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 xlate_ofproto_set(ofproto, ofproto->up.name,
802 ofproto->backer->dpif, ofproto->ml,
803 ofproto->stp, ofproto->mbridge,
804 ofproto->sflow, ofproto->ipfix,
805 ofproto->up.frag_handling,
806 ofproto->up.forward_bpdu,
807 connmgr_has_in_band(ofproto->up.connmgr),
808 ofproto->netflow != NULL);
810 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
811 xlate_bundle_set(ofproto, bundle, bundle->name,
812 bundle->vlan_mode, bundle->vlan,
813 bundle->trunks, bundle->use_priority_tags,
814 bundle->bond, bundle->lacp,
818 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
819 int stp_port = ofport->stp_port
820 ? stp_port_no(ofport->stp_port)
822 xlate_ofport_set(ofproto, ofport->bundle, ofport,
823 ofport->up.ofp_port, ofport->odp_port,
824 ofport->up.netdev, ofport->cfm,
825 ofport->bfd, ofport->peer, stp_port,
826 ofport->qdscp, ofport->n_qdscp,
827 ofport->up.pp.config, ofport->is_tunnel,
831 cls_cursor_init(&cursor, &ofproto->facets, NULL);
832 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
833 facet_revalidate(facet);
839 if (!backer->recv_set_enable) {
840 /* Wake up before a max of 1000ms. */
841 timer_set_duration(&backer->next_expiration, 1000);
842 } else if (timer_expired(&backer->next_expiration)) {
843 int delay = expire(backer);
844 timer_set_duration(&backer->next_expiration, delay);
847 process_dpif_port_changes(backer);
849 if (backer->governor) {
852 governor_run(backer->governor);
854 /* If the governor has shrunk to its minimum size and the number of
855 * subfacets has dwindled, then drop the governor entirely.
857 * For hysteresis, the number of subfacets to drop the governor is
858 * smaller than the number needed to trigger its creation. */
859 n_subfacets = hmap_count(&backer->subfacets);
860 if (n_subfacets * 4 < flow_eviction_threshold
861 && governor_is_idle(backer->governor)) {
862 governor_destroy(backer->governor);
863 backer->governor = NULL;
870 /* Check for and handle port changes in 'backer''s dpif. */
872 process_dpif_port_changes(struct dpif_backer *backer)
878 error = dpif_port_poll(backer->dpif, &devname);
884 process_dpif_all_ports_changed(backer);
888 process_dpif_port_change(backer, devname);
893 process_dpif_port_error(backer, error);
900 process_dpif_all_ports_changed(struct dpif_backer *backer)
902 struct ofproto_dpif *ofproto;
903 struct dpif_port dpif_port;
904 struct dpif_port_dump dump;
905 struct sset devnames;
908 sset_init(&devnames);
909 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
910 if (ofproto->backer == backer) {
911 struct ofport *ofport;
913 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
914 sset_add(&devnames, netdev_get_name(ofport->netdev));
918 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
919 sset_add(&devnames, dpif_port.name);
922 SSET_FOR_EACH (devname, &devnames) {
923 process_dpif_port_change(backer, devname);
925 sset_destroy(&devnames);
929 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
931 struct ofproto_dpif *ofproto;
932 struct dpif_port port;
934 /* Don't report on the datapath's device. */
935 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
939 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
940 &all_ofproto_dpifs) {
941 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
946 ofproto = lookup_ofproto_dpif_by_port_name(devname);
947 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
948 /* The port was removed. If we know the datapath,
949 * report it through poll_set(). If we don't, it may be
950 * notifying us of a removal we initiated, so ignore it.
951 * If there's a pending ENOBUFS, let it stand, since
952 * everything will be reevaluated. */
953 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
954 sset_add(&ofproto->port_poll_set, devname);
955 ofproto->port_poll_errno = 0;
957 } else if (!ofproto) {
958 /* The port was added, but we don't know with which
959 * ofproto we should associate it. Delete it. */
960 dpif_port_del(backer->dpif, port.port_no);
962 struct ofport_dpif *ofport;
964 ofport = ofport_dpif_cast(shash_find_data(
965 &ofproto->up.port_by_name, devname));
967 && ofport->odp_port != port.port_no
968 && !odp_port_to_ofport(backer, port.port_no))
970 /* 'ofport''s datapath port number has changed from
971 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
972 * structures to match. */
973 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
974 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
975 ofport->odp_port = port.port_no;
976 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
977 hash_odp_port(port.port_no));
978 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
979 backer->need_revalidate = REV_RECONFIGURE;
982 dpif_port_destroy(&port);
985 /* Propagate 'error' to all ofprotos based on 'backer'. */
987 process_dpif_port_error(struct dpif_backer *backer, int error)
989 struct ofproto_dpif *ofproto;
991 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
992 if (ofproto->backer == backer) {
993 sset_clear(&ofproto->port_poll_set);
994 ofproto->port_poll_errno = error;
1000 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1004 /* If recv_set_enable is false, we should not handle upcalls. */
1005 if (!backer->recv_set_enable) {
1009 /* Handle one or more batches of upcalls, until there's nothing left to do
1010 * or until we do a fixed total amount of work.
1012 * We do work in batches because it can be much cheaper to set up a number
1013 * of flows and fire off their patches all at once. We do multiple batches
1014 * because in some cases handling a packet can cause another packet to be
1015 * queued almost immediately as part of the return flow. Both
1016 * optimizations can make major improvements on some benchmarks and
1017 * presumably for real traffic as well. */
1019 while (work < max_batch) {
1020 int retval = handle_upcalls(backer, max_batch - work);
1031 type_run_fast(const char *type)
1033 struct dpif_backer *backer;
1035 backer = shash_find_data(&all_dpif_backers, type);
1037 /* This is not necessarily a problem, since backers are only
1038 * created on demand. */
1042 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1048 static long long int port_rl = LLONG_MIN;
1049 static unsigned int backer_rl = 0;
1051 if (time_msec() >= port_rl) {
1052 struct ofproto_dpif *ofproto;
1054 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1055 run_fast(&ofproto->up);
1057 port_rl = time_msec() + 200;
1060 /* XXX: We have to be careful not to do too much work in this function. If
1061 * we call dpif_backer_run_fast() too often, or with too large a batch,
1062 * performance improves signifcantly, but at a cost. It's possible for the
1063 * number of flows in the datapath to increase without bound, and for poll
1064 * loops to take 10s of seconds. The correct solution to this problem,
1065 * long term, is to separate flow miss handling into it's own thread so it
1066 * isn't affected by revalidations, and expirations. Until then, this is
1067 * the best we can do. */
1068 if (++backer_rl >= 10) {
1069 struct shash_node *node;
1072 SHASH_FOR_EACH (node, &all_dpif_backers) {
1073 dpif_backer_run_fast(node->data, 1);
1079 type_wait(const char *type)
1081 struct dpif_backer *backer;
1083 backer = shash_find_data(&all_dpif_backers, type);
1085 /* This is not necessarily a problem, since backers are only
1086 * created on demand. */
1090 if (backer->governor) {
1091 governor_wait(backer->governor);
1094 timer_wait(&backer->next_expiration);
1097 /* Basic life-cycle. */
1099 static int add_internal_flows(struct ofproto_dpif *);
1101 static struct ofproto *
1104 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1105 return &ofproto->up;
1109 dealloc(struct ofproto *ofproto_)
1111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1116 close_dpif_backer(struct dpif_backer *backer)
1118 struct shash_node *node;
1120 ovs_assert(backer->refcount > 0);
1122 if (--backer->refcount) {
1126 drop_key_clear(backer);
1127 hmap_destroy(&backer->drop_keys);
1129 simap_destroy(&backer->tnl_backers);
1130 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1131 hmap_destroy(&backer->odp_to_ofport_map);
1132 node = shash_find(&all_dpif_backers, backer->type);
1134 shash_delete(&all_dpif_backers, node);
1135 dpif_close(backer->dpif);
1137 ovs_assert(hmap_is_empty(&backer->subfacets));
1138 hmap_destroy(&backer->subfacets);
1139 governor_destroy(backer->governor);
1144 /* Datapath port slated for removal from datapath. */
1145 struct odp_garbage {
1146 struct list list_node;
1147 odp_port_t odp_port;
1151 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1153 struct dpif_backer *backer;
1154 struct dpif_port_dump port_dump;
1155 struct dpif_port port;
1156 struct shash_node *node;
1157 struct list garbage_list;
1158 struct odp_garbage *garbage, *next;
1164 backer = shash_find_data(&all_dpif_backers, type);
1171 backer_name = xasprintf("ovs-%s", type);
1173 /* Remove any existing datapaths, since we assume we're the only
1174 * userspace controlling the datapath. */
1176 dp_enumerate_names(type, &names);
1177 SSET_FOR_EACH(name, &names) {
1178 struct dpif *old_dpif;
1180 /* Don't remove our backer if it exists. */
1181 if (!strcmp(name, backer_name)) {
1185 if (dpif_open(name, type, &old_dpif)) {
1186 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1188 dpif_delete(old_dpif);
1189 dpif_close(old_dpif);
1192 sset_destroy(&names);
1194 backer = xmalloc(sizeof *backer);
1196 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1199 VLOG_ERR("failed to open datapath of type %s: %s", type,
1200 ovs_strerror(error));
1205 backer->type = xstrdup(type);
1206 backer->governor = NULL;
1207 backer->refcount = 1;
1208 hmap_init(&backer->odp_to_ofport_map);
1209 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1210 hmap_init(&backer->drop_keys);
1211 hmap_init(&backer->subfacets);
1212 timer_set_duration(&backer->next_expiration, 1000);
1213 backer->need_revalidate = 0;
1214 simap_init(&backer->tnl_backers);
1215 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1218 if (backer->recv_set_enable) {
1219 dpif_flow_flush(backer->dpif);
1222 /* Loop through the ports already on the datapath and remove any
1223 * that we don't need anymore. */
1224 list_init(&garbage_list);
1225 dpif_port_dump_start(&port_dump, backer->dpif);
1226 while (dpif_port_dump_next(&port_dump, &port)) {
1227 node = shash_find(&init_ofp_ports, port.name);
1228 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1229 garbage = xmalloc(sizeof *garbage);
1230 garbage->odp_port = port.port_no;
1231 list_push_front(&garbage_list, &garbage->list_node);
1234 dpif_port_dump_done(&port_dump);
1236 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1237 dpif_port_del(backer->dpif, garbage->odp_port);
1238 list_remove(&garbage->list_node);
1242 shash_add(&all_dpif_backers, type, backer);
1244 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1246 VLOG_ERR("failed to listen on datapath of type %s: %s",
1247 type, ovs_strerror(error));
1248 close_dpif_backer(backer);
1252 backer->max_n_subfacet = 0;
1253 backer->created = time_msec();
1254 backer->last_minute = backer->created;
1255 memset(&backer->hourly, 0, sizeof backer->hourly);
1256 memset(&backer->daily, 0, sizeof backer->daily);
1257 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1258 backer->subfacet_add_count = 0;
1259 backer->subfacet_del_count = 0;
1260 backer->total_subfacet_add_count = 0;
1261 backer->total_subfacet_del_count = 0;
1262 backer->avg_n_subfacet = 0;
1263 backer->avg_subfacet_life = 0;
1269 construct(struct ofproto *ofproto_)
1271 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1272 struct shash_node *node, *next;
1273 odp_port_t max_ports;
1276 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1281 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1282 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1283 ofp_to_u16(OFPP_MAX))));
1285 ofproto->netflow = NULL;
1286 ofproto->sflow = NULL;
1287 ofproto->ipfix = NULL;
1288 ofproto->stp = NULL;
1289 hmap_init(&ofproto->bundles);
1290 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1291 ofproto->mbridge = mbridge_create();
1292 ofproto->has_bonded_bundles = false;
1293 ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1295 classifier_init(&ofproto->facets);
1296 ofproto->consistency_rl = LLONG_MIN;
1298 list_init(&ofproto->completions);
1300 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1301 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1302 list_init(&ofproto->flow_mods);
1303 ofproto->n_flow_mods = 0;
1304 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1306 ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL);
1307 ovs_mutex_lock(&ofproto->pin_mutex);
1308 list_init(&ofproto->pins);
1309 ofproto->n_pins = 0;
1310 ovs_mutex_unlock(&ofproto->pin_mutex);
1312 ofproto_dpif_unixctl_init();
1314 hmap_init(&ofproto->vlandev_map);
1315 hmap_init(&ofproto->realdev_vid_map);
1317 sset_init(&ofproto->ports);
1318 sset_init(&ofproto->ghost_ports);
1319 sset_init(&ofproto->port_poll_set);
1320 ofproto->port_poll_errno = 0;
1322 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1323 struct iface_hint *iface_hint = node->data;
1325 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1326 /* Check if the datapath already has this port. */
1327 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1328 sset_add(&ofproto->ports, node->name);
1331 free(iface_hint->br_name);
1332 free(iface_hint->br_type);
1334 shash_delete(&init_ofp_ports, node);
1338 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1339 hash_string(ofproto->up.name, 0));
1340 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1342 ofproto_init_tables(ofproto_, N_TABLES);
1343 error = add_internal_flows(ofproto);
1344 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1347 ofproto->n_missed = 0;
1353 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1354 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1356 struct ofputil_flow_mod fm;
1359 match_init_catchall(&fm.match);
1361 match_set_reg(&fm.match, 0, id);
1362 fm.new_cookie = htonll(0);
1363 fm.cookie = htonll(0);
1364 fm.cookie_mask = htonll(0);
1365 fm.modify_cookie = false;
1366 fm.table_id = TBL_INTERNAL;
1367 fm.command = OFPFC_ADD;
1368 fm.idle_timeout = 0;
1369 fm.hard_timeout = 0;
1373 fm.ofpacts = ofpacts->data;
1374 fm.ofpacts_len = ofpacts->size;
1376 error = ofproto_flow_mod(&ofproto->up, &fm);
1378 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1379 id, ofperr_to_string(error));
1383 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1385 ovs_assert(*rulep != NULL);
1391 add_internal_flows(struct ofproto_dpif *ofproto)
1393 struct ofpact_controller *controller;
1394 uint64_t ofpacts_stub[128 / 8];
1395 struct ofpbuf ofpacts;
1399 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1402 controller = ofpact_put_CONTROLLER(&ofpacts);
1403 controller->max_len = UINT16_MAX;
1404 controller->controller_id = 0;
1405 controller->reason = OFPR_NO_MATCH;
1406 ofpact_pad(&ofpacts);
1408 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1413 ofpbuf_clear(&ofpacts);
1414 error = add_internal_flow(ofproto, id++, &ofpacts,
1415 &ofproto->no_packet_in_rule);
1420 error = add_internal_flow(ofproto, id++, &ofpacts,
1421 &ofproto->drop_frags_rule);
1426 complete_operations(struct ofproto_dpif *ofproto)
1428 struct dpif_completion *c, *next;
1430 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1431 ofoperation_complete(c->op, 0);
1432 list_remove(&c->list_node);
1438 destruct(struct ofproto *ofproto_)
1440 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1441 struct rule_dpif *rule, *next_rule;
1442 struct ofputil_flow_mod *pin, *next_pin;
1443 struct ofputil_flow_mod *fm, *next_fm;
1444 struct oftable *table;
1446 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1447 xlate_remove_ofproto(ofproto);
1449 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1450 complete_operations(ofproto);
1452 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1453 struct cls_cursor cursor;
1455 cls_cursor_init(&cursor, &table->cls, NULL);
1456 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1457 ofproto_rule_destroy(&rule->up);
1461 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1462 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1463 list_remove(&fm->list_node);
1464 ofproto->n_flow_mods--;
1468 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1469 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1471 ovs_mutex_lock(&ofproto->pin_mutex);
1472 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1473 list_remove(&pin->list_node);
1478 ovs_mutex_unlock(&ofproto->pin_mutex);
1479 ovs_mutex_destroy(&ofproto->pin_mutex);
1481 mbridge_unref(ofproto->mbridge);
1483 netflow_destroy(ofproto->netflow);
1484 dpif_sflow_unref(ofproto->sflow);
1485 hmap_destroy(&ofproto->bundles);
1486 mac_learning_unref(ofproto->ml);
1488 classifier_destroy(&ofproto->facets);
1490 hmap_destroy(&ofproto->vlandev_map);
1491 hmap_destroy(&ofproto->realdev_vid_map);
1493 sset_destroy(&ofproto->ports);
1494 sset_destroy(&ofproto->ghost_ports);
1495 sset_destroy(&ofproto->port_poll_set);
1497 ovs_mutex_destroy(&ofproto->vsp_mutex);
1499 close_dpif_backer(ofproto->backer);
1503 run_fast(struct ofproto *ofproto_)
1505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1506 struct ofputil_packet_in *pin, *next_pin;
1507 struct ofputil_flow_mod *fm, *next_fm;
1508 struct list flow_mods, pins;
1509 struct ofport_dpif *ofport;
1511 /* Do not perform any periodic activity required by 'ofproto' while
1512 * waiting for flow restore to complete. */
1513 if (ofproto_get_flow_restore_wait()) {
1517 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1518 if (ofproto->n_flow_mods) {
1519 flow_mods = ofproto->flow_mods;
1520 list_moved(&flow_mods);
1521 list_init(&ofproto->flow_mods);
1522 ofproto->n_flow_mods = 0;
1524 list_init(&flow_mods);
1526 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1528 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1529 int error = ofproto_flow_mod(&ofproto->up, fm);
1530 if (error && !VLOG_DROP_WARN(&rl)) {
1531 VLOG_WARN("learning action failed to modify flow table (%s)",
1532 ofperr_get_name(error));
1535 list_remove(&fm->list_node);
1540 ovs_mutex_lock(&ofproto->pin_mutex);
1541 if (ofproto->n_pins) {
1542 pins = ofproto->pins;
1544 list_init(&ofproto->pins);
1545 ofproto->n_pins = 0;
1549 ovs_mutex_unlock(&ofproto->pin_mutex);
1551 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1552 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1553 list_remove(&pin->list_node);
1554 free(CONST_CAST(void *, pin->packet));
1558 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1559 port_run_fast(ofport);
1566 run(struct ofproto *ofproto_)
1568 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1569 struct ofport_dpif *ofport;
1570 struct ofbundle *bundle;
1574 complete_operations(ofproto);
1577 if (mbridge_need_revalidate(ofproto->mbridge)) {
1578 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1579 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1580 mac_learning_flush(ofproto->ml);
1581 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1584 /* Do not perform any periodic activity below required by 'ofproto' while
1585 * waiting for flow restore to complete. */
1586 if (ofproto_get_flow_restore_wait()) {
1590 error = run_fast(ofproto_);
1595 if (ofproto->netflow) {
1596 if (netflow_run(ofproto->netflow)) {
1597 send_netflow_active_timeouts(ofproto);
1600 if (ofproto->sflow) {
1601 dpif_sflow_run(ofproto->sflow);
1604 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1607 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1612 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1613 if (mac_learning_run(ofproto->ml)) {
1614 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1616 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1618 /* Check the consistency of a random facet, to aid debugging. */
1619 if (time_msec() >= ofproto->consistency_rl
1620 && !classifier_is_empty(&ofproto->facets)
1621 && !ofproto->backer->need_revalidate) {
1622 struct cls_table *table;
1623 struct cls_rule *cr;
1624 struct facet *facet;
1626 ofproto->consistency_rl = time_msec() + 250;
1628 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1629 struct cls_table, hmap_node);
1630 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1632 facet = CONTAINER_OF(cr, struct facet, cr);
1634 if (!facet_check_consistency(facet)) {
1635 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1643 wait(struct ofproto *ofproto_)
1645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1646 struct ofport_dpif *ofport;
1647 struct ofbundle *bundle;
1649 if (!clogged && !list_is_empty(&ofproto->completions)) {
1650 poll_immediate_wake();
1653 if (ofproto_get_flow_restore_wait()) {
1657 dpif_wait(ofproto->backer->dpif);
1658 dpif_recv_wait(ofproto->backer->dpif);
1659 if (ofproto->sflow) {
1660 dpif_sflow_wait(ofproto->sflow);
1662 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1665 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1666 bundle_wait(bundle);
1668 if (ofproto->netflow) {
1669 netflow_wait(ofproto->netflow);
1671 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1672 mac_learning_wait(ofproto->ml);
1673 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1675 if (ofproto->backer->need_revalidate) {
1676 /* Shouldn't happen, but if it does just go around again. */
1677 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1678 poll_immediate_wake();
1683 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1685 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1686 struct cls_cursor cursor;
1687 size_t n_subfacets = 0;
1688 struct facet *facet;
1690 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1692 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1693 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1694 n_subfacets += list_size(&facet->subfacets);
1696 simap_increase(usage, "subfacets", n_subfacets);
1700 flush(struct ofproto *ofproto_)
1702 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1703 struct subfacet *subfacet, *next_subfacet;
1704 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1708 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1709 &ofproto->backer->subfacets) {
1710 if (subfacet->facet->ofproto != ofproto) {
1714 if (subfacet->path != SF_NOT_INSTALLED) {
1715 batch[n_batch++] = subfacet;
1716 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1717 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1721 subfacet_destroy(subfacet);
1726 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1731 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1732 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1734 *arp_match_ip = true;
1735 *actions = (OFPUTIL_A_OUTPUT |
1736 OFPUTIL_A_SET_VLAN_VID |
1737 OFPUTIL_A_SET_VLAN_PCP |
1738 OFPUTIL_A_STRIP_VLAN |
1739 OFPUTIL_A_SET_DL_SRC |
1740 OFPUTIL_A_SET_DL_DST |
1741 OFPUTIL_A_SET_NW_SRC |
1742 OFPUTIL_A_SET_NW_DST |
1743 OFPUTIL_A_SET_NW_TOS |
1744 OFPUTIL_A_SET_TP_SRC |
1745 OFPUTIL_A_SET_TP_DST |
1750 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1752 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1753 struct dpif_dp_stats s;
1754 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1757 strcpy(ots->name, "classifier");
1759 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1760 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1761 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1762 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1764 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1765 ots->lookup_count = htonll(n_lookup);
1766 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1769 static struct ofport *
1772 struct ofport_dpif *port = xmalloc(sizeof *port);
1777 port_dealloc(struct ofport *port_)
1779 struct ofport_dpif *port = ofport_dpif_cast(port_);
1784 port_construct(struct ofport *port_)
1786 struct ofport_dpif *port = ofport_dpif_cast(port_);
1787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1788 const struct netdev *netdev = port->up.netdev;
1789 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1790 struct dpif_port dpif_port;
1793 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1794 port->bundle = NULL;
1797 port->may_enable = true;
1798 port->stp_port = NULL;
1799 port->stp_state = STP_DISABLED;
1800 port->is_tunnel = false;
1804 port->realdev_ofp_port = 0;
1805 port->vlandev_vid = 0;
1806 port->carrier_seq = netdev_get_carrier_resets(netdev);
1808 if (netdev_vport_is_patch(netdev)) {
1809 /* By bailing out here, we don't submit the port to the sFlow module
1810 * to be considered for counter polling export. This is correct
1811 * because the patch port represents an interface that sFlow considers
1812 * to be "internal" to the switch as a whole, and therefore not an
1813 * candidate for counter polling. */
1814 port->odp_port = ODPP_NONE;
1815 ofport_update_peer(port);
1819 error = dpif_port_query_by_name(ofproto->backer->dpif,
1820 netdev_vport_get_dpif_port(netdev, namebuf,
1827 port->odp_port = dpif_port.port_no;
1829 if (netdev_get_tunnel_config(netdev)) {
1830 tnl_port_add(port, port->up.netdev, port->odp_port);
1831 port->is_tunnel = true;
1833 /* Sanity-check that a mapping doesn't already exist. This
1834 * shouldn't happen for non-tunnel ports. */
1835 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1836 VLOG_ERR("port %s already has an OpenFlow port number",
1838 dpif_port_destroy(&dpif_port);
1842 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1843 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1844 hash_odp_port(port->odp_port));
1845 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1847 dpif_port_destroy(&dpif_port);
1849 if (ofproto->sflow) {
1850 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1857 port_destruct(struct ofport *port_)
1859 struct ofport_dpif *port = ofport_dpif_cast(port_);
1860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1861 const char *devname = netdev_get_name(port->up.netdev);
1862 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1863 const char *dp_port_name;
1865 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1866 xlate_ofport_remove(port);
1868 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1870 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1871 /* The underlying device is still there, so delete it. This
1872 * happens when the ofproto is being destroyed, since the caller
1873 * assumes that removal of attached ports will happen as part of
1875 if (!port->is_tunnel) {
1876 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1881 port->peer->peer = NULL;
1885 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1886 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1887 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1888 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1892 sset_find_and_delete(&ofproto->ports, devname);
1893 sset_find_and_delete(&ofproto->ghost_ports, devname);
1894 bundle_remove(port_);
1895 set_cfm(port_, NULL);
1896 set_bfd(port_, NULL);
1897 if (ofproto->sflow) {
1898 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1905 port_modified(struct ofport *port_)
1907 struct ofport_dpif *port = ofport_dpif_cast(port_);
1909 if (port->bundle && port->bundle->bond) {
1910 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1914 cfm_set_netdev(port->cfm, port->up.netdev);
1917 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1919 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1923 ofport_update_peer(port);
1927 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1929 struct ofport_dpif *port = ofport_dpif_cast(port_);
1930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1931 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1933 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1934 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1935 OFPUTIL_PC_NO_PACKET_IN)) {
1936 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1938 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1939 bundle_update(port->bundle);
1945 set_sflow(struct ofproto *ofproto_,
1946 const struct ofproto_sflow_options *sflow_options)
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1949 struct dpif_sflow *ds = ofproto->sflow;
1951 if (sflow_options) {
1953 struct ofport_dpif *ofport;
1955 ds = ofproto->sflow = dpif_sflow_create();
1956 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1957 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961 dpif_sflow_set_options(ds, sflow_options);
1964 dpif_sflow_unref(ds);
1965 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966 ofproto->sflow = NULL;
1974 struct ofproto *ofproto_,
1975 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1976 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1977 size_t n_flow_exporters_options)
1979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1980 struct dpif_ipfix *di = ofproto->ipfix;
1982 if (bridge_exporter_options || flow_exporters_options) {
1984 di = ofproto->ipfix = dpif_ipfix_create();
1986 dpif_ipfix_set_options(
1987 di, bridge_exporter_options, flow_exporters_options,
1988 n_flow_exporters_options);
1991 dpif_ipfix_unref(di);
1992 ofproto->ipfix = NULL;
1999 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008 struct ofproto_dpif *ofproto;
2010 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2012 ofport->cfm = cfm_create(ofport->up.netdev);
2015 if (cfm_configure(ofport->cfm, s)) {
2021 cfm_unref(ofport->cfm);
2027 get_cfm_status(const struct ofport *ofport_,
2028 struct ofproto_cfm_status *status)
2030 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2033 status->faults = cfm_get_fault(ofport->cfm);
2034 status->remote_opstate = cfm_get_opup(ofport->cfm);
2035 status->health = cfm_get_health(ofport->cfm);
2036 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2044 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2047 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2051 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2052 if (ofport->bfd != old) {
2053 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2060 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2062 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2065 bfd_get_status(ofport->bfd, smap);
2072 /* Spanning Tree. */
2075 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2077 struct ofproto_dpif *ofproto = ofproto_;
2078 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2079 struct ofport_dpif *ofport;
2081 ofport = stp_port_get_aux(sp);
2083 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2084 ofproto->up.name, port_num);
2086 struct eth_header *eth = pkt->l2;
2088 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2089 if (eth_addr_is_zero(eth->eth_src)) {
2090 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2091 "with unknown MAC", ofproto->up.name, port_num);
2093 send_packet(ofport, pkt);
2099 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2101 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2105 /* Only revalidate flows if the configuration changed. */
2106 if (!s != !ofproto->stp) {
2107 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2111 if (!ofproto->stp) {
2112 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2113 send_bpdu_cb, ofproto);
2114 ofproto->stp_last_tick = time_msec();
2117 stp_set_bridge_id(ofproto->stp, s->system_id);
2118 stp_set_bridge_priority(ofproto->stp, s->priority);
2119 stp_set_hello_time(ofproto->stp, s->hello_time);
2120 stp_set_max_age(ofproto->stp, s->max_age);
2121 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2123 struct ofport *ofport;
2125 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2126 set_stp_port(ofport, NULL);
2129 stp_unref(ofproto->stp);
2130 ofproto->stp = NULL;
2137 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2139 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2143 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2144 s->designated_root = stp_get_designated_root(ofproto->stp);
2145 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2154 update_stp_port_state(struct ofport_dpif *ofport)
2156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2157 enum stp_state state;
2159 /* Figure out new state. */
2160 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2164 if (ofport->stp_state != state) {
2165 enum ofputil_port_state of_state;
2168 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2169 netdev_get_name(ofport->up.netdev),
2170 stp_state_name(ofport->stp_state),
2171 stp_state_name(state));
2172 if (stp_learn_in_state(ofport->stp_state)
2173 != stp_learn_in_state(state)) {
2174 /* xxx Learning action flows should also be flushed. */
2175 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2176 mac_learning_flush(ofproto->ml);
2177 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2179 fwd_change = stp_forward_in_state(ofport->stp_state)
2180 != stp_forward_in_state(state);
2182 ofproto->backer->need_revalidate = REV_STP;
2183 ofport->stp_state = state;
2184 ofport->stp_state_entered = time_msec();
2186 if (fwd_change && ofport->bundle) {
2187 bundle_update(ofport->bundle);
2190 /* Update the STP state bits in the OpenFlow port description. */
2191 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2192 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2193 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2194 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2195 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2197 ofproto_port_set_state(&ofport->up, of_state);
2201 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2202 * caller is responsible for assigning STP port numbers and ensuring
2203 * there are no duplicates. */
2205 set_stp_port(struct ofport *ofport_,
2206 const struct ofproto_port_stp_settings *s)
2208 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2210 struct stp_port *sp = ofport->stp_port;
2212 if (!s || !s->enable) {
2214 ofport->stp_port = NULL;
2215 stp_port_disable(sp);
2216 update_stp_port_state(ofport);
2219 } else if (sp && stp_port_no(sp) != s->port_num
2220 && ofport == stp_port_get_aux(sp)) {
2221 /* The port-id changed, so disable the old one if it's not
2222 * already in use by another port. */
2223 stp_port_disable(sp);
2226 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2227 stp_port_enable(sp);
2229 stp_port_set_aux(sp, ofport);
2230 stp_port_set_priority(sp, s->priority);
2231 stp_port_set_path_cost(sp, s->path_cost);
2233 update_stp_port_state(ofport);
2239 get_stp_port_status(struct ofport *ofport_,
2240 struct ofproto_port_stp_status *s)
2242 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2244 struct stp_port *sp = ofport->stp_port;
2246 if (!ofproto->stp || !sp) {
2252 s->port_id = stp_port_get_id(sp);
2253 s->state = stp_port_get_state(sp);
2254 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2255 s->role = stp_port_get_role(sp);
2256 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2262 stp_run(struct ofproto_dpif *ofproto)
2265 long long int now = time_msec();
2266 long long int elapsed = now - ofproto->stp_last_tick;
2267 struct stp_port *sp;
2270 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2271 ofproto->stp_last_tick = now;
2273 while (stp_get_changed_port(ofproto->stp, &sp)) {
2274 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2277 update_stp_port_state(ofport);
2281 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2282 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2283 mac_learning_flush(ofproto->ml);
2284 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2290 stp_wait(struct ofproto_dpif *ofproto)
2293 poll_timer_wait(1000);
2298 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2301 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2304 if (ofport->n_qdscp != n_qdscp
2305 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2306 n_qdscp * sizeof *qdscp))) {
2307 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2308 free(ofport->qdscp);
2309 ofport->qdscp = n_qdscp
2310 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2312 ofport->n_qdscp = n_qdscp;
2320 /* Expires all MAC learning entries associated with 'bundle' and forces its
2321 * ofproto to revalidate every flow.
2323 * Normally MAC learning entries are removed only from the ofproto associated
2324 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2325 * are removed from every ofproto. When patch ports and SLB bonds are in use
2326 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2327 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2328 * with the host from which it migrated. */
2330 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2332 struct ofproto_dpif *ofproto = bundle->ofproto;
2333 struct mac_learning *ml = ofproto->ml;
2334 struct mac_entry *mac, *next_mac;
2336 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2337 ovs_rwlock_wrlock(&ml->rwlock);
2338 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2339 if (mac->port.p == bundle) {
2341 struct ofproto_dpif *o;
2343 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2345 struct mac_entry *e;
2347 ovs_rwlock_wrlock(&o->ml->rwlock);
2348 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2350 mac_learning_expire(o->ml, e);
2352 ovs_rwlock_unlock(&o->ml->rwlock);
2357 mac_learning_expire(ml, mac);
2360 ovs_rwlock_unlock(&ml->rwlock);
2363 static struct ofbundle *
2364 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2366 struct ofbundle *bundle;
2368 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2369 &ofproto->bundles) {
2370 if (bundle->aux == aux) {
2378 bundle_update(struct ofbundle *bundle)
2380 struct ofport_dpif *port;
2382 bundle->floodable = true;
2383 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2384 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2385 || !stp_forward_in_state(port->stp_state)) {
2386 bundle->floodable = false;
2393 bundle_del_port(struct ofport_dpif *port)
2395 struct ofbundle *bundle = port->bundle;
2397 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2399 list_remove(&port->bundle_node);
2400 port->bundle = NULL;
2403 lacp_slave_unregister(bundle->lacp, port);
2406 bond_slave_unregister(bundle->bond, port);
2409 bundle_update(bundle);
2413 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2414 struct lacp_slave_settings *lacp)
2416 struct ofport_dpif *port;
2418 port = get_ofp_port(bundle->ofproto, ofp_port);
2423 if (port->bundle != bundle) {
2424 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2426 bundle_del_port(port);
2429 port->bundle = bundle;
2430 list_push_back(&bundle->ports, &port->bundle_node);
2431 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2432 || !stp_forward_in_state(port->stp_state)) {
2433 bundle->floodable = false;
2437 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2438 lacp_slave_register(bundle->lacp, port, lacp);
2445 bundle_destroy(struct ofbundle *bundle)
2447 struct ofproto_dpif *ofproto;
2448 struct ofport_dpif *port, *next_port;
2454 ofproto = bundle->ofproto;
2455 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2457 xlate_bundle_remove(bundle);
2459 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2460 bundle_del_port(port);
2463 bundle_flush_macs(bundle, true);
2464 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2466 free(bundle->trunks);
2467 lacp_unref(bundle->lacp);
2468 bond_unref(bundle->bond);
2473 bundle_set(struct ofproto *ofproto_, void *aux,
2474 const struct ofproto_bundle_settings *s)
2476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2477 bool need_flush = false;
2478 struct ofport_dpif *port;
2479 struct ofbundle *bundle;
2480 unsigned long *trunks;
2486 bundle_destroy(bundle_lookup(ofproto, aux));
2490 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2491 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2493 bundle = bundle_lookup(ofproto, aux);
2495 bundle = xmalloc(sizeof *bundle);
2497 bundle->ofproto = ofproto;
2498 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2499 hash_pointer(aux, 0));
2501 bundle->name = NULL;
2503 list_init(&bundle->ports);
2504 bundle->vlan_mode = PORT_VLAN_TRUNK;
2506 bundle->trunks = NULL;
2507 bundle->use_priority_tags = s->use_priority_tags;
2508 bundle->lacp = NULL;
2509 bundle->bond = NULL;
2511 bundle->floodable = true;
2512 mbridge_register_bundle(ofproto->mbridge, bundle);
2515 if (!bundle->name || strcmp(s->name, bundle->name)) {
2517 bundle->name = xstrdup(s->name);
2522 if (!bundle->lacp) {
2523 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524 bundle->lacp = lacp_create();
2526 lacp_configure(bundle->lacp, s->lacp);
2528 lacp_unref(bundle->lacp);
2529 bundle->lacp = NULL;
2532 /* Update set of ports. */
2534 for (i = 0; i < s->n_slaves; i++) {
2535 if (!bundle_add_port(bundle, s->slaves[i],
2536 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2540 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2541 struct ofport_dpif *next_port;
2543 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2544 for (i = 0; i < s->n_slaves; i++) {
2545 if (s->slaves[i] == port->up.ofp_port) {
2550 bundle_del_port(port);
2554 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2556 if (list_is_empty(&bundle->ports)) {
2557 bundle_destroy(bundle);
2561 /* Set VLAN tagging mode */
2562 if (s->vlan_mode != bundle->vlan_mode
2563 || s->use_priority_tags != bundle->use_priority_tags) {
2564 bundle->vlan_mode = s->vlan_mode;
2565 bundle->use_priority_tags = s->use_priority_tags;
2570 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2571 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2573 if (vlan != bundle->vlan) {
2574 bundle->vlan = vlan;
2578 /* Get trunked VLANs. */
2579 switch (s->vlan_mode) {
2580 case PORT_VLAN_ACCESS:
2584 case PORT_VLAN_TRUNK:
2585 trunks = CONST_CAST(unsigned long *, s->trunks);
2588 case PORT_VLAN_NATIVE_UNTAGGED:
2589 case PORT_VLAN_NATIVE_TAGGED:
2590 if (vlan != 0 && (!s->trunks
2591 || !bitmap_is_set(s->trunks, vlan)
2592 || bitmap_is_set(s->trunks, 0))) {
2593 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2595 trunks = bitmap_clone(s->trunks, 4096);
2597 trunks = bitmap_allocate1(4096);
2599 bitmap_set1(trunks, vlan);
2600 bitmap_set0(trunks, 0);
2602 trunks = CONST_CAST(unsigned long *, s->trunks);
2609 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2610 free(bundle->trunks);
2611 if (trunks == s->trunks) {
2612 bundle->trunks = vlan_bitmap_clone(trunks);
2614 bundle->trunks = trunks;
2619 if (trunks != s->trunks) {
2624 if (!list_is_short(&bundle->ports)) {
2625 bundle->ofproto->has_bonded_bundles = true;
2627 if (bond_reconfigure(bundle->bond, s->bond)) {
2628 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2631 bundle->bond = bond_create(s->bond);
2632 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2635 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2636 bond_slave_register(bundle->bond, port, port->up.netdev);
2639 bond_unref(bundle->bond);
2640 bundle->bond = NULL;
2643 /* If we changed something that would affect MAC learning, un-learn
2644 * everything on this port and force flow revalidation. */
2646 bundle_flush_macs(bundle, false);
2653 bundle_remove(struct ofport *port_)
2655 struct ofport_dpif *port = ofport_dpif_cast(port_);
2656 struct ofbundle *bundle = port->bundle;
2659 bundle_del_port(port);
2660 if (list_is_empty(&bundle->ports)) {
2661 bundle_destroy(bundle);
2662 } else if (list_is_short(&bundle->ports)) {
2663 bond_unref(bundle->bond);
2664 bundle->bond = NULL;
2670 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2673 struct ofport_dpif *port = port_;
2674 uint8_t ea[ETH_ADDR_LEN];
2677 error = netdev_get_etheraddr(port->up.netdev, ea);
2679 struct ofpbuf packet;
2682 ofpbuf_init(&packet, 0);
2683 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2685 memcpy(packet_pdu, pdu, pdu_size);
2687 send_packet(port, &packet);
2688 ofpbuf_uninit(&packet);
2690 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2691 "%s (%s)", port->bundle->name,
2692 netdev_get_name(port->up.netdev), ovs_strerror(error));
2697 bundle_send_learning_packets(struct ofbundle *bundle)
2699 struct ofproto_dpif *ofproto = bundle->ofproto;
2700 int error, n_packets, n_errors;
2701 struct mac_entry *e;
2703 error = n_packets = n_errors = 0;
2704 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2705 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2706 if (e->port.p != bundle) {
2707 struct ofpbuf *learning_packet;
2708 struct ofport_dpif *port;
2712 /* The assignment to "port" is unnecessary but makes "grep"ing for
2713 * struct ofport_dpif more effective. */
2714 learning_packet = bond_compose_learning_packet(bundle->bond,
2718 ret = send_packet(port, learning_packet);
2719 ofpbuf_delete(learning_packet);
2727 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2731 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2732 "packets, last error was: %s",
2733 bundle->name, n_errors, n_packets, ovs_strerror(error));
2735 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2736 bundle->name, n_packets);
2741 bundle_run(struct ofbundle *bundle)
2744 lacp_run(bundle->lacp, send_pdu_cb);
2747 struct ofport_dpif *port;
2749 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2750 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2753 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2754 bundle->ofproto->backer->need_revalidate = REV_BOND;
2757 if (bond_should_send_learning_packets(bundle->bond)) {
2758 bundle_send_learning_packets(bundle);
2764 bundle_wait(struct ofbundle *bundle)
2767 lacp_wait(bundle->lacp);
2770 bond_wait(bundle->bond);
2777 mirror_set__(struct ofproto *ofproto_, void *aux,
2778 const struct ofproto_mirror_settings *s)
2780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2781 struct ofbundle **srcs, **dsts;
2786 mirror_destroy(ofproto->mbridge, aux);
2790 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2791 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2793 for (i = 0; i < s->n_srcs; i++) {
2794 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2797 for (i = 0; i < s->n_dsts; i++) {
2798 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2801 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2802 s->n_dsts, s->src_vlans,
2803 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2810 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2811 uint64_t *packets, uint64_t *bytes)
2814 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2819 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2822 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2823 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2824 mac_learning_flush(ofproto->ml);
2826 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2831 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2834 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2835 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2839 forward_bpdu_changed(struct ofproto *ofproto_)
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2842 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2846 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2850 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2851 mac_learning_set_idle_time(ofproto->ml, idle_time);
2852 mac_learning_set_max_entries(ofproto->ml, max_entries);
2853 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2858 static struct ofport_dpif *
2859 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2861 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2862 return ofport ? ofport_dpif_cast(ofport) : NULL;
2865 static struct ofport_dpif *
2866 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2868 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2869 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2873 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2874 struct ofproto_port *ofproto_port,
2875 struct dpif_port *dpif_port)
2877 ofproto_port->name = dpif_port->name;
2878 ofproto_port->type = dpif_port->type;
2879 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2883 ofport_update_peer(struct ofport_dpif *ofport)
2885 const struct ofproto_dpif *ofproto;
2886 struct dpif_backer *backer;
2887 const char *peer_name;
2889 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2893 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2894 backer->need_revalidate = REV_RECONFIGURE;
2897 ofport->peer->peer = NULL;
2898 ofport->peer = NULL;
2901 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2906 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2907 struct ofport *peer_ofport;
2908 struct ofport_dpif *peer;
2909 const char *peer_peer;
2911 if (ofproto->backer != backer) {
2915 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2920 peer = ofport_dpif_cast(peer_ofport);
2921 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2922 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2924 ofport->peer = peer;
2925 ofport->peer->peer = ofport;
2933 port_run_fast(struct ofport_dpif *ofport)
2935 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2936 struct ofpbuf packet;
2938 ofpbuf_init(&packet, 0);
2939 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2940 send_packet(ofport, &packet);
2941 ofpbuf_uninit(&packet);
2944 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2945 struct ofpbuf packet;
2947 ofpbuf_init(&packet, 0);
2948 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2949 send_packet(ofport, &packet);
2950 ofpbuf_uninit(&packet);
2955 port_run(struct ofport_dpif *ofport)
2957 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2958 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2959 bool enable = netdev_get_carrier(ofport->up.netdev);
2961 ofport->carrier_seq = carrier_seq;
2963 port_run_fast(ofport);
2966 int cfm_opup = cfm_get_opup(ofport->cfm);
2968 cfm_run(ofport->cfm);
2969 enable = enable && !cfm_get_fault(ofport->cfm);
2971 if (cfm_opup >= 0) {
2972 enable = enable && cfm_opup;
2977 bfd_run(ofport->bfd);
2978 enable = enable && bfd_forwarding(ofport->bfd);
2981 if (ofport->bundle) {
2982 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2983 if (carrier_changed) {
2984 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2988 if (ofport->may_enable != enable) {
2989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2990 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2993 ofport->may_enable = enable;
2997 port_wait(struct ofport_dpif *ofport)
3000 cfm_wait(ofport->cfm);
3004 bfd_wait(ofport->bfd);
3009 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3010 struct ofproto_port *ofproto_port)
3012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3013 struct dpif_port dpif_port;
3016 if (sset_contains(&ofproto->ghost_ports, devname)) {
3017 const char *type = netdev_get_type_from_name(devname);
3019 /* We may be called before ofproto->up.port_by_name is populated with
3020 * the appropriate ofport. For this reason, we must get the name and
3021 * type from the netdev layer directly. */
3023 const struct ofport *ofport;
3025 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3026 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3027 ofproto_port->name = xstrdup(devname);
3028 ofproto_port->type = xstrdup(type);
3034 if (!sset_contains(&ofproto->ports, devname)) {
3037 error = dpif_port_query_by_name(ofproto->backer->dpif,
3038 devname, &dpif_port);
3040 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3046 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3049 const char *devname = netdev_get_name(netdev);
3050 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3051 const char *dp_port_name;
3053 if (netdev_vport_is_patch(netdev)) {
3054 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3058 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3059 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3060 odp_port_t port_no = ODPP_NONE;
3063 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3067 if (netdev_get_tunnel_config(netdev)) {
3068 simap_put(&ofproto->backer->tnl_backers,
3069 dp_port_name, odp_to_u32(port_no));
3073 if (netdev_get_tunnel_config(netdev)) {
3074 sset_add(&ofproto->ghost_ports, devname);
3076 sset_add(&ofproto->ports, devname);
3082 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3085 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3092 sset_find_and_delete(&ofproto->ghost_ports,
3093 netdev_get_name(ofport->up.netdev));
3094 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3095 if (!ofport->is_tunnel) {
3096 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3098 /* The caller is going to close ofport->up.netdev. If this is a
3099 * bonded port, then the bond is using that netdev, so remove it
3100 * from the bond. The client will need to reconfigure everything
3101 * after deleting ports, so then the slave will get re-added. */
3102 bundle_remove(&ofport->up);
3109 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3116 error = netdev_get_stats(ofport->up.netdev, stats);
3118 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3121 /* ofproto->stats.tx_packets represents packets that we created
3122 * internally and sent to some port (e.g. packets sent with
3123 * send_packet()). Account for them as if they had come from
3124 * OFPP_LOCAL and got forwarded. */
3126 if (stats->rx_packets != UINT64_MAX) {
3127 stats->rx_packets += ofproto->stats.tx_packets;
3130 if (stats->rx_bytes != UINT64_MAX) {
3131 stats->rx_bytes += ofproto->stats.tx_bytes;
3134 /* ofproto->stats.rx_packets represents packets that were received on
3135 * some port and we processed internally and dropped (e.g. STP).
3136 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3138 if (stats->tx_packets != UINT64_MAX) {
3139 stats->tx_packets += ofproto->stats.rx_packets;
3142 if (stats->tx_bytes != UINT64_MAX) {
3143 stats->tx_bytes += ofproto->stats.rx_bytes;
3150 struct port_dump_state {
3155 struct ofproto_port port;
3160 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3162 *statep = xzalloc(sizeof(struct port_dump_state));
3167 port_dump_next(const struct ofproto *ofproto_, void *state_,
3168 struct ofproto_port *port)
3170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3171 struct port_dump_state *state = state_;
3172 const struct sset *sset;
3173 struct sset_node *node;
3175 if (state->has_port) {
3176 ofproto_port_destroy(&state->port);
3177 state->has_port = false;
3179 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3180 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3183 error = port_query_by_name(ofproto_, node->name, &state->port);
3185 *port = state->port;
3186 state->has_port = true;
3188 } else if (error != ENODEV) {
3193 if (!state->ghost) {
3194 state->ghost = true;
3197 return port_dump_next(ofproto_, state_, port);
3204 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3206 struct port_dump_state *state = state_;
3208 if (state->has_port) {
3209 ofproto_port_destroy(&state->port);
3216 port_poll(const struct ofproto *ofproto_, char **devnamep)
3218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3220 if (ofproto->port_poll_errno) {
3221 int error = ofproto->port_poll_errno;
3222 ofproto->port_poll_errno = 0;
3226 if (sset_is_empty(&ofproto->port_poll_set)) {
3230 *devnamep = sset_pop(&ofproto->port_poll_set);
3235 port_poll_wait(const struct ofproto *ofproto_)
3237 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3238 dpif_port_poll_wait(ofproto->backer->dpif);
3242 port_is_lacp_current(const struct ofport *ofport_)
3244 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3245 return (ofport->bundle && ofport->bundle->lacp
3246 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3250 /* Upcall handling. */
3252 /* Flow miss batching.
3254 * Some dpifs implement operations faster when you hand them off in a batch.
3255 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3256 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3257 * more packets, plus possibly installing the flow in the dpif.
3259 * So far we only batch the operations that affect flow setup time the most.
3260 * It's possible to batch more than that, but the benefit might be minimal. */
3262 struct hmap_node hmap_node;
3263 struct ofproto_dpif *ofproto;
3265 enum odp_key_fitness key_fitness;
3266 const struct nlattr *key;
3268 struct list packets;
3269 enum dpif_upcall_type upcall_type;
3272 struct flow_miss_op {
3273 struct dpif_op dpif_op;
3275 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3276 struct xlate_out xout;
3277 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3279 struct ofpbuf mask; /* Flow mask for "put" ops. */
3280 struct odputil_keybuf maskbuf;
3282 /* If this is a "put" op, then a pointer to the subfacet that should
3283 * be marked as uninstalled if the operation fails. */
3284 struct subfacet *subfacet;
3287 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3288 * OpenFlow controller as necessary according to their individual
3289 * configurations. */
3291 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3292 const struct flow *flow)
3294 struct ofputil_packet_in pin;
3296 pin.packet = packet->data;
3297 pin.packet_len = packet->size;
3298 pin.reason = OFPR_NO_MATCH;
3299 pin.controller_id = 0;
3304 pin.send_len = 0; /* not used for flow table misses */
3306 flow_get_metadata(flow, &pin.fmd);
3308 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3311 static struct flow_miss *
3312 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3313 const struct flow *flow, uint32_t hash)
3315 struct flow_miss *miss;
3317 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3318 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3326 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3327 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3328 * 'miss' is associated with a subfacet the caller must also initialize the
3329 * returned op->subfacet, and if anything needs to be freed after processing
3330 * the op, the caller must initialize op->garbage also. */
3332 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3333 struct flow_miss_op *op)
3335 if (miss->flow.in_port.ofp_port
3336 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3337 miss->flow.vlan_tci)) {
3338 /* This packet was received on a VLAN splinter port. We
3339 * added a VLAN to the packet to make the packet resemble
3340 * the flow, but the actions were composed assuming that
3341 * the packet contained no VLAN. So, we must remove the
3342 * VLAN header from the packet before trying to execute the
3344 eth_pop_vlan(packet);
3347 op->subfacet = NULL;
3348 op->xout_garbage = false;
3349 op->dpif_op.type = DPIF_OP_EXECUTE;
3350 op->dpif_op.u.execute.key = miss->key;
3351 op->dpif_op.u.execute.key_len = miss->key_len;
3352 op->dpif_op.u.execute.packet = packet;
3353 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3356 /* Helper for handle_flow_miss_without_facet() and
3357 * handle_flow_miss_with_facet(). */
3359 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3360 const struct flow *flow, bool fail_open)
3364 * Extra-special case for fail-open mode.
3366 * We are in fail-open mode and the packet matched the fail-open
3367 * rule, but we are connected to a controller too. We should send
3368 * the packet up to the controller in the hope that it will try to
3369 * set up a flow and thereby allow us to exit fail-open.
3371 * See the top-level comment in fail-open.c for more information.
3373 send_packet_in_miss(ofproto, packet, flow);
3377 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3378 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3379 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3380 * return value of true). However, for short flows the cost of bookkeeping is
3381 * much higher than the benefits, so when the datapath holds a large number of
3382 * flows we impose some heuristics to decide which flows are likely to be worth
3385 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3387 struct dpif_backer *backer = miss->ofproto->backer;
3390 switch (flow_miss_model) {
3391 case OFPROTO_HANDLE_MISS_AUTO:
3393 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3395 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3399 if (!backer->governor) {
3402 n_subfacets = hmap_count(&backer->subfacets);
3403 if (n_subfacets * 2 <= flow_eviction_threshold) {
3407 backer->governor = governor_create();
3410 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3411 return governor_should_install_flow(backer->governor, hash,
3412 list_size(&miss->packets));
3415 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3416 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3417 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3419 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3420 struct flow_miss *miss,
3421 struct flow_miss_op *ops, size_t *n_ops)
3423 struct ofpbuf *packet;
3425 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3427 COVERAGE_INC(facet_suppress);
3429 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3430 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3433 struct xlate_in xin;
3435 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3436 xlate_actions_for_side_effects(&xin);
3439 if (xout->odp_actions.size) {
3440 struct flow_miss_op *op = &ops[*n_ops];
3441 struct dpif_execute *execute = &op->dpif_op.u.execute;
3443 init_flow_miss_execute_op(miss, packet, op);
3444 xlate_out_copy(&op->xout, xout);
3445 execute->actions = op->xout.odp_actions.data;
3446 execute->actions_len = op->xout.odp_actions.size;
3447 op->xout_garbage = true;
3454 /* Handles 'miss', which matches 'facet'. May add any required datapath
3455 * operations to 'ops', incrementing '*n_ops' for each new op.
3457 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3458 * This is really important only for new facets: if we just called time_msec()
3459 * here, then the new subfacet or its packets could look (occasionally) as
3460 * though it was used some time after the facet was used. That can make a
3461 * one-packet flow look like it has a nonzero duration, which looks odd in
3462 * e.g. NetFlow statistics.
3464 * If non-null, 'stats' will be folded into 'facet'. */
3466 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3467 long long int now, struct dpif_flow_stats *stats,
3468 struct flow_miss_op *ops, size_t *n_ops)
3470 enum subfacet_path want_path;
3471 struct subfacet *subfacet;
3472 struct ofpbuf *packet;
3474 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3476 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3477 struct flow_miss_op *op = &ops[*n_ops];
3479 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3482 if (want_path != SF_FAST_PATH) {
3483 struct rule_dpif *rule;
3484 struct xlate_in xin;
3486 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3487 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3488 xlate_actions_for_side_effects(&xin);
3491 if (facet->xout.odp_actions.size) {
3492 struct dpif_execute *execute = &op->dpif_op.u.execute;
3494 init_flow_miss_execute_op(miss, packet, op);
3495 execute->actions = facet->xout.odp_actions.data,
3496 execute->actions_len = facet->xout.odp_actions.size;
3501 /* Don't install the flow if it's the result of the "userspace"
3502 * action for an already installed facet. This can occur when a
3503 * datapath flow with wildcards has a "userspace" action and flows
3504 * sent to userspace result in a different subfacet, which will then
3505 * be rejected as overlapping by the datapath. */
3506 if (miss->upcall_type == DPIF_UC_ACTION
3507 && !list_is_empty(&facet->subfacets)) {
3509 facet->used = MAX(facet->used, stats->used);
3510 facet->packet_count += stats->n_packets;
3511 facet->byte_count += stats->n_bytes;
3512 facet->tcp_flags |= stats->tcp_flags;
3517 subfacet = subfacet_create(facet, miss, now);
3519 subfacet_update_stats(subfacet, stats);
3522 if (subfacet->path != want_path) {
3523 struct flow_miss_op *op = &ops[(*n_ops)++];
3524 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3526 subfacet->path = want_path;
3528 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3529 if (enable_megaflows) {
3530 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3531 &miss->flow, UINT32_MAX);
3534 op->xout_garbage = false;
3535 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3536 op->subfacet = subfacet;
3537 put->flags = DPIF_FP_CREATE;
3538 put->key = miss->key;
3539 put->key_len = miss->key_len;
3540 put->mask = op->mask.data;
3541 put->mask_len = op->mask.size;
3543 if (want_path == SF_FAST_PATH) {
3544 put->actions = facet->xout.odp_actions.data;
3545 put->actions_len = facet->xout.odp_actions.size;
3547 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3548 op->slow_stub, sizeof op->slow_stub,
3549 &put->actions, &put->actions_len);
3555 /* Handles flow miss 'miss'. May add any required datapath operations
3556 * to 'ops', incrementing '*n_ops' for each new op. */
3558 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3561 struct ofproto_dpif *ofproto = miss->ofproto;
3562 struct dpif_flow_stats stats__;
3563 struct dpif_flow_stats *stats = &stats__;
3564 struct ofpbuf *packet;
3565 struct facet *facet;
3569 memset(stats, 0, sizeof *stats);
3571 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3572 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3573 stats->n_bytes += packet->size;
3577 facet = facet_lookup_valid(ofproto, &miss->flow);
3579 struct flow_wildcards wc;
3580 struct rule_dpif *rule;
3581 struct xlate_out xout;
3582 struct xlate_in xin;
3584 flow_wildcards_init_catchall(&wc);
3585 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3586 rule_credit_stats(rule, stats);
3588 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3590 xin.resubmit_stats = stats;
3591 xin.may_learn = true;
3592 xlate_actions(&xin, &xout);
3593 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3595 /* There does not exist a bijection between 'struct flow' and datapath
3596 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3597 * assumption used throughout the facet and subfacet handling code.
3598 * Since we have to handle these misses in userspace anyway, we simply
3599 * skip facet creation, avoiding the problem altogether. */
3600 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3601 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3602 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3606 facet = facet_create(miss, rule, &xout, stats);
3609 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3612 static struct drop_key *
3613 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3616 struct drop_key *drop_key;
3618 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3619 &backer->drop_keys) {
3620 if (drop_key->key_len == key_len
3621 && !memcmp(drop_key->key, key, key_len)) {
3629 drop_key_clear(struct dpif_backer *backer)
3631 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3632 struct drop_key *drop_key, *next;
3634 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3637 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3639 if (error && !VLOG_DROP_WARN(&rl)) {
3640 struct ds ds = DS_EMPTY_INITIALIZER;
3641 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3642 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3643 ovs_strerror(error), ds_cstr(&ds));
3647 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3648 free(drop_key->key);
3654 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3657 struct dpif_upcall *upcall;
3658 struct flow_miss *miss;
3659 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3660 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3661 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3671 /* Construct the to-do list.
3673 * This just amounts to extracting the flow from each packet and sticking
3674 * the packets that have the same flow in the same "flow_miss" structure so
3675 * that we can process them together. */
3678 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3679 struct flow_miss *miss = &misses[n_misses];
3680 struct flow_miss *existing_miss;
3681 struct ofproto_dpif *ofproto;
3682 odp_port_t odp_in_port;
3687 error = xlate_receive(backer, upcall->packet, upcall->key,
3688 upcall->key_len, &flow, &miss->key_fitness,
3689 &ofproto, &odp_in_port);
3690 if (error == ENODEV) {
3691 struct drop_key *drop_key;
3693 /* Received packet on datapath port for which we couldn't
3694 * associate an ofproto. This can happen if a port is removed
3695 * while traffic is being received. Print a rate-limited message
3696 * in case it happens frequently. Install a drop flow so
3697 * that future packets of the flow are inexpensively dropped
3699 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3700 "%"PRIu32, odp_in_port);
3702 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3705 ret = dpif_flow_put(backer->dpif,
3706 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3707 upcall->key, upcall->key_len,
3708 NULL, 0, NULL, 0, NULL);
3711 drop_key = xmalloc(sizeof *drop_key);
3712 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3713 drop_key->key_len = upcall->key_len;
3715 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3716 hash_bytes(drop_key->key, drop_key->key_len, 0));
3725 ofproto->n_missed++;
3726 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3727 &flow.tunnel, &flow.in_port, &miss->flow);
3729 /* Add other packets to a to-do list. */
3730 hash = flow_hash(&miss->flow, 0);
3731 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3732 if (!existing_miss) {
3733 hmap_insert(&todo, &miss->hmap_node, hash);
3734 miss->ofproto = ofproto;
3735 miss->key = upcall->key;
3736 miss->key_len = upcall->key_len;
3737 miss->upcall_type = upcall->type;
3738 list_init(&miss->packets);
3742 miss = existing_miss;
3744 list_push_back(&miss->packets, &upcall->packet->list_node);
3747 /* Process each element in the to-do list, constructing the set of
3748 * operations to batch. */
3750 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3751 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3753 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3755 /* Execute batch. */
3756 for (i = 0; i < n_ops; i++) {
3757 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3759 dpif_operate(backer->dpif, dpif_ops, n_ops);
3761 for (i = 0; i < n_ops; i++) {
3762 if (dpif_ops[i]->error != 0
3763 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3764 && flow_miss_ops[i].subfacet) {
3765 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3767 COVERAGE_INC(subfacet_install_fail);
3769 /* Zero-out subfacet counters when installation failed, but
3770 * datapath reported hits. This should not happen and
3771 * indicates a bug, since if the datapath flow exists, we
3772 * should not be attempting to create a new subfacet. A
3773 * buggy datapath could trigger this, so just zero out the
3774 * counters and log an error. */
3775 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3776 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3777 "datapath reported hits");
3778 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3781 subfacet->path = SF_NOT_INSTALLED;
3785 if (flow_miss_ops[i].xout_garbage) {
3786 xlate_out_uninit(&flow_miss_ops[i].xout);
3789 hmap_destroy(&todo);
3792 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3794 classify_upcall(const struct dpif_upcall *upcall)
3796 size_t userdata_len;
3797 union user_action_cookie cookie;
3799 /* First look at the upcall type. */
3800 switch (upcall->type) {
3801 case DPIF_UC_ACTION:
3807 case DPIF_N_UC_TYPES:
3809 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3813 /* "action" upcalls need a closer look. */
3814 if (!upcall->userdata) {
3815 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3818 userdata_len = nl_attr_get_size(upcall->userdata);
3819 if (userdata_len < sizeof cookie.type
3820 || userdata_len > sizeof cookie) {
3821 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3825 memset(&cookie, 0, sizeof cookie);
3826 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3827 if (userdata_len == sizeof cookie.sflow
3828 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3829 return SFLOW_UPCALL;
3830 } else if (userdata_len == sizeof cookie.slow_path
3831 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3833 } else if (userdata_len == sizeof cookie.flow_sample
3834 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3835 return FLOW_SAMPLE_UPCALL;
3836 } else if (userdata_len == sizeof cookie.ipfix
3837 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3838 return IPFIX_UPCALL;
3840 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3841 " and size %zu", cookie.type, userdata_len);
3847 handle_sflow_upcall(struct dpif_backer *backer,
3848 const struct dpif_upcall *upcall)
3850 struct ofproto_dpif *ofproto;
3851 union user_action_cookie cookie;
3853 odp_port_t odp_in_port;
3855 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3856 &flow, NULL, &ofproto, &odp_in_port)
3857 || !ofproto->sflow) {
3861 memset(&cookie, 0, sizeof cookie);
3862 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3863 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3864 odp_in_port, &cookie);
3868 handle_flow_sample_upcall(struct dpif_backer *backer,
3869 const struct dpif_upcall *upcall)
3871 struct ofproto_dpif *ofproto;
3872 union user_action_cookie cookie;
3875 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3876 &flow, NULL, &ofproto, NULL)
3877 || !ofproto->ipfix) {
3881 memset(&cookie, 0, sizeof cookie);
3882 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3884 /* The flow reflects exactly the contents of the packet. Sample
3885 * the packet using it. */
3886 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3887 cookie.flow_sample.collector_set_id,
3888 cookie.flow_sample.probability,
3889 cookie.flow_sample.obs_domain_id,
3890 cookie.flow_sample.obs_point_id);
3894 handle_ipfix_upcall(struct dpif_backer *backer,
3895 const struct dpif_upcall *upcall)
3897 struct ofproto_dpif *ofproto;
3900 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3901 &flow, NULL, &ofproto, NULL)
3902 || !ofproto->ipfix) {
3906 /* The flow reflects exactly the contents of the packet. Sample
3907 * the packet using it. */
3908 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3912 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3914 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3915 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3916 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3921 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3924 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3925 struct dpif_upcall *upcall = &misses[n_misses];
3926 struct ofpbuf *buf = &miss_bufs[n_misses];
3929 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3930 sizeof miss_buf_stubs[n_misses]);
3931 error = dpif_recv(backer->dpif, upcall, buf);
3937 switch (classify_upcall(upcall)) {
3939 /* Handle it later. */
3944 handle_sflow_upcall(backer, upcall);
3948 case FLOW_SAMPLE_UPCALL:
3949 handle_flow_sample_upcall(backer, upcall);
3954 handle_ipfix_upcall(backer, upcall);
3964 /* Handle deferred MISS_UPCALL processing. */
3965 handle_miss_upcalls(backer, misses, n_misses);
3966 for (i = 0; i < n_misses; i++) {
3967 ofpbuf_uninit(&miss_bufs[i]);
3973 /* Flow expiration. */
3975 static int subfacet_max_idle(const struct dpif_backer *);
3976 static void update_stats(struct dpif_backer *);
3977 static void rule_expire(struct rule_dpif *);
3978 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3980 /* This function is called periodically by run(). Its job is to collect
3981 * updates for the flows that have been installed into the datapath, most
3982 * importantly when they last were used, and then use that information to
3983 * expire flows that have not been used recently.
3985 * Returns the number of milliseconds after which it should be called again. */
3987 expire(struct dpif_backer *backer)
3989 struct ofproto_dpif *ofproto;
3993 /* Periodically clear out the drop keys in an effort to keep them
3994 * relatively few. */
3995 drop_key_clear(backer);
3997 /* Update stats for each flow in the backer. */
3998 update_stats(backer);
4000 n_subfacets = hmap_count(&backer->subfacets);
4002 struct subfacet *subfacet;
4003 long long int total, now;
4007 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4008 total += now - subfacet->created;
4010 backer->avg_subfacet_life += total / n_subfacets;
4012 backer->avg_subfacet_life /= 2;
4014 backer->avg_n_subfacet += n_subfacets;
4015 backer->avg_n_subfacet /= 2;
4017 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4019 max_idle = subfacet_max_idle(backer);
4020 expire_subfacets(backer, max_idle);
4022 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4023 struct rule *rule, *next_rule;
4025 if (ofproto->backer != backer) {
4029 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4031 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4032 &ofproto->up.expirable) {
4033 rule_expire(rule_dpif_cast(rule));
4036 /* All outstanding data in existing flows has been accounted, so it's a
4037 * good time to do bond rebalancing. */
4038 if (ofproto->has_bonded_bundles) {
4039 struct ofbundle *bundle;
4041 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4043 bond_rebalance(bundle->bond);
4049 return MIN(max_idle, 1000);
4052 /* Updates flow table statistics given that the datapath just reported 'stats'
4053 * as 'subfacet''s statistics. */
4055 update_subfacet_stats(struct subfacet *subfacet,
4056 const struct dpif_flow_stats *stats)
4058 struct facet *facet = subfacet->facet;
4059 struct dpif_flow_stats diff;
4061 diff.tcp_flags = stats->tcp_flags;
4062 diff.used = stats->used;
4064 if (stats->n_packets >= subfacet->dp_packet_count) {
4065 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4067 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4071 if (stats->n_bytes >= subfacet->dp_byte_count) {
4072 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4074 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4078 facet->ofproto->n_hit += diff.n_packets;
4079 subfacet->dp_packet_count = stats->n_packets;
4080 subfacet->dp_byte_count = stats->n_bytes;
4081 subfacet_update_stats(subfacet, &diff);
4083 if (facet->accounted_bytes < facet->byte_count) {
4085 facet_account(facet);
4086 facet->accounted_bytes = facet->byte_count;
4090 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4091 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4093 delete_unexpected_flow(struct dpif_backer *backer,
4094 const struct nlattr *key, size_t key_len)
4096 if (!VLOG_DROP_WARN(&rl)) {
4100 odp_flow_key_format(key, key_len, &s);
4101 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4105 COVERAGE_INC(facet_unexpected);
4106 dpif_flow_del(backer->dpif, key, key_len, NULL);
4109 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4111 * This function also pushes statistics updates to rules which each facet
4112 * resubmits into. Generally these statistics will be accurate. However, if a
4113 * facet changes the rule it resubmits into at some time in between
4114 * update_stats() runs, it is possible that statistics accrued to the
4115 * old rule will be incorrectly attributed to the new rule. This could be
4116 * avoided by calling update_stats() whenever rules are created or
4117 * deleted. However, the performance impact of making so many calls to the
4118 * datapath do not justify the benefit of having perfectly accurate statistics.
4120 * In addition, this function maintains per ofproto flow hit counts. The patch
4121 * port is not treated specially. e.g. A packet ingress from br0 patched into
4122 * br1 will increase the hit count of br0 by 1, however, does not affect
4123 * the hit or miss counts of br1.
4126 update_stats(struct dpif_backer *backer)
4128 const struct dpif_flow_stats *stats;
4129 struct dpif_flow_dump dump;
4130 const struct nlattr *key, *mask;
4131 size_t key_len, mask_len;
4133 dpif_flow_dump_start(&dump, backer->dpif);
4134 while (dpif_flow_dump_next(&dump, &key, &key_len,
4135 &mask, &mask_len, NULL, NULL, &stats)) {
4136 struct subfacet *subfacet;
4139 key_hash = odp_flow_key_hash(key, key_len);
4140 subfacet = subfacet_find(backer, key, key_len, key_hash);
4141 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4143 update_subfacet_stats(subfacet, stats);
4147 /* Stats are updated per-packet. */
4150 case SF_NOT_INSTALLED:
4152 delete_unexpected_flow(backer, key, key_len);
4157 dpif_flow_dump_done(&dump);
4159 update_moving_averages(backer);
4162 /* Calculates and returns the number of milliseconds of idle time after which
4163 * subfacets should expire from the datapath. When a subfacet expires, we fold
4164 * its statistics into its facet, and when a facet's last subfacet expires, we
4165 * fold its statistic into its rule. */
4167 subfacet_max_idle(const struct dpif_backer *backer)
4170 * Idle time histogram.
4172 * Most of the time a switch has a relatively small number of subfacets.
4173 * When this is the case we might as well keep statistics for all of them
4174 * in userspace and to cache them in the kernel datapath for performance as
4177 * As the number of subfacets increases, the memory required to maintain
4178 * statistics about them in userspace and in the kernel becomes
4179 * significant. However, with a large number of subfacets it is likely
4180 * that only a few of them are "heavy hitters" that consume a large amount
4181 * of bandwidth. At this point, only heavy hitters are worth caching in
4182 * the kernel and maintaining in userspaces; other subfacets we can
4185 * The technique used to compute the idle time is to build a histogram with
4186 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4187 * that is installed in the kernel gets dropped in the appropriate bucket.
4188 * After the histogram has been built, we compute the cutoff so that only
4189 * the most-recently-used 1% of subfacets (but at least
4190 * flow_eviction_threshold flows) are kept cached. At least
4191 * the most-recently-used bucket of subfacets is kept, so actually an
4192 * arbitrary number of subfacets can be kept in any given expiration run
4193 * (though the next run will delete most of those unless they receive
4196 * This requires a second pass through the subfacets, in addition to the
4197 * pass made by update_stats(), because the former function never looks at
4198 * uninstallable subfacets.
4200 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4201 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4202 int buckets[N_BUCKETS] = { 0 };
4203 int total, subtotal, bucket;
4204 struct subfacet *subfacet;
4208 total = hmap_count(&backer->subfacets);
4209 if (total <= flow_eviction_threshold) {
4210 return N_BUCKETS * BUCKET_WIDTH;
4213 /* Build histogram. */
4215 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4216 long long int idle = now - subfacet->used;
4217 int bucket = (idle <= 0 ? 0
4218 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4219 : (unsigned int) idle / BUCKET_WIDTH);
4223 /* Find the first bucket whose flows should be expired. */
4224 subtotal = bucket = 0;
4226 subtotal += buckets[bucket++];
4227 } while (bucket < N_BUCKETS &&
4228 subtotal < MAX(flow_eviction_threshold, total / 100));
4230 if (VLOG_IS_DBG_ENABLED()) {
4234 ds_put_cstr(&s, "keep");
4235 for (i = 0; i < N_BUCKETS; i++) {
4237 ds_put_cstr(&s, ", drop");
4240 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4243 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4247 return bucket * BUCKET_WIDTH;
4251 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4253 /* Cutoff time for most flows. */
4254 long long int normal_cutoff = time_msec() - dp_max_idle;
4256 /* We really want to keep flows for special protocols around, so use a more
4257 * conservative cutoff. */
4258 long long int special_cutoff = time_msec() - 10000;
4260 struct subfacet *subfacet, *next_subfacet;
4261 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4265 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4266 &backer->subfacets) {
4267 long long int cutoff;
4269 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4273 if (subfacet->used < cutoff) {
4274 if (subfacet->path != SF_NOT_INSTALLED) {
4275 batch[n_batch++] = subfacet;
4276 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4277 subfacet_destroy_batch(backer, batch, n_batch);
4281 subfacet_destroy(subfacet);
4287 subfacet_destroy_batch(backer, batch, n_batch);
4291 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4292 * then delete it entirely. */
4294 rule_expire(struct rule_dpif *rule)
4299 if (rule->up.pending) {
4300 /* We'll have to expire it later. */
4304 /* Has 'rule' expired? */
4306 if (rule->up.hard_timeout
4307 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4308 reason = OFPRR_HARD_TIMEOUT;
4309 } else if (rule->up.idle_timeout
4310 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4311 reason = OFPRR_IDLE_TIMEOUT;
4316 COVERAGE_INC(ofproto_dpif_expired);
4318 /* Get rid of the rule. */
4319 ofproto_rule_expire(&rule->up, reason);
4324 /* Creates and returns a new facet based on 'miss'.
4326 * The caller must already have determined that no facet with an identical
4327 * 'miss->flow' exists in 'miss->ofproto'.
4329 * 'rule' and 'xout' must have been created based on 'miss'.
4331 * 'facet'' statistics are initialized based on 'stats'.
4333 * The facet will initially have no subfacets. The caller should create (at
4334 * least) one subfacet with subfacet_create(). */
4335 static struct facet *
4336 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4337 struct xlate_out *xout, struct dpif_flow_stats *stats)
4339 struct ofproto_dpif *ofproto = miss->ofproto;
4340 struct facet *facet;
4343 facet = xzalloc(sizeof *facet);
4344 facet->ofproto = miss->ofproto;
4345 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4346 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4347 facet->tcp_flags = stats->tcp_flags;
4348 facet->used = stats->used;
4349 facet->flow = miss->flow;
4350 facet->learn_rl = time_msec() + 500;
4352 list_init(&facet->subfacets);
4353 netflow_flow_init(&facet->nf_flow);
4354 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4356 xlate_out_copy(&facet->xout, xout);
4358 match_init(&match, &facet->flow, &facet->xout.wc);
4359 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4360 classifier_insert(&ofproto->facets, &facet->cr);
4362 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4363 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4369 facet_free(struct facet *facet)
4372 xlate_out_uninit(&facet->xout);
4377 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4378 * 'packet', which arrived on 'in_port'. */
4380 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4381 const struct nlattr *odp_actions, size_t actions_len,
4382 struct ofpbuf *packet)
4384 struct odputil_keybuf keybuf;
4388 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4389 odp_flow_key_from_flow(&key, flow,
4390 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4392 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4393 odp_actions, actions_len, packet);
4397 /* Remove 'facet' from its ofproto and free up the associated memory:
4399 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4400 * rule's statistics, via subfacet_uninstall().
4402 * - Removes 'facet' from its rule and from ofproto->facets.
4405 facet_remove(struct facet *facet)
4407 struct subfacet *subfacet, *next_subfacet;
4409 ovs_assert(!list_is_empty(&facet->subfacets));
4411 /* First uninstall all of the subfacets to get final statistics. */
4412 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4413 subfacet_uninstall(subfacet);
4416 /* Flush the final stats to the rule.
4418 * This might require us to have at least one subfacet around so that we
4419 * can use its actions for accounting in facet_account(), which is why we
4420 * have uninstalled but not yet destroyed the subfacets. */
4421 facet_flush_stats(facet);
4423 /* Now we're really all done so destroy everything. */
4424 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4425 &facet->subfacets) {
4426 subfacet_destroy__(subfacet);
4428 classifier_remove(&facet->ofproto->facets, &facet->cr);
4429 cls_rule_destroy(&facet->cr);
4433 /* Feed information from 'facet' back into the learning table to keep it in
4434 * sync with what is actually flowing through the datapath. */
4436 facet_learn(struct facet *facet)
4438 long long int now = time_msec();
4440 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4444 facet->learn_rl = now + 500;
4446 if (!facet->xout.has_learn
4447 && !facet->xout.has_normal
4448 && (!facet->xout.has_fin_timeout
4449 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4453 facet_push_stats(facet, true);
4457 facet_account(struct facet *facet)
4459 const struct nlattr *a;
4464 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4467 n_bytes = facet->byte_count - facet->accounted_bytes;
4469 /* This loop feeds byte counters to bond_account() for rebalancing to use
4470 * as a basis. We also need to track the actual VLAN on which the packet
4471 * is going to be sent to ensure that it matches the one passed to
4472 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4475 * We use the actions from an arbitrary subfacet because they should all
4476 * be equally valid for our purpose. */
4477 vlan_tci = facet->flow.vlan_tci;
4478 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4479 facet->xout.odp_actions.size) {
4480 const struct ovs_action_push_vlan *vlan;
4481 struct ofport_dpif *port;
4483 switch (nl_attr_type(a)) {
4484 case OVS_ACTION_ATTR_OUTPUT:
4485 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4486 if (port && port->bundle && port->bundle->bond) {
4487 bond_account(port->bundle->bond, &facet->flow,
4488 vlan_tci_to_vid(vlan_tci), n_bytes);
4492 case OVS_ACTION_ATTR_POP_VLAN:
4493 vlan_tci = htons(0);
4496 case OVS_ACTION_ATTR_PUSH_VLAN:
4497 vlan = nl_attr_get(a);
4498 vlan_tci = vlan->vlan_tci;
4504 /* Returns true if the only action for 'facet' is to send to the controller.
4505 * (We don't report NetFlow expiration messages for such facets because they
4506 * are just part of the control logic for the network, not real traffic). */
4508 facet_is_controller_flow(struct facet *facet)
4511 struct ofproto_dpif *ofproto = facet->ofproto;
4512 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4514 const struct ofpact *ofpacts = rule->up.ofpacts;
4515 size_t ofpacts_len = rule->up.ofpacts_len;
4517 if (ofpacts_len > 0 &&
4518 ofpacts->type == OFPACT_CONTROLLER &&
4519 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4526 /* Folds all of 'facet''s statistics into its rule. Also updates the
4527 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4528 * 'facet''s statistics in the datapath should have been zeroed and folded into
4529 * its packet and byte counts before this function is called. */
4531 facet_flush_stats(struct facet *facet)
4533 struct ofproto_dpif *ofproto = facet->ofproto;
4534 struct subfacet *subfacet;
4536 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4537 ovs_assert(!subfacet->dp_byte_count);
4538 ovs_assert(!subfacet->dp_packet_count);
4541 facet_push_stats(facet, false);
4542 if (facet->accounted_bytes < facet->byte_count) {
4543 facet_account(facet);
4544 facet->accounted_bytes = facet->byte_count;
4547 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4548 struct ofexpired expired;
4549 expired.flow = facet->flow;
4550 expired.packet_count = facet->packet_count;
4551 expired.byte_count = facet->byte_count;
4552 expired.used = facet->used;
4553 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4556 /* Reset counters to prevent double counting if 'facet' ever gets
4558 facet_reset_counters(facet);
4560 netflow_flow_clear(&facet->nf_flow);
4561 facet->tcp_flags = 0;
4564 /* Searches 'ofproto''s table of facets for one which would be responsible for
4565 * 'flow'. Returns it if found, otherwise a null pointer.
4567 * The returned facet might need revalidation; use facet_lookup_valid()
4568 * instead if that is important. */
4569 static struct facet *
4570 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4572 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4573 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4576 /* Searches 'ofproto''s table of facets for one capable that covers
4577 * 'flow'. Returns it if found, otherwise a null pointer.
4579 * The returned facet is guaranteed to be valid. */
4580 static struct facet *
4581 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4583 struct facet *facet;
4585 facet = facet_find(ofproto, flow);
4587 && ofproto->backer->need_revalidate
4588 && !facet_revalidate(facet)) {
4596 facet_check_consistency(struct facet *facet)
4598 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4600 struct xlate_out xout;
4601 struct xlate_in xin;
4603 struct rule_dpif *rule;
4606 /* Check the datapath actions for consistency. */
4607 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4608 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4609 xlate_actions(&xin, &xout);
4611 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4612 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4613 && facet->xout.slow == xout.slow
4614 && facet->fail_open == fail_open;
4615 if (!ok && !VLOG_DROP_WARN(&rl)) {
4616 struct ds s = DS_EMPTY_INITIALIZER;
4618 flow_format(&s, &facet->flow);
4619 ds_put_cstr(&s, ": inconsistency in facet");
4621 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4622 ds_put_cstr(&s, " (actions were: ");
4623 format_odp_actions(&s, facet->xout.odp_actions.data,
4624 facet->xout.odp_actions.size);
4625 ds_put_cstr(&s, ") (correct actions: ");
4626 format_odp_actions(&s, xout.odp_actions.data,
4627 xout.odp_actions.size);
4628 ds_put_char(&s, ')');
4631 if (facet->xout.slow != xout.slow) {
4632 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4635 if (facet->fail_open != fail_open) {
4636 ds_put_format(&s, " fail open incorrect. should be %s",
4637 fail_open ? "true" : "false");
4641 xlate_out_uninit(&xout);
4646 /* Re-searches the classifier for 'facet':
4648 * - If the rule found is different from 'facet''s current rule, moves
4649 * 'facet' to the new rule and recompiles its actions.
4651 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4652 * where it is and recompiles its actions anyway.
4654 * - If any of 'facet''s subfacets correspond to a new flow according to
4655 * xlate_receive(), 'facet' is removed.
4657 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4659 facet_revalidate(struct facet *facet)
4661 struct ofproto_dpif *ofproto = facet->ofproto;
4662 struct rule_dpif *new_rule;
4663 struct subfacet *subfacet;
4664 struct flow_wildcards wc;
4665 struct xlate_out xout;
4666 struct xlate_in xin;
4668 COVERAGE_INC(facet_revalidate);
4670 /* Check that child subfacets still correspond to this facet. Tunnel
4671 * configuration changes could cause a subfacet's OpenFlow in_port to
4673 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4674 struct ofproto_dpif *recv_ofproto;
4675 struct flow recv_flow;
4678 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4679 subfacet->key_len, &recv_flow, NULL,
4680 &recv_ofproto, NULL);
4682 || recv_ofproto != ofproto
4683 || facet != facet_find(ofproto, &recv_flow)) {
4684 facet_remove(facet);
4689 flow_wildcards_init_catchall(&wc);
4690 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4692 /* Calculate new datapath actions.
4694 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4695 * emit a NetFlow expiration and, if so, we need to have the old state
4696 * around to properly compose it. */
4697 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4698 xlate_actions(&xin, &xout);
4699 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4701 /* A facet's slow path reason should only change under dramatic
4702 * circumstances. Rather than try to update everything, it's simpler to
4703 * remove the facet and start over.
4705 * More importantly, if a facet's wildcards change, it will be relatively
4706 * difficult to figure out if its subfacets still belong to it, and if not
4707 * which facet they may belong to. Again, to avoid the complexity, we
4708 * simply give up instead. */
4709 if (facet->xout.slow != xout.slow
4710 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4711 facet_remove(facet);
4712 xlate_out_uninit(&xout);
4716 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4717 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4718 if (subfacet->path == SF_FAST_PATH) {
4719 struct dpif_flow_stats stats;
4721 subfacet_install(subfacet, &xout.odp_actions, &stats);
4722 subfacet_update_stats(subfacet, &stats);
4726 facet_flush_stats(facet);
4728 ofpbuf_clear(&facet->xout.odp_actions);
4729 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4730 xout.odp_actions.size);
4733 /* Update 'facet' now that we've taken care of all the old state. */
4734 facet->xout.slow = xout.slow;
4735 facet->xout.has_learn = xout.has_learn;
4736 facet->xout.has_normal = xout.has_normal;
4737 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4738 facet->xout.nf_output_iface = xout.nf_output_iface;
4739 facet->xout.mirrors = xout.mirrors;
4740 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4741 facet->used = MAX(facet->used, new_rule->up.created);
4742 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4744 xlate_out_uninit(&xout);
4749 facet_reset_counters(struct facet *facet)
4751 facet->packet_count = 0;
4752 facet->byte_count = 0;
4753 facet->prev_packet_count = 0;
4754 facet->prev_byte_count = 0;
4755 facet->accounted_bytes = 0;
4759 facet_push_stats(struct facet *facet, bool may_learn)
4761 struct dpif_flow_stats stats;
4763 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4764 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4765 ovs_assert(facet->used >= facet->prev_used);
4767 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4768 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4769 stats.used = facet->used;
4770 stats.tcp_flags = facet->tcp_flags;
4772 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4773 struct ofproto_dpif *ofproto = facet->ofproto;
4774 struct ofport_dpif *in_port;
4775 struct rule_dpif *rule;
4776 struct xlate_in xin;
4778 facet->prev_packet_count = facet->packet_count;
4779 facet->prev_byte_count = facet->byte_count;
4780 facet->prev_used = facet->used;
4782 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4783 if (in_port && in_port->is_tunnel) {
4784 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4787 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4788 rule_credit_stats(rule, &stats);
4789 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4791 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4792 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4793 stats.n_packets, stats.n_bytes);
4795 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4797 xin.resubmit_stats = &stats;
4798 xin.may_learn = may_learn;
4799 xlate_actions_for_side_effects(&xin);
4804 push_all_stats__(bool run_fast)
4806 static long long int rl = LLONG_MIN;
4807 struct ofproto_dpif *ofproto;
4809 if (time_msec() < rl) {
4813 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4814 struct cls_cursor cursor;
4815 struct facet *facet;
4817 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4818 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4819 facet_push_stats(facet, false);
4826 rl = time_msec() + 100;
4830 push_all_stats(void)
4832 push_all_stats__(true);
4836 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4838 ovs_mutex_lock(&rule->stats_mutex);
4839 rule->packet_count += stats->n_packets;
4840 rule->byte_count += stats->n_bytes;
4841 ofproto_rule_update_used(&rule->up, stats->used);
4842 ovs_mutex_unlock(&rule->stats_mutex);
4847 static struct subfacet *
4848 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4849 size_t key_len, uint32_t key_hash)
4851 struct subfacet *subfacet;
4853 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4854 &backer->subfacets) {
4855 if (subfacet->key_len == key_len
4856 && !memcmp(key, subfacet->key, key_len)) {
4864 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4865 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4866 * existing subfacet if there is one, otherwise creates and returns a
4868 static struct subfacet *
4869 subfacet_create(struct facet *facet, struct flow_miss *miss,
4872 struct dpif_backer *backer = miss->ofproto->backer;
4873 enum odp_key_fitness key_fitness = miss->key_fitness;
4874 const struct nlattr *key = miss->key;
4875 size_t key_len = miss->key_len;
4877 struct subfacet *subfacet;
4879 key_hash = odp_flow_key_hash(key, key_len);
4881 if (list_is_empty(&facet->subfacets)) {
4882 subfacet = &facet->one_subfacet;
4884 subfacet = subfacet_find(backer, key, key_len, key_hash);
4886 if (subfacet->facet == facet) {
4890 /* This shouldn't happen. */
4891 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4892 subfacet_destroy(subfacet);
4895 subfacet = xmalloc(sizeof *subfacet);
4898 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4899 list_push_back(&facet->subfacets, &subfacet->list_node);
4900 subfacet->facet = facet;
4901 subfacet->key_fitness = key_fitness;
4902 subfacet->key = xmemdup(key, key_len);
4903 subfacet->key_len = key_len;
4904 subfacet->used = now;
4905 subfacet->created = now;
4906 subfacet->dp_packet_count = 0;
4907 subfacet->dp_byte_count = 0;
4908 subfacet->path = SF_NOT_INSTALLED;
4909 subfacet->backer = backer;
4911 backer->subfacet_add_count++;
4915 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4916 * its facet within 'ofproto', and frees it. */
4918 subfacet_destroy__(struct subfacet *subfacet)
4920 struct facet *facet = subfacet->facet;
4921 struct ofproto_dpif *ofproto = facet->ofproto;
4923 /* Update ofproto stats before uninstall the subfacet. */
4924 ofproto->backer->subfacet_del_count++;
4926 subfacet_uninstall(subfacet);
4927 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4928 list_remove(&subfacet->list_node);
4929 free(subfacet->key);
4930 if (subfacet != &facet->one_subfacet) {
4935 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4936 * last remaining subfacet in its facet destroys the facet too. */
4938 subfacet_destroy(struct subfacet *subfacet)
4940 struct facet *facet = subfacet->facet;
4942 if (list_is_singleton(&facet->subfacets)) {
4943 /* facet_remove() needs at least one subfacet (it will remove it). */
4944 facet_remove(facet);
4946 subfacet_destroy__(subfacet);
4951 subfacet_destroy_batch(struct dpif_backer *backer,
4952 struct subfacet **subfacets, int n)
4954 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4955 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4956 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4959 for (i = 0; i < n; i++) {
4960 ops[i].type = DPIF_OP_FLOW_DEL;
4961 ops[i].u.flow_del.key = subfacets[i]->key;
4962 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4963 ops[i].u.flow_del.stats = &stats[i];
4967 dpif_operate(backer->dpif, opsp, n);
4968 for (i = 0; i < n; i++) {
4969 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4970 subfacets[i]->path = SF_NOT_INSTALLED;
4971 subfacet_destroy(subfacets[i]);
4976 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4977 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4978 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4979 * since 'subfacet' was last updated.
4981 * Returns 0 if successful, otherwise a positive errno value. */
4983 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4984 struct dpif_flow_stats *stats)
4986 struct facet *facet = subfacet->facet;
4987 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4988 const struct nlattr *actions = odp_actions->data;
4989 size_t actions_len = odp_actions->size;
4990 struct odputil_keybuf maskbuf;
4993 uint64_t slow_path_stub[128 / 8];
4994 enum dpif_flow_put_flags flags;
4997 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5000 flags |= DPIF_FP_ZERO_STATS;
5003 if (path == SF_SLOW_PATH) {
5004 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5005 slow_path_stub, sizeof slow_path_stub,
5006 &actions, &actions_len);
5009 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5010 if (enable_megaflows) {
5011 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5012 &facet->flow, UINT32_MAX);
5015 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5016 subfacet->key_len, mask.data, mask.size,
5017 actions, actions_len, stats);
5020 subfacet_reset_dp_stats(subfacet, stats);
5024 COVERAGE_INC(subfacet_install_fail);
5026 subfacet->path = path;
5031 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5033 subfacet_uninstall(struct subfacet *subfacet)
5035 if (subfacet->path != SF_NOT_INSTALLED) {
5036 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5037 struct dpif_flow_stats stats;
5040 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5041 subfacet->key_len, &stats);
5042 subfacet_reset_dp_stats(subfacet, &stats);
5044 subfacet_update_stats(subfacet, &stats);
5046 subfacet->path = SF_NOT_INSTALLED;
5048 ovs_assert(subfacet->dp_packet_count == 0);
5049 ovs_assert(subfacet->dp_byte_count == 0);
5053 /* Resets 'subfacet''s datapath statistics counters. This should be called
5054 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5055 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5056 * was reset in the datapath. 'stats' will be modified to include only
5057 * statistics new since 'subfacet' was last updated. */
5059 subfacet_reset_dp_stats(struct subfacet *subfacet,
5060 struct dpif_flow_stats *stats)
5063 && subfacet->dp_packet_count <= stats->n_packets
5064 && subfacet->dp_byte_count <= stats->n_bytes) {
5065 stats->n_packets -= subfacet->dp_packet_count;
5066 stats->n_bytes -= subfacet->dp_byte_count;
5069 subfacet->dp_packet_count = 0;
5070 subfacet->dp_byte_count = 0;
5073 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5075 * Because of the meaning of a subfacet's counters, it only makes sense to do
5076 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5077 * represents a packet that was sent by hand or if it represents statistics
5078 * that have been cleared out of the datapath. */
5080 subfacet_update_stats(struct subfacet *subfacet,
5081 const struct dpif_flow_stats *stats)
5083 if (stats->n_packets || stats->used > subfacet->used) {
5084 struct facet *facet = subfacet->facet;
5086 subfacet->used = MAX(subfacet->used, stats->used);
5087 facet->used = MAX(facet->used, stats->used);
5088 facet->packet_count += stats->n_packets;
5089 facet->byte_count += stats->n_bytes;
5090 facet->tcp_flags |= stats->tcp_flags;
5096 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5097 * the fields that were relevant as part of the lookup. */
5098 static struct rule_dpif *
5099 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5100 struct flow_wildcards *wc)
5102 struct rule_dpif *rule;
5104 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5109 return rule_dpif_miss_rule(ofproto, flow);
5113 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5114 const struct flow *flow, struct flow_wildcards *wc,
5117 struct cls_rule *cls_rule;
5118 struct classifier *cls;
5121 if (table_id >= N_TABLES) {
5126 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5127 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5130 cls = &ofproto->up.tables[table_id].cls;
5131 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5132 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5133 /* We must pretend that transport ports are unavailable. */
5134 struct flow ofpc_normal_flow = *flow;
5135 ofpc_normal_flow.tp_src = htons(0);
5136 ofpc_normal_flow.tp_dst = htons(0);
5137 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5138 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5139 cls_rule = &ofproto->drop_frags_rule->up.cr;
5141 flow_wildcards_init_exact(wc);
5144 cls_rule = classifier_lookup(cls, flow, wc);
5146 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5150 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5152 struct ofport_dpif *port;
5154 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5156 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5157 flow->in_port.ofp_port);
5158 return ofproto->miss_rule;
5161 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5162 return ofproto->no_packet_in_rule;
5164 return ofproto->miss_rule;
5168 complete_operation(struct rule_dpif *rule)
5170 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5172 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5174 struct dpif_completion *c = xmalloc(sizeof *c);
5175 c->op = rule->up.pending;
5176 list_push_back(&ofproto->completions, &c->list_node);
5178 ofoperation_complete(rule->up.pending, 0);
5182 static struct rule *
5185 struct rule_dpif *rule = xmalloc(sizeof *rule);
5190 rule_dealloc(struct rule *rule_)
5192 struct rule_dpif *rule = rule_dpif_cast(rule_);
5197 rule_construct(struct rule *rule_)
5199 struct rule_dpif *rule = rule_dpif_cast(rule_);
5200 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5201 ovs_mutex_lock(&rule->stats_mutex);
5202 rule->packet_count = 0;
5203 rule->byte_count = 0;
5204 ovs_mutex_unlock(&rule->stats_mutex);
5205 complete_operation(rule);
5210 rule_destruct(struct rule *rule_)
5212 struct rule_dpif *rule = rule_dpif_cast(rule_);
5213 complete_operation(rule);
5214 ovs_mutex_destroy(&rule->stats_mutex);
5218 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5220 struct rule_dpif *rule = rule_dpif_cast(rule_);
5222 /* push_all_stats() can handle flow misses which, when using the learn
5223 * action, can cause rules to be added and deleted. This can corrupt our
5224 * caller's datastructures which assume that rule_get_stats() doesn't have
5225 * an impact on the flow table. To be safe, we disable miss handling. */
5226 push_all_stats__(false);
5228 /* Start from historical data for 'rule' itself that are no longer tracked
5229 * in facets. This counts, for example, facets that have expired. */
5230 ovs_mutex_lock(&rule->stats_mutex);
5231 *packets = rule->packet_count;
5232 *bytes = rule->byte_count;
5233 ovs_mutex_unlock(&rule->stats_mutex);
5237 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5238 struct ofpbuf *packet)
5240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5241 struct dpif_flow_stats stats;
5242 struct xlate_out xout;
5243 struct xlate_in xin;
5245 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5246 rule_credit_stats(rule, &stats);
5248 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5249 xin.resubmit_stats = &stats;
5250 xlate_actions(&xin, &xout);
5252 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5253 xout.odp_actions.size, packet);
5255 xlate_out_uninit(&xout);
5259 rule_execute(struct rule *rule, const struct flow *flow,
5260 struct ofpbuf *packet)
5262 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5263 ofpbuf_delete(packet);
5268 rule_modify_actions(struct rule *rule_)
5270 struct rule_dpif *rule = rule_dpif_cast(rule_);
5272 complete_operation(rule);
5275 /* Sends 'packet' out 'ofport'.
5276 * May modify 'packet'.
5277 * Returns 0 if successful, otherwise a positive errno value. */
5279 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5281 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5282 uint64_t odp_actions_stub[1024 / 8];
5283 struct ofpbuf key, odp_actions;
5284 struct dpif_flow_stats stats;
5285 struct odputil_keybuf keybuf;
5286 struct ofpact_output output;
5287 struct xlate_out xout;
5288 struct xlate_in xin;
5290 union flow_in_port in_port_;
5293 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5294 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5296 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5297 in_port_.ofp_port = OFPP_NONE;
5298 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5299 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5301 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5303 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5304 output.port = ofport->up.ofp_port;
5307 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5308 xin.ofpacts_len = sizeof output;
5309 xin.ofpacts = &output.ofpact;
5310 xin.resubmit_stats = &stats;
5311 xlate_actions(&xin, &xout);
5313 error = dpif_execute(ofproto->backer->dpif,
5315 xout.odp_actions.data, xout.odp_actions.size,
5317 xlate_out_uninit(&xout);
5320 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5321 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5322 ovs_strerror(error));
5325 ofproto->stats.tx_packets++;
5326 ofproto->stats.tx_bytes += packet->size;
5330 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5331 * The action will state 'slow' as the reason that the action is in the slow
5332 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5333 * dump-flows" output to see why a flow is in the slow path.)
5335 * The 'stub_size' bytes in 'stub' will be used to store the action.
5336 * 'stub_size' must be large enough for the action.
5338 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5341 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5342 enum slow_path_reason slow,
5343 uint64_t *stub, size_t stub_size,
5344 const struct nlattr **actionsp, size_t *actions_lenp)
5346 union user_action_cookie cookie;
5349 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5350 cookie.slow_path.unused = 0;
5351 cookie.slow_path.reason = slow;
5353 ofpbuf_use_stack(&buf, stub, stub_size);
5354 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5355 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5357 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5359 odp_port_t odp_port;
5362 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5363 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5364 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5366 *actionsp = buf.data;
5367 *actions_lenp = buf.size;
5371 set_frag_handling(struct ofproto *ofproto_,
5372 enum ofp_config_flags frag_handling)
5374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5375 if (frag_handling != OFPC_FRAG_REASM) {
5376 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5384 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5385 const struct flow *flow,
5386 const struct ofpact *ofpacts, size_t ofpacts_len)
5388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5389 struct odputil_keybuf keybuf;
5390 struct dpif_flow_stats stats;
5391 struct xlate_out xout;
5392 struct xlate_in xin;
5396 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5397 odp_flow_key_from_flow(&key, flow,
5398 ofp_port_to_odp_port(ofproto,
5399 flow->in_port.ofp_port));
5401 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5403 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5404 xin.resubmit_stats = &stats;
5405 xin.ofpacts_len = ofpacts_len;
5406 xin.ofpacts = ofpacts;
5408 xlate_actions(&xin, &xout);
5409 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5410 xout.odp_actions.data, xout.odp_actions.size, packet);
5411 xlate_out_uninit(&xout);
5419 set_netflow(struct ofproto *ofproto_,
5420 const struct netflow_options *netflow_options)
5422 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5424 if (netflow_options) {
5425 if (!ofproto->netflow) {
5426 ofproto->netflow = netflow_create();
5427 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5429 return netflow_set_options(ofproto->netflow, netflow_options);
5430 } else if (ofproto->netflow) {
5431 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5432 netflow_destroy(ofproto->netflow);
5433 ofproto->netflow = NULL;
5440 get_netflow_ids(const struct ofproto *ofproto_,
5441 uint8_t *engine_type, uint8_t *engine_id)
5443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5445 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5449 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5451 if (!facet_is_controller_flow(facet) &&
5452 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5453 struct subfacet *subfacet;
5454 struct ofexpired expired;
5456 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5457 if (subfacet->path == SF_FAST_PATH) {
5458 struct dpif_flow_stats stats;
5460 subfacet_install(subfacet, &facet->xout.odp_actions,
5462 subfacet_update_stats(subfacet, &stats);
5466 expired.flow = facet->flow;
5467 expired.packet_count = facet->packet_count;
5468 expired.byte_count = facet->byte_count;
5469 expired.used = facet->used;
5470 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5475 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5477 struct cls_cursor cursor;
5478 struct facet *facet;
5480 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5481 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5482 send_active_timeout(ofproto, facet);
5486 static struct ofproto_dpif *
5487 ofproto_dpif_lookup(const char *name)
5489 struct ofproto_dpif *ofproto;
5491 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5492 hash_string(name, 0), &all_ofproto_dpifs) {
5493 if (!strcmp(ofproto->up.name, name)) {
5501 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5502 const char *argv[], void *aux OVS_UNUSED)
5504 struct ofproto_dpif *ofproto;
5507 ofproto = ofproto_dpif_lookup(argv[1]);
5509 unixctl_command_reply_error(conn, "no such bridge");
5512 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5513 mac_learning_flush(ofproto->ml);
5514 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5516 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5517 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5518 mac_learning_flush(ofproto->ml);
5519 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5523 unixctl_command_reply(conn, "table successfully flushed");
5526 static struct ofport_dpif *
5527 ofbundle_get_a_port(const struct ofbundle *bundle)
5529 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5534 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5535 const char *argv[], void *aux OVS_UNUSED)
5537 struct ds ds = DS_EMPTY_INITIALIZER;
5538 const struct ofproto_dpif *ofproto;
5539 const struct mac_entry *e;
5541 ofproto = ofproto_dpif_lookup(argv[1]);
5543 unixctl_command_reply_error(conn, "no such bridge");
5547 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5548 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5549 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5550 struct ofbundle *bundle = e->port.p;
5551 char name[OFP_MAX_PORT_NAME_LEN];
5553 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5555 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5556 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5557 mac_entry_age(ofproto->ml, e));
5559 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5560 unixctl_command_reply(conn, ds_cstr(&ds));
5565 struct xlate_out xout;
5566 struct xlate_in xin;
5572 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5574 ds_put_char_multiple(result, '\t', level);
5576 ds_put_cstr(result, "No match\n");
5580 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5581 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5582 cls_rule_format(&rule->up.cr, result);
5583 ds_put_char(result, '\n');
5585 ds_put_char_multiple(result, '\t', level);
5586 ds_put_cstr(result, "OpenFlow ");
5587 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5588 ds_put_char(result, '\n');
5592 trace_format_flow(struct ds *result, int level, const char *title,
5593 struct trace_ctx *trace)
5595 ds_put_char_multiple(result, '\t', level);
5596 ds_put_format(result, "%s: ", title);
5597 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5598 ds_put_cstr(result, "unchanged");
5600 flow_format(result, &trace->xin.flow);
5601 trace->flow = trace->xin.flow;
5603 ds_put_char(result, '\n');
5607 trace_format_regs(struct ds *result, int level, const char *title,
5608 struct trace_ctx *trace)
5612 ds_put_char_multiple(result, '\t', level);
5613 ds_put_format(result, "%s:", title);
5614 for (i = 0; i < FLOW_N_REGS; i++) {
5615 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5617 ds_put_char(result, '\n');
5621 trace_format_odp(struct ds *result, int level, const char *title,
5622 struct trace_ctx *trace)
5624 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5626 ds_put_char_multiple(result, '\t', level);
5627 ds_put_format(result, "%s: ", title);
5628 format_odp_actions(result, odp_actions->data, odp_actions->size);
5629 ds_put_char(result, '\n');
5633 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5635 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5636 struct ds *result = trace->result;
5638 ds_put_char(result, '\n');
5639 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5640 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5641 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5642 trace_format_rule(result, recurse + 1, rule);
5646 trace_report(struct xlate_in *xin, const char *s, int recurse)
5648 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5649 struct ds *result = trace->result;
5651 ds_put_char_multiple(result, '\t', recurse);
5652 ds_put_cstr(result, s);
5653 ds_put_char(result, '\n');
5657 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5658 void *aux OVS_UNUSED)
5660 const struct dpif_backer *backer;
5661 struct ofproto_dpif *ofproto;
5662 struct ofpbuf odp_key, odp_mask;
5663 struct ofpbuf *packet;
5671 ofpbuf_init(&odp_key, 0);
5672 ofpbuf_init(&odp_mask, 0);
5674 /* Handle "-generate" or a hex string as the last argument. */
5675 if (!strcmp(argv[argc - 1], "-generate")) {
5676 packet = ofpbuf_new(0);
5679 const char *error = eth_from_hex(argv[argc - 1], &packet);
5682 } else if (argc == 4) {
5683 /* The 3-argument form must end in "-generate' or a hex string. */
5684 unixctl_command_reply_error(conn, error);
5689 /* Parse the flow and determine whether a datapath or
5690 * bridge is specified. If function odp_flow_key_from_string()
5691 * returns 0, the flow is a odp_flow. If function
5692 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5693 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5694 /* If the odp_flow is the second argument,
5695 * the datapath name is the first argument. */
5697 const char *dp_type;
5698 if (!strncmp(argv[1], "ovs-", 4)) {
5699 dp_type = argv[1] + 4;
5703 backer = shash_find_data(&all_dpif_backers, dp_type);
5705 unixctl_command_reply_error(conn, "Cannot find datapath "
5710 /* No datapath name specified, so there should be only one
5712 struct shash_node *node;
5713 if (shash_count(&all_dpif_backers) != 1) {
5714 unixctl_command_reply_error(conn, "Must specify datapath "
5715 "name, there is more than one type of datapath");
5718 node = shash_first(&all_dpif_backers);
5719 backer = node->data;
5722 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5723 NULL, &ofproto, NULL)) {
5724 unixctl_command_reply_error(conn, "Invalid datapath flow");
5727 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5728 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5730 unixctl_command_reply_error(conn, "Must specify bridge name");
5734 ofproto = ofproto_dpif_lookup(argv[1]);
5736 unixctl_command_reply_error(conn, "Unknown bridge name");
5740 unixctl_command_reply_error(conn, "Bad flow syntax");
5744 /* Generate a packet, if requested. */
5746 if (!packet->size) {
5747 flow_compose(packet, &flow);
5749 union flow_in_port in_port_;
5751 in_port_ = flow.in_port;
5752 ds_put_cstr(&result, "Packet: ");
5753 s = ofp_packet_to_string(packet->data, packet->size);
5754 ds_put_cstr(&result, s);
5757 /* Use the metadata from the flow and the packet argument
5758 * to reconstruct the flow. */
5759 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5764 ofproto_trace(ofproto, &flow, packet, &result);
5765 unixctl_command_reply(conn, ds_cstr(&result));
5768 ds_destroy(&result);
5769 ofpbuf_delete(packet);
5770 ofpbuf_uninit(&odp_key);
5771 ofpbuf_uninit(&odp_mask);
5775 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5776 const struct ofpbuf *packet, struct ds *ds)
5778 struct rule_dpif *rule;
5780 ds_put_cstr(ds, "Flow: ");
5781 flow_format(ds, flow);
5782 ds_put_char(ds, '\n');
5784 rule = rule_dpif_lookup(ofproto, flow, NULL);
5786 trace_format_rule(ds, 0, rule);
5787 if (rule == ofproto->miss_rule) {
5788 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5789 } else if (rule == ofproto->no_packet_in_rule) {
5790 ds_put_cstr(ds, "\nNo match, packets dropped because "
5791 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5792 } else if (rule == ofproto->drop_frags_rule) {
5793 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5794 "and the fragment handling mode is \"drop\".\n");
5798 uint64_t odp_actions_stub[1024 / 8];
5799 struct ofpbuf odp_actions;
5800 struct trace_ctx trace;
5804 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5807 ofpbuf_use_stub(&odp_actions,
5808 odp_actions_stub, sizeof odp_actions_stub);
5809 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5810 trace.xin.resubmit_hook = trace_resubmit;
5811 trace.xin.report_hook = trace_report;
5813 xlate_actions(&trace.xin, &trace.xout);
5815 ds_put_char(ds, '\n');
5816 trace_format_flow(ds, 0, "Final flow", &trace);
5818 match_init(&match, flow, &trace.xout.wc);
5819 ds_put_cstr(ds, "Relevant fields: ");
5820 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5821 ds_put_char(ds, '\n');
5823 ds_put_cstr(ds, "Datapath actions: ");
5824 format_odp_actions(ds, trace.xout.odp_actions.data,
5825 trace.xout.odp_actions.size);
5827 if (trace.xout.slow) {
5828 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5829 "slow path because it:");
5830 switch (trace.xout.slow) {
5832 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5835 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5838 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5841 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5843 case SLOW_CONTROLLER:
5844 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5845 "to the OpenFlow controller.");
5852 xlate_out_uninit(&trace.xout);
5857 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5858 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5861 unixctl_command_reply(conn, NULL);
5865 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5866 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5869 unixctl_command_reply(conn, NULL);
5872 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5873 * 'reply' describing the results. */
5875 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5877 struct cls_cursor cursor;
5878 struct facet *facet;
5882 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5883 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5884 if (!facet_check_consistency(facet)) {
5889 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5893 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5894 ofproto->up.name, errors);
5896 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5901 ofproto_dpif_self_check(struct unixctl_conn *conn,
5902 int argc, const char *argv[], void *aux OVS_UNUSED)
5904 struct ds reply = DS_EMPTY_INITIALIZER;
5905 struct ofproto_dpif *ofproto;
5908 ofproto = ofproto_dpif_lookup(argv[1]);
5910 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5911 "ofproto/list for help)");
5914 ofproto_dpif_self_check__(ofproto, &reply);
5916 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5917 ofproto_dpif_self_check__(ofproto, &reply);
5921 unixctl_command_reply(conn, ds_cstr(&reply));
5925 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5926 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5927 * to destroy 'ofproto_shash' and free the returned value. */
5928 static const struct shash_node **
5929 get_ofprotos(struct shash *ofproto_shash)
5931 const struct ofproto_dpif *ofproto;
5933 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5934 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5935 shash_add_nocopy(ofproto_shash, name, ofproto);
5938 return shash_sort(ofproto_shash);
5942 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5943 const char *argv[] OVS_UNUSED,
5944 void *aux OVS_UNUSED)
5946 struct ds ds = DS_EMPTY_INITIALIZER;
5947 struct shash ofproto_shash;
5948 const struct shash_node **sorted_ofprotos;
5951 shash_init(&ofproto_shash);
5952 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5953 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5954 const struct shash_node *node = sorted_ofprotos[i];
5955 ds_put_format(&ds, "%s\n", node->name);
5958 shash_destroy(&ofproto_shash);
5959 free(sorted_ofprotos);
5961 unixctl_command_reply(conn, ds_cstr(&ds));
5966 show_dp_rates(struct ds *ds, const char *heading,
5967 const struct avg_subfacet_rates *rates)
5969 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5970 heading, rates->add_rate, rates->del_rate);
5974 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5976 const struct shash_node **ofprotos;
5977 struct ofproto_dpif *ofproto;
5978 struct shash ofproto_shash;
5979 uint64_t n_hit, n_missed;
5980 long long int minutes;
5983 n_hit = n_missed = 0;
5984 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5985 if (ofproto->backer == backer) {
5986 n_missed += ofproto->n_missed;
5987 n_hit += ofproto->n_hit;
5991 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5992 dpif_name(backer->dpif), n_hit, n_missed);
5993 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5994 " life span: %lldms\n", hmap_count(&backer->subfacets),
5995 backer->avg_n_subfacet, backer->max_n_subfacet,
5996 backer->avg_subfacet_life);
5998 minutes = (time_msec() - backer->created) / (1000 * 60);
5999 if (minutes >= 60) {
6000 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6002 if (minutes >= 60 * 24) {
6003 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6005 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6007 shash_init(&ofproto_shash);
6008 ofprotos = get_ofprotos(&ofproto_shash);
6009 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6010 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6011 const struct shash_node **ports;
6014 if (ofproto->backer != backer) {
6018 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6019 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6021 ports = shash_sort(&ofproto->up.port_by_name);
6022 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6023 const struct shash_node *node = ports[j];
6024 struct ofport *ofport = node->data;
6026 odp_port_t odp_port;
6028 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6031 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6032 if (odp_port != ODPP_NONE) {
6033 ds_put_format(ds, "%"PRIu32":", odp_port);
6035 ds_put_cstr(ds, "none:");
6038 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6041 if (!netdev_get_config(ofport->netdev, &config)) {
6042 const struct smap_node **nodes;
6045 nodes = smap_sort(&config);
6046 for (i = 0; i < smap_count(&config); i++) {
6047 const struct smap_node *node = nodes[i];
6048 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6049 node->key, node->value);
6053 smap_destroy(&config);
6055 ds_put_char(ds, ')');
6056 ds_put_char(ds, '\n');
6060 shash_destroy(&ofproto_shash);
6065 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6066 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6068 struct ds ds = DS_EMPTY_INITIALIZER;
6069 const struct shash_node **backers;
6072 backers = shash_sort(&all_dpif_backers);
6073 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6074 dpif_show_backer(backers[i]->data, &ds);
6078 unixctl_command_reply(conn, ds_cstr(&ds));
6082 /* Dump the megaflow (facet) cache. This is useful to check the
6083 * correctness of flow wildcarding, since the same mechanism is used for
6084 * both xlate caching and kernel wildcarding.
6086 * It's important to note that in the output the flow description uses
6087 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6089 * This command is only needed for advanced debugging, so it's not
6090 * documented in the man page. */
6092 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6093 int argc OVS_UNUSED, const char *argv[],
6094 void *aux OVS_UNUSED)
6096 struct ds ds = DS_EMPTY_INITIALIZER;
6097 const struct ofproto_dpif *ofproto;
6098 long long int now = time_msec();
6099 struct cls_cursor cursor;
6100 struct facet *facet;
6102 ofproto = ofproto_dpif_lookup(argv[1]);
6104 unixctl_command_reply_error(conn, "no such bridge");
6108 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6109 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6110 cls_rule_format(&facet->cr, &ds);
6111 ds_put_cstr(&ds, ", ");
6112 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6113 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6114 ds_put_cstr(&ds, "Datapath actions: ");
6115 if (facet->xout.slow) {
6116 uint64_t slow_path_stub[128 / 8];
6117 const struct nlattr *actions;
6120 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6121 slow_path_stub, sizeof slow_path_stub,
6122 &actions, &actions_len);
6123 format_odp_actions(&ds, actions, actions_len);
6125 format_odp_actions(&ds, facet->xout.odp_actions.data,
6126 facet->xout.odp_actions.size);
6128 ds_put_cstr(&ds, "\n");
6131 ds_chomp(&ds, '\n');
6132 unixctl_command_reply(conn, ds_cstr(&ds));
6136 /* Disable using the megaflows.
6138 * This command is only needed for advanced debugging, so it's not
6139 * documented in the man page. */
6141 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6142 int argc OVS_UNUSED,
6143 const char *argv[] OVS_UNUSED,
6144 void *aux OVS_UNUSED)
6146 struct ofproto_dpif *ofproto;
6148 enable_megaflows = false;
6150 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6151 flush(&ofproto->up);
6154 unixctl_command_reply(conn, "megaflows disabled");
6157 /* Re-enable using megaflows.
6159 * This command is only needed for advanced debugging, so it's not
6160 * documented in the man page. */
6162 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6163 int argc OVS_UNUSED,
6164 const char *argv[] OVS_UNUSED,
6165 void *aux OVS_UNUSED)
6167 struct ofproto_dpif *ofproto;
6169 enable_megaflows = true;
6171 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6172 flush(&ofproto->up);
6175 unixctl_command_reply(conn, "megaflows enabled");
6179 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6180 int argc OVS_UNUSED, const char *argv[],
6181 void *aux OVS_UNUSED)
6183 struct ds ds = DS_EMPTY_INITIALIZER;
6184 const struct ofproto_dpif *ofproto;
6185 struct subfacet *subfacet;
6187 ofproto = ofproto_dpif_lookup(argv[1]);
6189 unixctl_command_reply_error(conn, "no such bridge");
6193 update_stats(ofproto->backer);
6195 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6196 struct facet *facet = subfacet->facet;
6197 struct odputil_keybuf maskbuf;
6200 if (facet->ofproto != ofproto) {
6204 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6205 if (enable_megaflows) {
6206 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6207 &facet->flow, UINT32_MAX);
6210 odp_flow_format(subfacet->key, subfacet->key_len,
6211 mask.data, mask.size, &ds, false);
6213 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6214 subfacet->dp_packet_count, subfacet->dp_byte_count);
6215 if (subfacet->used) {
6216 ds_put_format(&ds, "%.3fs",
6217 (time_msec() - subfacet->used) / 1000.0);
6219 ds_put_format(&ds, "never");
6221 if (subfacet->facet->tcp_flags) {
6222 ds_put_cstr(&ds, ", flags:");
6223 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6226 ds_put_cstr(&ds, ", actions:");
6227 if (facet->xout.slow) {
6228 uint64_t slow_path_stub[128 / 8];
6229 const struct nlattr *actions;
6232 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6233 slow_path_stub, sizeof slow_path_stub,
6234 &actions, &actions_len);
6235 format_odp_actions(&ds, actions, actions_len);
6237 format_odp_actions(&ds, facet->xout.odp_actions.data,
6238 facet->xout.odp_actions.size);
6240 ds_put_char(&ds, '\n');
6243 unixctl_command_reply(conn, ds_cstr(&ds));
6248 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6249 int argc OVS_UNUSED, const char *argv[],
6250 void *aux OVS_UNUSED)
6252 struct ds ds = DS_EMPTY_INITIALIZER;
6253 struct ofproto_dpif *ofproto;
6255 ofproto = ofproto_dpif_lookup(argv[1]);
6257 unixctl_command_reply_error(conn, "no such bridge");
6261 flush(&ofproto->up);
6263 unixctl_command_reply(conn, ds_cstr(&ds));
6268 ofproto_dpif_unixctl_init(void)
6270 static bool registered;
6276 unixctl_command_register(
6278 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6279 1, 3, ofproto_unixctl_trace, NULL);
6280 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6281 ofproto_unixctl_fdb_flush, NULL);
6282 unixctl_command_register("fdb/show", "bridge", 1, 1,
6283 ofproto_unixctl_fdb_show, NULL);
6284 unixctl_command_register("ofproto/clog", "", 0, 0,
6285 ofproto_dpif_clog, NULL);
6286 unixctl_command_register("ofproto/unclog", "", 0, 0,
6287 ofproto_dpif_unclog, NULL);
6288 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6289 ofproto_dpif_self_check, NULL);
6290 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6291 ofproto_unixctl_dpif_dump_dps, NULL);
6292 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6294 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6295 ofproto_unixctl_dpif_dump_flows, NULL);
6296 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6297 ofproto_unixctl_dpif_del_flows, NULL);
6298 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6299 ofproto_unixctl_dpif_dump_megaflows, NULL);
6300 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6301 ofproto_unixctl_dpif_disable_megaflows, NULL);
6302 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6303 ofproto_unixctl_dpif_enable_megaflows, NULL);
6306 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6308 * This is deprecated. It is only for compatibility with broken device drivers
6309 * in old versions of Linux that do not properly support VLANs when VLAN
6310 * devices are not used. When broken device drivers are no longer in
6311 * widespread use, we will delete these interfaces. */
6314 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6316 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6317 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6319 if (realdev_ofp_port == ofport->realdev_ofp_port
6320 && vid == ofport->vlandev_vid) {
6324 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6326 if (ofport->realdev_ofp_port) {
6329 if (realdev_ofp_port && ofport->bundle) {
6330 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6331 * themselves be part of a bundle. */
6332 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6335 ofport->realdev_ofp_port = realdev_ofp_port;
6336 ofport->vlandev_vid = vid;
6338 if (realdev_ofp_port) {
6339 vsp_add(ofport, realdev_ofp_port, vid);
6346 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6348 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6352 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6353 OVS_EXCLUDED(ofproto->vsp_mutex)
6357 ovs_mutex_lock(&ofproto->vsp_mutex);
6358 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6359 ovs_mutex_unlock(&ofproto->vsp_mutex);
6364 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6365 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6366 OVS_REQUIRES(ofproto->vsp_mutex)
6368 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6369 int vid = vlan_tci_to_vid(vlan_tci);
6370 const struct vlan_splinter *vsp;
6372 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6373 hash_realdev_vid(realdev_ofp_port, vid),
6374 &ofproto->realdev_vid_map) {
6375 if (vsp->realdev_ofp_port == realdev_ofp_port
6376 && vsp->vid == vid) {
6377 return vsp->vlandev_ofp_port;
6381 return realdev_ofp_port;
6384 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6385 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6386 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6387 * 'vlan_tci' 9, it would return the port number of eth0.9.
6389 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6390 * function just returns its 'realdev_ofp_port' argument. */
6392 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6393 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6394 OVS_EXCLUDED(ofproto->vsp_mutex)
6398 ovs_mutex_lock(&ofproto->vsp_mutex);
6399 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6400 ovs_mutex_unlock(&ofproto->vsp_mutex);
6404 static struct vlan_splinter *
6405 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6407 struct vlan_splinter *vsp;
6409 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6410 hash_ofp_port(vlandev_ofp_port),
6411 &ofproto->vlandev_map) {
6412 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6420 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6421 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6422 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6423 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6424 * eth0 and store 9 in '*vid'.
6426 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6427 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6430 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6431 ofp_port_t vlandev_ofp_port, int *vid)
6432 OVS_REQ_WRLOCK(ofproto->vsp_mutex)
6434 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6435 const struct vlan_splinter *vsp;
6437 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6442 return vsp->realdev_ofp_port;
6448 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6449 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6450 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6451 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6452 * always the case unless VLAN splinters are enabled), returns false without
6453 * making any changes. */
6455 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6456 OVS_EXCLUDED(ofproto->vsp_mutex)
6461 ovs_mutex_lock(&ofproto->vsp_mutex);
6462 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6463 ovs_mutex_unlock(&ofproto->vsp_mutex);
6468 /* Cause the flow to be processed as if it came in on the real device with
6469 * the VLAN device's VLAN ID. */
6470 flow->in_port.ofp_port = realdev;
6471 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6476 vsp_remove(struct ofport_dpif *port)
6478 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6479 struct vlan_splinter *vsp;
6481 ovs_mutex_lock(&ofproto->vsp_mutex);
6482 vsp = vlandev_find(ofproto, port->up.ofp_port);
6484 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6485 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6488 port->realdev_ofp_port = 0;
6490 VLOG_ERR("missing vlan device record");
6492 ovs_mutex_unlock(&ofproto->vsp_mutex);
6496 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6500 ovs_mutex_lock(&ofproto->vsp_mutex);
6501 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6502 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6503 == realdev_ofp_port)) {
6504 struct vlan_splinter *vsp;
6506 vsp = xmalloc(sizeof *vsp);
6507 vsp->realdev_ofp_port = realdev_ofp_port;
6508 vsp->vlandev_ofp_port = port->up.ofp_port;
6511 port->realdev_ofp_port = realdev_ofp_port;
6513 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6514 hash_ofp_port(port->up.ofp_port));
6515 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6516 hash_realdev_vid(realdev_ofp_port, vid));
6518 VLOG_ERR("duplicate vlan device record");
6520 ovs_mutex_unlock(&ofproto->vsp_mutex);
6524 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6526 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6527 return ofport ? ofport->odp_port : ODPP_NONE;
6530 struct ofport_dpif *
6531 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6533 struct ofport_dpif *port;
6535 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6536 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6537 &backer->odp_to_ofport_map) {
6538 if (port->odp_port == odp_port) {
6539 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6544 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6549 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6551 struct ofport_dpif *port;
6553 port = odp_port_to_ofport(ofproto->backer, odp_port);
6554 if (port && &ofproto->up == port->up.ofproto) {
6555 return port->up.ofp_port;
6561 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6562 * most heavily weighted element. 'base' designates the rate of decay: after
6563 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6566 exp_mavg(double *avg, int base, double new)
6568 *avg = (*avg * (base - 1) + new) / base;
6572 update_moving_averages(struct dpif_backer *backer)
6574 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6575 long long int minutes = (time_msec() - backer->created) / min_ms;
6578 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6580 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6583 backer->lifetime.add_rate = 0.0;
6584 backer->lifetime.del_rate = 0.0;
6587 /* Update hourly averages on the minute boundaries. */
6588 if (time_msec() - backer->last_minute >= min_ms) {
6589 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6590 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6592 /* Update daily averages on the hour boundaries. */
6593 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6594 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6595 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6598 backer->total_subfacet_add_count += backer->subfacet_add_count;
6599 backer->total_subfacet_del_count += backer->subfacet_del_count;
6600 backer->subfacet_add_count = 0;
6601 backer->subfacet_del_count = 0;
6602 backer->last_minute += min_ms;
6606 const struct ofproto_class ofproto_dpif_class = {
6641 port_is_lacp_current,
6642 NULL, /* rule_choose_table */
6649 rule_modify_actions,
6663 get_stp_port_status,
6670 is_mirror_output_bundle,
6671 forward_bpdu_changed,
6672 set_mac_table_config,
6674 NULL, /* meter_get_features */
6675 NULL, /* meter_set */
6676 NULL, /* meter_get */
6677 NULL, /* meter_del */