dpif *port;
2409 struct ofbundle *bundle;
2410 unsigned long *trunks;
2416 bundle_destroy(bundle_lookup(ofproto, aux));
2420 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2421 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2423 bundle = bundle_lookup(ofproto, aux);
2425 bundle = xmalloc(sizeof *bundle);
2427 bundle->ofproto = ofproto;
2428 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2429 hash_pointer(aux, 0));
2431 bundle->name = NULL;
2433 list_init(&bundle->ports);
2434 bundle->vlan_mode = PORT_VLAN_TRUNK;
2436 bundle->trunks = NULL;
2437 bundle->use_priority_tags = s->use_priority_tags;
2438 bundle->lacp = NULL;
2439 bundle->bond = NULL;
2441 bundle->floodable = true;
2442 mbridge_register_bundle(ofproto->mbridge, bundle);
2445 if (!bundle->name || strcmp(s->name, bundle->name)) {
2447 bundle->name = xstrdup(s->name);
2452 if (!bundle->lacp) {
2453 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2454 bundle->lacp = lacp_create();
2456 lacp_configure(bundle->lacp, s->lacp);
2458 lacp_unref(bundle->lacp);
2459 bundle->lacp = NULL;
2462 /* Update set of ports. */
2464 for (i = 0; i < s->n_slaves; i++) {
2465 if (!bundle_add_port(bundle, s->slaves[i],
2466 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2470 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2471 struct ofport_dpif *next_port;
2473 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2474 for (i = 0; i < s->n_slaves; i++) {
2475 if (s->slaves[i] == port->up.ofp_port) {
2480 bundle_del_port(port);
2484 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2486 if (list_is_empty(&bundle->ports)) {
2487 bundle_destroy(bundle);
2491 /* Set VLAN tagging mode */
2492 if (s->vlan_mode != bundle->vlan_mode
2493 || s->use_priority_tags != bundle->use_priority_tags) {
2494 bundle->vlan_mode = s->vlan_mode;
2495 bundle->use_priority_tags = s->use_priority_tags;
2500 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2501 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2503 if (vlan != bundle->vlan) {
2504 bundle->vlan = vlan;
2508 /* Get trunked VLANs. */
2509 switch (s->vlan_mode) {
2510 case PORT_VLAN_ACCESS:
2514 case PORT_VLAN_TRUNK:
2515 trunks = CONST_CAST(unsigned long *, s->trunks);
2518 case PORT_VLAN_NATIVE_UNTAGGED:
2519 case PORT_VLAN_NATIVE_TAGGED:
2520 if (vlan != 0 && (!s->trunks
2521 || !bitmap_is_set(s->trunks, vlan)
2522 || bitmap_is_set(s->trunks, 0))) {
2523 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2525 trunks = bitmap_clone(s->trunks, 4096);
2527 trunks = bitmap_allocate1(4096);
2529 bitmap_set1(trunks, vlan);
2530 bitmap_set0(trunks, 0);
2532 trunks = CONST_CAST(unsigned long *, s->trunks);
2539 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2540 free(bundle->trunks);
2541 if (trunks == s->trunks) {
2542 bundle->trunks = vlan_bitmap_clone(trunks);
2544 bundle->trunks = trunks;
2549 if (trunks != s->trunks) {
2554 if (!list_is_short(&bundle->ports)) {
2555 bundle->ofproto->has_bonded_bundles = true;
2557 if (bond_reconfigure(bundle->bond, s->bond)) {
2558 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2561 bundle->bond = bond_create(s->bond);
2562 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2565 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2566 bond_slave_register(bundle->bond, port, port->up.netdev);
2569 bond_unref(bundle->bond);
2570 bundle->bond = NULL;
2573 /* If we changed something that would affect MAC learning, un-learn
2574 * everything on this port and force flow revalidation. */
2576 bundle_flush_macs(bundle, false);
2583 bundle_remove(struct ofport *port_)
2585 struct ofport_dpif *port = ofport_dpif_cast(port_);
2586 struct ofbundle *bundle = port->bundle;
2589 bundle_del_port(port);
2590 if (list_is_empty(&bundle->ports)) {
2591 bundle_destroy(bundle);
2592 } else if (list_is_short(&bundle->ports)) {
2593 bond_unref(bundle->bond);
2594 bundle->bond = NULL;
2600 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2602 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2603 struct ofport_dpif *port = port_;
2604 uint8_t ea[ETH_ADDR_LEN];
2607 error = netdev_get_etheraddr(port->up.netdev, ea);
2609 struct ofpbuf packet;
2612 ofpbuf_init(&packet, 0);
2613 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2615 memcpy(packet_pdu, pdu, pdu_size);
2617 ofproto_dpif_send_packet(port, &packet);
2618 ofpbuf_uninit(&packet);
2620 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2621 "%s (%s)", port->bundle->name,
2622 netdev_get_name(port->up.netdev), ovs_strerror(error));
2627 bundle_send_learning_packets(struct ofbundle *bundle)
2629 struct ofproto_dpif *ofproto = bundle->ofproto;
2630 struct ofpbuf *learning_packet;
2631 int error, n_packets, n_errors;
2632 struct mac_entry *e;
2633 struct list packets;
2635 list_init(&packets);
2636 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2637 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2638 if (e->port.p != bundle) {
2641 learning_packet = bond_compose_learning_packet(bundle->bond,
2644 learning_packet->private_p = port_void;
2645 list_push_back(&packets, &learning_packet->list_node);
2648 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2650 error = n_packets = n_errors = 0;
2651 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2654 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2661 ofpbuf_list_delete(&packets);
2664 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2665 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2666 "packets, last error was: %s",
2667 bundle->name, n_errors, n_packets, ovs_strerror(error));
2669 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2670 bundle->name, n_packets);
2675 bundle_run(struct ofbundle *bundle)
2678 lacp_run(bundle->lacp, send_pdu_cb);
2681 struct ofport_dpif *port;
2683 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2684 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2687 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2688 bundle->ofproto->backer->need_revalidate = REV_BOND;
2691 if (bond_should_send_learning_packets(bundle->bond)) {
2692 bundle_send_learning_packets(bundle);
2698 bundle_wait(struct ofbundle *bundle)
2701 lacp_wait(bundle->lacp);
2704 bond_wait(bundle->bond);
2711 mirror_set__(struct ofproto *ofproto_, void *aux,
2712 const struct ofproto_mirror_settings *s)
2714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2715 struct ofbundle **srcs, **dsts;
2720 mirror_destroy(ofproto->mbridge, aux);
2724 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2725 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2727 for (i = 0; i < s->n_srcs; i++) {
2728 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2731 for (i = 0; i < s->n_dsts; i++) {
2732 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2735 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2736 s->n_dsts, s->src_vlans,
2737 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2744 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2745 uint64_t *packets, uint64_t *bytes)
2748 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2753 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2756 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2757 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2758 mac_learning_flush(ofproto->ml);
2760 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2765 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2768 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2769 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2773 forward_bpdu_changed(struct ofproto *ofproto_)
2775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2776 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2780 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2783 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2784 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2785 mac_learning_set_idle_time(ofproto->ml, idle_time);
2786 mac_learning_set_max_entries(ofproto->ml, max_entries);
2787 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2792 static struct ofport_dpif *
2793 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2795 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2796 return ofport ? ofport_dpif_cast(ofport) : NULL;
2799 static struct ofport_dpif *
2800 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2802 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2803 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2807 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2808 struct ofproto_port *ofproto_port,
2809 struct dpif_port *dpif_port)
2811 ofproto_port->name = dpif_port->name;
2812 ofproto_port->type = dpif_port->type;
2813 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2817 ofport_update_peer(struct ofport_dpif *ofport)
2819 const struct ofproto_dpif *ofproto;
2820 struct dpif_backer *backer;
2823 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2827 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2828 backer->need_revalidate = REV_RECONFIGURE;
2831 ofport->peer->peer = NULL;
2832 ofport->peer = NULL;
2835 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2840 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2841 struct ofport *peer_ofport;
2842 struct ofport_dpif *peer;
2845 if (ofproto->backer != backer) {
2849 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2854 peer = ofport_dpif_cast(peer_ofport);
2855 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2856 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2858 ofport->peer = peer;
2859 ofport->peer->peer = ofport;
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;
2880 int cfm_opup = cfm_get_opup(ofport->cfm);
2882 cfm_enable = !cfm_get_fault(ofport->cfm);
2884 if (cfm_opup >= 0) {
2885 cfm_enable = cfm_enable && cfm_opup;
2890 bfd_enable = bfd_forwarding(ofport->bfd);
2893 if (ofport->bfd || ofport->cfm) {
2894 enable = enable && (cfm_enable || bfd_enable);
2897 if (ofport->bundle) {
2898 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2899 if (carrier_changed) {
2900 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2904 if (ofport->may_enable != enable) {
2905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2906 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2909 ofport->may_enable = enable;
2913 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2914 struct ofproto_port *ofproto_port)
2916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2917 struct dpif_port dpif_port;
2920 if (sset_contains(&ofproto->ghost_ports, devname)) {
2921 const char *type = netdev_get_type_from_name(devname);
2923 /* We may be called before ofproto->up.port_by_name is populated with
2924 * the appropriate ofport. For this reason, we must get the name and
2925 * type from the netdev layer directly. */
2927 const struct ofport *ofport;
2929 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2930 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2931 ofproto_port->name = xstrdup(devname);
2932 ofproto_port->type = xstrdup(type);
2938 if (!sset_contains(&ofproto->ports, devname)) {
2941 error = dpif_port_query_by_name(ofproto->backer->dpif,
2942 devname, &dpif_port);
2944 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2950 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2953 const char *devname = netdev_get_name(netdev);
2954 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2955 const char *dp_port_name;
2957 if (netdev_vport_is_patch(netdev)) {
2958 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2962 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2963 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2964 odp_port_t port_no = ODPP_NONE;
2967 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2971 if (netdev_get_tunnel_config(netdev)) {
2972 simap_put(&ofproto->backer->tnl_backers,
2973 dp_port_name, odp_to_u32(port_no));
2977 if (netdev_get_tunnel_config(netdev)) {
2978 sset_add(&ofproto->ghost_ports, devname);
2980 sset_add(&ofproto->ports, devname);
2986 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2989 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2996 sset_find_and_delete(&ofproto->ghost_ports,
2997 netdev_get_name(ofport->up.netdev));
2998 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2999 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3000 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3002 /* The caller is going to close ofport->up.netdev. If this is a
3003 * bonded port, then the bond is using that netdev, so remove it
3004 * from the bond. The client will need to reconfigure everything
3005 * after deleting ports, so then the slave will get re-added. */
3006 bundle_remove(&ofport->up);
3013 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3015 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3020 error = netdev_get_stats(ofport->up.netdev, stats);
3022 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3025 ovs_mutex_lock(&ofproto->stats_mutex);
3026 /* ofproto->stats.tx_packets represents packets that we created
3027 * internally and sent to some port (e.g. packets sent with
3028 * ofproto_dpif_send_packet()). Account for them as if they had
3029 * come from OFPP_LOCAL and got forwarded. */
3031 if (stats->rx_packets != UINT64_MAX) {
3032 stats->rx_packets += ofproto->stats.tx_packets;
3035 if (stats->rx_bytes != UINT64_MAX) {
3036 stats->rx_bytes += ofproto->stats.tx_bytes;
3039 /* ofproto->stats.rx_packets represents packets that were received on
3040 * some port and we processed internally and dropped (e.g. STP).
3041 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3043 if (stats->tx_packets != UINT64_MAX) {
3044 stats->tx_packets += ofproto->stats.rx_packets;
3047 if (stats->tx_bytes != UINT64_MAX) {
3048 stats->tx_bytes += ofproto->stats.rx_bytes;
3050 ovs_mutex_unlock(&ofproto->stats_mutex);
3056 struct port_dump_state {
3061 struct ofproto_port port;
3066 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3068 *statep = xzalloc(sizeof(struct port_dump_state));
3073 port_dump_next(const struct ofproto *ofproto_, void *state_,
3074 struct ofproto_port *port)
3076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3077 struct port_dump_state *state = state_;
3078 const struct sset *sset;
3079 struct sset_node *node;
3081 if (state->has_port) {
3082 ofproto_port_destroy(&state->port);
3083 state->has_port = false;
3085 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3086 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3089 error = port_query_by_name(ofproto_, node->name, &state->port);
3091 *port = state->port;
3092 state->has_port = true;
3094 } else if (error != ENODEV) {
3099 if (!state->ghost) {
3100 state->ghost = true;
3103 return port_dump_next(ofproto_, state_, port);
3110 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3112 struct port_dump_state *state = state_;
3114 if (state->has_port) {
3115 ofproto_port_destroy(&state->port);
3122 port_poll(const struct ofproto *ofproto_, char **devnamep)
3124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3126 if (ofproto->port_poll_errno) {
3127 int error = ofproto->port_poll_errno;
3128 ofproto->port_poll_errno = 0;
3132 if (sset_is_empty(&ofproto->port_poll_set)) {
3136 *devnamep = sset_pop(&ofproto->port_poll_set);
3141 port_poll_wait(const struct ofproto *ofproto_)
3143 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3144 dpif_port_poll_wait(ofproto->backer->dpif);
3148 port_is_lacp_current(const struct ofport *ofport_)
3150 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3151 return (ofport->bundle && ofport->bundle->lacp
3152 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3156 /* Upcall handling. */
3158 struct flow_miss_op {
3159 struct dpif_op dpif_op;
3161 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3162 struct xlate_out xout;
3163 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3165 struct ofpbuf mask; /* Flow mask for "put" ops. */
3166 struct odputil_keybuf maskbuf;
3168 /* If this is a "put" op, then a pointer to the subfacet that should
3169 * be marked as uninstalled if the operation fails. */
3170 struct subfacet *subfacet;
3173 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3174 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3175 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3176 * return value of true). However, for short flows the cost of bookkeeping is
3177 * much higher than the benefits, so when the datapath holds a large number of
3178 * flows we impose some heuristics to decide which flows are likely to be worth
3181 flow_miss_should_make_facet(struct flow_miss *miss)
3183 struct dpif_backer *backer = miss->ofproto->backer;
3186 switch (flow_miss_model) {
3187 case OFPROTO_HANDLE_MISS_AUTO:
3189 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3191 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3195 if (!backer->governor) {
3198 n_subfacets = hmap_count(&backer->subfacets);
3199 if (n_subfacets * 2 <= flow_eviction_threshold) {
3203 backer->governor = governor_create();
3206 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3207 return governor_should_install_flow(backer->governor, hash,
3208 miss->stats.n_packets);
3211 /* Handles 'miss', which matches 'facet'. May add any required datapath
3212 * operations to 'ops', incrementing '*n_ops' for each new op.
3214 * All of the packets in 'miss' are considered to have arrived at time
3215 * 'miss->stats.used'. This is really important only for new facets: if we
3216 * just called time_msec() here, then the new subfacet or its packets could
3217 * look (occasionally) as though it was used some time after the facet was
3218 * used. That can make a one-packet flow look like it has a nonzero duration,
3219 * which looks odd in e.g. NetFlow statistics. */
3221 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3222 struct flow_miss_op *ops, size_t *n_ops)
3224 enum subfacet_path want_path;
3225 struct subfacet *subfacet;
3228 /* Update facet stats. */
3229 facet->packet_count += miss->stats.n_packets;
3230 facet->prev_packet_count += miss->stats.n_packets;
3231 facet->byte_count += miss->stats.n_bytes;
3232 facet->prev_byte_count += miss->stats.n_bytes;
3234 /* Look for an existing subfacet. If we find one, update its used time. */
3235 key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3236 if (!list_is_empty(&facet->subfacets)) {
3237 subfacet = subfacet_find(miss->ofproto->backer,
3238 miss->key, miss->key_len, key_hash);
3240 if (subfacet->facet == facet) {
3241 subfacet->used = MAX(subfacet->used, miss->stats.used);
3243 /* This shouldn't happen. */
3244 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3245 subfacet_destroy(subfacet);
3253 /* Don't install the flow if it's the result of the "userspace"
3254 * action for an already installed facet. This can occur when a
3255 * datapath flow with wildcards has a "userspace" action and flows
3256 * sent to userspace result in a different subfacet, which will then
3257 * be rejected as overlapping by the datapath. */
3258 if (miss->upcall_type == DPIF_UC_ACTION
3259 && !list_is_empty(&facet->subfacets)) {
3263 /* Create a subfacet, if we don't already have one. */
3265 subfacet = subfacet_create(facet, miss, key_hash);
3268 /* Install the subfacet, if it's not already installed. */
3269 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3270 if (subfacet->path != want_path) {
3271 struct flow_miss_op *op = &ops[(*n_ops)++];
3272 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3274 subfacet->path = want_path;
3276 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3277 if (enable_megaflows) {
3278 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3279 &miss->flow, UINT32_MAX);
3282 op->xout_garbage = false;
3283 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3284 op->subfacet = subfacet;
3285 put->flags = DPIF_FP_CREATE;
3286 put->key = miss->key;
3287 put->key_len = miss->key_len;
3288 put->mask = op->mask.data;
3289 put->mask_len = op->mask.size;
3291 if (want_path == SF_FAST_PATH) {
3292 put->actions = facet->xout.odp_actions.data;
3293 put->actions_len = facet->xout.odp_actions.size;
3295 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3296 op->slow_stub, sizeof op->slow_stub,
3297 &put->actions, &put->actions_len);
3303 /* Handles flow miss 'miss'. May add any required datapath operations
3304 * to 'ops', incrementing '*n_ops' for each new op. */
3306 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3309 struct facet *facet;
3311 miss->ofproto->n_missed += miss->stats.n_packets;
3313 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3315 /* There does not exist a bijection between 'struct flow' and datapath
3316 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3317 * assumption used throughout the facet and subfacet handling code.
3318 * Since we have to handle these misses in userspace anyway, we simply
3319 * skip facet creation, avoiding the problem altogether. */
3320 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3321 || !flow_miss_should_make_facet(miss)) {
3325 facet = facet_create(miss);
3327 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3330 static struct drop_key *
3331 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3334 struct drop_key *drop_key;
3336 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3337 &backer->drop_keys) {
3338 if (drop_key->key_len == key_len
3339 && !memcmp(drop_key->key, key, key_len)) {
3347 drop_key_clear(struct dpif_backer *backer)
3349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3350 struct drop_key *drop_key, *next;
3352 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3355 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3357 if (error && !VLOG_DROP_WARN(&rl)) {
3358 struct ds ds = DS_EMPTY_INITIALIZER;
3359 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3360 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3361 ovs_strerror(error), ds_cstr(&ds));
3365 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3366 drop_key_destroy(drop_key);
3369 udpif_drop_key_clear(backer->udpif);
3373 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3375 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3376 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3377 struct flow_miss *miss;
3380 /* Process each element in the to-do list, constructing the set of
3381 * operations to batch. */
3383 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3384 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3386 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3388 /* Execute batch. */
3389 for (i = 0; i < n_ops; i++) {
3390 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3392 dpif_operate(backer->dpif, dpif_ops, n_ops);
3394 for (i = 0; i < n_ops; i++) {
3395 if (dpif_ops[i]->error != 0
3396 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3397 && flow_miss_ops[i].subfacet) {
3398 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3400 COVERAGE_INC(subfacet_install_fail);
3402 /* Zero-out subfacet counters when installation failed, but
3403 * datapath reported hits. This should not happen and
3404 * indicates a bug, since if the datapath flow exists, we
3405 * should not be attempting to create a new subfacet. A
3406 * buggy datapath could trigger this, so just zero out the
3407 * counters and log an error. */
3408 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3409 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3410 "datapath reported hits");
3411 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3414 subfacet->path = SF_NOT_INSTALLED;
3420 handle_upcalls(struct dpif_backer *backer)
3422 struct flow_miss_batch *fmb;
3425 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3426 struct drop_key *drop_key = drop_key_next(backer->udpif);
3431 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3432 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3433 hash_bytes(drop_key->key, drop_key->key_len, 0));
3434 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3435 drop_key->key, drop_key->key_len,
3436 NULL, 0, NULL, 0, NULL);
3438 drop_key_destroy(drop_key);
3442 fmb = flow_miss_batch_next(backer->udpif);
3444 handle_flow_misses(backer, fmb);
3445 flow_miss_batch_destroy(fmb);
3449 /* Flow expiration. */
3451 static int subfacet_max_idle(const struct dpif_backer *);
3452 static void update_stats(struct dpif_backer *);
3453 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3454 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3456 /* This function is called periodically by run(). Its job is to collect
3457 * updates for the flows that have been installed into the datapath, most
3458 * importantly when they last were used, and then use that information to
3459 * expire flows that have not been used recently.
3461 * Returns the number of milliseconds after which it should be called again. */
3463 expire(struct dpif_backer *backer)
3465 struct ofproto_dpif *ofproto;
3469 /* Periodically clear out the drop keys in an effort to keep them
3470 * relatively few. */
3471 drop_key_clear(backer);
3473 /* Update stats for each flow in the backer. */
3474 update_stats(backer);
3476 n_subfacets = hmap_count(&backer->subfacets);
3478 struct subfacet *subfacet;
3479 long long int total, now;
3483 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3484 total += now - subfacet->created;
3486 backer->avg_subfacet_life += total / n_subfacets;
3488 backer->avg_subfacet_life /= 2;
3490 backer->avg_n_subfacet += n_subfacets;
3491 backer->avg_n_subfacet /= 2;
3493 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3495 max_idle = subfacet_max_idle(backer);
3496 expire_subfacets(backer, max_idle);
3498 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3499 struct rule *rule, *next_rule;
3501 if (ofproto->backer != backer) {
3505 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3507 ovs_mutex_lock(&ofproto_mutex);
3508 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3509 &ofproto->up.expirable) {
3510 rule_expire(rule_dpif_cast(rule));
3512 ovs_mutex_unlock(&ofproto_mutex);
3514 /* All outstanding data in existing flows has been accounted, so it's a
3515 * good time to do bond rebalancing. */
3516 if (ofproto->has_bonded_bundles) {
3517 struct ofbundle *bundle;
3519 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3521 bond_rebalance(bundle->bond);
3527 return MIN(max_idle, 1000);
3530 /* Updates flow table statistics given that the datapath just reported 'stats'
3531 * as 'subfacet''s statistics. */
3533 update_subfacet_stats(struct subfacet *subfacet,
3534 const struct dpif_flow_stats *stats)
3536 struct facet *facet = subfacet->facet;
3537 struct dpif_flow_stats diff;
3539 diff.tcp_flags = stats->tcp_flags;
3540 diff.used = stats->used;
3542 if (stats->n_packets >= subfacet->dp_packet_count) {
3543 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3545 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3549 if (stats->n_bytes >= subfacet->dp_byte_count) {
3550 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3552 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3556 facet->ofproto->n_hit += diff.n_packets;
3557 subfacet->dp_packet_count = stats->n_packets;
3558 subfacet->dp_byte_count = stats->n_bytes;
3559 subfacet_update_stats(subfacet, &diff);
3561 if (facet->accounted_bytes < facet->byte_count) {
3563 facet_account(facet);
3564 facet->accounted_bytes = facet->byte_count;
3568 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3569 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3571 delete_unexpected_flow(struct dpif_backer *backer,
3572 const struct nlattr *key, size_t key_len)
3574 if (!VLOG_DROP_WARN(&rl)) {
3578 odp_flow_key_format(key, key_len, &s);
3579 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3583 COVERAGE_INC(facet_unexpected);
3584 dpif_flow_del(backer->dpif, key, key_len, NULL);
3587 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3589 * This function also pushes statistics updates to rules which each facet
3590 * resubmits into. Generally these statistics will be accurate. However, if a
3591 * facet changes the rule it resubmits into at some time in between
3592 * update_stats() runs, it is possible that statistics accrued to the
3593 * old rule will be incorrectly attributed to the new rule. This could be
3594 * avoided by calling update_stats() whenever rules are created or
3595 * deleted. However, the performance impact of making so many calls to the
3596 * datapath do not justify the benefit of having perfectly accurate statistics.
3598 * In addition, this function maintains per ofproto flow hit counts. The patch
3599 * port is not treated specially. e.g. A packet ingress from br0 patched into
3600 * br1 will increase the hit count of br0 by 1, however, does not affect
3601 * the hit or miss counts of br1.
3604 update_stats(struct dpif_backer *backer)
3606 const struct dpif_flow_stats *stats;
3607 struct dpif_flow_dump dump;
3608 const struct nlattr *key, *mask;
3609 size_t key_len, mask_len;
3611 dpif_flow_dump_start(&dump, backer->dpif);
3612 while (dpif_flow_dump_next(&dump, &key, &key_len,
3613 &mask, &mask_len, NULL, NULL, &stats)) {
3614 struct subfacet *subfacet;
3617 key_hash = odp_flow_key_hash(key, key_len);
3618 subfacet = subfacet_find(backer, key, key_len, key_hash);
3619 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3621 update_subfacet_stats(subfacet, stats);
3625 /* Stats are updated per-packet. */
3628 case SF_NOT_INSTALLED:
3630 delete_unexpected_flow(backer, key, key_len);
3635 dpif_flow_dump_done(&dump);
3638 /* Calculates and returns the number of milliseconds of idle time after which
3639 * subfacets should expire from the datapath. When a subfacet expires, we fold
3640 * its statistics into its facet, and when a facet's last subfacet expires, we
3641 * fold its statistic into its rule. */
3643 subfacet_max_idle(const struct dpif_backer *backer)
3646 * Idle time histogram.
3648 * Most of the time a switch has a relatively small number of subfacets.
3649 * When this is the case we might as well keep statistics for all of them
3650 * in userspace and to cache them in the kernel datapath for performance as
3653 * As the number of subfacets increases, the memory required to maintain
3654 * statistics about them in userspace and in the kernel becomes
3655 * significant. However, with a large number of subfacets it is likely
3656 * that only a few of them are "heavy hitters" that consume a large amount
3657 * of bandwidth. At this point, only heavy hitters are worth caching in
3658 * the kernel and maintaining in userspaces; other subfacets we can
3661 * The technique used to compute the idle time is to build a histogram with
3662 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3663 * that is installed in the kernel gets dropped in the appropriate bucket.
3664 * After the histogram has been built, we compute the cutoff so that only
3665 * the most-recently-used 1% of subfacets (but at least
3666 * flow_eviction_threshold flows) are kept cached. At least
3667 * the most-recently-used bucket of subfacets is kept, so actually an
3668 * arbitrary number of subfacets can be kept in any given expiration run
3669 * (though the next run will delete most of those unless they receive
3672 * This requires a second pass through the subfacets, in addition to the
3673 * pass made by update_stats(), because the former function never looks at
3674 * uninstallable subfacets.
3676 enum { BUCKET_WIDTH = 100 };
3677 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3678 int buckets[N_BUCKETS] = { 0 };
3679 int total, subtotal, bucket;
3680 struct subfacet *subfacet;
3684 total = hmap_count(&backer->subfacets);
3685 if (total <= flow_eviction_threshold) {
3686 return N_BUCKETS * BUCKET_WIDTH;
3689 /* Build histogram. */
3691 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3692 long long int idle = now - subfacet->used;
3693 int bucket = (idle <= 0 ? 0
3694 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3695 : (unsigned int) idle / BUCKET_WIDTH);
3699 /* Find the first bucket whose flows should be expired. */
3700 subtotal = bucket = 0;
3702 subtotal += buckets[bucket++];
3703 } while (bucket < N_BUCKETS &&
3704 subtotal < MAX(flow_eviction_threshold, total / 100));
3706 if (VLOG_IS_DBG_ENABLED()) {
3710 ds_put_cstr(&s, "keep");
3711 for (i = 0; i < N_BUCKETS; i++) {
3713 ds_put_cstr(&s, ", drop");
3716 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3719 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3723 return bucket * BUCKET_WIDTH;
3727 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3729 /* Cutoff time for most flows. */
3730 long long int normal_cutoff = time_msec() - dp_max_idle;
3732 /* We really want to keep flows for special protocols around, so use a more
3733 * conservative cutoff. */
3734 long long int special_cutoff = time_msec() - 10000;
3736 struct subfacet *subfacet, *next_subfacet;
3737 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3741 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3742 &backer->subfacets) {
3743 long long int cutoff;
3745 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3749 if (subfacet->used < cutoff) {
3750 if (subfacet->path != SF_NOT_INSTALLED) {
3751 batch[n_batch++] = subfacet;
3752 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3753 subfacet_destroy_batch(backer, batch, n_batch);
3757 subfacet_destroy(subfacet);
3763 subfacet_destroy_batch(backer, batch, n_batch);
3767 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3768 * then delete it entirely. */
3770 rule_expire(struct rule_dpif *rule)
3771 OVS_REQUIRES(ofproto_mutex)
3773 uint16_t idle_timeout, hard_timeout;
3774 long long int now = time_msec();
3777 ovs_assert(!rule->up.pending);
3779 /* Has 'rule' expired? */
3780 ovs_mutex_lock(&rule->up.mutex);
3781 hard_timeout = rule->up.hard_timeout;
3782 idle_timeout = rule->up.idle_timeout;
3783 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3784 reason = OFPRR_HARD_TIMEOUT;
3785 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3786 reason = OFPRR_IDLE_TIMEOUT;
3790 ovs_mutex_unlock(&rule->up.mutex);
3793 COVERAGE_INC(ofproto_dpif_expired);
3794 ofproto_rule_expire(&rule->up, reason);
3800 /* Creates and returns a new facet based on 'miss'.
3802 * The caller must already have determined that no facet with an identical
3803 * 'miss->flow' exists in 'miss->ofproto'.
3805 * 'rule' and 'xout' must have been created based on 'miss'.
3807 * 'facet'' statistics are initialized based on 'stats'.
3809 * The facet will initially have no subfacets. The caller should create (at
3810 * least) one subfacet with subfacet_create(). */
3811 static struct facet *
3812 facet_create(const struct flow_miss *miss)
3814 struct ofproto_dpif *ofproto = miss->ofproto;
3815 struct facet *facet;
3818 COVERAGE_INC(facet_create);
3819 facet = xzalloc(sizeof *facet);
3820 facet->ofproto = miss->ofproto;
3821 facet->used = miss->stats.used;
3822 facet->flow = miss->flow;
3823 facet->learn_rl = time_msec() + 500;
3825 list_init(&facet->subfacets);
3826 netflow_flow_init(&facet->nf_flow);
3827 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3829 xlate_out_copy(&facet->xout, &miss->xout);
3831 match_init(&match, &facet->flow, &facet->xout.wc);
3832 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3833 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3834 classifier_insert(&ofproto->facets, &facet->cr);
3835 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3837 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3842 facet_free(struct facet *facet)
3845 xlate_out_uninit(&facet->xout);
3850 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3851 * 'flow' must reflect the data in 'packet'. */
3853 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3854 const struct flow *flow,
3855 struct rule_dpif *rule,
3856 const struct ofpact *ofpacts, size_t ofpacts_len,
3857 struct ofpbuf *packet)
3859 struct odputil_keybuf keybuf;
3860 struct dpif_flow_stats stats;
3861 struct xlate_out xout;
3862 struct xlate_in xin;
3867 ovs_assert((rule != NULL) != (ofpacts != NULL));
3869 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3871 rule_dpif_credit_stats(rule, &stats);
3874 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3875 xin.ofpacts = ofpacts;
3876 xin.ofpacts_len = ofpacts_len;
3877 xin.resubmit_stats = &stats;
3878 xlate_actions(&xin, &xout);
3880 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3881 in_port = flow->in_port.ofp_port;
3882 if (in_port == OFPP_NONE) {
3883 in_port = OFPP_LOCAL;
3885 odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3887 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3888 xout.odp_actions.data, xout.odp_actions.size, packet,
3889 (xout.slow & SLOW_ACTION) != 0);
3890 xlate_out_uninit(&xout);
3895 /* Remove 'facet' from its ofproto and free up the associated memory:
3897 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3898 * rule's statistics, via subfacet_uninstall().
3900 * - Removes 'facet' from its rule and from ofproto->facets.
3903 facet_remove(struct facet *facet)
3905 struct subfacet *subfacet, *next_subfacet;
3907 COVERAGE_INC(facet_remove);
3908 ovs_assert(!list_is_empty(&facet->subfacets));
3910 /* First uninstall all of the subfacets to get final statistics. */
3911 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3912 subfacet_uninstall(subfacet);
3915 /* Flush the final stats to the rule.
3917 * This might require us to have at least one subfacet around so that we
3918 * can use its actions for accounting in facet_account(), which is why we
3919 * have uninstalled but not yet destroyed the subfacets. */
3920 facet_flush_stats(facet);
3922 /* Now we're really all done so destroy everything. */
3923 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3924 &facet->subfacets) {
3925 subfacet_destroy__(subfacet);
3927 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3928 classifier_remove(&facet->ofproto->facets, &facet->cr);
3929 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3930 cls_rule_destroy(&facet->cr);
3934 /* Feed information from 'facet' back into the learning table to keep it in
3935 * sync with what is actually flowing through the datapath. */
3937 facet_learn(struct facet *facet)
3939 long long int now = time_msec();
3941 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3945 facet->learn_rl = now + 500;
3947 if (!facet->xout.has_learn
3948 && !facet->xout.has_normal
3949 && (!facet->xout.has_fin_timeout
3950 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3954 facet_push_stats(facet, true);
3958 facet_account(struct facet *facet)
3960 const struct nlattr *a;
3965 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3968 n_bytes = facet->byte_count - facet->accounted_bytes;
3970 /* This loop feeds byte counters to bond_account() for rebalancing to use
3971 * as a basis. We also need to track the actual VLAN on which the packet
3972 * is going to be sent to ensure that it matches the one passed to
3973 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3976 * We use the actions from an arbitrary subfacet because they should all
3977 * be equally valid for our purpose. */
3978 vlan_tci = facet->flow.vlan_tci;
3979 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3980 facet->xout.odp_actions.size) {
3981 const struct ovs_action_push_vlan *vlan;
3982 struct ofport_dpif *port;
3984 switch (nl_attr_type(a)) {
3985 case OVS_ACTION_ATTR_OUTPUT:
3986 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3987 if (port && port->bundle && port->bundle->bond) {
3988 bond_account(port->bundle->bond, &facet->flow,
3989 vlan_tci_to_vid(vlan_tci), n_bytes);
3993 case OVS_ACTION_ATTR_POP_VLAN:
3994 vlan_tci = htons(0);
3997 case OVS_ACTION_ATTR_PUSH_VLAN:
3998 vlan = nl_attr_get(a);
3999 vlan_tci = vlan->vlan_tci;
4005 /* Returns true if the only action for 'facet' is to send to the controller.
4006 * (We don't report NetFlow expiration messages for such facets because they
4007 * are just part of the control logic for the network, not real traffic). */
4009 facet_is_controller_flow(struct facet *facet)
4012 struct ofproto_dpif *ofproto = facet->ofproto;
4013 const struct ofpact *ofpacts;
4014 struct rule_actions *actions;
4015 struct rule_dpif *rule;
4019 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4020 actions = rule_dpif_get_actions(rule);
4021 rule_dpif_unref(rule);
4023 ofpacts_len = actions->ofpacts_len;
4024 ofpacts = actions->ofpacts;
4025 is_controller = ofpacts_len > 0
4026 && ofpacts->type == OFPACT_CONTROLLER
4027 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4028 rule_actions_unref(actions);
4030 return is_controller;
4035 /* Folds all of 'facet''s statistics into its rule. Also updates the
4036 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4037 * 'facet''s statistics in the datapath should have been zeroed and folded into
4038 * its packet and byte counts before this function is called. */
4040 facet_flush_stats(struct facet *facet)
4042 struct ofproto_dpif *ofproto = facet->ofproto;
4043 struct subfacet *subfacet;
4045 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4046 ovs_assert(!subfacet->dp_byte_count);
4047 ovs_assert(!subfacet->dp_packet_count);
4050 facet_push_stats(facet, false);
4051 if (facet->accounted_bytes < facet->byte_count) {
4052 facet_account(facet);
4053 facet->accounted_bytes = facet->byte_count;
4056 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4057 struct ofexpired expired;
4058 expired.flow = facet->flow;
4059 expired.packet_count = facet->packet_count;
4060 expired.byte_count = facet->byte_count;
4061 expired.used = facet->used;
4062 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4065 /* Reset counters to prevent double counting if 'facet' ever gets
4067 facet_reset_counters(facet);
4069 netflow_flow_clear(&facet->nf_flow);
4070 facet->tcp_flags = 0;
4073 /* Searches 'ofproto''s table of facets for one which would be responsible for
4074 * 'flow'. Returns it if found, otherwise a null pointer.
4076 * The returned facet might need revalidation; use facet_lookup_valid()
4077 * instead if that is important. */
4078 static struct facet *
4079 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4081 struct cls_rule *cr;
4083 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4084 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4085 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4086 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4089 /* Searches 'ofproto''s table of facets for one capable that covers
4090 * 'flow'. Returns it if found, otherwise a null pointer.
4092 * The returned facet is guaranteed to be valid. */
4093 static struct facet *
4094 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4096 struct facet *facet;
4098 facet = facet_find(ofproto, flow);
4100 && ofproto->backer->need_revalidate
4101 && !facet_revalidate(facet)) {
4109 facet_check_consistency(struct facet *facet)
4111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4113 struct xlate_out xout;
4114 struct xlate_in xin;
4117 /* Check the datapath actions for consistency. */
4118 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4119 xlate_actions(&xin, &xout);
4121 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4122 && facet->xout.slow == xout.slow;
4123 if (!ok && !VLOG_DROP_WARN(&rl)) {
4124 struct ds s = DS_EMPTY_INITIALIZER;
4126 flow_format(&s, &facet->flow);
4127 ds_put_cstr(&s, ": inconsistency in facet");
4129 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4130 ds_put_cstr(&s, " (actions were: ");
4131 format_odp_actions(&s, facet->xout.odp_actions.data,
4132 facet->xout.odp_actions.size);
4133 ds_put_cstr(&s, ") (correct actions: ");
4134 format_odp_actions(&s, xout.odp_actions.data,
4135 xout.odp_actions.size);
4136 ds_put_char(&s, ')');
4139 if (facet->xout.slow != xout.slow) {
4140 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4145 xlate_out_uninit(&xout);
4150 /* Re-searches the classifier for 'facet':
4152 * - If the rule found is different from 'facet''s current rule, moves
4153 * 'facet' to the new rule and recompiles its actions.
4155 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4156 * where it is and recompiles its actions anyway.
4158 * - If any of 'facet''s subfacets correspond to a new flow according to
4159 * xlate_receive(), 'facet' is removed.
4161 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4163 facet_revalidate(struct facet *facet)
4165 struct ofproto_dpif *ofproto = facet->ofproto;
4166 struct rule_dpif *new_rule;
4167 struct subfacet *subfacet;
4168 struct flow_wildcards wc;
4169 struct xlate_out xout;
4170 struct xlate_in xin;
4172 COVERAGE_INC(facet_revalidate);
4174 /* Check that child subfacets still correspond to this facet. Tunnel
4175 * configuration changes could cause a subfacet's OpenFlow in_port to
4177 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4178 struct ofproto_dpif *recv_ofproto;
4179 struct flow recv_flow;
4182 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4183 subfacet->key_len, &recv_flow, NULL,
4184 &recv_ofproto, NULL);
4186 || recv_ofproto != ofproto
4187 || facet != facet_find(ofproto, &recv_flow)) {
4188 facet_remove(facet);
4193 flow_wildcards_init_catchall(&wc);
4194 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4196 /* Calculate new datapath actions.
4198 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4199 * emit a NetFlow expiration and, if so, we need to have the old state
4200 * around to properly compose it. */
4201 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4202 xlate_actions(&xin, &xout);
4203 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4205 /* A facet's slow path reason should only change under dramatic
4206 * circumstances. Rather than try to update everything, it's simpler to
4207 * remove the facet and start over.
4209 * More importantly, if a facet's wildcards change, it will be relatively
4210 * difficult to figure out if its subfacets still belong to it, and if not
4211 * which facet they may belong to. Again, to avoid the complexity, we
4212 * simply give up instead. */
4213 if (facet->xout.slow != xout.slow
4214 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4215 facet_remove(facet);
4216 xlate_out_uninit(&xout);
4217 rule_dpif_unref(new_rule);
4221 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4222 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4223 if (subfacet->path == SF_FAST_PATH) {
4224 struct dpif_flow_stats stats;
4226 subfacet_install(subfacet, &xout.odp_actions, &stats);
4227 subfacet_update_stats(subfacet, &stats);
4231 facet_flush_stats(facet);
4233 ofpbuf_clear(&facet->xout.odp_actions);
4234 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4235 xout.odp_actions.size);
4238 /* Update 'facet' now that we've taken care of all the old state. */
4239 facet->xout.slow = xout.slow;
4240 facet->xout.has_learn = xout.has_learn;
4241 facet->xout.has_normal = xout.has_normal;
4242 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4243 facet->xout.nf_output_iface = xout.nf_output_iface;
4244 facet->xout.mirrors = xout.mirrors;
4245 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4247 ovs_mutex_lock(&new_rule->up.mutex);
4248 facet->used = MAX(facet->used, new_rule->up.created);
4249 ovs_mutex_unlock(&new_rule->up.mutex);
4251 xlate_out_uninit(&xout);
4252 rule_dpif_unref(new_rule);
4257 facet_reset_counters(struct facet *facet)
4259 facet->packet_count = 0;
4260 facet->byte_count = 0;
4261 facet->prev_packet_count = 0;
4262 facet->prev_byte_count = 0;
4263 facet->accounted_bytes = 0;
4267 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4268 struct dpif_flow_stats *stats, bool may_learn)
4270 struct ofport_dpif *in_port;
4271 struct xlate_in xin;
4273 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4274 if (in_port && in_port->is_tunnel) {
4275 netdev_vport_inc_rx(in_port->up.netdev, stats);
4278 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4279 xin.resubmit_stats = stats;
4280 xin.may_learn = may_learn;
4281 xlate_actions_for_side_effects(&xin);
4285 facet_push_stats(struct facet *facet, bool may_learn)
4287 struct dpif_flow_stats stats;
4289 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4290 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4291 ovs_assert(facet->used >= facet->prev_used);
4293 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4294 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4295 stats.used = facet->used;
4296 stats.tcp_flags = facet->tcp_flags;
4298 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4299 facet->prev_packet_count = facet->packet_count;
4300 facet->prev_byte_count = facet->byte_count;
4301 facet->prev_used = facet->used;
4303 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4305 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4306 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4307 stats.n_packets, stats.n_bytes);
4308 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4313 push_all_stats__(bool run_fast)
4315 static long long int rl = LLONG_MIN;
4316 struct ofproto_dpif *ofproto;
4318 if (time_msec() < rl) {
4322 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4323 struct cls_cursor cursor;
4324 struct facet *facet;
4326 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4327 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4328 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4329 facet_push_stats(facet, false);
4334 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4337 rl = time_msec() + 100;
4341 push_all_stats(void)
4343 push_all_stats__(true);
4347 rule_dpif_credit_stats(struct rule_dpif *rule,
4348 const struct dpif_flow_stats *stats)
4350 ovs_mutex_lock(&rule->stats_mutex);
4351 rule->packet_count += stats->n_packets;
4352 rule->byte_count += stats->n_bytes;
4353 rule->up.used = MAX(rule->up.used, stats->used);
4354 ovs_mutex_unlock(&rule->stats_mutex);
4358 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4360 return is_fail_open_rule(&rule->up);
4364 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4366 return rule_is_table_miss(&rule->up);
4370 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4371 OVS_REQUIRES(rule->up.mutex)
4373 return rule->up.flow_cookie;
4377 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4378 uint16_t hard_timeout)
4380 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4383 /* Returns 'rule''s actions. The caller owns a reference on the returned
4384 * actions and must eventually release it (with rule_actions_unref()) to avoid
4386 struct rule_actions *
4387 rule_dpif_get_actions(const struct rule_dpif *rule)
4389 return rule_get_actions(&rule->up);
4394 static struct subfacet *
4395 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4396 size_t key_len, uint32_t key_hash)
4398 struct subfacet *subfacet;
4400 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4401 &backer->subfacets) {
4402 if (subfacet->key_len == key_len
4403 && !memcmp(key, subfacet->key, key_len)) {
4411 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4412 * 'key_hash' must be a hash over miss->key. The caller must have already
4413 * ensured that no subfacet subfacet already exists. */
4414 static struct subfacet *
4415 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4417 struct dpif_backer *backer = miss->ofproto->backer;
4418 const struct nlattr *key = miss->key;
4419 size_t key_len = miss->key_len;
4420 struct subfacet *subfacet;
4422 subfacet = (list_is_empty(&facet->subfacets)
4423 ? &facet->one_subfacet
4424 : xmalloc(sizeof *subfacet));
4426 COVERAGE_INC(subfacet_create);
4427 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4428 list_push_back(&facet->subfacets, &subfacet->list_node);
4429 subfacet->facet = facet;
4430 subfacet->key = xmemdup(key, key_len);
4431 subfacet->key_len = key_len;
4432 subfacet->used = miss->stats.used;
4433 subfacet->created = subfacet->used;
4434 subfacet->dp_packet_count = 0;
4435 subfacet->dp_byte_count = 0;
4436 subfacet->path = SF_NOT_INSTALLED;
4437 subfacet->backer = backer;
4442 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4443 * its facet within 'ofproto', and frees it. */
4445 subfacet_destroy__(struct subfacet *subfacet)
4447 struct facet *facet = subfacet->facet;
4449 COVERAGE_INC(subfacet_destroy);
4450 subfacet_uninstall(subfacet);
4451 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4452 list_remove(&subfacet->list_node);
4453 free(subfacet->key);
4454 if (subfacet != &facet->one_subfacet) {
4459 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4460 * last remaining subfacet in its facet destroys the facet too. */
4462 subfacet_destroy(struct subfacet *subfacet)
4464 struct facet *facet = subfacet->facet;
4466 if (list_is_singleton(&facet->subfacets)) {
4467 /* facet_remove() needs at least one subfacet (it will remove it). */
4468 facet_remove(facet);
4470 subfacet_destroy__(subfacet);
4475 subfacet_destroy_batch(struct dpif_backer *backer,
4476 struct subfacet **subfacets, int n)
4478 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4479 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4480 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4483 for (i = 0; i < n; i++) {
4484 ops[i].type = DPIF_OP_FLOW_DEL;
4485 ops[i].u.flow_del.key = subfacets[i]->key;
4486 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4487 ops[i].u.flow_del.stats = &stats[i];
4491 dpif_operate(backer->dpif, opsp, n);
4492 for (i = 0; i < n; i++) {
4493 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4494 subfacets[i]->path = SF_NOT_INSTALLED;
4495 subfacet_destroy(subfacets[i]);
4500 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4501 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4502 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4503 * since 'subfacet' was last updated.
4505 * Returns 0 if successful, otherwise a positive errno value. */
4507 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4508 struct dpif_flow_stats *stats)
4510 struct facet *facet = subfacet->facet;
4511 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4512 const struct nlattr *actions = odp_actions->data;
4513 size_t actions_len = odp_actions->size;
4514 struct odputil_keybuf maskbuf;
4517 uint64_t slow_path_stub[128 / 8];
4518 enum dpif_flow_put_flags flags;
4521 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4524 flags |= DPIF_FP_ZERO_STATS;
4527 if (path == SF_SLOW_PATH) {
4528 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4529 slow_path_stub, sizeof slow_path_stub,
4530 &actions, &actions_len);
4533 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4534 if (enable_megaflows) {
4535 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4536 &facet->flow, UINT32_MAX);
4539 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4540 subfacet->key_len, mask.data, mask.size,
4541 actions, actions_len, stats);
4544 subfacet_reset_dp_stats(subfacet, stats);
4548 COVERAGE_INC(subfacet_install_fail);
4550 subfacet->path = path;
4555 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4557 subfacet_uninstall(struct subfacet *subfacet)
4559 if (subfacet->path != SF_NOT_INSTALLED) {
4560 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4561 struct dpif_flow_stats stats;
4564 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4565 subfacet->key_len, &stats);
4566 subfacet_reset_dp_stats(subfacet, &stats);
4568 subfacet_update_stats(subfacet, &stats);
4570 subfacet->path = SF_NOT_INSTALLED;
4572 ovs_assert(subfacet->dp_packet_count == 0);
4573 ovs_assert(subfacet->dp_byte_count == 0);
4577 /* Resets 'subfacet''s datapath statistics counters. This should be called
4578 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4579 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4580 * was reset in the datapath. 'stats' will be modified to include only
4581 * statistics new since 'subfacet' was last updated. */
4583 subfacet_reset_dp_stats(struct subfacet *subfacet,
4584 struct dpif_flow_stats *stats)
4587 && subfacet->dp_packet_count <= stats->n_packets
4588 && subfacet->dp_byte_count <= stats->n_bytes) {
4589 stats->n_packets -= subfacet->dp_packet_count;
4590 stats->n_bytes -= subfacet->dp_byte_count;
4593 subfacet->dp_packet_count = 0;
4594 subfacet->dp_byte_count = 0;
4597 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4599 * Because of the meaning of a subfacet's counters, it only makes sense to do
4600 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4601 * represents a packet that was sent by hand or if it represents statistics
4602 * that have been cleared out of the datapath. */
4604 subfacet_update_stats(struct subfacet *subfacet,
4605 const struct dpif_flow_stats *stats)
4607 if (stats->n_packets || stats->used > subfacet->used) {
4608 struct facet *facet = subfacet->facet;
4610 subfacet->used = MAX(subfacet->used, stats->used);
4611 facet->used = MAX(facet->used, stats->used);
4612 facet->packet_count += stats->n_packets;
4613 facet->byte_count += stats->n_bytes;
4614 facet->tcp_flags |= stats->tcp_flags;
4620 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4621 * the fields that were relevant as part of the lookup. */
4623 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4624 struct flow_wildcards *wc, struct rule_dpif **rule)
4626 struct ofport_dpif *port;
4628 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4631 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4633 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4634 flow->in_port.ofp_port);
4637 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4638 ofproto->no_packet_in_rule, rule);
4642 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4643 const struct flow *flow, struct flow_wildcards *wc,
4644 uint8_t table_id, struct rule_dpif **rule)
4646 const struct cls_rule *cls_rule;
4647 struct classifier *cls;
4651 if (table_id >= N_TABLES) {
4656 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4657 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4660 cls = &ofproto->up.tables[table_id].cls;
4661 ovs_rwlock_rdlock(&cls->rwlock);
4662 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4663 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4664 /* We must pretend that transport ports are unavailable. */
4665 struct flow ofpc_normal_flow = *flow;
4666 ofpc_normal_flow.tp_src = htons(0);
4667 ofpc_normal_flow.tp_dst = htons(0);
4668 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4669 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4670 cls_rule = &ofproto->drop_frags_rule->up.cr;
4671 /* Frag mask in wc already set above. */
4673 cls_rule = classifier_lookup(cls, flow, wc);
4676 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4677 rule_dpif_ref(*rule);
4678 ovs_rwlock_unlock(&cls->rwlock);
4680 return *rule != NULL;
4683 /* Given a port configuration (specified as zero if there's no port), chooses
4684 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4685 * flow table miss. */
4687 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4688 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4690 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4691 rule_dpif_ref(*rule);
4695 rule_dpif_ref(struct rule_dpif *rule)
4698 ofproto_rule_ref(&rule->up);
4703 rule_dpif_unref(struct rule_dpif *rule)
4706 ofproto_rule_unref(&rule->up);
4711 complete_operation(struct rule_dpif *rule)
4712 OVS_REQUIRES(ofproto_mutex)
4714 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4716 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4717 ofoperation_complete(rule->up.pending, 0);
4720 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4722 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4725 static struct rule *
4728 struct rule_dpif *rule = xmalloc(sizeof *rule);
4733 rule_dealloc(struct rule *rule_)
4735 struct rule_dpif *rule = rule_dpif_cast(rule_);
4740 rule_construct(struct rule *rule_)
4742 struct rule_dpif *rule = rule_dpif_cast(rule_);
4743 ovs_mutex_init(&rule->stats_mutex);
4744 ovs_mutex_lock(&rule->stats_mutex);
4745 rule->packet_count = 0;
4746 rule->byte_count = 0;
4747 ovs_mutex_unlock(&rule->stats_mutex);
4752 rule_insert(struct rule *rule_)
4753 OVS_REQUIRES(ofproto_mutex)
4755 struct rule_dpif *rule = rule_dpif_cast(rule_);
4756 complete_operation(rule);
4760 rule_delete(struct rule *rule_)
4761 OVS_REQUIRES(ofproto_mutex)
4763 struct rule_dpif *rule = rule_dpif_cast(rule_);
4764 complete_operation(rule);
4768 rule_destruct(struct rule *rule_)
4770 struct rule_dpif *rule = rule_dpif_cast(rule_);
4771 ovs_mutex_destroy(&rule->stats_mutex);
4775 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4777 struct rule_dpif *rule = rule_dpif_cast(rule_);
4779 /* push_all_stats() can handle flow misses which, when using the learn
4780 * action, can cause rules to be added and deleted. This can corrupt our
4781 * caller's datastructures which assume that rule_get_stats() doesn't have
4782 * an impact on the flow table. To be safe, we disable miss handling. */
4783 push_all_stats__(false);
4785 /* Start from historical data for 'rule' itself that are no longer tracked
4786 * in facets. This counts, for example, facets that have expired. */
4787 ovs_mutex_lock(&rule->stats_mutex);
4788 *packets = rule->packet_count;
4789 *bytes = rule->byte_count;
4790 ovs_mutex_unlock(&rule->stats_mutex);
4794 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4795 struct ofpbuf *packet)
4797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4799 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4803 rule_execute(struct rule *rule, const struct flow *flow,
4804 struct ofpbuf *packet)
4806 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4807 ofpbuf_delete(packet);
4812 rule_modify_actions(struct rule *rule_, bool reset_counters)
4813 OVS_REQUIRES(ofproto_mutex)
4815 struct rule_dpif *rule = rule_dpif_cast(rule_);
4817 if (reset_counters) {
4818 ovs_mutex_lock(&rule->stats_mutex);
4819 rule->packet_count = 0;
4820 rule->byte_count = 0;
4821 ovs_mutex_unlock(&rule->stats_mutex);
4824 complete_operation(rule);
4827 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4829 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4832 static struct ofgroup *
4835 struct group_dpif *group = xzalloc(sizeof *group);
4840 group_dealloc(struct ofgroup *group_)
4842 struct group_dpif *group = group_dpif_cast(group_);
4847 group_construct_stats(struct group_dpif *group)
4848 OVS_REQUIRES(group->stats_mutex)
4850 group->packet_count = 0;
4851 group->byte_count = 0;
4852 if (!group->bucket_stats) {
4853 group->bucket_stats = xcalloc(group->up.n_buckets,
4854 sizeof *group->bucket_stats);
4856 memset(group->bucket_stats, 0, group->up.n_buckets *
4857 sizeof *group->bucket_stats);
4862 group_construct(struct ofgroup *group_)
4864 struct group_dpif *group = group_dpif_cast(group_);
4865 ovs_mutex_init(&group->stats_mutex);
4866 ovs_mutex_lock(&group->stats_mutex);
4867 group_construct_stats(group);
4868 ovs_mutex_unlock(&group->stats_mutex);
4873 group_destruct__(struct group_dpif *group)
4874 OVS_REQUIRES(group->stats_mutex)
4876 free(group->bucket_stats);
4877 group->bucket_stats = NULL;
4881 group_destruct(struct ofgroup *group_)
4883 struct group_dpif *group = group_dpif_cast(group_);
4884 ovs_mutex_lock(&group->stats_mutex);
4885 group_destruct__(group);
4886 ovs_mutex_unlock(&group->stats_mutex);
4887 ovs_mutex_destroy(&group->stats_mutex);
4891 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4893 struct group_dpif *group = group_dpif_cast(group_);
4894 struct group_dpif *victim = group_dpif_cast(victim_);
4896 ovs_mutex_lock(&group->stats_mutex);
4897 if (victim->up.n_buckets < group->up.n_buckets) {
4898 group_destruct__(group);
4900 group_construct_stats(group);
4901 ovs_mutex_unlock(&group->stats_mutex);
4907 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4909 struct group_dpif *group = group_dpif_cast(group_);
4911 /* Start from historical data for 'group' itself that are no longer tracked
4912 * in facets. This counts, for example, facets that have expired. */
4913 ovs_mutex_lock(&group->stats_mutex);
4914 ogs->packet_count = group->packet_count;
4915 ogs->byte_count = group->byte_count;
4916 memcpy(ogs->bucket_stats, group->bucket_stats,
4917 group->up.n_buckets * sizeof *group->bucket_stats);
4918 ovs_mutex_unlock(&group->stats_mutex);
4924 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4925 struct group_dpif **group)
4926 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4928 struct ofgroup *ofgroup;
4932 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4933 *group = found ? group_dpif_cast(ofgroup) : NULL;
4939 group_dpif_release(struct group_dpif *group)
4940 OVS_RELEASES(group->up.rwlock)
4942 ofproto_group_release(&group->up);
4946 group_dpif_get_buckets(const struct group_dpif *group,
4947 const struct list **buckets)
4949 *buckets = &group->up.buckets;
4952 enum ofp11_group_type
4953 group_dpif_get_type(const struct group_dpif *group)
4955 return group->up.type;
4958 /* Sends 'packet' out 'ofport'.
4959 * May modify 'packet'.
4960 * Returns 0 if successful, otherwise a positive errno value. */
4962 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4967 error = xlate_send_packet(ofport, packet);
4969 ovs_mutex_lock(&ofproto->stats_mutex);
4970 ofproto->stats.tx_packets++;
4971 ofproto->stats.tx_bytes += packet->size;
4972 ovs_mutex_unlock(&ofproto->stats_mutex);
4976 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4977 * The action will state 'slow' as the reason that the action is in the slow
4978 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4979 * dump-flows" output to see why a flow is in the slow path.)
4981 * The 'stub_size' bytes in 'stub' will be used to store the action.
4982 * 'stub_size' must be large enough for the action.
4984 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4987 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4988 enum slow_path_reason slow,
4989 uint64_t *stub, size_t stub_size,
4990 const struct nlattr **actionsp, size_t *actions_lenp)
4992 union user_action_cookie cookie;
4995 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4996 cookie.slow_path.unused = 0;
4997 cookie.slow_path.reason = slow;
4999 ofpbuf_use_stack(&buf, stub, stub_size);
5000 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5001 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5003 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5005 odp_port_t odp_port;
5008 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5009 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5010 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5012 *actionsp = buf.data;
5013 *actions_lenp = buf.size;
5017 set_frag_handling(struct ofproto *ofproto_,
5018 enum ofp_config_flags frag_handling)
5020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5021 if (frag_handling != OFPC_FRAG_REASM) {
5022 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5030 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5031 const struct flow *flow,
5032 const struct ofpact *ofpacts, size_t ofpacts_len)
5034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5036 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5037 ofpacts_len, packet);
5044 set_netflow(struct ofproto *ofproto_,
5045 const struct netflow_options *netflow_options)
5047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5049 if (netflow_options) {
5050 if (!ofproto->netflow) {
5051 ofproto->netflow = netflow_create();
5052 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5054 return netflow_set_options(ofproto->netflow, netflow_options);
5055 } else if (ofproto->netflow) {
5056 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5057 netflow_destroy(ofproto->netflow);
5058 ofproto->netflow = NULL;
5065 get_netflow_ids(const struct ofproto *ofproto_,
5066 uint8_t *engine_type, uint8_t *engine_id)
5068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5070 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5074 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5076 if (!facet_is_controller_flow(facet) &&
5077 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5078 struct subfacet *subfacet;
5079 struct ofexpired expired;
5081 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5082 if (subfacet->path == SF_FAST_PATH) {
5083 struct dpif_flow_stats stats;
5085 subfacet_install(subfacet, &facet->xout.odp_actions,
5087 subfacet_update_stats(subfacet, &stats);
5091 expired.flow = facet->flow;
5092 expired.packet_count = facet->packet_count;
5093 expired.byte_count = facet->byte_count;
5094 expired.used = facet->used;
5095 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5100 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5102 struct cls_cursor cursor;
5103 struct facet *facet;
5105 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5106 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5107 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5108 send_active_timeout(ofproto, facet);
5110 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5113 static struct ofproto_dpif *
5114 ofproto_dpif_lookup(const char *name)
5116 struct ofproto_dpif *ofproto;
5118 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5119 hash_string(name, 0), &all_ofproto_dpifs) {
5120 if (!strcmp(ofproto->up.name, name)) {
5128 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5129 const char *argv[], void *aux OVS_UNUSED)
5131 struct ofproto_dpif *ofproto;
5134 ofproto = ofproto_dpif_lookup(argv[1]);
5136 unixctl_command_reply_error(conn, "no such bridge");
5139 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5140 mac_learning_flush(ofproto->ml);
5141 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5143 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5144 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5145 mac_learning_flush(ofproto->ml);
5146 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5150 unixctl_command_reply(conn, "table successfully flushed");
5153 static struct ofport_dpif *
5154 ofbundle_get_a_port(const struct ofbundle *bundle)
5156 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5161 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5162 const char *argv[], void *aux OVS_UNUSED)
5164 struct ds ds = DS_EMPTY_INITIALIZER;
5165 const struct ofproto_dpif *ofproto;
5166 const struct mac_entry *e;
5168 ofproto = ofproto_dpif_lookup(argv[1]);
5170 unixctl_command_reply_error(conn, "no such bridge");
5174 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5175 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5176 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5177 struct ofbundle *bundle = e->port.p;
5178 char name[OFP_MAX_PORT_NAME_LEN];
5180 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5182 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5183 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5184 mac_entry_age(ofproto->ml, e));
5186 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5187 unixctl_command_reply(conn, ds_cstr(&ds));
5192 struct xlate_out xout;
5193 struct xlate_in xin;
5199 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5201 struct rule_actions *actions;
5204 ds_put_char_multiple(result, '\t', level);
5206 ds_put_cstr(result, "No match\n");
5210 ovs_mutex_lock(&rule->up.mutex);
5211 cookie = rule->up.flow_cookie;
5212 ovs_mutex_unlock(&rule->up.mutex);
5214 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5215 rule ? rule->up.table_id : 0, ntohll(cookie));
5216 cls_rule_format(&rule->up.cr, result);
5217 ds_put_char(result, '\n');
5219 actions = rule_dpif_get_actions(rule);
5221 ds_put_char_multiple(result, '\t', level);
5222 ds_put_cstr(result, "OpenFlow actions=");
5223 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5224 ds_put_char(result, '\n');
5226 rule_actions_unref(actions);
5230 trace_format_flow(struct ds *result, int level, const char *title,
5231 struct trace_ctx *trace)
5233 ds_put_char_multiple(result, '\t', level);
5234 ds_put_format(result, "%s: ", title);
5235 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5236 ds_put_cstr(result, "unchanged");
5238 flow_format(result, &trace->xin.flow);
5239 trace->flow = trace->xin.flow;
5241 ds_put_char(result, '\n');
5245 trace_format_regs(struct ds *result, int level, const char *title,
5246 struct trace_ctx *trace)
5250 ds_put_char_multiple(result, '\t', level);
5251 ds_put_format(result, "%s:", title);
5252 for (i = 0; i < FLOW_N_REGS; i++) {
5253 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5255 ds_put_char(result, '\n');
5259 trace_format_odp(struct ds *result, int level, const char *title,
5260 struct trace_ctx *trace)
5262 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5264 ds_put_char_multiple(result, '\t', level);
5265 ds_put_format(result, "%s: ", title);
5266 format_odp_actions(result, odp_actions->data, odp_actions->size);
5267 ds_put_char(result, '\n');
5271 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5273 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5274 struct ds *result = trace->result;
5276 ds_put_char(result, '\n');
5277 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5278 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5279 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5280 trace_format_rule(result, recurse + 1, rule);
5284 trace_report(struct xlate_in *xin, const char *s, int recurse)
5286 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5287 struct ds *result = trace->result;
5289 ds_put_char_multiple(result, '\t', recurse);
5290 ds_put_cstr(result, s);
5291 ds_put_char(result, '\n');
5294 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
5295 * forms are supported:
5297 * - [dpname] odp_flow [-generate | packet]
5298 * - bridge br_flow [-generate | packet]
5300 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
5301 * returns a nonnull error message. */
5303 parse_flow_and_packet(int argc, const char *argv[],
5304 struct ofproto_dpif **ofprotop, struct flow *flow,
5305 struct ofpbuf **packetp)
5307 const struct dpif_backer *backer = NULL;
5308 const char *error = NULL;
5309 struct simap port_names = SIMAP_INITIALIZER(&port_names);
5310 struct ofpbuf *packet;
5311 struct ofpbuf odp_key;
5312 struct ofpbuf odp_mask;
5314 ofpbuf_init(&odp_key, 0);
5315 ofpbuf_init(&odp_mask, 0);
5317 /* Handle "-generate" or a hex string as the last argument. */
5318 if (!strcmp(argv[argc - 1], "-generate")) {
5319 packet = ofpbuf_new(0);
5322 error = eth_from_hex(argv[argc - 1], &packet);
5325 } else if (argc == 4) {
5326 /* The 3-argument form must end in "-generate' or a hex string. */
5331 /* odp_flow can have its in_port specified as a name instead of port no.
5332 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5333 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5334 * we need to create a simap of name to port no. */
5336 const char *dp_type;
5337 if (!strncmp(argv[1], "ovs-", 4)) {
5338 dp_type = argv[1] + 4;
5342 backer = shash_find_data(&all_dpif_backers, dp_type);
5343 } else if (argc == 2) {
5344 struct shash_node *node;
5345 if (shash_count(&all_dpif_backers) == 1) {
5346 node = shash_first(&all_dpif_backers);
5347 backer = node->data;
5350 error = "Syntax error";
5353 if (backer && backer->dpif) {
5354 struct dpif_port dpif_port;
5355 struct dpif_port_dump port_dump;
5356 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5357 simap_put(&port_names, dpif_port.name,
5358 odp_to_u32(dpif_port.port_no));
5362 /* Parse the flow and determine whether a datapath or
5363 * bridge is specified. If function odp_flow_key_from_string()
5364 * returns 0, the flow is a odp_flow. If function
5365 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5366 if (!odp_flow_from_string(argv[argc - 1], &port_names,
5367 &odp_key, &odp_mask)) {
5369 error = "Cannot find the datapath";
5373 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5374 NULL, ofprotop, NULL)) {
5375 error = "Invalid datapath flow";
5378 } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5380 error = "Must specify bridge name";
5384 *ofprotop = ofproto_dpif_lookup(argv[1]);
5386 error = "Unknown bridge name";
5390 error = "Bad flow syntax";
5394 /* Generate a packet, if requested. */
5396 if (!packet->size) {
5397 flow_compose(packet, flow);
5399 union flow_in_port in_port = flow->in_port;
5401 /* Use the metadata from the flow and the packet argument
5402 * to reconstruct the flow. */
5403 flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5412 ofpbuf_delete(packet);
5416 ofpbuf_uninit(&odp_key);
5417 ofpbuf_uninit(&odp_mask);
5418 simap_destroy(&port_names);
5423 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5424 void *aux OVS_UNUSED)
5426 struct ofproto_dpif *ofproto;
5427 struct ofpbuf *packet;
5431 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5436 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5437 unixctl_command_reply(conn, ds_cstr(&result));
5438 ds_destroy(&result);
5439 ofpbuf_delete(packet);
5441 unixctl_command_reply_error(conn, error);
5446 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5447 const char *argv[], void *aux OVS_UNUSED)
5449 enum ofputil_protocol usable_protocols;
5450 struct ofproto_dpif *ofproto;
5451 bool enforce_consistency;
5452 struct ofpbuf ofpacts;
5453 struct ofpbuf *packet;
5458 /* Three kinds of error return values! */
5465 ofpbuf_init(&ofpacts, 0);
5467 /* Parse actions. */
5468 rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5470 unixctl_command_reply_error(conn, rw_error);
5475 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5476 * consistency between the flow and the actions. With -consistent, we
5477 * enforce consistency even for a flow supported in OpenFlow 1.0. */
5478 if (!strcmp(argv[1], "-consistent")) {
5479 enforce_consistency = true;
5483 enforce_consistency = false;
5486 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5488 unixctl_command_reply_error(conn, error);
5492 /* Do the same checks as handle_packet_out() in ofproto.c.
5494 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5495 * strictly correct because these actions aren't in any table, but it's OK
5496 * because it 'table_id' is used only to check goto_table instructions, but
5497 * packet-outs take a list of actions and therefore it can't include
5500 * We skip the "meter" check here because meter is an instruction, not an
5501 * action, and thus cannot appear in ofpacts. */
5502 in_port = ofp_to_u16(flow.in_port.ofp_port);
5503 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5504 unixctl_command_reply_error(conn, "invalid in_port");
5507 if (enforce_consistency) {
5508 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5509 u16_to_ofp(ofproto->up.max_ports),
5510 0, 0, usable_protocols);
5512 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5513 u16_to_ofp(ofproto->up.max_ports), 0, 0,
5519 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5520 unixctl_command_reply_error(conn, ds_cstr(&result));
5524 ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5525 unixctl_command_reply(conn, ds_cstr(&result));
5528 ds_destroy(&result);
5529 ofpbuf_delete(packet);
5530 ofpbuf_uninit(&ofpacts);
5533 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5534 * description of the results to 'ds'.
5536 * The trace follows a packet with the specified 'flow' through the flow
5537 * table. 'packet' may be nonnull to trace an actual packet, with consequent
5538 * side effects (if it is nonnull then its flow must be 'flow').
5540 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5541 * trace, otherwise the actions are determined by a flow table lookup. */
5543 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5544 const struct ofpbuf *packet,
5545 const struct ofpact ofpacts[], size_t ofpacts_len,
5548 struct rule_dpif *rule;
5549 struct flow_wildcards wc;
5551 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5552 ds_put_cstr(ds, "Flow: ");
5553 flow_format(ds, flow);
5554 ds_put_char(ds, '\n');
5556 flow_wildcards_init_catchall(&wc);
5560 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5562 trace_format_rule(ds, 0, rule);
5563 if (rule == ofproto->miss_rule) {
5564 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5565 } else if (rule == ofproto->no_packet_in_rule) {
5566 ds_put_cstr(ds, "\nNo match, packets dropped because "
5567 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5568 } else if (rule == ofproto->drop_frags_rule) {
5569 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5570 "and the fragment handling mode is \"drop\".\n");
5574 if (rule || ofpacts) {
5575 uint64_t odp_actions_stub[1024 / 8];
5576 struct ofpbuf odp_actions;
5577 struct trace_ctx trace;
5581 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5584 ofpbuf_use_stub(&odp_actions,
5585 odp_actions_stub, sizeof odp_actions_stub);
5586 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5588 trace.xin.ofpacts = ofpacts;
5589 trace.xin.ofpacts_len = ofpacts_len;
5591 trace.xin.resubmit_hook = trace_resubmit;
5592 trace.xin.report_hook = trace_report;
5594 xlate_actions(&trace.xin, &trace.xout);
5595 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5597 ds_put_char(ds, '\n');
5598 trace_format_flow(ds, 0, "Final flow", &trace);
5600 match_init(&match, flow, &trace.xout.wc);
5601 ds_put_cstr(ds, "Relevant fields: ");
5602 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5603 ds_put_char(ds, '\n');
5605 ds_put_cstr(ds, "Datapath actions: ");
5606 format_odp_actions(ds, trace.xout.odp_actions.data,
5607 trace.xout.odp_actions.size);
5609 if (trace.xout.slow) {
5610 enum slow_path_reason slow;
5612 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5613 "slow path because it:");
5615 slow = trace.xout.slow;
5617 enum slow_path_reason bit = rightmost_1bit(slow);
5619 ds_put_format(ds, "\n\t- %s.",
5620 slow_path_reason_to_explanation(bit));
5626 xlate_out_uninit(&trace.xout);
5629 rule_dpif_unref(rule);
5632 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5633 * 'reply' describing the results. */
5635 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5637 struct cls_cursor cursor;
5638 struct facet *facet;
5642 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5643 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5644 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5645 if (!facet_check_consistency(facet)) {
5649 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5651 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5655 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5656 ofproto->up.name, errors);
5658 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5663 ofproto_dpif_self_check(struct unixctl_conn *conn,
5664 int argc, const char *argv[], void *aux OVS_UNUSED)
5666 struct ds reply = DS_EMPTY_INITIALIZER;
5667 struct ofproto_dpif *ofproto;
5670 ofproto = ofproto_dpif_lookup(argv[1]);
5672 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5673 "ofproto/list for help)");
5676 ofproto_dpif_self_check__(ofproto, &reply);
5678 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5679 ofproto_dpif_self_check__(ofproto, &reply);
5683 unixctl_command_reply(conn, ds_cstr(&reply));
5687 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5688 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5689 * to destroy 'ofproto_shash' and free the returned value. */
5690 static const struct shash_node **
5691 get_ofprotos(struct shash *ofproto_shash)
5693 const struct ofproto_dpif *ofproto;
5695 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5696 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5697 shash_add_nocopy(ofproto_shash, name, ofproto);
5700 return shash_sort(ofproto_shash);
5704 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5705 const char *argv[] OVS_UNUSED,
5706 void *aux OVS_UNUSED)
5708 struct ds ds = DS_EMPTY_INITIALIZER;
5709 struct shash ofproto_shash;
5710 const struct shash_node **sorted_ofprotos;
5713 shash_init(&ofproto_shash);
5714 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5715 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5716 const struct shash_node *node = sorted_ofprotos[i];
5717 ds_put_format(&ds, "%s\n", node->name);
5720 shash_destroy(&ofproto_shash);
5721 free(sorted_ofprotos);
5723 unixctl_command_reply(conn, ds_cstr(&ds));
5728 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5730 const struct shash_node **ofprotos;
5731 struct ofproto_dpif *ofproto;
5732 struct shash ofproto_shash;
5733 uint64_t n_hit, n_missed;
5736 n_hit = n_missed = 0;
5737 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5738 if (ofproto->backer == backer) {
5739 n_missed += ofproto->n_missed;
5740 n_hit += ofproto->n_hit;
5744 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5745 dpif_name(backer->dpif), n_hit, n_missed);
5746 ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5747 " life span: %lldms\n", hmap_count(&backer->subfacets),
5748 backer->avg_n_subfacet, backer->max_n_subfacet,
5749 backer->avg_subfacet_life);
5751 shash_init(&ofproto_shash);
5752 ofprotos = get_ofprotos(&ofproto_shash);
5753 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5754 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5755 const struct shash_node **ports;
5758 if (ofproto->backer != backer) {
5762 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5763 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5765 ports = shash_sort(&ofproto->up.port_by_name);
5766 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5767 const struct shash_node *node = ports[j];
5768 struct ofport *ofport = node->data;
5770 odp_port_t odp_port;
5772 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5775 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5776 if (odp_port != ODPP_NONE) {
5777 ds_put_format(ds, "%"PRIu32":", odp_port);
5779 ds_put_cstr(ds, "none:");
5782 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5785 if (!netdev_get_config(ofport->netdev, &config)) {
5786 const struct smap_node **nodes;
5789 nodes = smap_sort(&config);
5790 for (i = 0; i < smap_count(&config); i++) {
5791 const struct smap_node *node = nodes[i];
5792 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5793 node->key, node->value);
5797 smap_destroy(&config);
5799 ds_put_char(ds, ')');
5800 ds_put_char(ds, '\n');
5804 shash_destroy(&ofproto_shash);
5809 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5810 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5812 struct ds ds = DS_EMPTY_INITIALIZER;
5813 const struct shash_node **backers;
5816 backers = shash_sort(&all_dpif_backers);
5817 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5818 dpif_show_backer(backers[i]->data, &ds);
5822 unixctl_command_reply(conn, ds_cstr(&ds));
5826 /* Dump the megaflow (facet) cache. This is useful to check the
5827 * correctness of flow wildcarding, since the same mechanism is used for
5828 * both xlate caching and kernel wildcarding.
5830 * It's important to note that in the output the flow description uses
5831 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5833 * This command is only needed for advanced debugging, so it's not
5834 * documented in the man page. */
5836 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5837 int argc OVS_UNUSED, const char *argv[],
5838 void *aux OVS_UNUSED)
5840 struct ds ds = DS_EMPTY_INITIALIZER;
5841 const struct ofproto_dpif *ofproto;
5842 long long int now = time_msec();
5843 struct cls_cursor cursor;
5844 struct facet *facet;
5846 ofproto = ofproto_dpif_lookup(argv[1]);
5848 unixctl_command_reply_error(conn, "no such bridge");
5852 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5853 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5854 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5855 cls_rule_format(&facet->cr, &ds);
5856 ds_put_cstr(&ds, ", ");
5857 ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5858 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5859 ds_put_cstr(&ds, "Datapath actions: ");
5860 if (facet->xout.slow) {
5861 uint64_t slow_path_stub[128 / 8];
5862 const struct nlattr *actions;
5865 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5866 slow_path_stub, sizeof slow_path_stub,
5867 &actions, &actions_len);
5868 format_odp_actions(&ds, actions, actions_len);
5870 format_odp_actions(&ds, facet->xout.odp_actions.data,
5871 facet->xout.odp_actions.size);
5873 ds_put_cstr(&ds, "\n");
5875 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5877 ds_chomp(&ds, '\n');
5878 unixctl_command_reply(conn, ds_cstr(&ds));
5882 /* Disable using the megaflows.
5884 * This command is only needed for advanced debugging, so it's not
5885 * documented in the man page. */
5887 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5888 int argc OVS_UNUSED,
5889 const char *argv[] OVS_UNUSED,
5890 void *aux OVS_UNUSED)
5892 struct ofproto_dpif *ofproto;
5894 enable_megaflows = false;
5896 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5897 flush(&ofproto->up);
5900 unixctl_command_reply(conn, "megaflows disabled");
5903 /* Re-enable using megaflows.
5905 * This command is only needed for advanced debugging, so it's not
5906 * documented in the man page. */
5908 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5909 int argc OVS_UNUSED,
5910 const char *argv[] OVS_UNUSED,
5911 void *aux OVS_UNUSED)
5913 struct ofproto_dpif *ofproto;
5915 enable_megaflows = true;
5917 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5918 flush(&ofproto->up);
5921 unixctl_command_reply(conn, "megaflows enabled");
5925 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5926 int argc OVS_UNUSED, const char *argv[],
5927 void *aux OVS_UNUSED)
5929 struct ds ds = DS_EMPTY_INITIALIZER;
5930 const struct ofproto_dpif *ofproto;
5931 struct subfacet *subfacet;
5933 ofproto = ofproto_dpif_lookup(argv[1]);
5935 unixctl_command_reply_error(conn, "no such bridge");
5939 update_stats(ofproto->backer);
5941 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5942 struct facet *facet = subfacet->facet;
5943 struct odputil_keybuf maskbuf;
5946 if (facet->ofproto != ofproto) {
5950 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5951 if (enable_megaflows) {
5952 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5953 &facet->flow, UINT32_MAX);
5956 odp_flow_format(subfacet->key, subfacet->key_len,
5957 mask.data, mask.size, NULL, &ds, false);
5959 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5960 subfacet->dp_packet_count, subfacet->dp_byte_count);
5961 if (subfacet->used) {
5962 ds_put_format(&ds, "%.3fs",
5963 (time_msec() - subfacet->used) / 1000.0);
5965 ds_put_format(&ds, "never");
5967 if (subfacet->facet->tcp_flags) {
5968 ds_put_cstr(&ds, ", flags:");
5969 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5972 ds_put_cstr(&ds, ", actions:");
5973 if (facet->xout.slow) {
5974 uint64_t slow_path_stub[128 / 8];
5975 const struct nlattr *actions;
5978 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5979 slow_path_stub, sizeof slow_path_stub,
5980 &actions, &actions_len);
5981 format_odp_actions(&ds, actions, actions_len);
5983 format_odp_actions(&ds, facet->xout.odp_actions.data,
5984 facet->xout.odp_actions.size);
5986 ds_put_char(&ds, '\n');
5989 unixctl_command_reply(conn, ds_cstr(&ds));
5994 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5995 int argc OVS_UNUSED, const char *argv[],
5996 void *aux OVS_UNUSED)
5998 struct ds ds = DS_EMPTY_INITIALIZER;
5999 struct ofproto_dpif *ofproto;
6001 ofproto = ofproto_dpif_lookup(argv[1]);
6003 unixctl_command_reply_error(conn, "no such bridge");
6007 flush(&ofproto->up);
6009 unixctl_command_reply(conn, ds_cstr(&ds));
6014 ofproto_dpif_unixctl_init(void)
6016 static bool registered;
6022 unixctl_command_register(
6024 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6025 1, 3, ofproto_unixctl_trace, NULL);
6026 unixctl_command_register(
6027 "ofproto/trace-packet-out",
6028 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6029 2, 6, ofproto_unixctl_trace_actions, NULL);
6030 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6031 ofproto_unixctl_fdb_flush, NULL);
6032 unixctl_command_register("fdb/show", "bridge", 1, 1,
6033 ofproto_unixctl_fdb_show, NULL);
6034 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6035 ofproto_dpif_self_check, NULL);
6036 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6037 ofproto_unixctl_dpif_dump_dps, NULL);
6038 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6040 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6041 ofproto_unixctl_dpif_dump_flows, NULL);
6042 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6043 ofproto_unixctl_dpif_del_flows, NULL);
6044 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6045 ofproto_unixctl_dpif_dump_megaflows, NULL);
6046 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6047 ofproto_unixctl_dpif_disable_megaflows, NULL);
6048 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6049 ofproto_unixctl_dpif_enable_megaflows, NULL);
6052 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6054 * This is deprecated. It is only for compatibility with broken device drivers
6055 * in old versions of Linux that do not properly support VLANs when VLAN
6056 * devices are not used. When broken device drivers are no longer in
6057 * widespread use, we will delete these interfaces. */
6060 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6063 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6065 if (realdev_ofp_port == ofport->realdev_ofp_port
6066 && vid == ofport->vlandev_vid) {
6070 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6072 if (ofport->realdev_ofp_port) {
6075 if (realdev_ofp_port && ofport->bundle) {
6076 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6077 * themselves be part of a bundle. */
6078 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6081 ofport->realdev_ofp_port = realdev_ofp_port;
6082 ofport->vlandev_vid = vid;
6084 if (realdev_ofp_port) {
6085 vsp_add(ofport, realdev_ofp_port, vid);
6092 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6094 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6098 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6099 OVS_EXCLUDED(ofproto->vsp_mutex)
6103 ovs_mutex_lock(&ofproto->vsp_mutex);
6104 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6105 ovs_mutex_unlock(&ofproto->vsp_mutex);
6110 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6111 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6112 OVS_REQUIRES(ofproto->vsp_mutex)
6114 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6115 int vid = vlan_tci_to_vid(vlan_tci);
6116 const struct vlan_splinter *vsp;
6118 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6119 hash_realdev_vid(realdev_ofp_port, vid),
6120 &ofproto->realdev_vid_map) {
6121 if (vsp->realdev_ofp_port == realdev_ofp_port
6122 && vsp->vid == vid) {
6123 return vsp->vlandev_ofp_port;
6127 return realdev_ofp_port;
6130 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6131 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6132 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6133 * 'vlan_tci' 9, it would return the port number of eth0.9.
6135 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6136 * function just returns its 'realdev_ofp_port' argument. */
6138 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6139 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6140 OVS_EXCLUDED(ofproto->vsp_mutex)
6144 ovs_mutex_lock(&ofproto->vsp_mutex);
6145 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6146 ovs_mutex_unlock(&ofproto->vsp_mutex);
6150 static struct vlan_splinter *
6151 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6153 struct vlan_splinter *vsp;
6155 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6156 hash_ofp_port(vlandev_ofp_port),
6157 &ofproto->vlandev_map) {
6158 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6166 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6167 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6168 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6169 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6170 * eth0 and store 9 in '*vid'.
6172 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6173 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6176 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6177 ofp_port_t vlandev_ofp_port, int *vid)
6178 OVS_REQUIRES(ofproto->vsp_mutex)
6180 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6181 const struct vlan_splinter *vsp;
6183 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6188 return vsp->realdev_ofp_port;
6194 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6195 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6196 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6197 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6198 * always the case unless VLAN splinters are enabled), returns false without
6199 * making any changes. */
6201 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6202 OVS_EXCLUDED(ofproto->vsp_mutex)
6207 ovs_mutex_lock(&ofproto->vsp_mutex);
6208 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6209 ovs_mutex_unlock(&ofproto->vsp_mutex);
6214 /* Cause the flow to be processed as if it came in on the real device with
6215 * the VLAN device's VLAN ID. */
6216 flow->in_port.ofp_port = realdev;
6217 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6222 vsp_remove(struct ofport_dpif *port)
6224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6225 struct vlan_splinter *vsp;
6227 ovs_mutex_lock(&ofproto->vsp_mutex);
6228 vsp = vlandev_find(ofproto, port->up.ofp_port);
6230 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6231 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6234 port->realdev_ofp_port = 0;
6236 VLOG_ERR("missing vlan device record");
6238 ovs_mutex_unlock(&ofproto->vsp_mutex);
6242 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6246 ovs_mutex_lock(&ofproto->vsp_mutex);
6247 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6248 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6249 == realdev_ofp_port)) {
6250 struct vlan_splinter *vsp;
6252 vsp = xmalloc(sizeof *vsp);
6253 vsp->realdev_ofp_port = realdev_ofp_port;
6254 vsp->vlandev_ofp_port = port->up.ofp_port;
6257 port->realdev_ofp_port = realdev_ofp_port;
6259 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6260 hash_ofp_port(port->up.ofp_port));
6261 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6262 hash_realdev_vid(realdev_ofp_port, vid));
6264 VLOG_ERR("duplicate vlan device record");
6266 ovs_mutex_unlock(&ofproto->vsp_mutex);
6270 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6272 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6273 return ofport ? ofport->odp_port : ODPP_NONE;
6276 struct ofport_dpif *
6277 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6279 struct ofport_dpif *port;
6281 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6282 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6283 &backer->odp_to_ofport_map) {
6284 if (port->odp_port == odp_port) {
6285 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6290 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6295 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6297 struct ofport_dpif *port;
6299 port = odp_port_to_ofport(ofproto->backer, odp_port);
6300 if (port && &ofproto->up == port->up.ofproto) {
6301 return port->up.ofp_port;
6307 const struct ofproto_class ofproto_dpif_class = {
6342 port_is_lacp_current,
6343 NULL, /* rule_choose_table */
6352 rule_modify_actions,
6366 get_stp_port_status,
6373 is_mirror_output_bundle,
6374 forward_bpdu_changed,
6375 set_mac_table_config,
6377 NULL, /* meter_get_features */
6378 NULL, /* meter_set */
6379 NULL, /* meter_get */
6380 NULL, /* meter_del */
6381 group_alloc, /* group_alloc */
6382 group_construct, /* group_construct */
6383 group_destruct, /* group_destruct */
6384 group_dealloc, /* group_dealloc */
6385 group_modify, /* group_modify */
6386 group_get_stats, /* group_get_stats */