2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
34 #include "guarded-list.h"
38 #include "mac-learning.h"
39 #include "meta-flow.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
46 #include "odp-execute.h"
49 #include "ofp-actions.h"
50 #include "ofp-parse.h"
51 #include "ofp-print.h"
52 #include "ofproto-dpif-governor.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-sflow.h"
56 #include "ofproto-dpif-upcall.h"
57 #include "ofproto-dpif-xlate.h"
58 #include "poll-loop.h"
63 #include "unaligned.h"
65 #include "vlan-bitmap.h"
68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
70 COVERAGE_DEFINE(ofproto_dpif_expired);
71 COVERAGE_DEFINE(facet_changed_rule);
72 COVERAGE_DEFINE(facet_revalidate);
73 COVERAGE_DEFINE(facet_unexpected);
74 COVERAGE_DEFINE(facet_suppress);
75 COVERAGE_DEFINE(subfacet_install_fail);
76 COVERAGE_DEFINE(packet_in_overflow);
77 COVERAGE_DEFINE(flow_mod_overflow);
79 /* Number of implemented OpenFlow tables. */
80 enum { N_TABLES = 255 };
81 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
82 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
92 * - Do include packets and bytes from facets that have been deleted or
93 * whose own statistics have been folded into the rule.
95 * - Do include packets and bytes sent "by hand" that were accounted to
96 * the rule without any facet being involved (this is a rare corner
97 * case in rule_execute()).
99 * - Do not include packet or bytes that can be obtained from any facet's
100 * packet_count or byte_count member or that can be obtained from the
101 * datapath by, e.g., dpif_flow_get() for any subfacet.
103 struct ovs_mutex stats_mutex;
104 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
105 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
108 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
109 static struct rule_dpif *rule_dpif_cast(const struct rule *);
112 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
113 struct ofproto_dpif *ofproto; /* Owning ofproto. */
114 void *aux; /* Key supplied by ofproto's client. */
115 char *name; /* Identifier for log messages. */
118 struct list ports; /* Contains "struct ofport"s. */
119 enum port_vlan_mode vlan_mode; /* VLAN mode */
120 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
121 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
122 * NULL if all VLANs are trunked. */
123 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
124 struct bond *bond; /* Nonnull iff more than one port. */
125 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
128 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
131 static void bundle_remove(struct ofport *);
132 static void bundle_update(struct ofbundle *);
133 static void bundle_destroy(struct ofbundle *);
134 static void bundle_del_port(struct ofport_dpif *);
135 static void bundle_run(struct ofbundle *);
136 static void bundle_wait(struct ofbundle *);
138 static void stp_run(struct ofproto_dpif *ofproto);
139 static void stp_wait(struct ofproto_dpif *ofproto);
140 static int set_stp_port(struct ofport *,
141 const struct ofproto_port_stp_settings *);
143 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
144 enum slow_path_reason,
145 uint64_t *stub, size_t stub_size,
146 const struct nlattr **actionsp,
147 size_t *actions_lenp);
149 /* A subfacet (see "struct subfacet" below) has three possible installation
152 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
153 * case just after the subfacet is created, just before the subfacet is
154 * destroyed, or if the datapath returns an error when we try to install a
157 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
159 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
160 * ofproto_dpif is installed in the datapath.
163 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
164 SF_FAST_PATH, /* Full actions are installed. */
165 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
168 /* A dpif flow and actions associated with a facet.
170 * See also the large comment on struct facet. */
173 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
174 struct list list_node; /* In struct facet's 'facets' list. */
175 struct facet *facet; /* Owning facet. */
176 struct dpif_backer *backer; /* Owning backer. */
178 enum odp_key_fitness key_fitness;
182 long long int used; /* Time last used; time created if not used. */
183 long long int created; /* Time created. */
185 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
186 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
188 enum subfacet_path path; /* Installed in datapath? */
191 #define SUBFACET_DESTROY_MAX_BATCH 50
193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
194 static struct subfacet *subfacet_find(struct dpif_backer *,
195 const struct nlattr *key, size_t key_len,
197 static void subfacet_destroy(struct subfacet *);
198 static void subfacet_destroy__(struct subfacet *);
199 static void subfacet_destroy_batch(struct dpif_backer *,
200 struct subfacet **, int n);
201 static void subfacet_reset_dp_stats(struct subfacet *,
202 struct dpif_flow_stats *);
203 static void subfacet_update_stats(struct subfacet *,
204 const struct dpif_flow_stats *);
205 static int subfacet_install(struct subfacet *,
206 const struct ofpbuf *odp_actions,
207 struct dpif_flow_stats *);
208 static void subfacet_uninstall(struct subfacet *);
210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
212 * A facet associates a "struct flow", which represents the Open vSwitch
213 * userspace idea of an exact-match flow, with one or more subfacets.
214 * While the facet is created based on an exact-match flow, it is stored
215 * within the ofproto based on the wildcards that could be expressed
216 * based on the flow table and other configuration. (See the 'wc'
217 * description in "struct xlate_out" for more details.)
219 * Each subfacet tracks the datapath's idea of the flow equivalent to
220 * the facet. When the kernel module (or other dpif implementation) and
221 * Open vSwitch userspace agree on the definition of a flow key, there
222 * is exactly one subfacet per facet. If the dpif implementation
223 * supports more-specific flow matching than userspace, however, a facet
224 * can have more than one subfacet. Examples include the dpif
225 * implementation not supporting the same wildcards as userspace or some
226 * distinction in flow that userspace simply doesn't understand.
228 * Flow expiration works in terms of subfacets, so a facet must have at
229 * least one subfacet or it will never expire, leaking memory. */
232 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
233 struct ofproto_dpif *ofproto;
236 struct list subfacets;
237 long long int used; /* Time last used; time created if not used. */
240 struct flow flow; /* Flow of the creating subfacet. */
241 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
245 * - Do include packets and bytes sent "by hand", e.g. with
248 * - Do include packets and bytes that were obtained from the datapath
249 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
250 * DPIF_FP_ZERO_STATS).
252 * - Do not include packets or bytes that can be obtained from the
253 * datapath for any existing subfacet.
255 uint64_t packet_count; /* Number of packets received. */
256 uint64_t byte_count; /* Number of bytes received. */
258 /* Resubmit statistics. */
259 uint64_t prev_packet_count; /* Number of packets from last stats push. */
260 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
261 long long int prev_used; /* Used time from last stats push. */
264 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
265 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
266 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
268 struct xlate_out xout;
270 /* Storage for a single subfacet, to reduce malloc() time and space
271 * overhead. (A facet always has at least one subfacet and in the common
272 * case has exactly one subfacet. However, 'one_subfacet' may not
273 * always be valid, since it could have been removed after newer
274 * subfacets were pushed onto the 'subfacets' list.) */
275 struct subfacet one_subfacet;
277 long long int learn_rl; /* Rate limiter for facet_learn(). */
280 static struct facet *facet_create(const struct flow_miss *);
281 static void facet_remove(struct facet *);
282 static void facet_free(struct facet *);
284 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
285 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
286 const struct flow *);
287 static bool facet_revalidate(struct facet *);
288 static bool facet_check_consistency(struct facet *);
290 static void facet_flush_stats(struct facet *);
292 static void facet_reset_counters(struct facet *);
293 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
294 struct dpif_flow_stats *, bool may_learn);
295 static void facet_push_stats(struct facet *, bool may_learn);
296 static void facet_learn(struct facet *);
297 static void facet_account(struct facet *);
298 static void push_all_stats(void);
300 static bool facet_is_controller_flow(struct facet *);
303 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
307 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
308 struct list bundle_node; /* In struct ofbundle's "ports" list. */
309 struct cfm *cfm; /* Connectivity Fault Management, if any. */
310 struct bfd *bfd; /* BFD, if any. */
311 bool may_enable; /* May be enabled in bonds. */
312 bool is_tunnel; /* This port is a tunnel. */
313 long long int carrier_seq; /* Carrier status changes. */
314 struct ofport_dpif *peer; /* Peer if patch port. */
317 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
318 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
319 long long int stp_state_entered;
321 /* Queue to DSCP mapping. */
322 struct ofproto_port_queue *qdscp;
325 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
327 * This is deprecated. It is only for compatibility with broken device
328 * drivers in old versions of Linux that do not properly support VLANs when
329 * VLAN devices are not used. When broken device drivers are no longer in
330 * widespread use, we will delete these interfaces. */
331 ofp_port_t realdev_ofp_port;
335 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
337 * This is deprecated. It is only for compatibility with broken device drivers
338 * in old versions of Linux that do not properly support VLANs when VLAN
339 * devices are not used. When broken device drivers are no longer in
340 * widespread use, we will delete these interfaces. */
341 struct vlan_splinter {
342 struct hmap_node realdev_vid_node;
343 struct hmap_node vlandev_node;
344 ofp_port_t realdev_ofp_port;
345 ofp_port_t vlandev_ofp_port;
349 static void vsp_remove(struct ofport_dpif *);
350 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
352 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
355 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
358 static struct ofport_dpif *
359 ofport_dpif_cast(const struct ofport *ofport)
361 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
364 static void port_run(struct ofport_dpif *);
365 static void port_run_fast(struct ofport_dpif *);
366 static void port_wait(struct ofport_dpif *);
367 static int set_bfd(struct ofport *, const struct smap *);
368 static int set_cfm(struct ofport *, const struct cfm_settings *);
369 static void ofport_update_peer(struct ofport_dpif *);
370 static void run_fast_rl(void);
371 static int run_fast(struct ofproto *);
373 struct dpif_completion {
374 struct list list_node;
375 struct ofoperation *op;
378 /* Reasons that we might need to revalidate every facet, and corresponding
381 * A value of 0 means that there is no need to revalidate.
383 * It would be nice to have some cleaner way to integrate with coverage
384 * counters, but with only a few reasons I guess this is good enough for
386 enum revalidate_reason {
387 REV_RECONFIGURE = 1, /* Switch configuration changed. */
388 REV_STP, /* Spanning tree protocol port status change. */
389 REV_BOND, /* Bonding changed. */
390 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
391 REV_FLOW_TABLE, /* Flow table changed. */
392 REV_MAC_LEARNING, /* Mac learning changed. */
393 REV_INCONSISTENCY /* Facet self-check failed. */
395 COVERAGE_DEFINE(rev_reconfigure);
396 COVERAGE_DEFINE(rev_stp);
397 COVERAGE_DEFINE(rev_bond);
398 COVERAGE_DEFINE(rev_port_toggled);
399 COVERAGE_DEFINE(rev_flow_table);
400 COVERAGE_DEFINE(rev_mac_learning);
401 COVERAGE_DEFINE(rev_inconsistency);
403 struct avg_subfacet_rates {
404 double add_rate; /* Moving average of new flows created per minute. */
405 double del_rate; /* Moving average of flows deleted per minute. */
408 /* All datapaths of a given type share a single dpif backer instance. */
414 struct timer next_expiration;
416 struct ovs_rwlock odp_to_ofport_lock;
417 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
419 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
421 /* Facet revalidation flags applying to facets which use this backer. */
422 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
424 struct hmap drop_keys; /* Set of dropped odp keys. */
425 bool recv_set_enable; /* Enables or disables receiving packets. */
427 struct hmap subfacets;
428 struct governor *governor;
430 /* Subfacet statistics.
432 * These keep track of the total number of subfacets added and deleted and
433 * flow life span. They are useful for computing the flow rates stats
434 * exposed via "ovs-appctl dpif/show". The goal is to learn about
435 * traffic patterns in ways that we can use later to improve Open vSwitch
436 * performance in new situations. */
437 long long int created; /* Time when it is created. */
438 unsigned max_n_subfacet; /* Maximum number of flows */
439 unsigned avg_n_subfacet; /* Average number of flows. */
440 long long int avg_subfacet_life; /* Average life span of subfacets. */
442 /* The average number of subfacets... */
443 struct avg_subfacet_rates hourly; /* ...over the last hour. */
444 struct avg_subfacet_rates daily; /* ...over the last day. */
445 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
446 long long int last_minute; /* Last time 'hourly' was updated. */
448 /* Number of subfacets added or deleted since 'last_minute'. */
449 unsigned subfacet_add_count;
450 unsigned subfacet_del_count;
452 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
453 unsigned long long int total_subfacet_add_count;
454 unsigned long long int total_subfacet_del_count;
456 /* Number of upcall handling threads. */
457 unsigned int n_handler_threads;
460 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
461 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
463 static void drop_key_clear(struct dpif_backer *);
464 static void update_moving_averages(struct dpif_backer *backer);
466 struct ofproto_dpif {
467 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
469 struct dpif_backer *backer;
471 /* Special OpenFlow rules. */
472 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
473 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
474 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
477 struct netflow *netflow;
478 struct dpif_sflow *sflow;
479 struct dpif_ipfix *ipfix;
480 struct hmap bundles; /* Contains "struct ofbundle"s. */
481 struct mac_learning *ml;
482 bool has_bonded_bundles;
483 struct mbridge *mbridge;
486 struct classifier facets; /* Contains 'struct facet's. */
487 long long int consistency_rl;
489 struct netdev_stats stats; /* To account packets generated and consumed in
494 long long int stp_last_tick;
496 /* VLAN splinters. */
497 struct ovs_mutex vsp_mutex;
498 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
499 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
502 struct sset ports; /* Set of standard port names. */
503 struct sset ghost_ports; /* Ports with no datapath port. */
504 struct sset port_poll_set; /* Queued names for port_poll() reply. */
505 int port_poll_errno; /* Last errno for port_poll() reply. */
507 /* Per ofproto's dpif stats. */
512 struct guarded_list flow_mods; /* Contains "struct flow_mod"s. */
513 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
516 /* By default, flows in the datapath are wildcarded (megaflows). They
517 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
518 static bool enable_megaflows = true;
520 /* All existing ofproto_dpif instances, indexed by ->up.name. */
521 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
523 static void ofproto_dpif_unixctl_init(void);
525 static inline struct ofproto_dpif *
526 ofproto_dpif_cast(const struct ofproto *ofproto)
528 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
529 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
532 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
533 ofp_port_t ofp_port);
534 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
535 const struct ofpbuf *packet, struct ds *);
538 static void handle_upcalls(struct dpif_backer *);
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 if (!guarded_list_push_back(&ofproto->flow_mods, &fm->list_node, 1024)) {
561 COVERAGE_INC(flow_mod_overflow);
567 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
568 * Takes ownership of 'pin' and pin->packet. */
570 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
571 struct ofputil_packet_in *pin)
573 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
574 COVERAGE_INC(packet_in_overflow);
575 free(CONST_CAST(void *, pin->packet));
580 /* Factory functions. */
583 init(const struct shash *iface_hints)
585 struct shash_node *node;
587 /* Make a local copy, since we don't own 'iface_hints' elements. */
588 SHASH_FOR_EACH(node, iface_hints) {
589 const struct iface_hint *orig_hint = node->data;
590 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
592 new_hint->br_name = xstrdup(orig_hint->br_name);
593 new_hint->br_type = xstrdup(orig_hint->br_type);
594 new_hint->ofp_port = orig_hint->ofp_port;
596 shash_add(&init_ofp_ports, node->name, new_hint);
601 enumerate_types(struct sset *types)
603 dp_enumerate_types(types);
607 enumerate_names(const char *type, struct sset *names)
609 struct ofproto_dpif *ofproto;
612 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
613 if (strcmp(type, ofproto->up.type)) {
616 sset_add(names, ofproto->up.name);
623 del(const char *type, const char *name)
628 error = dpif_open(name, type, &dpif);
630 error = dpif_delete(dpif);
637 port_open_type(const char *datapath_type, const char *port_type)
639 return dpif_port_open_type(datapath_type, port_type);
642 /* Type functions. */
644 static void process_dpif_port_changes(struct dpif_backer *);
645 static void process_dpif_all_ports_changed(struct dpif_backer *);
646 static void process_dpif_port_change(struct dpif_backer *,
647 const char *devname);
648 static void process_dpif_port_error(struct dpif_backer *, int error);
650 static struct ofproto_dpif *
651 lookup_ofproto_dpif_by_port_name(const char *name)
653 struct ofproto_dpif *ofproto;
655 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
656 if (sset_contains(&ofproto->ports, name)) {
665 type_run(const char *type)
667 static long long int push_timer = LLONG_MIN;
668 struct dpif_backer *backer;
670 backer = shash_find_data(&all_dpif_backers, type);
672 /* This is not necessarily a problem, since backers are only
673 * created on demand. */
677 dpif_run(backer->dpif);
679 /* The most natural place to push facet statistics is when they're pulled
680 * from the datapath. However, when there are many flows in the datapath,
681 * this expensive operation can occur so frequently, that it reduces our
682 * ability to quickly set up flows. To reduce the cost, we push statistics
684 if (time_msec() > push_timer) {
685 push_timer = time_msec() + 2000;
689 /* If vswitchd started with other_config:flow_restore_wait set as "true",
690 * and the configuration has now changed to "false", enable receiving
691 * packets from the datapath. */
692 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
695 backer->recv_set_enable = true;
697 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
699 udpif_recv_set(backer->udpif, 0, false);
700 VLOG_ERR("Failed to enable receiving packets in dpif.");
703 udpif_recv_set(backer->udpif, n_handler_threads,
704 backer->recv_set_enable);
705 dpif_flow_flush(backer->dpif);
706 backer->need_revalidate = REV_RECONFIGURE;
709 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
710 * to reset the handler threads. */
711 if (backer->n_handler_threads != n_handler_threads) {
712 udpif_recv_set(backer->udpif, n_handler_threads,
713 backer->recv_set_enable);
714 backer->n_handler_threads = n_handler_threads;
717 if (backer->need_revalidate) {
718 struct ofproto_dpif *ofproto;
719 struct simap_node *node;
720 struct simap tmp_backers;
722 /* Handle tunnel garbage collection. */
723 simap_init(&tmp_backers);
724 simap_swap(&backer->tnl_backers, &tmp_backers);
726 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
727 struct ofport_dpif *iter;
729 if (backer != ofproto->backer) {
733 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
734 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
737 if (!iter->is_tunnel) {
741 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
742 namebuf, sizeof namebuf);
743 node = simap_find(&tmp_backers, dp_port);
745 simap_put(&backer->tnl_backers, dp_port, node->data);
746 simap_delete(&tmp_backers, node);
747 node = simap_find(&backer->tnl_backers, dp_port);
749 node = simap_find(&backer->tnl_backers, dp_port);
751 odp_port_t odp_port = ODPP_NONE;
753 if (!dpif_port_add(backer->dpif, iter->up.netdev,
755 simap_put(&backer->tnl_backers, dp_port,
756 odp_to_u32(odp_port));
757 node = simap_find(&backer->tnl_backers, dp_port);
762 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
763 if (tnl_port_reconfigure(iter, iter->up.netdev,
765 backer->need_revalidate = REV_RECONFIGURE;
770 SIMAP_FOR_EACH (node, &tmp_backers) {
771 dpif_port_del(backer->dpif, u32_to_odp(node->data));
773 simap_destroy(&tmp_backers);
775 switch (backer->need_revalidate) {
776 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
777 case REV_STP: COVERAGE_INC(rev_stp); break;
778 case REV_BOND: COVERAGE_INC(rev_bond); break;
779 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
780 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
781 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
782 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
784 backer->need_revalidate = 0;
786 /* Clear the drop_keys in case we should now be accepting some
787 * formerly dropped flows. */
788 drop_key_clear(backer);
790 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
791 struct facet *facet, *next;
792 struct ofport_dpif *ofport;
793 struct cls_cursor cursor;
794 struct ofbundle *bundle;
796 if (ofproto->backer != backer) {
800 ovs_rwlock_wrlock(&xlate_rwlock);
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,
831 ovs_rwlock_unlock(&xlate_rwlock);
833 /* Only ofproto-dpif cares about the facet classifier so we just
834 * lock cls_cursor_init() to appease the thread safety analysis. */
835 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
836 cls_cursor_init(&cursor, &ofproto->facets, NULL);
837 ovs_rwlock_unlock(&ofproto->facets.rwlock);
838 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
839 facet_revalidate(facet);
844 udpif_revalidate(backer->udpif);
847 if (!backer->recv_set_enable) {
848 /* Wake up before a max of 1000ms. */
849 timer_set_duration(&backer->next_expiration, 1000);
850 } else if (timer_expired(&backer->next_expiration)) {
851 int delay = expire(backer);
852 timer_set_duration(&backer->next_expiration, delay);
855 process_dpif_port_changes(backer);
857 if (backer->governor) {
860 governor_run(backer->governor);
862 /* If the governor has shrunk to its minimum size and the number of
863 * subfacets has dwindled, then drop the governor entirely.
865 * For hysteresis, the number of subfacets to drop the governor is
866 * smaller than the number needed to trigger its creation. */
867 n_subfacets = hmap_count(&backer->subfacets);
868 if (n_subfacets * 4 < flow_eviction_threshold
869 && governor_is_idle(backer->governor)) {
870 governor_destroy(backer->governor);
871 backer->governor = NULL;
878 /* Check for and handle port changes in 'backer''s dpif. */
880 process_dpif_port_changes(struct dpif_backer *backer)
886 error = dpif_port_poll(backer->dpif, &devname);
892 process_dpif_all_ports_changed(backer);
896 process_dpif_port_change(backer, devname);
901 process_dpif_port_error(backer, error);
908 process_dpif_all_ports_changed(struct dpif_backer *backer)
910 struct ofproto_dpif *ofproto;
911 struct dpif_port dpif_port;
912 struct dpif_port_dump dump;
913 struct sset devnames;
916 sset_init(&devnames);
917 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
918 if (ofproto->backer == backer) {
919 struct ofport *ofport;
921 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
922 sset_add(&devnames, netdev_get_name(ofport->netdev));
926 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
927 sset_add(&devnames, dpif_port.name);
930 SSET_FOR_EACH (devname, &devnames) {
931 process_dpif_port_change(backer, devname);
933 sset_destroy(&devnames);
937 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
939 struct ofproto_dpif *ofproto;
940 struct dpif_port port;
942 /* Don't report on the datapath's device. */
943 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
947 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
948 &all_ofproto_dpifs) {
949 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
954 ofproto = lookup_ofproto_dpif_by_port_name(devname);
955 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
956 /* The port was removed. If we know the datapath,
957 * report it through poll_set(). If we don't, it may be
958 * notifying us of a removal we initiated, so ignore it.
959 * If there's a pending ENOBUFS, let it stand, since
960 * everything will be reevaluated. */
961 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
962 sset_add(&ofproto->port_poll_set, devname);
963 ofproto->port_poll_errno = 0;
965 } else if (!ofproto) {
966 /* The port was added, but we don't know with which
967 * ofproto we should associate it. Delete it. */
968 dpif_port_del(backer->dpif, port.port_no);
970 struct ofport_dpif *ofport;
972 ofport = ofport_dpif_cast(shash_find_data(
973 &ofproto->up.port_by_name, devname));
975 && ofport->odp_port != port.port_no
976 && !odp_port_to_ofport(backer, port.port_no))
978 /* 'ofport''s datapath port number has changed from
979 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
980 * structures to match. */
981 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
982 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
983 ofport->odp_port = port.port_no;
984 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
985 hash_odp_port(port.port_no));
986 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
987 backer->need_revalidate = REV_RECONFIGURE;
990 dpif_port_destroy(&port);
993 /* Propagate 'error' to all ofprotos based on 'backer'. */
995 process_dpif_port_error(struct dpif_backer *backer, int error)
997 struct ofproto_dpif *ofproto;
999 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1000 if (ofproto->backer == backer) {
1001 sset_clear(&ofproto->port_poll_set);
1002 ofproto->port_poll_errno = error;
1008 dpif_backer_run_fast(struct dpif_backer *backer)
1010 handle_upcalls(backer);
1016 type_run_fast(const char *type)
1018 struct dpif_backer *backer;
1020 backer = shash_find_data(&all_dpif_backers, type);
1022 /* This is not necessarily a problem, since backers are only
1023 * created on demand. */
1027 return dpif_backer_run_fast(backer);
1033 static long long int port_rl = LLONG_MIN;
1035 if (time_msec() >= port_rl) {
1036 struct ofproto_dpif *ofproto;
1038 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1039 run_fast(&ofproto->up);
1041 port_rl = time_msec() + 200;
1046 type_wait(const char *type)
1048 struct dpif_backer *backer;
1050 backer = shash_find_data(&all_dpif_backers, type);
1052 /* This is not necessarily a problem, since backers are only
1053 * created on demand. */
1057 if (backer->governor) {
1058 governor_wait(backer->governor);
1061 timer_wait(&backer->next_expiration);
1062 dpif_wait(backer->dpif);
1063 udpif_wait(backer->udpif);
1066 /* Basic life-cycle. */
1068 static int add_internal_flows(struct ofproto_dpif *);
1070 static struct ofproto *
1073 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1074 return &ofproto->up;
1078 dealloc(struct ofproto *ofproto_)
1080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085 close_dpif_backer(struct dpif_backer *backer)
1087 struct shash_node *node;
1089 ovs_assert(backer->refcount > 0);
1091 if (--backer->refcount) {
1095 drop_key_clear(backer);
1096 hmap_destroy(&backer->drop_keys);
1098 simap_destroy(&backer->tnl_backers);
1099 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1100 hmap_destroy(&backer->odp_to_ofport_map);
1101 node = shash_find(&all_dpif_backers, backer->type);
1103 shash_delete(&all_dpif_backers, node);
1104 udpif_destroy(backer->udpif);
1105 dpif_close(backer->dpif);
1107 ovs_assert(hmap_is_empty(&backer->subfacets));
1108 hmap_destroy(&backer->subfacets);
1109 governor_destroy(backer->governor);
1114 /* Datapath port slated for removal from datapath. */
1115 struct odp_garbage {
1116 struct list list_node;
1117 odp_port_t odp_port;
1121 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1123 struct dpif_backer *backer;
1124 struct dpif_port_dump port_dump;
1125 struct dpif_port port;
1126 struct shash_node *node;
1127 struct list garbage_list;
1128 struct odp_garbage *garbage, *next;
1134 backer = shash_find_data(&all_dpif_backers, type);
1141 backer_name = xasprintf("ovs-%s", type);
1143 /* Remove any existing datapaths, since we assume we're the only
1144 * userspace controlling the datapath. */
1146 dp_enumerate_names(type, &names);
1147 SSET_FOR_EACH(name, &names) {
1148 struct dpif *old_dpif;
1150 /* Don't remove our backer if it exists. */
1151 if (!strcmp(name, backer_name)) {
1155 if (dpif_open(name, type, &old_dpif)) {
1156 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1158 dpif_delete(old_dpif);
1159 dpif_close(old_dpif);
1162 sset_destroy(&names);
1164 backer = xmalloc(sizeof *backer);
1166 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1169 VLOG_ERR("failed to open datapath of type %s: %s", type,
1170 ovs_strerror(error));
1174 backer->udpif = udpif_create(backer, backer->dpif);
1176 backer->type = xstrdup(type);
1177 backer->governor = NULL;
1178 backer->refcount = 1;
1179 hmap_init(&backer->odp_to_ofport_map);
1180 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1181 hmap_init(&backer->drop_keys);
1182 hmap_init(&backer->subfacets);
1183 timer_set_duration(&backer->next_expiration, 1000);
1184 backer->need_revalidate = 0;
1185 simap_init(&backer->tnl_backers);
1186 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1189 if (backer->recv_set_enable) {
1190 dpif_flow_flush(backer->dpif);
1193 /* Loop through the ports already on the datapath and remove any
1194 * that we don't need anymore. */
1195 list_init(&garbage_list);
1196 dpif_port_dump_start(&port_dump, backer->dpif);
1197 while (dpif_port_dump_next(&port_dump, &port)) {
1198 node = shash_find(&init_ofp_ports, port.name);
1199 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1200 garbage = xmalloc(sizeof *garbage);
1201 garbage->odp_port = port.port_no;
1202 list_push_front(&garbage_list, &garbage->list_node);
1205 dpif_port_dump_done(&port_dump);
1207 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1208 dpif_port_del(backer->dpif, garbage->odp_port);
1209 list_remove(&garbage->list_node);
1213 shash_add(&all_dpif_backers, type, backer);
1215 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1217 VLOG_ERR("failed to listen on datapath of type %s: %s",
1218 type, ovs_strerror(error));
1219 close_dpif_backer(backer);
1222 udpif_recv_set(backer->udpif, n_handler_threads,
1223 backer->recv_set_enable);
1224 backer->n_handler_threads = n_handler_threads;
1226 backer->max_n_subfacet = 0;
1227 backer->created = time_msec();
1228 backer->last_minute = backer->created;
1229 memset(&backer->hourly, 0, sizeof backer->hourly);
1230 memset(&backer->daily, 0, sizeof backer->daily);
1231 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1232 backer->subfacet_add_count = 0;
1233 backer->subfacet_del_count = 0;
1234 backer->total_subfacet_add_count = 0;
1235 backer->total_subfacet_del_count = 0;
1236 backer->avg_n_subfacet = 0;
1237 backer->avg_subfacet_life = 0;
1243 construct(struct ofproto *ofproto_)
1245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1246 struct shash_node *node, *next;
1250 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1256 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1258 ofproto->netflow = NULL;
1259 ofproto->sflow = NULL;
1260 ofproto->ipfix = NULL;
1261 ofproto->stp = NULL;
1262 hmap_init(&ofproto->bundles);
1263 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1264 ofproto->mbridge = mbridge_create();
1265 ofproto->has_bonded_bundles = false;
1266 ovs_mutex_init(&ofproto->vsp_mutex);
1268 classifier_init(&ofproto->facets);
1269 ofproto->consistency_rl = LLONG_MIN;
1271 guarded_list_init(&ofproto->flow_mods);
1272 guarded_list_init(&ofproto->pins);
1274 ofproto_dpif_unixctl_init();
1276 hmap_init(&ofproto->vlandev_map);
1277 hmap_init(&ofproto->realdev_vid_map);
1279 sset_init(&ofproto->ports);
1280 sset_init(&ofproto->ghost_ports);
1281 sset_init(&ofproto->port_poll_set);
1282 ofproto->port_poll_errno = 0;
1284 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1285 struct iface_hint *iface_hint = node->data;
1287 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1288 /* Check if the datapath already has this port. */
1289 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1290 sset_add(&ofproto->ports, node->name);
1293 free(iface_hint->br_name);
1294 free(iface_hint->br_type);
1296 shash_delete(&init_ofp_ports, node);
1300 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1301 hash_string(ofproto->up.name, 0));
1302 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1304 ofproto_init_tables(ofproto_, N_TABLES);
1305 error = add_internal_flows(ofproto);
1306 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1309 ofproto->n_missed = 0;
1315 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1316 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1318 struct ofputil_flow_mod fm;
1321 match_init_catchall(&fm.match);
1323 match_set_reg(&fm.match, 0, id);
1324 fm.new_cookie = htonll(0);
1325 fm.cookie = htonll(0);
1326 fm.cookie_mask = htonll(0);
1327 fm.modify_cookie = false;
1328 fm.table_id = TBL_INTERNAL;
1329 fm.command = OFPFC_ADD;
1330 fm.idle_timeout = 0;
1331 fm.hard_timeout = 0;
1335 fm.ofpacts = ofpacts->data;
1336 fm.ofpacts_len = ofpacts->size;
1338 error = ofproto_flow_mod(&ofproto->up, &fm);
1340 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1341 id, ofperr_to_string(error));
1345 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1347 rule_dpif_unref(*rulep);
1356 add_internal_flows(struct ofproto_dpif *ofproto)
1358 struct ofpact_controller *controller;
1359 uint64_t ofpacts_stub[128 / 8];
1360 struct ofpbuf ofpacts;
1364 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1367 controller = ofpact_put_CONTROLLER(&ofpacts);
1368 controller->max_len = UINT16_MAX;
1369 controller->controller_id = 0;
1370 controller->reason = OFPR_NO_MATCH;
1371 ofpact_pad(&ofpacts);
1373 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1378 ofpbuf_clear(&ofpacts);
1379 error = add_internal_flow(ofproto, id++, &ofpacts,
1380 &ofproto->no_packet_in_rule);
1385 error = add_internal_flow(ofproto, id++, &ofpacts,
1386 &ofproto->drop_frags_rule);
1391 destruct(struct ofproto *ofproto_)
1393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1394 struct rule_dpif *rule, *next_rule;
1395 struct ofputil_packet_in *pin, *next_pin;
1396 struct ofputil_flow_mod *fm, *next_fm;
1397 struct facet *facet, *next_facet;
1398 struct list flow_mods, pins;
1399 struct cls_cursor cursor;
1400 struct oftable *table;
1402 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1403 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1404 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1405 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1406 facet_remove(facet);
1409 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1410 ovs_rwlock_wrlock(&xlate_rwlock);
1411 xlate_remove_ofproto(ofproto);
1412 ovs_rwlock_unlock(&xlate_rwlock);
1414 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1415 * use-after-free error. */
1416 udpif_revalidate(ofproto->backer->udpif);
1418 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1420 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1421 struct cls_cursor cursor;
1423 ovs_rwlock_rdlock(&table->cls.rwlock);
1424 cls_cursor_init(&cursor, &table->cls, NULL);
1425 ovs_rwlock_unlock(&table->cls.rwlock);
1426 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1427 ofproto_rule_delete(&ofproto->up, &rule->up);
1431 guarded_list_pop_all(&ofproto->flow_mods, &flow_mods);
1432 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1433 list_remove(&fm->list_node);
1437 guarded_list_destroy(&ofproto->flow_mods);
1439 guarded_list_pop_all(&ofproto->pins, &pins);
1440 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1441 list_remove(&pin->list_node);
1442 free(CONST_CAST(void *, pin->packet));
1445 guarded_list_destroy(&ofproto->pins);
1447 mbridge_unref(ofproto->mbridge);
1449 netflow_destroy(ofproto->netflow);
1450 dpif_sflow_unref(ofproto->sflow);
1451 hmap_destroy(&ofproto->bundles);
1452 mac_learning_unref(ofproto->ml);
1454 classifier_destroy(&ofproto->facets);
1456 hmap_destroy(&ofproto->vlandev_map);
1457 hmap_destroy(&ofproto->realdev_vid_map);
1459 sset_destroy(&ofproto->ports);
1460 sset_destroy(&ofproto->ghost_ports);
1461 sset_destroy(&ofproto->port_poll_set);
1463 ovs_mutex_destroy(&ofproto->vsp_mutex);
1465 close_dpif_backer(ofproto->backer);
1469 run_fast(struct ofproto *ofproto_)
1471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1472 struct ofputil_packet_in *pin, *next_pin;
1473 struct ofputil_flow_mod *fm, *next_fm;
1474 struct list flow_mods, pins;
1475 struct ofport_dpif *ofport;
1477 /* Do not perform any periodic activity required by 'ofproto' while
1478 * waiting for flow restore to complete. */
1479 if (ofproto_get_flow_restore_wait()) {
1483 guarded_list_pop_all(&ofproto->flow_mods, &flow_mods);
1484 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1485 int error = ofproto_flow_mod(&ofproto->up, fm);
1486 if (error && !VLOG_DROP_WARN(&rl)) {
1487 VLOG_WARN("learning action failed to modify flow table (%s)",
1488 ofperr_get_name(error));
1491 list_remove(&fm->list_node);
1496 guarded_list_pop_all(&ofproto->pins, &pins);
1497 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1498 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1499 list_remove(&pin->list_node);
1500 free(CONST_CAST(void *, pin->packet));
1504 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1505 port_run_fast(ofport);
1512 run(struct ofproto *ofproto_)
1514 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1515 struct ofport_dpif *ofport;
1516 struct ofbundle *bundle;
1519 if (mbridge_need_revalidate(ofproto->mbridge)) {
1520 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1521 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1522 mac_learning_flush(ofproto->ml);
1523 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1526 /* Do not perform any periodic activity below required by 'ofproto' while
1527 * waiting for flow restore to complete. */
1528 if (ofproto_get_flow_restore_wait()) {
1532 error = run_fast(ofproto_);
1537 if (ofproto->netflow) {
1538 if (netflow_run(ofproto->netflow)) {
1539 send_netflow_active_timeouts(ofproto);
1542 if (ofproto->sflow) {
1543 dpif_sflow_run(ofproto->sflow);
1545 if (ofproto->ipfix) {
1546 dpif_ipfix_run(ofproto->ipfix);
1549 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1552 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1557 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1558 if (mac_learning_run(ofproto->ml)) {
1559 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1561 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1563 /* Check the consistency of a random facet, to aid debugging. */
1564 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1565 if (time_msec() >= ofproto->consistency_rl
1566 && !classifier_is_empty(&ofproto->facets)
1567 && !ofproto->backer->need_revalidate) {
1568 struct cls_table *table;
1569 struct cls_rule *cr;
1570 struct facet *facet;
1572 ofproto->consistency_rl = time_msec() + 250;
1574 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1575 struct cls_table, hmap_node);
1576 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1578 facet = CONTAINER_OF(cr, struct facet, cr);
1580 if (!facet_check_consistency(facet)) {
1581 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1584 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1590 wait(struct ofproto *ofproto_)
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 struct ofport_dpif *ofport;
1594 struct ofbundle *bundle;
1596 if (ofproto_get_flow_restore_wait()) {
1600 if (ofproto->sflow) {
1601 dpif_sflow_wait(ofproto->sflow);
1603 if (ofproto->ipfix) {
1604 dpif_ipfix_wait(ofproto->ipfix);
1606 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1609 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1610 bundle_wait(bundle);
1612 if (ofproto->netflow) {
1613 netflow_wait(ofproto->netflow);
1615 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1616 mac_learning_wait(ofproto->ml);
1617 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1619 if (ofproto->backer->need_revalidate) {
1620 /* Shouldn't happen, but if it does just go around again. */
1621 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1622 poll_immediate_wake();
1627 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1629 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1630 struct cls_cursor cursor;
1631 size_t n_subfacets = 0;
1632 struct facet *facet;
1634 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1635 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1636 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1638 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1639 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1640 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1641 n_subfacets += list_size(&facet->subfacets);
1643 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1644 simap_increase(usage, "subfacets", n_subfacets);
1648 flush(struct ofproto *ofproto_)
1650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1651 struct subfacet *subfacet, *next_subfacet;
1652 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1656 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1657 &ofproto->backer->subfacets) {
1658 if (subfacet->facet->ofproto != ofproto) {
1662 if (subfacet->path != SF_NOT_INSTALLED) {
1663 batch[n_batch++] = subfacet;
1664 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1665 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1669 subfacet_destroy(subfacet);
1674 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1679 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1680 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1682 *arp_match_ip = true;
1683 *actions = (OFPUTIL_A_OUTPUT |
1684 OFPUTIL_A_SET_VLAN_VID |
1685 OFPUTIL_A_SET_VLAN_PCP |
1686 OFPUTIL_A_STRIP_VLAN |
1687 OFPUTIL_A_SET_DL_SRC |
1688 OFPUTIL_A_SET_DL_DST |
1689 OFPUTIL_A_SET_NW_SRC |
1690 OFPUTIL_A_SET_NW_DST |
1691 OFPUTIL_A_SET_NW_TOS |
1692 OFPUTIL_A_SET_TP_SRC |
1693 OFPUTIL_A_SET_TP_DST |
1698 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1701 struct dpif_dp_stats s;
1702 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1705 strcpy(ots->name, "classifier");
1707 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1708 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1709 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1710 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1712 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1713 ots->lookup_count = htonll(n_lookup);
1714 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1717 static struct ofport *
1720 struct ofport_dpif *port = xmalloc(sizeof *port);
1725 port_dealloc(struct ofport *port_)
1727 struct ofport_dpif *port = ofport_dpif_cast(port_);
1732 port_construct(struct ofport *port_)
1734 struct ofport_dpif *port = ofport_dpif_cast(port_);
1735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1736 const struct netdev *netdev = port->up.netdev;
1737 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1738 struct dpif_port dpif_port;
1741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1742 port->bundle = NULL;
1745 port->may_enable = true;
1746 port->stp_port = NULL;
1747 port->stp_state = STP_DISABLED;
1748 port->is_tunnel = false;
1752 port->realdev_ofp_port = 0;
1753 port->vlandev_vid = 0;
1754 port->carrier_seq = netdev_get_carrier_resets(netdev);
1756 if (netdev_vport_is_patch(netdev)) {
1757 /* By bailing out here, we don't submit the port to the sFlow module
1758 * to be considered for counter polling export. This is correct
1759 * because the patch port represents an interface that sFlow considers
1760 * to be "internal" to the switch as a whole, and therefore not an
1761 * candidate for counter polling. */
1762 port->odp_port = ODPP_NONE;
1763 ofport_update_peer(port);
1767 error = dpif_port_query_by_name(ofproto->backer->dpif,
1768 netdev_vport_get_dpif_port(netdev, namebuf,
1775 port->odp_port = dpif_port.port_no;
1777 if (netdev_get_tunnel_config(netdev)) {
1778 tnl_port_add(port, port->up.netdev, port->odp_port);
1779 port->is_tunnel = true;
1781 /* Sanity-check that a mapping doesn't already exist. This
1782 * shouldn't happen for non-tunnel ports. */
1783 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1784 VLOG_ERR("port %s already has an OpenFlow port number",
1786 dpif_port_destroy(&dpif_port);
1790 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1791 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1792 hash_odp_port(port->odp_port));
1793 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1795 dpif_port_destroy(&dpif_port);
1797 if (ofproto->sflow) {
1798 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1805 port_destruct(struct ofport *port_)
1807 struct ofport_dpif *port = ofport_dpif_cast(port_);
1808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1809 const char *devname = netdev_get_name(port->up.netdev);
1810 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1811 const char *dp_port_name;
1813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1814 ovs_rwlock_wrlock(&xlate_rwlock);
1815 xlate_ofport_remove(port);
1816 ovs_rwlock_unlock(&xlate_rwlock);
1818 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1820 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1821 /* The underlying device is still there, so delete it. This
1822 * happens when the ofproto is being destroyed, since the caller
1823 * assumes that removal of attached ports will happen as part of
1825 if (!port->is_tunnel) {
1826 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1831 port->peer->peer = NULL;
1835 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1836 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1837 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1838 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1842 sset_find_and_delete(&ofproto->ports, devname);
1843 sset_find_and_delete(&ofproto->ghost_ports, devname);
1844 bundle_remove(port_);
1845 set_cfm(port_, NULL);
1846 set_bfd(port_, NULL);
1847 if (ofproto->sflow) {
1848 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1855 port_modified(struct ofport *port_)
1857 struct ofport_dpif *port = ofport_dpif_cast(port_);
1859 if (port->bundle && port->bundle->bond) {
1860 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1864 cfm_set_netdev(port->cfm, port->up.netdev);
1868 bfd_set_netdev(port->bfd, port->up.netdev);
1871 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1873 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1877 ofport_update_peer(port);
1881 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1883 struct ofport_dpif *port = ofport_dpif_cast(port_);
1884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1885 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1887 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1888 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1889 OFPUTIL_PC_NO_PACKET_IN)) {
1890 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1893 bundle_update(port->bundle);
1899 set_sflow(struct ofproto *ofproto_,
1900 const struct ofproto_sflow_options *sflow_options)
1902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1903 struct dpif_sflow *ds = ofproto->sflow;
1905 if (sflow_options) {
1907 struct ofport_dpif *ofport;
1909 ds = ofproto->sflow = dpif_sflow_create();
1910 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1911 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1915 dpif_sflow_set_options(ds, sflow_options);
1918 dpif_sflow_unref(ds);
1919 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1920 ofproto->sflow = NULL;
1928 struct ofproto *ofproto_,
1929 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1930 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1931 size_t n_flow_exporters_options)
1933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1934 struct dpif_ipfix *di = ofproto->ipfix;
1935 bool has_options = bridge_exporter_options || flow_exporters_options;
1937 if (has_options && !di) {
1938 di = ofproto->ipfix = dpif_ipfix_create();
1942 /* Call set_options in any case to cleanly flush the flow
1943 * caches in the last exporters that are to be destroyed. */
1944 dpif_ipfix_set_options(
1945 di, bridge_exporter_options, flow_exporters_options,
1946 n_flow_exporters_options);
1949 dpif_ipfix_unref(di);
1950 ofproto->ipfix = NULL;
1958 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1960 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1967 struct ofproto_dpif *ofproto;
1969 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1970 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971 ofport->cfm = cfm_create(ofport->up.netdev);
1974 if (cfm_configure(ofport->cfm, s)) {
1980 cfm_unref(ofport->cfm);
1986 get_cfm_status(const struct ofport *ofport_,
1987 struct ofproto_cfm_status *status)
1989 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1992 status->faults = cfm_get_fault(ofport->cfm);
1993 status->remote_opstate = cfm_get_opup(ofport->cfm);
1994 status->health = cfm_get_health(ofport->cfm);
1995 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2003 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2006 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2010 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2011 cfg, ofport->up.netdev);
2012 if (ofport->bfd != old) {
2013 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2020 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2022 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2025 bfd_get_status(ofport->bfd, smap);
2032 /* Spanning Tree. */
2035 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2037 struct ofproto_dpif *ofproto = ofproto_;
2038 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2039 struct ofport_dpif *ofport;
2041 ofport = stp_port_get_aux(sp);
2043 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2044 ofproto->up.name, port_num);
2046 struct eth_header *eth = pkt->l2;
2048 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2049 if (eth_addr_is_zero(eth->eth_src)) {
2050 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2051 "with unknown MAC", ofproto->up.name, port_num);
2053 send_packet(ofport, pkt);
2059 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2061 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2065 /* Only revalidate flows if the configuration changed. */
2066 if (!s != !ofproto->stp) {
2067 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2071 if (!ofproto->stp) {
2072 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2073 send_bpdu_cb, ofproto);
2074 ofproto->stp_last_tick = time_msec();
2077 stp_set_bridge_id(ofproto->stp, s->system_id);
2078 stp_set_bridge_priority(ofproto->stp, s->priority);
2079 stp_set_hello_time(ofproto->stp, s->hello_time);
2080 stp_set_max_age(ofproto->stp, s->max_age);
2081 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2083 struct ofport *ofport;
2085 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2086 set_stp_port(ofport, NULL);
2089 stp_unref(ofproto->stp);
2090 ofproto->stp = NULL;
2097 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2103 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2104 s->designated_root = stp_get_designated_root(ofproto->stp);
2105 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2114 update_stp_port_state(struct ofport_dpif *ofport)
2116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2117 enum stp_state state;
2119 /* Figure out new state. */
2120 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2124 if (ofport->stp_state != state) {
2125 enum ofputil_port_state of_state;
2128 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2129 netdev_get_name(ofport->up.netdev),
2130 stp_state_name(ofport->stp_state),
2131 stp_state_name(state));
2132 if (stp_learn_in_state(ofport->stp_state)
2133 != stp_learn_in_state(state)) {
2134 /* xxx Learning action flows should also be flushed. */
2135 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2136 mac_learning_flush(ofproto->ml);
2137 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2139 fwd_change = stp_forward_in_state(ofport->stp_state)
2140 != stp_forward_in_state(state);
2142 ofproto->backer->need_revalidate = REV_STP;
2143 ofport->stp_state = state;
2144 ofport->stp_state_entered = time_msec();
2146 if (fwd_change && ofport->bundle) {
2147 bundle_update(ofport->bundle);
2150 /* Update the STP state bits in the OpenFlow port description. */
2151 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2152 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2153 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2154 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2155 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2157 ofproto_port_set_state(&ofport->up, of_state);
2161 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2162 * caller is responsible for assigning STP port numbers and ensuring
2163 * there are no duplicates. */
2165 set_stp_port(struct ofport *ofport_,
2166 const struct ofproto_port_stp_settings *s)
2168 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2170 struct stp_port *sp = ofport->stp_port;
2172 if (!s || !s->enable) {
2174 ofport->stp_port = NULL;
2175 stp_port_disable(sp);
2176 update_stp_port_state(ofport);
2179 } else if (sp && stp_port_no(sp) != s->port_num
2180 && ofport == stp_port_get_aux(sp)) {
2181 /* The port-id changed, so disable the old one if it's not
2182 * already in use by another port. */
2183 stp_port_disable(sp);
2186 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2187 stp_port_enable(sp);
2189 stp_port_set_aux(sp, ofport);
2190 stp_port_set_priority(sp, s->priority);
2191 stp_port_set_path_cost(sp, s->path_cost);
2193 update_stp_port_state(ofport);
2199 get_stp_port_status(struct ofport *ofport_,
2200 struct ofproto_port_stp_status *s)
2202 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2203 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2204 struct stp_port *sp = ofport->stp_port;
2206 if (!ofproto->stp || !sp) {
2212 s->port_id = stp_port_get_id(sp);
2213 s->state = stp_port_get_state(sp);
2214 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2215 s->role = stp_port_get_role(sp);
2216 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2222 stp_run(struct ofproto_dpif *ofproto)
2225 long long int now = time_msec();
2226 long long int elapsed = now - ofproto->stp_last_tick;
2227 struct stp_port *sp;
2230 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2231 ofproto->stp_last_tick = now;
2233 while (stp_get_changed_port(ofproto->stp, &sp)) {
2234 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2237 update_stp_port_state(ofport);
2241 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2242 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2243 mac_learning_flush(ofproto->ml);
2244 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2250 stp_wait(struct ofproto_dpif *ofproto)
2253 poll_timer_wait(1000);
2258 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2261 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2264 if (ofport->n_qdscp != n_qdscp
2265 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2266 n_qdscp * sizeof *qdscp))) {
2267 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2268 free(ofport->qdscp);
2269 ofport->qdscp = n_qdscp
2270 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2272 ofport->n_qdscp = n_qdscp;
2280 /* Expires all MAC learning entries associated with 'bundle' and forces its
2281 * ofproto to revalidate every flow.
2283 * Normally MAC learning entries are removed only from the ofproto associated
2284 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2285 * are removed from every ofproto. When patch ports and SLB bonds are in use
2286 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2287 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2288 * with the host from which it migrated. */
2290 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2292 struct ofproto_dpif *ofproto = bundle->ofproto;
2293 struct mac_learning *ml = ofproto->ml;
2294 struct mac_entry *mac, *next_mac;
2296 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2297 ovs_rwlock_wrlock(&ml->rwlock);
2298 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2299 if (mac->port.p == bundle) {
2301 struct ofproto_dpif *o;
2303 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2305 struct mac_entry *e;
2307 ovs_rwlock_wrlock(&o->ml->rwlock);
2308 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2310 mac_learning_expire(o->ml, e);
2312 ovs_rwlock_unlock(&o->ml->rwlock);
2317 mac_learning_expire(ml, mac);
2320 ovs_rwlock_unlock(&ml->rwlock);
2323 static struct ofbundle *
2324 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2326 struct ofbundle *bundle;
2328 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2329 &ofproto->bundles) {
2330 if (bundle->aux == aux) {
2338 bundle_update(struct ofbundle *bundle)
2340 struct ofport_dpif *port;
2342 bundle->floodable = true;
2343 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2344 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2345 || !stp_forward_in_state(port->stp_state)) {
2346 bundle->floodable = false;
2353 bundle_del_port(struct ofport_dpif *port)
2355 struct ofbundle *bundle = port->bundle;
2357 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2359 list_remove(&port->bundle_node);
2360 port->bundle = NULL;
2363 lacp_slave_unregister(bundle->lacp, port);
2366 bond_slave_unregister(bundle->bond, port);
2369 bundle_update(bundle);
2373 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2374 struct lacp_slave_settings *lacp)
2376 struct ofport_dpif *port;
2378 port = get_ofp_port(bundle->ofproto, ofp_port);
2383 if (port->bundle != bundle) {
2384 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 bundle_remove(&port->up);
2389 port->bundle = bundle;
2390 list_push_back(&bundle->ports, &port->bundle_node);
2391 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2392 || !stp_forward_in_state(port->stp_state)) {
2393 bundle->floodable = false;
2397 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2398 lacp_slave_register(bundle->lacp, port, lacp);
2405 bundle_destroy(struct ofbundle *bundle)
2407 struct ofproto_dpif *ofproto;
2408 struct ofport_dpif *port, *next_port;
2414 ofproto = bundle->ofproto;
2415 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2417 ovs_rwlock_wrlock(&xlate_rwlock);
2418 xlate_bundle_remove(bundle);
2419 ovs_rwlock_unlock(&xlate_rwlock);
2421 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2422 bundle_del_port(port);
2425 bundle_flush_macs(bundle, true);
2426 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2428 free(bundle->trunks);
2429 lacp_unref(bundle->lacp);
2430 bond_unref(bundle->bond);
2435 bundle_set(struct ofproto *ofproto_, void *aux,
2436 const struct ofproto_bundle_settings *s)
2438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2439 bool need_flush = false;
2440 struct ofport_dpif *port;
2441 struct ofbundle *bundle;
2442 unsigned long *trunks;
2448 bundle_destroy(bundle_lookup(ofproto, aux));
2452 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2453 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2455 bundle = bundle_lookup(ofproto, aux);
2457 bundle = xmalloc(sizeof *bundle);
2459 bundle->ofproto = ofproto;
2460 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2461 hash_pointer(aux, 0));
2463 bundle->name = NULL;
2465 list_init(&bundle->ports);
2466 bundle->vlan_mode = PORT_VLAN_TRUNK;
2468 bundle->trunks = NULL;
2469 bundle->use_priority_tags = s->use_priority_tags;
2470 bundle->lacp = NULL;
2471 bundle->bond = NULL;
2473 bundle->floodable = true;
2474 mbridge_register_bundle(ofproto->mbridge, bundle);
2477 if (!bundle->name || strcmp(s->name, bundle->name)) {
2479 bundle->name = xstrdup(s->name);
2484 if (!bundle->lacp) {
2485 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2486 bundle->lacp = lacp_create();
2488 lacp_configure(bundle->lacp, s->lacp);
2490 lacp_unref(bundle->lacp);
2491 bundle->lacp = NULL;
2494 /* Update set of ports. */
2496 for (i = 0; i < s->n_slaves; i++) {
2497 if (!bundle_add_port(bundle, s->slaves[i],
2498 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2502 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2503 struct ofport_dpif *next_port;
2505 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2506 for (i = 0; i < s->n_slaves; i++) {
2507 if (s->slaves[i] == port->up.ofp_port) {
2512 bundle_del_port(port);
2516 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2518 if (list_is_empty(&bundle->ports)) {
2519 bundle_destroy(bundle);
2523 /* Set VLAN tagging mode */
2524 if (s->vlan_mode != bundle->vlan_mode
2525 || s->use_priority_tags != bundle->use_priority_tags) {
2526 bundle->vlan_mode = s->vlan_mode;
2527 bundle->use_priority_tags = s->use_priority_tags;
2532 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2533 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2535 if (vlan != bundle->vlan) {
2536 bundle->vlan = vlan;
2540 /* Get trunked VLANs. */
2541 switch (s->vlan_mode) {
2542 case PORT_VLAN_ACCESS:
2546 case PORT_VLAN_TRUNK:
2547 trunks = CONST_CAST(unsigned long *, s->trunks);
2550 case PORT_VLAN_NATIVE_UNTAGGED:
2551 case PORT_VLAN_NATIVE_TAGGED:
2552 if (vlan != 0 && (!s->trunks
2553 || !bitmap_is_set(s->trunks, vlan)
2554 || bitmap_is_set(s->trunks, 0))) {
2555 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2557 trunks = bitmap_clone(s->trunks, 4096);
2559 trunks = bitmap_allocate1(4096);
2561 bitmap_set1(trunks, vlan);
2562 bitmap_set0(trunks, 0);
2564 trunks = CONST_CAST(unsigned long *, s->trunks);
2571 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2572 free(bundle->trunks);
2573 if (trunks == s->trunks) {
2574 bundle->trunks = vlan_bitmap_clone(trunks);
2576 bundle->trunks = trunks;
2581 if (trunks != s->trunks) {
2586 if (!list_is_short(&bundle->ports)) {
2587 bundle->ofproto->has_bonded_bundles = true;
2589 if (bond_reconfigure(bundle->bond, s->bond)) {
2590 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2593 bundle->bond = bond_create(s->bond);
2594 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2597 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2598 bond_slave_register(bundle->bond, port, port->up.netdev);
2601 bond_unref(bundle->bond);
2602 bundle->bond = NULL;
2605 /* If we changed something that would affect MAC learning, un-learn
2606 * everything on this port and force flow revalidation. */
2608 bundle_flush_macs(bundle, false);
2615 bundle_remove(struct ofport *port_)
2617 struct ofport_dpif *port = ofport_dpif_cast(port_);
2618 struct ofbundle *bundle = port->bundle;
2621 bundle_del_port(port);
2622 if (list_is_empty(&bundle->ports)) {
2623 bundle_destroy(bundle);
2624 } else if (list_is_short(&bundle->ports)) {
2625 bond_unref(bundle->bond);
2626 bundle->bond = NULL;
2632 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2634 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2635 struct ofport_dpif *port = port_;
2636 uint8_t ea[ETH_ADDR_LEN];
2639 error = netdev_get_etheraddr(port->up.netdev, ea);
2641 struct ofpbuf packet;
2644 ofpbuf_init(&packet, 0);
2645 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2647 memcpy(packet_pdu, pdu, pdu_size);
2649 send_packet(port, &packet);
2650 ofpbuf_uninit(&packet);
2652 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2653 "%s (%s)", port->bundle->name,
2654 netdev_get_name(port->up.netdev), ovs_strerror(error));
2659 bundle_send_learning_packets(struct ofbundle *bundle)
2661 struct ofproto_dpif *ofproto = bundle->ofproto;
2662 struct ofpbuf *learning_packet;
2663 int error, n_packets, n_errors;
2664 struct mac_entry *e;
2665 struct list packets;
2667 list_init(&packets);
2668 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2669 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2670 if (e->port.p != bundle) {
2673 learning_packet = bond_compose_learning_packet(bundle->bond,
2676 learning_packet->private_p = port_void;
2677 list_push_back(&packets, &learning_packet->list_node);
2680 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2682 error = n_packets = n_errors = 0;
2683 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2686 ret = send_packet(learning_packet->private_p, learning_packet);
2693 ofpbuf_list_delete(&packets);
2696 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2697 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2698 "packets, last error was: %s",
2699 bundle->name, n_errors, n_packets, ovs_strerror(error));
2701 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2702 bundle->name, n_packets);
2707 bundle_run(struct ofbundle *bundle)
2710 lacp_run(bundle->lacp, send_pdu_cb);
2713 struct ofport_dpif *port;
2715 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2716 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2719 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2720 bundle->ofproto->backer->need_revalidate = REV_BOND;
2723 if (bond_should_send_learning_packets(bundle->bond)) {
2724 bundle_send_learning_packets(bundle);
2730 bundle_wait(struct ofbundle *bundle)
2733 lacp_wait(bundle->lacp);
2736 bond_wait(bundle->bond);
2743 mirror_set__(struct ofproto *ofproto_, void *aux,
2744 const struct ofproto_mirror_settings *s)
2746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2747 struct ofbundle **srcs, **dsts;
2752 mirror_destroy(ofproto->mbridge, aux);
2756 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2757 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2759 for (i = 0; i < s->n_srcs; i++) {
2760 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2763 for (i = 0; i < s->n_dsts; i++) {
2764 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2767 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2768 s->n_dsts, s->src_vlans,
2769 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2776 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2777 uint64_t *packets, uint64_t *bytes)
2780 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2785 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2788 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2789 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2790 mac_learning_flush(ofproto->ml);
2792 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2797 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2800 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2801 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2805 forward_bpdu_changed(struct ofproto *ofproto_)
2807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2808 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2812 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2816 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2817 mac_learning_set_idle_time(ofproto->ml, idle_time);
2818 mac_learning_set_max_entries(ofproto->ml, max_entries);
2819 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2824 static struct ofport_dpif *
2825 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2827 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2828 return ofport ? ofport_dpif_cast(ofport) : NULL;
2831 static struct ofport_dpif *
2832 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2834 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2835 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2839 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2840 struct ofproto_port *ofproto_port,
2841 struct dpif_port *dpif_port)
2843 ofproto_port->name = dpif_port->name;
2844 ofproto_port->type = dpif_port->type;
2845 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2849 ofport_update_peer(struct ofport_dpif *ofport)
2851 const struct ofproto_dpif *ofproto;
2852 struct dpif_backer *backer;
2855 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2859 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2860 backer->need_revalidate = REV_RECONFIGURE;
2863 ofport->peer->peer = NULL;
2864 ofport->peer = NULL;
2867 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2872 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2873 struct ofport *peer_ofport;
2874 struct ofport_dpif *peer;
2877 if (ofproto->backer != backer) {
2881 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2886 peer = ofport_dpif_cast(peer_ofport);
2887 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2888 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2890 ofport->peer = peer;
2891 ofport->peer->peer = ofport;
2901 port_run_fast(struct ofport_dpif *ofport)
2903 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2904 struct ofpbuf packet;
2906 ofpbuf_init(&packet, 0);
2907 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2908 send_packet(ofport, &packet);
2909 ofpbuf_uninit(&packet);
2912 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2913 struct ofpbuf packet;
2915 ofpbuf_init(&packet, 0);
2916 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2917 send_packet(ofport, &packet);
2918 ofpbuf_uninit(&packet);
2923 port_run(struct ofport_dpif *ofport)
2925 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2926 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2927 bool enable = netdev_get_carrier(ofport->up.netdev);
2928 bool cfm_enable = false;
2929 bool bfd_enable = false;
2931 ofport->carrier_seq = carrier_seq;
2933 port_run_fast(ofport);
2936 int cfm_opup = cfm_get_opup(ofport->cfm);
2938 cfm_run(ofport->cfm);
2939 cfm_enable = !cfm_get_fault(ofport->cfm);
2941 if (cfm_opup >= 0) {
2942 cfm_enable = cfm_enable && cfm_opup;
2947 bfd_run(ofport->bfd);
2948 bfd_enable = bfd_forwarding(ofport->bfd);
2951 if (ofport->bfd || ofport->cfm) {
2952 enable = enable && (cfm_enable || bfd_enable);
2955 if (ofport->bundle) {
2956 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2957 if (carrier_changed) {
2958 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2962 if (ofport->may_enable != enable) {
2963 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2964 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2967 ofport->may_enable = enable;
2971 port_wait(struct ofport_dpif *ofport)
2974 cfm_wait(ofport->cfm);
2978 bfd_wait(ofport->bfd);
2983 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2984 struct ofproto_port *ofproto_port)
2986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2987 struct dpif_port dpif_port;
2990 if (sset_contains(&ofproto->ghost_ports, devname)) {
2991 const char *type = netdev_get_type_from_name(devname);
2993 /* We may be called before ofproto->up.port_by_name is populated with
2994 * the appropriate ofport. For this reason, we must get the name and
2995 * type from the netdev layer directly. */
2997 const struct ofport *ofport;
2999 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3000 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3001 ofproto_port->name = xstrdup(devname);
3002 ofproto_port->type = xstrdup(type);
3008 if (!sset_contains(&ofproto->ports, devname)) {
3011 error = dpif_port_query_by_name(ofproto->backer->dpif,
3012 devname, &dpif_port);
3014 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3020 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3023 const char *devname = netdev_get_name(netdev);
3024 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3025 const char *dp_port_name;
3027 if (netdev_vport_is_patch(netdev)) {
3028 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3032 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3033 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3034 odp_port_t port_no = ODPP_NONE;
3037 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3041 if (netdev_get_tunnel_config(netdev)) {
3042 simap_put(&ofproto->backer->tnl_backers,
3043 dp_port_name, odp_to_u32(port_no));
3047 if (netdev_get_tunnel_config(netdev)) {
3048 sset_add(&ofproto->ghost_ports, devname);
3050 sset_add(&ofproto->ports, devname);
3056 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3059 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3066 sset_find_and_delete(&ofproto->ghost_ports,
3067 netdev_get_name(ofport->up.netdev));
3068 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3069 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3070 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3072 /* The caller is going to close ofport->up.netdev. If this is a
3073 * bonded port, then the bond is using that netdev, so remove it
3074 * from the bond. The client will need to reconfigure everything
3075 * after deleting ports, so then the slave will get re-added. */
3076 bundle_remove(&ofport->up);
3083 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3085 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3090 error = netdev_get_stats(ofport->up.netdev, stats);
3092 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3095 /* ofproto->stats.tx_packets represents packets that we created
3096 * internally and sent to some port (e.g. packets sent with
3097 * send_packet()). Account for them as if they had come from
3098 * OFPP_LOCAL and got forwarded. */
3100 if (stats->rx_packets != UINT64_MAX) {
3101 stats->rx_packets += ofproto->stats.tx_packets;
3104 if (stats->rx_bytes != UINT64_MAX) {
3105 stats->rx_bytes += ofproto->stats.tx_bytes;
3108 /* ofproto->stats.rx_packets represents packets that were received on
3109 * some port and we processed internally and dropped (e.g. STP).
3110 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3112 if (stats->tx_packets != UINT64_MAX) {
3113 stats->tx_packets += ofproto->stats.rx_packets;
3116 if (stats->tx_bytes != UINT64_MAX) {
3117 stats->tx_bytes += ofproto->stats.rx_bytes;
3124 struct port_dump_state {
3129 struct ofproto_port port;
3134 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3136 *statep = xzalloc(sizeof(struct port_dump_state));
3141 port_dump_next(const struct ofproto *ofproto_, void *state_,
3142 struct ofproto_port *port)
3144 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3145 struct port_dump_state *state = state_;
3146 const struct sset *sset;
3147 struct sset_node *node;
3149 if (state->has_port) {
3150 ofproto_port_destroy(&state->port);
3151 state->has_port = false;
3153 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3154 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3157 error = port_query_by_name(ofproto_, node->name, &state->port);
3159 *port = state->port;
3160 state->has_port = true;
3162 } else if (error != ENODEV) {
3167 if (!state->ghost) {
3168 state->ghost = true;
3171 return port_dump_next(ofproto_, state_, port);
3178 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3180 struct port_dump_state *state = state_;
3182 if (state->has_port) {
3183 ofproto_port_destroy(&state->port);
3190 port_poll(const struct ofproto *ofproto_, char **devnamep)
3192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3194 if (ofproto->port_poll_errno) {
3195 int error = ofproto->port_poll_errno;
3196 ofproto->port_poll_errno = 0;
3200 if (sset_is_empty(&ofproto->port_poll_set)) {
3204 *devnamep = sset_pop(&ofproto->port_poll_set);
3209 port_poll_wait(const struct ofproto *ofproto_)
3211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3212 dpif_port_poll_wait(ofproto->backer->dpif);
3216 port_is_lacp_current(const struct ofport *ofport_)
3218 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3219 return (ofport->bundle && ofport->bundle->lacp
3220 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3224 /* Upcall handling. */
3226 struct flow_miss_op {
3227 struct dpif_op dpif_op;
3229 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3230 struct xlate_out xout;
3231 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3233 struct ofpbuf mask; /* Flow mask for "put" ops. */
3234 struct odputil_keybuf maskbuf;
3236 /* If this is a "put" op, then a pointer to the subfacet that should
3237 * be marked as uninstalled if the operation fails. */
3238 struct subfacet *subfacet;
3241 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3242 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3243 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3244 * return value of true). However, for short flows the cost of bookkeeping is
3245 * much higher than the benefits, so when the datapath holds a large number of
3246 * flows we impose some heuristics to decide which flows are likely to be worth
3249 flow_miss_should_make_facet(struct flow_miss *miss)
3251 struct dpif_backer *backer = miss->ofproto->backer;
3254 switch (flow_miss_model) {
3255 case OFPROTO_HANDLE_MISS_AUTO:
3257 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3259 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3263 if (!backer->governor) {
3266 n_subfacets = hmap_count(&backer->subfacets);
3267 if (n_subfacets * 2 <= flow_eviction_threshold) {
3271 backer->governor = governor_create();
3274 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3275 return governor_should_install_flow(backer->governor, hash,
3276 list_size(&miss->packets));
3279 /* Handles 'miss', which matches 'facet'. May add any required datapath
3280 * operations to 'ops', incrementing '*n_ops' for each new op.
3282 * All of the packets in 'miss' are considered to have arrived at time
3283 * 'miss->stats.used'. This is really important only for new facets: if we
3284 * just called time_msec() here, then the new subfacet or its packets could
3285 * look (occasionally) as though it was used some time after the facet was
3286 * used. That can make a one-packet flow look like it has a nonzero duration,
3287 * which looks odd in e.g. NetFlow statistics. */
3289 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3290 struct flow_miss_op *ops, size_t *n_ops)
3292 enum subfacet_path want_path;
3293 struct subfacet *subfacet;
3295 facet->packet_count += miss->stats.n_packets;
3296 facet->prev_packet_count += miss->stats.n_packets;
3297 facet->byte_count += miss->stats.n_bytes;
3298 facet->prev_byte_count += miss->stats.n_bytes;
3300 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3302 /* Don't install the flow if it's the result of the "userspace"
3303 * action for an already installed facet. This can occur when a
3304 * datapath flow with wildcards has a "userspace" action and flows
3305 * sent to userspace result in a different subfacet, which will then
3306 * be rejected as overlapping by the datapath. */
3307 if (miss->upcall_type == DPIF_UC_ACTION
3308 && !list_is_empty(&facet->subfacets)) {
3312 subfacet = subfacet_create(facet, miss);
3313 if (subfacet->path != want_path) {
3314 struct flow_miss_op *op = &ops[(*n_ops)++];
3315 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3317 subfacet->path = want_path;
3319 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3320 if (enable_megaflows) {
3321 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3322 &miss->flow, UINT32_MAX);
3325 op->xout_garbage = false;
3326 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3327 op->subfacet = subfacet;
3328 put->flags = DPIF_FP_CREATE;
3329 put->key = miss->key;
3330 put->key_len = miss->key_len;
3331 put->mask = op->mask.data;
3332 put->mask_len = op->mask.size;
3334 if (want_path == SF_FAST_PATH) {
3335 put->actions = facet->xout.odp_actions.data;
3336 put->actions_len = facet->xout.odp_actions.size;
3338 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3339 op->slow_stub, sizeof op->slow_stub,
3340 &put->actions, &put->actions_len);
3346 /* Handles flow miss 'miss'. May add any required datapath operations
3347 * to 'ops', incrementing '*n_ops' for each new op. */
3349 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3352 struct facet *facet;
3354 miss->ofproto->n_missed += list_size(&miss->packets);
3356 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3358 /* There does not exist a bijection between 'struct flow' and datapath
3359 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3360 * assumption used throughout the facet and subfacet handling code.
3361 * Since we have to handle these misses in userspace anyway, we simply
3362 * skip facet creation, avoiding the problem altogether. */
3363 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3364 || !flow_miss_should_make_facet(miss)) {
3368 facet = facet_create(miss);
3370 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3373 static struct drop_key *
3374 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3377 struct drop_key *drop_key;
3379 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3380 &backer->drop_keys) {
3381 if (drop_key->key_len == key_len
3382 && !memcmp(drop_key->key, key, key_len)) {
3390 drop_key_clear(struct dpif_backer *backer)
3392 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3393 struct drop_key *drop_key, *next;
3395 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3398 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3400 if (error && !VLOG_DROP_WARN(&rl)) {
3401 struct ds ds = DS_EMPTY_INITIALIZER;
3402 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3403 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3404 ovs_strerror(error), ds_cstr(&ds));
3408 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3409 drop_key_destroy(drop_key);
3412 udpif_drop_key_clear(backer->udpif);
3416 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3418 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3419 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3420 struct flow_miss *miss;
3423 /* Process each element in the to-do list, constructing the set of
3424 * operations to batch. */
3426 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3427 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3429 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3431 /* Execute batch. */
3432 for (i = 0; i < n_ops; i++) {
3433 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3435 dpif_operate(backer->dpif, dpif_ops, n_ops);
3437 for (i = 0; i < n_ops; i++) {
3438 if (dpif_ops[i]->error != 0
3439 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3440 && flow_miss_ops[i].subfacet) {
3441 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3443 COVERAGE_INC(subfacet_install_fail);
3445 /* Zero-out subfacet counters when installation failed, but
3446 * datapath reported hits. This should not happen and
3447 * indicates a bug, since if the datapath flow exists, we
3448 * should not be attempting to create a new subfacet. A
3449 * buggy datapath could trigger this, so just zero out the
3450 * counters and log an error. */
3451 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3452 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3453 "datapath reported hits");
3454 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3457 subfacet->path = SF_NOT_INSTALLED;
3463 handle_sflow_upcall(struct dpif_backer *backer,
3464 const struct dpif_upcall *upcall)
3466 struct ofproto_dpif *ofproto;
3467 union user_action_cookie cookie;
3469 odp_port_t odp_in_port;
3471 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3472 &flow, NULL, &ofproto, &odp_in_port)
3473 || !ofproto->sflow) {
3477 memset(&cookie, 0, sizeof cookie);
3478 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3479 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3480 odp_in_port, &cookie);
3484 handle_flow_sample_upcall(struct dpif_backer *backer,
3485 const struct dpif_upcall *upcall)
3487 struct ofproto_dpif *ofproto;
3488 union user_action_cookie cookie;
3491 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3492 &flow, NULL, &ofproto, NULL)
3493 || !ofproto->ipfix) {
3497 memset(&cookie, 0, sizeof cookie);
3498 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3500 /* The flow reflects exactly the contents of the packet. Sample
3501 * the packet using it. */
3502 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3503 cookie.flow_sample.collector_set_id,
3504 cookie.flow_sample.probability,
3505 cookie.flow_sample.obs_domain_id,
3506 cookie.flow_sample.obs_point_id);
3510 handle_ipfix_upcall(struct dpif_backer *backer,
3511 const struct dpif_upcall *upcall)
3513 struct ofproto_dpif *ofproto;
3516 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3517 &flow, NULL, &ofproto, NULL)
3518 || !ofproto->ipfix) {
3522 /* The flow reflects exactly the contents of the packet. Sample
3523 * the packet using it. */
3524 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3528 handle_upcalls(struct dpif_backer *backer)
3530 struct flow_miss_batch *fmb;
3533 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3534 struct upcall *upcall = upcall_next(backer->udpif);
3540 switch (upcall->type) {
3542 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3545 case FLOW_SAMPLE_UPCALL:
3546 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3550 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3560 upcall_destroy(upcall);
3563 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3564 struct drop_key *drop_key = drop_key_next(backer->udpif);
3569 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3570 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3571 hash_bytes(drop_key->key, drop_key->key_len, 0));
3572 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3573 drop_key->key, drop_key->key_len,
3574 NULL, 0, NULL, 0, NULL);
3576 drop_key_destroy(drop_key);
3580 fmb = flow_miss_batch_next(backer->udpif);
3582 handle_flow_misses(backer, fmb);
3583 flow_miss_batch_destroy(fmb);
3587 /* Flow expiration. */
3589 static int subfacet_max_idle(const struct dpif_backer *);
3590 static void update_stats(struct dpif_backer *);
3591 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3592 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3594 /* This function is called periodically by run(). Its job is to collect
3595 * updates for the flows that have been installed into the datapath, most
3596 * importantly when they last were used, and then use that information to
3597 * expire flows that have not been used recently.
3599 * Returns the number of milliseconds after which it should be called again. */
3601 expire(struct dpif_backer *backer)
3603 struct ofproto_dpif *ofproto;
3607 /* Periodically clear out the drop keys in an effort to keep them
3608 * relatively few. */
3609 drop_key_clear(backer);
3611 /* Update stats for each flow in the backer. */
3612 update_stats(backer);
3614 n_subfacets = hmap_count(&backer->subfacets);
3616 struct subfacet *subfacet;
3617 long long int total, now;
3621 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3622 total += now - subfacet->created;
3624 backer->avg_subfacet_life += total / n_subfacets;
3626 backer->avg_subfacet_life /= 2;
3628 backer->avg_n_subfacet += n_subfacets;
3629 backer->avg_n_subfacet /= 2;
3631 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3633 max_idle = subfacet_max_idle(backer);
3634 expire_subfacets(backer, max_idle);
3636 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3637 struct rule *rule, *next_rule;
3639 if (ofproto->backer != backer) {
3643 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3645 ovs_mutex_lock(&ofproto_mutex);
3646 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3647 &ofproto->up.expirable) {
3648 rule_expire(rule_dpif_cast(rule));
3650 ovs_mutex_unlock(&ofproto_mutex);
3652 /* All outstanding data in existing flows has been accounted, so it's a
3653 * good time to do bond rebalancing. */
3654 if (ofproto->has_bonded_bundles) {
3655 struct ofbundle *bundle;
3657 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3659 bond_rebalance(bundle->bond);
3665 return MIN(max_idle, 1000);
3668 /* Updates flow table statistics given that the datapath just reported 'stats'
3669 * as 'subfacet''s statistics. */
3671 update_subfacet_stats(struct subfacet *subfacet,
3672 const struct dpif_flow_stats *stats)
3674 struct facet *facet = subfacet->facet;
3675 struct dpif_flow_stats diff;
3677 diff.tcp_flags = stats->tcp_flags;
3678 diff.used = stats->used;
3680 if (stats->n_packets >= subfacet->dp_packet_count) {
3681 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3683 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3687 if (stats->n_bytes >= subfacet->dp_byte_count) {
3688 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3690 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3694 facet->ofproto->n_hit += diff.n_packets;
3695 subfacet->dp_packet_count = stats->n_packets;
3696 subfacet->dp_byte_count = stats->n_bytes;
3697 subfacet_update_stats(subfacet, &diff);
3699 if (facet->accounted_bytes < facet->byte_count) {
3701 facet_account(facet);
3702 facet->accounted_bytes = facet->byte_count;
3706 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3707 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3709 delete_unexpected_flow(struct dpif_backer *backer,
3710 const struct nlattr *key, size_t key_len)
3712 if (!VLOG_DROP_WARN(&rl)) {
3716 odp_flow_key_format(key, key_len, &s);
3717 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3721 COVERAGE_INC(facet_unexpected);
3722 dpif_flow_del(backer->dpif, key, key_len, NULL);
3725 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3727 * This function also pushes statistics updates to rules which each facet
3728 * resubmits into. Generally these statistics will be accurate. However, if a
3729 * facet changes the rule it resubmits into at some time in between
3730 * update_stats() runs, it is possible that statistics accrued to the
3731 * old rule will be incorrectly attributed to the new rule. This could be
3732 * avoided by calling update_stats() whenever rules are created or
3733 * deleted. However, the performance impact of making so many calls to the
3734 * datapath do not justify the benefit of having perfectly accurate statistics.
3736 * In addition, this function maintains per ofproto flow hit counts. The patch
3737 * port is not treated specially. e.g. A packet ingress from br0 patched into
3738 * br1 will increase the hit count of br0 by 1, however, does not affect
3739 * the hit or miss counts of br1.
3742 update_stats(struct dpif_backer *backer)
3744 const struct dpif_flow_stats *stats;
3745 struct dpif_flow_dump dump;
3746 const struct nlattr *key, *mask;
3747 size_t key_len, mask_len;
3749 dpif_flow_dump_start(&dump, backer->dpif);
3750 while (dpif_flow_dump_next(&dump, &key, &key_len,
3751 &mask, &mask_len, NULL, NULL, &stats)) {
3752 struct subfacet *subfacet;
3755 key_hash = odp_flow_key_hash(key, key_len);
3756 subfacet = subfacet_find(backer, key, key_len, key_hash);
3757 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3759 update_subfacet_stats(subfacet, stats);
3763 /* Stats are updated per-packet. */
3766 case SF_NOT_INSTALLED:
3768 delete_unexpected_flow(backer, key, key_len);
3773 dpif_flow_dump_done(&dump);
3775 update_moving_averages(backer);
3778 /* Calculates and returns the number of milliseconds of idle time after which
3779 * subfacets should expire from the datapath. When a subfacet expires, we fold
3780 * its statistics into its facet, and when a facet's last subfacet expires, we
3781 * fold its statistic into its rule. */
3783 subfacet_max_idle(const struct dpif_backer *backer)
3786 * Idle time histogram.
3788 * Most of the time a switch has a relatively small number of subfacets.
3789 * When this is the case we might as well keep statistics for all of them
3790 * in userspace and to cache them in the kernel datapath for performance as
3793 * As the number of subfacets increases, the memory required to maintain
3794 * statistics about them in userspace and in the kernel becomes
3795 * significant. However, with a large number of subfacets it is likely
3796 * that only a few of them are "heavy hitters" that consume a large amount
3797 * of bandwidth. At this point, only heavy hitters are worth caching in
3798 * the kernel and maintaining in userspaces; other subfacets we can
3801 * The technique used to compute the idle time is to build a histogram with
3802 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3803 * that is installed in the kernel gets dropped in the appropriate bucket.
3804 * After the histogram has been built, we compute the cutoff so that only
3805 * the most-recently-used 1% of subfacets (but at least
3806 * flow_eviction_threshold flows) are kept cached. At least
3807 * the most-recently-used bucket of subfacets is kept, so actually an
3808 * arbitrary number of subfacets can be kept in any given expiration run
3809 * (though the next run will delete most of those unless they receive
3812 * This requires a second pass through the subfacets, in addition to the
3813 * pass made by update_stats(), because the former function never looks at
3814 * uninstallable subfacets.
3816 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3817 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3818 int buckets[N_BUCKETS] = { 0 };
3819 int total, subtotal, bucket;
3820 struct subfacet *subfacet;
3824 total = hmap_count(&backer->subfacets);
3825 if (total <= flow_eviction_threshold) {
3826 return N_BUCKETS * BUCKET_WIDTH;
3829 /* Build histogram. */
3831 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3832 long long int idle = now - subfacet->used;
3833 int bucket = (idle <= 0 ? 0
3834 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3835 : (unsigned int) idle / BUCKET_WIDTH);
3839 /* Find the first bucket whose flows should be expired. */
3840 subtotal = bucket = 0;
3842 subtotal += buckets[bucket++];
3843 } while (bucket < N_BUCKETS &&
3844 subtotal < MAX(flow_eviction_threshold, total / 100));
3846 if (VLOG_IS_DBG_ENABLED()) {
3850 ds_put_cstr(&s, "keep");
3851 for (i = 0; i < N_BUCKETS; i++) {
3853 ds_put_cstr(&s, ", drop");
3856 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3859 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3863 return bucket * BUCKET_WIDTH;
3867 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3869 /* Cutoff time for most flows. */
3870 long long int normal_cutoff = time_msec() - dp_max_idle;
3872 /* We really want to keep flows for special protocols around, so use a more
3873 * conservative cutoff. */
3874 long long int special_cutoff = time_msec() - 10000;
3876 struct subfacet *subfacet, *next_subfacet;
3877 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3881 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3882 &backer->subfacets) {
3883 long long int cutoff;
3885 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3889 if (subfacet->used < cutoff) {
3890 if (subfacet->path != SF_NOT_INSTALLED) {
3891 batch[n_batch++] = subfacet;
3892 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3893 subfacet_destroy_batch(backer, batch, n_batch);
3897 subfacet_destroy(subfacet);
3903 subfacet_destroy_batch(backer, batch, n_batch);
3907 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3908 * then delete it entirely. */
3910 rule_expire(struct rule_dpif *rule)
3911 OVS_REQUIRES(ofproto_mutex)
3913 uint16_t idle_timeout, hard_timeout;
3914 long long int now = time_msec();
3917 ovs_assert(!rule->up.pending);
3919 /* Has 'rule' expired? */
3920 ovs_mutex_lock(&rule->up.mutex);
3921 hard_timeout = rule->up.hard_timeout;
3922 idle_timeout = rule->up.idle_timeout;
3923 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3924 reason = OFPRR_HARD_TIMEOUT;
3925 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3926 reason = OFPRR_IDLE_TIMEOUT;
3930 ovs_mutex_unlock(&rule->up.mutex);
3933 COVERAGE_INC(ofproto_dpif_expired);
3934 ofproto_rule_expire(&rule->up, reason);
3940 /* Creates and returns a new facet based on 'miss'.
3942 * The caller must already have determined that no facet with an identical
3943 * 'miss->flow' exists in 'miss->ofproto'.
3945 * 'rule' and 'xout' must have been created based on 'miss'.
3947 * 'facet'' statistics are initialized based on 'stats'.
3949 * The facet will initially have no subfacets. The caller should create (at
3950 * least) one subfacet with subfacet_create(). */
3951 static struct facet *
3952 facet_create(const struct flow_miss *miss)
3954 struct ofproto_dpif *ofproto = miss->ofproto;
3955 struct facet *facet;
3958 facet = xzalloc(sizeof *facet);
3959 facet->ofproto = miss->ofproto;
3960 facet->used = miss->stats.used;
3961 facet->flow = miss->flow;
3962 facet->learn_rl = time_msec() + 500;
3964 list_init(&facet->subfacets);
3965 netflow_flow_init(&facet->nf_flow);
3966 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3968 xlate_out_copy(&facet->xout, &miss->xout);
3970 match_init(&match, &facet->flow, &facet->xout.wc);
3971 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3972 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3973 classifier_insert(&ofproto->facets, &facet->cr);
3974 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3976 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3981 facet_free(struct facet *facet)
3984 xlate_out_uninit(&facet->xout);
3989 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3990 * 'packet', which arrived on 'in_port'. */
3992 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3993 const struct nlattr *odp_actions, size_t actions_len,
3994 struct ofpbuf *packet)
3996 struct odputil_keybuf keybuf;
4000 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4001 odp_flow_key_from_flow(&key, flow,
4002 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4004 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4005 odp_actions, actions_len, packet);
4009 /* Remove 'facet' from its ofproto and free up the associated memory:
4011 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4012 * rule's statistics, via subfacet_uninstall().
4014 * - Removes 'facet' from its rule and from ofproto->facets.
4017 facet_remove(struct facet *facet)
4019 struct subfacet *subfacet, *next_subfacet;
4021 ovs_assert(!list_is_empty(&facet->subfacets));
4023 /* First uninstall all of the subfacets to get final statistics. */
4024 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4025 subfacet_uninstall(subfacet);
4028 /* Flush the final stats to the rule.
4030 * This might require us to have at least one subfacet around so that we
4031 * can use its actions for accounting in facet_account(), which is why we
4032 * have uninstalled but not yet destroyed the subfacets. */
4033 facet_flush_stats(facet);
4035 /* Now we're really all done so destroy everything. */
4036 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4037 &facet->subfacets) {
4038 subfacet_destroy__(subfacet);
4040 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4041 classifier_remove(&facet->ofproto->facets, &facet->cr);
4042 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4043 cls_rule_destroy(&facet->cr);
4047 /* Feed information from 'facet' back into the learning table to keep it in
4048 * sync with what is actually flowing through the datapath. */
4050 facet_learn(struct facet *facet)
4052 long long int now = time_msec();
4054 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4058 facet->learn_rl = now + 500;
4060 if (!facet->xout.has_learn
4061 && !facet->xout.has_normal
4062 && (!facet->xout.has_fin_timeout
4063 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4067 facet_push_stats(facet, true);
4071 facet_account(struct facet *facet)
4073 const struct nlattr *a;
4078 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4081 n_bytes = facet->byte_count - facet->accounted_bytes;
4083 /* This loop feeds byte counters to bond_account() for rebalancing to use
4084 * as a basis. We also need to track the actual VLAN on which the packet
4085 * is going to be sent to ensure that it matches the one passed to
4086 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4089 * We use the actions from an arbitrary subfacet because they should all
4090 * be equally valid for our purpose. */
4091 vlan_tci = facet->flow.vlan_tci;
4092 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4093 facet->xout.odp_actions.size) {
4094 const struct ovs_action_push_vlan *vlan;
4095 struct ofport_dpif *port;
4097 switch (nl_attr_type(a)) {
4098 case OVS_ACTION_ATTR_OUTPUT:
4099 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4100 if (port && port->bundle && port->bundle->bond) {
4101 bond_account(port->bundle->bond, &facet->flow,
4102 vlan_tci_to_vid(vlan_tci), n_bytes);
4106 case OVS_ACTION_ATTR_POP_VLAN:
4107 vlan_tci = htons(0);
4110 case OVS_ACTION_ATTR_PUSH_VLAN:
4111 vlan = nl_attr_get(a);
4112 vlan_tci = vlan->vlan_tci;
4118 /* Returns true if the only action for 'facet' is to send to the controller.
4119 * (We don't report NetFlow expiration messages for such facets because they
4120 * are just part of the control logic for the network, not real traffic). */
4122 facet_is_controller_flow(struct facet *facet)
4125 struct ofproto_dpif *ofproto = facet->ofproto;
4126 const struct ofpact *ofpacts;
4127 struct rule_actions *actions;
4128 struct rule_dpif *rule;
4132 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4133 actions = rule_dpif_get_actions(rule);
4134 rule_dpif_unref(rule);
4136 ofpacts_len = actions->ofpacts_len;
4137 ofpacts = actions->ofpacts;
4138 is_controller = ofpacts_len > 0
4139 && ofpacts->type == OFPACT_CONTROLLER
4140 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4141 rule_actions_unref(actions);
4143 return is_controller;
4148 /* Folds all of 'facet''s statistics into its rule. Also updates the
4149 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4150 * 'facet''s statistics in the datapath should have been zeroed and folded into
4151 * its packet and byte counts before this function is called. */
4153 facet_flush_stats(struct facet *facet)
4155 struct ofproto_dpif *ofproto = facet->ofproto;
4156 struct subfacet *subfacet;
4158 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4159 ovs_assert(!subfacet->dp_byte_count);
4160 ovs_assert(!subfacet->dp_packet_count);
4163 facet_push_stats(facet, false);
4164 if (facet->accounted_bytes < facet->byte_count) {
4165 facet_account(facet);
4166 facet->accounted_bytes = facet->byte_count;
4169 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4170 struct ofexpired expired;
4171 expired.flow = facet->flow;
4172 expired.packet_count = facet->packet_count;
4173 expired.byte_count = facet->byte_count;
4174 expired.used = facet->used;
4175 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4178 /* Reset counters to prevent double counting if 'facet' ever gets
4180 facet_reset_counters(facet);
4182 netflow_flow_clear(&facet->nf_flow);
4183 facet->tcp_flags = 0;
4186 /* Searches 'ofproto''s table of facets for one which would be responsible for
4187 * 'flow'. Returns it if found, otherwise a null pointer.
4189 * The returned facet might need revalidation; use facet_lookup_valid()
4190 * instead if that is important. */
4191 static struct facet *
4192 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4194 struct cls_rule *cr;
4196 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4197 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4198 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4199 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4202 /* Searches 'ofproto''s table of facets for one capable that covers
4203 * 'flow'. Returns it if found, otherwise a null pointer.
4205 * The returned facet is guaranteed to be valid. */
4206 static struct facet *
4207 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4209 struct facet *facet;
4211 facet = facet_find(ofproto, flow);
4213 && ofproto->backer->need_revalidate
4214 && !facet_revalidate(facet)) {
4222 facet_check_consistency(struct facet *facet)
4224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4226 struct xlate_out xout;
4227 struct xlate_in xin;
4229 struct rule_dpif *rule;
4232 /* Check the datapath actions for consistency. */
4233 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4234 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4235 xlate_actions(&xin, &xout);
4236 rule_dpif_unref(rule);
4238 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4239 && facet->xout.slow == xout.slow;
4240 if (!ok && !VLOG_DROP_WARN(&rl)) {
4241 struct ds s = DS_EMPTY_INITIALIZER;
4243 flow_format(&s, &facet->flow);
4244 ds_put_cstr(&s, ": inconsistency in facet");
4246 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4247 ds_put_cstr(&s, " (actions were: ");
4248 format_odp_actions(&s, facet->xout.odp_actions.data,
4249 facet->xout.odp_actions.size);
4250 ds_put_cstr(&s, ") (correct actions: ");
4251 format_odp_actions(&s, xout.odp_actions.data,
4252 xout.odp_actions.size);
4253 ds_put_char(&s, ')');
4256 if (facet->xout.slow != xout.slow) {
4257 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4262 xlate_out_uninit(&xout);
4267 /* Re-searches the classifier for 'facet':
4269 * - If the rule found is different from 'facet''s current rule, moves
4270 * 'facet' to the new rule and recompiles its actions.
4272 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4273 * where it is and recompiles its actions anyway.
4275 * - If any of 'facet''s subfacets correspond to a new flow according to
4276 * xlate_receive(), 'facet' is removed.
4278 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4280 facet_revalidate(struct facet *facet)
4282 struct ofproto_dpif *ofproto = facet->ofproto;
4283 struct rule_dpif *new_rule;
4284 struct subfacet *subfacet;
4285 struct flow_wildcards wc;
4286 struct xlate_out xout;
4287 struct xlate_in xin;
4289 COVERAGE_INC(facet_revalidate);
4291 /* Check that child subfacets still correspond to this facet. Tunnel
4292 * configuration changes could cause a subfacet's OpenFlow in_port to
4294 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4295 struct ofproto_dpif *recv_ofproto;
4296 struct flow recv_flow;
4299 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4300 subfacet->key_len, &recv_flow, NULL,
4301 &recv_ofproto, NULL);
4303 || recv_ofproto != ofproto
4304 || facet != facet_find(ofproto, &recv_flow)) {
4305 facet_remove(facet);
4310 flow_wildcards_init_catchall(&wc);
4311 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4313 /* Calculate new datapath actions.
4315 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4316 * emit a NetFlow expiration and, if so, we need to have the old state
4317 * around to properly compose it. */
4318 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4319 xlate_actions(&xin, &xout);
4320 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4322 /* A facet's slow path reason should only change under dramatic
4323 * circumstances. Rather than try to update everything, it's simpler to
4324 * remove the facet and start over.
4326 * More importantly, if a facet's wildcards change, it will be relatively
4327 * difficult to figure out if its subfacets still belong to it, and if not
4328 * which facet they may belong to. Again, to avoid the complexity, we
4329 * simply give up instead. */
4330 if (facet->xout.slow != xout.slow
4331 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4332 facet_remove(facet);
4333 xlate_out_uninit(&xout);
4334 rule_dpif_unref(new_rule);
4338 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4339 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4340 if (subfacet->path == SF_FAST_PATH) {
4341 struct dpif_flow_stats stats;
4343 subfacet_install(subfacet, &xout.odp_actions, &stats);
4344 subfacet_update_stats(subfacet, &stats);
4348 facet_flush_stats(facet);
4350 ofpbuf_clear(&facet->xout.odp_actions);
4351 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4352 xout.odp_actions.size);
4355 /* Update 'facet' now that we've taken care of all the old state. */
4356 facet->xout.slow = xout.slow;
4357 facet->xout.has_learn = xout.has_learn;
4358 facet->xout.has_normal = xout.has_normal;
4359 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4360 facet->xout.nf_output_iface = xout.nf_output_iface;
4361 facet->xout.mirrors = xout.mirrors;
4362 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4364 ovs_mutex_lock(&new_rule->up.mutex);
4365 facet->used = MAX(facet->used, new_rule->up.created);
4366 ovs_mutex_unlock(&new_rule->up.mutex);
4368 xlate_out_uninit(&xout);
4369 rule_dpif_unref(new_rule);
4374 facet_reset_counters(struct facet *facet)
4376 facet->packet_count = 0;
4377 facet->byte_count = 0;
4378 facet->prev_packet_count = 0;
4379 facet->prev_byte_count = 0;
4380 facet->accounted_bytes = 0;
4384 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4385 struct dpif_flow_stats *stats, bool may_learn)
4387 struct ofport_dpif *in_port;
4388 struct rule_dpif *rule;
4389 struct xlate_in xin;
4391 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4392 if (in_port && in_port->is_tunnel) {
4393 netdev_vport_inc_rx(in_port->up.netdev, stats);
4396 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4397 rule_dpif_credit_stats(rule, stats);
4398 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4399 xin.resubmit_stats = stats;
4400 xin.may_learn = may_learn;
4401 xlate_actions_for_side_effects(&xin);
4402 rule_dpif_unref(rule);
4406 facet_push_stats(struct facet *facet, bool may_learn)
4408 struct dpif_flow_stats stats;
4410 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4411 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4412 ovs_assert(facet->used >= facet->prev_used);
4414 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4415 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4416 stats.used = facet->used;
4417 stats.tcp_flags = facet->tcp_flags;
4419 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4420 facet->prev_packet_count = facet->packet_count;
4421 facet->prev_byte_count = facet->byte_count;
4422 facet->prev_used = facet->used;
4424 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4426 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4427 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4428 stats.n_packets, stats.n_bytes);
4429 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4434 push_all_stats__(bool run_fast)
4436 static long long int rl = LLONG_MIN;
4437 struct ofproto_dpif *ofproto;
4439 if (time_msec() < rl) {
4443 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4444 struct cls_cursor cursor;
4445 struct facet *facet;
4447 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4448 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4449 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4450 facet_push_stats(facet, false);
4455 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4458 rl = time_msec() + 100;
4462 push_all_stats(void)
4464 push_all_stats__(true);
4468 rule_dpif_credit_stats(struct rule_dpif *rule,
4469 const struct dpif_flow_stats *stats)
4471 ovs_mutex_lock(&rule->stats_mutex);
4472 rule->packet_count += stats->n_packets;
4473 rule->byte_count += stats->n_bytes;
4474 rule->up.used = MAX(rule->up.used, stats->used);
4475 ovs_mutex_unlock(&rule->stats_mutex);
4479 rule_dpif_fail_open(const struct rule_dpif *rule)
4481 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4485 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4486 OVS_REQUIRES(rule->up.mutex)
4488 return rule->up.flow_cookie;
4492 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4493 uint16_t hard_timeout)
4495 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4498 /* Returns 'rule''s actions. The caller owns a reference on the returned
4499 * actions and must eventually release it (with rule_actions_unref()) to avoid
4501 struct rule_actions *
4502 rule_dpif_get_actions(const struct rule_dpif *rule)
4504 return rule_get_actions(&rule->up);
4509 static struct subfacet *
4510 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4511 size_t key_len, uint32_t key_hash)
4513 struct subfacet *subfacet;
4515 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4516 &backer->subfacets) {
4517 if (subfacet->key_len == key_len
4518 && !memcmp(key, subfacet->key, key_len)) {
4526 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4527 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4528 * existing subfacet if there is one, otherwise creates and returns a
4530 static struct subfacet *
4531 subfacet_create(struct facet *facet, struct flow_miss *miss)
4533 struct dpif_backer *backer = miss->ofproto->backer;
4534 enum odp_key_fitness key_fitness = miss->key_fitness;
4535 const struct nlattr *key = miss->key;
4536 size_t key_len = miss->key_len;
4538 struct subfacet *subfacet;
4540 key_hash = odp_flow_key_hash(key, key_len);
4542 if (list_is_empty(&facet->subfacets)) {
4543 subfacet = &facet->one_subfacet;
4545 subfacet = subfacet_find(backer, key, key_len, key_hash);
4547 if (subfacet->facet == facet) {
4551 /* This shouldn't happen. */
4552 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4553 subfacet_destroy(subfacet);
4556 subfacet = xmalloc(sizeof *subfacet);
4559 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4560 list_push_back(&facet->subfacets, &subfacet->list_node);
4561 subfacet->facet = facet;
4562 subfacet->key_fitness = key_fitness;
4563 subfacet->key = xmemdup(key, key_len);
4564 subfacet->key_len = key_len;
4565 subfacet->used = miss->stats.used;
4566 subfacet->created = subfacet->used;
4567 subfacet->dp_packet_count = 0;
4568 subfacet->dp_byte_count = 0;
4569 subfacet->path = SF_NOT_INSTALLED;
4570 subfacet->backer = backer;
4572 backer->subfacet_add_count++;
4576 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4577 * its facet within 'ofproto', and frees it. */
4579 subfacet_destroy__(struct subfacet *subfacet)
4581 struct facet *facet = subfacet->facet;
4582 struct ofproto_dpif *ofproto = facet->ofproto;
4584 /* Update ofproto stats before uninstall the subfacet. */
4585 ofproto->backer->subfacet_del_count++;
4587 subfacet_uninstall(subfacet);
4588 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4589 list_remove(&subfacet->list_node);
4590 free(subfacet->key);
4591 if (subfacet != &facet->one_subfacet) {
4596 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4597 * last remaining subfacet in its facet destroys the facet too. */
4599 subfacet_destroy(struct subfacet *subfacet)
4601 struct facet *facet = subfacet->facet;
4603 if (list_is_singleton(&facet->subfacets)) {
4604 /* facet_remove() needs at least one subfacet (it will remove it). */
4605 facet_remove(facet);
4607 subfacet_destroy__(subfacet);
4612 subfacet_destroy_batch(struct dpif_backer *backer,
4613 struct subfacet **subfacets, int n)
4615 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4616 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4617 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4620 for (i = 0; i < n; i++) {
4621 ops[i].type = DPIF_OP_FLOW_DEL;
4622 ops[i].u.flow_del.key = subfacets[i]->key;
4623 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4624 ops[i].u.flow_del.stats = &stats[i];
4628 dpif_operate(backer->dpif, opsp, n);
4629 for (i = 0; i < n; i++) {
4630 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4631 subfacets[i]->path = SF_NOT_INSTALLED;
4632 subfacet_destroy(subfacets[i]);
4637 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4638 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4639 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4640 * since 'subfacet' was last updated.
4642 * Returns 0 if successful, otherwise a positive errno value. */
4644 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4645 struct dpif_flow_stats *stats)
4647 struct facet *facet = subfacet->facet;
4648 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4649 const struct nlattr *actions = odp_actions->data;
4650 size_t actions_len = odp_actions->size;
4651 struct odputil_keybuf maskbuf;
4654 uint64_t slow_path_stub[128 / 8];
4655 enum dpif_flow_put_flags flags;
4658 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4661 flags |= DPIF_FP_ZERO_STATS;
4664 if (path == SF_SLOW_PATH) {
4665 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4666 slow_path_stub, sizeof slow_path_stub,
4667 &actions, &actions_len);
4670 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4671 if (enable_megaflows) {
4672 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4673 &facet->flow, UINT32_MAX);
4676 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4677 subfacet->key_len, mask.data, mask.size,
4678 actions, actions_len, stats);
4681 subfacet_reset_dp_stats(subfacet, stats);
4685 COVERAGE_INC(subfacet_install_fail);
4687 subfacet->path = path;
4692 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4694 subfacet_uninstall(struct subfacet *subfacet)
4696 if (subfacet->path != SF_NOT_INSTALLED) {
4697 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4698 struct dpif_flow_stats stats;
4701 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4702 subfacet->key_len, &stats);
4703 subfacet_reset_dp_stats(subfacet, &stats);
4705 subfacet_update_stats(subfacet, &stats);
4707 subfacet->path = SF_NOT_INSTALLED;
4709 ovs_assert(subfacet->dp_packet_count == 0);
4710 ovs_assert(subfacet->dp_byte_count == 0);
4714 /* Resets 'subfacet''s datapath statistics counters. This should be called
4715 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4716 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4717 * was reset in the datapath. 'stats' will be modified to include only
4718 * statistics new since 'subfacet' was last updated. */
4720 subfacet_reset_dp_stats(struct subfacet *subfacet,
4721 struct dpif_flow_stats *stats)
4724 && subfacet->dp_packet_count <= stats->n_packets
4725 && subfacet->dp_byte_count <= stats->n_bytes) {
4726 stats->n_packets -= subfacet->dp_packet_count;
4727 stats->n_bytes -= subfacet->dp_byte_count;
4730 subfacet->dp_packet_count = 0;
4731 subfacet->dp_byte_count = 0;
4734 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4736 * Because of the meaning of a subfacet's counters, it only makes sense to do
4737 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4738 * represents a packet that was sent by hand or if it represents statistics
4739 * that have been cleared out of the datapath. */
4741 subfacet_update_stats(struct subfacet *subfacet,
4742 const struct dpif_flow_stats *stats)
4744 if (stats->n_packets || stats->used > subfacet->used) {
4745 struct facet *facet = subfacet->facet;
4747 subfacet->used = MAX(subfacet->used, stats->used);
4748 facet->used = MAX(facet->used, stats->used);
4749 facet->packet_count += stats->n_packets;
4750 facet->byte_count += stats->n_bytes;
4751 facet->tcp_flags |= stats->tcp_flags;
4757 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4758 * the fields that were relevant as part of the lookup. */
4760 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4761 struct flow_wildcards *wc, struct rule_dpif **rule)
4763 struct ofport_dpif *port;
4765 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4768 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4770 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4771 flow->in_port.ofp_port);
4774 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4775 ofproto->no_packet_in_rule, rule);
4779 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4780 const struct flow *flow, struct flow_wildcards *wc,
4781 uint8_t table_id, struct rule_dpif **rule)
4783 const struct cls_rule *cls_rule;
4784 struct classifier *cls;
4788 if (table_id >= N_TABLES) {
4793 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4794 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4797 cls = &ofproto->up.tables[table_id].cls;
4798 ovs_rwlock_rdlock(&cls->rwlock);
4799 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4800 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4801 /* We must pretend that transport ports are unavailable. */
4802 struct flow ofpc_normal_flow = *flow;
4803 ofpc_normal_flow.tp_src = htons(0);
4804 ofpc_normal_flow.tp_dst = htons(0);
4805 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4806 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4807 cls_rule = &ofproto->drop_frags_rule->up.cr;
4809 flow_wildcards_init_exact(wc);
4812 cls_rule = classifier_lookup(cls, flow, wc);
4815 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4816 rule_dpif_ref(*rule);
4817 ovs_rwlock_unlock(&cls->rwlock);
4819 return *rule != NULL;
4822 /* Given a port configuration (specified as zero if there's no port), chooses
4823 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4824 * flow table miss. */
4826 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4827 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4829 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4830 rule_dpif_ref(*rule);
4834 rule_dpif_ref(struct rule_dpif *rule)
4837 ofproto_rule_ref(&rule->up);
4842 rule_dpif_unref(struct rule_dpif *rule)
4845 ofproto_rule_unref(&rule->up);
4850 complete_operation(struct rule_dpif *rule)
4851 OVS_REQUIRES(ofproto_mutex)
4853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4855 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4856 ofoperation_complete(rule->up.pending, 0);
4859 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4861 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4864 static struct rule *
4867 struct rule_dpif *rule = xmalloc(sizeof *rule);
4872 rule_dealloc(struct rule *rule_)
4874 struct rule_dpif *rule = rule_dpif_cast(rule_);
4879 rule_construct(struct rule *rule_)
4881 struct rule_dpif *rule = rule_dpif_cast(rule_);
4882 ovs_mutex_init(&rule->stats_mutex);
4883 ovs_mutex_lock(&rule->stats_mutex);
4884 rule->packet_count = 0;
4885 rule->byte_count = 0;
4886 ovs_mutex_unlock(&rule->stats_mutex);
4891 rule_insert(struct rule *rule_)
4892 OVS_REQUIRES(ofproto_mutex)
4894 struct rule_dpif *rule = rule_dpif_cast(rule_);
4895 complete_operation(rule);
4899 rule_delete(struct rule *rule_)
4900 OVS_REQUIRES(ofproto_mutex)
4902 struct rule_dpif *rule = rule_dpif_cast(rule_);
4903 complete_operation(rule);
4907 rule_destruct(struct rule *rule_)
4909 struct rule_dpif *rule = rule_dpif_cast(rule_);
4910 ovs_mutex_destroy(&rule->stats_mutex);
4914 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4916 struct rule_dpif *rule = rule_dpif_cast(rule_);
4918 /* push_all_stats() can handle flow misses which, when using the learn
4919 * action, can cause rules to be added and deleted. This can corrupt our
4920 * caller's datastructures which assume that rule_get_stats() doesn't have
4921 * an impact on the flow table. To be safe, we disable miss handling. */
4922 push_all_stats__(false);
4924 /* Start from historical data for 'rule' itself that are no longer tracked
4925 * in facets. This counts, for example, facets that have expired. */
4926 ovs_mutex_lock(&rule->stats_mutex);
4927 *packets = rule->packet_count;
4928 *bytes = rule->byte_count;
4929 ovs_mutex_unlock(&rule->stats_mutex);
4933 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4934 struct ofpbuf *packet)
4936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4937 struct dpif_flow_stats stats;
4938 struct xlate_out xout;
4939 struct xlate_in xin;
4941 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4942 rule_dpif_credit_stats(rule, &stats);
4944 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4945 xin.resubmit_stats = &stats;
4946 xlate_actions(&xin, &xout);
4948 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4949 xout.odp_actions.size, packet);
4951 xlate_out_uninit(&xout);
4955 rule_execute(struct rule *rule, const struct flow *flow,
4956 struct ofpbuf *packet)
4958 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4959 ofpbuf_delete(packet);
4964 rule_modify_actions(struct rule *rule_, bool reset_counters)
4965 OVS_REQUIRES(ofproto_mutex)
4967 struct rule_dpif *rule = rule_dpif_cast(rule_);
4969 if (reset_counters) {
4970 ovs_mutex_lock(&rule->stats_mutex);
4971 rule->packet_count = 0;
4972 rule->byte_count = 0;
4973 ovs_mutex_unlock(&rule->stats_mutex);
4976 complete_operation(rule);
4979 /* Sends 'packet' out 'ofport'.
4980 * May modify 'packet'.
4981 * Returns 0 if successful, otherwise a positive errno value. */
4983 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4986 uint64_t odp_actions_stub[1024 / 8];
4987 struct ofpbuf key, odp_actions;
4988 struct dpif_flow_stats stats;
4989 struct odputil_keybuf keybuf;
4990 struct ofpact_output output;
4991 struct xlate_out xout;
4992 struct xlate_in xin;
4994 union flow_in_port in_port_;
4997 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4998 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5000 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5001 in_port_.ofp_port = OFPP_NONE;
5002 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5003 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5005 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5007 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5008 output.port = ofport->up.ofp_port;
5011 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5012 xin.ofpacts_len = sizeof output;
5013 xin.ofpacts = &output.ofpact;
5014 xin.resubmit_stats = &stats;
5015 xlate_actions(&xin, &xout);
5017 error = dpif_execute(ofproto->backer->dpif,
5019 xout.odp_actions.data, xout.odp_actions.size,
5021 xlate_out_uninit(&xout);
5024 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5025 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5026 ovs_strerror(error));
5029 ofproto->stats.tx_packets++;
5030 ofproto->stats.tx_bytes += packet->size;
5034 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5035 * The action will state 'slow' as the reason that the action is in the slow
5036 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5037 * dump-flows" output to see why a flow is in the slow path.)
5039 * The 'stub_size' bytes in 'stub' will be used to store the action.
5040 * 'stub_size' must be large enough for the action.
5042 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5045 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5046 enum slow_path_reason slow,
5047 uint64_t *stub, size_t stub_size,
5048 const struct nlattr **actionsp, size_t *actions_lenp)
5050 union user_action_cookie cookie;
5053 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5054 cookie.slow_path.unused = 0;
5055 cookie.slow_path.reason = slow;
5057 ofpbuf_use_stack(&buf, stub, stub_size);
5058 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5059 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5061 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5063 odp_port_t odp_port;
5066 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5067 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5068 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5070 *actionsp = buf.data;
5071 *actions_lenp = buf.size;
5075 set_frag_handling(struct ofproto *ofproto_,
5076 enum ofp_config_flags frag_handling)
5078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5079 if (frag_handling != OFPC_FRAG_REASM) {
5080 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5088 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5089 const struct flow *flow,
5090 const struct ofpact *ofpacts, size_t ofpacts_len)
5092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5093 struct odputil_keybuf keybuf;
5094 struct dpif_flow_stats stats;
5095 struct xlate_out xout;
5096 struct xlate_in xin;
5100 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5101 odp_flow_key_from_flow(&key, flow,
5102 ofp_port_to_odp_port(ofproto,
5103 flow->in_port.ofp_port));
5105 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5107 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5108 xin.resubmit_stats = &stats;
5109 xin.ofpacts_len = ofpacts_len;
5110 xin.ofpacts = ofpacts;
5112 xlate_actions(&xin, &xout);
5113 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5114 xout.odp_actions.data, xout.odp_actions.size, packet);
5115 xlate_out_uninit(&xout);
5123 set_netflow(struct ofproto *ofproto_,
5124 const struct netflow_options *netflow_options)
5126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5128 if (netflow_options) {
5129 if (!ofproto->netflow) {
5130 ofproto->netflow = netflow_create();
5131 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5133 return netflow_set_options(ofproto->netflow, netflow_options);
5134 } else if (ofproto->netflow) {
5135 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5136 netflow_destroy(ofproto->netflow);
5137 ofproto->netflow = NULL;
5144 get_netflow_ids(const struct ofproto *ofproto_,
5145 uint8_t *engine_type, uint8_t *engine_id)
5147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5149 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5153 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5155 if (!facet_is_controller_flow(facet) &&
5156 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5157 struct subfacet *subfacet;
5158 struct ofexpired expired;
5160 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5161 if (subfacet->path == SF_FAST_PATH) {
5162 struct dpif_flow_stats stats;
5164 subfacet_install(subfacet, &facet->xout.odp_actions,
5166 subfacet_update_stats(subfacet, &stats);
5170 expired.flow = facet->flow;
5171 expired.packet_count = facet->packet_count;
5172 expired.byte_count = facet->byte_count;
5173 expired.used = facet->used;
5174 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5179 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5181 struct cls_cursor cursor;
5182 struct facet *facet;
5184 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5185 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5186 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5187 send_active_timeout(ofproto, facet);
5189 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5192 static struct ofproto_dpif *
5193 ofproto_dpif_lookup(const char *name)
5195 struct ofproto_dpif *ofproto;
5197 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5198 hash_string(name, 0), &all_ofproto_dpifs) {
5199 if (!strcmp(ofproto->up.name, name)) {
5207 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5208 const char *argv[], void *aux OVS_UNUSED)
5210 struct ofproto_dpif *ofproto;
5213 ofproto = ofproto_dpif_lookup(argv[1]);
5215 unixctl_command_reply_error(conn, "no such bridge");
5218 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5219 mac_learning_flush(ofproto->ml);
5220 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5222 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5223 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5224 mac_learning_flush(ofproto->ml);
5225 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5229 unixctl_command_reply(conn, "table successfully flushed");
5232 static struct ofport_dpif *
5233 ofbundle_get_a_port(const struct ofbundle *bundle)
5235 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5240 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5241 const char *argv[], void *aux OVS_UNUSED)
5243 struct ds ds = DS_EMPTY_INITIALIZER;
5244 const struct ofproto_dpif *ofproto;
5245 const struct mac_entry *e;
5247 ofproto = ofproto_dpif_lookup(argv[1]);
5249 unixctl_command_reply_error(conn, "no such bridge");
5253 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5254 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5255 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5256 struct ofbundle *bundle = e->port.p;
5257 char name[OFP_MAX_PORT_NAME_LEN];
5259 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5261 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5262 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5263 mac_entry_age(ofproto->ml, e));
5265 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5266 unixctl_command_reply(conn, ds_cstr(&ds));
5271 struct xlate_out xout;
5272 struct xlate_in xin;
5278 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5280 struct rule_actions *actions;
5283 ds_put_char_multiple(result, '\t', level);
5285 ds_put_cstr(result, "No match\n");
5289 ovs_mutex_lock(&rule->up.mutex);
5290 cookie = rule->up.flow_cookie;
5291 ovs_mutex_unlock(&rule->up.mutex);
5293 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5294 rule ? rule->up.table_id : 0, ntohll(cookie));
5295 cls_rule_format(&rule->up.cr, result);
5296 ds_put_char(result, '\n');
5298 actions = rule_dpif_get_actions(rule);
5300 ds_put_char_multiple(result, '\t', level);
5301 ds_put_cstr(result, "OpenFlow ");
5302 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5303 ds_put_char(result, '\n');
5305 rule_actions_unref(actions);
5309 trace_format_flow(struct ds *result, int level, const char *title,
5310 struct trace_ctx *trace)
5312 ds_put_char_multiple(result, '\t', level);
5313 ds_put_format(result, "%s: ", title);
5314 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5315 ds_put_cstr(result, "unchanged");
5317 flow_format(result, &trace->xin.flow);
5318 trace->flow = trace->xin.flow;
5320 ds_put_char(result, '\n');
5324 trace_format_regs(struct ds *result, int level, const char *title,
5325 struct trace_ctx *trace)
5329 ds_put_char_multiple(result, '\t', level);
5330 ds_put_format(result, "%s:", title);
5331 for (i = 0; i < FLOW_N_REGS; i++) {
5332 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5334 ds_put_char(result, '\n');
5338 trace_format_odp(struct ds *result, int level, const char *title,
5339 struct trace_ctx *trace)
5341 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5343 ds_put_char_multiple(result, '\t', level);
5344 ds_put_format(result, "%s: ", title);
5345 format_odp_actions(result, odp_actions->data, odp_actions->size);
5346 ds_put_char(result, '\n');
5350 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5352 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5353 struct ds *result = trace->result;
5355 ds_put_char(result, '\n');
5356 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5357 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5358 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5359 trace_format_rule(result, recurse + 1, rule);
5363 trace_report(struct xlate_in *xin, const char *s, int recurse)
5365 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5366 struct ds *result = trace->result;
5368 ds_put_char_multiple(result, '\t', recurse);
5369 ds_put_cstr(result, s);
5370 ds_put_char(result, '\n');
5374 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5375 void *aux OVS_UNUSED)
5377 const struct dpif_backer *backer;
5378 struct ofproto_dpif *ofproto;
5379 struct ofpbuf odp_key, odp_mask;
5380 struct ofpbuf *packet;
5388 ofpbuf_init(&odp_key, 0);
5389 ofpbuf_init(&odp_mask, 0);
5391 /* Handle "-generate" or a hex string as the last argument. */
5392 if (!strcmp(argv[argc - 1], "-generate")) {
5393 packet = ofpbuf_new(0);
5396 const char *error = eth_from_hex(argv[argc - 1], &packet);
5399 } else if (argc == 4) {
5400 /* The 3-argument form must end in "-generate' or a hex string. */
5401 unixctl_command_reply_error(conn, error);
5406 /* Parse the flow and determine whether a datapath or
5407 * bridge is specified. If function odp_flow_key_from_string()
5408 * returns 0, the flow is a odp_flow. If function
5409 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5410 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5411 /* If the odp_flow is the second argument,
5412 * the datapath name is the first argument. */
5414 const char *dp_type;
5415 if (!strncmp(argv[1], "ovs-", 4)) {
5416 dp_type = argv[1] + 4;
5420 backer = shash_find_data(&all_dpif_backers, dp_type);
5422 unixctl_command_reply_error(conn, "Cannot find datapath "
5427 /* No datapath name specified, so there should be only one
5429 struct shash_node *node;
5430 if (shash_count(&all_dpif_backers) != 1) {
5431 unixctl_command_reply_error(conn, "Must specify datapath "
5432 "name, there is more than one type of datapath");
5435 node = shash_first(&all_dpif_backers);
5436 backer = node->data;
5439 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5440 NULL, &ofproto, NULL)) {
5441 unixctl_command_reply_error(conn, "Invalid datapath flow");
5444 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5445 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5447 unixctl_command_reply_error(conn, "Must specify bridge name");
5451 ofproto = ofproto_dpif_lookup(argv[1]);
5453 unixctl_command_reply_error(conn, "Unknown bridge name");
5457 unixctl_command_reply_error(conn, "Bad flow syntax");
5461 /* Generate a packet, if requested. */
5463 if (!packet->size) {
5464 flow_compose(packet, &flow);
5466 union flow_in_port in_port_;
5468 in_port_ = flow.in_port;
5469 ds_put_cstr(&result, "Packet: ");
5470 s = ofp_packet_to_string(packet->data, packet->size);
5471 ds_put_cstr(&result, s);
5474 /* Use the metadata from the flow and the packet argument
5475 * to reconstruct the flow. */
5476 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5481 ofproto_trace(ofproto, &flow, packet, &result);
5482 unixctl_command_reply(conn, ds_cstr(&result));
5485 ds_destroy(&result);
5486 ofpbuf_delete(packet);
5487 ofpbuf_uninit(&odp_key);
5488 ofpbuf_uninit(&odp_mask);
5492 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5493 const struct ofpbuf *packet, struct ds *ds)
5495 struct rule_dpif *rule;
5496 struct flow_wildcards wc;
5498 ds_put_cstr(ds, "Flow: ");
5499 flow_format(ds, flow);
5500 ds_put_char(ds, '\n');
5502 flow_wildcards_init_catchall(&wc);
5503 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5505 trace_format_rule(ds, 0, rule);
5506 if (rule == ofproto->miss_rule) {
5507 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5508 } else if (rule == ofproto->no_packet_in_rule) {
5509 ds_put_cstr(ds, "\nNo match, packets dropped because "
5510 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5511 } else if (rule == ofproto->drop_frags_rule) {
5512 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5513 "and the fragment handling mode is \"drop\".\n");
5517 uint64_t odp_actions_stub[1024 / 8];
5518 struct ofpbuf odp_actions;
5519 struct trace_ctx trace;
5523 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5526 ofpbuf_use_stub(&odp_actions,
5527 odp_actions_stub, sizeof odp_actions_stub);
5528 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5529 trace.xin.resubmit_hook = trace_resubmit;
5530 trace.xin.report_hook = trace_report;
5532 xlate_actions(&trace.xin, &trace.xout);
5533 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5535 ds_put_char(ds, '\n');
5536 trace_format_flow(ds, 0, "Final flow", &trace);
5538 match_init(&match, flow, &trace.xout.wc);
5539 ds_put_cstr(ds, "Relevant fields: ");
5540 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5541 ds_put_char(ds, '\n');
5543 ds_put_cstr(ds, "Datapath actions: ");
5544 format_odp_actions(ds, trace.xout.odp_actions.data,
5545 trace.xout.odp_actions.size);
5547 if (trace.xout.slow) {
5548 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5549 "slow path because it:");
5550 switch (trace.xout.slow) {
5552 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5555 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5558 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5561 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5563 case SLOW_CONTROLLER:
5564 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5565 "to the OpenFlow controller.");
5572 xlate_out_uninit(&trace.xout);
5575 rule_dpif_unref(rule);
5578 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5579 * 'reply' describing the results. */
5581 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5583 struct cls_cursor cursor;
5584 struct facet *facet;
5588 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5589 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5590 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5591 if (!facet_check_consistency(facet)) {
5595 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5597 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5601 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5602 ofproto->up.name, errors);
5604 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5609 ofproto_dpif_self_check(struct unixctl_conn *conn,
5610 int argc, const char *argv[], void *aux OVS_UNUSED)
5612 struct ds reply = DS_EMPTY_INITIALIZER;
5613 struct ofproto_dpif *ofproto;
5616 ofproto = ofproto_dpif_lookup(argv[1]);
5618 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5619 "ofproto/list for help)");
5622 ofproto_dpif_self_check__(ofproto, &reply);
5624 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5625 ofproto_dpif_self_check__(ofproto, &reply);
5629 unixctl_command_reply(conn, ds_cstr(&reply));
5633 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5634 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5635 * to destroy 'ofproto_shash' and free the returned value. */
5636 static const struct shash_node **
5637 get_ofprotos(struct shash *ofproto_shash)
5639 const struct ofproto_dpif *ofproto;
5641 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5642 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5643 shash_add_nocopy(ofproto_shash, name, ofproto);
5646 return shash_sort(ofproto_shash);
5650 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5651 const char *argv[] OVS_UNUSED,
5652 void *aux OVS_UNUSED)
5654 struct ds ds = DS_EMPTY_INITIALIZER;
5655 struct shash ofproto_shash;
5656 const struct shash_node **sorted_ofprotos;
5659 shash_init(&ofproto_shash);
5660 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5661 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5662 const struct shash_node *node = sorted_ofprotos[i];
5663 ds_put_format(&ds, "%s\n", node->name);
5666 shash_destroy(&ofproto_shash);
5667 free(sorted_ofprotos);
5669 unixctl_command_reply(conn, ds_cstr(&ds));
5674 show_dp_rates(struct ds *ds, const char *heading,
5675 const struct avg_subfacet_rates *rates)
5677 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5678 heading, rates->add_rate, rates->del_rate);
5682 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5684 const struct shash_node **ofprotos;
5685 struct ofproto_dpif *ofproto;
5686 struct shash ofproto_shash;
5687 uint64_t n_hit, n_missed;
5688 long long int minutes;
5691 n_hit = n_missed = 0;
5692 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5693 if (ofproto->backer == backer) {
5694 n_missed += ofproto->n_missed;
5695 n_hit += ofproto->n_hit;
5699 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5700 dpif_name(backer->dpif), n_hit, n_missed);
5701 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5702 " life span: %lldms\n", hmap_count(&backer->subfacets),
5703 backer->avg_n_subfacet, backer->max_n_subfacet,
5704 backer->avg_subfacet_life);
5706 minutes = (time_msec() - backer->created) / (1000 * 60);
5707 if (minutes >= 60) {
5708 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5710 if (minutes >= 60 * 24) {
5711 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5713 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5715 shash_init(&ofproto_shash);
5716 ofprotos = get_ofprotos(&ofproto_shash);
5717 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5718 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5719 const struct shash_node **ports;
5722 if (ofproto->backer != backer) {
5726 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5727 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5729 ports = shash_sort(&ofproto->up.port_by_name);
5730 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5731 const struct shash_node *node = ports[j];
5732 struct ofport *ofport = node->data;
5734 odp_port_t odp_port;
5736 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5739 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5740 if (odp_port != ODPP_NONE) {
5741 ds_put_format(ds, "%"PRIu32":", odp_port);
5743 ds_put_cstr(ds, "none:");
5746 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5749 if (!netdev_get_config(ofport->netdev, &config)) {
5750 const struct smap_node **nodes;
5753 nodes = smap_sort(&config);
5754 for (i = 0; i < smap_count(&config); i++) {
5755 const struct smap_node *node = nodes[i];
5756 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5757 node->key, node->value);
5761 smap_destroy(&config);
5763 ds_put_char(ds, ')');
5764 ds_put_char(ds, '\n');
5768 shash_destroy(&ofproto_shash);
5773 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5774 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5776 struct ds ds = DS_EMPTY_INITIALIZER;
5777 const struct shash_node **backers;
5780 backers = shash_sort(&all_dpif_backers);
5781 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5782 dpif_show_backer(backers[i]->data, &ds);
5786 unixctl_command_reply(conn, ds_cstr(&ds));
5790 /* Dump the megaflow (facet) cache. This is useful to check the
5791 * correctness of flow wildcarding, since the same mechanism is used for
5792 * both xlate caching and kernel wildcarding.
5794 * It's important to note that in the output the flow description uses
5795 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5797 * This command is only needed for advanced debugging, so it's not
5798 * documented in the man page. */
5800 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5801 int argc OVS_UNUSED, const char *argv[],
5802 void *aux OVS_UNUSED)
5804 struct ds ds = DS_EMPTY_INITIALIZER;
5805 const struct ofproto_dpif *ofproto;
5806 long long int now = time_msec();
5807 struct cls_cursor cursor;
5808 struct facet *facet;
5810 ofproto = ofproto_dpif_lookup(argv[1]);
5812 unixctl_command_reply_error(conn, "no such bridge");
5816 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5817 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5818 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5819 cls_rule_format(&facet->cr, &ds);
5820 ds_put_cstr(&ds, ", ");
5821 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5822 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5823 ds_put_cstr(&ds, "Datapath actions: ");
5824 if (facet->xout.slow) {
5825 uint64_t slow_path_stub[128 / 8];
5826 const struct nlattr *actions;
5829 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5830 slow_path_stub, sizeof slow_path_stub,
5831 &actions, &actions_len);
5832 format_odp_actions(&ds, actions, actions_len);
5834 format_odp_actions(&ds, facet->xout.odp_actions.data,
5835 facet->xout.odp_actions.size);
5837 ds_put_cstr(&ds, "\n");
5839 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5841 ds_chomp(&ds, '\n');
5842 unixctl_command_reply(conn, ds_cstr(&ds));
5846 /* Disable using the megaflows.
5848 * This command is only needed for advanced debugging, so it's not
5849 * documented in the man page. */
5851 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5852 int argc OVS_UNUSED,
5853 const char *argv[] OVS_UNUSED,
5854 void *aux OVS_UNUSED)
5856 struct ofproto_dpif *ofproto;
5858 enable_megaflows = false;
5860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5861 flush(&ofproto->up);
5864 unixctl_command_reply(conn, "megaflows disabled");
5867 /* Re-enable using megaflows.
5869 * This command is only needed for advanced debugging, so it's not
5870 * documented in the man page. */
5872 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5873 int argc OVS_UNUSED,
5874 const char *argv[] OVS_UNUSED,
5875 void *aux OVS_UNUSED)
5877 struct ofproto_dpif *ofproto;
5879 enable_megaflows = true;
5881 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5882 flush(&ofproto->up);
5885 unixctl_command_reply(conn, "megaflows enabled");
5889 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5890 int argc OVS_UNUSED, const char *argv[],
5891 void *aux OVS_UNUSED)
5893 struct ds ds = DS_EMPTY_INITIALIZER;
5894 const struct ofproto_dpif *ofproto;
5895 struct subfacet *subfacet;
5897 ofproto = ofproto_dpif_lookup(argv[1]);
5899 unixctl_command_reply_error(conn, "no such bridge");
5903 update_stats(ofproto->backer);
5905 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5906 struct facet *facet = subfacet->facet;
5907 struct odputil_keybuf maskbuf;
5910 if (facet->ofproto != ofproto) {
5914 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5915 if (enable_megaflows) {
5916 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5917 &facet->flow, UINT32_MAX);
5920 odp_flow_format(subfacet->key, subfacet->key_len,
5921 mask.data, mask.size, &ds, false);
5923 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5924 subfacet->dp_packet_count, subfacet->dp_byte_count);
5925 if (subfacet->used) {
5926 ds_put_format(&ds, "%.3fs",
5927 (time_msec() - subfacet->used) / 1000.0);
5929 ds_put_format(&ds, "never");
5931 if (subfacet->facet->tcp_flags) {
5932 ds_put_cstr(&ds, ", flags:");
5933 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5936 ds_put_cstr(&ds, ", actions:");
5937 if (facet->xout.slow) {
5938 uint64_t slow_path_stub[128 / 8];
5939 const struct nlattr *actions;
5942 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5943 slow_path_stub, sizeof slow_path_stub,
5944 &actions, &actions_len);
5945 format_odp_actions(&ds, actions, actions_len);
5947 format_odp_actions(&ds, facet->xout.odp_actions.data,
5948 facet->xout.odp_actions.size);
5950 ds_put_char(&ds, '\n');
5953 unixctl_command_reply(conn, ds_cstr(&ds));
5958 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5959 int argc OVS_UNUSED, const char *argv[],
5960 void *aux OVS_UNUSED)
5962 struct ds ds = DS_EMPTY_INITIALIZER;
5963 struct ofproto_dpif *ofproto;
5965 ofproto = ofproto_dpif_lookup(argv[1]);
5967 unixctl_command_reply_error(conn, "no such bridge");
5971 flush(&ofproto->up);
5973 unixctl_command_reply(conn, ds_cstr(&ds));
5978 ofproto_dpif_unixctl_init(void)
5980 static bool registered;
5986 unixctl_command_register(
5988 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5989 1, 3, ofproto_unixctl_trace, NULL);
5990 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5991 ofproto_unixctl_fdb_flush, NULL);
5992 unixctl_command_register("fdb/show", "bridge", 1, 1,
5993 ofproto_unixctl_fdb_show, NULL);
5994 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5995 ofproto_dpif_self_check, NULL);
5996 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5997 ofproto_unixctl_dpif_dump_dps, NULL);
5998 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6000 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6001 ofproto_unixctl_dpif_dump_flows, NULL);
6002 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6003 ofproto_unixctl_dpif_del_flows, NULL);
6004 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6005 ofproto_unixctl_dpif_dump_megaflows, NULL);
6006 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6007 ofproto_unixctl_dpif_disable_megaflows, NULL);
6008 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6009 ofproto_unixctl_dpif_enable_megaflows, NULL);
6012 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6014 * This is deprecated. It is only for compatibility with broken device drivers
6015 * in old versions of Linux that do not properly support VLANs when VLAN
6016 * devices are not used. When broken device drivers are no longer in
6017 * widespread use, we will delete these interfaces. */
6020 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6022 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6023 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6025 if (realdev_ofp_port == ofport->realdev_ofp_port
6026 && vid == ofport->vlandev_vid) {
6030 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6032 if (ofport->realdev_ofp_port) {
6035 if (realdev_ofp_port && ofport->bundle) {
6036 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6037 * themselves be part of a bundle. */
6038 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6041 ofport->realdev_ofp_port = realdev_ofp_port;
6042 ofport->vlandev_vid = vid;
6044 if (realdev_ofp_port) {
6045 vsp_add(ofport, realdev_ofp_port, vid);
6052 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6054 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6058 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6059 OVS_EXCLUDED(ofproto->vsp_mutex)
6063 ovs_mutex_lock(&ofproto->vsp_mutex);
6064 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6065 ovs_mutex_unlock(&ofproto->vsp_mutex);
6070 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6071 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6072 OVS_REQUIRES(ofproto->vsp_mutex)
6074 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6075 int vid = vlan_tci_to_vid(vlan_tci);
6076 const struct vlan_splinter *vsp;
6078 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6079 hash_realdev_vid(realdev_ofp_port, vid),
6080 &ofproto->realdev_vid_map) {
6081 if (vsp->realdev_ofp_port == realdev_ofp_port
6082 && vsp->vid == vid) {
6083 return vsp->vlandev_ofp_port;
6087 return realdev_ofp_port;
6090 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6091 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6092 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6093 * 'vlan_tci' 9, it would return the port number of eth0.9.
6095 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6096 * function just returns its 'realdev_ofp_port' argument. */
6098 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6099 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6100 OVS_EXCLUDED(ofproto->vsp_mutex)
6104 ovs_mutex_lock(&ofproto->vsp_mutex);
6105 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6106 ovs_mutex_unlock(&ofproto->vsp_mutex);
6110 static struct vlan_splinter *
6111 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6113 struct vlan_splinter *vsp;
6115 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6116 hash_ofp_port(vlandev_ofp_port),
6117 &ofproto->vlandev_map) {
6118 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6126 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6127 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6128 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6129 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6130 * eth0 and store 9 in '*vid'.
6132 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6133 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6136 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6137 ofp_port_t vlandev_ofp_port, int *vid)
6138 OVS_REQUIRES(ofproto->vsp_mutex)
6140 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6141 const struct vlan_splinter *vsp;
6143 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6148 return vsp->realdev_ofp_port;
6154 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6155 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6156 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6157 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6158 * always the case unless VLAN splinters are enabled), returns false without
6159 * making any changes. */
6161 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6162 OVS_EXCLUDED(ofproto->vsp_mutex)
6167 ovs_mutex_lock(&ofproto->vsp_mutex);
6168 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6169 ovs_mutex_unlock(&ofproto->vsp_mutex);
6174 /* Cause the flow to be processed as if it came in on the real device with
6175 * the VLAN device's VLAN ID. */
6176 flow->in_port.ofp_port = realdev;
6177 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6182 vsp_remove(struct ofport_dpif *port)
6184 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6185 struct vlan_splinter *vsp;
6187 ovs_mutex_lock(&ofproto->vsp_mutex);
6188 vsp = vlandev_find(ofproto, port->up.ofp_port);
6190 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6191 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6194 port->realdev_ofp_port = 0;
6196 VLOG_ERR("missing vlan device record");
6198 ovs_mutex_unlock(&ofproto->vsp_mutex);
6202 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6206 ovs_mutex_lock(&ofproto->vsp_mutex);
6207 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6208 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6209 == realdev_ofp_port)) {
6210 struct vlan_splinter *vsp;
6212 vsp = xmalloc(sizeof *vsp);
6213 vsp->realdev_ofp_port = realdev_ofp_port;
6214 vsp->vlandev_ofp_port = port->up.ofp_port;
6217 port->realdev_ofp_port = realdev_ofp_port;
6219 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6220 hash_ofp_port(port->up.ofp_port));
6221 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6222 hash_realdev_vid(realdev_ofp_port, vid));
6224 VLOG_ERR("duplicate vlan device record");
6226 ovs_mutex_unlock(&ofproto->vsp_mutex);
6230 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6232 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6233 return ofport ? ofport->odp_port : ODPP_NONE;
6236 struct ofport_dpif *
6237 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6239 struct ofport_dpif *port;
6241 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6242 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6243 &backer->odp_to_ofport_map) {
6244 if (port->odp_port == odp_port) {
6245 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6250 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6255 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6257 struct ofport_dpif *port;
6259 port = odp_port_to_ofport(ofproto->backer, odp_port);
6260 if (port && &ofproto->up == port->up.ofproto) {
6261 return port->up.ofp_port;
6267 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6268 * most heavily weighted element. 'base' designates the rate of decay: after
6269 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6272 exp_mavg(double *avg, int base, double new)
6274 *avg = (*avg * (base - 1) + new) / base;
6278 update_moving_averages(struct dpif_backer *backer)
6280 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6281 long long int minutes = (time_msec() - backer->created) / min_ms;
6284 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6286 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6289 backer->lifetime.add_rate = 0.0;
6290 backer->lifetime.del_rate = 0.0;
6293 /* Update hourly averages on the minute boundaries. */
6294 if (time_msec() - backer->last_minute >= min_ms) {
6295 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6296 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6298 /* Update daily averages on the hour boundaries. */
6299 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6300 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6301 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6304 backer->total_subfacet_add_count += backer->subfacet_add_count;
6305 backer->total_subfacet_del_count += backer->subfacet_del_count;
6306 backer->subfacet_add_count = 0;
6307 backer->subfacet_del_count = 0;
6308 backer->last_minute += min_ms;
6312 const struct ofproto_class ofproto_dpif_class = {
6347 port_is_lacp_current,
6348 NULL, /* rule_choose_table */
6357 rule_modify_actions,
6371 get_stp_port_status,
6378 is_mirror_output_bundle,
6379 forward_bpdu_changed,
6380 set_mac_table_config,
6382 NULL, /* meter_get_features */
6383 NULL, /* meter_set */
6384 NULL, /* meter_get */
6385 NULL, /* meter_del */