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"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
74 COVERAGE_DEFINE(packet_in_overflow);
75 COVERAGE_DEFINE(flow_mod_overflow);
77 /* Number of implemented OpenFlow tables. */
78 enum { N_TABLES = 255 };
79 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
80 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
85 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
87 struct flow_wildcards *wc);
89 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
92 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
93 struct ofproto_dpif *ofproto; /* Owning ofproto. */
94 void *aux; /* Key supplied by ofproto's client. */
95 char *name; /* Identifier for log messages. */
98 struct list ports; /* Contains "struct ofport"s. */
99 enum port_vlan_mode vlan_mode; /* VLAN mode */
100 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
101 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
102 * NULL if all VLANs are trunked. */
103 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
104 struct bond *bond; /* Nonnull iff more than one port. */
105 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
108 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
111 static void bundle_remove(struct ofport *);
112 static void bundle_update(struct ofbundle *);
113 static void bundle_destroy(struct ofbundle *);
114 static void bundle_del_port(struct ofport_dpif *);
115 static void bundle_run(struct ofbundle *);
116 static void bundle_wait(struct ofbundle *);
118 static void stp_run(struct ofproto_dpif *ofproto);
119 static void stp_wait(struct ofproto_dpif *ofproto);
120 static int set_stp_port(struct ofport *,
121 const struct ofproto_port_stp_settings *);
123 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
124 enum slow_path_reason,
125 uint64_t *stub, size_t stub_size,
126 const struct nlattr **actionsp,
127 size_t *actions_lenp);
129 /* A subfacet (see "struct subfacet" below) has three possible installation
132 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
133 * case just after the subfacet is created, just before the subfacet is
134 * destroyed, or if the datapath returns an error when we try to install a
137 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
139 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
140 * ofproto_dpif is installed in the datapath.
143 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
144 SF_FAST_PATH, /* Full actions are installed. */
145 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
148 /* A dpif flow and actions associated with a facet.
150 * See also the large comment on struct facet. */
153 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
154 struct list list_node; /* In struct facet's 'facets' list. */
155 struct facet *facet; /* Owning facet. */
156 struct dpif_backer *backer; /* Owning backer. */
158 enum odp_key_fitness key_fitness;
162 long long int used; /* Time last used; time created if not used. */
163 long long int created; /* Time created. */
165 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
166 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
168 enum subfacet_path path; /* Installed in datapath? */
171 #define SUBFACET_DESTROY_MAX_BATCH 50
173 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
175 static struct subfacet *subfacet_find(struct dpif_backer *,
176 const struct nlattr *key, size_t key_len,
178 static void subfacet_destroy(struct subfacet *);
179 static void subfacet_destroy__(struct subfacet *);
180 static void subfacet_destroy_batch(struct dpif_backer *,
181 struct subfacet **, int n);
182 static void subfacet_reset_dp_stats(struct subfacet *,
183 struct dpif_flow_stats *);
184 static void subfacet_update_stats(struct subfacet *,
185 const struct dpif_flow_stats *);
186 static int subfacet_install(struct subfacet *,
187 const struct ofpbuf *odp_actions,
188 struct dpif_flow_stats *);
189 static void subfacet_uninstall(struct subfacet *);
191 /* A unique, non-overlapping instantiation of an OpenFlow flow.
193 * A facet associates a "struct flow", which represents the Open vSwitch
194 * userspace idea of an exact-match flow, with one or more subfacets.
195 * While the facet is created based on an exact-match flow, it is stored
196 * within the ofproto based on the wildcards that could be expressed
197 * based on the flow table and other configuration. (See the 'wc'
198 * description in "struct xlate_out" for more details.)
200 * Each subfacet tracks the datapath's idea of the flow equivalent to
201 * the facet. When the kernel module (or other dpif implementation) and
202 * Open vSwitch userspace agree on the definition of a flow key, there
203 * is exactly one subfacet per facet. If the dpif implementation
204 * supports more-specific flow matching than userspace, however, a facet
205 * can have more than one subfacet. Examples include the dpif
206 * implementation not supporting the same wildcards as userspace or some
207 * distinction in flow that userspace simply doesn't understand.
209 * Flow expiration works in terms of subfacets, so a facet must have at
210 * least one subfacet or it will never expire, leaking memory. */
213 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
214 struct ofproto_dpif *ofproto;
217 struct list subfacets;
218 long long int used; /* Time last used; time created if not used. */
221 struct flow flow; /* Flow of the creating subfacet. */
222 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
226 * - Do include packets and bytes sent "by hand", e.g. with
229 * - Do include packets and bytes that were obtained from the datapath
230 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
231 * DPIF_FP_ZERO_STATS).
233 * - Do not include packets or bytes that can be obtained from the
234 * datapath for any existing subfacet.
236 uint64_t packet_count; /* Number of packets received. */
237 uint64_t byte_count; /* Number of bytes received. */
239 /* Resubmit statistics. */
240 uint64_t prev_packet_count; /* Number of packets from last stats push. */
241 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
242 long long int prev_used; /* Used time from last stats push. */
245 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
246 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
247 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
249 struct xlate_out xout;
250 bool fail_open; /* Facet matched the fail open rule. */
252 /* Storage for a single subfacet, to reduce malloc() time and space
253 * overhead. (A facet always has at least one subfacet and in the common
254 * case has exactly one subfacet. However, 'one_subfacet' may not
255 * always be valid, since it could have been removed after newer
256 * subfacets were pushed onto the 'subfacets' list.) */
257 struct subfacet one_subfacet;
259 long long int learn_rl; /* Rate limiter for facet_learn(). */
262 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
264 struct dpif_flow_stats *);
265 static void facet_remove(struct facet *);
266 static void facet_free(struct facet *);
268 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
269 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
270 const struct flow *);
271 static bool facet_revalidate(struct facet *);
272 static bool facet_check_consistency(struct facet *);
274 static void facet_flush_stats(struct facet *);
276 static void facet_reset_counters(struct facet *);
277 static void facet_push_stats(struct facet *, bool may_learn);
278 static void facet_learn(struct facet *);
279 static void facet_account(struct facet *);
280 static void push_all_stats(void);
282 static bool facet_is_controller_flow(struct facet *);
285 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
289 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
290 struct list bundle_node; /* In struct ofbundle's "ports" list. */
291 struct cfm *cfm; /* Connectivity Fault Management, if any. */
292 struct bfd *bfd; /* BFD, if any. */
293 bool may_enable; /* May be enabled in bonds. */
294 bool is_tunnel; /* This port is a tunnel. */
295 long long int carrier_seq; /* Carrier status changes. */
296 struct ofport_dpif *peer; /* Peer if patch port. */
299 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
300 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
301 long long int stp_state_entered;
303 /* Queue to DSCP mapping. */
304 struct ofproto_port_queue *qdscp;
307 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
309 * This is deprecated. It is only for compatibility with broken device
310 * drivers in old versions of Linux that do not properly support VLANs when
311 * VLAN devices are not used. When broken device drivers are no longer in
312 * widespread use, we will delete these interfaces. */
313 ofp_port_t realdev_ofp_port;
317 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
319 * This is deprecated. It is only for compatibility with broken device drivers
320 * in old versions of Linux that do not properly support VLANs when VLAN
321 * devices are not used. When broken device drivers are no longer in
322 * widespread use, we will delete these interfaces. */
323 struct vlan_splinter {
324 struct hmap_node realdev_vid_node;
325 struct hmap_node vlandev_node;
326 ofp_port_t realdev_ofp_port;
327 ofp_port_t vlandev_ofp_port;
331 static void vsp_remove(struct ofport_dpif *);
332 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
334 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
337 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
340 static struct ofport_dpif *
341 ofport_dpif_cast(const struct ofport *ofport)
343 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
346 static void port_run(struct ofport_dpif *);
347 static void port_run_fast(struct ofport_dpif *);
348 static void port_wait(struct ofport_dpif *);
349 static int set_bfd(struct ofport *, const struct smap *);
350 static int set_cfm(struct ofport *, const struct cfm_settings *);
351 static void ofport_update_peer(struct ofport_dpif *);
352 static void run_fast_rl(void);
353 static int run_fast(struct ofproto *);
355 struct dpif_completion {
356 struct list list_node;
357 struct ofoperation *op;
360 /* Reasons that we might need to revalidate every facet, and corresponding
363 * A value of 0 means that there is no need to revalidate.
365 * It would be nice to have some cleaner way to integrate with coverage
366 * counters, but with only a few reasons I guess this is good enough for
368 enum revalidate_reason {
369 REV_RECONFIGURE = 1, /* Switch configuration changed. */
370 REV_STP, /* Spanning tree protocol port status change. */
371 REV_BOND, /* Bonding changed. */
372 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
373 REV_FLOW_TABLE, /* Flow table changed. */
374 REV_MAC_LEARNING, /* Mac learning changed. */
375 REV_INCONSISTENCY /* Facet self-check failed. */
377 COVERAGE_DEFINE(rev_reconfigure);
378 COVERAGE_DEFINE(rev_stp);
379 COVERAGE_DEFINE(rev_bond);
380 COVERAGE_DEFINE(rev_port_toggled);
381 COVERAGE_DEFINE(rev_flow_table);
382 COVERAGE_DEFINE(rev_mac_learning);
383 COVERAGE_DEFINE(rev_inconsistency);
385 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
386 * These are datapath flows which have no associated ofproto, if they did we
387 * would use facets. */
389 struct hmap_node hmap_node;
394 struct avg_subfacet_rates {
395 double add_rate; /* Moving average of new flows created per minute. */
396 double del_rate; /* Moving average of flows deleted per minute. */
399 /* All datapaths of a given type share a single dpif backer instance. */
404 struct timer next_expiration;
406 struct ovs_rwlock odp_to_ofport_lock;
407 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
409 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
411 /* Facet revalidation flags applying to facets which use this backer. */
412 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
414 struct hmap drop_keys; /* Set of dropped odp keys. */
415 bool recv_set_enable; /* Enables or disables receiving packets. */
417 struct hmap subfacets;
418 struct governor *governor;
420 /* Subfacet statistics.
422 * These keep track of the total number of subfacets added and deleted and
423 * flow life span. They are useful for computing the flow rates stats
424 * exposed via "ovs-appctl dpif/show". The goal is to learn about
425 * traffic patterns in ways that we can use later to improve Open vSwitch
426 * performance in new situations. */
427 long long int created; /* Time when it is created. */
428 unsigned max_n_subfacet; /* Maximum number of flows */
429 unsigned avg_n_subfacet; /* Average number of flows. */
430 long long int avg_subfacet_life; /* Average life span of subfacets. */
432 /* The average number of subfacets... */
433 struct avg_subfacet_rates hourly; /* ...over the last hour. */
434 struct avg_subfacet_rates daily; /* ...over the last day. */
435 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
436 long long int last_minute; /* Last time 'hourly' was updated. */
438 /* Number of subfacets added or deleted since 'last_minute'. */
439 unsigned subfacet_add_count;
440 unsigned subfacet_del_count;
442 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
443 unsigned long long int total_subfacet_add_count;
444 unsigned long long int total_subfacet_del_count;
447 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
448 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
450 static void drop_key_clear(struct dpif_backer *);
451 static void update_moving_averages(struct dpif_backer *backer);
453 struct ofproto_dpif {
454 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
456 struct dpif_backer *backer;
458 /* Special OpenFlow rules. */
459 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
460 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
461 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
464 struct netflow *netflow;
465 struct dpif_sflow *sflow;
466 struct dpif_ipfix *ipfix;
467 struct hmap bundles; /* Contains "struct ofbundle"s. */
468 struct mac_learning *ml;
469 bool has_bonded_bundles;
470 struct mbridge *mbridge;
473 struct classifier facets; /* Contains 'struct facet's. */
474 long long int consistency_rl;
476 /* Support for debugging async flow mods. */
477 struct list completions;
479 struct netdev_stats stats; /* To account packets generated and consumed in
484 long long int stp_last_tick;
486 /* VLAN splinters. */
487 struct ovs_mutex vsp_mutex;
488 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
489 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
492 struct sset ports; /* Set of standard port names. */
493 struct sset ghost_ports; /* Ports with no datapath port. */
494 struct sset port_poll_set; /* Queued names for port_poll() reply. */
495 int port_poll_errno; /* Last errno for port_poll() reply. */
497 /* Per ofproto's dpif stats. */
502 struct ovs_mutex flow_mod_mutex;
503 struct list flow_mods OVS_GUARDED;
504 size_t n_flow_mods OVS_GUARDED;
506 struct ovs_mutex pin_mutex;
507 struct list pins OVS_GUARDED;
508 size_t n_pins OVS_GUARDED;
511 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
512 * for debugging the asynchronous flow_mod implementation.) */
515 /* By default, flows in the datapath are wildcarded (megaflows). They
516 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
517 static bool enable_megaflows = true;
519 /* All existing ofproto_dpif instances, indexed by ->up.name. */
520 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
522 static void ofproto_dpif_unixctl_init(void);
524 static inline struct ofproto_dpif *
525 ofproto_dpif_cast(const struct ofproto *ofproto)
527 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
528 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
531 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
532 ofp_port_t ofp_port);
533 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
534 const struct ofpbuf *packet, struct ds *);
537 #define FLOW_MISS_MAX_BATCH 50
538 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
540 /* Flow expiration. */
541 static int expire(struct dpif_backer *);
544 static void send_netflow_active_timeouts(struct ofproto_dpif *);
547 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
549 /* Global variables. */
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
552 /* Initial mappings of port to bridge mappings. */
553 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
555 /* Executes and takes ownership of 'fm'. */
557 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
558 struct ofputil_flow_mod *fm)
560 ovs_mutex_lock(&ofproto->flow_mod_mutex);
561 if (ofproto->n_flow_mods > 1024) {
562 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
563 COVERAGE_INC(flow_mod_overflow);
569 list_push_back(&ofproto->flow_mods, &fm->list_node);
570 ofproto->n_flow_mods++;
571 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
575 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
576 struct ofputil_packet_in *pin)
578 ovs_mutex_lock(&ofproto->pin_mutex);
579 if (ofproto->n_pins > 1024) {
580 ovs_mutex_unlock(&ofproto->pin_mutex);
581 COVERAGE_INC(packet_in_overflow);
582 free(CONST_CAST(void *, pin->packet));
587 list_push_back(&ofproto->pins, &pin->list_node);
589 ovs_mutex_unlock(&ofproto->pin_mutex);
592 /* Factory functions. */
595 init(const struct shash *iface_hints)
597 struct shash_node *node;
599 /* Make a local copy, since we don't own 'iface_hints' elements. */
600 SHASH_FOR_EACH(node, iface_hints) {
601 const struct iface_hint *orig_hint = node->data;
602 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
604 new_hint->br_name = xstrdup(orig_hint->br_name);
605 new_hint->br_type = xstrdup(orig_hint->br_type);
606 new_hint->ofp_port = orig_hint->ofp_port;
608 shash_add(&init_ofp_ports, node->name, new_hint);
613 enumerate_types(struct sset *types)
615 dp_enumerate_types(types);
619 enumerate_names(const char *type, struct sset *names)
621 struct ofproto_dpif *ofproto;
624 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
625 if (strcmp(type, ofproto->up.type)) {
628 sset_add(names, ofproto->up.name);
635 del(const char *type, const char *name)
640 error = dpif_open(name, type, &dpif);
642 error = dpif_delete(dpif);
649 port_open_type(const char *datapath_type, const char *port_type)
651 return dpif_port_open_type(datapath_type, port_type);
654 /* Type functions. */
656 static void process_dpif_port_changes(struct dpif_backer *);
657 static void process_dpif_all_ports_changed(struct dpif_backer *);
658 static void process_dpif_port_change(struct dpif_backer *,
659 const char *devname);
660 static void process_dpif_port_error(struct dpif_backer *, int error);
662 static struct ofproto_dpif *
663 lookup_ofproto_dpif_by_port_name(const char *name)
665 struct ofproto_dpif *ofproto;
667 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
668 if (sset_contains(&ofproto->ports, name)) {
677 type_run(const char *type)
679 static long long int push_timer = LLONG_MIN;
680 struct dpif_backer *backer;
682 backer = shash_find_data(&all_dpif_backers, type);
684 /* This is not necessarily a problem, since backers are only
685 * created on demand. */
689 dpif_run(backer->dpif);
691 /* The most natural place to push facet statistics is when they're pulled
692 * from the datapath. However, when there are many flows in the datapath,
693 * this expensive operation can occur so frequently, that it reduces our
694 * ability to quickly set up flows. To reduce the cost, we push statistics
696 if (time_msec() > push_timer) {
697 push_timer = time_msec() + 2000;
701 /* If vswitchd started with other_config:flow_restore_wait set as "true",
702 * and the configuration has now changed to "false", enable receiving
703 * packets from the datapath. */
704 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
707 backer->recv_set_enable = true;
709 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
711 VLOG_ERR("Failed to enable receiving packets in dpif.");
714 dpif_flow_flush(backer->dpif);
715 backer->need_revalidate = REV_RECONFIGURE;
718 if (backer->need_revalidate) {
719 struct ofproto_dpif *ofproto;
720 struct simap_node *node;
721 struct simap tmp_backers;
723 /* Handle tunnel garbage collection. */
724 simap_init(&tmp_backers);
725 simap_swap(&backer->tnl_backers, &tmp_backers);
727 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
728 struct ofport_dpif *iter;
730 if (backer != ofproto->backer) {
734 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
735 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
738 if (!iter->is_tunnel) {
742 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
743 namebuf, sizeof namebuf);
744 node = simap_find(&tmp_backers, dp_port);
746 simap_put(&backer->tnl_backers, dp_port, node->data);
747 simap_delete(&tmp_backers, node);
748 node = simap_find(&backer->tnl_backers, dp_port);
750 node = simap_find(&backer->tnl_backers, dp_port);
752 odp_port_t odp_port = ODPP_NONE;
754 if (!dpif_port_add(backer->dpif, iter->up.netdev,
756 simap_put(&backer->tnl_backers, dp_port,
757 odp_to_u32(odp_port));
758 node = simap_find(&backer->tnl_backers, dp_port);
763 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
764 if (tnl_port_reconfigure(iter, iter->up.netdev,
766 backer->need_revalidate = REV_RECONFIGURE;
771 SIMAP_FOR_EACH (node, &tmp_backers) {
772 dpif_port_del(backer->dpif, u32_to_odp(node->data));
774 simap_destroy(&tmp_backers);
776 switch (backer->need_revalidate) {
777 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
778 case REV_STP: COVERAGE_INC(rev_stp); break;
779 case REV_BOND: COVERAGE_INC(rev_bond); break;
780 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
781 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
782 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
783 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
785 backer->need_revalidate = 0;
787 /* Clear the drop_keys in case we should now be accepting some
788 * formerly dropped flows. */
789 drop_key_clear(backer);
791 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
792 struct facet *facet, *next;
793 struct ofport_dpif *ofport;
794 struct cls_cursor cursor;
795 struct ofbundle *bundle;
797 if (ofproto->backer != backer) {
801 ovs_rwlock_wrlock(&xlate_rwlock);
802 xlate_ofproto_set(ofproto, ofproto->up.name,
803 ofproto->backer->dpif, ofproto->miss_rule,
804 ofproto->no_packet_in_rule, ofproto->ml,
805 ofproto->stp, ofproto->mbridge,
806 ofproto->sflow, ofproto->ipfix,
807 ofproto->up.frag_handling,
808 ofproto->up.forward_bpdu,
809 connmgr_has_in_band(ofproto->up.connmgr),
810 ofproto->netflow != NULL);
812 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
813 xlate_bundle_set(ofproto, bundle, bundle->name,
814 bundle->vlan_mode, bundle->vlan,
815 bundle->trunks, bundle->use_priority_tags,
816 bundle->bond, bundle->lacp,
820 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
821 int stp_port = ofport->stp_port
822 ? stp_port_no(ofport->stp_port)
824 xlate_ofport_set(ofproto, ofport->bundle, ofport,
825 ofport->up.ofp_port, ofport->odp_port,
826 ofport->up.netdev, ofport->cfm,
827 ofport->bfd, ofport->peer, stp_port,
828 ofport->qdscp, ofport->n_qdscp,
829 ofport->up.pp.config, ofport->is_tunnel,
832 ovs_rwlock_unlock(&xlate_rwlock);
834 /* Only ofproto-dpif cares about the facet classifier so we just
835 * lock cls_cursor_init() to appease the thread safety analysis. */
836 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
837 cls_cursor_init(&cursor, &ofproto->facets, NULL);
838 ovs_rwlock_unlock(&ofproto->facets.rwlock);
839 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
840 facet_revalidate(facet);
846 if (!backer->recv_set_enable) {
847 /* Wake up before a max of 1000ms. */
848 timer_set_duration(&backer->next_expiration, 1000);
849 } else if (timer_expired(&backer->next_expiration)) {
850 int delay = expire(backer);
851 timer_set_duration(&backer->next_expiration, delay);
854 process_dpif_port_changes(backer);
856 if (backer->governor) {
859 governor_run(backer->governor);
861 /* If the governor has shrunk to its minimum size and the number of
862 * subfacets has dwindled, then drop the governor entirely.
864 * For hysteresis, the number of subfacets to drop the governor is
865 * smaller than the number needed to trigger its creation. */
866 n_subfacets = hmap_count(&backer->subfacets);
867 if (n_subfacets * 4 < flow_eviction_threshold
868 && governor_is_idle(backer->governor)) {
869 governor_destroy(backer->governor);
870 backer->governor = NULL;
877 /* Check for and handle port changes in 'backer''s dpif. */
879 process_dpif_port_changes(struct dpif_backer *backer)
885 error = dpif_port_poll(backer->dpif, &devname);
891 process_dpif_all_ports_changed(backer);
895 process_dpif_port_change(backer, devname);
900 process_dpif_port_error(backer, error);
907 process_dpif_all_ports_changed(struct dpif_backer *backer)
909 struct ofproto_dpif *ofproto;
910 struct dpif_port dpif_port;
911 struct dpif_port_dump dump;
912 struct sset devnames;
915 sset_init(&devnames);
916 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
917 if (ofproto->backer == backer) {
918 struct ofport *ofport;
920 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
921 sset_add(&devnames, netdev_get_name(ofport->netdev));
925 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
926 sset_add(&devnames, dpif_port.name);
929 SSET_FOR_EACH (devname, &devnames) {
930 process_dpif_port_change(backer, devname);
932 sset_destroy(&devnames);
936 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
938 struct ofproto_dpif *ofproto;
939 struct dpif_port port;
941 /* Don't report on the datapath's device. */
942 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
946 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
947 &all_ofproto_dpifs) {
948 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
953 ofproto = lookup_ofproto_dpif_by_port_name(devname);
954 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
955 /* The port was removed. If we know the datapath,
956 * report it through poll_set(). If we don't, it may be
957 * notifying us of a removal we initiated, so ignore it.
958 * If there's a pending ENOBUFS, let it stand, since
959 * everything will be reevaluated. */
960 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
961 sset_add(&ofproto->port_poll_set, devname);
962 ofproto->port_poll_errno = 0;
964 } else if (!ofproto) {
965 /* The port was added, but we don't know with which
966 * ofproto we should associate it. Delete it. */
967 dpif_port_del(backer->dpif, port.port_no);
969 struct ofport_dpif *ofport;
971 ofport = ofport_dpif_cast(shash_find_data(
972 &ofproto->up.port_by_name, devname));
974 && ofport->odp_port != port.port_no
975 && !odp_port_to_ofport(backer, port.port_no))
977 /* 'ofport''s datapath port number has changed from
978 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
979 * structures to match. */
980 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
981 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
982 ofport->odp_port = port.port_no;
983 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
984 hash_odp_port(port.port_no));
985 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
986 backer->need_revalidate = REV_RECONFIGURE;
989 dpif_port_destroy(&port);
992 /* Propagate 'error' to all ofprotos based on 'backer'. */
994 process_dpif_port_error(struct dpif_backer *backer, int error)
996 struct ofproto_dpif *ofproto;
998 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
999 if (ofproto->backer == backer) {
1000 sset_clear(&ofproto->port_poll_set);
1001 ofproto->port_poll_errno = error;
1007 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1011 /* If recv_set_enable is false, we should not handle upcalls. */
1012 if (!backer->recv_set_enable) {
1016 /* Handle one or more batches of upcalls, until there's nothing left to do
1017 * or until we do a fixed total amount of work.
1019 * We do work in batches because it can be much cheaper to set up a number
1020 * of flows and fire off their patches all at once. We do multiple batches
1021 * because in some cases handling a packet can cause another packet to be
1022 * queued almost immediately as part of the return flow. Both
1023 * optimizations can make major improvements on some benchmarks and
1024 * presumably for real traffic as well. */
1026 while (work < max_batch) {
1027 int retval = handle_upcalls(backer, max_batch - work);
1038 type_run_fast(const char *type)
1040 struct dpif_backer *backer;
1042 backer = shash_find_data(&all_dpif_backers, type);
1044 /* This is not necessarily a problem, since backers are only
1045 * created on demand. */
1049 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1055 static long long int port_rl = LLONG_MIN;
1056 static unsigned int backer_rl = 0;
1058 if (time_msec() >= port_rl) {
1059 struct ofproto_dpif *ofproto;
1061 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1062 run_fast(&ofproto->up);
1064 port_rl = time_msec() + 200;
1067 /* XXX: We have to be careful not to do too much work in this function. If
1068 * we call dpif_backer_run_fast() too often, or with too large a batch,
1069 * performance improves signifcantly, but at a cost. It's possible for the
1070 * number of flows in the datapath to increase without bound, and for poll
1071 * loops to take 10s of seconds. The correct solution to this problem,
1072 * long term, is to separate flow miss handling into it's own thread so it
1073 * isn't affected by revalidations, and expirations. Until then, this is
1074 * the best we can do. */
1075 if (++backer_rl >= 10) {
1076 struct shash_node *node;
1079 SHASH_FOR_EACH (node, &all_dpif_backers) {
1080 dpif_backer_run_fast(node->data, 1);
1086 type_wait(const char *type)
1088 struct dpif_backer *backer;
1090 backer = shash_find_data(&all_dpif_backers, type);
1092 /* This is not necessarily a problem, since backers are only
1093 * created on demand. */
1097 if (backer->governor) {
1098 governor_wait(backer->governor);
1101 timer_wait(&backer->next_expiration);
1104 /* Basic life-cycle. */
1106 static int add_internal_flows(struct ofproto_dpif *);
1108 static struct ofproto *
1111 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1112 return &ofproto->up;
1116 dealloc(struct ofproto *ofproto_)
1118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1123 close_dpif_backer(struct dpif_backer *backer)
1125 struct shash_node *node;
1127 ovs_assert(backer->refcount > 0);
1129 if (--backer->refcount) {
1133 drop_key_clear(backer);
1134 hmap_destroy(&backer->drop_keys);
1136 simap_destroy(&backer->tnl_backers);
1137 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1138 hmap_destroy(&backer->odp_to_ofport_map);
1139 node = shash_find(&all_dpif_backers, backer->type);
1141 shash_delete(&all_dpif_backers, node);
1142 dpif_close(backer->dpif);
1144 ovs_assert(hmap_is_empty(&backer->subfacets));
1145 hmap_destroy(&backer->subfacets);
1146 governor_destroy(backer->governor);
1151 /* Datapath port slated for removal from datapath. */
1152 struct odp_garbage {
1153 struct list list_node;
1154 odp_port_t odp_port;
1158 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1160 struct dpif_backer *backer;
1161 struct dpif_port_dump port_dump;
1162 struct dpif_port port;
1163 struct shash_node *node;
1164 struct list garbage_list;
1165 struct odp_garbage *garbage, *next;
1171 backer = shash_find_data(&all_dpif_backers, type);
1178 backer_name = xasprintf("ovs-%s", type);
1180 /* Remove any existing datapaths, since we assume we're the only
1181 * userspace controlling the datapath. */
1183 dp_enumerate_names(type, &names);
1184 SSET_FOR_EACH(name, &names) {
1185 struct dpif *old_dpif;
1187 /* Don't remove our backer if it exists. */
1188 if (!strcmp(name, backer_name)) {
1192 if (dpif_open(name, type, &old_dpif)) {
1193 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1195 dpif_delete(old_dpif);
1196 dpif_close(old_dpif);
1199 sset_destroy(&names);
1201 backer = xmalloc(sizeof *backer);
1203 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1206 VLOG_ERR("failed to open datapath of type %s: %s", type,
1207 ovs_strerror(error));
1212 backer->type = xstrdup(type);
1213 backer->governor = NULL;
1214 backer->refcount = 1;
1215 hmap_init(&backer->odp_to_ofport_map);
1216 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1217 hmap_init(&backer->drop_keys);
1218 hmap_init(&backer->subfacets);
1219 timer_set_duration(&backer->next_expiration, 1000);
1220 backer->need_revalidate = 0;
1221 simap_init(&backer->tnl_backers);
1222 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1225 if (backer->recv_set_enable) {
1226 dpif_flow_flush(backer->dpif);
1229 /* Loop through the ports already on the datapath and remove any
1230 * that we don't need anymore. */
1231 list_init(&garbage_list);
1232 dpif_port_dump_start(&port_dump, backer->dpif);
1233 while (dpif_port_dump_next(&port_dump, &port)) {
1234 node = shash_find(&init_ofp_ports, port.name);
1235 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1236 garbage = xmalloc(sizeof *garbage);
1237 garbage->odp_port = port.port_no;
1238 list_push_front(&garbage_list, &garbage->list_node);
1241 dpif_port_dump_done(&port_dump);
1243 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1244 dpif_port_del(backer->dpif, garbage->odp_port);
1245 list_remove(&garbage->list_node);
1249 shash_add(&all_dpif_backers, type, backer);
1251 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1253 VLOG_ERR("failed to listen on datapath of type %s: %s",
1254 type, ovs_strerror(error));
1255 close_dpif_backer(backer);
1259 backer->max_n_subfacet = 0;
1260 backer->created = time_msec();
1261 backer->last_minute = backer->created;
1262 memset(&backer->hourly, 0, sizeof backer->hourly);
1263 memset(&backer->daily, 0, sizeof backer->daily);
1264 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1265 backer->subfacet_add_count = 0;
1266 backer->subfacet_del_count = 0;
1267 backer->total_subfacet_add_count = 0;
1268 backer->total_subfacet_del_count = 0;
1269 backer->avg_n_subfacet = 0;
1270 backer->avg_subfacet_life = 0;
1276 construct(struct ofproto *ofproto_)
1278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1279 struct shash_node *node, *next;
1280 odp_port_t max_ports;
1283 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1288 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1289 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1290 ofp_to_u16(OFPP_MAX))));
1292 ofproto->netflow = NULL;
1293 ofproto->sflow = NULL;
1294 ofproto->ipfix = NULL;
1295 ofproto->stp = NULL;
1296 hmap_init(&ofproto->bundles);
1297 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1298 ofproto->mbridge = mbridge_create();
1299 ofproto->has_bonded_bundles = false;
1300 ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1302 classifier_init(&ofproto->facets);
1303 ofproto->consistency_rl = LLONG_MIN;
1305 list_init(&ofproto->completions);
1307 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1308 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1309 list_init(&ofproto->flow_mods);
1310 ofproto->n_flow_mods = 0;
1311 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1313 ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL);
1314 ovs_mutex_lock(&ofproto->pin_mutex);
1315 list_init(&ofproto->pins);
1316 ofproto->n_pins = 0;
1317 ovs_mutex_unlock(&ofproto->pin_mutex);
1319 ofproto_dpif_unixctl_init();
1321 hmap_init(&ofproto->vlandev_map);
1322 hmap_init(&ofproto->realdev_vid_map);
1324 sset_init(&ofproto->ports);
1325 sset_init(&ofproto->ghost_ports);
1326 sset_init(&ofproto->port_poll_set);
1327 ofproto->port_poll_errno = 0;
1329 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1330 struct iface_hint *iface_hint = node->data;
1332 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1333 /* Check if the datapath already has this port. */
1334 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1335 sset_add(&ofproto->ports, node->name);
1338 free(iface_hint->br_name);
1339 free(iface_hint->br_type);
1341 shash_delete(&init_ofp_ports, node);
1345 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1346 hash_string(ofproto->up.name, 0));
1347 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1349 ofproto_init_tables(ofproto_, N_TABLES);
1350 error = add_internal_flows(ofproto);
1351 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1354 ofproto->n_missed = 0;
1360 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1361 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1363 struct ofputil_flow_mod fm;
1366 match_init_catchall(&fm.match);
1368 match_set_reg(&fm.match, 0, id);
1369 fm.new_cookie = htonll(0);
1370 fm.cookie = htonll(0);
1371 fm.cookie_mask = htonll(0);
1372 fm.modify_cookie = false;
1373 fm.table_id = TBL_INTERNAL;
1374 fm.command = OFPFC_ADD;
1375 fm.idle_timeout = 0;
1376 fm.hard_timeout = 0;
1380 fm.ofpacts = ofpacts->data;
1381 fm.ofpacts_len = ofpacts->size;
1383 error = ofproto_flow_mod(&ofproto->up, &fm);
1385 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1386 id, ofperr_to_string(error));
1390 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1392 ovs_assert(*rulep != NULL);
1398 add_internal_flows(struct ofproto_dpif *ofproto)
1400 struct ofpact_controller *controller;
1401 uint64_t ofpacts_stub[128 / 8];
1402 struct ofpbuf ofpacts;
1406 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1409 controller = ofpact_put_CONTROLLER(&ofpacts);
1410 controller->max_len = UINT16_MAX;
1411 controller->controller_id = 0;
1412 controller->reason = OFPR_NO_MATCH;
1413 ofpact_pad(&ofpacts);
1415 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1420 ofpbuf_clear(&ofpacts);
1421 error = add_internal_flow(ofproto, id++, &ofpacts,
1422 &ofproto->no_packet_in_rule);
1427 error = add_internal_flow(ofproto, id++, &ofpacts,
1428 &ofproto->drop_frags_rule);
1433 complete_operations(struct ofproto_dpif *ofproto)
1435 struct dpif_completion *c, *next;
1437 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1438 ofoperation_complete(c->op, 0);
1439 list_remove(&c->list_node);
1445 destruct(struct ofproto *ofproto_)
1447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1448 struct rule_dpif *rule, *next_rule;
1449 struct ofputil_flow_mod *pin, *next_pin;
1450 struct ofputil_flow_mod *fm, *next_fm;
1451 struct oftable *table;
1453 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1454 ovs_rwlock_wrlock(&xlate_rwlock);
1455 xlate_remove_ofproto(ofproto);
1456 ovs_rwlock_unlock(&xlate_rwlock);
1458 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1459 complete_operations(ofproto);
1461 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1462 struct cls_cursor cursor;
1464 ovs_rwlock_wrlock(&table->cls.rwlock);
1465 cls_cursor_init(&cursor, &table->cls, NULL);
1466 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1467 ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up);
1469 ovs_rwlock_unlock(&table->cls.rwlock);
1472 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1473 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1474 list_remove(&fm->list_node);
1475 ofproto->n_flow_mods--;
1479 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1480 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1482 ovs_mutex_lock(&ofproto->pin_mutex);
1483 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1484 list_remove(&pin->list_node);
1489 ovs_mutex_unlock(&ofproto->pin_mutex);
1490 ovs_mutex_destroy(&ofproto->pin_mutex);
1492 mbridge_unref(ofproto->mbridge);
1494 netflow_destroy(ofproto->netflow);
1495 dpif_sflow_unref(ofproto->sflow);
1496 hmap_destroy(&ofproto->bundles);
1497 mac_learning_unref(ofproto->ml);
1499 classifier_destroy(&ofproto->facets);
1501 hmap_destroy(&ofproto->vlandev_map);
1502 hmap_destroy(&ofproto->realdev_vid_map);
1504 sset_destroy(&ofproto->ports);
1505 sset_destroy(&ofproto->ghost_ports);
1506 sset_destroy(&ofproto->port_poll_set);
1508 ovs_mutex_destroy(&ofproto->vsp_mutex);
1510 close_dpif_backer(ofproto->backer);
1514 run_fast(struct ofproto *ofproto_)
1516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1517 struct ofputil_packet_in *pin, *next_pin;
1518 struct ofputil_flow_mod *fm, *next_fm;
1519 struct list flow_mods, pins;
1520 struct ofport_dpif *ofport;
1522 /* Do not perform any periodic activity required by 'ofproto' while
1523 * waiting for flow restore to complete. */
1524 if (ofproto_get_flow_restore_wait()) {
1528 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1529 if (ofproto->n_flow_mods) {
1530 flow_mods = ofproto->flow_mods;
1531 list_moved(&flow_mods);
1532 list_init(&ofproto->flow_mods);
1533 ofproto->n_flow_mods = 0;
1535 list_init(&flow_mods);
1537 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1539 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1540 int error = ofproto_flow_mod(&ofproto->up, fm);
1541 if (error && !VLOG_DROP_WARN(&rl)) {
1542 VLOG_WARN("learning action failed to modify flow table (%s)",
1543 ofperr_get_name(error));
1546 list_remove(&fm->list_node);
1551 ovs_mutex_lock(&ofproto->pin_mutex);
1552 if (ofproto->n_pins) {
1553 pins = ofproto->pins;
1555 list_init(&ofproto->pins);
1556 ofproto->n_pins = 0;
1560 ovs_mutex_unlock(&ofproto->pin_mutex);
1562 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1563 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1564 list_remove(&pin->list_node);
1565 free(CONST_CAST(void *, pin->packet));
1569 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1570 port_run_fast(ofport);
1577 run(struct ofproto *ofproto_)
1579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1580 struct ofport_dpif *ofport;
1581 struct ofbundle *bundle;
1585 complete_operations(ofproto);
1588 if (mbridge_need_revalidate(ofproto->mbridge)) {
1589 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1590 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1591 mac_learning_flush(ofproto->ml);
1592 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1595 /* Do not perform any periodic activity below required by 'ofproto' while
1596 * waiting for flow restore to complete. */
1597 if (ofproto_get_flow_restore_wait()) {
1601 error = run_fast(ofproto_);
1606 if (ofproto->netflow) {
1607 if (netflow_run(ofproto->netflow)) {
1608 send_netflow_active_timeouts(ofproto);
1611 if (ofproto->sflow) {
1612 dpif_sflow_run(ofproto->sflow);
1615 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1618 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1623 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1624 if (mac_learning_run(ofproto->ml)) {
1625 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1627 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1629 /* Check the consistency of a random facet, to aid debugging. */
1630 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1631 if (time_msec() >= ofproto->consistency_rl
1632 && !classifier_is_empty(&ofproto->facets)
1633 && !ofproto->backer->need_revalidate) {
1634 struct cls_table *table;
1635 struct cls_rule *cr;
1636 struct facet *facet;
1638 ofproto->consistency_rl = time_msec() + 250;
1640 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1641 struct cls_table, hmap_node);
1642 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1644 facet = CONTAINER_OF(cr, struct facet, cr);
1646 if (!facet_check_consistency(facet)) {
1647 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1650 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1656 wait(struct ofproto *ofproto_)
1658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659 struct ofport_dpif *ofport;
1660 struct ofbundle *bundle;
1662 if (!clogged && !list_is_empty(&ofproto->completions)) {
1663 poll_immediate_wake();
1666 if (ofproto_get_flow_restore_wait()) {
1670 dpif_wait(ofproto->backer->dpif);
1671 dpif_recv_wait(ofproto->backer->dpif);
1672 if (ofproto->sflow) {
1673 dpif_sflow_wait(ofproto->sflow);
1675 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1678 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1679 bundle_wait(bundle);
1681 if (ofproto->netflow) {
1682 netflow_wait(ofproto->netflow);
1684 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1685 mac_learning_wait(ofproto->ml);
1686 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1688 if (ofproto->backer->need_revalidate) {
1689 /* Shouldn't happen, but if it does just go around again. */
1690 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1691 poll_immediate_wake();
1696 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1698 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699 struct cls_cursor cursor;
1700 size_t n_subfacets = 0;
1701 struct facet *facet;
1703 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1704 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1705 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1707 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1708 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1709 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1710 n_subfacets += list_size(&facet->subfacets);
1712 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1713 simap_increase(usage, "subfacets", n_subfacets);
1717 flush(struct ofproto *ofproto_)
1719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1720 struct subfacet *subfacet, *next_subfacet;
1721 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1725 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1726 &ofproto->backer->subfacets) {
1727 if (subfacet->facet->ofproto != ofproto) {
1731 if (subfacet->path != SF_NOT_INSTALLED) {
1732 batch[n_batch++] = subfacet;
1733 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1734 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1738 subfacet_destroy(subfacet);
1743 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1748 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1749 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1751 *arp_match_ip = true;
1752 *actions = (OFPUTIL_A_OUTPUT |
1753 OFPUTIL_A_SET_VLAN_VID |
1754 OFPUTIL_A_SET_VLAN_PCP |
1755 OFPUTIL_A_STRIP_VLAN |
1756 OFPUTIL_A_SET_DL_SRC |
1757 OFPUTIL_A_SET_DL_DST |
1758 OFPUTIL_A_SET_NW_SRC |
1759 OFPUTIL_A_SET_NW_DST |
1760 OFPUTIL_A_SET_NW_TOS |
1761 OFPUTIL_A_SET_TP_SRC |
1762 OFPUTIL_A_SET_TP_DST |
1767 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1770 struct dpif_dp_stats s;
1771 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1774 strcpy(ots->name, "classifier");
1776 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1777 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1778 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1779 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1781 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1782 ots->lookup_count = htonll(n_lookup);
1783 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1786 static struct ofport *
1789 struct ofport_dpif *port = xmalloc(sizeof *port);
1794 port_dealloc(struct ofport *port_)
1796 struct ofport_dpif *port = ofport_dpif_cast(port_);
1801 port_construct(struct ofport *port_)
1803 struct ofport_dpif *port = ofport_dpif_cast(port_);
1804 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1805 const struct netdev *netdev = port->up.netdev;
1806 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1807 struct dpif_port dpif_port;
1810 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1811 port->bundle = NULL;
1814 port->may_enable = true;
1815 port->stp_port = NULL;
1816 port->stp_state = STP_DISABLED;
1817 port->is_tunnel = false;
1821 port->realdev_ofp_port = 0;
1822 port->vlandev_vid = 0;
1823 port->carrier_seq = netdev_get_carrier_resets(netdev);
1825 if (netdev_vport_is_patch(netdev)) {
1826 /* By bailing out here, we don't submit the port to the sFlow module
1827 * to be considered for counter polling export. This is correct
1828 * because the patch port represents an interface that sFlow considers
1829 * to be "internal" to the switch as a whole, and therefore not an
1830 * candidate for counter polling. */
1831 port->odp_port = ODPP_NONE;
1832 ofport_update_peer(port);
1836 error = dpif_port_query_by_name(ofproto->backer->dpif,
1837 netdev_vport_get_dpif_port(netdev, namebuf,
1844 port->odp_port = dpif_port.port_no;
1846 if (netdev_get_tunnel_config(netdev)) {
1847 tnl_port_add(port, port->up.netdev, port->odp_port);
1848 port->is_tunnel = true;
1850 /* Sanity-check that a mapping doesn't already exist. This
1851 * shouldn't happen for non-tunnel ports. */
1852 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1853 VLOG_ERR("port %s already has an OpenFlow port number",
1855 dpif_port_destroy(&dpif_port);
1859 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1860 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1861 hash_odp_port(port->odp_port));
1862 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1864 dpif_port_destroy(&dpif_port);
1866 if (ofproto->sflow) {
1867 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1874 port_destruct(struct ofport *port_)
1876 struct ofport_dpif *port = ofport_dpif_cast(port_);
1877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1878 const char *devname = netdev_get_name(port->up.netdev);
1879 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1880 const char *dp_port_name;
1882 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883 ovs_rwlock_wrlock(&xlate_rwlock);
1884 xlate_ofport_remove(port);
1885 ovs_rwlock_unlock(&xlate_rwlock);
1887 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1889 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1890 /* The underlying device is still there, so delete it. This
1891 * happens when the ofproto is being destroyed, since the caller
1892 * assumes that removal of attached ports will happen as part of
1894 if (!port->is_tunnel) {
1895 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1900 port->peer->peer = NULL;
1904 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1905 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1906 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1907 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1911 sset_find_and_delete(&ofproto->ports, devname);
1912 sset_find_and_delete(&ofproto->ghost_ports, devname);
1913 bundle_remove(port_);
1914 set_cfm(port_, NULL);
1915 set_bfd(port_, NULL);
1916 if (ofproto->sflow) {
1917 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1924 port_modified(struct ofport *port_)
1926 struct ofport_dpif *port = ofport_dpif_cast(port_);
1928 if (port->bundle && port->bundle->bond) {
1929 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1933 cfm_set_netdev(port->cfm, port->up.netdev);
1936 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1938 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1942 ofport_update_peer(port);
1946 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1948 struct ofport_dpif *port = ofport_dpif_cast(port_);
1949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1950 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1952 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1953 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1954 OFPUTIL_PC_NO_PACKET_IN)) {
1955 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1957 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1958 bundle_update(port->bundle);
1964 set_sflow(struct ofproto *ofproto_,
1965 const struct ofproto_sflow_options *sflow_options)
1967 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1968 struct dpif_sflow *ds = ofproto->sflow;
1970 if (sflow_options) {
1972 struct ofport_dpif *ofport;
1974 ds = ofproto->sflow = dpif_sflow_create();
1975 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1976 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1978 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1980 dpif_sflow_set_options(ds, sflow_options);
1983 dpif_sflow_unref(ds);
1984 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1985 ofproto->sflow = NULL;
1993 struct ofproto *ofproto_,
1994 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1995 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1996 size_t n_flow_exporters_options)
1998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1999 struct dpif_ipfix *di = ofproto->ipfix;
2001 if (bridge_exporter_options || flow_exporters_options) {
2003 di = ofproto->ipfix = dpif_ipfix_create();
2005 dpif_ipfix_set_options(
2006 di, bridge_exporter_options, flow_exporters_options,
2007 n_flow_exporters_options);
2010 dpif_ipfix_unref(di);
2011 ofproto->ipfix = NULL;
2018 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2020 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2027 struct ofproto_dpif *ofproto;
2029 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2030 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2031 ofport->cfm = cfm_create(ofport->up.netdev);
2034 if (cfm_configure(ofport->cfm, s)) {
2040 cfm_unref(ofport->cfm);
2046 get_cfm_status(const struct ofport *ofport_,
2047 struct ofproto_cfm_status *status)
2049 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052 status->faults = cfm_get_fault(ofport->cfm);
2053 status->remote_opstate = cfm_get_opup(ofport->cfm);
2054 status->health = cfm_get_health(ofport->cfm);
2055 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2063 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2066 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2071 if (ofport->bfd != old) {
2072 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2079 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2081 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2084 bfd_get_status(ofport->bfd, smap);
2091 /* Spanning Tree. */
2094 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2096 struct ofproto_dpif *ofproto = ofproto_;
2097 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2098 struct ofport_dpif *ofport;
2100 ofport = stp_port_get_aux(sp);
2102 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2103 ofproto->up.name, port_num);
2105 struct eth_header *eth = pkt->l2;
2107 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2108 if (eth_addr_is_zero(eth->eth_src)) {
2109 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2110 "with unknown MAC", ofproto->up.name, port_num);
2112 send_packet(ofport, pkt);
2118 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2120 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2124 /* Only revalidate flows if the configuration changed. */
2125 if (!s != !ofproto->stp) {
2126 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2130 if (!ofproto->stp) {
2131 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2132 send_bpdu_cb, ofproto);
2133 ofproto->stp_last_tick = time_msec();
2136 stp_set_bridge_id(ofproto->stp, s->system_id);
2137 stp_set_bridge_priority(ofproto->stp, s->priority);
2138 stp_set_hello_time(ofproto->stp, s->hello_time);
2139 stp_set_max_age(ofproto->stp, s->max_age);
2140 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2142 struct ofport *ofport;
2144 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2145 set_stp_port(ofport, NULL);
2148 stp_unref(ofproto->stp);
2149 ofproto->stp = NULL;
2156 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2162 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2163 s->designated_root = stp_get_designated_root(ofproto->stp);
2164 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2173 update_stp_port_state(struct ofport_dpif *ofport)
2175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2176 enum stp_state state;
2178 /* Figure out new state. */
2179 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2183 if (ofport->stp_state != state) {
2184 enum ofputil_port_state of_state;
2187 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2188 netdev_get_name(ofport->up.netdev),
2189 stp_state_name(ofport->stp_state),
2190 stp_state_name(state));
2191 if (stp_learn_in_state(ofport->stp_state)
2192 != stp_learn_in_state(state)) {
2193 /* xxx Learning action flows should also be flushed. */
2194 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2195 mac_learning_flush(ofproto->ml);
2196 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2198 fwd_change = stp_forward_in_state(ofport->stp_state)
2199 != stp_forward_in_state(state);
2201 ofproto->backer->need_revalidate = REV_STP;
2202 ofport->stp_state = state;
2203 ofport->stp_state_entered = time_msec();
2205 if (fwd_change && ofport->bundle) {
2206 bundle_update(ofport->bundle);
2209 /* Update the STP state bits in the OpenFlow port description. */
2210 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2211 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2212 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2213 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2214 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2216 ofproto_port_set_state(&ofport->up, of_state);
2220 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2221 * caller is responsible for assigning STP port numbers and ensuring
2222 * there are no duplicates. */
2224 set_stp_port(struct ofport *ofport_,
2225 const struct ofproto_port_stp_settings *s)
2227 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2229 struct stp_port *sp = ofport->stp_port;
2231 if (!s || !s->enable) {
2233 ofport->stp_port = NULL;
2234 stp_port_disable(sp);
2235 update_stp_port_state(ofport);
2238 } else if (sp && stp_port_no(sp) != s->port_num
2239 && ofport == stp_port_get_aux(sp)) {
2240 /* The port-id changed, so disable the old one if it's not
2241 * already in use by another port. */
2242 stp_port_disable(sp);
2245 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2246 stp_port_enable(sp);
2248 stp_port_set_aux(sp, ofport);
2249 stp_port_set_priority(sp, s->priority);
2250 stp_port_set_path_cost(sp, s->path_cost);
2252 update_stp_port_state(ofport);
2258 get_stp_port_status(struct ofport *ofport_,
2259 struct ofproto_port_stp_status *s)
2261 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2263 struct stp_port *sp = ofport->stp_port;
2265 if (!ofproto->stp || !sp) {
2271 s->port_id = stp_port_get_id(sp);
2272 s->state = stp_port_get_state(sp);
2273 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2274 s->role = stp_port_get_role(sp);
2275 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2281 stp_run(struct ofproto_dpif *ofproto)
2284 long long int now = time_msec();
2285 long long int elapsed = now - ofproto->stp_last_tick;
2286 struct stp_port *sp;
2289 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2290 ofproto->stp_last_tick = now;
2292 while (stp_get_changed_port(ofproto->stp, &sp)) {
2293 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2296 update_stp_port_state(ofport);
2300 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2301 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2302 mac_learning_flush(ofproto->ml);
2303 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2309 stp_wait(struct ofproto_dpif *ofproto)
2312 poll_timer_wait(1000);
2317 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2320 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2323 if (ofport->n_qdscp != n_qdscp
2324 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2325 n_qdscp * sizeof *qdscp))) {
2326 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2327 free(ofport->qdscp);
2328 ofport->qdscp = n_qdscp
2329 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2331 ofport->n_qdscp = n_qdscp;
2339 /* Expires all MAC learning entries associated with 'bundle' and forces its
2340 * ofproto to revalidate every flow.
2342 * Normally MAC learning entries are removed only from the ofproto associated
2343 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2344 * are removed from every ofproto. When patch ports and SLB bonds are in use
2345 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2346 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2347 * with the host from which it migrated. */
2349 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2351 struct ofproto_dpif *ofproto = bundle->ofproto;
2352 struct mac_learning *ml = ofproto->ml;
2353 struct mac_entry *mac, *next_mac;
2355 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2356 ovs_rwlock_wrlock(&ml->rwlock);
2357 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2358 if (mac->port.p == bundle) {
2360 struct ofproto_dpif *o;
2362 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2364 struct mac_entry *e;
2366 ovs_rwlock_wrlock(&o->ml->rwlock);
2367 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2369 mac_learning_expire(o->ml, e);
2371 ovs_rwlock_unlock(&o->ml->rwlock);
2376 mac_learning_expire(ml, mac);
2379 ovs_rwlock_unlock(&ml->rwlock);
2382 static struct ofbundle *
2383 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2385 struct ofbundle *bundle;
2387 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2388 &ofproto->bundles) {
2389 if (bundle->aux == aux) {
2397 bundle_update(struct ofbundle *bundle)
2399 struct ofport_dpif *port;
2401 bundle->floodable = true;
2402 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2403 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2404 || !stp_forward_in_state(port->stp_state)) {
2405 bundle->floodable = false;
2412 bundle_del_port(struct ofport_dpif *port)
2414 struct ofbundle *bundle = port->bundle;
2416 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2418 list_remove(&port->bundle_node);
2419 port->bundle = NULL;
2422 lacp_slave_unregister(bundle->lacp, port);
2425 bond_slave_unregister(bundle->bond, port);
2428 bundle_update(bundle);
2432 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2433 struct lacp_slave_settings *lacp)
2435 struct ofport_dpif *port;
2437 port = get_ofp_port(bundle->ofproto, ofp_port);
2442 if (port->bundle != bundle) {
2443 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2445 bundle_del_port(port);
2448 port->bundle = bundle;
2449 list_push_back(&bundle->ports, &port->bundle_node);
2450 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2451 || !stp_forward_in_state(port->stp_state)) {
2452 bundle->floodable = false;
2456 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2457 lacp_slave_register(bundle->lacp, port, lacp);
2464 bundle_destroy(struct ofbundle *bundle)
2466 struct ofproto_dpif *ofproto;
2467 struct ofport_dpif *port, *next_port;
2473 ofproto = bundle->ofproto;
2474 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2476 ovs_rwlock_wrlock(&xlate_rwlock);
2477 xlate_bundle_remove(bundle);
2478 ovs_rwlock_unlock(&xlate_rwlock);
2480 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2481 bundle_del_port(port);
2484 bundle_flush_macs(bundle, true);
2485 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2487 free(bundle->trunks);
2488 lacp_unref(bundle->lacp);
2489 bond_unref(bundle->bond);
2494 bundle_set(struct ofproto *ofproto_, void *aux,
2495 const struct ofproto_bundle_settings *s)
2497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2498 bool need_flush = false;
2499 struct ofport_dpif *port;
2500 struct ofbundle *bundle;
2501 unsigned long *trunks;
2507 bundle_destroy(bundle_lookup(ofproto, aux));
2511 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2512 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2514 bundle = bundle_lookup(ofproto, aux);
2516 bundle = xmalloc(sizeof *bundle);
2518 bundle->ofproto = ofproto;
2519 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2520 hash_pointer(aux, 0));
2522 bundle->name = NULL;
2524 list_init(&bundle->ports);
2525 bundle->vlan_mode = PORT_VLAN_TRUNK;
2527 bundle->trunks = NULL;
2528 bundle->use_priority_tags = s->use_priority_tags;
2529 bundle->lacp = NULL;
2530 bundle->bond = NULL;
2532 bundle->floodable = true;
2533 mbridge_register_bundle(ofproto->mbridge, bundle);
2536 if (!bundle->name || strcmp(s->name, bundle->name)) {
2538 bundle->name = xstrdup(s->name);
2543 if (!bundle->lacp) {
2544 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2545 bundle->lacp = lacp_create();
2547 lacp_configure(bundle->lacp, s->lacp);
2549 lacp_unref(bundle->lacp);
2550 bundle->lacp = NULL;
2553 /* Update set of ports. */
2555 for (i = 0; i < s->n_slaves; i++) {
2556 if (!bundle_add_port(bundle, s->slaves[i],
2557 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2561 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2562 struct ofport_dpif *next_port;
2564 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2565 for (i = 0; i < s->n_slaves; i++) {
2566 if (s->slaves[i] == port->up.ofp_port) {
2571 bundle_del_port(port);
2575 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2577 if (list_is_empty(&bundle->ports)) {
2578 bundle_destroy(bundle);
2582 /* Set VLAN tagging mode */
2583 if (s->vlan_mode != bundle->vlan_mode
2584 || s->use_priority_tags != bundle->use_priority_tags) {
2585 bundle->vlan_mode = s->vlan_mode;
2586 bundle->use_priority_tags = s->use_priority_tags;
2591 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2592 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2594 if (vlan != bundle->vlan) {
2595 bundle->vlan = vlan;
2599 /* Get trunked VLANs. */
2600 switch (s->vlan_mode) {
2601 case PORT_VLAN_ACCESS:
2605 case PORT_VLAN_TRUNK:
2606 trunks = CONST_CAST(unsigned long *, s->trunks);
2609 case PORT_VLAN_NATIVE_UNTAGGED:
2610 case PORT_VLAN_NATIVE_TAGGED:
2611 if (vlan != 0 && (!s->trunks
2612 || !bitmap_is_set(s->trunks, vlan)
2613 || bitmap_is_set(s->trunks, 0))) {
2614 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2616 trunks = bitmap_clone(s->trunks, 4096);
2618 trunks = bitmap_allocate1(4096);
2620 bitmap_set1(trunks, vlan);
2621 bitmap_set0(trunks, 0);
2623 trunks = CONST_CAST(unsigned long *, s->trunks);
2630 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2631 free(bundle->trunks);
2632 if (trunks == s->trunks) {
2633 bundle->trunks = vlan_bitmap_clone(trunks);
2635 bundle->trunks = trunks;
2640 if (trunks != s->trunks) {
2645 if (!list_is_short(&bundle->ports)) {
2646 bundle->ofproto->has_bonded_bundles = true;
2648 if (bond_reconfigure(bundle->bond, s->bond)) {
2649 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2652 bundle->bond = bond_create(s->bond);
2653 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2656 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2657 bond_slave_register(bundle->bond, port, port->up.netdev);
2660 bond_unref(bundle->bond);
2661 bundle->bond = NULL;
2664 /* If we changed something that would affect MAC learning, un-learn
2665 * everything on this port and force flow revalidation. */
2667 bundle_flush_macs(bundle, false);
2674 bundle_remove(struct ofport *port_)
2676 struct ofport_dpif *port = ofport_dpif_cast(port_);
2677 struct ofbundle *bundle = port->bundle;
2680 bundle_del_port(port);
2681 if (list_is_empty(&bundle->ports)) {
2682 bundle_destroy(bundle);
2683 } else if (list_is_short(&bundle->ports)) {
2684 bond_unref(bundle->bond);
2685 bundle->bond = NULL;
2691 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2694 struct ofport_dpif *port = port_;
2695 uint8_t ea[ETH_ADDR_LEN];
2698 error = netdev_get_etheraddr(port->up.netdev, ea);
2700 struct ofpbuf packet;
2703 ofpbuf_init(&packet, 0);
2704 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2706 memcpy(packet_pdu, pdu, pdu_size);
2708 send_packet(port, &packet);
2709 ofpbuf_uninit(&packet);
2711 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2712 "%s (%s)", port->bundle->name,
2713 netdev_get_name(port->up.netdev), ovs_strerror(error));
2718 bundle_send_learning_packets(struct ofbundle *bundle)
2720 struct ofproto_dpif *ofproto = bundle->ofproto;
2721 int error, n_packets, n_errors;
2722 struct mac_entry *e;
2724 error = n_packets = n_errors = 0;
2725 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2726 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2727 if (e->port.p != bundle) {
2728 struct ofpbuf *learning_packet;
2729 struct ofport_dpif *port;
2733 /* The assignment to "port" is unnecessary but makes "grep"ing for
2734 * struct ofport_dpif more effective. */
2735 learning_packet = bond_compose_learning_packet(bundle->bond,
2739 ret = send_packet(port, learning_packet);
2740 ofpbuf_delete(learning_packet);
2748 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2752 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2753 "packets, last error was: %s",
2754 bundle->name, n_errors, n_packets, ovs_strerror(error));
2756 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2757 bundle->name, n_packets);
2762 bundle_run(struct ofbundle *bundle)
2765 lacp_run(bundle->lacp, send_pdu_cb);
2768 struct ofport_dpif *port;
2770 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2771 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2774 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2775 bundle->ofproto->backer->need_revalidate = REV_BOND;
2778 if (bond_should_send_learning_packets(bundle->bond)) {
2779 bundle_send_learning_packets(bundle);
2785 bundle_wait(struct ofbundle *bundle)
2788 lacp_wait(bundle->lacp);
2791 bond_wait(bundle->bond);
2798 mirror_set__(struct ofproto *ofproto_, void *aux,
2799 const struct ofproto_mirror_settings *s)
2801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2802 struct ofbundle **srcs, **dsts;
2807 mirror_destroy(ofproto->mbridge, aux);
2811 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2812 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2814 for (i = 0; i < s->n_srcs; i++) {
2815 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2818 for (i = 0; i < s->n_dsts; i++) {
2819 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2822 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2823 s->n_dsts, s->src_vlans,
2824 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2831 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2832 uint64_t *packets, uint64_t *bytes)
2835 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2840 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2843 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2844 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2845 mac_learning_flush(ofproto->ml);
2847 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2852 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2855 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2856 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2860 forward_bpdu_changed(struct ofproto *ofproto_)
2862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2863 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2867 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2871 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2872 mac_learning_set_idle_time(ofproto->ml, idle_time);
2873 mac_learning_set_max_entries(ofproto->ml, max_entries);
2874 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2879 static struct ofport_dpif *
2880 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2882 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2883 return ofport ? ofport_dpif_cast(ofport) : NULL;
2886 static struct ofport_dpif *
2887 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2889 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2890 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2894 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2895 struct ofproto_port *ofproto_port,
2896 struct dpif_port *dpif_port)
2898 ofproto_port->name = dpif_port->name;
2899 ofproto_port->type = dpif_port->type;
2900 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2904 ofport_update_peer(struct ofport_dpif *ofport)
2906 const struct ofproto_dpif *ofproto;
2907 struct dpif_backer *backer;
2910 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2914 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2915 backer->need_revalidate = REV_RECONFIGURE;
2918 ofport->peer->peer = NULL;
2919 ofport->peer = NULL;
2922 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2927 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2928 struct ofport *peer_ofport;
2929 struct ofport_dpif *peer;
2932 if (ofproto->backer != backer) {
2936 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2941 peer = ofport_dpif_cast(peer_ofport);
2942 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2943 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2945 ofport->peer = peer;
2946 ofport->peer->peer = ofport;
2956 port_run_fast(struct ofport_dpif *ofport)
2958 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2959 struct ofpbuf packet;
2961 ofpbuf_init(&packet, 0);
2962 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2963 send_packet(ofport, &packet);
2964 ofpbuf_uninit(&packet);
2967 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2968 struct ofpbuf packet;
2970 ofpbuf_init(&packet, 0);
2971 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2972 send_packet(ofport, &packet);
2973 ofpbuf_uninit(&packet);
2978 port_run(struct ofport_dpif *ofport)
2980 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2981 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2982 bool enable = netdev_get_carrier(ofport->up.netdev);
2984 ofport->carrier_seq = carrier_seq;
2986 port_run_fast(ofport);
2989 int cfm_opup = cfm_get_opup(ofport->cfm);
2991 cfm_run(ofport->cfm);
2992 enable = enable && !cfm_get_fault(ofport->cfm);
2994 if (cfm_opup >= 0) {
2995 enable = enable && cfm_opup;
3000 bfd_run(ofport->bfd);
3001 enable = enable && bfd_forwarding(ofport->bfd);
3004 if (ofport->bundle) {
3005 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3006 if (carrier_changed) {
3007 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3011 if (ofport->may_enable != enable) {
3012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3013 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3016 ofport->may_enable = enable;
3020 port_wait(struct ofport_dpif *ofport)
3023 cfm_wait(ofport->cfm);
3027 bfd_wait(ofport->bfd);
3032 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3033 struct ofproto_port *ofproto_port)
3035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3036 struct dpif_port dpif_port;
3039 if (sset_contains(&ofproto->ghost_ports, devname)) {
3040 const char *type = netdev_get_type_from_name(devname);
3042 /* We may be called before ofproto->up.port_by_name is populated with
3043 * the appropriate ofport. For this reason, we must get the name and
3044 * type from the netdev layer directly. */
3046 const struct ofport *ofport;
3048 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3049 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3050 ofproto_port->name = xstrdup(devname);
3051 ofproto_port->type = xstrdup(type);
3057 if (!sset_contains(&ofproto->ports, devname)) {
3060 error = dpif_port_query_by_name(ofproto->backer->dpif,
3061 devname, &dpif_port);
3063 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3069 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3072 const char *devname = netdev_get_name(netdev);
3073 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3074 const char *dp_port_name;
3076 if (netdev_vport_is_patch(netdev)) {
3077 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3081 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3082 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3083 odp_port_t port_no = ODPP_NONE;
3086 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3090 if (netdev_get_tunnel_config(netdev)) {
3091 simap_put(&ofproto->backer->tnl_backers,
3092 dp_port_name, odp_to_u32(port_no));
3096 if (netdev_get_tunnel_config(netdev)) {
3097 sset_add(&ofproto->ghost_ports, devname);
3099 sset_add(&ofproto->ports, devname);
3105 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3108 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3115 sset_find_and_delete(&ofproto->ghost_ports,
3116 netdev_get_name(ofport->up.netdev));
3117 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3118 if (!ofport->is_tunnel) {
3119 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3121 /* The caller is going to close ofport->up.netdev. If this is a
3122 * bonded port, then the bond is using that netdev, so remove it
3123 * from the bond. The client will need to reconfigure everything
3124 * after deleting ports, so then the slave will get re-added. */
3125 bundle_remove(&ofport->up);
3132 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3134 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3139 error = netdev_get_stats(ofport->up.netdev, stats);
3141 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3144 /* ofproto->stats.tx_packets represents packets that we created
3145 * internally and sent to some port (e.g. packets sent with
3146 * send_packet()). Account for them as if they had come from
3147 * OFPP_LOCAL and got forwarded. */
3149 if (stats->rx_packets != UINT64_MAX) {
3150 stats->rx_packets += ofproto->stats.tx_packets;
3153 if (stats->rx_bytes != UINT64_MAX) {
3154 stats->rx_bytes += ofproto->stats.tx_bytes;
3157 /* ofproto->stats.rx_packets represents packets that were received on
3158 * some port and we processed internally and dropped (e.g. STP).
3159 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3161 if (stats->tx_packets != UINT64_MAX) {
3162 stats->tx_packets += ofproto->stats.rx_packets;
3165 if (stats->tx_bytes != UINT64_MAX) {
3166 stats->tx_bytes += ofproto->stats.rx_bytes;
3173 struct port_dump_state {
3178 struct ofproto_port port;
3183 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3185 *statep = xzalloc(sizeof(struct port_dump_state));
3190 port_dump_next(const struct ofproto *ofproto_, void *state_,
3191 struct ofproto_port *port)
3193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3194 struct port_dump_state *state = state_;
3195 const struct sset *sset;
3196 struct sset_node *node;
3198 if (state->has_port) {
3199 ofproto_port_destroy(&state->port);
3200 state->has_port = false;
3202 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3203 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3206 error = port_query_by_name(ofproto_, node->name, &state->port);
3208 *port = state->port;
3209 state->has_port = true;
3211 } else if (error != ENODEV) {
3216 if (!state->ghost) {
3217 state->ghost = true;
3220 return port_dump_next(ofproto_, state_, port);
3227 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3229 struct port_dump_state *state = state_;
3231 if (state->has_port) {
3232 ofproto_port_destroy(&state->port);
3239 port_poll(const struct ofproto *ofproto_, char **devnamep)
3241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3243 if (ofproto->port_poll_errno) {
3244 int error = ofproto->port_poll_errno;
3245 ofproto->port_poll_errno = 0;
3249 if (sset_is_empty(&ofproto->port_poll_set)) {
3253 *devnamep = sset_pop(&ofproto->port_poll_set);
3258 port_poll_wait(const struct ofproto *ofproto_)
3260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3261 dpif_port_poll_wait(ofproto->backer->dpif);
3265 port_is_lacp_current(const struct ofport *ofport_)
3267 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3268 return (ofport->bundle && ofport->bundle->lacp
3269 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3273 /* Upcall handling. */
3275 /* Flow miss batching.
3277 * Some dpifs implement operations faster when you hand them off in a batch.
3278 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3279 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3280 * more packets, plus possibly installing the flow in the dpif.
3282 * So far we only batch the operations that affect flow setup time the most.
3283 * It's possible to batch more than that, but the benefit might be minimal. */
3285 struct hmap_node hmap_node;
3286 struct ofproto_dpif *ofproto;
3288 enum odp_key_fitness key_fitness;
3289 const struct nlattr *key;
3291 struct list packets;
3292 enum dpif_upcall_type upcall_type;
3295 struct flow_miss_op {
3296 struct dpif_op dpif_op;
3298 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3299 struct xlate_out xout;
3300 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3302 struct ofpbuf mask; /* Flow mask for "put" ops. */
3303 struct odputil_keybuf maskbuf;
3305 /* If this is a "put" op, then a pointer to the subfacet that should
3306 * be marked as uninstalled if the operation fails. */
3307 struct subfacet *subfacet;
3310 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3311 * OpenFlow controller as necessary according to their individual
3312 * configurations. */
3314 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3315 const struct flow *flow)
3317 struct ofputil_packet_in pin;
3319 pin.packet = packet->data;
3320 pin.packet_len = packet->size;
3321 pin.reason = OFPR_NO_MATCH;
3322 pin.controller_id = 0;
3327 pin.send_len = 0; /* not used for flow table misses */
3329 flow_get_metadata(flow, &pin.fmd);
3331 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3334 static struct flow_miss *
3335 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3336 const struct flow *flow, uint32_t hash)
3338 struct flow_miss *miss;
3340 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3341 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3349 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3350 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3351 * 'miss' is associated with a subfacet the caller must also initialize the
3352 * returned op->subfacet, and if anything needs to be freed after processing
3353 * the op, the caller must initialize op->garbage also. */
3355 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3356 struct flow_miss_op *op)
3358 if (miss->flow.in_port.ofp_port
3359 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3360 miss->flow.vlan_tci)) {
3361 /* This packet was received on a VLAN splinter port. We
3362 * added a VLAN to the packet to make the packet resemble
3363 * the flow, but the actions were composed assuming that
3364 * the packet contained no VLAN. So, we must remove the
3365 * VLAN header from the packet before trying to execute the
3367 eth_pop_vlan(packet);
3370 op->subfacet = NULL;
3371 op->xout_garbage = false;
3372 op->dpif_op.type = DPIF_OP_EXECUTE;
3373 op->dpif_op.u.execute.key = miss->key;
3374 op->dpif_op.u.execute.key_len = miss->key_len;
3375 op->dpif_op.u.execute.packet = packet;
3376 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3379 /* Helper for handle_flow_miss_without_facet() and
3380 * handle_flow_miss_with_facet(). */
3382 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3383 const struct flow *flow, bool fail_open)
3387 * Extra-special case for fail-open mode.
3389 * We are in fail-open mode and the packet matched the fail-open
3390 * rule, but we are connected to a controller too. We should send
3391 * the packet up to the controller in the hope that it will try to
3392 * set up a flow and thereby allow us to exit fail-open.
3394 * See the top-level comment in fail-open.c for more information.
3396 send_packet_in_miss(ofproto, packet, flow);
3400 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3401 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3402 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3403 * return value of true). However, for short flows the cost of bookkeeping is
3404 * much higher than the benefits, so when the datapath holds a large number of
3405 * flows we impose some heuristics to decide which flows are likely to be worth
3408 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3410 struct dpif_backer *backer = miss->ofproto->backer;
3413 switch (flow_miss_model) {
3414 case OFPROTO_HANDLE_MISS_AUTO:
3416 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3418 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3422 if (!backer->governor) {
3425 n_subfacets = hmap_count(&backer->subfacets);
3426 if (n_subfacets * 2 <= flow_eviction_threshold) {
3430 backer->governor = governor_create();
3433 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3434 return governor_should_install_flow(backer->governor, hash,
3435 list_size(&miss->packets));
3438 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3439 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3440 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3442 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3443 struct flow_miss *miss,
3444 struct flow_miss_op *ops, size_t *n_ops)
3446 struct ofpbuf *packet;
3448 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3450 COVERAGE_INC(facet_suppress);
3452 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3453 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3456 struct xlate_in xin;
3458 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3459 xlate_actions_for_side_effects(&xin);
3462 if (xout->odp_actions.size) {
3463 struct flow_miss_op *op = &ops[*n_ops];
3464 struct dpif_execute *execute = &op->dpif_op.u.execute;
3466 init_flow_miss_execute_op(miss, packet, op);
3467 xlate_out_copy(&op->xout, xout);
3468 execute->actions = op->xout.odp_actions.data;
3469 execute->actions_len = op->xout.odp_actions.size;
3470 op->xout_garbage = true;
3477 /* Handles 'miss', which matches 'facet'. May add any required datapath
3478 * operations to 'ops', incrementing '*n_ops' for each new op.
3480 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3481 * This is really important only for new facets: if we just called time_msec()
3482 * here, then the new subfacet or its packets could look (occasionally) as
3483 * though it was used some time after the facet was used. That can make a
3484 * one-packet flow look like it has a nonzero duration, which looks odd in
3485 * e.g. NetFlow statistics.
3487 * If non-null, 'stats' will be folded into 'facet'. */
3489 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3490 long long int now, struct dpif_flow_stats *stats,
3491 struct flow_miss_op *ops, size_t *n_ops)
3493 enum subfacet_path want_path;
3494 struct subfacet *subfacet;
3495 struct ofpbuf *packet;
3497 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3499 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3500 struct flow_miss_op *op = &ops[*n_ops];
3502 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3505 if (want_path != SF_FAST_PATH) {
3506 struct rule_dpif *rule;
3507 struct xlate_in xin;
3509 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3510 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3511 xlate_actions_for_side_effects(&xin);
3514 if (facet->xout.odp_actions.size) {
3515 struct dpif_execute *execute = &op->dpif_op.u.execute;
3517 init_flow_miss_execute_op(miss, packet, op);
3518 execute->actions = facet->xout.odp_actions.data,
3519 execute->actions_len = facet->xout.odp_actions.size;
3524 /* Don't install the flow if it's the result of the "userspace"
3525 * action for an already installed facet. This can occur when a
3526 * datapath flow with wildcards has a "userspace" action and flows
3527 * sent to userspace result in a different subfacet, which will then
3528 * be rejected as overlapping by the datapath. */
3529 if (miss->upcall_type == DPIF_UC_ACTION
3530 && !list_is_empty(&facet->subfacets)) {
3532 facet->used = MAX(facet->used, stats->used);
3533 facet->packet_count += stats->n_packets;
3534 facet->byte_count += stats->n_bytes;
3535 facet->tcp_flags |= stats->tcp_flags;
3540 subfacet = subfacet_create(facet, miss, now);
3542 subfacet_update_stats(subfacet, stats);
3545 if (subfacet->path != want_path) {
3546 struct flow_miss_op *op = &ops[(*n_ops)++];
3547 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3549 subfacet->path = want_path;
3551 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3552 if (enable_megaflows) {
3553 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3554 &miss->flow, UINT32_MAX);
3557 op->xout_garbage = false;
3558 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3559 op->subfacet = subfacet;
3560 put->flags = DPIF_FP_CREATE;
3561 put->key = miss->key;
3562 put->key_len = miss->key_len;
3563 put->mask = op->mask.data;
3564 put->mask_len = op->mask.size;
3566 if (want_path == SF_FAST_PATH) {
3567 put->actions = facet->xout.odp_actions.data;
3568 put->actions_len = facet->xout.odp_actions.size;
3570 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3571 op->slow_stub, sizeof op->slow_stub,
3572 &put->actions, &put->actions_len);
3578 /* Handles flow miss 'miss'. May add any required datapath operations
3579 * to 'ops', incrementing '*n_ops' for each new op. */
3581 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3584 struct ofproto_dpif *ofproto = miss->ofproto;
3585 struct dpif_flow_stats stats__;
3586 struct dpif_flow_stats *stats = &stats__;
3587 struct ofpbuf *packet;
3588 struct facet *facet;
3592 memset(stats, 0, sizeof *stats);
3594 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3595 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3596 stats->n_bytes += packet->size;
3600 facet = facet_lookup_valid(ofproto, &miss->flow);
3602 struct flow_wildcards wc;
3603 struct rule_dpif *rule;
3604 struct xlate_out xout;
3605 struct xlate_in xin;
3607 flow_wildcards_init_catchall(&wc);
3608 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3609 rule_credit_stats(rule, stats);
3611 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3613 xin.resubmit_stats = stats;
3614 xin.may_learn = true;
3615 xlate_actions(&xin, &xout);
3616 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3618 /* There does not exist a bijection between 'struct flow' and datapath
3619 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3620 * assumption used throughout the facet and subfacet handling code.
3621 * Since we have to handle these misses in userspace anyway, we simply
3622 * skip facet creation, avoiding the problem altogether. */
3623 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3624 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3625 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3629 facet = facet_create(miss, rule, &xout, stats);
3632 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3635 static struct drop_key *
3636 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3639 struct drop_key *drop_key;
3641 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3642 &backer->drop_keys) {
3643 if (drop_key->key_len == key_len
3644 && !memcmp(drop_key->key, key, key_len)) {
3652 drop_key_clear(struct dpif_backer *backer)
3654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3655 struct drop_key *drop_key, *next;
3657 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3660 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3662 if (error && !VLOG_DROP_WARN(&rl)) {
3663 struct ds ds = DS_EMPTY_INITIALIZER;
3664 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3665 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3666 ovs_strerror(error), ds_cstr(&ds));
3670 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3671 free(drop_key->key);
3677 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3680 struct dpif_upcall *upcall;
3681 struct flow_miss *miss;
3682 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3683 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3684 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3694 /* Construct the to-do list.
3696 * This just amounts to extracting the flow from each packet and sticking
3697 * the packets that have the same flow in the same "flow_miss" structure so
3698 * that we can process them together. */
3701 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3702 struct flow_miss *miss = &misses[n_misses];
3703 struct flow_miss *existing_miss;
3704 struct ofproto_dpif *ofproto;
3705 odp_port_t odp_in_port;
3710 error = xlate_receive(backer, upcall->packet, upcall->key,
3711 upcall->key_len, &flow, &miss->key_fitness,
3712 &ofproto, &odp_in_port);
3713 if (error == ENODEV) {
3714 struct drop_key *drop_key;
3716 /* Received packet on datapath port for which we couldn't
3717 * associate an ofproto. This can happen if a port is removed
3718 * while traffic is being received. Print a rate-limited message
3719 * in case it happens frequently. Install a drop flow so
3720 * that future packets of the flow are inexpensively dropped
3722 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3723 "%"PRIu32, odp_in_port);
3725 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3728 ret = dpif_flow_put(backer->dpif,
3729 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3730 upcall->key, upcall->key_len,
3731 NULL, 0, NULL, 0, NULL);
3734 drop_key = xmalloc(sizeof *drop_key);
3735 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3736 drop_key->key_len = upcall->key_len;
3738 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3739 hash_bytes(drop_key->key, drop_key->key_len, 0));
3748 ofproto->n_missed++;
3749 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3750 &flow.tunnel, &flow.in_port, &miss->flow);
3752 /* Add other packets to a to-do list. */
3753 hash = flow_hash(&miss->flow, 0);
3754 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3755 if (!existing_miss) {
3756 hmap_insert(&todo, &miss->hmap_node, hash);
3757 miss->ofproto = ofproto;
3758 miss->key = upcall->key;
3759 miss->key_len = upcall->key_len;
3760 miss->upcall_type = upcall->type;
3761 list_init(&miss->packets);
3765 miss = existing_miss;
3767 list_push_back(&miss->packets, &upcall->packet->list_node);
3770 /* Process each element in the to-do list, constructing the set of
3771 * operations to batch. */
3773 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3774 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3776 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3778 /* Execute batch. */
3779 for (i = 0; i < n_ops; i++) {
3780 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3782 dpif_operate(backer->dpif, dpif_ops, n_ops);
3784 for (i = 0; i < n_ops; i++) {
3785 if (dpif_ops[i]->error != 0
3786 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3787 && flow_miss_ops[i].subfacet) {
3788 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3790 COVERAGE_INC(subfacet_install_fail);
3792 /* Zero-out subfacet counters when installation failed, but
3793 * datapath reported hits. This should not happen and
3794 * indicates a bug, since if the datapath flow exists, we
3795 * should not be attempting to create a new subfacet. A
3796 * buggy datapath could trigger this, so just zero out the
3797 * counters and log an error. */
3798 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3799 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3800 "datapath reported hits");
3801 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3804 subfacet->path = SF_NOT_INSTALLED;
3808 if (flow_miss_ops[i].xout_garbage) {
3809 xlate_out_uninit(&flow_miss_ops[i].xout);
3812 hmap_destroy(&todo);
3815 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3817 classify_upcall(const struct dpif_upcall *upcall)
3819 size_t userdata_len;
3820 union user_action_cookie cookie;
3822 /* First look at the upcall type. */
3823 switch (upcall->type) {
3824 case DPIF_UC_ACTION:
3830 case DPIF_N_UC_TYPES:
3832 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3836 /* "action" upcalls need a closer look. */
3837 if (!upcall->userdata) {
3838 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3841 userdata_len = nl_attr_get_size(upcall->userdata);
3842 if (userdata_len < sizeof cookie.type
3843 || userdata_len > sizeof cookie) {
3844 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3848 memset(&cookie, 0, sizeof cookie);
3849 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3850 if (userdata_len == sizeof cookie.sflow
3851 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3852 return SFLOW_UPCALL;
3853 } else if (userdata_len == sizeof cookie.slow_path
3854 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3856 } else if (userdata_len == sizeof cookie.flow_sample
3857 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3858 return FLOW_SAMPLE_UPCALL;
3859 } else if (userdata_len == sizeof cookie.ipfix
3860 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3861 return IPFIX_UPCALL;
3863 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3864 " and size %zu", cookie.type, userdata_len);
3870 handle_sflow_upcall(struct dpif_backer *backer,
3871 const struct dpif_upcall *upcall)
3873 struct ofproto_dpif *ofproto;
3874 union user_action_cookie cookie;
3876 odp_port_t odp_in_port;
3878 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3879 &flow, NULL, &ofproto, &odp_in_port)
3880 || !ofproto->sflow) {
3884 memset(&cookie, 0, sizeof cookie);
3885 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3886 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3887 odp_in_port, &cookie);
3891 handle_flow_sample_upcall(struct dpif_backer *backer,
3892 const struct dpif_upcall *upcall)
3894 struct ofproto_dpif *ofproto;
3895 union user_action_cookie cookie;
3898 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3899 &flow, NULL, &ofproto, NULL)
3900 || !ofproto->ipfix) {
3904 memset(&cookie, 0, sizeof cookie);
3905 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3907 /* The flow reflects exactly the contents of the packet. Sample
3908 * the packet using it. */
3909 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3910 cookie.flow_sample.collector_set_id,
3911 cookie.flow_sample.probability,
3912 cookie.flow_sample.obs_domain_id,
3913 cookie.flow_sample.obs_point_id);
3917 handle_ipfix_upcall(struct dpif_backer *backer,
3918 const struct dpif_upcall *upcall)
3920 struct ofproto_dpif *ofproto;
3923 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3924 &flow, NULL, &ofproto, NULL)
3925 || !ofproto->ipfix) {
3929 /* The flow reflects exactly the contents of the packet. Sample
3930 * the packet using it. */
3931 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3935 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3937 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3938 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3939 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3944 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3947 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3948 struct dpif_upcall *upcall = &misses[n_misses];
3949 struct ofpbuf *buf = &miss_bufs[n_misses];
3952 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3953 sizeof miss_buf_stubs[n_misses]);
3954 error = dpif_recv(backer->dpif, upcall, buf);
3960 switch (classify_upcall(upcall)) {
3962 /* Handle it later. */
3967 handle_sflow_upcall(backer, upcall);
3971 case FLOW_SAMPLE_UPCALL:
3972 handle_flow_sample_upcall(backer, upcall);
3977 handle_ipfix_upcall(backer, upcall);
3987 /* Handle deferred MISS_UPCALL processing. */
3988 handle_miss_upcalls(backer, misses, n_misses);
3989 for (i = 0; i < n_misses; i++) {
3990 ofpbuf_uninit(&miss_bufs[i]);
3996 /* Flow expiration. */
3998 static int subfacet_max_idle(const struct dpif_backer *);
3999 static void update_stats(struct dpif_backer *);
4000 static void rule_expire(struct rule_dpif *);
4001 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4003 /* This function is called periodically by run(). Its job is to collect
4004 * updates for the flows that have been installed into the datapath, most
4005 * importantly when they last were used, and then use that information to
4006 * expire flows that have not been used recently.
4008 * Returns the number of milliseconds after which it should be called again. */
4010 expire(struct dpif_backer *backer)
4012 struct ofproto_dpif *ofproto;
4016 /* Periodically clear out the drop keys in an effort to keep them
4017 * relatively few. */
4018 drop_key_clear(backer);
4020 /* Update stats for each flow in the backer. */
4021 update_stats(backer);
4023 n_subfacets = hmap_count(&backer->subfacets);
4025 struct subfacet *subfacet;
4026 long long int total, now;
4030 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4031 total += now - subfacet->created;
4033 backer->avg_subfacet_life += total / n_subfacets;
4035 backer->avg_subfacet_life /= 2;
4037 backer->avg_n_subfacet += n_subfacets;
4038 backer->avg_n_subfacet /= 2;
4040 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4042 max_idle = subfacet_max_idle(backer);
4043 expire_subfacets(backer, max_idle);
4045 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4046 struct rule *rule, *next_rule;
4048 if (ofproto->backer != backer) {
4052 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4054 ovs_mutex_lock(&ofproto->up.expirable_mutex);
4055 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4056 &ofproto->up.expirable) {
4057 rule_expire(rule_dpif_cast(rule));
4059 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
4061 /* All outstanding data in existing flows has been accounted, so it's a
4062 * good time to do bond rebalancing. */
4063 if (ofproto->has_bonded_bundles) {
4064 struct ofbundle *bundle;
4066 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4068 bond_rebalance(bundle->bond);
4074 return MIN(max_idle, 1000);
4077 /* Updates flow table statistics given that the datapath just reported 'stats'
4078 * as 'subfacet''s statistics. */
4080 update_subfacet_stats(struct subfacet *subfacet,
4081 const struct dpif_flow_stats *stats)
4083 struct facet *facet = subfacet->facet;
4084 struct dpif_flow_stats diff;
4086 diff.tcp_flags = stats->tcp_flags;
4087 diff.used = stats->used;
4089 if (stats->n_packets >= subfacet->dp_packet_count) {
4090 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4092 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4096 if (stats->n_bytes >= subfacet->dp_byte_count) {
4097 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4099 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4103 facet->ofproto->n_hit += diff.n_packets;
4104 subfacet->dp_packet_count = stats->n_packets;
4105 subfacet->dp_byte_count = stats->n_bytes;
4106 subfacet_update_stats(subfacet, &diff);
4108 if (facet->accounted_bytes < facet->byte_count) {
4110 facet_account(facet);
4111 facet->accounted_bytes = facet->byte_count;
4115 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4116 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4118 delete_unexpected_flow(struct dpif_backer *backer,
4119 const struct nlattr *key, size_t key_len)
4121 if (!VLOG_DROP_WARN(&rl)) {
4125 odp_flow_key_format(key, key_len, &s);
4126 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4130 COVERAGE_INC(facet_unexpected);
4131 dpif_flow_del(backer->dpif, key, key_len, NULL);
4134 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4136 * This function also pushes statistics updates to rules which each facet
4137 * resubmits into. Generally these statistics will be accurate. However, if a
4138 * facet changes the rule it resubmits into at some time in between
4139 * update_stats() runs, it is possible that statistics accrued to the
4140 * old rule will be incorrectly attributed to the new rule. This could be
4141 * avoided by calling update_stats() whenever rules are created or
4142 * deleted. However, the performance impact of making so many calls to the
4143 * datapath do not justify the benefit of having perfectly accurate statistics.
4145 * In addition, this function maintains per ofproto flow hit counts. The patch
4146 * port is not treated specially. e.g. A packet ingress from br0 patched into
4147 * br1 will increase the hit count of br0 by 1, however, does not affect
4148 * the hit or miss counts of br1.
4151 update_stats(struct dpif_backer *backer)
4153 const struct dpif_flow_stats *stats;
4154 struct dpif_flow_dump dump;
4155 const struct nlattr *key, *mask;
4156 size_t key_len, mask_len;
4158 dpif_flow_dump_start(&dump, backer->dpif);
4159 while (dpif_flow_dump_next(&dump, &key, &key_len,
4160 &mask, &mask_len, NULL, NULL, &stats)) {
4161 struct subfacet *subfacet;
4164 key_hash = odp_flow_key_hash(key, key_len);
4165 subfacet = subfacet_find(backer, key, key_len, key_hash);
4166 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4168 update_subfacet_stats(subfacet, stats);
4172 /* Stats are updated per-packet. */
4175 case SF_NOT_INSTALLED:
4177 delete_unexpected_flow(backer, key, key_len);
4182 dpif_flow_dump_done(&dump);
4184 update_moving_averages(backer);
4187 /* Calculates and returns the number of milliseconds of idle time after which
4188 * subfacets should expire from the datapath. When a subfacet expires, we fold
4189 * its statistics into its facet, and when a facet's last subfacet expires, we
4190 * fold its statistic into its rule. */
4192 subfacet_max_idle(const struct dpif_backer *backer)
4195 * Idle time histogram.
4197 * Most of the time a switch has a relatively small number of subfacets.
4198 * When this is the case we might as well keep statistics for all of them
4199 * in userspace and to cache them in the kernel datapath for performance as
4202 * As the number of subfacets increases, the memory required to maintain
4203 * statistics about them in userspace and in the kernel becomes
4204 * significant. However, with a large number of subfacets it is likely
4205 * that only a few of them are "heavy hitters" that consume a large amount
4206 * of bandwidth. At this point, only heavy hitters are worth caching in
4207 * the kernel and maintaining in userspaces; other subfacets we can
4210 * The technique used to compute the idle time is to build a histogram with
4211 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4212 * that is installed in the kernel gets dropped in the appropriate bucket.
4213 * After the histogram has been built, we compute the cutoff so that only
4214 * the most-recently-used 1% of subfacets (but at least
4215 * flow_eviction_threshold flows) are kept cached. At least
4216 * the most-recently-used bucket of subfacets is kept, so actually an
4217 * arbitrary number of subfacets can be kept in any given expiration run
4218 * (though the next run will delete most of those unless they receive
4221 * This requires a second pass through the subfacets, in addition to the
4222 * pass made by update_stats(), because the former function never looks at
4223 * uninstallable subfacets.
4225 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4226 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4227 int buckets[N_BUCKETS] = { 0 };
4228 int total, subtotal, bucket;
4229 struct subfacet *subfacet;
4233 total = hmap_count(&backer->subfacets);
4234 if (total <= flow_eviction_threshold) {
4235 return N_BUCKETS * BUCKET_WIDTH;
4238 /* Build histogram. */
4240 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4241 long long int idle = now - subfacet->used;
4242 int bucket = (idle <= 0 ? 0
4243 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4244 : (unsigned int) idle / BUCKET_WIDTH);
4248 /* Find the first bucket whose flows should be expired. */
4249 subtotal = bucket = 0;
4251 subtotal += buckets[bucket++];
4252 } while (bucket < N_BUCKETS &&
4253 subtotal < MAX(flow_eviction_threshold, total / 100));
4255 if (VLOG_IS_DBG_ENABLED()) {
4259 ds_put_cstr(&s, "keep");
4260 for (i = 0; i < N_BUCKETS; i++) {
4262 ds_put_cstr(&s, ", drop");
4265 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4268 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4272 return bucket * BUCKET_WIDTH;
4276 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4278 /* Cutoff time for most flows. */
4279 long long int normal_cutoff = time_msec() - dp_max_idle;
4281 /* We really want to keep flows for special protocols around, so use a more
4282 * conservative cutoff. */
4283 long long int special_cutoff = time_msec() - 10000;
4285 struct subfacet *subfacet, *next_subfacet;
4286 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4290 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4291 &backer->subfacets) {
4292 long long int cutoff;
4294 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4298 if (subfacet->used < cutoff) {
4299 if (subfacet->path != SF_NOT_INSTALLED) {
4300 batch[n_batch++] = subfacet;
4301 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4302 subfacet_destroy_batch(backer, batch, n_batch);
4306 subfacet_destroy(subfacet);
4312 subfacet_destroy_batch(backer, batch, n_batch);
4316 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4317 * then delete it entirely. */
4319 rule_expire(struct rule_dpif *rule)
4321 uint16_t idle_timeout, hard_timeout;
4325 if (rule->up.pending) {
4326 /* We'll have to expire it later. */
4330 ovs_mutex_lock(&rule->up.timeout_mutex);
4331 hard_timeout = rule->up.hard_timeout;
4332 idle_timeout = rule->up.idle_timeout;
4333 ovs_mutex_unlock(&rule->up.timeout_mutex);
4335 /* Has 'rule' expired? */
4337 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
4338 reason = OFPRR_HARD_TIMEOUT;
4339 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
4340 reason = OFPRR_IDLE_TIMEOUT;
4345 COVERAGE_INC(ofproto_dpif_expired);
4347 /* Get rid of the rule. */
4348 ofproto_rule_expire(&rule->up, reason);
4353 /* Creates and returns a new facet based on 'miss'.
4355 * The caller must already have determined that no facet with an identical
4356 * 'miss->flow' exists in 'miss->ofproto'.
4358 * 'rule' and 'xout' must have been created based on 'miss'.
4360 * 'facet'' statistics are initialized based on 'stats'.
4362 * The facet will initially have no subfacets. The caller should create (at
4363 * least) one subfacet with subfacet_create(). */
4364 static struct facet *
4365 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4366 struct xlate_out *xout, struct dpif_flow_stats *stats)
4368 struct ofproto_dpif *ofproto = miss->ofproto;
4369 struct facet *facet;
4372 facet = xzalloc(sizeof *facet);
4373 facet->ofproto = miss->ofproto;
4374 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4375 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4376 facet->tcp_flags = stats->tcp_flags;
4377 facet->used = stats->used;
4378 facet->flow = miss->flow;
4379 facet->learn_rl = time_msec() + 500;
4381 list_init(&facet->subfacets);
4382 netflow_flow_init(&facet->nf_flow);
4383 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4385 xlate_out_copy(&facet->xout, xout);
4387 match_init(&match, &facet->flow, &facet->xout.wc);
4388 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4389 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4390 classifier_insert(&ofproto->facets, &facet->cr);
4391 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4393 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4394 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4400 facet_free(struct facet *facet)
4403 xlate_out_uninit(&facet->xout);
4408 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4409 * 'packet', which arrived on 'in_port'. */
4411 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4412 const struct nlattr *odp_actions, size_t actions_len,
4413 struct ofpbuf *packet)
4415 struct odputil_keybuf keybuf;
4419 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4420 odp_flow_key_from_flow(&key, flow,
4421 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4423 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4424 odp_actions, actions_len, packet);
4428 /* Remove 'facet' from its ofproto and free up the associated memory:
4430 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4431 * rule's statistics, via subfacet_uninstall().
4433 * - Removes 'facet' from its rule and from ofproto->facets.
4436 facet_remove(struct facet *facet)
4438 struct subfacet *subfacet, *next_subfacet;
4440 ovs_assert(!list_is_empty(&facet->subfacets));
4442 /* First uninstall all of the subfacets to get final statistics. */
4443 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4444 subfacet_uninstall(subfacet);
4447 /* Flush the final stats to the rule.
4449 * This might require us to have at least one subfacet around so that we
4450 * can use its actions for accounting in facet_account(), which is why we
4451 * have uninstalled but not yet destroyed the subfacets. */
4452 facet_flush_stats(facet);
4454 /* Now we're really all done so destroy everything. */
4455 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4456 &facet->subfacets) {
4457 subfacet_destroy__(subfacet);
4459 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4460 classifier_remove(&facet->ofproto->facets, &facet->cr);
4461 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4462 cls_rule_destroy(&facet->cr);
4466 /* Feed information from 'facet' back into the learning table to keep it in
4467 * sync with what is actually flowing through the datapath. */
4469 facet_learn(struct facet *facet)
4471 long long int now = time_msec();
4473 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4477 facet->learn_rl = now + 500;
4479 if (!facet->xout.has_learn
4480 && !facet->xout.has_normal
4481 && (!facet->xout.has_fin_timeout
4482 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4486 facet_push_stats(facet, true);
4490 facet_account(struct facet *facet)
4492 const struct nlattr *a;
4497 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4500 n_bytes = facet->byte_count - facet->accounted_bytes;
4502 /* This loop feeds byte counters to bond_account() for rebalancing to use
4503 * as a basis. We also need to track the actual VLAN on which the packet
4504 * is going to be sent to ensure that it matches the one passed to
4505 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4508 * We use the actions from an arbitrary subfacet because they should all
4509 * be equally valid for our purpose. */
4510 vlan_tci = facet->flow.vlan_tci;
4511 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4512 facet->xout.odp_actions.size) {
4513 const struct ovs_action_push_vlan *vlan;
4514 struct ofport_dpif *port;
4516 switch (nl_attr_type(a)) {
4517 case OVS_ACTION_ATTR_OUTPUT:
4518 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4519 if (port && port->bundle && port->bundle->bond) {
4520 bond_account(port->bundle->bond, &facet->flow,
4521 vlan_tci_to_vid(vlan_tci), n_bytes);
4525 case OVS_ACTION_ATTR_POP_VLAN:
4526 vlan_tci = htons(0);
4529 case OVS_ACTION_ATTR_PUSH_VLAN:
4530 vlan = nl_attr_get(a);
4531 vlan_tci = vlan->vlan_tci;
4537 /* Returns true if the only action for 'facet' is to send to the controller.
4538 * (We don't report NetFlow expiration messages for such facets because they
4539 * are just part of the control logic for the network, not real traffic). */
4541 facet_is_controller_flow(struct facet *facet)
4544 struct ofproto_dpif *ofproto = facet->ofproto;
4545 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4547 const struct ofpact *ofpacts = rule->up.ofpacts;
4548 size_t ofpacts_len = rule->up.ofpacts_len;
4550 if (ofpacts_len > 0 &&
4551 ofpacts->type == OFPACT_CONTROLLER &&
4552 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4559 /* Folds all of 'facet''s statistics into its rule. Also updates the
4560 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4561 * 'facet''s statistics in the datapath should have been zeroed and folded into
4562 * its packet and byte counts before this function is called. */
4564 facet_flush_stats(struct facet *facet)
4566 struct ofproto_dpif *ofproto = facet->ofproto;
4567 struct subfacet *subfacet;
4569 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4570 ovs_assert(!subfacet->dp_byte_count);
4571 ovs_assert(!subfacet->dp_packet_count);
4574 facet_push_stats(facet, false);
4575 if (facet->accounted_bytes < facet->byte_count) {
4576 facet_account(facet);
4577 facet->accounted_bytes = facet->byte_count;
4580 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4581 struct ofexpired expired;
4582 expired.flow = facet->flow;
4583 expired.packet_count = facet->packet_count;
4584 expired.byte_count = facet->byte_count;
4585 expired.used = facet->used;
4586 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4589 /* Reset counters to prevent double counting if 'facet' ever gets
4591 facet_reset_counters(facet);
4593 netflow_flow_clear(&facet->nf_flow);
4594 facet->tcp_flags = 0;
4597 /* Searches 'ofproto''s table of facets for one which would be responsible for
4598 * 'flow'. Returns it if found, otherwise a null pointer.
4600 * The returned facet might need revalidation; use facet_lookup_valid()
4601 * instead if that is important. */
4602 static struct facet *
4603 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4605 struct cls_rule *cr;
4607 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4608 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4609 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4610 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4613 /* Searches 'ofproto''s table of facets for one capable that covers
4614 * 'flow'. Returns it if found, otherwise a null pointer.
4616 * The returned facet is guaranteed to be valid. */
4617 static struct facet *
4618 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4620 struct facet *facet;
4622 facet = facet_find(ofproto, flow);
4624 && ofproto->backer->need_revalidate
4625 && !facet_revalidate(facet)) {
4633 facet_check_consistency(struct facet *facet)
4635 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4637 struct xlate_out xout;
4638 struct xlate_in xin;
4640 struct rule_dpif *rule;
4643 /* Check the datapath actions for consistency. */
4644 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4645 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4646 xlate_actions(&xin, &xout);
4648 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4649 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4650 && facet->xout.slow == xout.slow
4651 && facet->fail_open == fail_open;
4652 if (!ok && !VLOG_DROP_WARN(&rl)) {
4653 struct ds s = DS_EMPTY_INITIALIZER;
4655 flow_format(&s, &facet->flow);
4656 ds_put_cstr(&s, ": inconsistency in facet");
4658 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4659 ds_put_cstr(&s, " (actions were: ");
4660 format_odp_actions(&s, facet->xout.odp_actions.data,
4661 facet->xout.odp_actions.size);
4662 ds_put_cstr(&s, ") (correct actions: ");
4663 format_odp_actions(&s, xout.odp_actions.data,
4664 xout.odp_actions.size);
4665 ds_put_char(&s, ')');
4668 if (facet->xout.slow != xout.slow) {
4669 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4672 if (facet->fail_open != fail_open) {
4673 ds_put_format(&s, " fail open incorrect. should be %s",
4674 fail_open ? "true" : "false");
4678 xlate_out_uninit(&xout);
4683 /* Re-searches the classifier for 'facet':
4685 * - If the rule found is different from 'facet''s current rule, moves
4686 * 'facet' to the new rule and recompiles its actions.
4688 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4689 * where it is and recompiles its actions anyway.
4691 * - If any of 'facet''s subfacets correspond to a new flow according to
4692 * xlate_receive(), 'facet' is removed.
4694 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4696 facet_revalidate(struct facet *facet)
4698 struct ofproto_dpif *ofproto = facet->ofproto;
4699 struct rule_dpif *new_rule;
4700 struct subfacet *subfacet;
4701 struct flow_wildcards wc;
4702 struct xlate_out xout;
4703 struct xlate_in xin;
4705 COVERAGE_INC(facet_revalidate);
4707 /* Check that child subfacets still correspond to this facet. Tunnel
4708 * configuration changes could cause a subfacet's OpenFlow in_port to
4710 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4711 struct ofproto_dpif *recv_ofproto;
4712 struct flow recv_flow;
4715 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4716 subfacet->key_len, &recv_flow, NULL,
4717 &recv_ofproto, NULL);
4719 || recv_ofproto != ofproto
4720 || facet != facet_find(ofproto, &recv_flow)) {
4721 facet_remove(facet);
4726 flow_wildcards_init_catchall(&wc);
4727 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4729 /* Calculate new datapath actions.
4731 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4732 * emit a NetFlow expiration and, if so, we need to have the old state
4733 * around to properly compose it. */
4734 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4735 xlate_actions(&xin, &xout);
4736 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4738 /* A facet's slow path reason should only change under dramatic
4739 * circumstances. Rather than try to update everything, it's simpler to
4740 * remove the facet and start over.
4742 * More importantly, if a facet's wildcards change, it will be relatively
4743 * difficult to figure out if its subfacets still belong to it, and if not
4744 * which facet they may belong to. Again, to avoid the complexity, we
4745 * simply give up instead. */
4746 if (facet->xout.slow != xout.slow
4747 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4748 facet_remove(facet);
4749 xlate_out_uninit(&xout);
4753 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4754 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4755 if (subfacet->path == SF_FAST_PATH) {
4756 struct dpif_flow_stats stats;
4758 subfacet_install(subfacet, &xout.odp_actions, &stats);
4759 subfacet_update_stats(subfacet, &stats);
4763 facet_flush_stats(facet);
4765 ofpbuf_clear(&facet->xout.odp_actions);
4766 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4767 xout.odp_actions.size);
4770 /* Update 'facet' now that we've taken care of all the old state. */
4771 facet->xout.slow = xout.slow;
4772 facet->xout.has_learn = xout.has_learn;
4773 facet->xout.has_normal = xout.has_normal;
4774 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4775 facet->xout.nf_output_iface = xout.nf_output_iface;
4776 facet->xout.mirrors = xout.mirrors;
4777 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4778 facet->used = MAX(facet->used, new_rule->up.created);
4779 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4781 xlate_out_uninit(&xout);
4786 facet_reset_counters(struct facet *facet)
4788 facet->packet_count = 0;
4789 facet->byte_count = 0;
4790 facet->prev_packet_count = 0;
4791 facet->prev_byte_count = 0;
4792 facet->accounted_bytes = 0;
4796 facet_push_stats(struct facet *facet, bool may_learn)
4798 struct dpif_flow_stats stats;
4800 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4801 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4802 ovs_assert(facet->used >= facet->prev_used);
4804 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4805 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4806 stats.used = facet->used;
4807 stats.tcp_flags = facet->tcp_flags;
4809 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4810 struct ofproto_dpif *ofproto = facet->ofproto;
4811 struct ofport_dpif *in_port;
4812 struct rule_dpif *rule;
4813 struct xlate_in xin;
4815 facet->prev_packet_count = facet->packet_count;
4816 facet->prev_byte_count = facet->byte_count;
4817 facet->prev_used = facet->used;
4819 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4820 if (in_port && in_port->is_tunnel) {
4821 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4824 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4825 rule_credit_stats(rule, &stats);
4826 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4828 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4829 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4830 stats.n_packets, stats.n_bytes);
4832 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4834 xin.resubmit_stats = &stats;
4835 xin.may_learn = may_learn;
4836 xlate_actions_for_side_effects(&xin);
4841 push_all_stats__(bool run_fast)
4843 static long long int rl = LLONG_MIN;
4844 struct ofproto_dpif *ofproto;
4846 if (time_msec() < rl) {
4850 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4851 struct cls_cursor cursor;
4852 struct facet *facet;
4854 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4855 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4856 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4857 facet_push_stats(facet, false);
4862 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4865 rl = time_msec() + 100;
4869 push_all_stats(void)
4871 push_all_stats__(true);
4875 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4877 ovs_mutex_lock(&rule->stats_mutex);
4878 rule->packet_count += stats->n_packets;
4879 rule->byte_count += stats->n_bytes;
4880 ofproto_rule_update_used(&rule->up, stats->used);
4881 ovs_mutex_unlock(&rule->stats_mutex);
4886 static struct subfacet *
4887 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4888 size_t key_len, uint32_t key_hash)
4890 struct subfacet *subfacet;
4892 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4893 &backer->subfacets) {
4894 if (subfacet->key_len == key_len
4895 && !memcmp(key, subfacet->key, key_len)) {
4903 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4904 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4905 * existing subfacet if there is one, otherwise creates and returns a
4907 static struct subfacet *
4908 subfacet_create(struct facet *facet, struct flow_miss *miss,
4911 struct dpif_backer *backer = miss->ofproto->backer;
4912 enum odp_key_fitness key_fitness = miss->key_fitness;
4913 const struct nlattr *key = miss->key;
4914 size_t key_len = miss->key_len;
4916 struct subfacet *subfacet;
4918 key_hash = odp_flow_key_hash(key, key_len);
4920 if (list_is_empty(&facet->subfacets)) {
4921 subfacet = &facet->one_subfacet;
4923 subfacet = subfacet_find(backer, key, key_len, key_hash);
4925 if (subfacet->facet == facet) {
4929 /* This shouldn't happen. */
4930 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4931 subfacet_destroy(subfacet);
4934 subfacet = xmalloc(sizeof *subfacet);
4937 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4938 list_push_back(&facet->subfacets, &subfacet->list_node);
4939 subfacet->facet = facet;
4940 subfacet->key_fitness = key_fitness;
4941 subfacet->key = xmemdup(key, key_len);
4942 subfacet->key_len = key_len;
4943 subfacet->used = now;
4944 subfacet->created = now;
4945 subfacet->dp_packet_count = 0;
4946 subfacet->dp_byte_count = 0;
4947 subfacet->path = SF_NOT_INSTALLED;
4948 subfacet->backer = backer;
4950 backer->subfacet_add_count++;
4954 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4955 * its facet within 'ofproto', and frees it. */
4957 subfacet_destroy__(struct subfacet *subfacet)
4959 struct facet *facet = subfacet->facet;
4960 struct ofproto_dpif *ofproto = facet->ofproto;
4962 /* Update ofproto stats before uninstall the subfacet. */
4963 ofproto->backer->subfacet_del_count++;
4965 subfacet_uninstall(subfacet);
4966 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4967 list_remove(&subfacet->list_node);
4968 free(subfacet->key);
4969 if (subfacet != &facet->one_subfacet) {
4974 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4975 * last remaining subfacet in its facet destroys the facet too. */
4977 subfacet_destroy(struct subfacet *subfacet)
4979 struct facet *facet = subfacet->facet;
4981 if (list_is_singleton(&facet->subfacets)) {
4982 /* facet_remove() needs at least one subfacet (it will remove it). */
4983 facet_remove(facet);
4985 subfacet_destroy__(subfacet);
4990 subfacet_destroy_batch(struct dpif_backer *backer,
4991 struct subfacet **subfacets, int n)
4993 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4994 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4995 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4998 for (i = 0; i < n; i++) {
4999 ops[i].type = DPIF_OP_FLOW_DEL;
5000 ops[i].u.flow_del.key = subfacets[i]->key;
5001 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5002 ops[i].u.flow_del.stats = &stats[i];
5006 dpif_operate(backer->dpif, opsp, n);
5007 for (i = 0; i < n; i++) {
5008 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5009 subfacets[i]->path = SF_NOT_INSTALLED;
5010 subfacet_destroy(subfacets[i]);
5015 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5016 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5017 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5018 * since 'subfacet' was last updated.
5020 * Returns 0 if successful, otherwise a positive errno value. */
5022 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5023 struct dpif_flow_stats *stats)
5025 struct facet *facet = subfacet->facet;
5026 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5027 const struct nlattr *actions = odp_actions->data;
5028 size_t actions_len = odp_actions->size;
5029 struct odputil_keybuf maskbuf;
5032 uint64_t slow_path_stub[128 / 8];
5033 enum dpif_flow_put_flags flags;
5036 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5039 flags |= DPIF_FP_ZERO_STATS;
5042 if (path == SF_SLOW_PATH) {
5043 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5044 slow_path_stub, sizeof slow_path_stub,
5045 &actions, &actions_len);
5048 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5049 if (enable_megaflows) {
5050 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5051 &facet->flow, UINT32_MAX);
5054 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5055 subfacet->key_len, mask.data, mask.size,
5056 actions, actions_len, stats);
5059 subfacet_reset_dp_stats(subfacet, stats);
5063 COVERAGE_INC(subfacet_install_fail);
5065 subfacet->path = path;
5070 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5072 subfacet_uninstall(struct subfacet *subfacet)
5074 if (subfacet->path != SF_NOT_INSTALLED) {
5075 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5076 struct dpif_flow_stats stats;
5079 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5080 subfacet->key_len, &stats);
5081 subfacet_reset_dp_stats(subfacet, &stats);
5083 subfacet_update_stats(subfacet, &stats);
5085 subfacet->path = SF_NOT_INSTALLED;
5087 ovs_assert(subfacet->dp_packet_count == 0);
5088 ovs_assert(subfacet->dp_byte_count == 0);
5092 /* Resets 'subfacet''s datapath statistics counters. This should be called
5093 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5094 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5095 * was reset in the datapath. 'stats' will be modified to include only
5096 * statistics new since 'subfacet' was last updated. */
5098 subfacet_reset_dp_stats(struct subfacet *subfacet,
5099 struct dpif_flow_stats *stats)
5102 && subfacet->dp_packet_count <= stats->n_packets
5103 && subfacet->dp_byte_count <= stats->n_bytes) {
5104 stats->n_packets -= subfacet->dp_packet_count;
5105 stats->n_bytes -= subfacet->dp_byte_count;
5108 subfacet->dp_packet_count = 0;
5109 subfacet->dp_byte_count = 0;
5112 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5114 * Because of the meaning of a subfacet's counters, it only makes sense to do
5115 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5116 * represents a packet that was sent by hand or if it represents statistics
5117 * that have been cleared out of the datapath. */
5119 subfacet_update_stats(struct subfacet *subfacet,
5120 const struct dpif_flow_stats *stats)
5122 if (stats->n_packets || stats->used > subfacet->used) {
5123 struct facet *facet = subfacet->facet;
5125 subfacet->used = MAX(subfacet->used, stats->used);
5126 facet->used = MAX(facet->used, stats->used);
5127 facet->packet_count += stats->n_packets;
5128 facet->byte_count += stats->n_bytes;
5129 facet->tcp_flags |= stats->tcp_flags;
5135 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5136 * the fields that were relevant as part of the lookup. */
5137 static struct rule_dpif *
5138 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5139 struct flow_wildcards *wc)
5141 struct ofport_dpif *port;
5142 struct rule_dpif *rule;
5144 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5148 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5150 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5151 flow->in_port.ofp_port);
5154 return choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
5155 ofproto->no_packet_in_rule);
5159 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5160 const struct flow *flow, struct flow_wildcards *wc,
5163 struct cls_rule *cls_rule;
5164 struct classifier *cls;
5167 if (table_id >= N_TABLES) {
5172 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5173 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5176 cls = &ofproto->up.tables[table_id].cls;
5177 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5178 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5179 /* We must pretend that transport ports are unavailable. */
5180 struct flow ofpc_normal_flow = *flow;
5181 ofpc_normal_flow.tp_src = htons(0);
5182 ofpc_normal_flow.tp_dst = htons(0);
5183 ovs_rwlock_rdlock(&cls->rwlock);
5184 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5185 ovs_rwlock_unlock(&cls->rwlock);
5186 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5187 cls_rule = &ofproto->drop_frags_rule->up.cr;
5189 flow_wildcards_init_exact(wc);
5192 ovs_rwlock_rdlock(&cls->rwlock);
5193 cls_rule = classifier_lookup(cls, flow, wc);
5194 ovs_rwlock_unlock(&cls->rwlock);
5196 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5199 /* Given a port configuration (specified as zero if there's no port), chooses
5200 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
5201 * flow table miss. */
5203 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
5204 struct rule_dpif *no_packet_in_rule)
5206 return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
5210 complete_operation(struct rule_dpif *rule)
5212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5214 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5216 struct dpif_completion *c = xmalloc(sizeof *c);
5217 c->op = rule->up.pending;
5218 list_push_back(&ofproto->completions, &c->list_node);
5220 ofoperation_complete(rule->up.pending, 0);
5224 static struct rule *
5227 struct rule_dpif *rule = xmalloc(sizeof *rule);
5232 rule_dealloc(struct rule *rule_)
5234 struct rule_dpif *rule = rule_dpif_cast(rule_);
5239 rule_construct(struct rule *rule_)
5241 struct rule_dpif *rule = rule_dpif_cast(rule_);
5242 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5243 ovs_mutex_lock(&rule->stats_mutex);
5244 rule->packet_count = 0;
5245 rule->byte_count = 0;
5246 ovs_mutex_unlock(&rule->stats_mutex);
5247 complete_operation(rule);
5252 rule_destruct(struct rule *rule_)
5254 struct rule_dpif *rule = rule_dpif_cast(rule_);
5255 complete_operation(rule);
5256 ovs_mutex_destroy(&rule->stats_mutex);
5260 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5262 struct rule_dpif *rule = rule_dpif_cast(rule_);
5264 /* push_all_stats() can handle flow misses which, when using the learn
5265 * action, can cause rules to be added and deleted. This can corrupt our
5266 * caller's datastructures which assume that rule_get_stats() doesn't have
5267 * an impact on the flow table. To be safe, we disable miss handling. */
5268 push_all_stats__(false);
5270 /* Start from historical data for 'rule' itself that are no longer tracked
5271 * in facets. This counts, for example, facets that have expired. */
5272 ovs_mutex_lock(&rule->stats_mutex);
5273 *packets = rule->packet_count;
5274 *bytes = rule->byte_count;
5275 ovs_mutex_unlock(&rule->stats_mutex);
5279 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5280 struct ofpbuf *packet)
5282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5283 struct dpif_flow_stats stats;
5284 struct xlate_out xout;
5285 struct xlate_in xin;
5287 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5288 rule_credit_stats(rule, &stats);
5290 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5291 xin.resubmit_stats = &stats;
5292 xlate_actions(&xin, &xout);
5294 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5295 xout.odp_actions.size, packet);
5297 xlate_out_uninit(&xout);
5301 rule_execute(struct rule *rule, const struct flow *flow,
5302 struct ofpbuf *packet)
5304 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5305 ofpbuf_delete(packet);
5310 rule_modify_actions(struct rule *rule_)
5312 struct rule_dpif *rule = rule_dpif_cast(rule_);
5314 complete_operation(rule);
5317 /* Sends 'packet' out 'ofport'.
5318 * May modify 'packet'.
5319 * Returns 0 if successful, otherwise a positive errno value. */
5321 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5323 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5324 uint64_t odp_actions_stub[1024 / 8];
5325 struct ofpbuf key, odp_actions;
5326 struct dpif_flow_stats stats;
5327 struct odputil_keybuf keybuf;
5328 struct ofpact_output output;
5329 struct xlate_out xout;
5330 struct xlate_in xin;
5332 union flow_in_port in_port_;
5335 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5336 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5338 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5339 in_port_.ofp_port = OFPP_NONE;
5340 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5341 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5343 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5345 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5346 output.port = ofport->up.ofp_port;
5349 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5350 xin.ofpacts_len = sizeof output;
5351 xin.ofpacts = &output.ofpact;
5352 xin.resubmit_stats = &stats;
5353 xlate_actions(&xin, &xout);
5355 error = dpif_execute(ofproto->backer->dpif,
5357 xout.odp_actions.data, xout.odp_actions.size,
5359 xlate_out_uninit(&xout);
5362 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5363 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5364 ovs_strerror(error));
5367 ofproto->stats.tx_packets++;
5368 ofproto->stats.tx_bytes += packet->size;
5372 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5373 * The action will state 'slow' as the reason that the action is in the slow
5374 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5375 * dump-flows" output to see why a flow is in the slow path.)
5377 * The 'stub_size' bytes in 'stub' will be used to store the action.
5378 * 'stub_size' must be large enough for the action.
5380 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5383 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5384 enum slow_path_reason slow,
5385 uint64_t *stub, size_t stub_size,
5386 const struct nlattr **actionsp, size_t *actions_lenp)
5388 union user_action_cookie cookie;
5391 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5392 cookie.slow_path.unused = 0;
5393 cookie.slow_path.reason = slow;
5395 ofpbuf_use_stack(&buf, stub, stub_size);
5396 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5397 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5399 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5401 odp_port_t odp_port;
5404 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5405 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5406 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5408 *actionsp = buf.data;
5409 *actions_lenp = buf.size;
5413 set_frag_handling(struct ofproto *ofproto_,
5414 enum ofp_config_flags frag_handling)
5416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5417 if (frag_handling != OFPC_FRAG_REASM) {
5418 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5426 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5427 const struct flow *flow,
5428 const struct ofpact *ofpacts, size_t ofpacts_len)
5430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5431 struct odputil_keybuf keybuf;
5432 struct dpif_flow_stats stats;
5433 struct xlate_out xout;
5434 struct xlate_in xin;
5438 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5439 odp_flow_key_from_flow(&key, flow,
5440 ofp_port_to_odp_port(ofproto,
5441 flow->in_port.ofp_port));
5443 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5445 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5446 xin.resubmit_stats = &stats;
5447 xin.ofpacts_len = ofpacts_len;
5448 xin.ofpacts = ofpacts;
5450 xlate_actions(&xin, &xout);
5451 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5452 xout.odp_actions.data, xout.odp_actions.size, packet);
5453 xlate_out_uninit(&xout);
5461 set_netflow(struct ofproto *ofproto_,
5462 const struct netflow_options *netflow_options)
5464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5466 if (netflow_options) {
5467 if (!ofproto->netflow) {
5468 ofproto->netflow = netflow_create();
5469 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5471 return netflow_set_options(ofproto->netflow, netflow_options);
5472 } else if (ofproto->netflow) {
5473 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5474 netflow_destroy(ofproto->netflow);
5475 ofproto->netflow = NULL;
5482 get_netflow_ids(const struct ofproto *ofproto_,
5483 uint8_t *engine_type, uint8_t *engine_id)
5485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5487 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5491 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5493 if (!facet_is_controller_flow(facet) &&
5494 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5495 struct subfacet *subfacet;
5496 struct ofexpired expired;
5498 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5499 if (subfacet->path == SF_FAST_PATH) {
5500 struct dpif_flow_stats stats;
5502 subfacet_install(subfacet, &facet->xout.odp_actions,
5504 subfacet_update_stats(subfacet, &stats);
5508 expired.flow = facet->flow;
5509 expired.packet_count = facet->packet_count;
5510 expired.byte_count = facet->byte_count;
5511 expired.used = facet->used;
5512 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5517 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5519 struct cls_cursor cursor;
5520 struct facet *facet;
5522 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5523 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5524 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5525 send_active_timeout(ofproto, facet);
5527 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5530 static struct ofproto_dpif *
5531 ofproto_dpif_lookup(const char *name)
5533 struct ofproto_dpif *ofproto;
5535 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5536 hash_string(name, 0), &all_ofproto_dpifs) {
5537 if (!strcmp(ofproto->up.name, name)) {
5545 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5546 const char *argv[], void *aux OVS_UNUSED)
5548 struct ofproto_dpif *ofproto;
5551 ofproto = ofproto_dpif_lookup(argv[1]);
5553 unixctl_command_reply_error(conn, "no such bridge");
5556 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5557 mac_learning_flush(ofproto->ml);
5558 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5560 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5561 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5562 mac_learning_flush(ofproto->ml);
5563 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5567 unixctl_command_reply(conn, "table successfully flushed");
5570 static struct ofport_dpif *
5571 ofbundle_get_a_port(const struct ofbundle *bundle)
5573 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5578 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5579 const char *argv[], void *aux OVS_UNUSED)
5581 struct ds ds = DS_EMPTY_INITIALIZER;
5582 const struct ofproto_dpif *ofproto;
5583 const struct mac_entry *e;
5585 ofproto = ofproto_dpif_lookup(argv[1]);
5587 unixctl_command_reply_error(conn, "no such bridge");
5591 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5592 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5593 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5594 struct ofbundle *bundle = e->port.p;
5595 char name[OFP_MAX_PORT_NAME_LEN];
5597 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5599 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5600 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5601 mac_entry_age(ofproto->ml, e));
5603 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5604 unixctl_command_reply(conn, ds_cstr(&ds));
5609 struct xlate_out xout;
5610 struct xlate_in xin;
5616 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5618 ds_put_char_multiple(result, '\t', level);
5620 ds_put_cstr(result, "No match\n");
5624 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5625 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5626 cls_rule_format(&rule->up.cr, result);
5627 ds_put_char(result, '\n');
5629 ds_put_char_multiple(result, '\t', level);
5630 ds_put_cstr(result, "OpenFlow ");
5631 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5632 ds_put_char(result, '\n');
5636 trace_format_flow(struct ds *result, int level, const char *title,
5637 struct trace_ctx *trace)
5639 ds_put_char_multiple(result, '\t', level);
5640 ds_put_format(result, "%s: ", title);
5641 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5642 ds_put_cstr(result, "unchanged");
5644 flow_format(result, &trace->xin.flow);
5645 trace->flow = trace->xin.flow;
5647 ds_put_char(result, '\n');
5651 trace_format_regs(struct ds *result, int level, const char *title,
5652 struct trace_ctx *trace)
5656 ds_put_char_multiple(result, '\t', level);
5657 ds_put_format(result, "%s:", title);
5658 for (i = 0; i < FLOW_N_REGS; i++) {
5659 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5661 ds_put_char(result, '\n');
5665 trace_format_odp(struct ds *result, int level, const char *title,
5666 struct trace_ctx *trace)
5668 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5670 ds_put_char_multiple(result, '\t', level);
5671 ds_put_format(result, "%s: ", title);
5672 format_odp_actions(result, odp_actions->data, odp_actions->size);
5673 ds_put_char(result, '\n');
5677 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5679 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5680 struct ds *result = trace->result;
5682 ds_put_char(result, '\n');
5683 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5684 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5685 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5686 trace_format_rule(result, recurse + 1, rule);
5690 trace_report(struct xlate_in *xin, const char *s, int recurse)
5692 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5693 struct ds *result = trace->result;
5695 ds_put_char_multiple(result, '\t', recurse);
5696 ds_put_cstr(result, s);
5697 ds_put_char(result, '\n');
5701 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5702 void *aux OVS_UNUSED)
5704 const struct dpif_backer *backer;
5705 struct ofproto_dpif *ofproto;
5706 struct ofpbuf odp_key, odp_mask;
5707 struct ofpbuf *packet;
5715 ofpbuf_init(&odp_key, 0);
5716 ofpbuf_init(&odp_mask, 0);
5718 /* Handle "-generate" or a hex string as the last argument. */
5719 if (!strcmp(argv[argc - 1], "-generate")) {
5720 packet = ofpbuf_new(0);
5723 const char *error = eth_from_hex(argv[argc - 1], &packet);
5726 } else if (argc == 4) {
5727 /* The 3-argument form must end in "-generate' or a hex string. */
5728 unixctl_command_reply_error(conn, error);
5733 /* Parse the flow and determine whether a datapath or
5734 * bridge is specified. If function odp_flow_key_from_string()
5735 * returns 0, the flow is a odp_flow. If function
5736 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5737 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5738 /* If the odp_flow is the second argument,
5739 * the datapath name is the first argument. */
5741 const char *dp_type;
5742 if (!strncmp(argv[1], "ovs-", 4)) {
5743 dp_type = argv[1] + 4;
5747 backer = shash_find_data(&all_dpif_backers, dp_type);
5749 unixctl_command_reply_error(conn, "Cannot find datapath "
5754 /* No datapath name specified, so there should be only one
5756 struct shash_node *node;
5757 if (shash_count(&all_dpif_backers) != 1) {
5758 unixctl_command_reply_error(conn, "Must specify datapath "
5759 "name, there is more than one type of datapath");
5762 node = shash_first(&all_dpif_backers);
5763 backer = node->data;
5766 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5767 NULL, &ofproto, NULL)) {
5768 unixctl_command_reply_error(conn, "Invalid datapath flow");
5771 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5772 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5774 unixctl_command_reply_error(conn, "Must specify bridge name");
5778 ofproto = ofproto_dpif_lookup(argv[1]);
5780 unixctl_command_reply_error(conn, "Unknown bridge name");
5784 unixctl_command_reply_error(conn, "Bad flow syntax");
5788 /* Generate a packet, if requested. */
5790 if (!packet->size) {
5791 flow_compose(packet, &flow);
5793 union flow_in_port in_port_;
5795 in_port_ = flow.in_port;
5796 ds_put_cstr(&result, "Packet: ");
5797 s = ofp_packet_to_string(packet->data, packet->size);
5798 ds_put_cstr(&result, s);
5801 /* Use the metadata from the flow and the packet argument
5802 * to reconstruct the flow. */
5803 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5808 ofproto_trace(ofproto, &flow, packet, &result);
5809 unixctl_command_reply(conn, ds_cstr(&result));
5812 ds_destroy(&result);
5813 ofpbuf_delete(packet);
5814 ofpbuf_uninit(&odp_key);
5815 ofpbuf_uninit(&odp_mask);
5819 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5820 const struct ofpbuf *packet, struct ds *ds)
5822 struct rule_dpif *rule;
5824 ds_put_cstr(ds, "Flow: ");
5825 flow_format(ds, flow);
5826 ds_put_char(ds, '\n');
5828 rule = rule_dpif_lookup(ofproto, flow, NULL);
5830 trace_format_rule(ds, 0, rule);
5831 if (rule == ofproto->miss_rule) {
5832 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5833 } else if (rule == ofproto->no_packet_in_rule) {
5834 ds_put_cstr(ds, "\nNo match, packets dropped because "
5835 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5836 } else if (rule == ofproto->drop_frags_rule) {
5837 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5838 "and the fragment handling mode is \"drop\".\n");
5842 uint64_t odp_actions_stub[1024 / 8];
5843 struct ofpbuf odp_actions;
5844 struct trace_ctx trace;
5848 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5851 ofpbuf_use_stub(&odp_actions,
5852 odp_actions_stub, sizeof odp_actions_stub);
5853 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5854 trace.xin.resubmit_hook = trace_resubmit;
5855 trace.xin.report_hook = trace_report;
5857 xlate_actions(&trace.xin, &trace.xout);
5859 ds_put_char(ds, '\n');
5860 trace_format_flow(ds, 0, "Final flow", &trace);
5862 match_init(&match, flow, &trace.xout.wc);
5863 ds_put_cstr(ds, "Relevant fields: ");
5864 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5865 ds_put_char(ds, '\n');
5867 ds_put_cstr(ds, "Datapath actions: ");
5868 format_odp_actions(ds, trace.xout.odp_actions.data,
5869 trace.xout.odp_actions.size);
5871 if (trace.xout.slow) {
5872 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5873 "slow path because it:");
5874 switch (trace.xout.slow) {
5876 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5879 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5882 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5885 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5887 case SLOW_CONTROLLER:
5888 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5889 "to the OpenFlow controller.");
5896 xlate_out_uninit(&trace.xout);
5901 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5902 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5905 unixctl_command_reply(conn, NULL);
5909 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5910 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5913 unixctl_command_reply(conn, NULL);
5916 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5917 * 'reply' describing the results. */
5919 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5921 struct cls_cursor cursor;
5922 struct facet *facet;
5926 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5927 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5928 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5929 if (!facet_check_consistency(facet)) {
5933 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5935 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5939 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5940 ofproto->up.name, errors);
5942 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5947 ofproto_dpif_self_check(struct unixctl_conn *conn,
5948 int argc, const char *argv[], void *aux OVS_UNUSED)
5950 struct ds reply = DS_EMPTY_INITIALIZER;
5951 struct ofproto_dpif *ofproto;
5954 ofproto = ofproto_dpif_lookup(argv[1]);
5956 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5957 "ofproto/list for help)");
5960 ofproto_dpif_self_check__(ofproto, &reply);
5962 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5963 ofproto_dpif_self_check__(ofproto, &reply);
5967 unixctl_command_reply(conn, ds_cstr(&reply));
5971 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5972 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5973 * to destroy 'ofproto_shash' and free the returned value. */
5974 static const struct shash_node **
5975 get_ofprotos(struct shash *ofproto_shash)
5977 const struct ofproto_dpif *ofproto;
5979 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5980 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5981 shash_add_nocopy(ofproto_shash, name, ofproto);
5984 return shash_sort(ofproto_shash);
5988 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5989 const char *argv[] OVS_UNUSED,
5990 void *aux OVS_UNUSED)
5992 struct ds ds = DS_EMPTY_INITIALIZER;
5993 struct shash ofproto_shash;
5994 const struct shash_node **sorted_ofprotos;
5997 shash_init(&ofproto_shash);
5998 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5999 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6000 const struct shash_node *node = sorted_ofprotos[i];
6001 ds_put_format(&ds, "%s\n", node->name);
6004 shash_destroy(&ofproto_shash);
6005 free(sorted_ofprotos);
6007 unixctl_command_reply(conn, ds_cstr(&ds));
6012 show_dp_rates(struct ds *ds, const char *heading,
6013 const struct avg_subfacet_rates *rates)
6015 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6016 heading, rates->add_rate, rates->del_rate);
6020 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6022 const struct shash_node **ofprotos;
6023 struct ofproto_dpif *ofproto;
6024 struct shash ofproto_shash;
6025 uint64_t n_hit, n_missed;
6026 long long int minutes;
6029 n_hit = n_missed = 0;
6030 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6031 if (ofproto->backer == backer) {
6032 n_missed += ofproto->n_missed;
6033 n_hit += ofproto->n_hit;
6037 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6038 dpif_name(backer->dpif), n_hit, n_missed);
6039 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6040 " life span: %lldms\n", hmap_count(&backer->subfacets),
6041 backer->avg_n_subfacet, backer->max_n_subfacet,
6042 backer->avg_subfacet_life);
6044 minutes = (time_msec() - backer->created) / (1000 * 60);
6045 if (minutes >= 60) {
6046 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6048 if (minutes >= 60 * 24) {
6049 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6051 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6053 shash_init(&ofproto_shash);
6054 ofprotos = get_ofprotos(&ofproto_shash);
6055 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6056 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6057 const struct shash_node **ports;
6060 if (ofproto->backer != backer) {
6064 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6065 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6067 ports = shash_sort(&ofproto->up.port_by_name);
6068 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6069 const struct shash_node *node = ports[j];
6070 struct ofport *ofport = node->data;
6072 odp_port_t odp_port;
6074 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6077 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6078 if (odp_port != ODPP_NONE) {
6079 ds_put_format(ds, "%"PRIu32":", odp_port);
6081 ds_put_cstr(ds, "none:");
6084 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6087 if (!netdev_get_config(ofport->netdev, &config)) {
6088 const struct smap_node **nodes;
6091 nodes = smap_sort(&config);
6092 for (i = 0; i < smap_count(&config); i++) {
6093 const struct smap_node *node = nodes[i];
6094 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6095 node->key, node->value);
6099 smap_destroy(&config);
6101 ds_put_char(ds, ')');
6102 ds_put_char(ds, '\n');
6106 shash_destroy(&ofproto_shash);
6111 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6112 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6114 struct ds ds = DS_EMPTY_INITIALIZER;
6115 const struct shash_node **backers;
6118 backers = shash_sort(&all_dpif_backers);
6119 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6120 dpif_show_backer(backers[i]->data, &ds);
6124 unixctl_command_reply(conn, ds_cstr(&ds));
6128 /* Dump the megaflow (facet) cache. This is useful to check the
6129 * correctness of flow wildcarding, since the same mechanism is used for
6130 * both xlate caching and kernel wildcarding.
6132 * It's important to note that in the output the flow description uses
6133 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6135 * This command is only needed for advanced debugging, so it's not
6136 * documented in the man page. */
6138 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6139 int argc OVS_UNUSED, const char *argv[],
6140 void *aux OVS_UNUSED)
6142 struct ds ds = DS_EMPTY_INITIALIZER;
6143 const struct ofproto_dpif *ofproto;
6144 long long int now = time_msec();
6145 struct cls_cursor cursor;
6146 struct facet *facet;
6148 ofproto = ofproto_dpif_lookup(argv[1]);
6150 unixctl_command_reply_error(conn, "no such bridge");
6154 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
6155 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6156 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6157 cls_rule_format(&facet->cr, &ds);
6158 ds_put_cstr(&ds, ", ");
6159 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6160 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6161 ds_put_cstr(&ds, "Datapath actions: ");
6162 if (facet->xout.slow) {
6163 uint64_t slow_path_stub[128 / 8];
6164 const struct nlattr *actions;
6167 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6168 slow_path_stub, sizeof slow_path_stub,
6169 &actions, &actions_len);
6170 format_odp_actions(&ds, actions, actions_len);
6172 format_odp_actions(&ds, facet->xout.odp_actions.data,
6173 facet->xout.odp_actions.size);
6175 ds_put_cstr(&ds, "\n");
6177 ovs_rwlock_unlock(&ofproto->facets.rwlock);
6179 ds_chomp(&ds, '\n');
6180 unixctl_command_reply(conn, ds_cstr(&ds));
6184 /* Disable using the megaflows.
6186 * This command is only needed for advanced debugging, so it's not
6187 * documented in the man page. */
6189 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6190 int argc OVS_UNUSED,
6191 const char *argv[] OVS_UNUSED,
6192 void *aux OVS_UNUSED)
6194 struct ofproto_dpif *ofproto;
6196 enable_megaflows = false;
6198 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6199 flush(&ofproto->up);
6202 unixctl_command_reply(conn, "megaflows disabled");
6205 /* Re-enable using megaflows.
6207 * This command is only needed for advanced debugging, so it's not
6208 * documented in the man page. */
6210 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6211 int argc OVS_UNUSED,
6212 const char *argv[] OVS_UNUSED,
6213 void *aux OVS_UNUSED)
6215 struct ofproto_dpif *ofproto;
6217 enable_megaflows = true;
6219 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6220 flush(&ofproto->up);
6223 unixctl_command_reply(conn, "megaflows enabled");
6227 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6228 int argc OVS_UNUSED, const char *argv[],
6229 void *aux OVS_UNUSED)
6231 struct ds ds = DS_EMPTY_INITIALIZER;
6232 const struct ofproto_dpif *ofproto;
6233 struct subfacet *subfacet;
6235 ofproto = ofproto_dpif_lookup(argv[1]);
6237 unixctl_command_reply_error(conn, "no such bridge");
6241 update_stats(ofproto->backer);
6243 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6244 struct facet *facet = subfacet->facet;
6245 struct odputil_keybuf maskbuf;
6248 if (facet->ofproto != ofproto) {
6252 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6253 if (enable_megaflows) {
6254 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6255 &facet->flow, UINT32_MAX);
6258 odp_flow_format(subfacet->key, subfacet->key_len,
6259 mask.data, mask.size, &ds, false);
6261 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6262 subfacet->dp_packet_count, subfacet->dp_byte_count);
6263 if (subfacet->used) {
6264 ds_put_format(&ds, "%.3fs",
6265 (time_msec() - subfacet->used) / 1000.0);
6267 ds_put_format(&ds, "never");
6269 if (subfacet->facet->tcp_flags) {
6270 ds_put_cstr(&ds, ", flags:");
6271 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6274 ds_put_cstr(&ds, ", actions:");
6275 if (facet->xout.slow) {
6276 uint64_t slow_path_stub[128 / 8];
6277 const struct nlattr *actions;
6280 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6281 slow_path_stub, sizeof slow_path_stub,
6282 &actions, &actions_len);
6283 format_odp_actions(&ds, actions, actions_len);
6285 format_odp_actions(&ds, facet->xout.odp_actions.data,
6286 facet->xout.odp_actions.size);
6288 ds_put_char(&ds, '\n');
6291 unixctl_command_reply(conn, ds_cstr(&ds));
6296 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6297 int argc OVS_UNUSED, const char *argv[],
6298 void *aux OVS_UNUSED)
6300 struct ds ds = DS_EMPTY_INITIALIZER;
6301 struct ofproto_dpif *ofproto;
6303 ofproto = ofproto_dpif_lookup(argv[1]);
6305 unixctl_command_reply_error(conn, "no such bridge");
6309 flush(&ofproto->up);
6311 unixctl_command_reply(conn, ds_cstr(&ds));
6316 ofproto_dpif_unixctl_init(void)
6318 static bool registered;
6324 unixctl_command_register(
6326 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6327 1, 3, ofproto_unixctl_trace, NULL);
6328 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6329 ofproto_unixctl_fdb_flush, NULL);
6330 unixctl_command_register("fdb/show", "bridge", 1, 1,
6331 ofproto_unixctl_fdb_show, NULL);
6332 unixctl_command_register("ofproto/clog", "", 0, 0,
6333 ofproto_dpif_clog, NULL);
6334 unixctl_command_register("ofproto/unclog", "", 0, 0,
6335 ofproto_dpif_unclog, NULL);
6336 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6337 ofproto_dpif_self_check, NULL);
6338 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6339 ofproto_unixctl_dpif_dump_dps, NULL);
6340 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6342 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6343 ofproto_unixctl_dpif_dump_flows, NULL);
6344 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6345 ofproto_unixctl_dpif_del_flows, NULL);
6346 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6347 ofproto_unixctl_dpif_dump_megaflows, NULL);
6348 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6349 ofproto_unixctl_dpif_disable_megaflows, NULL);
6350 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6351 ofproto_unixctl_dpif_enable_megaflows, NULL);
6354 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6356 * This is deprecated. It is only for compatibility with broken device drivers
6357 * in old versions of Linux that do not properly support VLANs when VLAN
6358 * devices are not used. When broken device drivers are no longer in
6359 * widespread use, we will delete these interfaces. */
6362 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6364 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6365 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6367 if (realdev_ofp_port == ofport->realdev_ofp_port
6368 && vid == ofport->vlandev_vid) {
6372 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6374 if (ofport->realdev_ofp_port) {
6377 if (realdev_ofp_port && ofport->bundle) {
6378 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6379 * themselves be part of a bundle. */
6380 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6383 ofport->realdev_ofp_port = realdev_ofp_port;
6384 ofport->vlandev_vid = vid;
6386 if (realdev_ofp_port) {
6387 vsp_add(ofport, realdev_ofp_port, vid);
6394 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6396 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6400 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6401 OVS_EXCLUDED(ofproto->vsp_mutex)
6405 ovs_mutex_lock(&ofproto->vsp_mutex);
6406 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6407 ovs_mutex_unlock(&ofproto->vsp_mutex);
6412 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6413 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6414 OVS_REQUIRES(ofproto->vsp_mutex)
6416 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6417 int vid = vlan_tci_to_vid(vlan_tci);
6418 const struct vlan_splinter *vsp;
6420 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6421 hash_realdev_vid(realdev_ofp_port, vid),
6422 &ofproto->realdev_vid_map) {
6423 if (vsp->realdev_ofp_port == realdev_ofp_port
6424 && vsp->vid == vid) {
6425 return vsp->vlandev_ofp_port;
6429 return realdev_ofp_port;
6432 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6433 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6434 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6435 * 'vlan_tci' 9, it would return the port number of eth0.9.
6437 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6438 * function just returns its 'realdev_ofp_port' argument. */
6440 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6441 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6442 OVS_EXCLUDED(ofproto->vsp_mutex)
6446 ovs_mutex_lock(&ofproto->vsp_mutex);
6447 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6448 ovs_mutex_unlock(&ofproto->vsp_mutex);
6452 static struct vlan_splinter *
6453 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6455 struct vlan_splinter *vsp;
6457 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6458 hash_ofp_port(vlandev_ofp_port),
6459 &ofproto->vlandev_map) {
6460 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6468 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6469 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6470 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6471 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6472 * eth0 and store 9 in '*vid'.
6474 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6475 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6478 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6479 ofp_port_t vlandev_ofp_port, int *vid)
6480 OVS_REQUIRES(ofproto->vsp_mutex)
6482 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6483 const struct vlan_splinter *vsp;
6485 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6490 return vsp->realdev_ofp_port;
6496 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6497 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6498 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6499 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6500 * always the case unless VLAN splinters are enabled), returns false without
6501 * making any changes. */
6503 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6504 OVS_EXCLUDED(ofproto->vsp_mutex)
6509 ovs_mutex_lock(&ofproto->vsp_mutex);
6510 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6511 ovs_mutex_unlock(&ofproto->vsp_mutex);
6516 /* Cause the flow to be processed as if it came in on the real device with
6517 * the VLAN device's VLAN ID. */
6518 flow->in_port.ofp_port = realdev;
6519 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6524 vsp_remove(struct ofport_dpif *port)
6526 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6527 struct vlan_splinter *vsp;
6529 ovs_mutex_lock(&ofproto->vsp_mutex);
6530 vsp = vlandev_find(ofproto, port->up.ofp_port);
6532 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6533 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6536 port->realdev_ofp_port = 0;
6538 VLOG_ERR("missing vlan device record");
6540 ovs_mutex_unlock(&ofproto->vsp_mutex);
6544 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6548 ovs_mutex_lock(&ofproto->vsp_mutex);
6549 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6550 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6551 == realdev_ofp_port)) {
6552 struct vlan_splinter *vsp;
6554 vsp = xmalloc(sizeof *vsp);
6555 vsp->realdev_ofp_port = realdev_ofp_port;
6556 vsp->vlandev_ofp_port = port->up.ofp_port;
6559 port->realdev_ofp_port = realdev_ofp_port;
6561 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6562 hash_ofp_port(port->up.ofp_port));
6563 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6564 hash_realdev_vid(realdev_ofp_port, vid));
6566 VLOG_ERR("duplicate vlan device record");
6568 ovs_mutex_unlock(&ofproto->vsp_mutex);
6572 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6574 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6575 return ofport ? ofport->odp_port : ODPP_NONE;
6578 struct ofport_dpif *
6579 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6581 struct ofport_dpif *port;
6583 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6584 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6585 &backer->odp_to_ofport_map) {
6586 if (port->odp_port == odp_port) {
6587 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6592 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6597 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6599 struct ofport_dpif *port;
6601 port = odp_port_to_ofport(ofproto->backer, odp_port);
6602 if (port && &ofproto->up == port->up.ofproto) {
6603 return port->up.ofp_port;
6609 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6610 * most heavily weighted element. 'base' designates the rate of decay: after
6611 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6614 exp_mavg(double *avg, int base, double new)
6616 *avg = (*avg * (base - 1) + new) / base;
6620 update_moving_averages(struct dpif_backer *backer)
6622 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6623 long long int minutes = (time_msec() - backer->created) / min_ms;
6626 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6628 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6631 backer->lifetime.add_rate = 0.0;
6632 backer->lifetime.del_rate = 0.0;
6635 /* Update hourly averages on the minute boundaries. */
6636 if (time_msec() - backer->last_minute >= min_ms) {
6637 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6638 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6640 /* Update daily averages on the hour boundaries. */
6641 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6642 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6643 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6646 backer->total_subfacet_add_count += backer->subfacet_add_count;
6647 backer->total_subfacet_del_count += backer->subfacet_del_count;
6648 backer->subfacet_add_count = 0;
6649 backer->subfacet_del_count = 0;
6650 backer->last_minute += min_ms;
6654 const struct ofproto_class ofproto_dpif_class = {
6689 port_is_lacp_current,
6690 NULL, /* rule_choose_table */
6697 rule_modify_actions,
6711 get_stp_port_status,
6718 is_mirror_output_bundle,
6719 forward_bpdu_changed,
6720 set_mac_table_config,
6722 NULL, /* meter_get_features */
6723 NULL, /* meter_set */
6724 NULL, /* meter_get */
6725 NULL, /* meter_del */