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 pins; /* Contains "struct ofputil_packet_in"s. */
515 /* By default, flows in the datapath are wildcarded (megaflows). They
516 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
517 static bool enable_megaflows = true;
519 /* All existing ofproto_dpif instances, indexed by ->up.name. */
520 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
522 static void ofproto_dpif_unixctl_init(void);
524 static inline struct ofproto_dpif *
525 ofproto_dpif_cast(const struct ofproto *ofproto)
527 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
528 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
531 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
532 ofp_port_t ofp_port);
533 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
534 const struct ofpbuf *packet, struct ds *);
537 static void handle_upcalls(struct dpif_backer *);
539 /* Flow expiration. */
540 static int expire(struct dpif_backer *);
543 static void send_netflow_active_timeouts(struct ofproto_dpif *);
546 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
548 /* Global variables. */
549 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
551 /* Initial mappings of port to bridge mappings. */
552 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
554 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
557 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
558 struct ofputil_flow_mod *fm)
560 ofproto_flow_mod(&ofproto->up, fm);
563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
564 * Takes ownership of 'pin' and pin->packet. */
566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
567 struct ofputil_packet_in *pin)
569 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
570 COVERAGE_INC(packet_in_overflow);
571 free(CONST_CAST(void *, pin->packet));
576 /* Factory functions. */
579 init(const struct shash *iface_hints)
581 struct shash_node *node;
583 /* Make a local copy, since we don't own 'iface_hints' elements. */
584 SHASH_FOR_EACH(node, iface_hints) {
585 const struct iface_hint *orig_hint = node->data;
586 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
588 new_hint->br_name = xstrdup(orig_hint->br_name);
589 new_hint->br_type = xstrdup(orig_hint->br_type);
590 new_hint->ofp_port = orig_hint->ofp_port;
592 shash_add(&init_ofp_ports, node->name, new_hint);
597 enumerate_types(struct sset *types)
599 dp_enumerate_types(types);
603 enumerate_names(const char *type, struct sset *names)
605 struct ofproto_dpif *ofproto;
608 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
609 if (strcmp(type, ofproto->up.type)) {
612 sset_add(names, ofproto->up.name);
619 del(const char *type, const char *name)
624 error = dpif_open(name, type, &dpif);
626 error = dpif_delete(dpif);
633 port_open_type(const char *datapath_type, const char *port_type)
635 return dpif_port_open_type(datapath_type, port_type);
638 /* Type functions. */
640 static void process_dpif_port_changes(struct dpif_backer *);
641 static void process_dpif_all_ports_changed(struct dpif_backer *);
642 static void process_dpif_port_change(struct dpif_backer *,
643 const char *devname);
644 static void process_dpif_port_error(struct dpif_backer *, int error);
646 static struct ofproto_dpif *
647 lookup_ofproto_dpif_by_port_name(const char *name)
649 struct ofproto_dpif *ofproto;
651 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
652 if (sset_contains(&ofproto->ports, name)) {
661 type_run(const char *type)
663 static long long int push_timer = LLONG_MIN;
664 struct dpif_backer *backer;
666 backer = shash_find_data(&all_dpif_backers, type);
668 /* This is not necessarily a problem, since backers are only
669 * created on demand. */
673 dpif_run(backer->dpif);
675 /* The most natural place to push facet statistics is when they're pulled
676 * from the datapath. However, when there are many flows in the datapath,
677 * this expensive operation can occur so frequently, that it reduces our
678 * ability to quickly set up flows. To reduce the cost, we push statistics
680 if (time_msec() > push_timer) {
681 push_timer = time_msec() + 2000;
685 /* If vswitchd started with other_config:flow_restore_wait set as "true",
686 * and the configuration has now changed to "false", enable receiving
687 * packets from the datapath. */
688 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
691 backer->recv_set_enable = true;
693 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
695 udpif_recv_set(backer->udpif, 0, false);
696 VLOG_ERR("Failed to enable receiving packets in dpif.");
699 udpif_recv_set(backer->udpif, n_handler_threads,
700 backer->recv_set_enable);
701 dpif_flow_flush(backer->dpif);
702 backer->need_revalidate = REV_RECONFIGURE;
705 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
706 * to reset the handler threads. */
707 if (backer->n_handler_threads != n_handler_threads) {
708 udpif_recv_set(backer->udpif, n_handler_threads,
709 backer->recv_set_enable);
710 backer->n_handler_threads = n_handler_threads;
713 if (backer->need_revalidate) {
714 struct ofproto_dpif *ofproto;
715 struct simap_node *node;
716 struct simap tmp_backers;
718 /* Handle tunnel garbage collection. */
719 simap_init(&tmp_backers);
720 simap_swap(&backer->tnl_backers, &tmp_backers);
722 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
723 struct ofport_dpif *iter;
725 if (backer != ofproto->backer) {
729 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
730 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
733 if (!iter->is_tunnel) {
737 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
738 namebuf, sizeof namebuf);
739 node = simap_find(&tmp_backers, dp_port);
741 simap_put(&backer->tnl_backers, dp_port, node->data);
742 simap_delete(&tmp_backers, node);
743 node = simap_find(&backer->tnl_backers, dp_port);
745 node = simap_find(&backer->tnl_backers, dp_port);
747 odp_port_t odp_port = ODPP_NONE;
749 if (!dpif_port_add(backer->dpif, iter->up.netdev,
751 simap_put(&backer->tnl_backers, dp_port,
752 odp_to_u32(odp_port));
753 node = simap_find(&backer->tnl_backers, dp_port);
758 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
759 if (tnl_port_reconfigure(iter, iter->up.netdev,
761 backer->need_revalidate = REV_RECONFIGURE;
766 SIMAP_FOR_EACH (node, &tmp_backers) {
767 dpif_port_del(backer->dpif, u32_to_odp(node->data));
769 simap_destroy(&tmp_backers);
771 switch (backer->need_revalidate) {
772 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
773 case REV_STP: COVERAGE_INC(rev_stp); break;
774 case REV_BOND: COVERAGE_INC(rev_bond); break;
775 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
776 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
777 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
778 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
780 backer->need_revalidate = 0;
782 /* Clear the drop_keys in case we should now be accepting some
783 * formerly dropped flows. */
784 drop_key_clear(backer);
786 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
787 struct facet *facet, *next;
788 struct ofport_dpif *ofport;
789 struct cls_cursor cursor;
790 struct ofbundle *bundle;
792 if (ofproto->backer != backer) {
796 ovs_rwlock_wrlock(&xlate_rwlock);
797 xlate_ofproto_set(ofproto, ofproto->up.name,
798 ofproto->backer->dpif, ofproto->miss_rule,
799 ofproto->no_packet_in_rule, ofproto->ml,
800 ofproto->stp, ofproto->mbridge,
801 ofproto->sflow, ofproto->ipfix,
802 ofproto->up.frag_handling,
803 ofproto->up.forward_bpdu,
804 connmgr_has_in_band(ofproto->up.connmgr),
805 ofproto->netflow != NULL);
807 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
808 xlate_bundle_set(ofproto, bundle, bundle->name,
809 bundle->vlan_mode, bundle->vlan,
810 bundle->trunks, bundle->use_priority_tags,
811 bundle->bond, bundle->lacp,
815 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
816 int stp_port = ofport->stp_port
817 ? stp_port_no(ofport->stp_port)
819 xlate_ofport_set(ofproto, ofport->bundle, ofport,
820 ofport->up.ofp_port, ofport->odp_port,
821 ofport->up.netdev, ofport->cfm,
822 ofport->bfd, ofport->peer, stp_port,
823 ofport->qdscp, ofport->n_qdscp,
824 ofport->up.pp.config, ofport->is_tunnel,
827 ovs_rwlock_unlock(&xlate_rwlock);
829 /* Only ofproto-dpif cares about the facet classifier so we just
830 * lock cls_cursor_init() to appease the thread safety analysis. */
831 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
832 cls_cursor_init(&cursor, &ofproto->facets, NULL);
833 ovs_rwlock_unlock(&ofproto->facets.rwlock);
834 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
835 facet_revalidate(facet);
840 udpif_revalidate(backer->udpif);
843 if (!backer->recv_set_enable) {
844 /* Wake up before a max of 1000ms. */
845 timer_set_duration(&backer->next_expiration, 1000);
846 } else if (timer_expired(&backer->next_expiration)) {
847 int delay = expire(backer);
848 timer_set_duration(&backer->next_expiration, delay);
851 process_dpif_port_changes(backer);
853 if (backer->governor) {
856 governor_run(backer->governor);
858 /* If the governor has shrunk to its minimum size and the number of
859 * subfacets has dwindled, then drop the governor entirely.
861 * For hysteresis, the number of subfacets to drop the governor is
862 * smaller than the number needed to trigger its creation. */
863 n_subfacets = hmap_count(&backer->subfacets);
864 if (n_subfacets * 4 < flow_eviction_threshold
865 && governor_is_idle(backer->governor)) {
866 governor_destroy(backer->governor);
867 backer->governor = NULL;
874 /* Check for and handle port changes in 'backer''s dpif. */
876 process_dpif_port_changes(struct dpif_backer *backer)
882 error = dpif_port_poll(backer->dpif, &devname);
888 process_dpif_all_ports_changed(backer);
892 process_dpif_port_change(backer, devname);
897 process_dpif_port_error(backer, error);
904 process_dpif_all_ports_changed(struct dpif_backer *backer)
906 struct ofproto_dpif *ofproto;
907 struct dpif_port dpif_port;
908 struct dpif_port_dump dump;
909 struct sset devnames;
912 sset_init(&devnames);
913 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
914 if (ofproto->backer == backer) {
915 struct ofport *ofport;
917 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
918 sset_add(&devnames, netdev_get_name(ofport->netdev));
922 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
923 sset_add(&devnames, dpif_port.name);
926 SSET_FOR_EACH (devname, &devnames) {
927 process_dpif_port_change(backer, devname);
929 sset_destroy(&devnames);
933 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
935 struct ofproto_dpif *ofproto;
936 struct dpif_port port;
938 /* Don't report on the datapath's device. */
939 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
943 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
944 &all_ofproto_dpifs) {
945 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
950 ofproto = lookup_ofproto_dpif_by_port_name(devname);
951 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
952 /* The port was removed. If we know the datapath,
953 * report it through poll_set(). If we don't, it may be
954 * notifying us of a removal we initiated, so ignore it.
955 * If there's a pending ENOBUFS, let it stand, since
956 * everything will be reevaluated. */
957 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
958 sset_add(&ofproto->port_poll_set, devname);
959 ofproto->port_poll_errno = 0;
961 } else if (!ofproto) {
962 /* The port was added, but we don't know with which
963 * ofproto we should associate it. Delete it. */
964 dpif_port_del(backer->dpif, port.port_no);
966 struct ofport_dpif *ofport;
968 ofport = ofport_dpif_cast(shash_find_data(
969 &ofproto->up.port_by_name, devname));
971 && ofport->odp_port != port.port_no
972 && !odp_port_to_ofport(backer, port.port_no))
974 /* 'ofport''s datapath port number has changed from
975 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
976 * structures to match. */
977 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
978 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
979 ofport->odp_port = port.port_no;
980 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
981 hash_odp_port(port.port_no));
982 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
983 backer->need_revalidate = REV_RECONFIGURE;
986 dpif_port_destroy(&port);
989 /* Propagate 'error' to all ofprotos based on 'backer'. */
991 process_dpif_port_error(struct dpif_backer *backer, int error)
993 struct ofproto_dpif *ofproto;
995 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
996 if (ofproto->backer == backer) {
997 sset_clear(&ofproto->port_poll_set);
998 ofproto->port_poll_errno = error;
1004 dpif_backer_run_fast(struct dpif_backer *backer)
1006 handle_upcalls(backer);
1012 type_run_fast(const char *type)
1014 struct dpif_backer *backer;
1016 backer = shash_find_data(&all_dpif_backers, type);
1018 /* This is not necessarily a problem, since backers are only
1019 * created on demand. */
1023 return dpif_backer_run_fast(backer);
1029 static long long int port_rl = LLONG_MIN;
1031 if (time_msec() >= port_rl) {
1032 struct ofproto_dpif *ofproto;
1034 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1035 run_fast(&ofproto->up);
1037 port_rl = time_msec() + 200;
1042 type_wait(const char *type)
1044 struct dpif_backer *backer;
1046 backer = shash_find_data(&all_dpif_backers, type);
1048 /* This is not necessarily a problem, since backers are only
1049 * created on demand. */
1053 if (backer->governor) {
1054 governor_wait(backer->governor);
1057 timer_wait(&backer->next_expiration);
1058 dpif_wait(backer->dpif);
1059 udpif_wait(backer->udpif);
1062 /* Basic life-cycle. */
1064 static int add_internal_flows(struct ofproto_dpif *);
1066 static struct ofproto *
1069 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1070 return &ofproto->up;
1074 dealloc(struct ofproto *ofproto_)
1076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1081 close_dpif_backer(struct dpif_backer *backer)
1083 struct shash_node *node;
1085 ovs_assert(backer->refcount > 0);
1087 if (--backer->refcount) {
1091 drop_key_clear(backer);
1092 hmap_destroy(&backer->drop_keys);
1094 simap_destroy(&backer->tnl_backers);
1095 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1096 hmap_destroy(&backer->odp_to_ofport_map);
1097 node = shash_find(&all_dpif_backers, backer->type);
1099 shash_delete(&all_dpif_backers, node);
1100 udpif_destroy(backer->udpif);
1101 dpif_close(backer->dpif);
1103 ovs_assert(hmap_is_empty(&backer->subfacets));
1104 hmap_destroy(&backer->subfacets);
1105 governor_destroy(backer->governor);
1110 /* Datapath port slated for removal from datapath. */
1111 struct odp_garbage {
1112 struct list list_node;
1113 odp_port_t odp_port;
1117 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1119 struct dpif_backer *backer;
1120 struct dpif_port_dump port_dump;
1121 struct dpif_port port;
1122 struct shash_node *node;
1123 struct list garbage_list;
1124 struct odp_garbage *garbage, *next;
1130 backer = shash_find_data(&all_dpif_backers, type);
1137 backer_name = xasprintf("ovs-%s", type);
1139 /* Remove any existing datapaths, since we assume we're the only
1140 * userspace controlling the datapath. */
1142 dp_enumerate_names(type, &names);
1143 SSET_FOR_EACH(name, &names) {
1144 struct dpif *old_dpif;
1146 /* Don't remove our backer if it exists. */
1147 if (!strcmp(name, backer_name)) {
1151 if (dpif_open(name, type, &old_dpif)) {
1152 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1154 dpif_delete(old_dpif);
1155 dpif_close(old_dpif);
1158 sset_destroy(&names);
1160 backer = xmalloc(sizeof *backer);
1162 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1165 VLOG_ERR("failed to open datapath of type %s: %s", type,
1166 ovs_strerror(error));
1170 backer->udpif = udpif_create(backer, backer->dpif);
1172 backer->type = xstrdup(type);
1173 backer->governor = NULL;
1174 backer->refcount = 1;
1175 hmap_init(&backer->odp_to_ofport_map);
1176 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1177 hmap_init(&backer->drop_keys);
1178 hmap_init(&backer->subfacets);
1179 timer_set_duration(&backer->next_expiration, 1000);
1180 backer->need_revalidate = 0;
1181 simap_init(&backer->tnl_backers);
1182 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1185 if (backer->recv_set_enable) {
1186 dpif_flow_flush(backer->dpif);
1189 /* Loop through the ports already on the datapath and remove any
1190 * that we don't need anymore. */
1191 list_init(&garbage_list);
1192 dpif_port_dump_start(&port_dump, backer->dpif);
1193 while (dpif_port_dump_next(&port_dump, &port)) {
1194 node = shash_find(&init_ofp_ports, port.name);
1195 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1196 garbage = xmalloc(sizeof *garbage);
1197 garbage->odp_port = port.port_no;
1198 list_push_front(&garbage_list, &garbage->list_node);
1201 dpif_port_dump_done(&port_dump);
1203 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1204 dpif_port_del(backer->dpif, garbage->odp_port);
1205 list_remove(&garbage->list_node);
1209 shash_add(&all_dpif_backers, type, backer);
1211 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1213 VLOG_ERR("failed to listen on datapath of type %s: %s",
1214 type, ovs_strerror(error));
1215 close_dpif_backer(backer);
1218 udpif_recv_set(backer->udpif, n_handler_threads,
1219 backer->recv_set_enable);
1220 backer->n_handler_threads = n_handler_threads;
1222 backer->max_n_subfacet = 0;
1223 backer->created = time_msec();
1224 backer->last_minute = backer->created;
1225 memset(&backer->hourly, 0, sizeof backer->hourly);
1226 memset(&backer->daily, 0, sizeof backer->daily);
1227 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1228 backer->subfacet_add_count = 0;
1229 backer->subfacet_del_count = 0;
1230 backer->total_subfacet_add_count = 0;
1231 backer->total_subfacet_del_count = 0;
1232 backer->avg_n_subfacet = 0;
1233 backer->avg_subfacet_life = 0;
1239 construct(struct ofproto *ofproto_)
1241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1242 struct shash_node *node, *next;
1246 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1251 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1252 ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1254 ofproto->netflow = NULL;
1255 ofproto->sflow = NULL;
1256 ofproto->ipfix = NULL;
1257 ofproto->stp = NULL;
1258 hmap_init(&ofproto->bundles);
1259 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1260 ofproto->mbridge = mbridge_create();
1261 ofproto->has_bonded_bundles = false;
1262 ovs_mutex_init(&ofproto->vsp_mutex);
1264 classifier_init(&ofproto->facets);
1265 ofproto->consistency_rl = LLONG_MIN;
1267 guarded_list_init(&ofproto->pins);
1269 ofproto_dpif_unixctl_init();
1271 hmap_init(&ofproto->vlandev_map);
1272 hmap_init(&ofproto->realdev_vid_map);
1274 sset_init(&ofproto->ports);
1275 sset_init(&ofproto->ghost_ports);
1276 sset_init(&ofproto->port_poll_set);
1277 ofproto->port_poll_errno = 0;
1279 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1280 struct iface_hint *iface_hint = node->data;
1282 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1283 /* Check if the datapath already has this port. */
1284 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1285 sset_add(&ofproto->ports, node->name);
1288 free(iface_hint->br_name);
1289 free(iface_hint->br_type);
1291 shash_delete(&init_ofp_ports, node);
1295 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1296 hash_string(ofproto->up.name, 0));
1297 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1299 ofproto_init_tables(ofproto_, N_TABLES);
1300 error = add_internal_flows(ofproto);
1301 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1304 ofproto->n_missed = 0;
1310 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1311 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1313 struct ofputil_flow_mod fm;
1316 match_init_catchall(&fm.match);
1318 match_set_reg(&fm.match, 0, id);
1319 fm.new_cookie = htonll(0);
1320 fm.cookie = htonll(0);
1321 fm.cookie_mask = htonll(0);
1322 fm.modify_cookie = false;
1323 fm.table_id = TBL_INTERNAL;
1324 fm.command = OFPFC_ADD;
1325 fm.idle_timeout = 0;
1326 fm.hard_timeout = 0;
1330 fm.ofpacts = ofpacts->data;
1331 fm.ofpacts_len = ofpacts->size;
1333 error = ofproto_flow_mod(&ofproto->up, &fm);
1335 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1336 id, ofperr_to_string(error));
1340 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1342 rule_dpif_unref(*rulep);
1351 add_internal_flows(struct ofproto_dpif *ofproto)
1353 struct ofpact_controller *controller;
1354 uint64_t ofpacts_stub[128 / 8];
1355 struct ofpbuf ofpacts;
1359 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1362 controller = ofpact_put_CONTROLLER(&ofpacts);
1363 controller->max_len = UINT16_MAX;
1364 controller->controller_id = 0;
1365 controller->reason = OFPR_NO_MATCH;
1366 ofpact_pad(&ofpacts);
1368 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1373 ofpbuf_clear(&ofpacts);
1374 error = add_internal_flow(ofproto, id++, &ofpacts,
1375 &ofproto->no_packet_in_rule);
1380 error = add_internal_flow(ofproto, id++, &ofpacts,
1381 &ofproto->drop_frags_rule);
1386 destruct(struct ofproto *ofproto_)
1388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1389 struct rule_dpif *rule, *next_rule;
1390 struct ofputil_packet_in *pin, *next_pin;
1391 struct facet *facet, *next_facet;
1392 struct cls_cursor cursor;
1393 struct oftable *table;
1396 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1397 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1398 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1399 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1400 facet_remove(facet);
1403 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1404 ovs_rwlock_wrlock(&xlate_rwlock);
1405 xlate_remove_ofproto(ofproto);
1406 ovs_rwlock_unlock(&xlate_rwlock);
1408 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1409 * use-after-free error. */
1410 udpif_revalidate(ofproto->backer->udpif);
1412 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1414 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1415 struct cls_cursor cursor;
1417 ovs_rwlock_rdlock(&table->cls.rwlock);
1418 cls_cursor_init(&cursor, &table->cls, NULL);
1419 ovs_rwlock_unlock(&table->cls.rwlock);
1420 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1421 ofproto_rule_delete(&ofproto->up, &rule->up);
1425 guarded_list_pop_all(&ofproto->pins, &pins);
1426 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1427 list_remove(&pin->list_node);
1428 free(CONST_CAST(void *, pin->packet));
1431 guarded_list_destroy(&ofproto->pins);
1433 mbridge_unref(ofproto->mbridge);
1435 netflow_destroy(ofproto->netflow);
1436 dpif_sflow_unref(ofproto->sflow);
1437 hmap_destroy(&ofproto->bundles);
1438 mac_learning_unref(ofproto->ml);
1440 classifier_destroy(&ofproto->facets);
1442 hmap_destroy(&ofproto->vlandev_map);
1443 hmap_destroy(&ofproto->realdev_vid_map);
1445 sset_destroy(&ofproto->ports);
1446 sset_destroy(&ofproto->ghost_ports);
1447 sset_destroy(&ofproto->port_poll_set);
1449 ovs_mutex_destroy(&ofproto->vsp_mutex);
1451 close_dpif_backer(ofproto->backer);
1455 run_fast(struct ofproto *ofproto_)
1457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1458 struct ofputil_packet_in *pin, *next_pin;
1459 struct ofport_dpif *ofport;
1462 /* Do not perform any periodic activity required by 'ofproto' while
1463 * waiting for flow restore to complete. */
1464 if (ofproto_get_flow_restore_wait()) {
1468 guarded_list_pop_all(&ofproto->pins, &pins);
1469 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1470 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1471 list_remove(&pin->list_node);
1472 free(CONST_CAST(void *, pin->packet));
1476 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1477 port_run_fast(ofport);
1484 run(struct ofproto *ofproto_)
1486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1487 struct ofport_dpif *ofport;
1488 struct ofbundle *bundle;
1491 if (mbridge_need_revalidate(ofproto->mbridge)) {
1492 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1493 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1494 mac_learning_flush(ofproto->ml);
1495 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1498 /* Do not perform any periodic activity below required by 'ofproto' while
1499 * waiting for flow restore to complete. */
1500 if (ofproto_get_flow_restore_wait()) {
1504 error = run_fast(ofproto_);
1509 if (ofproto->netflow) {
1510 if (netflow_run(ofproto->netflow)) {
1511 send_netflow_active_timeouts(ofproto);
1514 if (ofproto->sflow) {
1515 dpif_sflow_run(ofproto->sflow);
1517 if (ofproto->ipfix) {
1518 dpif_ipfix_run(ofproto->ipfix);
1521 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1524 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1529 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1530 if (mac_learning_run(ofproto->ml)) {
1531 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1533 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1535 /* Check the consistency of a random facet, to aid debugging. */
1536 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1537 if (time_msec() >= ofproto->consistency_rl
1538 && !classifier_is_empty(&ofproto->facets)
1539 && !ofproto->backer->need_revalidate) {
1540 struct cls_table *table;
1541 struct cls_rule *cr;
1542 struct facet *facet;
1544 ofproto->consistency_rl = time_msec() + 250;
1546 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1547 struct cls_table, hmap_node);
1548 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1550 facet = CONTAINER_OF(cr, struct facet, cr);
1552 if (!facet_check_consistency(facet)) {
1553 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1556 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1562 wait(struct ofproto *ofproto_)
1564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1565 struct ofport_dpif *ofport;
1566 struct ofbundle *bundle;
1568 if (ofproto_get_flow_restore_wait()) {
1572 if (ofproto->sflow) {
1573 dpif_sflow_wait(ofproto->sflow);
1575 if (ofproto->ipfix) {
1576 dpif_ipfix_wait(ofproto->ipfix);
1578 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1581 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1582 bundle_wait(bundle);
1584 if (ofproto->netflow) {
1585 netflow_wait(ofproto->netflow);
1587 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1588 mac_learning_wait(ofproto->ml);
1589 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1591 if (ofproto->backer->need_revalidate) {
1592 /* Shouldn't happen, but if it does just go around again. */
1593 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1594 poll_immediate_wake();
1599 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1601 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1602 struct cls_cursor cursor;
1603 size_t n_subfacets = 0;
1604 struct facet *facet;
1606 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1607 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1608 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1610 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1611 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1612 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1613 n_subfacets += list_size(&facet->subfacets);
1615 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1616 simap_increase(usage, "subfacets", n_subfacets);
1620 flush(struct ofproto *ofproto_)
1622 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1623 struct subfacet *subfacet, *next_subfacet;
1624 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1628 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1629 &ofproto->backer->subfacets) {
1630 if (subfacet->facet->ofproto != ofproto) {
1634 if (subfacet->path != SF_NOT_INSTALLED) {
1635 batch[n_batch++] = subfacet;
1636 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1637 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1641 subfacet_destroy(subfacet);
1646 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1651 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1652 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1654 *arp_match_ip = true;
1655 *actions = (OFPUTIL_A_OUTPUT |
1656 OFPUTIL_A_SET_VLAN_VID |
1657 OFPUTIL_A_SET_VLAN_PCP |
1658 OFPUTIL_A_STRIP_VLAN |
1659 OFPUTIL_A_SET_DL_SRC |
1660 OFPUTIL_A_SET_DL_DST |
1661 OFPUTIL_A_SET_NW_SRC |
1662 OFPUTIL_A_SET_NW_DST |
1663 OFPUTIL_A_SET_NW_TOS |
1664 OFPUTIL_A_SET_TP_SRC |
1665 OFPUTIL_A_SET_TP_DST |
1670 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1673 struct dpif_dp_stats s;
1674 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1677 strcpy(ots->name, "classifier");
1679 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1680 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1681 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1682 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1684 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1685 ots->lookup_count = htonll(n_lookup);
1686 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1689 static struct ofport *
1692 struct ofport_dpif *port = xmalloc(sizeof *port);
1697 port_dealloc(struct ofport *port_)
1699 struct ofport_dpif *port = ofport_dpif_cast(port_);
1704 port_construct(struct ofport *port_)
1706 struct ofport_dpif *port = ofport_dpif_cast(port_);
1707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1708 const struct netdev *netdev = port->up.netdev;
1709 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1710 struct dpif_port dpif_port;
1713 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1714 port->bundle = NULL;
1717 port->may_enable = true;
1718 port->stp_port = NULL;
1719 port->stp_state = STP_DISABLED;
1720 port->is_tunnel = false;
1724 port->realdev_ofp_port = 0;
1725 port->vlandev_vid = 0;
1726 port->carrier_seq = netdev_get_carrier_resets(netdev);
1728 if (netdev_vport_is_patch(netdev)) {
1729 /* By bailing out here, we don't submit the port to the sFlow module
1730 * to be considered for counter polling export. This is correct
1731 * because the patch port represents an interface that sFlow considers
1732 * to be "internal" to the switch as a whole, and therefore not an
1733 * candidate for counter polling. */
1734 port->odp_port = ODPP_NONE;
1735 ofport_update_peer(port);
1739 error = dpif_port_query_by_name(ofproto->backer->dpif,
1740 netdev_vport_get_dpif_port(netdev, namebuf,
1747 port->odp_port = dpif_port.port_no;
1749 if (netdev_get_tunnel_config(netdev)) {
1750 tnl_port_add(port, port->up.netdev, port->odp_port);
1751 port->is_tunnel = true;
1753 /* Sanity-check that a mapping doesn't already exist. This
1754 * shouldn't happen for non-tunnel ports. */
1755 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1756 VLOG_ERR("port %s already has an OpenFlow port number",
1758 dpif_port_destroy(&dpif_port);
1762 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1763 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1764 hash_odp_port(port->odp_port));
1765 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1767 dpif_port_destroy(&dpif_port);
1769 if (ofproto->sflow) {
1770 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1777 port_destruct(struct ofport *port_)
1779 struct ofport_dpif *port = ofport_dpif_cast(port_);
1780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1781 const char *devname = netdev_get_name(port->up.netdev);
1782 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1783 const char *dp_port_name;
1785 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1786 ovs_rwlock_wrlock(&xlate_rwlock);
1787 xlate_ofport_remove(port);
1788 ovs_rwlock_unlock(&xlate_rwlock);
1790 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1792 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1793 /* The underlying device is still there, so delete it. This
1794 * happens when the ofproto is being destroyed, since the caller
1795 * assumes that removal of attached ports will happen as part of
1797 if (!port->is_tunnel) {
1798 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1803 port->peer->peer = NULL;
1807 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1808 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1809 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1810 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1814 sset_find_and_delete(&ofproto->ports, devname);
1815 sset_find_and_delete(&ofproto->ghost_ports, devname);
1816 bundle_remove(port_);
1817 set_cfm(port_, NULL);
1818 set_bfd(port_, NULL);
1819 if (ofproto->sflow) {
1820 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1827 port_modified(struct ofport *port_)
1829 struct ofport_dpif *port = ofport_dpif_cast(port_);
1831 if (port->bundle && port->bundle->bond) {
1832 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1836 cfm_set_netdev(port->cfm, port->up.netdev);
1840 bfd_set_netdev(port->bfd, port->up.netdev);
1843 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1845 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1849 ofport_update_peer(port);
1853 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1855 struct ofport_dpif *port = ofport_dpif_cast(port_);
1856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1857 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1859 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1860 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1861 OFPUTIL_PC_NO_PACKET_IN)) {
1862 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1864 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1865 bundle_update(port->bundle);
1871 set_sflow(struct ofproto *ofproto_,
1872 const struct ofproto_sflow_options *sflow_options)
1874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1875 struct dpif_sflow *ds = ofproto->sflow;
1877 if (sflow_options) {
1879 struct ofport_dpif *ofport;
1881 ds = ofproto->sflow = dpif_sflow_create();
1882 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1883 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1885 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1887 dpif_sflow_set_options(ds, sflow_options);
1890 dpif_sflow_unref(ds);
1891 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892 ofproto->sflow = NULL;
1900 struct ofproto *ofproto_,
1901 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1902 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1903 size_t n_flow_exporters_options)
1905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1906 struct dpif_ipfix *di = ofproto->ipfix;
1907 bool has_options = bridge_exporter_options || flow_exporters_options;
1909 if (has_options && !di) {
1910 di = ofproto->ipfix = dpif_ipfix_create();
1914 /* Call set_options in any case to cleanly flush the flow
1915 * caches in the last exporters that are to be destroyed. */
1916 dpif_ipfix_set_options(
1917 di, bridge_exporter_options, flow_exporters_options,
1918 n_flow_exporters_options);
1921 dpif_ipfix_unref(di);
1922 ofproto->ipfix = NULL;
1930 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1932 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1939 struct ofproto_dpif *ofproto;
1941 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1942 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1943 ofport->cfm = cfm_create(ofport->up.netdev);
1946 if (cfm_configure(ofport->cfm, s)) {
1952 cfm_unref(ofport->cfm);
1958 get_cfm_status(const struct ofport *ofport_,
1959 struct ofproto_cfm_status *status)
1961 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1964 status->faults = cfm_get_fault(ofport->cfm);
1965 status->remote_opstate = cfm_get_opup(ofport->cfm);
1966 status->health = cfm_get_health(ofport->cfm);
1967 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1975 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1978 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1982 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1983 cfg, ofport->up.netdev);
1984 if (ofport->bfd != old) {
1985 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1992 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1994 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997 bfd_get_status(ofport->bfd, smap);
2004 /* Spanning Tree. */
2007 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2009 struct ofproto_dpif *ofproto = ofproto_;
2010 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2011 struct ofport_dpif *ofport;
2013 ofport = stp_port_get_aux(sp);
2015 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2016 ofproto->up.name, port_num);
2018 struct eth_header *eth = pkt->l2;
2020 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2021 if (eth_addr_is_zero(eth->eth_src)) {
2022 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2023 "with unknown MAC", ofproto->up.name, port_num);
2025 send_packet(ofport, pkt);
2031 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2033 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2037 /* Only revalidate flows if the configuration changed. */
2038 if (!s != !ofproto->stp) {
2039 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043 if (!ofproto->stp) {
2044 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2045 send_bpdu_cb, ofproto);
2046 ofproto->stp_last_tick = time_msec();
2049 stp_set_bridge_id(ofproto->stp, s->system_id);
2050 stp_set_bridge_priority(ofproto->stp, s->priority);
2051 stp_set_hello_time(ofproto->stp, s->hello_time);
2052 stp_set_max_age(ofproto->stp, s->max_age);
2053 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2055 struct ofport *ofport;
2057 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2058 set_stp_port(ofport, NULL);
2061 stp_unref(ofproto->stp);
2062 ofproto->stp = NULL;
2069 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2075 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2076 s->designated_root = stp_get_designated_root(ofproto->stp);
2077 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2086 update_stp_port_state(struct ofport_dpif *ofport)
2088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2089 enum stp_state state;
2091 /* Figure out new state. */
2092 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2096 if (ofport->stp_state != state) {
2097 enum ofputil_port_state of_state;
2100 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2101 netdev_get_name(ofport->up.netdev),
2102 stp_state_name(ofport->stp_state),
2103 stp_state_name(state));
2104 if (stp_learn_in_state(ofport->stp_state)
2105 != stp_learn_in_state(state)) {
2106 /* xxx Learning action flows should also be flushed. */
2107 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2108 mac_learning_flush(ofproto->ml);
2109 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2111 fwd_change = stp_forward_in_state(ofport->stp_state)
2112 != stp_forward_in_state(state);
2114 ofproto->backer->need_revalidate = REV_STP;
2115 ofport->stp_state = state;
2116 ofport->stp_state_entered = time_msec();
2118 if (fwd_change && ofport->bundle) {
2119 bundle_update(ofport->bundle);
2122 /* Update the STP state bits in the OpenFlow port description. */
2123 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2124 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2125 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2126 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2127 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2129 ofproto_port_set_state(&ofport->up, of_state);
2133 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2134 * caller is responsible for assigning STP port numbers and ensuring
2135 * there are no duplicates. */
2137 set_stp_port(struct ofport *ofport_,
2138 const struct ofproto_port_stp_settings *s)
2140 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2142 struct stp_port *sp = ofport->stp_port;
2144 if (!s || !s->enable) {
2146 ofport->stp_port = NULL;
2147 stp_port_disable(sp);
2148 update_stp_port_state(ofport);
2151 } else if (sp && stp_port_no(sp) != s->port_num
2152 && ofport == stp_port_get_aux(sp)) {
2153 /* The port-id changed, so disable the old one if it's not
2154 * already in use by another port. */
2155 stp_port_disable(sp);
2158 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2159 stp_port_enable(sp);
2161 stp_port_set_aux(sp, ofport);
2162 stp_port_set_priority(sp, s->priority);
2163 stp_port_set_path_cost(sp, s->path_cost);
2165 update_stp_port_state(ofport);
2171 get_stp_port_status(struct ofport *ofport_,
2172 struct ofproto_port_stp_status *s)
2174 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2176 struct stp_port *sp = ofport->stp_port;
2178 if (!ofproto->stp || !sp) {
2184 s->port_id = stp_port_get_id(sp);
2185 s->state = stp_port_get_state(sp);
2186 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2187 s->role = stp_port_get_role(sp);
2188 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2194 stp_run(struct ofproto_dpif *ofproto)
2197 long long int now = time_msec();
2198 long long int elapsed = now - ofproto->stp_last_tick;
2199 struct stp_port *sp;
2202 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2203 ofproto->stp_last_tick = now;
2205 while (stp_get_changed_port(ofproto->stp, &sp)) {
2206 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2209 update_stp_port_state(ofport);
2213 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2214 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2215 mac_learning_flush(ofproto->ml);
2216 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2222 stp_wait(struct ofproto_dpif *ofproto)
2225 poll_timer_wait(1000);
2230 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2233 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2236 if (ofport->n_qdscp != n_qdscp
2237 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2238 n_qdscp * sizeof *qdscp))) {
2239 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2240 free(ofport->qdscp);
2241 ofport->qdscp = n_qdscp
2242 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2244 ofport->n_qdscp = n_qdscp;
2252 /* Expires all MAC learning entries associated with 'bundle' and forces its
2253 * ofproto to revalidate every flow.
2255 * Normally MAC learning entries are removed only from the ofproto associated
2256 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2257 * are removed from every ofproto. When patch ports and SLB bonds are in use
2258 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2259 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2260 * with the host from which it migrated. */
2262 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2264 struct ofproto_dpif *ofproto = bundle->ofproto;
2265 struct mac_learning *ml = ofproto->ml;
2266 struct mac_entry *mac, *next_mac;
2268 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2269 ovs_rwlock_wrlock(&ml->rwlock);
2270 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2271 if (mac->port.p == bundle) {
2273 struct ofproto_dpif *o;
2275 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2277 struct mac_entry *e;
2279 ovs_rwlock_wrlock(&o->ml->rwlock);
2280 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2282 mac_learning_expire(o->ml, e);
2284 ovs_rwlock_unlock(&o->ml->rwlock);
2289 mac_learning_expire(ml, mac);
2292 ovs_rwlock_unlock(&ml->rwlock);
2295 static struct ofbundle *
2296 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2298 struct ofbundle *bundle;
2300 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2301 &ofproto->bundles) {
2302 if (bundle->aux == aux) {
2310 bundle_update(struct ofbundle *bundle)
2312 struct ofport_dpif *port;
2314 bundle->floodable = true;
2315 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2316 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2317 || !stp_forward_in_state(port->stp_state)) {
2318 bundle->floodable = false;
2325 bundle_del_port(struct ofport_dpif *port)
2327 struct ofbundle *bundle = port->bundle;
2329 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2331 list_remove(&port->bundle_node);
2332 port->bundle = NULL;
2335 lacp_slave_unregister(bundle->lacp, port);
2338 bond_slave_unregister(bundle->bond, port);
2341 bundle_update(bundle);
2345 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2346 struct lacp_slave_settings *lacp)
2348 struct ofport_dpif *port;
2350 port = get_ofp_port(bundle->ofproto, ofp_port);
2355 if (port->bundle != bundle) {
2356 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2358 bundle_remove(&port->up);
2361 port->bundle = bundle;
2362 list_push_back(&bundle->ports, &port->bundle_node);
2363 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2364 || !stp_forward_in_state(port->stp_state)) {
2365 bundle->floodable = false;
2369 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370 lacp_slave_register(bundle->lacp, port, lacp);
2377 bundle_destroy(struct ofbundle *bundle)
2379 struct ofproto_dpif *ofproto;
2380 struct ofport_dpif *port, *next_port;
2386 ofproto = bundle->ofproto;
2387 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2389 ovs_rwlock_wrlock(&xlate_rwlock);
2390 xlate_bundle_remove(bundle);
2391 ovs_rwlock_unlock(&xlate_rwlock);
2393 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2394 bundle_del_port(port);
2397 bundle_flush_macs(bundle, true);
2398 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2400 free(bundle->trunks);
2401 lacp_unref(bundle->lacp);
2402 bond_unref(bundle->bond);
2407 bundle_set(struct ofproto *ofproto_, void *aux,
2408 const struct ofproto_bundle_settings *s)
2410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2411 bool need_flush = false;
2412 struct ofport_dpif *port;
2413 struct ofbundle *bundle;
2414 unsigned long *trunks;
2420 bundle_destroy(bundle_lookup(ofproto, aux));
2424 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2425 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2427 bundle = bundle_lookup(ofproto, aux);
2429 bundle = xmalloc(sizeof *bundle);
2431 bundle->ofproto = ofproto;
2432 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2433 hash_pointer(aux, 0));
2435 bundle->name = NULL;
2437 list_init(&bundle->ports);
2438 bundle->vlan_mode = PORT_VLAN_TRUNK;
2440 bundle->trunks = NULL;
2441 bundle->use_priority_tags = s->use_priority_tags;
2442 bundle->lacp = NULL;
2443 bundle->bond = NULL;
2445 bundle->floodable = true;
2446 mbridge_register_bundle(ofproto->mbridge, bundle);
2449 if (!bundle->name || strcmp(s->name, bundle->name)) {
2451 bundle->name = xstrdup(s->name);
2456 if (!bundle->lacp) {
2457 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2458 bundle->lacp = lacp_create();
2460 lacp_configure(bundle->lacp, s->lacp);
2462 lacp_unref(bundle->lacp);
2463 bundle->lacp = NULL;
2466 /* Update set of ports. */
2468 for (i = 0; i < s->n_slaves; i++) {
2469 if (!bundle_add_port(bundle, s->slaves[i],
2470 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2474 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2475 struct ofport_dpif *next_port;
2477 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2478 for (i = 0; i < s->n_slaves; i++) {
2479 if (s->slaves[i] == port->up.ofp_port) {
2484 bundle_del_port(port);
2488 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2490 if (list_is_empty(&bundle->ports)) {
2491 bundle_destroy(bundle);
2495 /* Set VLAN tagging mode */
2496 if (s->vlan_mode != bundle->vlan_mode
2497 || s->use_priority_tags != bundle->use_priority_tags) {
2498 bundle->vlan_mode = s->vlan_mode;
2499 bundle->use_priority_tags = s->use_priority_tags;
2504 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2505 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2507 if (vlan != bundle->vlan) {
2508 bundle->vlan = vlan;
2512 /* Get trunked VLANs. */
2513 switch (s->vlan_mode) {
2514 case PORT_VLAN_ACCESS:
2518 case PORT_VLAN_TRUNK:
2519 trunks = CONST_CAST(unsigned long *, s->trunks);
2522 case PORT_VLAN_NATIVE_UNTAGGED:
2523 case PORT_VLAN_NATIVE_TAGGED:
2524 if (vlan != 0 && (!s->trunks
2525 || !bitmap_is_set(s->trunks, vlan)
2526 || bitmap_is_set(s->trunks, 0))) {
2527 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2529 trunks = bitmap_clone(s->trunks, 4096);
2531 trunks = bitmap_allocate1(4096);
2533 bitmap_set1(trunks, vlan);
2534 bitmap_set0(trunks, 0);
2536 trunks = CONST_CAST(unsigned long *, s->trunks);
2543 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2544 free(bundle->trunks);
2545 if (trunks == s->trunks) {
2546 bundle->trunks = vlan_bitmap_clone(trunks);
2548 bundle->trunks = trunks;
2553 if (trunks != s->trunks) {
2558 if (!list_is_short(&bundle->ports)) {
2559 bundle->ofproto->has_bonded_bundles = true;
2561 if (bond_reconfigure(bundle->bond, s->bond)) {
2562 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2565 bundle->bond = bond_create(s->bond);
2566 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2569 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2570 bond_slave_register(bundle->bond, port, port->up.netdev);
2573 bond_unref(bundle->bond);
2574 bundle->bond = NULL;
2577 /* If we changed something that would affect MAC learning, un-learn
2578 * everything on this port and force flow revalidation. */
2580 bundle_flush_macs(bundle, false);
2587 bundle_remove(struct ofport *port_)
2589 struct ofport_dpif *port = ofport_dpif_cast(port_);
2590 struct ofbundle *bundle = port->bundle;
2593 bundle_del_port(port);
2594 if (list_is_empty(&bundle->ports)) {
2595 bundle_destroy(bundle);
2596 } else if (list_is_short(&bundle->ports)) {
2597 bond_unref(bundle->bond);
2598 bundle->bond = NULL;
2604 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2606 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2607 struct ofport_dpif *port = port_;
2608 uint8_t ea[ETH_ADDR_LEN];
2611 error = netdev_get_etheraddr(port->up.netdev, ea);
2613 struct ofpbuf packet;
2616 ofpbuf_init(&packet, 0);
2617 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2619 memcpy(packet_pdu, pdu, pdu_size);
2621 send_packet(port, &packet);
2622 ofpbuf_uninit(&packet);
2624 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2625 "%s (%s)", port->bundle->name,
2626 netdev_get_name(port->up.netdev), ovs_strerror(error));
2631 bundle_send_learning_packets(struct ofbundle *bundle)
2633 struct ofproto_dpif *ofproto = bundle->ofproto;
2634 struct ofpbuf *learning_packet;
2635 int error, n_packets, n_errors;
2636 struct mac_entry *e;
2637 struct list packets;
2639 list_init(&packets);
2640 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2641 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2642 if (e->port.p != bundle) {
2645 learning_packet = bond_compose_learning_packet(bundle->bond,
2648 learning_packet->private_p = port_void;
2649 list_push_back(&packets, &learning_packet->list_node);
2652 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2654 error = n_packets = n_errors = 0;
2655 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2658 ret = send_packet(learning_packet->private_p, learning_packet);
2665 ofpbuf_list_delete(&packets);
2668 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2669 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2670 "packets, last error was: %s",
2671 bundle->name, n_errors, n_packets, ovs_strerror(error));
2673 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2674 bundle->name, n_packets);
2679 bundle_run(struct ofbundle *bundle)
2682 lacp_run(bundle->lacp, send_pdu_cb);
2685 struct ofport_dpif *port;
2687 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2688 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2691 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2692 bundle->ofproto->backer->need_revalidate = REV_BOND;
2695 if (bond_should_send_learning_packets(bundle->bond)) {
2696 bundle_send_learning_packets(bundle);
2702 bundle_wait(struct ofbundle *bundle)
2705 lacp_wait(bundle->lacp);
2708 bond_wait(bundle->bond);
2715 mirror_set__(struct ofproto *ofproto_, void *aux,
2716 const struct ofproto_mirror_settings *s)
2718 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2719 struct ofbundle **srcs, **dsts;
2724 mirror_destroy(ofproto->mbridge, aux);
2728 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2729 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2731 for (i = 0; i < s->n_srcs; i++) {
2732 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2735 for (i = 0; i < s->n_dsts; i++) {
2736 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2739 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2740 s->n_dsts, s->src_vlans,
2741 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2748 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2749 uint64_t *packets, uint64_t *bytes)
2752 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2757 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2760 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2761 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2762 mac_learning_flush(ofproto->ml);
2764 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2769 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2773 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2777 forward_bpdu_changed(struct ofproto *ofproto_)
2779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2784 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2788 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2789 mac_learning_set_idle_time(ofproto->ml, idle_time);
2790 mac_learning_set_max_entries(ofproto->ml, max_entries);
2791 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2796 static struct ofport_dpif *
2797 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2799 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2800 return ofport ? ofport_dpif_cast(ofport) : NULL;
2803 static struct ofport_dpif *
2804 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2806 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2807 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2811 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2812 struct ofproto_port *ofproto_port,
2813 struct dpif_port *dpif_port)
2815 ofproto_port->name = dpif_port->name;
2816 ofproto_port->type = dpif_port->type;
2817 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2821 ofport_update_peer(struct ofport_dpif *ofport)
2823 const struct ofproto_dpif *ofproto;
2824 struct dpif_backer *backer;
2827 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2831 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2832 backer->need_revalidate = REV_RECONFIGURE;
2835 ofport->peer->peer = NULL;
2836 ofport->peer = NULL;
2839 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2844 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2845 struct ofport *peer_ofport;
2846 struct ofport_dpif *peer;
2849 if (ofproto->backer != backer) {
2853 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2858 peer = ofport_dpif_cast(peer_ofport);
2859 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2860 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2862 ofport->peer = peer;
2863 ofport->peer->peer = ofport;
2873 port_run_fast(struct ofport_dpif *ofport)
2875 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2876 struct ofpbuf packet;
2878 ofpbuf_init(&packet, 0);
2879 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2880 send_packet(ofport, &packet);
2881 ofpbuf_uninit(&packet);
2884 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2885 struct ofpbuf packet;
2887 ofpbuf_init(&packet, 0);
2888 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2889 send_packet(ofport, &packet);
2890 ofpbuf_uninit(&packet);
2895 port_run(struct ofport_dpif *ofport)
2897 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2898 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2899 bool enable = netdev_get_carrier(ofport->up.netdev);
2900 bool cfm_enable = false;
2901 bool bfd_enable = false;
2903 ofport->carrier_seq = carrier_seq;
2905 port_run_fast(ofport);
2908 int cfm_opup = cfm_get_opup(ofport->cfm);
2910 cfm_run(ofport->cfm);
2911 cfm_enable = !cfm_get_fault(ofport->cfm);
2913 if (cfm_opup >= 0) {
2914 cfm_enable = cfm_enable && cfm_opup;
2919 bfd_run(ofport->bfd);
2920 bfd_enable = bfd_forwarding(ofport->bfd);
2923 if (ofport->bfd || ofport->cfm) {
2924 enable = enable && (cfm_enable || bfd_enable);
2927 if (ofport->bundle) {
2928 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2929 if (carrier_changed) {
2930 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2934 if (ofport->may_enable != enable) {
2935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2936 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2939 ofport->may_enable = enable;
2943 port_wait(struct ofport_dpif *ofport)
2946 cfm_wait(ofport->cfm);
2950 bfd_wait(ofport->bfd);
2955 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2956 struct ofproto_port *ofproto_port)
2958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2959 struct dpif_port dpif_port;
2962 if (sset_contains(&ofproto->ghost_ports, devname)) {
2963 const char *type = netdev_get_type_from_name(devname);
2965 /* We may be called before ofproto->up.port_by_name is populated with
2966 * the appropriate ofport. For this reason, we must get the name and
2967 * type from the netdev layer directly. */
2969 const struct ofport *ofport;
2971 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2972 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2973 ofproto_port->name = xstrdup(devname);
2974 ofproto_port->type = xstrdup(type);
2980 if (!sset_contains(&ofproto->ports, devname)) {
2983 error = dpif_port_query_by_name(ofproto->backer->dpif,
2984 devname, &dpif_port);
2986 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2992 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2995 const char *devname = netdev_get_name(netdev);
2996 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2997 const char *dp_port_name;
2999 if (netdev_vport_is_patch(netdev)) {
3000 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3004 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3005 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3006 odp_port_t port_no = ODPP_NONE;
3009 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3013 if (netdev_get_tunnel_config(netdev)) {
3014 simap_put(&ofproto->backer->tnl_backers,
3015 dp_port_name, odp_to_u32(port_no));
3019 if (netdev_get_tunnel_config(netdev)) {
3020 sset_add(&ofproto->ghost_ports, devname);
3022 sset_add(&ofproto->ports, devname);
3028 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3031 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3038 sset_find_and_delete(&ofproto->ghost_ports,
3039 netdev_get_name(ofport->up.netdev));
3040 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3041 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3042 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3044 /* The caller is going to close ofport->up.netdev. If this is a
3045 * bonded port, then the bond is using that netdev, so remove it
3046 * from the bond. The client will need to reconfigure everything
3047 * after deleting ports, so then the slave will get re-added. */
3048 bundle_remove(&ofport->up);
3055 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3057 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3062 error = netdev_get_stats(ofport->up.netdev, stats);
3064 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3067 /* ofproto->stats.tx_packets represents packets that we created
3068 * internally and sent to some port (e.g. packets sent with
3069 * send_packet()). Account for them as if they had come from
3070 * OFPP_LOCAL and got forwarded. */
3072 if (stats->rx_packets != UINT64_MAX) {
3073 stats->rx_packets += ofproto->stats.tx_packets;
3076 if (stats->rx_bytes != UINT64_MAX) {
3077 stats->rx_bytes += ofproto->stats.tx_bytes;
3080 /* ofproto->stats.rx_packets represents packets that were received on
3081 * some port and we processed internally and dropped (e.g. STP).
3082 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3084 if (stats->tx_packets != UINT64_MAX) {
3085 stats->tx_packets += ofproto->stats.rx_packets;
3088 if (stats->tx_bytes != UINT64_MAX) {
3089 stats->tx_bytes += ofproto->stats.rx_bytes;
3096 struct port_dump_state {
3101 struct ofproto_port port;
3106 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3108 *statep = xzalloc(sizeof(struct port_dump_state));
3113 port_dump_next(const struct ofproto *ofproto_, void *state_,
3114 struct ofproto_port *port)
3116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3117 struct port_dump_state *state = state_;
3118 const struct sset *sset;
3119 struct sset_node *node;
3121 if (state->has_port) {
3122 ofproto_port_destroy(&state->port);
3123 state->has_port = false;
3125 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3126 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3129 error = port_query_by_name(ofproto_, node->name, &state->port);
3131 *port = state->port;
3132 state->has_port = true;
3134 } else if (error != ENODEV) {
3139 if (!state->ghost) {
3140 state->ghost = true;
3143 return port_dump_next(ofproto_, state_, port);
3150 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3152 struct port_dump_state *state = state_;
3154 if (state->has_port) {
3155 ofproto_port_destroy(&state->port);
3162 port_poll(const struct ofproto *ofproto_, char **devnamep)
3164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3166 if (ofproto->port_poll_errno) {
3167 int error = ofproto->port_poll_errno;
3168 ofproto->port_poll_errno = 0;
3172 if (sset_is_empty(&ofproto->port_poll_set)) {
3176 *devnamep = sset_pop(&ofproto->port_poll_set);
3181 port_poll_wait(const struct ofproto *ofproto_)
3183 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3184 dpif_port_poll_wait(ofproto->backer->dpif);
3188 port_is_lacp_current(const struct ofport *ofport_)
3190 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3191 return (ofport->bundle && ofport->bundle->lacp
3192 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3196 /* Upcall handling. */
3198 struct flow_miss_op {
3199 struct dpif_op dpif_op;
3201 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3202 struct xlate_out xout;
3203 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3205 struct ofpbuf mask; /* Flow mask for "put" ops. */
3206 struct odputil_keybuf maskbuf;
3208 /* If this is a "put" op, then a pointer to the subfacet that should
3209 * be marked as uninstalled if the operation fails. */
3210 struct subfacet *subfacet;
3213 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3214 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3215 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3216 * return value of true). However, for short flows the cost of bookkeeping is
3217 * much higher than the benefits, so when the datapath holds a large number of
3218 * flows we impose some heuristics to decide which flows are likely to be worth
3221 flow_miss_should_make_facet(struct flow_miss *miss)
3223 struct dpif_backer *backer = miss->ofproto->backer;
3226 switch (flow_miss_model) {
3227 case OFPROTO_HANDLE_MISS_AUTO:
3229 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3231 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3235 if (!backer->governor) {
3238 n_subfacets = hmap_count(&backer->subfacets);
3239 if (n_subfacets * 2 <= flow_eviction_threshold) {
3243 backer->governor = governor_create();
3246 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3247 return governor_should_install_flow(backer->governor, hash,
3248 list_size(&miss->packets));
3251 /* Handles 'miss', which matches 'facet'. May add any required datapath
3252 * operations to 'ops', incrementing '*n_ops' for each new op.
3254 * All of the packets in 'miss' are considered to have arrived at time
3255 * 'miss->stats.used'. This is really important only for new facets: if we
3256 * just called time_msec() here, then the new subfacet or its packets could
3257 * look (occasionally) as though it was used some time after the facet was
3258 * used. That can make a one-packet flow look like it has a nonzero duration,
3259 * which looks odd in e.g. NetFlow statistics. */
3261 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3262 struct flow_miss_op *ops, size_t *n_ops)
3264 enum subfacet_path want_path;
3265 struct subfacet *subfacet;
3267 facet->packet_count += miss->stats.n_packets;
3268 facet->prev_packet_count += miss->stats.n_packets;
3269 facet->byte_count += miss->stats.n_bytes;
3270 facet->prev_byte_count += miss->stats.n_bytes;
3272 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3274 /* Don't install the flow if it's the result of the "userspace"
3275 * action for an already installed facet. This can occur when a
3276 * datapath flow with wildcards has a "userspace" action and flows
3277 * sent to userspace result in a different subfacet, which will then
3278 * be rejected as overlapping by the datapath. */
3279 if (miss->upcall_type == DPIF_UC_ACTION
3280 && !list_is_empty(&facet->subfacets)) {
3284 subfacet = subfacet_create(facet, miss);
3285 if (subfacet->path != want_path) {
3286 struct flow_miss_op *op = &ops[(*n_ops)++];
3287 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3289 subfacet->path = want_path;
3291 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3292 if (enable_megaflows) {
3293 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3294 &miss->flow, UINT32_MAX);
3297 op->xout_garbage = false;
3298 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3299 op->subfacet = subfacet;
3300 put->flags = DPIF_FP_CREATE;
3301 put->key = miss->key;
3302 put->key_len = miss->key_len;
3303 put->mask = op->mask.data;
3304 put->mask_len = op->mask.size;
3306 if (want_path == SF_FAST_PATH) {
3307 put->actions = facet->xout.odp_actions.data;
3308 put->actions_len = facet->xout.odp_actions.size;
3310 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3311 op->slow_stub, sizeof op->slow_stub,
3312 &put->actions, &put->actions_len);
3318 /* Handles flow miss 'miss'. May add any required datapath operations
3319 * to 'ops', incrementing '*n_ops' for each new op. */
3321 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3324 struct facet *facet;
3326 miss->ofproto->n_missed += list_size(&miss->packets);
3328 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3330 /* There does not exist a bijection between 'struct flow' and datapath
3331 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3332 * assumption used throughout the facet and subfacet handling code.
3333 * Since we have to handle these misses in userspace anyway, we simply
3334 * skip facet creation, avoiding the problem altogether. */
3335 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3336 || !flow_miss_should_make_facet(miss)) {
3340 facet = facet_create(miss);
3342 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3345 static struct drop_key *
3346 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3349 struct drop_key *drop_key;
3351 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3352 &backer->drop_keys) {
3353 if (drop_key->key_len == key_len
3354 && !memcmp(drop_key->key, key, key_len)) {
3362 drop_key_clear(struct dpif_backer *backer)
3364 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3365 struct drop_key *drop_key, *next;
3367 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3370 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3372 if (error && !VLOG_DROP_WARN(&rl)) {
3373 struct ds ds = DS_EMPTY_INITIALIZER;
3374 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3375 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3376 ovs_strerror(error), ds_cstr(&ds));
3380 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3381 drop_key_destroy(drop_key);
3384 udpif_drop_key_clear(backer->udpif);
3388 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3390 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3391 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3392 struct flow_miss *miss;
3395 /* Process each element in the to-do list, constructing the set of
3396 * operations to batch. */
3398 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3399 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3401 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3403 /* Execute batch. */
3404 for (i = 0; i < n_ops; i++) {
3405 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3407 dpif_operate(backer->dpif, dpif_ops, n_ops);
3409 for (i = 0; i < n_ops; i++) {
3410 if (dpif_ops[i]->error != 0
3411 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3412 && flow_miss_ops[i].subfacet) {
3413 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3415 COVERAGE_INC(subfacet_install_fail);
3417 /* Zero-out subfacet counters when installation failed, but
3418 * datapath reported hits. This should not happen and
3419 * indicates a bug, since if the datapath flow exists, we
3420 * should not be attempting to create a new subfacet. A
3421 * buggy datapath could trigger this, so just zero out the
3422 * counters and log an error. */
3423 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3424 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3425 "datapath reported hits");
3426 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3429 subfacet->path = SF_NOT_INSTALLED;
3435 handle_sflow_upcall(struct dpif_backer *backer,
3436 const struct dpif_upcall *upcall)
3438 struct ofproto_dpif *ofproto;
3439 union user_action_cookie cookie;
3441 odp_port_t odp_in_port;
3443 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3444 &flow, NULL, &ofproto, &odp_in_port)
3445 || !ofproto->sflow) {
3449 memset(&cookie, 0, sizeof cookie);
3450 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3451 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3452 odp_in_port, &cookie);
3456 handle_flow_sample_upcall(struct dpif_backer *backer,
3457 const struct dpif_upcall *upcall)
3459 struct ofproto_dpif *ofproto;
3460 union user_action_cookie cookie;
3463 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3464 &flow, NULL, &ofproto, NULL)
3465 || !ofproto->ipfix) {
3469 memset(&cookie, 0, sizeof cookie);
3470 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3472 /* The flow reflects exactly the contents of the packet. Sample
3473 * the packet using it. */
3474 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3475 cookie.flow_sample.collector_set_id,
3476 cookie.flow_sample.probability,
3477 cookie.flow_sample.obs_domain_id,
3478 cookie.flow_sample.obs_point_id);
3482 handle_ipfix_upcall(struct dpif_backer *backer,
3483 const struct dpif_upcall *upcall)
3485 struct ofproto_dpif *ofproto;
3488 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3489 &flow, NULL, &ofproto, NULL)
3490 || !ofproto->ipfix) {
3494 /* The flow reflects exactly the contents of the packet. Sample
3495 * the packet using it. */
3496 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3500 handle_upcalls(struct dpif_backer *backer)
3502 struct flow_miss_batch *fmb;
3505 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3506 struct upcall *upcall = upcall_next(backer->udpif);
3512 switch (upcall->type) {
3514 handle_sflow_upcall(backer, &upcall->dpif_upcall);
3517 case FLOW_SAMPLE_UPCALL:
3518 handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3522 handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3532 upcall_destroy(upcall);
3535 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3536 struct drop_key *drop_key = drop_key_next(backer->udpif);
3541 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3542 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3543 hash_bytes(drop_key->key, drop_key->key_len, 0));
3544 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3545 drop_key->key, drop_key->key_len,
3546 NULL, 0, NULL, 0, NULL);
3548 drop_key_destroy(drop_key);
3552 fmb = flow_miss_batch_next(backer->udpif);
3554 handle_flow_misses(backer, fmb);
3555 flow_miss_batch_destroy(fmb);
3559 /* Flow expiration. */
3561 static int subfacet_max_idle(const struct dpif_backer *);
3562 static void update_stats(struct dpif_backer *);
3563 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3564 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3566 /* This function is called periodically by run(). Its job is to collect
3567 * updates for the flows that have been installed into the datapath, most
3568 * importantly when they last were used, and then use that information to
3569 * expire flows that have not been used recently.
3571 * Returns the number of milliseconds after which it should be called again. */
3573 expire(struct dpif_backer *backer)
3575 struct ofproto_dpif *ofproto;
3579 /* Periodically clear out the drop keys in an effort to keep them
3580 * relatively few. */
3581 drop_key_clear(backer);
3583 /* Update stats for each flow in the backer. */
3584 update_stats(backer);
3586 n_subfacets = hmap_count(&backer->subfacets);
3588 struct subfacet *subfacet;
3589 long long int total, now;
3593 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3594 total += now - subfacet->created;
3596 backer->avg_subfacet_life += total / n_subfacets;
3598 backer->avg_subfacet_life /= 2;
3600 backer->avg_n_subfacet += n_subfacets;
3601 backer->avg_n_subfacet /= 2;
3603 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3605 max_idle = subfacet_max_idle(backer);
3606 expire_subfacets(backer, max_idle);
3608 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3609 struct rule *rule, *next_rule;
3611 if (ofproto->backer != backer) {
3615 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3617 ovs_mutex_lock(&ofproto_mutex);
3618 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3619 &ofproto->up.expirable) {
3620 rule_expire(rule_dpif_cast(rule));
3622 ovs_mutex_unlock(&ofproto_mutex);
3624 /* All outstanding data in existing flows has been accounted, so it's a
3625 * good time to do bond rebalancing. */
3626 if (ofproto->has_bonded_bundles) {
3627 struct ofbundle *bundle;
3629 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3631 bond_rebalance(bundle->bond);
3637 return MIN(max_idle, 1000);
3640 /* Updates flow table statistics given that the datapath just reported 'stats'
3641 * as 'subfacet''s statistics. */
3643 update_subfacet_stats(struct subfacet *subfacet,
3644 const struct dpif_flow_stats *stats)
3646 struct facet *facet = subfacet->facet;
3647 struct dpif_flow_stats diff;
3649 diff.tcp_flags = stats->tcp_flags;
3650 diff.used = stats->used;
3652 if (stats->n_packets >= subfacet->dp_packet_count) {
3653 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3655 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3659 if (stats->n_bytes >= subfacet->dp_byte_count) {
3660 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3662 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3666 facet->ofproto->n_hit += diff.n_packets;
3667 subfacet->dp_packet_count = stats->n_packets;
3668 subfacet->dp_byte_count = stats->n_bytes;
3669 subfacet_update_stats(subfacet, &diff);
3671 if (facet->accounted_bytes < facet->byte_count) {
3673 facet_account(facet);
3674 facet->accounted_bytes = facet->byte_count;
3678 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3679 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3681 delete_unexpected_flow(struct dpif_backer *backer,
3682 const struct nlattr *key, size_t key_len)
3684 if (!VLOG_DROP_WARN(&rl)) {
3688 odp_flow_key_format(key, key_len, &s);
3689 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3693 COVERAGE_INC(facet_unexpected);
3694 dpif_flow_del(backer->dpif, key, key_len, NULL);
3697 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3699 * This function also pushes statistics updates to rules which each facet
3700 * resubmits into. Generally these statistics will be accurate. However, if a
3701 * facet changes the rule it resubmits into at some time in between
3702 * update_stats() runs, it is possible that statistics accrued to the
3703 * old rule will be incorrectly attributed to the new rule. This could be
3704 * avoided by calling update_stats() whenever rules are created or
3705 * deleted. However, the performance impact of making so many calls to the
3706 * datapath do not justify the benefit of having perfectly accurate statistics.
3708 * In addition, this function maintains per ofproto flow hit counts. The patch
3709 * port is not treated specially. e.g. A packet ingress from br0 patched into
3710 * br1 will increase the hit count of br0 by 1, however, does not affect
3711 * the hit or miss counts of br1.
3714 update_stats(struct dpif_backer *backer)
3716 const struct dpif_flow_stats *stats;
3717 struct dpif_flow_dump dump;
3718 const struct nlattr *key, *mask;
3719 size_t key_len, mask_len;
3721 dpif_flow_dump_start(&dump, backer->dpif);
3722 while (dpif_flow_dump_next(&dump, &key, &key_len,
3723 &mask, &mask_len, NULL, NULL, &stats)) {
3724 struct subfacet *subfacet;
3727 key_hash = odp_flow_key_hash(key, key_len);
3728 subfacet = subfacet_find(backer, key, key_len, key_hash);
3729 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3731 update_subfacet_stats(subfacet, stats);
3735 /* Stats are updated per-packet. */
3738 case SF_NOT_INSTALLED:
3740 delete_unexpected_flow(backer, key, key_len);
3745 dpif_flow_dump_done(&dump);
3747 update_moving_averages(backer);
3750 /* Calculates and returns the number of milliseconds of idle time after which
3751 * subfacets should expire from the datapath. When a subfacet expires, we fold
3752 * its statistics into its facet, and when a facet's last subfacet expires, we
3753 * fold its statistic into its rule. */
3755 subfacet_max_idle(const struct dpif_backer *backer)
3758 * Idle time histogram.
3760 * Most of the time a switch has a relatively small number of subfacets.
3761 * When this is the case we might as well keep statistics for all of them
3762 * in userspace and to cache them in the kernel datapath for performance as
3765 * As the number of subfacets increases, the memory required to maintain
3766 * statistics about them in userspace and in the kernel becomes
3767 * significant. However, with a large number of subfacets it is likely
3768 * that only a few of them are "heavy hitters" that consume a large amount
3769 * of bandwidth. At this point, only heavy hitters are worth caching in
3770 * the kernel and maintaining in userspaces; other subfacets we can
3773 * The technique used to compute the idle time is to build a histogram with
3774 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3775 * that is installed in the kernel gets dropped in the appropriate bucket.
3776 * After the histogram has been built, we compute the cutoff so that only
3777 * the most-recently-used 1% of subfacets (but at least
3778 * flow_eviction_threshold flows) are kept cached. At least
3779 * the most-recently-used bucket of subfacets is kept, so actually an
3780 * arbitrary number of subfacets can be kept in any given expiration run
3781 * (though the next run will delete most of those unless they receive
3784 * This requires a second pass through the subfacets, in addition to the
3785 * pass made by update_stats(), because the former function never looks at
3786 * uninstallable subfacets.
3788 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3789 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3790 int buckets[N_BUCKETS] = { 0 };
3791 int total, subtotal, bucket;
3792 struct subfacet *subfacet;
3796 total = hmap_count(&backer->subfacets);
3797 if (total <= flow_eviction_threshold) {
3798 return N_BUCKETS * BUCKET_WIDTH;
3801 /* Build histogram. */
3803 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3804 long long int idle = now - subfacet->used;
3805 int bucket = (idle <= 0 ? 0
3806 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3807 : (unsigned int) idle / BUCKET_WIDTH);
3811 /* Find the first bucket whose flows should be expired. */
3812 subtotal = bucket = 0;
3814 subtotal += buckets[bucket++];
3815 } while (bucket < N_BUCKETS &&
3816 subtotal < MAX(flow_eviction_threshold, total / 100));
3818 if (VLOG_IS_DBG_ENABLED()) {
3822 ds_put_cstr(&s, "keep");
3823 for (i = 0; i < N_BUCKETS; i++) {
3825 ds_put_cstr(&s, ", drop");
3828 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3831 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3835 return bucket * BUCKET_WIDTH;
3839 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3841 /* Cutoff time for most flows. */
3842 long long int normal_cutoff = time_msec() - dp_max_idle;
3844 /* We really want to keep flows for special protocols around, so use a more
3845 * conservative cutoff. */
3846 long long int special_cutoff = time_msec() - 10000;
3848 struct subfacet *subfacet, *next_subfacet;
3849 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3853 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3854 &backer->subfacets) {
3855 long long int cutoff;
3857 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3861 if (subfacet->used < cutoff) {
3862 if (subfacet->path != SF_NOT_INSTALLED) {
3863 batch[n_batch++] = subfacet;
3864 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3865 subfacet_destroy_batch(backer, batch, n_batch);
3869 subfacet_destroy(subfacet);
3875 subfacet_destroy_batch(backer, batch, n_batch);
3879 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3880 * then delete it entirely. */
3882 rule_expire(struct rule_dpif *rule)
3883 OVS_REQUIRES(ofproto_mutex)
3885 uint16_t idle_timeout, hard_timeout;
3886 long long int now = time_msec();
3889 ovs_assert(!rule->up.pending);
3891 /* Has 'rule' expired? */
3892 ovs_mutex_lock(&rule->up.mutex);
3893 hard_timeout = rule->up.hard_timeout;
3894 idle_timeout = rule->up.idle_timeout;
3895 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3896 reason = OFPRR_HARD_TIMEOUT;
3897 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3898 reason = OFPRR_IDLE_TIMEOUT;
3902 ovs_mutex_unlock(&rule->up.mutex);
3905 COVERAGE_INC(ofproto_dpif_expired);
3906 ofproto_rule_expire(&rule->up, reason);
3912 /* Creates and returns a new facet based on 'miss'.
3914 * The caller must already have determined that no facet with an identical
3915 * 'miss->flow' exists in 'miss->ofproto'.
3917 * 'rule' and 'xout' must have been created based on 'miss'.
3919 * 'facet'' statistics are initialized based on 'stats'.
3921 * The facet will initially have no subfacets. The caller should create (at
3922 * least) one subfacet with subfacet_create(). */
3923 static struct facet *
3924 facet_create(const struct flow_miss *miss)
3926 struct ofproto_dpif *ofproto = miss->ofproto;
3927 struct facet *facet;
3930 facet = xzalloc(sizeof *facet);
3931 facet->ofproto = miss->ofproto;
3932 facet->used = miss->stats.used;
3933 facet->flow = miss->flow;
3934 facet->learn_rl = time_msec() + 500;
3936 list_init(&facet->subfacets);
3937 netflow_flow_init(&facet->nf_flow);
3938 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3940 xlate_out_copy(&facet->xout, &miss->xout);
3942 match_init(&match, &facet->flow, &facet->xout.wc);
3943 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3944 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3945 classifier_insert(&ofproto->facets, &facet->cr);
3946 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3948 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3953 facet_free(struct facet *facet)
3956 xlate_out_uninit(&facet->xout);
3961 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3962 * 'packet', which arrived on 'in_port'. */
3964 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3965 const struct nlattr *odp_actions, size_t actions_len,
3966 struct ofpbuf *packet)
3968 struct odputil_keybuf keybuf;
3972 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3973 odp_flow_key_from_flow(&key, flow,
3974 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3976 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3977 odp_actions, actions_len, packet);
3981 /* Remove 'facet' from its ofproto and free up the associated memory:
3983 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3984 * rule's statistics, via subfacet_uninstall().
3986 * - Removes 'facet' from its rule and from ofproto->facets.
3989 facet_remove(struct facet *facet)
3991 struct subfacet *subfacet, *next_subfacet;
3993 ovs_assert(!list_is_empty(&facet->subfacets));
3995 /* First uninstall all of the subfacets to get final statistics. */
3996 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3997 subfacet_uninstall(subfacet);
4000 /* Flush the final stats to the rule.
4002 * This might require us to have at least one subfacet around so that we
4003 * can use its actions for accounting in facet_account(), which is why we
4004 * have uninstalled but not yet destroyed the subfacets. */
4005 facet_flush_stats(facet);
4007 /* Now we're really all done so destroy everything. */
4008 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4009 &facet->subfacets) {
4010 subfacet_destroy__(subfacet);
4012 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4013 classifier_remove(&facet->ofproto->facets, &facet->cr);
4014 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4015 cls_rule_destroy(&facet->cr);
4019 /* Feed information from 'facet' back into the learning table to keep it in
4020 * sync with what is actually flowing through the datapath. */
4022 facet_learn(struct facet *facet)
4024 long long int now = time_msec();
4026 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4030 facet->learn_rl = now + 500;
4032 if (!facet->xout.has_learn
4033 && !facet->xout.has_normal
4034 && (!facet->xout.has_fin_timeout
4035 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4039 facet_push_stats(facet, true);
4043 facet_account(struct facet *facet)
4045 const struct nlattr *a;
4050 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4053 n_bytes = facet->byte_count - facet->accounted_bytes;
4055 /* This loop feeds byte counters to bond_account() for rebalancing to use
4056 * as a basis. We also need to track the actual VLAN on which the packet
4057 * is going to be sent to ensure that it matches the one passed to
4058 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4061 * We use the actions from an arbitrary subfacet because they should all
4062 * be equally valid for our purpose. */
4063 vlan_tci = facet->flow.vlan_tci;
4064 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4065 facet->xout.odp_actions.size) {
4066 const struct ovs_action_push_vlan *vlan;
4067 struct ofport_dpif *port;
4069 switch (nl_attr_type(a)) {
4070 case OVS_ACTION_ATTR_OUTPUT:
4071 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4072 if (port && port->bundle && port->bundle->bond) {
4073 bond_account(port->bundle->bond, &facet->flow,
4074 vlan_tci_to_vid(vlan_tci), n_bytes);
4078 case OVS_ACTION_ATTR_POP_VLAN:
4079 vlan_tci = htons(0);
4082 case OVS_ACTION_ATTR_PUSH_VLAN:
4083 vlan = nl_attr_get(a);
4084 vlan_tci = vlan->vlan_tci;
4090 /* Returns true if the only action for 'facet' is to send to the controller.
4091 * (We don't report NetFlow expiration messages for such facets because they
4092 * are just part of the control logic for the network, not real traffic). */
4094 facet_is_controller_flow(struct facet *facet)
4097 struct ofproto_dpif *ofproto = facet->ofproto;
4098 const struct ofpact *ofpacts;
4099 struct rule_actions *actions;
4100 struct rule_dpif *rule;
4104 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4105 actions = rule_dpif_get_actions(rule);
4106 rule_dpif_unref(rule);
4108 ofpacts_len = actions->ofpacts_len;
4109 ofpacts = actions->ofpacts;
4110 is_controller = ofpacts_len > 0
4111 && ofpacts->type == OFPACT_CONTROLLER
4112 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4113 rule_actions_unref(actions);
4115 return is_controller;
4120 /* Folds all of 'facet''s statistics into its rule. Also updates the
4121 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4122 * 'facet''s statistics in the datapath should have been zeroed and folded into
4123 * its packet and byte counts before this function is called. */
4125 facet_flush_stats(struct facet *facet)
4127 struct ofproto_dpif *ofproto = facet->ofproto;
4128 struct subfacet *subfacet;
4130 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4131 ovs_assert(!subfacet->dp_byte_count);
4132 ovs_assert(!subfacet->dp_packet_count);
4135 facet_push_stats(facet, false);
4136 if (facet->accounted_bytes < facet->byte_count) {
4137 facet_account(facet);
4138 facet->accounted_bytes = facet->byte_count;
4141 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4142 struct ofexpired expired;
4143 expired.flow = facet->flow;
4144 expired.packet_count = facet->packet_count;
4145 expired.byte_count = facet->byte_count;
4146 expired.used = facet->used;
4147 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4150 /* Reset counters to prevent double counting if 'facet' ever gets
4152 facet_reset_counters(facet);
4154 netflow_flow_clear(&facet->nf_flow);
4155 facet->tcp_flags = 0;
4158 /* Searches 'ofproto''s table of facets for one which would be responsible for
4159 * 'flow'. Returns it if found, otherwise a null pointer.
4161 * The returned facet might need revalidation; use facet_lookup_valid()
4162 * instead if that is important. */
4163 static struct facet *
4164 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4166 struct cls_rule *cr;
4168 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4169 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4170 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4171 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4174 /* Searches 'ofproto''s table of facets for one capable that covers
4175 * 'flow'. Returns it if found, otherwise a null pointer.
4177 * The returned facet is guaranteed to be valid. */
4178 static struct facet *
4179 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4181 struct facet *facet;
4183 facet = facet_find(ofproto, flow);
4185 && ofproto->backer->need_revalidate
4186 && !facet_revalidate(facet)) {
4194 facet_check_consistency(struct facet *facet)
4196 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4198 struct xlate_out xout;
4199 struct xlate_in xin;
4201 struct rule_dpif *rule;
4204 /* Check the datapath actions for consistency. */
4205 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4206 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4207 xlate_actions(&xin, &xout);
4208 rule_dpif_unref(rule);
4210 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4211 && facet->xout.slow == xout.slow;
4212 if (!ok && !VLOG_DROP_WARN(&rl)) {
4213 struct ds s = DS_EMPTY_INITIALIZER;
4215 flow_format(&s, &facet->flow);
4216 ds_put_cstr(&s, ": inconsistency in facet");
4218 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4219 ds_put_cstr(&s, " (actions were: ");
4220 format_odp_actions(&s, facet->xout.odp_actions.data,
4221 facet->xout.odp_actions.size);
4222 ds_put_cstr(&s, ") (correct actions: ");
4223 format_odp_actions(&s, xout.odp_actions.data,
4224 xout.odp_actions.size);
4225 ds_put_char(&s, ')');
4228 if (facet->xout.slow != xout.slow) {
4229 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4234 xlate_out_uninit(&xout);
4239 /* Re-searches the classifier for 'facet':
4241 * - If the rule found is different from 'facet''s current rule, moves
4242 * 'facet' to the new rule and recompiles its actions.
4244 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4245 * where it is and recompiles its actions anyway.
4247 * - If any of 'facet''s subfacets correspond to a new flow according to
4248 * xlate_receive(), 'facet' is removed.
4250 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4252 facet_revalidate(struct facet *facet)
4254 struct ofproto_dpif *ofproto = facet->ofproto;
4255 struct rule_dpif *new_rule;
4256 struct subfacet *subfacet;
4257 struct flow_wildcards wc;
4258 struct xlate_out xout;
4259 struct xlate_in xin;
4261 COVERAGE_INC(facet_revalidate);
4263 /* Check that child subfacets still correspond to this facet. Tunnel
4264 * configuration changes could cause a subfacet's OpenFlow in_port to
4266 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4267 struct ofproto_dpif *recv_ofproto;
4268 struct flow recv_flow;
4271 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4272 subfacet->key_len, &recv_flow, NULL,
4273 &recv_ofproto, NULL);
4275 || recv_ofproto != ofproto
4276 || facet != facet_find(ofproto, &recv_flow)) {
4277 facet_remove(facet);
4282 flow_wildcards_init_catchall(&wc);
4283 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4285 /* Calculate new datapath actions.
4287 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4288 * emit a NetFlow expiration and, if so, we need to have the old state
4289 * around to properly compose it. */
4290 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4291 xlate_actions(&xin, &xout);
4292 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4294 /* A facet's slow path reason should only change under dramatic
4295 * circumstances. Rather than try to update everything, it's simpler to
4296 * remove the facet and start over.
4298 * More importantly, if a facet's wildcards change, it will be relatively
4299 * difficult to figure out if its subfacets still belong to it, and if not
4300 * which facet they may belong to. Again, to avoid the complexity, we
4301 * simply give up instead. */
4302 if (facet->xout.slow != xout.slow
4303 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4304 facet_remove(facet);
4305 xlate_out_uninit(&xout);
4306 rule_dpif_unref(new_rule);
4310 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4311 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4312 if (subfacet->path == SF_FAST_PATH) {
4313 struct dpif_flow_stats stats;
4315 subfacet_install(subfacet, &xout.odp_actions, &stats);
4316 subfacet_update_stats(subfacet, &stats);
4320 facet_flush_stats(facet);
4322 ofpbuf_clear(&facet->xout.odp_actions);
4323 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4324 xout.odp_actions.size);
4327 /* Update 'facet' now that we've taken care of all the old state. */
4328 facet->xout.slow = xout.slow;
4329 facet->xout.has_learn = xout.has_learn;
4330 facet->xout.has_normal = xout.has_normal;
4331 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4332 facet->xout.nf_output_iface = xout.nf_output_iface;
4333 facet->xout.mirrors = xout.mirrors;
4334 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4336 ovs_mutex_lock(&new_rule->up.mutex);
4337 facet->used = MAX(facet->used, new_rule->up.created);
4338 ovs_mutex_unlock(&new_rule->up.mutex);
4340 xlate_out_uninit(&xout);
4341 rule_dpif_unref(new_rule);
4346 facet_reset_counters(struct facet *facet)
4348 facet->packet_count = 0;
4349 facet->byte_count = 0;
4350 facet->prev_packet_count = 0;
4351 facet->prev_byte_count = 0;
4352 facet->accounted_bytes = 0;
4356 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4357 struct dpif_flow_stats *stats, bool may_learn)
4359 struct ofport_dpif *in_port;
4360 struct rule_dpif *rule;
4361 struct xlate_in xin;
4363 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4364 if (in_port && in_port->is_tunnel) {
4365 netdev_vport_inc_rx(in_port->up.netdev, stats);
4368 rule_dpif_lookup(ofproto, flow, NULL, &rule);
4369 rule_dpif_credit_stats(rule, stats);
4370 xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4371 xin.resubmit_stats = stats;
4372 xin.may_learn = may_learn;
4373 xlate_actions_for_side_effects(&xin);
4374 rule_dpif_unref(rule);
4378 facet_push_stats(struct facet *facet, bool may_learn)
4380 struct dpif_flow_stats stats;
4382 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4383 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4384 ovs_assert(facet->used >= facet->prev_used);
4386 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4387 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4388 stats.used = facet->used;
4389 stats.tcp_flags = facet->tcp_flags;
4391 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4392 facet->prev_packet_count = facet->packet_count;
4393 facet->prev_byte_count = facet->byte_count;
4394 facet->prev_used = facet->used;
4396 netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4398 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4399 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4400 stats.n_packets, stats.n_bytes);
4401 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4406 push_all_stats__(bool run_fast)
4408 static long long int rl = LLONG_MIN;
4409 struct ofproto_dpif *ofproto;
4411 if (time_msec() < rl) {
4415 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4416 struct cls_cursor cursor;
4417 struct facet *facet;
4419 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4420 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4421 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4422 facet_push_stats(facet, false);
4427 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4430 rl = time_msec() + 100;
4434 push_all_stats(void)
4436 push_all_stats__(true);
4440 rule_dpif_credit_stats(struct rule_dpif *rule,
4441 const struct dpif_flow_stats *stats)
4443 ovs_mutex_lock(&rule->stats_mutex);
4444 rule->packet_count += stats->n_packets;
4445 rule->byte_count += stats->n_bytes;
4446 rule->up.used = MAX(rule->up.used, stats->used);
4447 ovs_mutex_unlock(&rule->stats_mutex);
4451 rule_dpif_fail_open(const struct rule_dpif *rule)
4453 return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4457 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4458 OVS_REQUIRES(rule->up.mutex)
4460 return rule->up.flow_cookie;
4464 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4465 uint16_t hard_timeout)
4467 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4470 /* Returns 'rule''s actions. The caller owns a reference on the returned
4471 * actions and must eventually release it (with rule_actions_unref()) to avoid
4473 struct rule_actions *
4474 rule_dpif_get_actions(const struct rule_dpif *rule)
4476 return rule_get_actions(&rule->up);
4481 static struct subfacet *
4482 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4483 size_t key_len, uint32_t key_hash)
4485 struct subfacet *subfacet;
4487 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4488 &backer->subfacets) {
4489 if (subfacet->key_len == key_len
4490 && !memcmp(key, subfacet->key, key_len)) {
4498 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4499 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4500 * existing subfacet if there is one, otherwise creates and returns a
4502 static struct subfacet *
4503 subfacet_create(struct facet *facet, struct flow_miss *miss)
4505 struct dpif_backer *backer = miss->ofproto->backer;
4506 enum odp_key_fitness key_fitness = miss->key_fitness;
4507 const struct nlattr *key = miss->key;
4508 size_t key_len = miss->key_len;
4510 struct subfacet *subfacet;
4512 key_hash = odp_flow_key_hash(key, key_len);
4514 if (list_is_empty(&facet->subfacets)) {
4515 subfacet = &facet->one_subfacet;
4517 subfacet = subfacet_find(backer, key, key_len, key_hash);
4519 if (subfacet->facet == facet) {
4523 /* This shouldn't happen. */
4524 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4525 subfacet_destroy(subfacet);
4528 subfacet = xmalloc(sizeof *subfacet);
4531 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4532 list_push_back(&facet->subfacets, &subfacet->list_node);
4533 subfacet->facet = facet;
4534 subfacet->key_fitness = key_fitness;
4535 subfacet->key = xmemdup(key, key_len);
4536 subfacet->key_len = key_len;
4537 subfacet->used = miss->stats.used;
4538 subfacet->created = subfacet->used;
4539 subfacet->dp_packet_count = 0;
4540 subfacet->dp_byte_count = 0;
4541 subfacet->path = SF_NOT_INSTALLED;
4542 subfacet->backer = backer;
4544 backer->subfacet_add_count++;
4548 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4549 * its facet within 'ofproto', and frees it. */
4551 subfacet_destroy__(struct subfacet *subfacet)
4553 struct facet *facet = subfacet->facet;
4554 struct ofproto_dpif *ofproto = facet->ofproto;
4556 /* Update ofproto stats before uninstall the subfacet. */
4557 ofproto->backer->subfacet_del_count++;
4559 subfacet_uninstall(subfacet);
4560 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4561 list_remove(&subfacet->list_node);
4562 free(subfacet->key);
4563 if (subfacet != &facet->one_subfacet) {
4568 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4569 * last remaining subfacet in its facet destroys the facet too. */
4571 subfacet_destroy(struct subfacet *subfacet)
4573 struct facet *facet = subfacet->facet;
4575 if (list_is_singleton(&facet->subfacets)) {
4576 /* facet_remove() needs at least one subfacet (it will remove it). */
4577 facet_remove(facet);
4579 subfacet_destroy__(subfacet);
4584 subfacet_destroy_batch(struct dpif_backer *backer,
4585 struct subfacet **subfacets, int n)
4587 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4588 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4589 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4592 for (i = 0; i < n; i++) {
4593 ops[i].type = DPIF_OP_FLOW_DEL;
4594 ops[i].u.flow_del.key = subfacets[i]->key;
4595 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4596 ops[i].u.flow_del.stats = &stats[i];
4600 dpif_operate(backer->dpif, opsp, n);
4601 for (i = 0; i < n; i++) {
4602 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4603 subfacets[i]->path = SF_NOT_INSTALLED;
4604 subfacet_destroy(subfacets[i]);
4609 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4610 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4611 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4612 * since 'subfacet' was last updated.
4614 * Returns 0 if successful, otherwise a positive errno value. */
4616 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4617 struct dpif_flow_stats *stats)
4619 struct facet *facet = subfacet->facet;
4620 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4621 const struct nlattr *actions = odp_actions->data;
4622 size_t actions_len = odp_actions->size;
4623 struct odputil_keybuf maskbuf;
4626 uint64_t slow_path_stub[128 / 8];
4627 enum dpif_flow_put_flags flags;
4630 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4633 flags |= DPIF_FP_ZERO_STATS;
4636 if (path == SF_SLOW_PATH) {
4637 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4638 slow_path_stub, sizeof slow_path_stub,
4639 &actions, &actions_len);
4642 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4643 if (enable_megaflows) {
4644 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4645 &facet->flow, UINT32_MAX);
4648 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4649 subfacet->key_len, mask.data, mask.size,
4650 actions, actions_len, stats);
4653 subfacet_reset_dp_stats(subfacet, stats);
4657 COVERAGE_INC(subfacet_install_fail);
4659 subfacet->path = path;
4664 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4666 subfacet_uninstall(struct subfacet *subfacet)
4668 if (subfacet->path != SF_NOT_INSTALLED) {
4669 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4670 struct dpif_flow_stats stats;
4673 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4674 subfacet->key_len, &stats);
4675 subfacet_reset_dp_stats(subfacet, &stats);
4677 subfacet_update_stats(subfacet, &stats);
4679 subfacet->path = SF_NOT_INSTALLED;
4681 ovs_assert(subfacet->dp_packet_count == 0);
4682 ovs_assert(subfacet->dp_byte_count == 0);
4686 /* Resets 'subfacet''s datapath statistics counters. This should be called
4687 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4688 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4689 * was reset in the datapath. 'stats' will be modified to include only
4690 * statistics new since 'subfacet' was last updated. */
4692 subfacet_reset_dp_stats(struct subfacet *subfacet,
4693 struct dpif_flow_stats *stats)
4696 && subfacet->dp_packet_count <= stats->n_packets
4697 && subfacet->dp_byte_count <= stats->n_bytes) {
4698 stats->n_packets -= subfacet->dp_packet_count;
4699 stats->n_bytes -= subfacet->dp_byte_count;
4702 subfacet->dp_packet_count = 0;
4703 subfacet->dp_byte_count = 0;
4706 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4708 * Because of the meaning of a subfacet's counters, it only makes sense to do
4709 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4710 * represents a packet that was sent by hand or if it represents statistics
4711 * that have been cleared out of the datapath. */
4713 subfacet_update_stats(struct subfacet *subfacet,
4714 const struct dpif_flow_stats *stats)
4716 if (stats->n_packets || stats->used > subfacet->used) {
4717 struct facet *facet = subfacet->facet;
4719 subfacet->used = MAX(subfacet->used, stats->used);
4720 facet->used = MAX(facet->used, stats->used);
4721 facet->packet_count += stats->n_packets;
4722 facet->byte_count += stats->n_bytes;
4723 facet->tcp_flags |= stats->tcp_flags;
4729 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4730 * the fields that were relevant as part of the lookup. */
4732 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4733 struct flow_wildcards *wc, struct rule_dpif **rule)
4735 struct ofport_dpif *port;
4737 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4740 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4742 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4743 flow->in_port.ofp_port);
4746 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4747 ofproto->no_packet_in_rule, rule);
4751 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4752 const struct flow *flow, struct flow_wildcards *wc,
4753 uint8_t table_id, struct rule_dpif **rule)
4755 const struct cls_rule *cls_rule;
4756 struct classifier *cls;
4760 if (table_id >= N_TABLES) {
4765 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4766 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4769 cls = &ofproto->up.tables[table_id].cls;
4770 ovs_rwlock_rdlock(&cls->rwlock);
4771 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4772 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4773 /* We must pretend that transport ports are unavailable. */
4774 struct flow ofpc_normal_flow = *flow;
4775 ofpc_normal_flow.tp_src = htons(0);
4776 ofpc_normal_flow.tp_dst = htons(0);
4777 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4778 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4779 cls_rule = &ofproto->drop_frags_rule->up.cr;
4781 flow_wildcards_init_exact(wc);
4784 cls_rule = classifier_lookup(cls, flow, wc);
4787 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4788 rule_dpif_ref(*rule);
4789 ovs_rwlock_unlock(&cls->rwlock);
4791 return *rule != NULL;
4794 /* Given a port configuration (specified as zero if there's no port), chooses
4795 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4796 * flow table miss. */
4798 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4799 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4801 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4802 rule_dpif_ref(*rule);
4806 rule_dpif_ref(struct rule_dpif *rule)
4809 ofproto_rule_ref(&rule->up);
4814 rule_dpif_unref(struct rule_dpif *rule)
4817 ofproto_rule_unref(&rule->up);
4822 complete_operation(struct rule_dpif *rule)
4823 OVS_REQUIRES(ofproto_mutex)
4825 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4827 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4828 ofoperation_complete(rule->up.pending, 0);
4831 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4833 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4836 static struct rule *
4839 struct rule_dpif *rule = xmalloc(sizeof *rule);
4844 rule_dealloc(struct rule *rule_)
4846 struct rule_dpif *rule = rule_dpif_cast(rule_);
4851 rule_construct(struct rule *rule_)
4853 struct rule_dpif *rule = rule_dpif_cast(rule_);
4854 ovs_mutex_init(&rule->stats_mutex);
4855 ovs_mutex_lock(&rule->stats_mutex);
4856 rule->packet_count = 0;
4857 rule->byte_count = 0;
4858 ovs_mutex_unlock(&rule->stats_mutex);
4863 rule_insert(struct rule *rule_)
4864 OVS_REQUIRES(ofproto_mutex)
4866 struct rule_dpif *rule = rule_dpif_cast(rule_);
4867 complete_operation(rule);
4871 rule_delete(struct rule *rule_)
4872 OVS_REQUIRES(ofproto_mutex)
4874 struct rule_dpif *rule = rule_dpif_cast(rule_);
4875 complete_operation(rule);
4879 rule_destruct(struct rule *rule_)
4881 struct rule_dpif *rule = rule_dpif_cast(rule_);
4882 ovs_mutex_destroy(&rule->stats_mutex);
4886 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4888 struct rule_dpif *rule = rule_dpif_cast(rule_);
4890 /* push_all_stats() can handle flow misses which, when using the learn
4891 * action, can cause rules to be added and deleted. This can corrupt our
4892 * caller's datastructures which assume that rule_get_stats() doesn't have
4893 * an impact on the flow table. To be safe, we disable miss handling. */
4894 push_all_stats__(false);
4896 /* Start from historical data for 'rule' itself that are no longer tracked
4897 * in facets. This counts, for example, facets that have expired. */
4898 ovs_mutex_lock(&rule->stats_mutex);
4899 *packets = rule->packet_count;
4900 *bytes = rule->byte_count;
4901 ovs_mutex_unlock(&rule->stats_mutex);
4905 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4906 struct ofpbuf *packet)
4908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4909 struct dpif_flow_stats stats;
4910 struct xlate_out xout;
4911 struct xlate_in xin;
4913 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4914 rule_dpif_credit_stats(rule, &stats);
4916 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4917 xin.resubmit_stats = &stats;
4918 xlate_actions(&xin, &xout);
4920 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4921 xout.odp_actions.size, packet);
4923 xlate_out_uninit(&xout);
4927 rule_execute(struct rule *rule, const struct flow *flow,
4928 struct ofpbuf *packet)
4930 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4931 ofpbuf_delete(packet);
4936 rule_modify_actions(struct rule *rule_, bool reset_counters)
4937 OVS_REQUIRES(ofproto_mutex)
4939 struct rule_dpif *rule = rule_dpif_cast(rule_);
4941 if (reset_counters) {
4942 ovs_mutex_lock(&rule->stats_mutex);
4943 rule->packet_count = 0;
4944 rule->byte_count = 0;
4945 ovs_mutex_unlock(&rule->stats_mutex);
4948 complete_operation(rule);
4951 /* Sends 'packet' out 'ofport'.
4952 * May modify 'packet'.
4953 * Returns 0 if successful, otherwise a positive errno value. */
4955 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4958 uint64_t odp_actions_stub[1024 / 8];
4959 struct ofpbuf key, odp_actions;
4960 struct dpif_flow_stats stats;
4961 struct odputil_keybuf keybuf;
4962 struct ofpact_output output;
4963 struct xlate_out xout;
4964 struct xlate_in xin;
4966 union flow_in_port in_port_;
4969 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4970 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4972 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4973 in_port_.ofp_port = OFPP_NONE;
4974 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4975 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4977 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4979 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4980 output.port = ofport->up.ofp_port;
4983 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4984 xin.ofpacts_len = sizeof output;
4985 xin.ofpacts = &output.ofpact;
4986 xin.resubmit_stats = &stats;
4987 xlate_actions(&xin, &xout);
4989 error = dpif_execute(ofproto->backer->dpif,
4991 xout.odp_actions.data, xout.odp_actions.size,
4993 xlate_out_uninit(&xout);
4996 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4997 ofproto->up.name, netdev_get_name(ofport->up.netdev),
4998 ovs_strerror(error));
5001 ofproto->stats.tx_packets++;
5002 ofproto->stats.tx_bytes += packet->size;
5006 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5007 * The action will state 'slow' as the reason that the action is in the slow
5008 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5009 * dump-flows" output to see why a flow is in the slow path.)
5011 * The 'stub_size' bytes in 'stub' will be used to store the action.
5012 * 'stub_size' must be large enough for the action.
5014 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5017 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5018 enum slow_path_reason slow,
5019 uint64_t *stub, size_t stub_size,
5020 const struct nlattr **actionsp, size_t *actions_lenp)
5022 union user_action_cookie cookie;
5025 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5026 cookie.slow_path.unused = 0;
5027 cookie.slow_path.reason = slow;
5029 ofpbuf_use_stack(&buf, stub, stub_size);
5030 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5031 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5033 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5035 odp_port_t odp_port;
5038 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5039 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5040 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5042 *actionsp = buf.data;
5043 *actions_lenp = buf.size;
5047 set_frag_handling(struct ofproto *ofproto_,
5048 enum ofp_config_flags frag_handling)
5050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5051 if (frag_handling != OFPC_FRAG_REASM) {
5052 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5060 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5061 const struct flow *flow,
5062 const struct ofpact *ofpacts, size_t ofpacts_len)
5064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5065 struct odputil_keybuf keybuf;
5066 struct dpif_flow_stats stats;
5067 struct xlate_out xout;
5068 struct xlate_in xin;
5072 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5073 odp_flow_key_from_flow(&key, flow,
5074 ofp_port_to_odp_port(ofproto,
5075 flow->in_port.ofp_port));
5077 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5079 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5080 xin.resubmit_stats = &stats;
5081 xin.ofpacts_len = ofpacts_len;
5082 xin.ofpacts = ofpacts;
5084 xlate_actions(&xin, &xout);
5085 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5086 xout.odp_actions.data, xout.odp_actions.size, packet);
5087 xlate_out_uninit(&xout);
5095 set_netflow(struct ofproto *ofproto_,
5096 const struct netflow_options *netflow_options)
5098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5100 if (netflow_options) {
5101 if (!ofproto->netflow) {
5102 ofproto->netflow = netflow_create();
5103 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5105 return netflow_set_options(ofproto->netflow, netflow_options);
5106 } else if (ofproto->netflow) {
5107 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5108 netflow_destroy(ofproto->netflow);
5109 ofproto->netflow = NULL;
5116 get_netflow_ids(const struct ofproto *ofproto_,
5117 uint8_t *engine_type, uint8_t *engine_id)
5119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5121 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5125 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5127 if (!facet_is_controller_flow(facet) &&
5128 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5129 struct subfacet *subfacet;
5130 struct ofexpired expired;
5132 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5133 if (subfacet->path == SF_FAST_PATH) {
5134 struct dpif_flow_stats stats;
5136 subfacet_install(subfacet, &facet->xout.odp_actions,
5138 subfacet_update_stats(subfacet, &stats);
5142 expired.flow = facet->flow;
5143 expired.packet_count = facet->packet_count;
5144 expired.byte_count = facet->byte_count;
5145 expired.used = facet->used;
5146 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5151 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5153 struct cls_cursor cursor;
5154 struct facet *facet;
5156 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5157 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5158 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5159 send_active_timeout(ofproto, facet);
5161 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5164 static struct ofproto_dpif *
5165 ofproto_dpif_lookup(const char *name)
5167 struct ofproto_dpif *ofproto;
5169 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5170 hash_string(name, 0), &all_ofproto_dpifs) {
5171 if (!strcmp(ofproto->up.name, name)) {
5179 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5180 const char *argv[], void *aux OVS_UNUSED)
5182 struct ofproto_dpif *ofproto;
5185 ofproto = ofproto_dpif_lookup(argv[1]);
5187 unixctl_command_reply_error(conn, "no such bridge");
5190 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5191 mac_learning_flush(ofproto->ml);
5192 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5194 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5195 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5196 mac_learning_flush(ofproto->ml);
5197 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5201 unixctl_command_reply(conn, "table successfully flushed");
5204 static struct ofport_dpif *
5205 ofbundle_get_a_port(const struct ofbundle *bundle)
5207 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5212 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5213 const char *argv[], void *aux OVS_UNUSED)
5215 struct ds ds = DS_EMPTY_INITIALIZER;
5216 const struct ofproto_dpif *ofproto;
5217 const struct mac_entry *e;
5219 ofproto = ofproto_dpif_lookup(argv[1]);
5221 unixctl_command_reply_error(conn, "no such bridge");
5225 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5226 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5227 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5228 struct ofbundle *bundle = e->port.p;
5229 char name[OFP_MAX_PORT_NAME_LEN];
5231 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5233 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5234 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5235 mac_entry_age(ofproto->ml, e));
5237 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5238 unixctl_command_reply(conn, ds_cstr(&ds));
5243 struct xlate_out xout;
5244 struct xlate_in xin;
5250 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5252 struct rule_actions *actions;
5255 ds_put_char_multiple(result, '\t', level);
5257 ds_put_cstr(result, "No match\n");
5261 ovs_mutex_lock(&rule->up.mutex);
5262 cookie = rule->up.flow_cookie;
5263 ovs_mutex_unlock(&rule->up.mutex);
5265 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5266 rule ? rule->up.table_id : 0, ntohll(cookie));
5267 cls_rule_format(&rule->up.cr, result);
5268 ds_put_char(result, '\n');
5270 actions = rule_dpif_get_actions(rule);
5272 ds_put_char_multiple(result, '\t', level);
5273 ds_put_cstr(result, "OpenFlow ");
5274 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5275 ds_put_char(result, '\n');
5277 rule_actions_unref(actions);
5281 trace_format_flow(struct ds *result, int level, const char *title,
5282 struct trace_ctx *trace)
5284 ds_put_char_multiple(result, '\t', level);
5285 ds_put_format(result, "%s: ", title);
5286 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5287 ds_put_cstr(result, "unchanged");
5289 flow_format(result, &trace->xin.flow);
5290 trace->flow = trace->xin.flow;
5292 ds_put_char(result, '\n');
5296 trace_format_regs(struct ds *result, int level, const char *title,
5297 struct trace_ctx *trace)
5301 ds_put_char_multiple(result, '\t', level);
5302 ds_put_format(result, "%s:", title);
5303 for (i = 0; i < FLOW_N_REGS; i++) {
5304 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5306 ds_put_char(result, '\n');
5310 trace_format_odp(struct ds *result, int level, const char *title,
5311 struct trace_ctx *trace)
5313 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5315 ds_put_char_multiple(result, '\t', level);
5316 ds_put_format(result, "%s: ", title);
5317 format_odp_actions(result, odp_actions->data, odp_actions->size);
5318 ds_put_char(result, '\n');
5322 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5324 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5325 struct ds *result = trace->result;
5327 ds_put_char(result, '\n');
5328 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5329 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5330 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5331 trace_format_rule(result, recurse + 1, rule);
5335 trace_report(struct xlate_in *xin, const char *s, int recurse)
5337 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5338 struct ds *result = trace->result;
5340 ds_put_char_multiple(result, '\t', recurse);
5341 ds_put_cstr(result, s);
5342 ds_put_char(result, '\n');
5346 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5347 void *aux OVS_UNUSED)
5349 const struct dpif_backer *backer;
5350 struct ofproto_dpif *ofproto;
5351 struct ofpbuf odp_key, odp_mask;
5352 struct ofpbuf *packet;
5360 ofpbuf_init(&odp_key, 0);
5361 ofpbuf_init(&odp_mask, 0);
5363 /* Handle "-generate" or a hex string as the last argument. */
5364 if (!strcmp(argv[argc - 1], "-generate")) {
5365 packet = ofpbuf_new(0);
5368 const char *error = eth_from_hex(argv[argc - 1], &packet);
5371 } else if (argc == 4) {
5372 /* The 3-argument form must end in "-generate' or a hex string. */
5373 unixctl_command_reply_error(conn, error);
5378 /* Parse the flow and determine whether a datapath or
5379 * bridge is specified. If function odp_flow_key_from_string()
5380 * returns 0, the flow is a odp_flow. If function
5381 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5382 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5383 /* If the odp_flow is the second argument,
5384 * the datapath name is the first argument. */
5386 const char *dp_type;
5387 if (!strncmp(argv[1], "ovs-", 4)) {
5388 dp_type = argv[1] + 4;
5392 backer = shash_find_data(&all_dpif_backers, dp_type);
5394 unixctl_command_reply_error(conn, "Cannot find datapath "
5399 /* No datapath name specified, so there should be only one
5401 struct shash_node *node;
5402 if (shash_count(&all_dpif_backers) != 1) {
5403 unixctl_command_reply_error(conn, "Must specify datapath "
5404 "name, there is more than one type of datapath");
5407 node = shash_first(&all_dpif_backers);
5408 backer = node->data;
5411 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5412 NULL, &ofproto, NULL)) {
5413 unixctl_command_reply_error(conn, "Invalid datapath flow");
5416 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5417 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5419 unixctl_command_reply_error(conn, "Must specify bridge name");
5423 ofproto = ofproto_dpif_lookup(argv[1]);
5425 unixctl_command_reply_error(conn, "Unknown bridge name");
5429 unixctl_command_reply_error(conn, "Bad flow syntax");
5433 /* Generate a packet, if requested. */
5435 if (!packet->size) {
5436 flow_compose(packet, &flow);
5438 union flow_in_port in_port_;
5440 in_port_ = flow.in_port;
5441 ds_put_cstr(&result, "Packet: ");
5442 s = ofp_packet_to_string(packet->data, packet->size);
5443 ds_put_cstr(&result, s);
5446 /* Use the metadata from the flow and the packet argument
5447 * to reconstruct the flow. */
5448 flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5453 ofproto_trace(ofproto, &flow, packet, &result);
5454 unixctl_command_reply(conn, ds_cstr(&result));
5457 ds_destroy(&result);
5458 ofpbuf_delete(packet);
5459 ofpbuf_uninit(&odp_key);
5460 ofpbuf_uninit(&odp_mask);
5464 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5465 const struct ofpbuf *packet, struct ds *ds)
5467 struct rule_dpif *rule;
5468 struct flow_wildcards wc;
5470 ds_put_cstr(ds, "Flow: ");
5471 flow_format(ds, flow);
5472 ds_put_char(ds, '\n');
5474 flow_wildcards_init_catchall(&wc);
5475 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5477 trace_format_rule(ds, 0, rule);
5478 if (rule == ofproto->miss_rule) {
5479 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5480 } else if (rule == ofproto->no_packet_in_rule) {
5481 ds_put_cstr(ds, "\nNo match, packets dropped because "
5482 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5483 } else if (rule == ofproto->drop_frags_rule) {
5484 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5485 "and the fragment handling mode is \"drop\".\n");
5489 uint64_t odp_actions_stub[1024 / 8];
5490 struct ofpbuf odp_actions;
5491 struct trace_ctx trace;
5495 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5498 ofpbuf_use_stub(&odp_actions,
5499 odp_actions_stub, sizeof odp_actions_stub);
5500 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5501 trace.xin.resubmit_hook = trace_resubmit;
5502 trace.xin.report_hook = trace_report;
5504 xlate_actions(&trace.xin, &trace.xout);
5505 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5507 ds_put_char(ds, '\n');
5508 trace_format_flow(ds, 0, "Final flow", &trace);
5510 match_init(&match, flow, &trace.xout.wc);
5511 ds_put_cstr(ds, "Relevant fields: ");
5512 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5513 ds_put_char(ds, '\n');
5515 ds_put_cstr(ds, "Datapath actions: ");
5516 format_odp_actions(ds, trace.xout.odp_actions.data,
5517 trace.xout.odp_actions.size);
5519 if (trace.xout.slow) {
5520 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5521 "slow path because it:");
5522 switch (trace.xout.slow) {
5524 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5527 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5530 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5533 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5535 case SLOW_CONTROLLER:
5536 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5537 "to the OpenFlow controller.");
5544 xlate_out_uninit(&trace.xout);
5547 rule_dpif_unref(rule);
5550 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5551 * 'reply' describing the results. */
5553 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5555 struct cls_cursor cursor;
5556 struct facet *facet;
5560 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5561 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5562 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5563 if (!facet_check_consistency(facet)) {
5567 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5569 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5573 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5574 ofproto->up.name, errors);
5576 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5581 ofproto_dpif_self_check(struct unixctl_conn *conn,
5582 int argc, const char *argv[], void *aux OVS_UNUSED)
5584 struct ds reply = DS_EMPTY_INITIALIZER;
5585 struct ofproto_dpif *ofproto;
5588 ofproto = ofproto_dpif_lookup(argv[1]);
5590 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5591 "ofproto/list for help)");
5594 ofproto_dpif_self_check__(ofproto, &reply);
5596 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5597 ofproto_dpif_self_check__(ofproto, &reply);
5601 unixctl_command_reply(conn, ds_cstr(&reply));
5605 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5606 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5607 * to destroy 'ofproto_shash' and free the returned value. */
5608 static const struct shash_node **
5609 get_ofprotos(struct shash *ofproto_shash)
5611 const struct ofproto_dpif *ofproto;
5613 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5614 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5615 shash_add_nocopy(ofproto_shash, name, ofproto);
5618 return shash_sort(ofproto_shash);
5622 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5623 const char *argv[] OVS_UNUSED,
5624 void *aux OVS_UNUSED)
5626 struct ds ds = DS_EMPTY_INITIALIZER;
5627 struct shash ofproto_shash;
5628 const struct shash_node **sorted_ofprotos;
5631 shash_init(&ofproto_shash);
5632 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5633 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5634 const struct shash_node *node = sorted_ofprotos[i];
5635 ds_put_format(&ds, "%s\n", node->name);
5638 shash_destroy(&ofproto_shash);
5639 free(sorted_ofprotos);
5641 unixctl_command_reply(conn, ds_cstr(&ds));
5646 show_dp_rates(struct ds *ds, const char *heading,
5647 const struct avg_subfacet_rates *rates)
5649 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5650 heading, rates->add_rate, rates->del_rate);
5654 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5656 const struct shash_node **ofprotos;
5657 struct ofproto_dpif *ofproto;
5658 struct shash ofproto_shash;
5659 uint64_t n_hit, n_missed;
5660 long long int minutes;
5663 n_hit = n_missed = 0;
5664 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5665 if (ofproto->backer == backer) {
5666 n_missed += ofproto->n_missed;
5667 n_hit += ofproto->n_hit;
5671 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5672 dpif_name(backer->dpif), n_hit, n_missed);
5673 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5674 " life span: %lldms\n", hmap_count(&backer->subfacets),
5675 backer->avg_n_subfacet, backer->max_n_subfacet,
5676 backer->avg_subfacet_life);
5678 minutes = (time_msec() - backer->created) / (1000 * 60);
5679 if (minutes >= 60) {
5680 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5682 if (minutes >= 60 * 24) {
5683 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5685 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5687 shash_init(&ofproto_shash);
5688 ofprotos = get_ofprotos(&ofproto_shash);
5689 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5690 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5691 const struct shash_node **ports;
5694 if (ofproto->backer != backer) {
5698 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5699 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5701 ports = shash_sort(&ofproto->up.port_by_name);
5702 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5703 const struct shash_node *node = ports[j];
5704 struct ofport *ofport = node->data;
5706 odp_port_t odp_port;
5708 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5711 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5712 if (odp_port != ODPP_NONE) {
5713 ds_put_format(ds, "%"PRIu32":", odp_port);
5715 ds_put_cstr(ds, "none:");
5718 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5721 if (!netdev_get_config(ofport->netdev, &config)) {
5722 const struct smap_node **nodes;
5725 nodes = smap_sort(&config);
5726 for (i = 0; i < smap_count(&config); i++) {
5727 const struct smap_node *node = nodes[i];
5728 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5729 node->key, node->value);
5733 smap_destroy(&config);
5735 ds_put_char(ds, ')');
5736 ds_put_char(ds, '\n');
5740 shash_destroy(&ofproto_shash);
5745 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5746 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5748 struct ds ds = DS_EMPTY_INITIALIZER;
5749 const struct shash_node **backers;
5752 backers = shash_sort(&all_dpif_backers);
5753 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5754 dpif_show_backer(backers[i]->data, &ds);
5758 unixctl_command_reply(conn, ds_cstr(&ds));
5762 /* Dump the megaflow (facet) cache. This is useful to check the
5763 * correctness of flow wildcarding, since the same mechanism is used for
5764 * both xlate caching and kernel wildcarding.
5766 * It's important to note that in the output the flow description uses
5767 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5769 * This command is only needed for advanced debugging, so it's not
5770 * documented in the man page. */
5772 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5773 int argc OVS_UNUSED, const char *argv[],
5774 void *aux OVS_UNUSED)
5776 struct ds ds = DS_EMPTY_INITIALIZER;
5777 const struct ofproto_dpif *ofproto;
5778 long long int now = time_msec();
5779 struct cls_cursor cursor;
5780 struct facet *facet;
5782 ofproto = ofproto_dpif_lookup(argv[1]);
5784 unixctl_command_reply_error(conn, "no such bridge");
5788 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5789 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5790 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5791 cls_rule_format(&facet->cr, &ds);
5792 ds_put_cstr(&ds, ", ");
5793 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5794 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5795 ds_put_cstr(&ds, "Datapath actions: ");
5796 if (facet->xout.slow) {
5797 uint64_t slow_path_stub[128 / 8];
5798 const struct nlattr *actions;
5801 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5802 slow_path_stub, sizeof slow_path_stub,
5803 &actions, &actions_len);
5804 format_odp_actions(&ds, actions, actions_len);
5806 format_odp_actions(&ds, facet->xout.odp_actions.data,
5807 facet->xout.odp_actions.size);
5809 ds_put_cstr(&ds, "\n");
5811 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5813 ds_chomp(&ds, '\n');
5814 unixctl_command_reply(conn, ds_cstr(&ds));
5818 /* Disable using the megaflows.
5820 * This command is only needed for advanced debugging, so it's not
5821 * documented in the man page. */
5823 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5824 int argc OVS_UNUSED,
5825 const char *argv[] OVS_UNUSED,
5826 void *aux OVS_UNUSED)
5828 struct ofproto_dpif *ofproto;
5830 enable_megaflows = false;
5832 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5833 flush(&ofproto->up);
5836 unixctl_command_reply(conn, "megaflows disabled");
5839 /* Re-enable using megaflows.
5841 * This command is only needed for advanced debugging, so it's not
5842 * documented in the man page. */
5844 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5845 int argc OVS_UNUSED,
5846 const char *argv[] OVS_UNUSED,
5847 void *aux OVS_UNUSED)
5849 struct ofproto_dpif *ofproto;
5851 enable_megaflows = true;
5853 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5854 flush(&ofproto->up);
5857 unixctl_command_reply(conn, "megaflows enabled");
5861 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5862 int argc OVS_UNUSED, const char *argv[],
5863 void *aux OVS_UNUSED)
5865 struct ds ds = DS_EMPTY_INITIALIZER;
5866 const struct ofproto_dpif *ofproto;
5867 struct subfacet *subfacet;
5869 ofproto = ofproto_dpif_lookup(argv[1]);
5871 unixctl_command_reply_error(conn, "no such bridge");
5875 update_stats(ofproto->backer);
5877 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5878 struct facet *facet = subfacet->facet;
5879 struct odputil_keybuf maskbuf;
5882 if (facet->ofproto != ofproto) {
5886 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5887 if (enable_megaflows) {
5888 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5889 &facet->flow, UINT32_MAX);
5892 odp_flow_format(subfacet->key, subfacet->key_len,
5893 mask.data, mask.size, &ds, false);
5895 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5896 subfacet->dp_packet_count, subfacet->dp_byte_count);
5897 if (subfacet->used) {
5898 ds_put_format(&ds, "%.3fs",
5899 (time_msec() - subfacet->used) / 1000.0);
5901 ds_put_format(&ds, "never");
5903 if (subfacet->facet->tcp_flags) {
5904 ds_put_cstr(&ds, ", flags:");
5905 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5908 ds_put_cstr(&ds, ", actions:");
5909 if (facet->xout.slow) {
5910 uint64_t slow_path_stub[128 / 8];
5911 const struct nlattr *actions;
5914 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5915 slow_path_stub, sizeof slow_path_stub,
5916 &actions, &actions_len);
5917 format_odp_actions(&ds, actions, actions_len);
5919 format_odp_actions(&ds, facet->xout.odp_actions.data,
5920 facet->xout.odp_actions.size);
5922 ds_put_char(&ds, '\n');
5925 unixctl_command_reply(conn, ds_cstr(&ds));
5930 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5931 int argc OVS_UNUSED, const char *argv[],
5932 void *aux OVS_UNUSED)
5934 struct ds ds = DS_EMPTY_INITIALIZER;
5935 struct ofproto_dpif *ofproto;
5937 ofproto = ofproto_dpif_lookup(argv[1]);
5939 unixctl_command_reply_error(conn, "no such bridge");
5943 flush(&ofproto->up);
5945 unixctl_command_reply(conn, ds_cstr(&ds));
5950 ofproto_dpif_unixctl_init(void)
5952 static bool registered;
5958 unixctl_command_register(
5960 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5961 1, 3, ofproto_unixctl_trace, NULL);
5962 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5963 ofproto_unixctl_fdb_flush, NULL);
5964 unixctl_command_register("fdb/show", "bridge", 1, 1,
5965 ofproto_unixctl_fdb_show, NULL);
5966 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5967 ofproto_dpif_self_check, NULL);
5968 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5969 ofproto_unixctl_dpif_dump_dps, NULL);
5970 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5972 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5973 ofproto_unixctl_dpif_dump_flows, NULL);
5974 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5975 ofproto_unixctl_dpif_del_flows, NULL);
5976 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5977 ofproto_unixctl_dpif_dump_megaflows, NULL);
5978 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5979 ofproto_unixctl_dpif_disable_megaflows, NULL);
5980 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5981 ofproto_unixctl_dpif_enable_megaflows, NULL);
5984 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5986 * This is deprecated. It is only for compatibility with broken device drivers
5987 * in old versions of Linux that do not properly support VLANs when VLAN
5988 * devices are not used. When broken device drivers are no longer in
5989 * widespread use, we will delete these interfaces. */
5992 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5995 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5997 if (realdev_ofp_port == ofport->realdev_ofp_port
5998 && vid == ofport->vlandev_vid) {
6002 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6004 if (ofport->realdev_ofp_port) {
6007 if (realdev_ofp_port && ofport->bundle) {
6008 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6009 * themselves be part of a bundle. */
6010 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6013 ofport->realdev_ofp_port = realdev_ofp_port;
6014 ofport->vlandev_vid = vid;
6016 if (realdev_ofp_port) {
6017 vsp_add(ofport, realdev_ofp_port, vid);
6024 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6026 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6030 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6031 OVS_EXCLUDED(ofproto->vsp_mutex)
6035 ovs_mutex_lock(&ofproto->vsp_mutex);
6036 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6037 ovs_mutex_unlock(&ofproto->vsp_mutex);
6042 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6043 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6044 OVS_REQUIRES(ofproto->vsp_mutex)
6046 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6047 int vid = vlan_tci_to_vid(vlan_tci);
6048 const struct vlan_splinter *vsp;
6050 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6051 hash_realdev_vid(realdev_ofp_port, vid),
6052 &ofproto->realdev_vid_map) {
6053 if (vsp->realdev_ofp_port == realdev_ofp_port
6054 && vsp->vid == vid) {
6055 return vsp->vlandev_ofp_port;
6059 return realdev_ofp_port;
6062 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6063 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6064 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6065 * 'vlan_tci' 9, it would return the port number of eth0.9.
6067 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6068 * function just returns its 'realdev_ofp_port' argument. */
6070 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6071 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6072 OVS_EXCLUDED(ofproto->vsp_mutex)
6076 ovs_mutex_lock(&ofproto->vsp_mutex);
6077 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6078 ovs_mutex_unlock(&ofproto->vsp_mutex);
6082 static struct vlan_splinter *
6083 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6085 struct vlan_splinter *vsp;
6087 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6088 hash_ofp_port(vlandev_ofp_port),
6089 &ofproto->vlandev_map) {
6090 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6098 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6099 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6100 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6101 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6102 * eth0 and store 9 in '*vid'.
6104 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6105 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6108 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6109 ofp_port_t vlandev_ofp_port, int *vid)
6110 OVS_REQUIRES(ofproto->vsp_mutex)
6112 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6113 const struct vlan_splinter *vsp;
6115 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6120 return vsp->realdev_ofp_port;
6126 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6127 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6128 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6129 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6130 * always the case unless VLAN splinters are enabled), returns false without
6131 * making any changes. */
6133 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6134 OVS_EXCLUDED(ofproto->vsp_mutex)
6139 ovs_mutex_lock(&ofproto->vsp_mutex);
6140 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6141 ovs_mutex_unlock(&ofproto->vsp_mutex);
6146 /* Cause the flow to be processed as if it came in on the real device with
6147 * the VLAN device's VLAN ID. */
6148 flow->in_port.ofp_port = realdev;
6149 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6154 vsp_remove(struct ofport_dpif *port)
6156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6157 struct vlan_splinter *vsp;
6159 ovs_mutex_lock(&ofproto->vsp_mutex);
6160 vsp = vlandev_find(ofproto, port->up.ofp_port);
6162 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6163 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6166 port->realdev_ofp_port = 0;
6168 VLOG_ERR("missing vlan device record");
6170 ovs_mutex_unlock(&ofproto->vsp_mutex);
6174 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6176 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6178 ovs_mutex_lock(&ofproto->vsp_mutex);
6179 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6180 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6181 == realdev_ofp_port)) {
6182 struct vlan_splinter *vsp;
6184 vsp = xmalloc(sizeof *vsp);
6185 vsp->realdev_ofp_port = realdev_ofp_port;
6186 vsp->vlandev_ofp_port = port->up.ofp_port;
6189 port->realdev_ofp_port = realdev_ofp_port;
6191 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6192 hash_ofp_port(port->up.ofp_port));
6193 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6194 hash_realdev_vid(realdev_ofp_port, vid));
6196 VLOG_ERR("duplicate vlan device record");
6198 ovs_mutex_unlock(&ofproto->vsp_mutex);
6202 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6204 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6205 return ofport ? ofport->odp_port : ODPP_NONE;
6208 struct ofport_dpif *
6209 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6211 struct ofport_dpif *port;
6213 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6214 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6215 &backer->odp_to_ofport_map) {
6216 if (port->odp_port == odp_port) {
6217 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6222 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6227 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6229 struct ofport_dpif *port;
6231 port = odp_port_to_ofport(ofproto->backer, odp_port);
6232 if (port && &ofproto->up == port->up.ofproto) {
6233 return port->up.ofp_port;
6239 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6240 * most heavily weighted element. 'base' designates the rate of decay: after
6241 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6244 exp_mavg(double *avg, int base, double new)
6246 *avg = (*avg * (base - 1) + new) / base;
6250 update_moving_averages(struct dpif_backer *backer)
6252 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6253 long long int minutes = (time_msec() - backer->created) / min_ms;
6256 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6258 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6261 backer->lifetime.add_rate = 0.0;
6262 backer->lifetime.del_rate = 0.0;
6265 /* Update hourly averages on the minute boundaries. */
6266 if (time_msec() - backer->last_minute >= min_ms) {
6267 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6268 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6270 /* Update daily averages on the hour boundaries. */
6271 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6272 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6273 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6276 backer->total_subfacet_add_count += backer->subfacet_add_count;
6277 backer->total_subfacet_del_count += backer->subfacet_del_count;
6278 backer->subfacet_add_count = 0;
6279 backer->subfacet_del_count = 0;
6280 backer->last_minute += min_ms;
6284 const struct ofproto_class ofproto_dpif_class = {
6319 port_is_lacp_current,
6320 NULL, /* rule_choose_table */
6329 rule_modify_actions,
6343 get_stp_port_status,
6350 is_mirror_output_bundle,
6351 forward_bpdu_changed,
6352 set_mac_table_config,
6354 NULL, /* meter_get_features */
6355 NULL, /* meter_set */
6356 NULL, /* meter_get */
6357 NULL, /* meter_del */