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->miss_rule,
803 ofproto->no_packet_in_rule, ofproto->ml,
804 ofproto->stp, ofproto->mbridge,
805 ofproto->sflow, ofproto->ipfix,
806 ofproto->up.frag_handling,
807 ofproto->up.forward_bpdu,
808 connmgr_has_in_band(ofproto->up.connmgr),
809 ofproto->netflow != NULL);
811 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
812 xlate_bundle_set(ofproto, bundle, bundle->name,
813 bundle->vlan_mode, bundle->vlan,
814 bundle->trunks, bundle->use_priority_tags,
815 bundle->bond, bundle->lacp,
819 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
820 int stp_port = ofport->stp_port
821 ? stp_port_no(ofport->stp_port)
823 xlate_ofport_set(ofproto, ofport->bundle, ofport,
824 ofport->up.ofp_port, ofport->odp_port,
825 ofport->up.netdev, ofport->cfm,
826 ofport->bfd, ofport->peer, stp_port,
827 ofport->qdscp, ofport->n_qdscp,
828 ofport->up.pp.config, ofport->is_tunnel,
832 cls_cursor_init(&cursor, &ofproto->facets, NULL);
833 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
834 facet_revalidate(facet);
840 if (!backer->recv_set_enable) {
841 /* Wake up before a max of 1000ms. */
842 timer_set_duration(&backer->next_expiration, 1000);
843 } else if (timer_expired(&backer->next_expiration)) {
844 int delay = expire(backer);
845 timer_set_duration(&backer->next_expiration, delay);
848 process_dpif_port_changes(backer);
850 if (backer->governor) {
853 governor_run(backer->governor);
855 /* If the governor has shrunk to its minimum size and the number of
856 * subfacets has dwindled, then drop the governor entirely.
858 * For hysteresis, the number of subfacets to drop the governor is
859 * smaller than the number needed to trigger its creation. */
860 n_subfacets = hmap_count(&backer->subfacets);
861 if (n_subfacets * 4 < flow_eviction_threshold
862 && governor_is_idle(backer->governor)) {
863 governor_destroy(backer->governor);
864 backer->governor = NULL;
871 /* Check for and handle port changes in 'backer''s dpif. */
873 process_dpif_port_changes(struct dpif_backer *backer)
879 error = dpif_port_poll(backer->dpif, &devname);
885 process_dpif_all_ports_changed(backer);
889 process_dpif_port_change(backer, devname);
894 process_dpif_port_error(backer, error);
901 process_dpif_all_ports_changed(struct dpif_backer *backer)
903 struct ofproto_dpif *ofproto;
904 struct dpif_port dpif_port;
905 struct dpif_port_dump dump;
906 struct sset devnames;
909 sset_init(&devnames);
910 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
911 if (ofproto->backer == backer) {
912 struct ofport *ofport;
914 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
915 sset_add(&devnames, netdev_get_name(ofport->netdev));
919 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
920 sset_add(&devnames, dpif_port.name);
923 SSET_FOR_EACH (devname, &devnames) {
924 process_dpif_port_change(backer, devname);
926 sset_destroy(&devnames);
930 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
932 struct ofproto_dpif *ofproto;
933 struct dpif_port port;
935 /* Don't report on the datapath's device. */
936 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
940 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
941 &all_ofproto_dpifs) {
942 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
947 ofproto = lookup_ofproto_dpif_by_port_name(devname);
948 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
949 /* The port was removed. If we know the datapath,
950 * report it through poll_set(). If we don't, it may be
951 * notifying us of a removal we initiated, so ignore it.
952 * If there's a pending ENOBUFS, let it stand, since
953 * everything will be reevaluated. */
954 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
955 sset_add(&ofproto->port_poll_set, devname);
956 ofproto->port_poll_errno = 0;
958 } else if (!ofproto) {
959 /* The port was added, but we don't know with which
960 * ofproto we should associate it. Delete it. */
961 dpif_port_del(backer->dpif, port.port_no);
963 struct ofport_dpif *ofport;
965 ofport = ofport_dpif_cast(shash_find_data(
966 &ofproto->up.port_by_name, devname));
968 && ofport->odp_port != port.port_no
969 && !odp_port_to_ofport(backer, port.port_no))
971 /* 'ofport''s datapath port number has changed from
972 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
973 * structures to match. */
974 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
975 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
976 ofport->odp_port = port.port_no;
977 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
978 hash_odp_port(port.port_no));
979 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
980 backer->need_revalidate = REV_RECONFIGURE;
983 dpif_port_destroy(&port);
986 /* Propagate 'error' to all ofprotos based on 'backer'. */
988 process_dpif_port_error(struct dpif_backer *backer, int error)
990 struct ofproto_dpif *ofproto;
992 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
993 if (ofproto->backer == backer) {
994 sset_clear(&ofproto->port_poll_set);
995 ofproto->port_poll_errno = error;
1001 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1005 /* If recv_set_enable is false, we should not handle upcalls. */
1006 if (!backer->recv_set_enable) {
1010 /* Handle one or more batches of upcalls, until there's nothing left to do
1011 * or until we do a fixed total amount of work.
1013 * We do work in batches because it can be much cheaper to set up a number
1014 * of flows and fire off their patches all at once. We do multiple batches
1015 * because in some cases handling a packet can cause another packet to be
1016 * queued almost immediately as part of the return flow. Both
1017 * optimizations can make major improvements on some benchmarks and
1018 * presumably for real traffic as well. */
1020 while (work < max_batch) {
1021 int retval = handle_upcalls(backer, max_batch - work);
1032 type_run_fast(const char *type)
1034 struct dpif_backer *backer;
1036 backer = shash_find_data(&all_dpif_backers, type);
1038 /* This is not necessarily a problem, since backers are only
1039 * created on demand. */
1043 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1049 static long long int port_rl = LLONG_MIN;
1050 static unsigned int backer_rl = 0;
1052 if (time_msec() >= port_rl) {
1053 struct ofproto_dpif *ofproto;
1055 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1056 run_fast(&ofproto->up);
1058 port_rl = time_msec() + 200;
1061 /* XXX: We have to be careful not to do too much work in this function. If
1062 * we call dpif_backer_run_fast() too often, or with too large a batch,
1063 * performance improves signifcantly, but at a cost. It's possible for the
1064 * number of flows in the datapath to increase without bound, and for poll
1065 * loops to take 10s of seconds. The correct solution to this problem,
1066 * long term, is to separate flow miss handling into it's own thread so it
1067 * isn't affected by revalidations, and expirations. Until then, this is
1068 * the best we can do. */
1069 if (++backer_rl >= 10) {
1070 struct shash_node *node;
1073 SHASH_FOR_EACH (node, &all_dpif_backers) {
1074 dpif_backer_run_fast(node->data, 1);
1080 type_wait(const char *type)
1082 struct dpif_backer *backer;
1084 backer = shash_find_data(&all_dpif_backers, type);
1086 /* This is not necessarily a problem, since backers are only
1087 * created on demand. */
1091 if (backer->governor) {
1092 governor_wait(backer->governor);
1095 timer_wait(&backer->next_expiration);
1098 /* Basic life-cycle. */
1100 static int add_internal_flows(struct ofproto_dpif *);
1102 static struct ofproto *
1105 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1106 return &ofproto->up;
1110 dealloc(struct ofproto *ofproto_)
1112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1117 close_dpif_backer(struct dpif_backer *backer)
1119 struct shash_node *node;
1121 ovs_assert(backer->refcount > 0);
1123 if (--backer->refcount) {
1127 drop_key_clear(backer);
1128 hmap_destroy(&backer->drop_keys);
1130 simap_destroy(&backer->tnl_backers);
1131 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1132 hmap_destroy(&backer->odp_to_ofport_map);
1133 node = shash_find(&all_dpif_backers, backer->type);
1135 shash_delete(&all_dpif_backers, node);
1136 dpif_close(backer->dpif);
1138 ovs_assert(hmap_is_empty(&backer->subfacets));
1139 hmap_destroy(&backer->subfacets);
1140 governor_destroy(backer->governor);
1145 /* Datapath port slated for removal from datapath. */
1146 struct odp_garbage {
1147 struct list list_node;
1148 odp_port_t odp_port;
1152 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1154 struct dpif_backer *backer;
1155 struct dpif_port_dump port_dump;
1156 struct dpif_port port;
1157 struct shash_node *node;
1158 struct list garbage_list;
1159 struct odp_garbage *garbage, *next;
1165 backer = shash_find_data(&all_dpif_backers, type);
1172 backer_name = xasprintf("ovs-%s", type);
1174 /* Remove any existing datapaths, since we assume we're the only
1175 * userspace controlling the datapath. */
1177 dp_enumerate_names(type, &names);
1178 SSET_FOR_EACH(name, &names) {
1179 struct dpif *old_dpif;
1181 /* Don't remove our backer if it exists. */
1182 if (!strcmp(name, backer_name)) {
1186 if (dpif_open(name, type, &old_dpif)) {
1187 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1189 dpif_delete(old_dpif);
1190 dpif_close(old_dpif);
1193 sset_destroy(&names);
1195 backer = xmalloc(sizeof *backer);
1197 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1200 VLOG_ERR("failed to open datapath of type %s: %s", type,
1201 ovs_strerror(error));
1206 backer->type = xstrdup(type);
1207 backer->governor = NULL;
1208 backer->refcount = 1;
1209 hmap_init(&backer->odp_to_ofport_map);
1210 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1211 hmap_init(&backer->drop_keys);
1212 hmap_init(&backer->subfacets);
1213 timer_set_duration(&backer->next_expiration, 1000);
1214 backer->need_revalidate = 0;
1215 simap_init(&backer->tnl_backers);
1216 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1219 if (backer->recv_set_enable) {
1220 dpif_flow_flush(backer->dpif);
1223 /* Loop through the ports already on the datapath and remove any
1224 * that we don't need anymore. */
1225 list_init(&garbage_list);
1226 dpif_port_dump_start(&port_dump, backer->dpif);
1227 while (dpif_port_dump_next(&port_dump, &port)) {
1228 node = shash_find(&init_ofp_ports, port.name);
1229 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1230 garbage = xmalloc(sizeof *garbage);
1231 garbage->odp_port = port.port_no;
1232 list_push_front(&garbage_list, &garbage->list_node);
1235 dpif_port_dump_done(&port_dump);
1237 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1238 dpif_port_del(backer->dpif, garbage->odp_port);
1239 list_remove(&garbage->list_node);
1243 shash_add(&all_dpif_backers, type, backer);
1245 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1247 VLOG_ERR("failed to listen on datapath of type %s: %s",
1248 type, ovs_strerror(error));
1249 close_dpif_backer(backer);
1253 backer->max_n_subfacet = 0;
1254 backer->created = time_msec();
1255 backer->last_minute = backer->created;
1256 memset(&backer->hourly, 0, sizeof backer->hourly);
1257 memset(&backer->daily, 0, sizeof backer->daily);
1258 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1259 backer->subfacet_add_count = 0;
1260 backer->subfacet_del_count = 0;
1261 backer->total_subfacet_add_count = 0;
1262 backer->total_subfacet_del_count = 0;
1263 backer->avg_n_subfacet = 0;
1264 backer->avg_subfacet_life = 0;
1270 construct(struct ofproto *ofproto_)
1272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1273 struct shash_node *node, *next;
1274 odp_port_t max_ports;
1277 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1282 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1283 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1284 ofp_to_u16(OFPP_MAX))));
1286 ofproto->netflow = NULL;
1287 ofproto->sflow = NULL;
1288 ofproto->ipfix = NULL;
1289 ofproto->stp = NULL;
1290 hmap_init(&ofproto->bundles);
1291 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1292 ofproto->mbridge = mbridge_create();
1293 ofproto->has_bonded_bundles = false;
1294 ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1296 classifier_init(&ofproto->facets);
1297 ofproto->consistency_rl = LLONG_MIN;
1299 list_init(&ofproto->completions);
1301 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1302 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1303 list_init(&ofproto->flow_mods);
1304 ofproto->n_flow_mods = 0;
1305 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1307 ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL);
1308 ovs_mutex_lock(&ofproto->pin_mutex);
1309 list_init(&ofproto->pins);
1310 ofproto->n_pins = 0;
1311 ovs_mutex_unlock(&ofproto->pin_mutex);
1313 ofproto_dpif_unixctl_init();
1315 hmap_init(&ofproto->vlandev_map);
1316 hmap_init(&ofproto->realdev_vid_map);
1318 sset_init(&ofproto->ports);
1319 sset_init(&ofproto->ghost_ports);
1320 sset_init(&ofproto->port_poll_set);
1321 ofproto->port_poll_errno = 0;
1323 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1324 struct iface_hint *iface_hint = node->data;
1326 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1327 /* Check if the datapath already has this port. */
1328 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1329 sset_add(&ofproto->ports, node->name);
1332 free(iface_hint->br_name);
1333 free(iface_hint->br_type);
1335 shash_delete(&init_ofp_ports, node);
1339 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1340 hash_string(ofproto->up.name, 0));
1341 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1343 ofproto_init_tables(ofproto_, N_TABLES);
1344 error = add_internal_flows(ofproto);
1345 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1348 ofproto->n_missed = 0;
1354 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1355 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1357 struct ofputil_flow_mod fm;
1360 match_init_catchall(&fm.match);
1362 match_set_reg(&fm.match, 0, id);
1363 fm.new_cookie = htonll(0);
1364 fm.cookie = htonll(0);
1365 fm.cookie_mask = htonll(0);
1366 fm.modify_cookie = false;
1367 fm.table_id = TBL_INTERNAL;
1368 fm.command = OFPFC_ADD;
1369 fm.idle_timeout = 0;
1370 fm.hard_timeout = 0;
1374 fm.ofpacts = ofpacts->data;
1375 fm.ofpacts_len = ofpacts->size;
1377 error = ofproto_flow_mod(&ofproto->up, &fm);
1379 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1380 id, ofperr_to_string(error));
1384 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1386 ovs_assert(*rulep != NULL);
1392 add_internal_flows(struct ofproto_dpif *ofproto)
1394 struct ofpact_controller *controller;
1395 uint64_t ofpacts_stub[128 / 8];
1396 struct ofpbuf ofpacts;
1400 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1403 controller = ofpact_put_CONTROLLER(&ofpacts);
1404 controller->max_len = UINT16_MAX;
1405 controller->controller_id = 0;
1406 controller->reason = OFPR_NO_MATCH;
1407 ofpact_pad(&ofpacts);
1409 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1414 ofpbuf_clear(&ofpacts);
1415 error = add_internal_flow(ofproto, id++, &ofpacts,
1416 &ofproto->no_packet_in_rule);
1421 error = add_internal_flow(ofproto, id++, &ofpacts,
1422 &ofproto->drop_frags_rule);
1427 complete_operations(struct ofproto_dpif *ofproto)
1429 struct dpif_completion *c, *next;
1431 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1432 ofoperation_complete(c->op, 0);
1433 list_remove(&c->list_node);
1439 destruct(struct ofproto *ofproto_)
1441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1442 struct rule_dpif *rule, *next_rule;
1443 struct ofputil_flow_mod *pin, *next_pin;
1444 struct ofputil_flow_mod *fm, *next_fm;
1445 struct oftable *table;
1447 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1448 xlate_remove_ofproto(ofproto);
1450 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1451 complete_operations(ofproto);
1453 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1454 struct cls_cursor cursor;
1456 cls_cursor_init(&cursor, &table->cls, NULL);
1457 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1458 ofproto_rule_destroy(&rule->up);
1462 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1463 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1464 list_remove(&fm->list_node);
1465 ofproto->n_flow_mods--;
1469 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1470 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1472 ovs_mutex_lock(&ofproto->pin_mutex);
1473 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1474 list_remove(&pin->list_node);
1479 ovs_mutex_unlock(&ofproto->pin_mutex);
1480 ovs_mutex_destroy(&ofproto->pin_mutex);
1482 mbridge_unref(ofproto->mbridge);
1484 netflow_destroy(ofproto->netflow);
1485 dpif_sflow_unref(ofproto->sflow);
1486 hmap_destroy(&ofproto->bundles);
1487 mac_learning_unref(ofproto->ml);
1489 classifier_destroy(&ofproto->facets);
1491 hmap_destroy(&ofproto->vlandev_map);
1492 hmap_destroy(&ofproto->realdev_vid_map);
1494 sset_destroy(&ofproto->ports);
1495 sset_destroy(&ofproto->ghost_ports);
1496 sset_destroy(&ofproto->port_poll_set);
1498 ovs_mutex_destroy(&ofproto->vsp_mutex);
1500 close_dpif_backer(ofproto->backer);
1504 run_fast(struct ofproto *ofproto_)
1506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1507 struct ofputil_packet_in *pin, *next_pin;
1508 struct ofputil_flow_mod *fm, *next_fm;
1509 struct list flow_mods, pins;
1510 struct ofport_dpif *ofport;
1512 /* Do not perform any periodic activity required by 'ofproto' while
1513 * waiting for flow restore to complete. */
1514 if (ofproto_get_flow_restore_wait()) {
1518 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1519 if (ofproto->n_flow_mods) {
1520 flow_mods = ofproto->flow_mods;
1521 list_moved(&flow_mods);
1522 list_init(&ofproto->flow_mods);
1523 ofproto->n_flow_mods = 0;
1525 list_init(&flow_mods);
1527 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1529 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1530 int error = ofproto_flow_mod(&ofproto->up, fm);
1531 if (error && !VLOG_DROP_WARN(&rl)) {
1532 VLOG_WARN("learning action failed to modify flow table (%s)",
1533 ofperr_get_name(error));
1536 list_remove(&fm->list_node);
1541 ovs_mutex_lock(&ofproto->pin_mutex);
1542 if (ofproto->n_pins) {
1543 pins = ofproto->pins;
1545 list_init(&ofproto->pins);
1546 ofproto->n_pins = 0;
1550 ovs_mutex_unlock(&ofproto->pin_mutex);
1552 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1553 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1554 list_remove(&pin->list_node);
1555 free(CONST_CAST(void *, pin->packet));
1559 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1560 port_run_fast(ofport);
1567 run(struct ofproto *ofproto_)
1569 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1570 struct ofport_dpif *ofport;
1571 struct ofbundle *bundle;
1575 complete_operations(ofproto);
1578 if (mbridge_need_revalidate(ofproto->mbridge)) {
1579 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1580 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1581 mac_learning_flush(ofproto->ml);
1582 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1585 /* Do not perform any periodic activity below required by 'ofproto' while
1586 * waiting for flow restore to complete. */
1587 if (ofproto_get_flow_restore_wait()) {
1591 error = run_fast(ofproto_);
1596 if (ofproto->netflow) {
1597 if (netflow_run(ofproto->netflow)) {
1598 send_netflow_active_timeouts(ofproto);
1601 if (ofproto->sflow) {
1602 dpif_sflow_run(ofproto->sflow);
1605 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1608 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1613 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1614 if (mac_learning_run(ofproto->ml)) {
1615 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1617 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1619 /* Check the consistency of a random facet, to aid debugging. */
1620 if (time_msec() >= ofproto->consistency_rl
1621 && !classifier_is_empty(&ofproto->facets)
1622 && !ofproto->backer->need_revalidate) {
1623 struct cls_table *table;
1624 struct cls_rule *cr;
1625 struct facet *facet;
1627 ofproto->consistency_rl = time_msec() + 250;
1629 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1630 struct cls_table, hmap_node);
1631 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1633 facet = CONTAINER_OF(cr, struct facet, cr);
1635 if (!facet_check_consistency(facet)) {
1636 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1644 wait(struct ofproto *ofproto_)
1646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1647 struct ofport_dpif *ofport;
1648 struct ofbundle *bundle;
1650 if (!clogged && !list_is_empty(&ofproto->completions)) {
1651 poll_immediate_wake();
1654 if (ofproto_get_flow_restore_wait()) {
1658 dpif_wait(ofproto->backer->dpif);
1659 dpif_recv_wait(ofproto->backer->dpif);
1660 if (ofproto->sflow) {
1661 dpif_sflow_wait(ofproto->sflow);
1663 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1666 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1667 bundle_wait(bundle);
1669 if (ofproto->netflow) {
1670 netflow_wait(ofproto->netflow);
1672 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1673 mac_learning_wait(ofproto->ml);
1674 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1676 if (ofproto->backer->need_revalidate) {
1677 /* Shouldn't happen, but if it does just go around again. */
1678 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1679 poll_immediate_wake();
1684 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1686 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1687 struct cls_cursor cursor;
1688 size_t n_subfacets = 0;
1689 struct facet *facet;
1691 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1693 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1694 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1695 n_subfacets += list_size(&facet->subfacets);
1697 simap_increase(usage, "subfacets", n_subfacets);
1701 flush(struct ofproto *ofproto_)
1703 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1704 struct subfacet *subfacet, *next_subfacet;
1705 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1709 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1710 &ofproto->backer->subfacets) {
1711 if (subfacet->facet->ofproto != ofproto) {
1715 if (subfacet->path != SF_NOT_INSTALLED) {
1716 batch[n_batch++] = subfacet;
1717 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1718 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1722 subfacet_destroy(subfacet);
1727 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1732 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1733 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1735 *arp_match_ip = true;
1736 *actions = (OFPUTIL_A_OUTPUT |
1737 OFPUTIL_A_SET_VLAN_VID |
1738 OFPUTIL_A_SET_VLAN_PCP |
1739 OFPUTIL_A_STRIP_VLAN |
1740 OFPUTIL_A_SET_DL_SRC |
1741 OFPUTIL_A_SET_DL_DST |
1742 OFPUTIL_A_SET_NW_SRC |
1743 OFPUTIL_A_SET_NW_DST |
1744 OFPUTIL_A_SET_NW_TOS |
1745 OFPUTIL_A_SET_TP_SRC |
1746 OFPUTIL_A_SET_TP_DST |
1751 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1754 struct dpif_dp_stats s;
1755 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1758 strcpy(ots->name, "classifier");
1760 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1761 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1762 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1763 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1765 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1766 ots->lookup_count = htonll(n_lookup);
1767 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1770 static struct ofport *
1773 struct ofport_dpif *port = xmalloc(sizeof *port);
1778 port_dealloc(struct ofport *port_)
1780 struct ofport_dpif *port = ofport_dpif_cast(port_);
1785 port_construct(struct ofport *port_)
1787 struct ofport_dpif *port = ofport_dpif_cast(port_);
1788 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1789 const struct netdev *netdev = port->up.netdev;
1790 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1791 struct dpif_port dpif_port;
1794 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1795 port->bundle = NULL;
1798 port->may_enable = true;
1799 port->stp_port = NULL;
1800 port->stp_state = STP_DISABLED;
1801 port->is_tunnel = false;
1805 port->realdev_ofp_port = 0;
1806 port->vlandev_vid = 0;
1807 port->carrier_seq = netdev_get_carrier_resets(netdev);
1809 if (netdev_vport_is_patch(netdev)) {
1810 /* By bailing out here, we don't submit the port to the sFlow module
1811 * to be considered for counter polling export. This is correct
1812 * because the patch port represents an interface that sFlow considers
1813 * to be "internal" to the switch as a whole, and therefore not an
1814 * candidate for counter polling. */
1815 port->odp_port = ODPP_NONE;
1816 ofport_update_peer(port);
1820 error = dpif_port_query_by_name(ofproto->backer->dpif,
1821 netdev_vport_get_dpif_port(netdev, namebuf,
1828 port->odp_port = dpif_port.port_no;
1830 if (netdev_get_tunnel_config(netdev)) {
1831 tnl_port_add(port, port->up.netdev, port->odp_port);
1832 port->is_tunnel = true;
1834 /* Sanity-check that a mapping doesn't already exist. This
1835 * shouldn't happen for non-tunnel ports. */
1836 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1837 VLOG_ERR("port %s already has an OpenFlow port number",
1839 dpif_port_destroy(&dpif_port);
1843 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1844 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1845 hash_odp_port(port->odp_port));
1846 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1848 dpif_port_destroy(&dpif_port);
1850 if (ofproto->sflow) {
1851 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1858 port_destruct(struct ofport *port_)
1860 struct ofport_dpif *port = ofport_dpif_cast(port_);
1861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1862 const char *devname = netdev_get_name(port->up.netdev);
1863 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1864 const char *dp_port_name;
1866 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1867 xlate_ofport_remove(port);
1869 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1871 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1872 /* The underlying device is still there, so delete it. This
1873 * happens when the ofproto is being destroyed, since the caller
1874 * assumes that removal of attached ports will happen as part of
1876 if (!port->is_tunnel) {
1877 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1882 port->peer->peer = NULL;
1886 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1887 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1888 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1889 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1893 sset_find_and_delete(&ofproto->ports, devname);
1894 sset_find_and_delete(&ofproto->ghost_ports, devname);
1895 bundle_remove(port_);
1896 set_cfm(port_, NULL);
1897 set_bfd(port_, NULL);
1898 if (ofproto->sflow) {
1899 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1906 port_modified(struct ofport *port_)
1908 struct ofport_dpif *port = ofport_dpif_cast(port_);
1910 if (port->bundle && port->bundle->bond) {
1911 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1915 cfm_set_netdev(port->cfm, port->up.netdev);
1918 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1920 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1924 ofport_update_peer(port);
1928 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1930 struct ofport_dpif *port = ofport_dpif_cast(port_);
1931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1932 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1934 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1935 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1936 OFPUTIL_PC_NO_PACKET_IN)) {
1937 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1939 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1940 bundle_update(port->bundle);
1946 set_sflow(struct ofproto *ofproto_,
1947 const struct ofproto_sflow_options *sflow_options)
1949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1950 struct dpif_sflow *ds = ofproto->sflow;
1952 if (sflow_options) {
1954 struct ofport_dpif *ofport;
1956 ds = ofproto->sflow = dpif_sflow_create();
1957 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1958 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1960 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1962 dpif_sflow_set_options(ds, sflow_options);
1965 dpif_sflow_unref(ds);
1966 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1967 ofproto->sflow = NULL;
1975 struct ofproto *ofproto_,
1976 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1977 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1978 size_t n_flow_exporters_options)
1980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1981 struct dpif_ipfix *di = ofproto->ipfix;
1983 if (bridge_exporter_options || flow_exporters_options) {
1985 di = ofproto->ipfix = dpif_ipfix_create();
1987 dpif_ipfix_set_options(
1988 di, bridge_exporter_options, flow_exporters_options,
1989 n_flow_exporters_options);
1992 dpif_ipfix_unref(di);
1993 ofproto->ipfix = NULL;
2000 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2002 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 struct ofproto_dpif *ofproto;
2011 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2012 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2013 ofport->cfm = cfm_create(ofport->up.netdev);
2016 if (cfm_configure(ofport->cfm, s)) {
2022 cfm_unref(ofport->cfm);
2028 get_cfm_status(const struct ofport *ofport_,
2029 struct ofproto_cfm_status *status)
2031 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2034 status->faults = cfm_get_fault(ofport->cfm);
2035 status->remote_opstate = cfm_get_opup(ofport->cfm);
2036 status->health = cfm_get_health(ofport->cfm);
2037 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2045 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2048 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2053 if (ofport->bfd != old) {
2054 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2061 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2063 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2066 bfd_get_status(ofport->bfd, smap);
2073 /* Spanning Tree. */
2076 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2078 struct ofproto_dpif *ofproto = ofproto_;
2079 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2080 struct ofport_dpif *ofport;
2082 ofport = stp_port_get_aux(sp);
2084 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2085 ofproto->up.name, port_num);
2087 struct eth_header *eth = pkt->l2;
2089 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2090 if (eth_addr_is_zero(eth->eth_src)) {
2091 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2092 "with unknown MAC", ofproto->up.name, port_num);
2094 send_packet(ofport, pkt);
2100 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2102 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2106 /* Only revalidate flows if the configuration changed. */
2107 if (!s != !ofproto->stp) {
2108 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2112 if (!ofproto->stp) {
2113 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2114 send_bpdu_cb, ofproto);
2115 ofproto->stp_last_tick = time_msec();
2118 stp_set_bridge_id(ofproto->stp, s->system_id);
2119 stp_set_bridge_priority(ofproto->stp, s->priority);
2120 stp_set_hello_time(ofproto->stp, s->hello_time);
2121 stp_set_max_age(ofproto->stp, s->max_age);
2122 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2124 struct ofport *ofport;
2126 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2127 set_stp_port(ofport, NULL);
2130 stp_unref(ofproto->stp);
2131 ofproto->stp = NULL;
2138 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2144 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2145 s->designated_root = stp_get_designated_root(ofproto->stp);
2146 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2155 update_stp_port_state(struct ofport_dpif *ofport)
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2158 enum stp_state state;
2160 /* Figure out new state. */
2161 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2165 if (ofport->stp_state != state) {
2166 enum ofputil_port_state of_state;
2169 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2170 netdev_get_name(ofport->up.netdev),
2171 stp_state_name(ofport->stp_state),
2172 stp_state_name(state));
2173 if (stp_learn_in_state(ofport->stp_state)
2174 != stp_learn_in_state(state)) {
2175 /* xxx Learning action flows should also be flushed. */
2176 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2177 mac_learning_flush(ofproto->ml);
2178 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2180 fwd_change = stp_forward_in_state(ofport->stp_state)
2181 != stp_forward_in_state(state);
2183 ofproto->backer->need_revalidate = REV_STP;
2184 ofport->stp_state = state;
2185 ofport->stp_state_entered = time_msec();
2187 if (fwd_change && ofport->bundle) {
2188 bundle_update(ofport->bundle);
2191 /* Update the STP state bits in the OpenFlow port description. */
2192 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2193 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2194 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2195 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2196 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2198 ofproto_port_set_state(&ofport->up, of_state);
2202 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2203 * caller is responsible for assigning STP port numbers and ensuring
2204 * there are no duplicates. */
2206 set_stp_port(struct ofport *ofport_,
2207 const struct ofproto_port_stp_settings *s)
2209 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2210 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2211 struct stp_port *sp = ofport->stp_port;
2213 if (!s || !s->enable) {
2215 ofport->stp_port = NULL;
2216 stp_port_disable(sp);
2217 update_stp_port_state(ofport);
2220 } else if (sp && stp_port_no(sp) != s->port_num
2221 && ofport == stp_port_get_aux(sp)) {
2222 /* The port-id changed, so disable the old one if it's not
2223 * already in use by another port. */
2224 stp_port_disable(sp);
2227 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2228 stp_port_enable(sp);
2230 stp_port_set_aux(sp, ofport);
2231 stp_port_set_priority(sp, s->priority);
2232 stp_port_set_path_cost(sp, s->path_cost);
2234 update_stp_port_state(ofport);
2240 get_stp_port_status(struct ofport *ofport_,
2241 struct ofproto_port_stp_status *s)
2243 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2245 struct stp_port *sp = ofport->stp_port;
2247 if (!ofproto->stp || !sp) {
2253 s->port_id = stp_port_get_id(sp);
2254 s->state = stp_port_get_state(sp);
2255 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2256 s->role = stp_port_get_role(sp);
2257 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2263 stp_run(struct ofproto_dpif *ofproto)
2266 long long int now = time_msec();
2267 long long int elapsed = now - ofproto->stp_last_tick;
2268 struct stp_port *sp;
2271 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2272 ofproto->stp_last_tick = now;
2274 while (stp_get_changed_port(ofproto->stp, &sp)) {
2275 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2278 update_stp_port_state(ofport);
2282 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2283 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2284 mac_learning_flush(ofproto->ml);
2285 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2291 stp_wait(struct ofproto_dpif *ofproto)
2294 poll_timer_wait(1000);
2299 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2302 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2303 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2305 if (ofport->n_qdscp != n_qdscp
2306 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2307 n_qdscp * sizeof *qdscp))) {
2308 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2309 free(ofport->qdscp);
2310 ofport->qdscp = n_qdscp
2311 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2313 ofport->n_qdscp = n_qdscp;
2321 /* Expires all MAC learning entries associated with 'bundle' and forces its
2322 * ofproto to revalidate every flow.
2324 * Normally MAC learning entries are removed only from the ofproto associated
2325 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2326 * are removed from every ofproto. When patch ports and SLB bonds are in use
2327 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2328 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2329 * with the host from which it migrated. */
2331 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2333 struct ofproto_dpif *ofproto = bundle->ofproto;
2334 struct mac_learning *ml = ofproto->ml;
2335 struct mac_entry *mac, *next_mac;
2337 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2338 ovs_rwlock_wrlock(&ml->rwlock);
2339 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2340 if (mac->port.p == bundle) {
2342 struct ofproto_dpif *o;
2344 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2346 struct mac_entry *e;
2348 ovs_rwlock_wrlock(&o->ml->rwlock);
2349 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2351 mac_learning_expire(o->ml, e);
2353 ovs_rwlock_unlock(&o->ml->rwlock);
2358 mac_learning_expire(ml, mac);
2361 ovs_rwlock_unlock(&ml->rwlock);
2364 static struct ofbundle *
2365 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2367 struct ofbundle *bundle;
2369 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2370 &ofproto->bundles) {
2371 if (bundle->aux == aux) {
2379 bundle_update(struct ofbundle *bundle)
2381 struct ofport_dpif *port;
2383 bundle->floodable = true;
2384 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2385 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2386 || !stp_forward_in_state(port->stp_state)) {
2387 bundle->floodable = false;
2394 bundle_del_port(struct ofport_dpif *port)
2396 struct ofbundle *bundle = port->bundle;
2398 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2400 list_remove(&port->bundle_node);
2401 port->bundle = NULL;
2404 lacp_slave_unregister(bundle->lacp, port);
2407 bond_slave_unregister(bundle->bond, port);
2410 bundle_update(bundle);
2414 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2415 struct lacp_slave_settings *lacp)
2417 struct ofport_dpif *port;
2419 port = get_ofp_port(bundle->ofproto, ofp_port);
2424 if (port->bundle != bundle) {
2425 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2427 bundle_del_port(port);
2430 port->bundle = bundle;
2431 list_push_back(&bundle->ports, &port->bundle_node);
2432 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2433 || !stp_forward_in_state(port->stp_state)) {
2434 bundle->floodable = false;
2438 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2439 lacp_slave_register(bundle->lacp, port, lacp);
2446 bundle_destroy(struct ofbundle *bundle)
2448 struct ofproto_dpif *ofproto;
2449 struct ofport_dpif *port, *next_port;
2455 ofproto = bundle->ofproto;
2456 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2458 xlate_bundle_remove(bundle);
2460 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2461 bundle_del_port(port);
2464 bundle_flush_macs(bundle, true);
2465 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2467 free(bundle->trunks);
2468 lacp_unref(bundle->lacp);
2469 bond_unref(bundle->bond);
2474 bundle_set(struct ofproto *ofproto_, void *aux,
2475 const struct ofproto_bundle_settings *s)
2477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2478 bool need_flush = false;
2479 struct ofport_dpif *port;
2480 struct ofbundle *bundle;
2481 unsigned long *trunks;
2487 bundle_destroy(bundle_lookup(ofproto, aux));
2491 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2492 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2494 bundle = bundle_lookup(ofproto, aux);
2496 bundle = xmalloc(sizeof *bundle);
2498 bundle->ofproto = ofproto;
2499 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2500 hash_pointer(aux, 0));
2502 bundle->name = NULL;
2504 list_init(&bundle->ports);
2505 bundle->vlan_mode = PORT_VLAN_TRUNK;
2507 bundle->trunks = NULL;
2508 bundle->use_priority_tags = s->use_priority_tags;
2509 bundle->lacp = NULL;
2510 bundle->bond = NULL;
2512 bundle->floodable = true;
2513 mbridge_register_bundle(ofproto->mbridge, bundle);
2516 if (!bundle->name || strcmp(s->name, bundle->name)) {
2518 bundle->name = xstrdup(s->name);
2523 if (!bundle->lacp) {
2524 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2525 bundle->lacp = lacp_create();
2527 lacp_configure(bundle->lacp, s->lacp);
2529 lacp_unref(bundle->lacp);
2530 bundle->lacp = NULL;
2533 /* Update set of ports. */
2535 for (i = 0; i < s->n_slaves; i++) {
2536 if (!bundle_add_port(bundle, s->slaves[i],
2537 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2541 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2542 struct ofport_dpif *next_port;
2544 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2545 for (i = 0; i < s->n_slaves; i++) {
2546 if (s->slaves[i] == port->up.ofp_port) {
2551 bundle_del_port(port);
2555 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2557 if (list_is_empty(&bundle->ports)) {
2558 bundle_destroy(bundle);
2562 /* Set VLAN tagging mode */
2563 if (s->vlan_mode != bundle->vlan_mode
2564 || s->use_priority_tags != bundle->use_priority_tags) {
2565 bundle->vlan_mode = s->vlan_mode;
2566 bundle->use_priority_tags = s->use_priority_tags;
2571 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2572 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2574 if (vlan != bundle->vlan) {
2575 bundle->vlan = vlan;
2579 /* Get trunked VLANs. */
2580 switch (s->vlan_mode) {
2581 case PORT_VLAN_ACCESS:
2585 case PORT_VLAN_TRUNK:
2586 trunks = CONST_CAST(unsigned long *, s->trunks);
2589 case PORT_VLAN_NATIVE_UNTAGGED:
2590 case PORT_VLAN_NATIVE_TAGGED:
2591 if (vlan != 0 && (!s->trunks
2592 || !bitmap_is_set(s->trunks, vlan)
2593 || bitmap_is_set(s->trunks, 0))) {
2594 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2596 trunks = bitmap_clone(s->trunks, 4096);
2598 trunks = bitmap_allocate1(4096);
2600 bitmap_set1(trunks, vlan);
2601 bitmap_set0(trunks, 0);
2603 trunks = CONST_CAST(unsigned long *, s->trunks);
2610 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2611 free(bundle->trunks);
2612 if (trunks == s->trunks) {
2613 bundle->trunks = vlan_bitmap_clone(trunks);
2615 bundle->trunks = trunks;
2620 if (trunks != s->trunks) {
2625 if (!list_is_short(&bundle->ports)) {
2626 bundle->ofproto->has_bonded_bundles = true;
2628 if (bond_reconfigure(bundle->bond, s->bond)) {
2629 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2632 bundle->bond = bond_create(s->bond);
2633 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2636 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2637 bond_slave_register(bundle->bond, port, port->up.netdev);
2640 bond_unref(bundle->bond);
2641 bundle->bond = NULL;
2644 /* If we changed something that would affect MAC learning, un-learn
2645 * everything on this port and force flow revalidation. */
2647 bundle_flush_macs(bundle, false);
2654 bundle_remove(struct ofport *port_)
2656 struct ofport_dpif *port = ofport_dpif_cast(port_);
2657 struct ofbundle *bundle = port->bundle;
2660 bundle_del_port(port);
2661 if (list_is_empty(&bundle->ports)) {
2662 bundle_destroy(bundle);
2663 } else if (list_is_short(&bundle->ports)) {
2664 bond_unref(bundle->bond);
2665 bundle->bond = NULL;
2671 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2674 struct ofport_dpif *port = port_;
2675 uint8_t ea[ETH_ADDR_LEN];
2678 error = netdev_get_etheraddr(port->up.netdev, ea);
2680 struct ofpbuf packet;
2683 ofpbuf_init(&packet, 0);
2684 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2686 memcpy(packet_pdu, pdu, pdu_size);
2688 send_packet(port, &packet);
2689 ofpbuf_uninit(&packet);
2691 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2692 "%s (%s)", port->bundle->name,
2693 netdev_get_name(port->up.netdev), ovs_strerror(error));
2698 bundle_send_learning_packets(struct ofbundle *bundle)
2700 struct ofproto_dpif *ofproto = bundle->ofproto;
2701 int error, n_packets, n_errors;
2702 struct mac_entry *e;
2704 error = n_packets = n_errors = 0;
2705 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2706 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2707 if (e->port.p != bundle) {
2708 struct ofpbuf *learning_packet;
2709 struct ofport_dpif *port;
2713 /* The assignment to "port" is unnecessary but makes "grep"ing for
2714 * struct ofport_dpif more effective. */
2715 learning_packet = bond_compose_learning_packet(bundle->bond,
2719 ret = send_packet(port, learning_packet);
2720 ofpbuf_delete(learning_packet);
2728 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2731 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2732 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2733 "packets, last error was: %s",
2734 bundle->name, n_errors, n_packets, ovs_strerror(error));
2736 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2737 bundle->name, n_packets);
2742 bundle_run(struct ofbundle *bundle)
2745 lacp_run(bundle->lacp, send_pdu_cb);
2748 struct ofport_dpif *port;
2750 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2751 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2754 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2755 bundle->ofproto->backer->need_revalidate = REV_BOND;
2758 if (bond_should_send_learning_packets(bundle->bond)) {
2759 bundle_send_learning_packets(bundle);
2765 bundle_wait(struct ofbundle *bundle)
2768 lacp_wait(bundle->lacp);
2771 bond_wait(bundle->bond);
2778 mirror_set__(struct ofproto *ofproto_, void *aux,
2779 const struct ofproto_mirror_settings *s)
2781 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2782 struct ofbundle **srcs, **dsts;
2787 mirror_destroy(ofproto->mbridge, aux);
2791 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2792 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2794 for (i = 0; i < s->n_srcs; i++) {
2795 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2798 for (i = 0; i < s->n_dsts; i++) {
2799 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2802 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2803 s->n_dsts, s->src_vlans,
2804 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2811 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2812 uint64_t *packets, uint64_t *bytes)
2815 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2820 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2822 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2823 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2824 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2825 mac_learning_flush(ofproto->ml);
2827 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2832 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2835 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2836 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2840 forward_bpdu_changed(struct ofproto *ofproto_)
2842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2843 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2847 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2851 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2852 mac_learning_set_idle_time(ofproto->ml, idle_time);
2853 mac_learning_set_max_entries(ofproto->ml, max_entries);
2854 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2859 static struct ofport_dpif *
2860 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2862 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2863 return ofport ? ofport_dpif_cast(ofport) : NULL;
2866 static struct ofport_dpif *
2867 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2869 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2870 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2874 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2875 struct ofproto_port *ofproto_port,
2876 struct dpif_port *dpif_port)
2878 ofproto_port->name = dpif_port->name;
2879 ofproto_port->type = dpif_port->type;
2880 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2884 ofport_update_peer(struct ofport_dpif *ofport)
2886 const struct ofproto_dpif *ofproto;
2887 struct dpif_backer *backer;
2888 const char *peer_name;
2890 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2894 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2895 backer->need_revalidate = REV_RECONFIGURE;
2898 ofport->peer->peer = NULL;
2899 ofport->peer = NULL;
2902 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2907 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2908 struct ofport *peer_ofport;
2909 struct ofport_dpif *peer;
2910 const char *peer_peer;
2912 if (ofproto->backer != backer) {
2916 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2921 peer = ofport_dpif_cast(peer_ofport);
2922 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2923 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2925 ofport->peer = peer;
2926 ofport->peer->peer = ofport;
2934 port_run_fast(struct ofport_dpif *ofport)
2936 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2937 struct ofpbuf packet;
2939 ofpbuf_init(&packet, 0);
2940 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2941 send_packet(ofport, &packet);
2942 ofpbuf_uninit(&packet);
2945 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2946 struct ofpbuf packet;
2948 ofpbuf_init(&packet, 0);
2949 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2950 send_packet(ofport, &packet);
2951 ofpbuf_uninit(&packet);
2956 port_run(struct ofport_dpif *ofport)
2958 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2959 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2960 bool enable = netdev_get_carrier(ofport->up.netdev);
2962 ofport->carrier_seq = carrier_seq;
2964 port_run_fast(ofport);
2967 int cfm_opup = cfm_get_opup(ofport->cfm);
2969 cfm_run(ofport->cfm);
2970 enable = enable && !cfm_get_fault(ofport->cfm);
2972 if (cfm_opup >= 0) {
2973 enable = enable && cfm_opup;
2978 bfd_run(ofport->bfd);
2979 enable = enable && bfd_forwarding(ofport->bfd);
2982 if (ofport->bundle) {
2983 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2984 if (carrier_changed) {
2985 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2989 if (ofport->may_enable != enable) {
2990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2991 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2994 ofport->may_enable = enable;
2998 port_wait(struct ofport_dpif *ofport)
3001 cfm_wait(ofport->cfm);
3005 bfd_wait(ofport->bfd);
3010 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3011 struct ofproto_port *ofproto_port)
3013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3014 struct dpif_port dpif_port;
3017 if (sset_contains(&ofproto->ghost_ports, devname)) {
3018 const char *type = netdev_get_type_from_name(devname);
3020 /* We may be called before ofproto->up.port_by_name is populated with
3021 * the appropriate ofport. For this reason, we must get the name and
3022 * type from the netdev layer directly. */
3024 const struct ofport *ofport;
3026 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3027 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3028 ofproto_port->name = xstrdup(devname);
3029 ofproto_port->type = xstrdup(type);
3035 if (!sset_contains(&ofproto->ports, devname)) {
3038 error = dpif_port_query_by_name(ofproto->backer->dpif,
3039 devname, &dpif_port);
3041 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3047 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3050 const char *devname = netdev_get_name(netdev);
3051 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3052 const char *dp_port_name;
3054 if (netdev_vport_is_patch(netdev)) {
3055 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3059 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3060 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3061 odp_port_t port_no = ODPP_NONE;
3064 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3068 if (netdev_get_tunnel_config(netdev)) {
3069 simap_put(&ofproto->backer->tnl_backers,
3070 dp_port_name, odp_to_u32(port_no));
3074 if (netdev_get_tunnel_config(netdev)) {
3075 sset_add(&ofproto->ghost_ports, devname);
3077 sset_add(&ofproto->ports, devname);
3083 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3086 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3093 sset_find_and_delete(&ofproto->ghost_ports,
3094 netdev_get_name(ofport->up.netdev));
3095 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3096 if (!ofport->is_tunnel) {
3097 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3099 /* The caller is going to close ofport->up.netdev. If this is a
3100 * bonded port, then the bond is using that netdev, so remove it
3101 * from the bond. The client will need to reconfigure everything
3102 * after deleting ports, so then the slave will get re-added. */
3103 bundle_remove(&ofport->up);
3110 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3112 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3117 error = netdev_get_stats(ofport->up.netdev, stats);
3119 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3122 /* ofproto->stats.tx_packets represents packets that we created
3123 * internally and sent to some port (e.g. packets sent with
3124 * send_packet()). Account for them as if they had come from
3125 * OFPP_LOCAL and got forwarded. */
3127 if (stats->rx_packets != UINT64_MAX) {
3128 stats->rx_packets += ofproto->stats.tx_packets;
3131 if (stats->rx_bytes != UINT64_MAX) {
3132 stats->rx_bytes += ofproto->stats.tx_bytes;
3135 /* ofproto->stats.rx_packets represents packets that were received on
3136 * some port and we processed internally and dropped (e.g. STP).
3137 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3139 if (stats->tx_packets != UINT64_MAX) {
3140 stats->tx_packets += ofproto->stats.rx_packets;
3143 if (stats->tx_bytes != UINT64_MAX) {
3144 stats->tx_bytes += ofproto->stats.rx_bytes;
3151 struct port_dump_state {
3156 struct ofproto_port port;
3161 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3163 *statep = xzalloc(sizeof(struct port_dump_state));
3168 port_dump_next(const struct ofproto *ofproto_, void *state_,
3169 struct ofproto_port *port)
3171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3172 struct port_dump_state *state = state_;
3173 const struct sset *sset;
3174 struct sset_node *node;
3176 if (state->has_port) {
3177 ofproto_port_destroy(&state->port);
3178 state->has_port = false;
3180 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3181 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3184 error = port_query_by_name(ofproto_, node->name, &state->port);
3186 *port = state->port;
3187 state->has_port = true;
3189 } else if (error != ENODEV) {
3194 if (!state->ghost) {
3195 state->ghost = true;
3198 return port_dump_next(ofproto_, state_, port);
3205 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3207 struct port_dump_state *state = state_;
3209 if (state->has_port) {
3210 ofproto_port_destroy(&state->port);
3217 port_poll(const struct ofproto *ofproto_, char **devnamep)
3219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3221 if (ofproto->port_poll_errno) {
3222 int error = ofproto->port_poll_errno;
3223 ofproto->port_poll_errno = 0;
3227 if (sset_is_empty(&ofproto->port_poll_set)) {
3231 *devnamep = sset_pop(&ofproto->port_poll_set);
3236 port_poll_wait(const struct ofproto *ofproto_)
3238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3239 dpif_port_poll_wait(ofproto->backer->dpif);
3243 port_is_lacp_current(const struct ofport *ofport_)
3245 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3246 return (ofport->bundle && ofport->bundle->lacp
3247 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3251 /* Upcall handling. */
3253 /* Flow miss batching.
3255 * Some dpifs implement operations faster when you hand them off in a batch.
3256 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3257 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3258 * more packets, plus possibly installing the flow in the dpif.
3260 * So far we only batch the operations that affect flow setup time the most.
3261 * It's possible to batch more than that, but the benefit might be minimal. */
3263 struct hmap_node hmap_node;
3264 struct ofproto_dpif *ofproto;
3266 enum odp_key_fitness key_fitness;
3267 const struct nlattr *key;
3269 struct list packets;
3270 enum dpif_upcall_type upcall_type;
3273 struct flow_miss_op {
3274 struct dpif_op dpif_op;
3276 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3277 struct xlate_out xout;
3278 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3280 struct ofpbuf mask; /* Flow mask for "put" ops. */
3281 struct odputil_keybuf maskbuf;
3283 /* If this is a "put" op, then a pointer to the subfacet that should
3284 * be marked as uninstalled if the operation fails. */
3285 struct subfacet *subfacet;
3288 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3289 * OpenFlow controller as necessary according to their individual
3290 * configurations. */
3292 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3293 const struct flow *flow)
3295 struct ofputil_packet_in pin;
3297 pin.packet = packet->data;
3298 pin.packet_len = packet->size;
3299 pin.reason = OFPR_NO_MATCH;
3300 pin.controller_id = 0;
3305 pin.send_len = 0; /* not used for flow table misses */
3307 flow_get_metadata(flow, &pin.fmd);
3309 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3312 static struct flow_miss *
3313 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3314 const struct flow *flow, uint32_t hash)
3316 struct flow_miss *miss;
3318 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3319 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3327 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3328 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3329 * 'miss' is associated with a subfacet the caller must also initialize the
3330 * returned op->subfacet, and if anything needs to be freed after processing
3331 * the op, the caller must initialize op->garbage also. */
3333 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3334 struct flow_miss_op *op)
3336 if (miss->flow.in_port.ofp_port
3337 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3338 miss->flow.vlan_tci)) {
3339 /* This packet was received on a VLAN splinter port. We
3340 * added a VLAN to the packet to make the packet resemble
3341 * the flow, but the actions were composed assuming that
3342 * the packet contained no VLAN. So, we must remove the
3343 * VLAN header from the packet before trying to execute the
3345 eth_pop_vlan(packet);
3348 op->subfacet = NULL;
3349 op->xout_garbage = false;
3350 op->dpif_op.type = DPIF_OP_EXECUTE;
3351 op->dpif_op.u.execute.key = miss->key;
3352 op->dpif_op.u.execute.key_len = miss->key_len;
3353 op->dpif_op.u.execute.packet = packet;
3354 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3357 /* Helper for handle_flow_miss_without_facet() and
3358 * handle_flow_miss_with_facet(). */
3360 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3361 const struct flow *flow, bool fail_open)
3365 * Extra-special case for fail-open mode.
3367 * We are in fail-open mode and the packet matched the fail-open
3368 * rule, but we are connected to a controller too. We should send
3369 * the packet up to the controller in the hope that it will try to
3370 * set up a flow and thereby allow us to exit fail-open.
3372 * See the top-level comment in fail-open.c for more information.
3374 send_packet_in_miss(ofproto, packet, flow);
3378 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3379 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3380 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3381 * return value of true). However, for short flows the cost of bookkeeping is
3382 * much higher than the benefits, so when the datapath holds a large number of
3383 * flows we impose some heuristics to decide which flows are likely to be worth
3386 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3388 struct dpif_backer *backer = miss->ofproto->backer;
3391 switch (flow_miss_model) {
3392 case OFPROTO_HANDLE_MISS_AUTO:
3394 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3396 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3400 if (!backer->governor) {
3403 n_subfacets = hmap_count(&backer->subfacets);
3404 if (n_subfacets * 2 <= flow_eviction_threshold) {
3408 backer->governor = governor_create();
3411 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3412 return governor_should_install_flow(backer->governor, hash,
3413 list_size(&miss->packets));
3416 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3417 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3418 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3420 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3421 struct flow_miss *miss,
3422 struct flow_miss_op *ops, size_t *n_ops)
3424 struct ofpbuf *packet;
3426 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3428 COVERAGE_INC(facet_suppress);
3430 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3431 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3434 struct xlate_in xin;
3436 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3437 xlate_actions_for_side_effects(&xin);
3440 if (xout->odp_actions.size) {
3441 struct flow_miss_op *op = &ops[*n_ops];
3442 struct dpif_execute *execute = &op->dpif_op.u.execute;
3444 init_flow_miss_execute_op(miss, packet, op);
3445 xlate_out_copy(&op->xout, xout);
3446 execute->actions = op->xout.odp_actions.data;
3447 execute->actions_len = op->xout.odp_actions.size;
3448 op->xout_garbage = true;
3455 /* Handles 'miss', which matches 'facet'. May add any required datapath
3456 * operations to 'ops', incrementing '*n_ops' for each new op.
3458 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3459 * This is really important only for new facets: if we just called time_msec()
3460 * here, then the new subfacet or its packets could look (occasionally) as
3461 * though it was used some time after the facet was used. That can make a
3462 * one-packet flow look like it has a nonzero duration, which looks odd in
3463 * e.g. NetFlow statistics.
3465 * If non-null, 'stats' will be folded into 'facet'. */
3467 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3468 long long int now, struct dpif_flow_stats *stats,
3469 struct flow_miss_op *ops, size_t *n_ops)
3471 enum subfacet_path want_path;
3472 struct subfacet *subfacet;
3473 struct ofpbuf *packet;
3475 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3477 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3478 struct flow_miss_op *op = &ops[*n_ops];
3480 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3483 if (want_path != SF_FAST_PATH) {
3484 struct rule_dpif *rule;
3485 struct xlate_in xin;
3487 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3488 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3489 xlate_actions_for_side_effects(&xin);
3492 if (facet->xout.odp_actions.size) {
3493 struct dpif_execute *execute = &op->dpif_op.u.execute;
3495 init_flow_miss_execute_op(miss, packet, op);
3496 execute->actions = facet->xout.odp_actions.data,
3497 execute->actions_len = facet->xout.odp_actions.size;
3502 /* Don't install the flow if it's the result of the "userspace"
3503 * action for an already installed facet. This can occur when a
3504 * datapath flow with wildcards has a "userspace" action and flows
3505 * sent to userspace result in a different subfacet, which will then
3506 * be rejected as overlapping by the datapath. */
3507 if (miss->upcall_type == DPIF_UC_ACTION
3508 && !list_is_empty(&facet->subfacets)) {
3510 facet->used = MAX(facet->used, stats->used);
3511 facet->packet_count += stats->n_packets;
3512 facet->byte_count += stats->n_bytes;
3513 facet->tcp_flags |= stats->tcp_flags;
3518 subfacet = subfacet_create(facet, miss, now);
3520 subfacet_update_stats(subfacet, stats);
3523 if (subfacet->path != want_path) {
3524 struct flow_miss_op *op = &ops[(*n_ops)++];
3525 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3527 subfacet->path = want_path;
3529 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3530 if (enable_megaflows) {
3531 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3532 &miss->flow, UINT32_MAX);
3535 op->xout_garbage = false;
3536 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3537 op->subfacet = subfacet;
3538 put->flags = DPIF_FP_CREATE;
3539 put->key = miss->key;
3540 put->key_len = miss->key_len;
3541 put->mask = op->mask.data;
3542 put->mask_len = op->mask.size;
3544 if (want_path == SF_FAST_PATH) {
3545 put->actions = facet->xout.odp_actions.data;
3546 put->actions_len = facet->xout.odp_actions.size;
3548 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3549 op->slow_stub, sizeof op->slow_stub,
3550 &put->actions, &put->actions_len);
3556 /* Handles flow miss 'miss'. May add any required datapath operations
3557 * to 'ops', incrementing '*n_ops' for each new op. */
3559 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3562 struct ofproto_dpif *ofproto = miss->ofproto;
3563 struct dpif_flow_stats stats__;
3564 struct dpif_flow_stats *stats = &stats__;
3565 struct ofpbuf *packet;
3566 struct facet *facet;
3570 memset(stats, 0, sizeof *stats);
3572 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3573 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3574 stats->n_bytes += packet->size;
3578 facet = facet_lookup_valid(ofproto, &miss->flow);
3580 struct flow_wildcards wc;
3581 struct rule_dpif *rule;
3582 struct xlate_out xout;
3583 struct xlate_in xin;
3585 flow_wildcards_init_catchall(&wc);
3586 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3587 rule_credit_stats(rule, stats);
3589 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3591 xin.resubmit_stats = stats;
3592 xin.may_learn = true;
3593 xlate_actions(&xin, &xout);
3594 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3596 /* There does not exist a bijection between 'struct flow' and datapath
3597 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3598 * assumption used throughout the facet and subfacet handling code.
3599 * Since we have to handle these misses in userspace anyway, we simply
3600 * skip facet creation, avoiding the problem altogether. */
3601 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3602 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3603 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3607 facet = facet_create(miss, rule, &xout, stats);
3610 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3613 static struct drop_key *
3614 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3617 struct drop_key *drop_key;
3619 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3620 &backer->drop_keys) {
3621 if (drop_key->key_len == key_len
3622 && !memcmp(drop_key->key, key, key_len)) {
3630 drop_key_clear(struct dpif_backer *backer)
3632 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3633 struct drop_key *drop_key, *next;
3635 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3638 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3640 if (error && !VLOG_DROP_WARN(&rl)) {
3641 struct ds ds = DS_EMPTY_INITIALIZER;
3642 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3643 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3644 ovs_strerror(error), ds_cstr(&ds));
3648 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3649 free(drop_key->key);
3655 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3658 struct dpif_upcall *upcall;
3659 struct flow_miss *miss;
3660 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3661 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3662 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3672 /* Construct the to-do list.
3674 * This just amounts to extracting the flow from each packet and sticking
3675 * the packets that have the same flow in the same "flow_miss" structure so
3676 * that we can process them together. */
3679 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3680 struct flow_miss *miss = &misses[n_misses];
3681 struct flow_miss *existing_miss;
3682 struct ofproto_dpif *ofproto;
3683 odp_port_t odp_in_port;
3688 error = xlate_receive(backer, upcall->packet, upcall->key,
3689 upcall->key_len, &flow, &miss->key_fitness,
3690 &ofproto, &odp_in_port);
3691 if (error == ENODEV) {
3692 struct drop_key *drop_key;
3694 /* Received packet on datapath port for which we couldn't
3695 * associate an ofproto. This can happen if a port is removed
3696 * while traffic is being received. Print a rate-limited message
3697 * in case it happens frequently. Install a drop flow so
3698 * that future packets of the flow are inexpensively dropped
3700 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3701 "%"PRIu32, odp_in_port);
3703 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3706 ret = dpif_flow_put(backer->dpif,
3707 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3708 upcall->key, upcall->key_len,
3709 NULL, 0, NULL, 0, NULL);
3712 drop_key = xmalloc(sizeof *drop_key);
3713 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3714 drop_key->key_len = upcall->key_len;
3716 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3717 hash_bytes(drop_key->key, drop_key->key_len, 0));
3726 ofproto->n_missed++;
3727 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3728 &flow.tunnel, &flow.in_port, &miss->flow);
3730 /* Add other packets to a to-do list. */
3731 hash = flow_hash(&miss->flow, 0);
3732 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3733 if (!existing_miss) {
3734 hmap_insert(&todo, &miss->hmap_node, hash);
3735 miss->ofproto = ofproto;
3736 miss->key = upcall->key;
3737 miss->key_len = upcall->key_len;
3738 miss->upcall_type = upcall->type;
3739 list_init(&miss->packets);
3743 miss = existing_miss;
3745 list_push_back(&miss->packets, &upcall->packet->list_node);
3748 /* Process each element in the to-do list, constructing the set of
3749 * operations to batch. */
3751 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3752 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3754 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3756 /* Execute batch. */
3757 for (i = 0; i < n_ops; i++) {
3758 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3760 dpif_operate(backer->dpif, dpif_ops, n_ops);
3762 for (i = 0; i < n_ops; i++) {
3763 if (dpif_ops[i]->error != 0
3764 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3765 && flow_miss_ops[i].subfacet) {
3766 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3768 COVERAGE_INC(subfacet_install_fail);
3770 /* Zero-out subfacet counters when installation failed, but
3771 * datapath reported hits. This should not happen and
3772 * indicates a bug, since if the datapath flow exists, we
3773 * should not be attempting to create a new subfacet. A
3774 * buggy datapath could trigger this, so just zero out the
3775 * counters and log an error. */
3776 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3777 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3778 "datapath reported hits");
3779 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3782 subfacet->path = SF_NOT_INSTALLED;
3786 if (flow_miss_ops[i].xout_garbage) {
3787 xlate_out_uninit(&flow_miss_ops[i].xout);
3790 hmap_destroy(&todo);
3793 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3795 classify_upcall(const struct dpif_upcall *upcall)
3797 size_t userdata_len;
3798 union user_action_cookie cookie;
3800 /* First look at the upcall type. */
3801 switch (upcall->type) {
3802 case DPIF_UC_ACTION:
3808 case DPIF_N_UC_TYPES:
3810 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3814 /* "action" upcalls need a closer look. */
3815 if (!upcall->userdata) {
3816 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3819 userdata_len = nl_attr_get_size(upcall->userdata);
3820 if (userdata_len < sizeof cookie.type
3821 || userdata_len > sizeof cookie) {
3822 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3826 memset(&cookie, 0, sizeof cookie);
3827 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3828 if (userdata_len == sizeof cookie.sflow
3829 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3830 return SFLOW_UPCALL;
3831 } else if (userdata_len == sizeof cookie.slow_path
3832 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3834 } else if (userdata_len == sizeof cookie.flow_sample
3835 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3836 return FLOW_SAMPLE_UPCALL;
3837 } else if (userdata_len == sizeof cookie.ipfix
3838 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3839 return IPFIX_UPCALL;
3841 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3842 " and size %zu", cookie.type, userdata_len);
3848 handle_sflow_upcall(struct dpif_backer *backer,
3849 const struct dpif_upcall *upcall)
3851 struct ofproto_dpif *ofproto;
3852 union user_action_cookie cookie;
3854 odp_port_t odp_in_port;
3856 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3857 &flow, NULL, &ofproto, &odp_in_port)
3858 || !ofproto->sflow) {
3862 memset(&cookie, 0, sizeof cookie);
3863 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3864 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3865 odp_in_port, &cookie);
3869 handle_flow_sample_upcall(struct dpif_backer *backer,
3870 const struct dpif_upcall *upcall)
3872 struct ofproto_dpif *ofproto;
3873 union user_action_cookie cookie;
3876 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3877 &flow, NULL, &ofproto, NULL)
3878 || !ofproto->ipfix) {
3882 memset(&cookie, 0, sizeof cookie);
3883 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3885 /* The flow reflects exactly the contents of the packet. Sample
3886 * the packet using it. */
3887 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3888 cookie.flow_sample.collector_set_id,
3889 cookie.flow_sample.probability,
3890 cookie.flow_sample.obs_domain_id,
3891 cookie.flow_sample.obs_point_id);
3895 handle_ipfix_upcall(struct dpif_backer *backer,
3896 const struct dpif_upcall *upcall)
3898 struct ofproto_dpif *ofproto;
3901 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3902 &flow, NULL, &ofproto, NULL)
3903 || !ofproto->ipfix) {
3907 /* The flow reflects exactly the contents of the packet. Sample
3908 * the packet using it. */
3909 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3913 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3915 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3916 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3917 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3922 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3925 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3926 struct dpif_upcall *upcall = &misses[n_misses];
3927 struct ofpbuf *buf = &miss_bufs[n_misses];
3930 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3931 sizeof miss_buf_stubs[n_misses]);
3932 error = dpif_recv(backer->dpif, upcall, buf);
3938 switch (classify_upcall(upcall)) {
3940 /* Handle it later. */
3945 handle_sflow_upcall(backer, upcall);
3949 case FLOW_SAMPLE_UPCALL:
3950 handle_flow_sample_upcall(backer, upcall);
3955 handle_ipfix_upcall(backer, upcall);
3965 /* Handle deferred MISS_UPCALL processing. */
3966 handle_miss_upcalls(backer, misses, n_misses);
3967 for (i = 0; i < n_misses; i++) {
3968 ofpbuf_uninit(&miss_bufs[i]);
3974 /* Flow expiration. */
3976 static int subfacet_max_idle(const struct dpif_backer *);
3977 static void update_stats(struct dpif_backer *);
3978 static void rule_expire(struct rule_dpif *);
3979 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3981 /* This function is called periodically by run(). Its job is to collect
3982 * updates for the flows that have been installed into the datapath, most
3983 * importantly when they last were used, and then use that information to
3984 * expire flows that have not been used recently.
3986 * Returns the number of milliseconds after which it should be called again. */
3988 expire(struct dpif_backer *backer)
3990 struct ofproto_dpif *ofproto;
3994 /* Periodically clear out the drop keys in an effort to keep them
3995 * relatively few. */
3996 drop_key_clear(backer);
3998 /* Update stats for each flow in the backer. */
3999 update_stats(backer);
4001 n_subfacets = hmap_count(&backer->subfacets);
4003 struct subfacet *subfacet;
4004 long long int total, now;
4008 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4009 total += now - subfacet->created;
4011 backer->avg_subfacet_life += total / n_subfacets;
4013 backer->avg_subfacet_life /= 2;
4015 backer->avg_n_subfacet += n_subfacets;
4016 backer->avg_n_subfacet /= 2;
4018 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4020 max_idle = subfacet_max_idle(backer);
4021 expire_subfacets(backer, max_idle);
4023 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4024 struct rule *rule, *next_rule;
4026 if (ofproto->backer != backer) {
4030 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4032 ovs_mutex_lock(&ofproto->up.expirable_mutex);
4033 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4034 &ofproto->up.expirable) {
4035 rule_expire(rule_dpif_cast(rule));
4037 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
4039 /* All outstanding data in existing flows has been accounted, so it's a
4040 * good time to do bond rebalancing. */
4041 if (ofproto->has_bonded_bundles) {
4042 struct ofbundle *bundle;
4044 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4046 bond_rebalance(bundle->bond);
4052 return MIN(max_idle, 1000);
4055 /* Updates flow table statistics given that the datapath just reported 'stats'
4056 * as 'subfacet''s statistics. */
4058 update_subfacet_stats(struct subfacet *subfacet,
4059 const struct dpif_flow_stats *stats)
4061 struct facet *facet = subfacet->facet;
4062 struct dpif_flow_stats diff;
4064 diff.tcp_flags = stats->tcp_flags;
4065 diff.used = stats->used;
4067 if (stats->n_packets >= subfacet->dp_packet_count) {
4068 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4070 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4074 if (stats->n_bytes >= subfacet->dp_byte_count) {
4075 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4077 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4081 facet->ofproto->n_hit += diff.n_packets;
4082 subfacet->dp_packet_count = stats->n_packets;
4083 subfacet->dp_byte_count = stats->n_bytes;
4084 subfacet_update_stats(subfacet, &diff);
4086 if (facet->accounted_bytes < facet->byte_count) {
4088 facet_account(facet);
4089 facet->accounted_bytes = facet->byte_count;
4093 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4094 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4096 delete_unexpected_flow(struct dpif_backer *backer,
4097 const struct nlattr *key, size_t key_len)
4099 if (!VLOG_DROP_WARN(&rl)) {
4103 odp_flow_key_format(key, key_len, &s);
4104 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4108 COVERAGE_INC(facet_unexpected);
4109 dpif_flow_del(backer->dpif, key, key_len, NULL);
4112 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4114 * This function also pushes statistics updates to rules which each facet
4115 * resubmits into. Generally these statistics will be accurate. However, if a
4116 * facet changes the rule it resubmits into at some time in between
4117 * update_stats() runs, it is possible that statistics accrued to the
4118 * old rule will be incorrectly attributed to the new rule. This could be
4119 * avoided by calling update_stats() whenever rules are created or
4120 * deleted. However, the performance impact of making so many calls to the
4121 * datapath do not justify the benefit of having perfectly accurate statistics.
4123 * In addition, this function maintains per ofproto flow hit counts. The patch
4124 * port is not treated specially. e.g. A packet ingress from br0 patched into
4125 * br1 will increase the hit count of br0 by 1, however, does not affect
4126 * the hit or miss counts of br1.
4129 update_stats(struct dpif_backer *backer)
4131 const struct dpif_flow_stats *stats;
4132 struct dpif_flow_dump dump;
4133 const struct nlattr *key, *mask;
4134 size_t key_len, mask_len;
4136 dpif_flow_dump_start(&dump, backer->dpif);
4137 while (dpif_flow_dump_next(&dump, &key, &key_len,
4138 &mask, &mask_len, NULL, NULL, &stats)) {
4139 struct subfacet *subfacet;
4142 key_hash = odp_flow_key_hash(key, key_len);
4143 subfacet = subfacet_find(backer, key, key_len, key_hash);
4144 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4146 update_subfacet_stats(subfacet, stats);
4150 /* Stats are updated per-packet. */
4153 case SF_NOT_INSTALLED:
4155 delete_unexpected_flow(backer, key, key_len);
4160 dpif_flow_dump_done(&dump);
4162 update_moving_averages(backer);
4165 /* Calculates and returns the number of milliseconds of idle time after which
4166 * subfacets should expire from the datapath. When a subfacet expires, we fold
4167 * its statistics into its facet, and when a facet's last subfacet expires, we
4168 * fold its statistic into its rule. */
4170 subfacet_max_idle(const struct dpif_backer *backer)
4173 * Idle time histogram.
4175 * Most of the time a switch has a relatively small number of subfacets.
4176 * When this is the case we might as well keep statistics for all of them
4177 * in userspace and to cache them in the kernel datapath for performance as
4180 * As the number of subfacets increases, the memory required to maintain
4181 * statistics about them in userspace and in the kernel becomes
4182 * significant. However, with a large number of subfacets it is likely
4183 * that only a few of them are "heavy hitters" that consume a large amount
4184 * of bandwidth. At this point, only heavy hitters are worth caching in
4185 * the kernel and maintaining in userspaces; other subfacets we can
4188 * The technique used to compute the idle time is to build a histogram with
4189 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4190 * that is installed in the kernel gets dropped in the appropriate bucket.
4191 * After the histogram has been built, we compute the cutoff so that only
4192 * the most-recently-used 1% of subfacets (but at least
4193 * flow_eviction_threshold flows) are kept cached. At least
4194 * the most-recently-used bucket of subfacets is kept, so actually an
4195 * arbitrary number of subfacets can be kept in any given expiration run
4196 * (though the next run will delete most of those unless they receive
4199 * This requires a second pass through the subfacets, in addition to the
4200 * pass made by update_stats(), because the former function never looks at
4201 * uninstallable subfacets.
4203 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4204 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4205 int buckets[N_BUCKETS] = { 0 };
4206 int total, subtotal, bucket;
4207 struct subfacet *subfacet;
4211 total = hmap_count(&backer->subfacets);
4212 if (total <= flow_eviction_threshold) {
4213 return N_BUCKETS * BUCKET_WIDTH;
4216 /* Build histogram. */
4218 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4219 long long int idle = now - subfacet->used;
4220 int bucket = (idle <= 0 ? 0
4221 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4222 : (unsigned int) idle / BUCKET_WIDTH);
4226 /* Find the first bucket whose flows should be expired. */
4227 subtotal = bucket = 0;
4229 subtotal += buckets[bucket++];
4230 } while (bucket < N_BUCKETS &&
4231 subtotal < MAX(flow_eviction_threshold, total / 100));
4233 if (VLOG_IS_DBG_ENABLED()) {
4237 ds_put_cstr(&s, "keep");
4238 for (i = 0; i < N_BUCKETS; i++) {
4240 ds_put_cstr(&s, ", drop");
4243 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4246 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4250 return bucket * BUCKET_WIDTH;
4254 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4256 /* Cutoff time for most flows. */
4257 long long int normal_cutoff = time_msec() - dp_max_idle;
4259 /* We really want to keep flows for special protocols around, so use a more
4260 * conservative cutoff. */
4261 long long int special_cutoff = time_msec() - 10000;
4263 struct subfacet *subfacet, *next_subfacet;
4264 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4268 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4269 &backer->subfacets) {
4270 long long int cutoff;
4272 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4276 if (subfacet->used < cutoff) {
4277 if (subfacet->path != SF_NOT_INSTALLED) {
4278 batch[n_batch++] = subfacet;
4279 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4280 subfacet_destroy_batch(backer, batch, n_batch);
4284 subfacet_destroy(subfacet);
4290 subfacet_destroy_batch(backer, batch, n_batch);
4294 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4295 * then delete it entirely. */
4297 rule_expire(struct rule_dpif *rule)
4299 uint16_t idle_timeout, hard_timeout;
4303 if (rule->up.pending) {
4304 /* We'll have to expire it later. */
4308 ovs_mutex_lock(&rule->up.timeout_mutex);
4309 hard_timeout = rule->up.hard_timeout;
4310 idle_timeout = rule->up.idle_timeout;
4311 ovs_mutex_unlock(&rule->up.timeout_mutex);
4313 /* Has 'rule' expired? */
4315 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
4316 reason = OFPRR_HARD_TIMEOUT;
4317 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
4318 reason = OFPRR_IDLE_TIMEOUT;
4323 COVERAGE_INC(ofproto_dpif_expired);
4325 /* Get rid of the rule. */
4326 ofproto_rule_expire(&rule->up, reason);
4331 /* Creates and returns a new facet based on 'miss'.
4333 * The caller must already have determined that no facet with an identical
4334 * 'miss->flow' exists in 'miss->ofproto'.
4336 * 'rule' and 'xout' must have been created based on 'miss'.
4338 * 'facet'' statistics are initialized based on 'stats'.
4340 * The facet will initially have no subfacets. The caller should create (at
4341 * least) one subfacet with subfacet_create(). */
4342 static struct facet *
4343 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4344 struct xlate_out *xout, struct dpif_flow_stats *stats)
4346 struct ofproto_dpif *ofproto = miss->ofproto;
4347 struct facet *facet;
4350 facet = xzalloc(sizeof *facet);
4351 facet->ofproto = miss->ofproto;
4352 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4353 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4354 facet->tcp_flags = stats->tcp_flags;
4355 facet->used = stats->used;
4356 facet->flow = miss->flow;
4357 facet->learn_rl = time_msec() + 500;
4359 list_init(&facet->subfacets);
4360 netflow_flow_init(&facet->nf_flow);
4361 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4363 xlate_out_copy(&facet->xout, xout);
4365 match_init(&match, &facet->flow, &facet->xout.wc);
4366 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4367 classifier_insert(&ofproto->facets, &facet->cr);
4369 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4370 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4376 facet_free(struct facet *facet)
4379 xlate_out_uninit(&facet->xout);
4384 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4385 * 'packet', which arrived on 'in_port'. */
4387 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4388 const struct nlattr *odp_actions, size_t actions_len,
4389 struct ofpbuf *packet)
4391 struct odputil_keybuf keybuf;
4395 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4396 odp_flow_key_from_flow(&key, flow,
4397 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4399 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4400 odp_actions, actions_len, packet);
4404 /* Remove 'facet' from its ofproto and free up the associated memory:
4406 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4407 * rule's statistics, via subfacet_uninstall().
4409 * - Removes 'facet' from its rule and from ofproto->facets.
4412 facet_remove(struct facet *facet)
4414 struct subfacet *subfacet, *next_subfacet;
4416 ovs_assert(!list_is_empty(&facet->subfacets));
4418 /* First uninstall all of the subfacets to get final statistics. */
4419 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4420 subfacet_uninstall(subfacet);
4423 /* Flush the final stats to the rule.
4425 * This might require us to have at least one subfacet around so that we
4426 * can use its actions for accounting in facet_account(), which is why we
4427 * have uninstalled but not yet destroyed the subfacets. */
4428 facet_flush_stats(facet);
4430 /* Now we're really all done so destroy everything. */
4431 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4432 &facet->subfacets) {
4433 subfacet_destroy__(subfacet);
4435 classifier_remove(&facet->ofproto->facets, &facet->cr);
4436 cls_rule_destroy(&facet->cr);
4440 /* Feed information from 'facet' back into the learning table to keep it in
4441 * sync with what is actually flowing through the datapath. */
4443 facet_learn(struct facet *facet)
4445 long long int now = time_msec();
4447 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4451 facet->learn_rl = now + 500;
4453 if (!facet->xout.has_learn
4454 && !facet->xout.has_normal
4455 && (!facet->xout.has_fin_timeout
4456 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4460 facet_push_stats(facet, true);
4464 facet_account(struct facet *facet)
4466 const struct nlattr *a;
4471 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4474 n_bytes = facet->byte_count - facet->accounted_bytes;
4476 /* This loop feeds byte counters to bond_account() for rebalancing to use
4477 * as a basis. We also need to track the actual VLAN on which the packet
4478 * is going to be sent to ensure that it matches the one passed to
4479 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4482 * We use the actions from an arbitrary subfacet because they should all
4483 * be equally valid for our purpose. */
4484 vlan_tci = facet->flow.vlan_tci;
4485 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4486 facet->xout.odp_actions.size) {
4487 const struct ovs_action_push_vlan *vlan;
4488 struct ofport_dpif *port;
4490 switch (nl_attr_type(a)) {
4491 case OVS_ACTION_ATTR_OUTPUT:
4492 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4493 if (port && port->bundle && port->bundle->bond) {
4494 bond_account(port->bundle->bond, &facet->flow,
4495 vlan_tci_to_vid(vlan_tci), n_bytes);
4499 case OVS_ACTION_ATTR_POP_VLAN:
4500 vlan_tci = htons(0);
4503 case OVS_ACTION_ATTR_PUSH_VLAN:
4504 vlan = nl_attr_get(a);
4505 vlan_tci = vlan->vlan_tci;
4511 /* Returns true if the only action for 'facet' is to send to the controller.
4512 * (We don't report NetFlow expiration messages for such facets because they
4513 * are just part of the control logic for the network, not real traffic). */
4515 facet_is_controller_flow(struct facet *facet)
4518 struct ofproto_dpif *ofproto = facet->ofproto;
4519 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4521 const struct ofpact *ofpacts = rule->up.ofpacts;
4522 size_t ofpacts_len = rule->up.ofpacts_len;
4524 if (ofpacts_len > 0 &&
4525 ofpacts->type == OFPACT_CONTROLLER &&
4526 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4533 /* Folds all of 'facet''s statistics into its rule. Also updates the
4534 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4535 * 'facet''s statistics in the datapath should have been zeroed and folded into
4536 * its packet and byte counts before this function is called. */
4538 facet_flush_stats(struct facet *facet)
4540 struct ofproto_dpif *ofproto = facet->ofproto;
4541 struct subfacet *subfacet;
4543 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4544 ovs_assert(!subfacet->dp_byte_count);
4545 ovs_assert(!subfacet->dp_packet_count);
4548 facet_push_stats(facet, false);
4549 if (facet->accounted_bytes < facet->byte_count) {
4550 facet_account(facet);
4551 facet->accounted_bytes = facet->byte_count;
4554 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4555 struct ofexpired expired;
4556 expired.flow = facet->flow;
4557 expired.packet_count = facet->packet_count;
4558 expired.byte_count = facet->byte_count;
4559 expired.used = facet->used;
4560 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4563 /* Reset counters to prevent double counting if 'facet' ever gets
4565 facet_reset_counters(facet);
4567 netflow_flow_clear(&facet->nf_flow);
4568 facet->tcp_flags = 0;
4571 /* Searches 'ofproto''s table of facets for one which would be responsible for
4572 * 'flow'. Returns it if found, otherwise a null pointer.
4574 * The returned facet might need revalidation; use facet_lookup_valid()
4575 * instead if that is important. */
4576 static struct facet *
4577 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4579 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4580 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4583 /* Searches 'ofproto''s table of facets for one capable that covers
4584 * 'flow'. Returns it if found, otherwise a null pointer.
4586 * The returned facet is guaranteed to be valid. */
4587 static struct facet *
4588 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4590 struct facet *facet;
4592 facet = facet_find(ofproto, flow);
4594 && ofproto->backer->need_revalidate
4595 && !facet_revalidate(facet)) {
4603 facet_check_consistency(struct facet *facet)
4605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4607 struct xlate_out xout;
4608 struct xlate_in xin;
4610 struct rule_dpif *rule;
4613 /* Check the datapath actions for consistency. */
4614 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4615 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4616 xlate_actions(&xin, &xout);
4618 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4619 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4620 && facet->xout.slow == xout.slow
4621 && facet->fail_open == fail_open;
4622 if (!ok && !VLOG_DROP_WARN(&rl)) {
4623 struct ds s = DS_EMPTY_INITIALIZER;
4625 flow_format(&s, &facet->flow);
4626 ds_put_cstr(&s, ": inconsistency in facet");
4628 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4629 ds_put_cstr(&s, " (actions were: ");
4630 format_odp_actions(&s, facet->xout.odp_actions.data,
4631 facet->xout.odp_actions.size);
4632 ds_put_cstr(&s, ") (correct actions: ");
4633 format_odp_actions(&s, xout.odp_actions.data,
4634 xout.odp_actions.size);
4635 ds_put_char(&s, ')');
4638 if (facet->xout.slow != xout.slow) {
4639 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4642 if (facet->fail_open != fail_open) {
4643 ds_put_format(&s, " fail open incorrect. should be %s",
4644 fail_open ? "true" : "false");
4648 xlate_out_uninit(&xout);
4653 /* Re-searches the classifier for 'facet':
4655 * - If the rule found is different from 'facet''s current rule, moves
4656 * 'facet' to the new rule and recompiles its actions.
4658 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4659 * where it is and recompiles its actions anyway.
4661 * - If any of 'facet''s subfacets correspond to a new flow according to
4662 * xlate_receive(), 'facet' is removed.
4664 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4666 facet_revalidate(struct facet *facet)
4668 struct ofproto_dpif *ofproto = facet->ofproto;
4669 struct rule_dpif *new_rule;
4670 struct subfacet *subfacet;
4671 struct flow_wildcards wc;
4672 struct xlate_out xout;
4673 struct xlate_in xin;
4675 COVERAGE_INC(facet_revalidate);
4677 /* Check that child subfacets still correspond to this facet. Tunnel
4678 * configuration changes could cause a subfacet's OpenFlow in_port to
4680 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4681 struct ofproto_dpif *recv_ofproto;
4682 struct flow recv_flow;
4685 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4686 subfacet->key_len, &recv_flow, NULL,
4687 &recv_ofproto, NULL);
4689 || recv_ofproto != ofproto
4690 || facet != facet_find(ofproto, &recv_flow)) {
4691 facet_remove(facet);
4696 flow_wildcards_init_catchall(&wc);
4697 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4699 /* Calculate new datapath actions.
4701 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4702 * emit a NetFlow expiration and, if so, we need to have the old state
4703 * around to properly compose it. */
4704 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4705 xlate_actions(&xin, &xout);
4706 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4708 /* A facet's slow path reason should only change under dramatic
4709 * circumstances. Rather than try to update everything, it's simpler to
4710 * remove the facet and start over.
4712 * More importantly, if a facet's wildcards change, it will be relatively
4713 * difficult to figure out if its subfacets still belong to it, and if not
4714 * which facet they may belong to. Again, to avoid the complexity, we
4715 * simply give up instead. */
4716 if (facet->xout.slow != xout.slow
4717 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4718 facet_remove(facet);
4719 xlate_out_uninit(&xout);
4723 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4724 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4725 if (subfacet->path == SF_FAST_PATH) {
4726 struct dpif_flow_stats stats;
4728 subfacet_install(subfacet, &xout.odp_actions, &stats);
4729 subfacet_update_stats(subfacet, &stats);
4733 facet_flush_stats(facet);
4735 ofpbuf_clear(&facet->xout.odp_actions);
4736 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4737 xout.odp_actions.size);
4740 /* Update 'facet' now that we've taken care of all the old state. */
4741 facet->xout.slow = xout.slow;
4742 facet->xout.has_learn = xout.has_learn;
4743 facet->xout.has_normal = xout.has_normal;
4744 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4745 facet->xout.nf_output_iface = xout.nf_output_iface;
4746 facet->xout.mirrors = xout.mirrors;
4747 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4748 facet->used = MAX(facet->used, new_rule->up.created);
4749 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4751 xlate_out_uninit(&xout);
4756 facet_reset_counters(struct facet *facet)
4758 facet->packet_count = 0;
4759 facet->byte_count = 0;
4760 facet->prev_packet_count = 0;
4761 facet->prev_byte_count = 0;
4762 facet->accounted_bytes = 0;
4766 facet_push_stats(struct facet *facet, bool may_learn)
4768 struct dpif_flow_stats stats;
4770 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4771 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4772 ovs_assert(facet->used >= facet->prev_used);
4774 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4775 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4776 stats.used = facet->used;
4777 stats.tcp_flags = facet->tcp_flags;
4779 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4780 struct ofproto_dpif *ofproto = facet->ofproto;
4781 struct ofport_dpif *in_port;
4782 struct rule_dpif *rule;
4783 struct xlate_in xin;
4785 facet->prev_packet_count = facet->packet_count;
4786 facet->prev_byte_count = facet->byte_count;
4787 facet->prev_used = facet->used;
4789 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4790 if (in_port && in_port->is_tunnel) {
4791 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4794 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4795 rule_credit_stats(rule, &stats);
4796 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4798 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4799 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4800 stats.n_packets, stats.n_bytes);
4802 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4804 xin.resubmit_stats = &stats;
4805 xin.may_learn = may_learn;
4806 xlate_actions_for_side_effects(&xin);
4811 push_all_stats__(bool run_fast)
4813 static long long int rl = LLONG_MIN;
4814 struct ofproto_dpif *ofproto;
4816 if (time_msec() < rl) {
4820 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4821 struct cls_cursor cursor;
4822 struct facet *facet;
4824 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4825 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4826 facet_push_stats(facet, false);
4833 rl = time_msec() + 100;
4837 push_all_stats(void)
4839 push_all_stats__(true);
4843 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4845 ovs_mutex_lock(&rule->stats_mutex);
4846 rule->packet_count += stats->n_packets;
4847 rule->byte_count += stats->n_bytes;
4848 ofproto_rule_update_used(&rule->up, stats->used);
4849 ovs_mutex_unlock(&rule->stats_mutex);
4854 static struct subfacet *
4855 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4856 size_t key_len, uint32_t key_hash)
4858 struct subfacet *subfacet;
4860 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4861 &backer->subfacets) {
4862 if (subfacet->key_len == key_len
4863 && !memcmp(key, subfacet->key, key_len)) {
4871 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4872 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4873 * existing subfacet if there is one, otherwise creates and returns a
4875 static struct subfacet *
4876 subfacet_create(struct facet *facet, struct flow_miss *miss,
4879 struct dpif_backer *backer = miss->ofproto->backer;
4880 enum odp_key_fitness key_fitness = miss->key_fitness;
4881 const struct nlattr *key = miss->key;
4882 size_t key_len = miss->key_len;
4884 struct subfacet *subfacet;
4886 key_hash = odp_flow_key_hash(key, key_len);
4888 if (list_is_empty(&facet->subfacets)) {
4889 subfacet = &facet->one_subfacet;
4891 subfacet = subfacet_find(backer, key, key_len, key_hash);
4893 if (subfacet->facet == facet) {
4897 /* This shouldn't happen. */
4898 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4899 subfacet_destroy(subfacet);
4902 subfacet = xmalloc(sizeof *subfacet);
4905 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4906 list_push_back(&facet->subfacets, &subfacet->list_node);
4907 subfacet->facet = facet;
4908 subfacet->key_fitness = key_fitness;
4909 subfacet->key = xmemdup(key, key_len);
4910 subfacet->key_len = key_len;
4911 subfacet->used = now;
4912 subfacet->created = now;
4913 subfacet->dp_packet_count = 0;
4914 subfacet->dp_byte_count = 0;
4915 subfacet->path = SF_NOT_INSTALLED;
4916 subfacet->backer = backer;
4918 backer->subfacet_add_count++;
4922 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4923 * its facet within 'ofproto', and frees it. */
4925 subfacet_destroy__(struct subfacet *subfacet)
4927 struct facet *facet = subfacet->facet;
4928 struct ofproto_dpif *ofproto = facet->ofproto;
4930 /* Update ofproto stats before uninstall the subfacet. */
4931 ofproto->backer->subfacet_del_count++;
4933 subfacet_uninstall(subfacet);
4934 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4935 list_remove(&subfacet->list_node);
4936 free(subfacet->key);
4937 if (subfacet != &facet->one_subfacet) {
4942 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4943 * last remaining subfacet in its facet destroys the facet too. */
4945 subfacet_destroy(struct subfacet *subfacet)
4947 struct facet *facet = subfacet->facet;
4949 if (list_is_singleton(&facet->subfacets)) {
4950 /* facet_remove() needs at least one subfacet (it will remove it). */
4951 facet_remove(facet);
4953 subfacet_destroy__(subfacet);
4958 subfacet_destroy_batch(struct dpif_backer *backer,
4959 struct subfacet **subfacets, int n)
4961 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4962 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4963 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4966 for (i = 0; i < n; i++) {
4967 ops[i].type = DPIF_OP_FLOW_DEL;
4968 ops[i].u.flow_del.key = subfacets[i]->key;
4969 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4970 ops[i].u.flow_del.stats = &stats[i];
4974 dpif_operate(backer->dpif, opsp, n);
4975 for (i = 0; i < n; i++) {
4976 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4977 subfacets[i]->path = SF_NOT_INSTALLED;
4978 subfacet_destroy(subfacets[i]);
4983 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4984 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4985 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4986 * since 'subfacet' was last updated.
4988 * Returns 0 if successful, otherwise a positive errno value. */
4990 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4991 struct dpif_flow_stats *stats)
4993 struct facet *facet = subfacet->facet;
4994 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4995 const struct nlattr *actions = odp_actions->data;
4996 size_t actions_len = odp_actions->size;
4997 struct odputil_keybuf maskbuf;
5000 uint64_t slow_path_stub[128 / 8];
5001 enum dpif_flow_put_flags flags;
5004 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5007 flags |= DPIF_FP_ZERO_STATS;
5010 if (path == SF_SLOW_PATH) {
5011 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5012 slow_path_stub, sizeof slow_path_stub,
5013 &actions, &actions_len);
5016 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5017 if (enable_megaflows) {
5018 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5019 &facet->flow, UINT32_MAX);
5022 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5023 subfacet->key_len, mask.data, mask.size,
5024 actions, actions_len, stats);
5027 subfacet_reset_dp_stats(subfacet, stats);
5031 COVERAGE_INC(subfacet_install_fail);
5033 subfacet->path = path;
5038 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5040 subfacet_uninstall(struct subfacet *subfacet)
5042 if (subfacet->path != SF_NOT_INSTALLED) {
5043 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5044 struct dpif_flow_stats stats;
5047 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5048 subfacet->key_len, &stats);
5049 subfacet_reset_dp_stats(subfacet, &stats);
5051 subfacet_update_stats(subfacet, &stats);
5053 subfacet->path = SF_NOT_INSTALLED;
5055 ovs_assert(subfacet->dp_packet_count == 0);
5056 ovs_assert(subfacet->dp_byte_count == 0);
5060 /* Resets 'subfacet''s datapath statistics counters. This should be called
5061 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5062 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5063 * was reset in the datapath. 'stats' will be modified to include only
5064 * statistics new since 'subfacet' was last updated. */
5066 subfacet_reset_dp_stats(struct subfacet *subfacet,
5067 struct dpif_flow_stats *stats)
5070 && subfacet->dp_packet_count <= stats->n_packets
5071 && subfacet->dp_byte_count <= stats->n_bytes) {
5072 stats->n_packets -= subfacet->dp_packet_count;
5073 stats->n_bytes -= subfacet->dp_byte_count;
5076 subfacet->dp_packet_count = 0;
5077 subfacet->dp_byte_count = 0;
5080 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5082 * Because of the meaning of a subfacet's counters, it only makes sense to do
5083 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5084 * represents a packet that was sent by hand or if it represents statistics
5085 * that have been cleared out of the datapath. */
5087 subfacet_update_stats(struct subfacet *subfacet,
5088 const struct dpif_flow_stats *stats)
5090 if (stats->n_packets || stats->used > subfacet->used) {
5091 struct facet *facet = subfacet->facet;
5093 subfacet->used = MAX(subfacet->used, stats->used);
5094 facet->used = MAX(facet->used, stats->used);
5095 facet->packet_count += stats->n_packets;
5096 facet->byte_count += stats->n_bytes;
5097 facet->tcp_flags |= stats->tcp_flags;
5103 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5104 * the fields that were relevant as part of the lookup. */
5105 static struct rule_dpif *
5106 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5107 struct flow_wildcards *wc)
5109 struct ofport_dpif *port;
5110 struct rule_dpif *rule;
5112 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5116 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5118 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5119 flow->in_port.ofp_port);
5122 return choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
5123 ofproto->no_packet_in_rule);
5127 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5128 const struct flow *flow, struct flow_wildcards *wc,
5131 struct cls_rule *cls_rule;
5132 struct classifier *cls;
5135 if (table_id >= N_TABLES) {
5140 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5141 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5144 cls = &ofproto->up.tables[table_id].cls;
5145 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5146 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5147 /* We must pretend that transport ports are unavailable. */
5148 struct flow ofpc_normal_flow = *flow;
5149 ofpc_normal_flow.tp_src = htons(0);
5150 ofpc_normal_flow.tp_dst = htons(0);
5151 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5152 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5153 cls_rule = &ofproto->drop_frags_rule->up.cr;
5155 flow_wildcards_init_exact(wc);
5158 cls_rule = classifier_lookup(cls, flow, wc);
5160 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5163 /* Given a port configuration (specified as zero if there's no port), chooses
5164 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
5165 * flow table miss. */
5167 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
5168 struct rule_dpif *no_packet_in_rule)
5170 return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
5174 complete_operation(struct rule_dpif *rule)
5176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5178 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5180 struct dpif_completion *c = xmalloc(sizeof *c);
5181 c->op = rule->up.pending;
5182 list_push_back(&ofproto->completions, &c->list_node);
5184 ofoperation_complete(rule->up.pending, 0);
5188 static struct rule *
5191 struct rule_dpif *rule = xmalloc(sizeof *rule);
5196 rule_dealloc(struct rule *rule_)
5198 struct rule_dpif *rule = rule_dpif_cast(rule_);
5203 rule_construct(struct rule *rule_)
5205 struct rule_dpif *rule = rule_dpif_cast(rule_);
5206 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5207 ovs_mutex_lock(&rule->stats_mutex);
5208 rule->packet_count = 0;
5209 rule->byte_count = 0;
5210 ovs_mutex_unlock(&rule->stats_mutex);
5211 complete_operation(rule);
5216 rule_destruct(struct rule *rule_)
5218 struct rule_dpif *rule = rule_dpif_cast(rule_);
5219 complete_operation(rule);
5220 ovs_mutex_destroy(&rule->stats_mutex);
5224 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5226 struct rule_dpif *rule = rule_dpif_cast(rule_);
5228 /* push_all_stats() can handle flow misses which, when using the learn
5229 * action, can cause rules to be added and deleted. This can corrupt our
5230 * caller's datastructures which assume that rule_get_stats() doesn't have
5231 * an impact on the flow table. To be safe, we disable miss handling. */
5232 push_all_stats__(false);
5234 /* Start from historical data for 'rule' itself that are no longer tracked
5235 * in facets. This counts, for example, facets that have expired. */
5236 ovs_mutex_lock(&rule->stats_mutex);
5237 *packets = rule->packet_count;
5238 *bytes = rule->byte_count;
5239 ovs_mutex_unlock(&rule->stats_mutex);
5243 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5244 struct ofpbuf *packet)
5246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5247 struct dpif_flow_stats stats;
5248 struct xlate_out xout;
5249 struct xlate_in xin;
5251 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5252 rule_credit_stats(rule, &stats);
5254 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5255 xin.resubmit_stats = &stats;
5256 xlate_actions(&xin, &xout);
5258 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5259 xout.odp_actions.size, packet);
5261 xlate_out_uninit(&xout);
5265 rule_execute(struct rule *rule, const struct flow *flow,
5266 struct ofpbuf *packet)
5268 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5269 ofpbuf_delete(packet);
5274 rule_modify_actions(struct rule *rule_)
5276 struct rule_dpif *rule = rule_dpif_cast(rule_);
5278 complete_operation(rule);
5281 /* Sends 'packet' out 'ofport'.
5282 * May modify 'packet'.
5283 * Returns 0 if successful, otherwise a positive errno value. */
5285 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5287 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5288 uint64_t odp_actions_stub[1024 / 8];
5289 struct ofpbuf key, odp_actions;
5290 struct dpif_flow_stats stats;
5291 struct odputil_keybuf keybuf;
5292 struct ofpact_output output;
5293 struct xlate_out xout;
5294 struct xlate_in xin;
5296 union flow_in_port in_port_;
5299 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5300 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5302 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5303 in_port_.ofp_port = OFPP_NONE;
5304 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5305 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5307 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5309 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5310 output.port = ofport->up.ofp_port;
5313 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5314 xin.ofpacts_len = sizeof output;
5315 xin.ofpacts = &output.ofpact;
5316 xin.resubmit_stats = &stats;
5317 xlate_actions(&xin, &xout);
5319 error = dpif_execute(ofproto->backer->dpif,
5321 xout.odp_actions.data, xout.odp_actions.size,
5323 xlate_out_uninit(&xout);
5326 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5327 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5328 ovs_strerror(error));
5331 ofproto->stats.tx_packets++;
5332 ofproto->stats.tx_bytes += packet->size;
5336 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5337 * The action will state 'slow' as the reason that the action is in the slow
5338 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5339 * dump-flows" output to see why a flow is in the slow path.)
5341 * The 'stub_size' bytes in 'stub' will be used to store the action.
5342 * 'stub_size' must be large enough for the action.
5344 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5347 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5348 enum slow_path_reason slow,
5349 uint64_t *stub, size_t stub_size,
5350 const struct nlattr **actionsp, size_t *actions_lenp)
5352 union user_action_cookie cookie;
5355 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5356 cookie.slow_path.unused = 0;
5357 cookie.slow_path.reason = slow;
5359 ofpbuf_use_stack(&buf, stub, stub_size);
5360 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5361 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5363 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5365 odp_port_t odp_port;
5368 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5369 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5370 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5372 *actionsp = buf.data;
5373 *actions_lenp = buf.size;
5377 set_frag_handling(struct ofproto *ofproto_,
5378 enum ofp_config_flags frag_handling)
5380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5381 if (frag_handling != OFPC_FRAG_REASM) {
5382 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5390 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5391 const struct flow *flow,
5392 const struct ofpact *ofpacts, size_t ofpacts_len)
5394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5395 struct odputil_keybuf keybuf;
5396 struct dpif_flow_stats stats;
5397 struct xlate_out xout;
5398 struct xlate_in xin;
5402 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5403 odp_flow_key_from_flow(&key, flow,
5404 ofp_port_to_odp_port(ofproto,
5405 flow->in_port.ofp_port));
5407 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5409 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5410 xin.resubmit_stats = &stats;
5411 xin.ofpacts_len = ofpacts_len;
5412 xin.ofpacts = ofpacts;
5414 xlate_actions(&xin, &xout);
5415 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5416 xout.odp_actions.data, xout.odp_actions.size, packet);
5417 xlate_out_uninit(&xout);
5425 set_netflow(struct ofproto *ofproto_,
5426 const struct netflow_options *netflow_options)
5428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5430 if (netflow_options) {
5431 if (!ofproto->netflow) {
5432 ofproto->netflow = netflow_create();
5433 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5435 return netflow_set_options(ofproto->netflow, netflow_options);
5436 } else if (ofproto->netflow) {
5437 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5438 netflow_destroy(ofproto->netflow);
5439 ofproto->netflow = NULL;
5446 get_netflow_ids(const struct ofproto *ofproto_,
5447 uint8_t *engine_type, uint8_t *engine_id)
5449 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5451 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5455 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5457 if (!facet_is_controller_flow(facet) &&
5458 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5459 struct subfacet *subfacet;
5460 struct ofexpired expired;
5462 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5463 if (subfacet->path == SF_FAST_PATH) {
5464 struct dpif_flow_stats stats;
5466 subfacet_install(subfacet, &facet->xout.odp_actions,
5468 subfacet_update_stats(subfacet, &stats);
5472 expired.flow = facet->flow;
5473 expired.packet_count = facet->packet_count;
5474 expired.byte_count = facet->byte_count;
5475 expired.used = facet->used;
5476 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5481 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5483 struct cls_cursor cursor;
5484 struct facet *facet;
5486 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5487 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5488 send_active_timeout(ofproto, facet);
5492 static struct ofproto_dpif *
5493 ofproto_dpif_lookup(const char *name)
5495 struct ofproto_dpif *ofproto;
5497 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5498 hash_string(name, 0), &all_ofproto_dpifs) {
5499 if (!strcmp(ofproto->up.name, name)) {
5507 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5508 const char *argv[], void *aux OVS_UNUSED)
5510 struct ofproto_dpif *ofproto;
5513 ofproto = ofproto_dpif_lookup(argv[1]);
5515 unixctl_command_reply_error(conn, "no such bridge");
5518 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5519 mac_learning_flush(ofproto->ml);
5520 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5522 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5523 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5524 mac_learning_flush(ofproto->ml);
5525 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5529 unixctl_command_reply(conn, "table successfully flushed");
5532 static struct ofport_dpif *
5533 ofbundle_get_a_port(const struct ofbundle *bundle)
5535 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5540 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5541 const char *argv[], void *aux OVS_UNUSED)
5543 struct ds ds = DS_EMPTY_INITIALIZER;
5544 const struct ofproto_dpif *ofproto;
5545 const struct mac_entry *e;
5547 ofproto = ofproto_dpif_lookup(argv[1]);
5549 unixctl_command_reply_error(conn, "no such bridge");
5553 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5554 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5555 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5556 struct ofbundle *bundle = e->port.p;
5557 char name[OFP_MAX_PORT_NAME_LEN];
5559 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5561 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5562 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5563 mac_entry_age(ofproto->ml, e));
5565 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5566 unixctl_command_reply(conn, ds_cstr(&ds));
5571 struct xlate_out xout;
5572 struct xlate_in xin;
5578 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5580 ds_put_char_multiple(result, '\t', level);
5582 ds_put_cstr(result, "No match\n");
5586 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5587 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5588 cls_rule_format(&rule->up.cr, result);
5589 ds_put_char(result, '\n');
5591 ds_put_char_multiple(result, '\t', level);
5592 ds_put_cstr(result, "OpenFlow ");
5593 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5594 ds_put_char(result, '\n');
5598 trace_format_flow(struct ds *result, int level, const char *title,
5599 struct trace_ctx *trace)
5601 ds_put_char_multiple(result, '\t', level);
5602 ds_put_format(result, "%s: ", title);
5603 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5604 ds_put_cstr(result, "unchanged");
5606 flow_format(result, &trace->xin.flow);
5607 trace->flow = trace->xin.flow;
5609 ds_put_char(result, '\n');
5613 trace_format_regs(struct ds *result, int level, const char *title,
5614 struct trace_ctx *trace)
5618 ds_put_char_multiple(result, '\t', level);
5619 ds_put_format(result, "%s:", title);
5620 for (i = 0; i < FLOW_N_REGS; i++) {
5621 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5623 ds_put_char(result, '\n');
5627 trace_format_odp(struct ds *result, int level, const char *title,
5628 struct trace_ctx *trace)
5630 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5632 ds_put_char_multiple(result, '\t', level);
5633 ds_put_format(result, "%s: ", title);
5634 format_odp_actions(result, odp_actions->data, odp_actions->size);
5635 ds_put_char(result, '\n');
5639 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5641 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5642 struct ds *result = trace->result;
5644 ds_put_char(result, '\n');
5645 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5646 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5647 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5648 trace_format_rule(result, recurse + 1, rule);
5652 trace_report(struct xlate_in *xin, const char *s, int recurse)
5654 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5655 struct ds *result = trace->result;
5657 ds_put_char_multiple(result, '\t', recurse);
5658 ds_put_cstr(result, s);
5659 ds_put_char(result, '\n');
5663 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5664 void *aux OVS_UNUSED)
5666 const struct dpif_backer *backer;
5667 struct ofproto_dpif *ofproto;
5668 struct ofpbuf odp_key, odp_mask;
5669 struct ofpbuf *packet;
5677 ofpbuf_init(&odp_key, 0);
5678 ofpbuf_init(&odp_mask, 0);
5680 /* Handle "-generate" or a hex string as the last argument. */
5681 if (!strcmp(argv[argc - 1], "-generate")) {
5682 packet = ofpbuf_new(0);
5685 const char *error = eth_from_hex(argv[argc - 1], &packet);
5688 } else if (argc == 4) {
5689 /* The 3-argument form must end in "-generate' or a hex string. */
5690 unixctl_command_reply_error(conn, error);
5695 /* Parse the flow and determine whether a datapath or
5696 * bridge is specified. If function odp_flow_key_from_string()
5697 * returns 0, the flow is a odp_flow. If function
5698 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5699 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5700 /* If the odp_flow is the second argument,
5701 * the datapath name is the first argument. */
5703 const char *dp_type;
5704 if (!strncmp(argv[1], "ovs-", 4)) {
5705 dp_type = argv[1] + 4;
5709 backer = shash_find_data(&all_dpif_backers, dp_type);
5711 unixctl_command_reply_error(conn, "Cannot find datapath "
5716 /* No datapath name specified, so there should be only one
5718 struct shash_node *node;
5719 if (shash_count(&all_dpif_backers) != 1) {
5720 unixctl_command_reply_error(conn, "Must specify datapath "
5721 "name, there is more than one type of datapath");
5724 node = shash_first(&all_dpif_backers);
5725 backer = node->data;
5728 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5729 NULL, &ofproto, NULL)) {
5730 unixctl_command_reply_error(conn, "Invalid datapath flow");
5733 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5734 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5736 unixctl_command_reply_error(conn, "Must specify bridge name");
5740 ofproto = ofproto_dpif_lookup(argv[1]);
5742 unixctl_command_reply_error(conn, "Unknown bridge name");
5746 unixctl_command_reply_error(conn, "Bad flow syntax");
5750 /* Generate a packet, if requested. */
5752 if (!packet->size) {
5753 flow_compose(packet, &flow);
5755 union flow_in_port in_port_;
5757 in_port_ = flow.in_port;
5758 ds_put_cstr(&result, "Packet: ");
5759 s = ofp_packet_to_string(packet->data, packet->size);
5760 ds_put_cstr(&result, s);
5763 /* Use the metadata from the flow and the packet argument
5764 * to reconstruct the flow. */
5765 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5770 ofproto_trace(ofproto, &flow, packet, &result);
5771 unixctl_command_reply(conn, ds_cstr(&result));
5774 ds_destroy(&result);
5775 ofpbuf_delete(packet);
5776 ofpbuf_uninit(&odp_key);
5777 ofpbuf_uninit(&odp_mask);
5781 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5782 const struct ofpbuf *packet, struct ds *ds)
5784 struct rule_dpif *rule;
5786 ds_put_cstr(ds, "Flow: ");
5787 flow_format(ds, flow);
5788 ds_put_char(ds, '\n');
5790 rule = rule_dpif_lookup(ofproto, flow, NULL);
5792 trace_format_rule(ds, 0, rule);
5793 if (rule == ofproto->miss_rule) {
5794 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5795 } else if (rule == ofproto->no_packet_in_rule) {
5796 ds_put_cstr(ds, "\nNo match, packets dropped because "
5797 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5798 } else if (rule == ofproto->drop_frags_rule) {
5799 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5800 "and the fragment handling mode is \"drop\".\n");
5804 uint64_t odp_actions_stub[1024 / 8];
5805 struct ofpbuf odp_actions;
5806 struct trace_ctx trace;
5810 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5813 ofpbuf_use_stub(&odp_actions,
5814 odp_actions_stub, sizeof odp_actions_stub);
5815 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5816 trace.xin.resubmit_hook = trace_resubmit;
5817 trace.xin.report_hook = trace_report;
5819 xlate_actions(&trace.xin, &trace.xout);
5821 ds_put_char(ds, '\n');
5822 trace_format_flow(ds, 0, "Final flow", &trace);
5824 match_init(&match, flow, &trace.xout.wc);
5825 ds_put_cstr(ds, "Relevant fields: ");
5826 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5827 ds_put_char(ds, '\n');
5829 ds_put_cstr(ds, "Datapath actions: ");
5830 format_odp_actions(ds, trace.xout.odp_actions.data,
5831 trace.xout.odp_actions.size);
5833 if (trace.xout.slow) {
5834 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5835 "slow path because it:");
5836 switch (trace.xout.slow) {
5838 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5841 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5844 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5847 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5849 case SLOW_CONTROLLER:
5850 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5851 "to the OpenFlow controller.");
5858 xlate_out_uninit(&trace.xout);
5863 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5864 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5867 unixctl_command_reply(conn, NULL);
5871 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5872 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5875 unixctl_command_reply(conn, NULL);
5878 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5879 * 'reply' describing the results. */
5881 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5883 struct cls_cursor cursor;
5884 struct facet *facet;
5888 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5889 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5890 if (!facet_check_consistency(facet)) {
5895 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5899 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5900 ofproto->up.name, errors);
5902 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5907 ofproto_dpif_self_check(struct unixctl_conn *conn,
5908 int argc, const char *argv[], void *aux OVS_UNUSED)
5910 struct ds reply = DS_EMPTY_INITIALIZER;
5911 struct ofproto_dpif *ofproto;
5914 ofproto = ofproto_dpif_lookup(argv[1]);
5916 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5917 "ofproto/list for help)");
5920 ofproto_dpif_self_check__(ofproto, &reply);
5922 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5923 ofproto_dpif_self_check__(ofproto, &reply);
5927 unixctl_command_reply(conn, ds_cstr(&reply));
5931 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5932 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5933 * to destroy 'ofproto_shash' and free the returned value. */
5934 static const struct shash_node **
5935 get_ofprotos(struct shash *ofproto_shash)
5937 const struct ofproto_dpif *ofproto;
5939 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5940 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5941 shash_add_nocopy(ofproto_shash, name, ofproto);
5944 return shash_sort(ofproto_shash);
5948 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5949 const char *argv[] OVS_UNUSED,
5950 void *aux OVS_UNUSED)
5952 struct ds ds = DS_EMPTY_INITIALIZER;
5953 struct shash ofproto_shash;
5954 const struct shash_node **sorted_ofprotos;
5957 shash_init(&ofproto_shash);
5958 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5959 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5960 const struct shash_node *node = sorted_ofprotos[i];
5961 ds_put_format(&ds, "%s\n", node->name);
5964 shash_destroy(&ofproto_shash);
5965 free(sorted_ofprotos);
5967 unixctl_command_reply(conn, ds_cstr(&ds));
5972 show_dp_rates(struct ds *ds, const char *heading,
5973 const struct avg_subfacet_rates *rates)
5975 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5976 heading, rates->add_rate, rates->del_rate);
5980 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5982 const struct shash_node **ofprotos;
5983 struct ofproto_dpif *ofproto;
5984 struct shash ofproto_shash;
5985 uint64_t n_hit, n_missed;
5986 long long int minutes;
5989 n_hit = n_missed = 0;
5990 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5991 if (ofproto->backer == backer) {
5992 n_missed += ofproto->n_missed;
5993 n_hit += ofproto->n_hit;
5997 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5998 dpif_name(backer->dpif), n_hit, n_missed);
5999 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6000 " life span: %lldms\n", hmap_count(&backer->subfacets),
6001 backer->avg_n_subfacet, backer->max_n_subfacet,
6002 backer->avg_subfacet_life);
6004 minutes = (time_msec() - backer->created) / (1000 * 60);
6005 if (minutes >= 60) {
6006 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6008 if (minutes >= 60 * 24) {
6009 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6011 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6013 shash_init(&ofproto_shash);
6014 ofprotos = get_ofprotos(&ofproto_shash);
6015 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6016 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6017 const struct shash_node **ports;
6020 if (ofproto->backer != backer) {
6024 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6025 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6027 ports = shash_sort(&ofproto->up.port_by_name);
6028 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6029 const struct shash_node *node = ports[j];
6030 struct ofport *ofport = node->data;
6032 odp_port_t odp_port;
6034 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6037 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6038 if (odp_port != ODPP_NONE) {
6039 ds_put_format(ds, "%"PRIu32":", odp_port);
6041 ds_put_cstr(ds, "none:");
6044 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6047 if (!netdev_get_config(ofport->netdev, &config)) {
6048 const struct smap_node **nodes;
6051 nodes = smap_sort(&config);
6052 for (i = 0; i < smap_count(&config); i++) {
6053 const struct smap_node *node = nodes[i];
6054 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6055 node->key, node->value);
6059 smap_destroy(&config);
6061 ds_put_char(ds, ')');
6062 ds_put_char(ds, '\n');
6066 shash_destroy(&ofproto_shash);
6071 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6072 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6074 struct ds ds = DS_EMPTY_INITIALIZER;
6075 const struct shash_node **backers;
6078 backers = shash_sort(&all_dpif_backers);
6079 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6080 dpif_show_backer(backers[i]->data, &ds);
6084 unixctl_command_reply(conn, ds_cstr(&ds));
6088 /* Dump the megaflow (facet) cache. This is useful to check the
6089 * correctness of flow wildcarding, since the same mechanism is used for
6090 * both xlate caching and kernel wildcarding.
6092 * It's important to note that in the output the flow description uses
6093 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6095 * This command is only needed for advanced debugging, so it's not
6096 * documented in the man page. */
6098 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6099 int argc OVS_UNUSED, const char *argv[],
6100 void *aux OVS_UNUSED)
6102 struct ds ds = DS_EMPTY_INITIALIZER;
6103 const struct ofproto_dpif *ofproto;
6104 long long int now = time_msec();
6105 struct cls_cursor cursor;
6106 struct facet *facet;
6108 ofproto = ofproto_dpif_lookup(argv[1]);
6110 unixctl_command_reply_error(conn, "no such bridge");
6114 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6115 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6116 cls_rule_format(&facet->cr, &ds);
6117 ds_put_cstr(&ds, ", ");
6118 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6119 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6120 ds_put_cstr(&ds, "Datapath actions: ");
6121 if (facet->xout.slow) {
6122 uint64_t slow_path_stub[128 / 8];
6123 const struct nlattr *actions;
6126 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6127 slow_path_stub, sizeof slow_path_stub,
6128 &actions, &actions_len);
6129 format_odp_actions(&ds, actions, actions_len);
6131 format_odp_actions(&ds, facet->xout.odp_actions.data,
6132 facet->xout.odp_actions.size);
6134 ds_put_cstr(&ds, "\n");
6137 ds_chomp(&ds, '\n');
6138 unixctl_command_reply(conn, ds_cstr(&ds));
6142 /* Disable using the megaflows.
6144 * This command is only needed for advanced debugging, so it's not
6145 * documented in the man page. */
6147 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6148 int argc OVS_UNUSED,
6149 const char *argv[] OVS_UNUSED,
6150 void *aux OVS_UNUSED)
6152 struct ofproto_dpif *ofproto;
6154 enable_megaflows = false;
6156 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6157 flush(&ofproto->up);
6160 unixctl_command_reply(conn, "megaflows disabled");
6163 /* Re-enable using megaflows.
6165 * This command is only needed for advanced debugging, so it's not
6166 * documented in the man page. */
6168 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6169 int argc OVS_UNUSED,
6170 const char *argv[] OVS_UNUSED,
6171 void *aux OVS_UNUSED)
6173 struct ofproto_dpif *ofproto;
6175 enable_megaflows = true;
6177 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6178 flush(&ofproto->up);
6181 unixctl_command_reply(conn, "megaflows enabled");
6185 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6186 int argc OVS_UNUSED, const char *argv[],
6187 void *aux OVS_UNUSED)
6189 struct ds ds = DS_EMPTY_INITIALIZER;
6190 const struct ofproto_dpif *ofproto;
6191 struct subfacet *subfacet;
6193 ofproto = ofproto_dpif_lookup(argv[1]);
6195 unixctl_command_reply_error(conn, "no such bridge");
6199 update_stats(ofproto->backer);
6201 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6202 struct facet *facet = subfacet->facet;
6203 struct odputil_keybuf maskbuf;
6206 if (facet->ofproto != ofproto) {
6210 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6211 if (enable_megaflows) {
6212 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6213 &facet->flow, UINT32_MAX);
6216 odp_flow_format(subfacet->key, subfacet->key_len,
6217 mask.data, mask.size, &ds, false);
6219 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6220 subfacet->dp_packet_count, subfacet->dp_byte_count);
6221 if (subfacet->used) {
6222 ds_put_format(&ds, "%.3fs",
6223 (time_msec() - subfacet->used) / 1000.0);
6225 ds_put_format(&ds, "never");
6227 if (subfacet->facet->tcp_flags) {
6228 ds_put_cstr(&ds, ", flags:");
6229 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6232 ds_put_cstr(&ds, ", actions:");
6233 if (facet->xout.slow) {
6234 uint64_t slow_path_stub[128 / 8];
6235 const struct nlattr *actions;
6238 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6239 slow_path_stub, sizeof slow_path_stub,
6240 &actions, &actions_len);
6241 format_odp_actions(&ds, actions, actions_len);
6243 format_odp_actions(&ds, facet->xout.odp_actions.data,
6244 facet->xout.odp_actions.size);
6246 ds_put_char(&ds, '\n');
6249 unixctl_command_reply(conn, ds_cstr(&ds));
6254 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6255 int argc OVS_UNUSED, const char *argv[],
6256 void *aux OVS_UNUSED)
6258 struct ds ds = DS_EMPTY_INITIALIZER;
6259 struct ofproto_dpif *ofproto;
6261 ofproto = ofproto_dpif_lookup(argv[1]);
6263 unixctl_command_reply_error(conn, "no such bridge");
6267 flush(&ofproto->up);
6269 unixctl_command_reply(conn, ds_cstr(&ds));
6274 ofproto_dpif_unixctl_init(void)
6276 static bool registered;
6282 unixctl_command_register(
6284 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6285 1, 3, ofproto_unixctl_trace, NULL);
6286 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6287 ofproto_unixctl_fdb_flush, NULL);
6288 unixctl_command_register("fdb/show", "bridge", 1, 1,
6289 ofproto_unixctl_fdb_show, NULL);
6290 unixctl_command_register("ofproto/clog", "", 0, 0,
6291 ofproto_dpif_clog, NULL);
6292 unixctl_command_register("ofproto/unclog", "", 0, 0,
6293 ofproto_dpif_unclog, NULL);
6294 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6295 ofproto_dpif_self_check, NULL);
6296 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6297 ofproto_unixctl_dpif_dump_dps, NULL);
6298 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6300 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6301 ofproto_unixctl_dpif_dump_flows, NULL);
6302 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6303 ofproto_unixctl_dpif_del_flows, NULL);
6304 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6305 ofproto_unixctl_dpif_dump_megaflows, NULL);
6306 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6307 ofproto_unixctl_dpif_disable_megaflows, NULL);
6308 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6309 ofproto_unixctl_dpif_enable_megaflows, NULL);
6312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6314 * This is deprecated. It is only for compatibility with broken device drivers
6315 * in old versions of Linux that do not properly support VLANs when VLAN
6316 * devices are not used. When broken device drivers are no longer in
6317 * widespread use, we will delete these interfaces. */
6320 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6323 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6325 if (realdev_ofp_port == ofport->realdev_ofp_port
6326 && vid == ofport->vlandev_vid) {
6330 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6332 if (ofport->realdev_ofp_port) {
6335 if (realdev_ofp_port && ofport->bundle) {
6336 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6337 * themselves be part of a bundle. */
6338 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6341 ofport->realdev_ofp_port = realdev_ofp_port;
6342 ofport->vlandev_vid = vid;
6344 if (realdev_ofp_port) {
6345 vsp_add(ofport, realdev_ofp_port, vid);
6352 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6354 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6358 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6359 OVS_EXCLUDED(ofproto->vsp_mutex)
6363 ovs_mutex_lock(&ofproto->vsp_mutex);
6364 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6365 ovs_mutex_unlock(&ofproto->vsp_mutex);
6370 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6371 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6372 OVS_REQUIRES(ofproto->vsp_mutex)
6374 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6375 int vid = vlan_tci_to_vid(vlan_tci);
6376 const struct vlan_splinter *vsp;
6378 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6379 hash_realdev_vid(realdev_ofp_port, vid),
6380 &ofproto->realdev_vid_map) {
6381 if (vsp->realdev_ofp_port == realdev_ofp_port
6382 && vsp->vid == vid) {
6383 return vsp->vlandev_ofp_port;
6387 return realdev_ofp_port;
6390 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6391 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6392 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6393 * 'vlan_tci' 9, it would return the port number of eth0.9.
6395 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6396 * function just returns its 'realdev_ofp_port' argument. */
6398 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6399 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6400 OVS_EXCLUDED(ofproto->vsp_mutex)
6404 ovs_mutex_lock(&ofproto->vsp_mutex);
6405 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6406 ovs_mutex_unlock(&ofproto->vsp_mutex);
6410 static struct vlan_splinter *
6411 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6413 struct vlan_splinter *vsp;
6415 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6416 hash_ofp_port(vlandev_ofp_port),
6417 &ofproto->vlandev_map) {
6418 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6426 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6427 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6428 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6429 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6430 * eth0 and store 9 in '*vid'.
6432 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6433 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6436 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6437 ofp_port_t vlandev_ofp_port, int *vid)
6438 OVS_REQ_WRLOCK(ofproto->vsp_mutex)
6440 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6441 const struct vlan_splinter *vsp;
6443 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6448 return vsp->realdev_ofp_port;
6454 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6455 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6456 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6457 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6458 * always the case unless VLAN splinters are enabled), returns false without
6459 * making any changes. */
6461 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6462 OVS_EXCLUDED(ofproto->vsp_mutex)
6467 ovs_mutex_lock(&ofproto->vsp_mutex);
6468 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6469 ovs_mutex_unlock(&ofproto->vsp_mutex);
6474 /* Cause the flow to be processed as if it came in on the real device with
6475 * the VLAN device's VLAN ID. */
6476 flow->in_port.ofp_port = realdev;
6477 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6482 vsp_remove(struct ofport_dpif *port)
6484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6485 struct vlan_splinter *vsp;
6487 ovs_mutex_lock(&ofproto->vsp_mutex);
6488 vsp = vlandev_find(ofproto, port->up.ofp_port);
6490 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6491 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6494 port->realdev_ofp_port = 0;
6496 VLOG_ERR("missing vlan device record");
6498 ovs_mutex_unlock(&ofproto->vsp_mutex);
6502 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6506 ovs_mutex_lock(&ofproto->vsp_mutex);
6507 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6508 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6509 == realdev_ofp_port)) {
6510 struct vlan_splinter *vsp;
6512 vsp = xmalloc(sizeof *vsp);
6513 vsp->realdev_ofp_port = realdev_ofp_port;
6514 vsp->vlandev_ofp_port = port->up.ofp_port;
6517 port->realdev_ofp_port = realdev_ofp_port;
6519 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6520 hash_ofp_port(port->up.ofp_port));
6521 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6522 hash_realdev_vid(realdev_ofp_port, vid));
6524 VLOG_ERR("duplicate vlan device record");
6526 ovs_mutex_unlock(&ofproto->vsp_mutex);
6530 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6532 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6533 return ofport ? ofport->odp_port : ODPP_NONE;
6536 struct ofport_dpif *
6537 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6539 struct ofport_dpif *port;
6541 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6542 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6543 &backer->odp_to_ofport_map) {
6544 if (port->odp_port == odp_port) {
6545 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6550 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6555 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6557 struct ofport_dpif *port;
6559 port = odp_port_to_ofport(ofproto->backer, odp_port);
6560 if (port && &ofproto->up == port->up.ofproto) {
6561 return port->up.ofp_port;
6567 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6568 * most heavily weighted element. 'base' designates the rate of decay: after
6569 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6572 exp_mavg(double *avg, int base, double new)
6574 *avg = (*avg * (base - 1) + new) / base;
6578 update_moving_averages(struct dpif_backer *backer)
6580 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6581 long long int minutes = (time_msec() - backer->created) / min_ms;
6584 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6586 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6589 backer->lifetime.add_rate = 0.0;
6590 backer->lifetime.del_rate = 0.0;
6593 /* Update hourly averages on the minute boundaries. */
6594 if (time_msec() - backer->last_minute >= min_ms) {
6595 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6596 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6598 /* Update daily averages on the hour boundaries. */
6599 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6600 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6601 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6604 backer->total_subfacet_add_count += backer->subfacet_add_count;
6605 backer->total_subfacet_del_count += backer->subfacet_del_count;
6606 backer->subfacet_add_count = 0;
6607 backer->subfacet_del_count = 0;
6608 backer->last_minute += min_ms;
6612 const struct ofproto_class ofproto_dpif_class = {
6647 port_is_lacp_current,
6648 NULL, /* rule_choose_table */
6655 rule_modify_actions,
6669 get_stp_port_status,
6676 is_mirror_output_bundle,
6677 forward_bpdu_changed,
6678 set_mac_table_config,
6680 NULL, /* meter_get_features */
6681 NULL, /* meter_set */
6682 NULL, /* meter_get */
6683 NULL, /* meter_del */