2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
74 COVERAGE_DEFINE(flow_mod_overflow);
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
86 struct flow_wildcards *wc);
88 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
91 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
92 struct ofproto_dpif *ofproto; /* Owning ofproto. */
93 void *aux; /* Key supplied by ofproto's client. */
94 char *name; /* Identifier for log messages. */
97 struct list ports; /* Contains "struct ofport"s. */
98 enum port_vlan_mode vlan_mode; /* VLAN mode */
99 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
100 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
101 * NULL if all VLANs are trunked. */
102 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
103 struct bond *bond; /* Nonnull iff more than one port. */
104 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
107 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
110 static void bundle_remove(struct ofport *);
111 static void bundle_update(struct ofbundle *);
112 static void bundle_destroy(struct ofbundle *);
113 static void bundle_del_port(struct ofport_dpif *);
114 static void bundle_run(struct ofbundle *);
115 static void bundle_wait(struct ofbundle *);
117 static void stp_run(struct ofproto_dpif *ofproto);
118 static void stp_wait(struct ofproto_dpif *ofproto);
119 static int set_stp_port(struct ofport *,
120 const struct ofproto_port_stp_settings *);
122 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
123 enum slow_path_reason,
124 uint64_t *stub, size_t stub_size,
125 const struct nlattr **actionsp,
126 size_t *actions_lenp);
128 /* A subfacet (see "struct subfacet" below) has three possible installation
131 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
132 * case just after the subfacet is created, just before the subfacet is
133 * destroyed, or if the datapath returns an error when we try to install a
136 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
138 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
139 * ofproto_dpif is installed in the datapath.
142 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
143 SF_FAST_PATH, /* Full actions are installed. */
144 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
147 /* A dpif flow and actions associated with a facet.
149 * See also the large comment on struct facet. */
152 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
153 struct list list_node; /* In struct facet's 'facets' list. */
154 struct facet *facet; /* Owning facet. */
155 struct dpif_backer *backer; /* Owning backer. */
157 enum odp_key_fitness key_fitness;
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Time created. */
164 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
165 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
167 enum subfacet_path path; /* Installed in datapath? */
170 #define SUBFACET_DESTROY_MAX_BATCH 50
172 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
174 static struct subfacet *subfacet_find(struct dpif_backer *,
175 const struct nlattr *key, size_t key_len,
177 static void subfacet_destroy(struct subfacet *);
178 static void subfacet_destroy__(struct subfacet *);
179 static void subfacet_destroy_batch(struct dpif_backer *,
180 struct subfacet **, int n);
181 static void subfacet_reset_dp_stats(struct subfacet *,
182 struct dpif_flow_stats *);
183 static void subfacet_update_stats(struct subfacet *,
184 const struct dpif_flow_stats *);
185 static int subfacet_install(struct subfacet *,
186 const struct ofpbuf *odp_actions,
187 struct dpif_flow_stats *);
188 static void subfacet_uninstall(struct subfacet *);
190 /* A unique, non-overlapping instantiation of an OpenFlow flow.
192 * A facet associates a "struct flow", which represents the Open vSwitch
193 * userspace idea of an exact-match flow, with one or more subfacets.
194 * While the facet is created based on an exact-match flow, it is stored
195 * within the ofproto based on the wildcards that could be expressed
196 * based on the flow table and other configuration. (See the 'wc'
197 * description in "struct xlate_out" for more details.)
199 * Each subfacet tracks the datapath's idea of the flow equivalent to
200 * the facet. When the kernel module (or other dpif implementation) and
201 * Open vSwitch userspace agree on the definition of a flow key, there
202 * is exactly one subfacet per facet. If the dpif implementation
203 * supports more-specific flow matching than userspace, however, a facet
204 * can have more than one subfacet. Examples include the dpif
205 * implementation not supporting the same wildcards as userspace or some
206 * distinction in flow that userspace simply doesn't understand.
208 * Flow expiration works in terms of subfacets, so a facet must have at
209 * least one subfacet or it will never expire, leaking memory. */
212 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
213 struct ofproto_dpif *ofproto;
216 struct list subfacets;
217 long long int used; /* Time last used; time created if not used. */
220 struct flow flow; /* Flow of the creating subfacet. */
221 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
225 * - Do include packets and bytes sent "by hand", e.g. with
228 * - Do include packets and bytes that were obtained from the datapath
229 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
230 * DPIF_FP_ZERO_STATS).
232 * - Do not include packets or bytes that can be obtained from the
233 * datapath for any existing subfacet.
235 uint64_t packet_count; /* Number of packets received. */
236 uint64_t byte_count; /* Number of bytes received. */
238 /* Resubmit statistics. */
239 uint64_t prev_packet_count; /* Number of packets from last stats push. */
240 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
241 long long int prev_used; /* Used time from last stats push. */
244 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
245 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
246 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
248 struct xlate_out xout;
249 bool fail_open; /* Facet matched the fail open rule. */
251 /* Storage for a single subfacet, to reduce malloc() time and space
252 * overhead. (A facet always has at least one subfacet and in the common
253 * case has exactly one subfacet. However, 'one_subfacet' may not
254 * always be valid, since it could have been removed after newer
255 * subfacets were pushed onto the 'subfacets' list.) */
256 struct subfacet one_subfacet;
258 long long int learn_rl; /* Rate limiter for facet_learn(). */
261 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
263 struct dpif_flow_stats *);
264 static void facet_remove(struct facet *);
265 static void facet_free(struct facet *);
267 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
268 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
269 const struct flow *);
270 static bool facet_revalidate(struct facet *);
271 static bool facet_check_consistency(struct facet *);
273 static void facet_flush_stats(struct facet *);
275 static void facet_reset_counters(struct facet *);
276 static void facet_push_stats(struct facet *, bool may_learn);
277 static void facet_learn(struct facet *);
278 static void facet_account(struct facet *);
279 static void push_all_stats(void);
281 static bool facet_is_controller_flow(struct facet *);
284 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
288 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
289 struct list bundle_node; /* In struct ofbundle's "ports" list. */
290 struct cfm *cfm; /* Connectivity Fault Management, if any. */
291 struct bfd *bfd; /* BFD, if any. */
292 bool may_enable; /* May be enabled in bonds. */
293 bool is_tunnel; /* This port is a tunnel. */
294 long long int carrier_seq; /* Carrier status changes. */
295 struct ofport_dpif *peer; /* Peer if patch port. */
298 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
299 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
300 long long int stp_state_entered;
302 /* Queue to DSCP mapping. */
303 struct ofproto_port_queue *qdscp;
306 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
308 * This is deprecated. It is only for compatibility with broken device
309 * drivers in old versions of Linux that do not properly support VLANs when
310 * VLAN devices are not used. When broken device drivers are no longer in
311 * widespread use, we will delete these interfaces. */
312 ofp_port_t realdev_ofp_port;
316 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
318 * This is deprecated. It is only for compatibility with broken device drivers
319 * in old versions of Linux that do not properly support VLANs when VLAN
320 * devices are not used. When broken device drivers are no longer in
321 * widespread use, we will delete these interfaces. */
322 struct vlan_splinter {
323 struct hmap_node realdev_vid_node;
324 struct hmap_node vlandev_node;
325 ofp_port_t realdev_ofp_port;
326 ofp_port_t vlandev_ofp_port;
330 static void vsp_remove(struct ofport_dpif *);
331 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
333 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
336 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
339 static struct ofport_dpif *
340 ofport_dpif_cast(const struct ofport *ofport)
342 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
345 static void port_run(struct ofport_dpif *);
346 static void port_run_fast(struct ofport_dpif *);
347 static void port_wait(struct ofport_dpif *);
348 static int set_bfd(struct ofport *, const struct smap *);
349 static int set_cfm(struct ofport *, const struct cfm_settings *);
350 static void ofport_update_peer(struct ofport_dpif *);
351 static void run_fast_rl(void);
352 static int run_fast(struct ofproto *);
354 struct dpif_completion {
355 struct list list_node;
356 struct ofoperation *op;
359 /* Reasons that we might need to revalidate every facet, and corresponding
362 * A value of 0 means that there is no need to revalidate.
364 * It would be nice to have some cleaner way to integrate with coverage
365 * counters, but with only a few reasons I guess this is good enough for
367 enum revalidate_reason {
368 REV_RECONFIGURE = 1, /* Switch configuration changed. */
369 REV_STP, /* Spanning tree protocol port status change. */
370 REV_BOND, /* Bonding changed. */
371 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
372 REV_FLOW_TABLE, /* Flow table changed. */
373 REV_MAC_LEARNING, /* Mac learning changed. */
374 REV_INCONSISTENCY /* Facet self-check failed. */
376 COVERAGE_DEFINE(rev_reconfigure);
377 COVERAGE_DEFINE(rev_stp);
378 COVERAGE_DEFINE(rev_bond);
379 COVERAGE_DEFINE(rev_port_toggled);
380 COVERAGE_DEFINE(rev_flow_table);
381 COVERAGE_DEFINE(rev_mac_learning);
382 COVERAGE_DEFINE(rev_inconsistency);
384 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
385 * These are datapath flows which have no associated ofproto, if they did we
386 * would use facets. */
388 struct hmap_node hmap_node;
393 struct avg_subfacet_rates {
394 double add_rate; /* Moving average of new flows created per minute. */
395 double del_rate; /* Moving average of flows deleted per minute. */
398 /* All datapaths of a given type share a single dpif backer instance. */
403 struct timer next_expiration;
405 struct ovs_rwlock odp_to_ofport_lock;
406 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
408 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
410 /* Facet revalidation flags applying to facets which use this backer. */
411 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
413 struct hmap drop_keys; /* Set of dropped odp keys. */
414 bool recv_set_enable; /* Enables or disables receiving packets. */
416 struct hmap subfacets;
417 struct governor *governor;
419 /* Subfacet statistics.
421 * These keep track of the total number of subfacets added and deleted and
422 * flow life span. They are useful for computing the flow rates stats
423 * exposed via "ovs-appctl dpif/show". The goal is to learn about
424 * traffic patterns in ways that we can use later to improve Open vSwitch
425 * performance in new situations. */
426 long long int created; /* Time when it is created. */
427 unsigned max_n_subfacet; /* Maximum number of flows */
428 unsigned avg_n_subfacet; /* Average number of flows. */
429 long long int avg_subfacet_life; /* Average life span of subfacets. */
431 /* The average number of subfacets... */
432 struct avg_subfacet_rates hourly; /* ...over the last hour. */
433 struct avg_subfacet_rates daily; /* ...over the last day. */
434 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
435 long long int last_minute; /* Last time 'hourly' was updated. */
437 /* Number of subfacets added or deleted since 'last_minute'. */
438 unsigned subfacet_add_count;
439 unsigned subfacet_del_count;
441 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
442 unsigned long long int total_subfacet_add_count;
443 unsigned long long int total_subfacet_del_count;
446 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
447 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
449 static void drop_key_clear(struct dpif_backer *);
450 static void update_moving_averages(struct dpif_backer *backer);
452 struct ofproto_dpif {
453 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
455 struct dpif_backer *backer;
457 /* Special OpenFlow rules. */
458 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
459 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
460 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
463 struct netflow *netflow;
464 struct dpif_sflow *sflow;
465 struct dpif_ipfix *ipfix;
466 struct hmap bundles; /* Contains "struct ofbundle"s. */
467 struct mac_learning *ml;
468 bool has_bonded_bundles;
469 struct mbridge *mbridge;
472 struct classifier facets; /* Contains 'struct facet's. */
473 long long int consistency_rl;
475 /* Support for debugging async flow mods. */
476 struct list completions;
478 struct netdev_stats stats; /* To account packets generated and consumed in
483 long long int stp_last_tick;
485 /* VLAN splinters. */
486 struct ovs_mutex vsp_mutex;
487 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
488 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
491 struct sset ports; /* Set of standard port names. */
492 struct sset ghost_ports; /* Ports with no datapath port. */
493 struct sset port_poll_set; /* Queued names for port_poll() reply. */
494 int port_poll_errno; /* Last errno for port_poll() reply. */
496 /* Per ofproto's dpif stats. */
501 struct ovs_mutex flow_mod_mutex;
502 struct list flow_mods OVS_GUARDED;
503 size_t n_flow_mods OVS_GUARDED;
506 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
507 * for debugging the asynchronous flow_mod implementation.) */
510 /* By default, flows in the datapath are wildcarded (megaflows). They
511 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
512 static bool enable_megaflows = true;
514 /* All existing ofproto_dpif instances, indexed by ->up.name. */
515 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
517 static void ofproto_dpif_unixctl_init(void);
519 static inline struct ofproto_dpif *
520 ofproto_dpif_cast(const struct ofproto *ofproto)
522 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
523 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
526 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
527 ofp_port_t ofp_port);
528 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
529 const struct ofpbuf *packet, struct ds *);
532 #define FLOW_MISS_MAX_BATCH 50
533 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
535 /* Flow expiration. */
536 static int expire(struct dpif_backer *);
539 static void send_netflow_active_timeouts(struct ofproto_dpif *);
542 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
544 /* Global variables. */
545 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
547 /* Initial mappings of port to bridge mappings. */
548 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
550 /* Executes and takes ownership of 'fm'. */
552 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
553 struct ofputil_flow_mod *fm)
555 ovs_mutex_lock(&ofproto->flow_mod_mutex);
556 if (ofproto->n_flow_mods > 1024) {
557 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
558 COVERAGE_INC(flow_mod_overflow);
564 list_push_back(&ofproto->flow_mods, &fm->list_node);
565 ofproto->n_flow_mods++;
566 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
570 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
571 struct ofputil_packet_in *pin)
573 connmgr_send_packet_in(ofproto->up.connmgr, pin);
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 VLOG_ERR("Failed to enable receiving packets in dpif.");
698 dpif_flow_flush(backer->dpif);
699 backer->need_revalidate = REV_RECONFIGURE;
702 if (backer->need_revalidate) {
703 struct ofproto_dpif *ofproto;
704 struct simap_node *node;
705 struct simap tmp_backers;
707 /* Handle tunnel garbage collection. */
708 simap_init(&tmp_backers);
709 simap_swap(&backer->tnl_backers, &tmp_backers);
711 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
712 struct ofport_dpif *iter;
714 if (backer != ofproto->backer) {
718 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
719 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
722 if (!iter->is_tunnel) {
726 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
727 namebuf, sizeof namebuf);
728 node = simap_find(&tmp_backers, dp_port);
730 simap_put(&backer->tnl_backers, dp_port, node->data);
731 simap_delete(&tmp_backers, node);
732 node = simap_find(&backer->tnl_backers, dp_port);
734 node = simap_find(&backer->tnl_backers, dp_port);
736 odp_port_t odp_port = ODPP_NONE;
738 if (!dpif_port_add(backer->dpif, iter->up.netdev,
740 simap_put(&backer->tnl_backers, dp_port,
741 odp_to_u32(odp_port));
742 node = simap_find(&backer->tnl_backers, dp_port);
747 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
748 if (tnl_port_reconfigure(iter, iter->up.netdev,
750 backer->need_revalidate = REV_RECONFIGURE;
755 SIMAP_FOR_EACH (node, &tmp_backers) {
756 dpif_port_del(backer->dpif, u32_to_odp(node->data));
758 simap_destroy(&tmp_backers);
760 switch (backer->need_revalidate) {
761 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
762 case REV_STP: COVERAGE_INC(rev_stp); break;
763 case REV_BOND: COVERAGE_INC(rev_bond); break;
764 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
765 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
766 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
767 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
769 backer->need_revalidate = 0;
771 /* Clear the drop_keys in case we should now be accepting some
772 * formerly dropped flows. */
773 drop_key_clear(backer);
775 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
776 struct facet *facet, *next;
777 struct ofport_dpif *ofport;
778 struct cls_cursor cursor;
779 struct ofbundle *bundle;
781 if (ofproto->backer != backer) {
785 xlate_ofproto_set(ofproto, ofproto->up.name,
786 ofproto->backer->dpif, ofproto->ml,
787 ofproto->stp, ofproto->mbridge,
788 ofproto->sflow, ofproto->ipfix,
789 ofproto->up.frag_handling,
790 ofproto->up.forward_bpdu,
791 connmgr_has_in_band(ofproto->up.connmgr),
792 ofproto->netflow != NULL);
794 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
795 xlate_bundle_set(ofproto, bundle, bundle->name,
796 bundle->vlan_mode, bundle->vlan,
797 bundle->trunks, bundle->use_priority_tags,
798 bundle->bond, bundle->lacp,
802 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
803 int stp_port = ofport->stp_port
804 ? stp_port_no(ofport->stp_port)
806 xlate_ofport_set(ofproto, ofport->bundle, ofport,
807 ofport->up.ofp_port, ofport->odp_port,
808 ofport->up.netdev, ofport->cfm,
809 ofport->bfd, ofport->peer, stp_port,
810 ofport->qdscp, ofport->n_qdscp,
811 ofport->up.pp.config, ofport->is_tunnel,
815 cls_cursor_init(&cursor, &ofproto->facets, NULL);
816 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
817 facet_revalidate(facet);
823 if (!backer->recv_set_enable) {
824 /* Wake up before a max of 1000ms. */
825 timer_set_duration(&backer->next_expiration, 1000);
826 } else if (timer_expired(&backer->next_expiration)) {
827 int delay = expire(backer);
828 timer_set_duration(&backer->next_expiration, delay);
831 process_dpif_port_changes(backer);
833 if (backer->governor) {
836 governor_run(backer->governor);
838 /* If the governor has shrunk to its minimum size and the number of
839 * subfacets has dwindled, then drop the governor entirely.
841 * For hysteresis, the number of subfacets to drop the governor is
842 * smaller than the number needed to trigger its creation. */
843 n_subfacets = hmap_count(&backer->subfacets);
844 if (n_subfacets * 4 < flow_eviction_threshold
845 && governor_is_idle(backer->governor)) {
846 governor_destroy(backer->governor);
847 backer->governor = NULL;
854 /* Check for and handle port changes in 'backer''s dpif. */
856 process_dpif_port_changes(struct dpif_backer *backer)
862 error = dpif_port_poll(backer->dpif, &devname);
868 process_dpif_all_ports_changed(backer);
872 process_dpif_port_change(backer, devname);
877 process_dpif_port_error(backer, error);
884 process_dpif_all_ports_changed(struct dpif_backer *backer)
886 struct ofproto_dpif *ofproto;
887 struct dpif_port dpif_port;
888 struct dpif_port_dump dump;
889 struct sset devnames;
892 sset_init(&devnames);
893 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
894 if (ofproto->backer == backer) {
895 struct ofport *ofport;
897 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
898 sset_add(&devnames, netdev_get_name(ofport->netdev));
902 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
903 sset_add(&devnames, dpif_port.name);
906 SSET_FOR_EACH (devname, &devnames) {
907 process_dpif_port_change(backer, devname);
909 sset_destroy(&devnames);
913 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
915 struct ofproto_dpif *ofproto;
916 struct dpif_port port;
918 /* Don't report on the datapath's device. */
919 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
923 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
924 &all_ofproto_dpifs) {
925 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
930 ofproto = lookup_ofproto_dpif_by_port_name(devname);
931 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
932 /* The port was removed. If we know the datapath,
933 * report it through poll_set(). If we don't, it may be
934 * notifying us of a removal we initiated, so ignore it.
935 * If there's a pending ENOBUFS, let it stand, since
936 * everything will be reevaluated. */
937 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
938 sset_add(&ofproto->port_poll_set, devname);
939 ofproto->port_poll_errno = 0;
941 } else if (!ofproto) {
942 /* The port was added, but we don't know with which
943 * ofproto we should associate it. Delete it. */
944 dpif_port_del(backer->dpif, port.port_no);
946 struct ofport_dpif *ofport;
948 ofport = ofport_dpif_cast(shash_find_data(
949 &ofproto->up.port_by_name, devname));
951 && ofport->odp_port != port.port_no
952 && !odp_port_to_ofport(backer, port.port_no))
954 /* 'ofport''s datapath port number has changed from
955 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
956 * structures to match. */
957 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
958 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
959 ofport->odp_port = port.port_no;
960 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
961 hash_odp_port(port.port_no));
962 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
963 backer->need_revalidate = REV_RECONFIGURE;
966 dpif_port_destroy(&port);
969 /* Propagate 'error' to all ofprotos based on 'backer'. */
971 process_dpif_port_error(struct dpif_backer *backer, int error)
973 struct ofproto_dpif *ofproto;
975 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
976 if (ofproto->backer == backer) {
977 sset_clear(&ofproto->port_poll_set);
978 ofproto->port_poll_errno = error;
984 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
988 /* If recv_set_enable is false, we should not handle upcalls. */
989 if (!backer->recv_set_enable) {
993 /* Handle one or more batches of upcalls, until there's nothing left to do
994 * or until we do a fixed total amount of work.
996 * We do work in batches because it can be much cheaper to set up a number
997 * of flows and fire off their patches all at once. We do multiple batches
998 * because in some cases handling a packet can cause another packet to be
999 * queued almost immediately as part of the return flow. Both
1000 * optimizations can make major improvements on some benchmarks and
1001 * presumably for real traffic as well. */
1003 while (work < max_batch) {
1004 int retval = handle_upcalls(backer, max_batch - work);
1015 type_run_fast(const char *type)
1017 struct dpif_backer *backer;
1019 backer = shash_find_data(&all_dpif_backers, type);
1021 /* This is not necessarily a problem, since backers are only
1022 * created on demand. */
1026 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1032 static long long int port_rl = LLONG_MIN;
1033 static unsigned int backer_rl = 0;
1035 if (time_msec() >= port_rl) {
1036 struct ofproto_dpif *ofproto;
1038 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1039 run_fast(&ofproto->up);
1041 port_rl = time_msec() + 200;
1044 /* XXX: We have to be careful not to do too much work in this function. If
1045 * we call dpif_backer_run_fast() too often, or with too large a batch,
1046 * performance improves signifcantly, but at a cost. It's possible for the
1047 * number of flows in the datapath to increase without bound, and for poll
1048 * loops to take 10s of seconds. The correct solution to this problem,
1049 * long term, is to separate flow miss handling into it's own thread so it
1050 * isn't affected by revalidations, and expirations. Until then, this is
1051 * the best we can do. */
1052 if (++backer_rl >= 10) {
1053 struct shash_node *node;
1056 SHASH_FOR_EACH (node, &all_dpif_backers) {
1057 dpif_backer_run_fast(node->data, 1);
1063 type_wait(const char *type)
1065 struct dpif_backer *backer;
1067 backer = shash_find_data(&all_dpif_backers, type);
1069 /* This is not necessarily a problem, since backers are only
1070 * created on demand. */
1074 if (backer->governor) {
1075 governor_wait(backer->governor);
1078 timer_wait(&backer->next_expiration);
1081 /* Basic life-cycle. */
1083 static int add_internal_flows(struct ofproto_dpif *);
1085 static struct ofproto *
1088 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1089 return &ofproto->up;
1093 dealloc(struct ofproto *ofproto_)
1095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1100 close_dpif_backer(struct dpif_backer *backer)
1102 struct shash_node *node;
1104 ovs_assert(backer->refcount > 0);
1106 if (--backer->refcount) {
1110 drop_key_clear(backer);
1111 hmap_destroy(&backer->drop_keys);
1113 simap_destroy(&backer->tnl_backers);
1114 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1115 hmap_destroy(&backer->odp_to_ofport_map);
1116 node = shash_find(&all_dpif_backers, backer->type);
1118 shash_delete(&all_dpif_backers, node);
1119 dpif_close(backer->dpif);
1121 ovs_assert(hmap_is_empty(&backer->subfacets));
1122 hmap_destroy(&backer->subfacets);
1123 governor_destroy(backer->governor);
1128 /* Datapath port slated for removal from datapath. */
1129 struct odp_garbage {
1130 struct list list_node;
1131 odp_port_t odp_port;
1135 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1137 struct dpif_backer *backer;
1138 struct dpif_port_dump port_dump;
1139 struct dpif_port port;
1140 struct shash_node *node;
1141 struct list garbage_list;
1142 struct odp_garbage *garbage, *next;
1148 backer = shash_find_data(&all_dpif_backers, type);
1155 backer_name = xasprintf("ovs-%s", type);
1157 /* Remove any existing datapaths, since we assume we're the only
1158 * userspace controlling the datapath. */
1160 dp_enumerate_names(type, &names);
1161 SSET_FOR_EACH(name, &names) {
1162 struct dpif *old_dpif;
1164 /* Don't remove our backer if it exists. */
1165 if (!strcmp(name, backer_name)) {
1169 if (dpif_open(name, type, &old_dpif)) {
1170 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1172 dpif_delete(old_dpif);
1173 dpif_close(old_dpif);
1176 sset_destroy(&names);
1178 backer = xmalloc(sizeof *backer);
1180 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1183 VLOG_ERR("failed to open datapath of type %s: %s", type,
1184 ovs_strerror(error));
1189 backer->type = xstrdup(type);
1190 backer->governor = NULL;
1191 backer->refcount = 1;
1192 hmap_init(&backer->odp_to_ofport_map);
1193 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1194 hmap_init(&backer->drop_keys);
1195 hmap_init(&backer->subfacets);
1196 timer_set_duration(&backer->next_expiration, 1000);
1197 backer->need_revalidate = 0;
1198 simap_init(&backer->tnl_backers);
1199 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1202 if (backer->recv_set_enable) {
1203 dpif_flow_flush(backer->dpif);
1206 /* Loop through the ports already on the datapath and remove any
1207 * that we don't need anymore. */
1208 list_init(&garbage_list);
1209 dpif_port_dump_start(&port_dump, backer->dpif);
1210 while (dpif_port_dump_next(&port_dump, &port)) {
1211 node = shash_find(&init_ofp_ports, port.name);
1212 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1213 garbage = xmalloc(sizeof *garbage);
1214 garbage->odp_port = port.port_no;
1215 list_push_front(&garbage_list, &garbage->list_node);
1218 dpif_port_dump_done(&port_dump);
1220 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1221 dpif_port_del(backer->dpif, garbage->odp_port);
1222 list_remove(&garbage->list_node);
1226 shash_add(&all_dpif_backers, type, backer);
1228 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1230 VLOG_ERR("failed to listen on datapath of type %s: %s",
1231 type, ovs_strerror(error));
1232 close_dpif_backer(backer);
1236 backer->max_n_subfacet = 0;
1237 backer->created = time_msec();
1238 backer->last_minute = backer->created;
1239 memset(&backer->hourly, 0, sizeof backer->hourly);
1240 memset(&backer->daily, 0, sizeof backer->daily);
1241 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1242 backer->subfacet_add_count = 0;
1243 backer->subfacet_del_count = 0;
1244 backer->total_subfacet_add_count = 0;
1245 backer->total_subfacet_del_count = 0;
1246 backer->avg_n_subfacet = 0;
1247 backer->avg_subfacet_life = 0;
1253 construct(struct ofproto *ofproto_)
1255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1256 struct shash_node *node, *next;
1257 odp_port_t max_ports;
1260 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1265 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1266 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1267 ofp_to_u16(OFPP_MAX))));
1269 ofproto->netflow = NULL;
1270 ofproto->sflow = NULL;
1271 ofproto->ipfix = NULL;
1272 ofproto->stp = NULL;
1273 hmap_init(&ofproto->bundles);
1274 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1275 ofproto->mbridge = mbridge_create();
1276 ofproto->has_bonded_bundles = false;
1277 ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1279 classifier_init(&ofproto->facets);
1280 ofproto->consistency_rl = LLONG_MIN;
1282 list_init(&ofproto->completions);
1284 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1285 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1286 list_init(&ofproto->flow_mods);
1287 ofproto->n_flow_mods = 0;
1288 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1290 ofproto_dpif_unixctl_init();
1292 hmap_init(&ofproto->vlandev_map);
1293 hmap_init(&ofproto->realdev_vid_map);
1295 sset_init(&ofproto->ports);
1296 sset_init(&ofproto->ghost_ports);
1297 sset_init(&ofproto->port_poll_set);
1298 ofproto->port_poll_errno = 0;
1300 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1301 struct iface_hint *iface_hint = node->data;
1303 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1304 /* Check if the datapath already has this port. */
1305 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1306 sset_add(&ofproto->ports, node->name);
1309 free(iface_hint->br_name);
1310 free(iface_hint->br_type);
1312 shash_delete(&init_ofp_ports, node);
1316 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1317 hash_string(ofproto->up.name, 0));
1318 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1320 ofproto_init_tables(ofproto_, N_TABLES);
1321 error = add_internal_flows(ofproto);
1322 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1325 ofproto->n_missed = 0;
1331 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1332 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1334 struct ofputil_flow_mod fm;
1337 match_init_catchall(&fm.match);
1339 match_set_reg(&fm.match, 0, id);
1340 fm.new_cookie = htonll(0);
1341 fm.cookie = htonll(0);
1342 fm.cookie_mask = htonll(0);
1343 fm.modify_cookie = false;
1344 fm.table_id = TBL_INTERNAL;
1345 fm.command = OFPFC_ADD;
1346 fm.idle_timeout = 0;
1347 fm.hard_timeout = 0;
1351 fm.ofpacts = ofpacts->data;
1352 fm.ofpacts_len = ofpacts->size;
1354 error = ofproto_flow_mod(&ofproto->up, &fm);
1356 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1357 id, ofperr_to_string(error));
1361 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1363 ovs_assert(*rulep != NULL);
1369 add_internal_flows(struct ofproto_dpif *ofproto)
1371 struct ofpact_controller *controller;
1372 uint64_t ofpacts_stub[128 / 8];
1373 struct ofpbuf ofpacts;
1377 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1380 controller = ofpact_put_CONTROLLER(&ofpacts);
1381 controller->max_len = UINT16_MAX;
1382 controller->controller_id = 0;
1383 controller->reason = OFPR_NO_MATCH;
1384 ofpact_pad(&ofpacts);
1386 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1391 ofpbuf_clear(&ofpacts);
1392 error = add_internal_flow(ofproto, id++, &ofpacts,
1393 &ofproto->no_packet_in_rule);
1398 error = add_internal_flow(ofproto, id++, &ofpacts,
1399 &ofproto->drop_frags_rule);
1404 complete_operations(struct ofproto_dpif *ofproto)
1406 struct dpif_completion *c, *next;
1408 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1409 ofoperation_complete(c->op, 0);
1410 list_remove(&c->list_node);
1416 destruct(struct ofproto *ofproto_)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1419 struct rule_dpif *rule, *next_rule;
1420 struct ofputil_flow_mod *fm, *next_fm;
1421 struct oftable *table;
1423 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1424 xlate_remove_ofproto(ofproto);
1426 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1427 complete_operations(ofproto);
1429 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1430 struct cls_cursor cursor;
1432 cls_cursor_init(&cursor, &table->cls, NULL);
1433 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1434 ofproto_rule_destroy(&rule->up);
1438 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1439 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1440 list_remove(&fm->list_node);
1441 ofproto->n_flow_mods--;
1445 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1446 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1448 mbridge_unref(ofproto->mbridge);
1450 netflow_destroy(ofproto->netflow);
1451 dpif_sflow_unref(ofproto->sflow);
1452 hmap_destroy(&ofproto->bundles);
1453 mac_learning_unref(ofproto->ml);
1455 classifier_destroy(&ofproto->facets);
1457 hmap_destroy(&ofproto->vlandev_map);
1458 hmap_destroy(&ofproto->realdev_vid_map);
1460 sset_destroy(&ofproto->ports);
1461 sset_destroy(&ofproto->ghost_ports);
1462 sset_destroy(&ofproto->port_poll_set);
1464 ovs_mutex_destroy(&ofproto->vsp_mutex);
1466 close_dpif_backer(ofproto->backer);
1470 run_fast(struct ofproto *ofproto_)
1472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1473 struct ofputil_flow_mod *fm, *next;
1474 struct ofport_dpif *ofport;
1475 struct list flow_mods;
1477 /* Do not perform any periodic activity required by 'ofproto' while
1478 * waiting for flow restore to complete. */
1479 if (ofproto_get_flow_restore_wait()) {
1483 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1484 if (ofproto->n_flow_mods) {
1485 flow_mods = ofproto->flow_mods;
1486 list_moved(&flow_mods);
1487 list_init(&ofproto->flow_mods);
1488 ofproto->n_flow_mods = 0;
1490 list_init(&flow_mods);
1492 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1494 LIST_FOR_EACH_SAFE (fm, next, list_node, &flow_mods) {
1495 int error = ofproto_flow_mod(&ofproto->up, fm);
1496 if (error && !VLOG_DROP_WARN(&rl)) {
1497 VLOG_WARN("learning action failed to modify flow table (%s)",
1498 ofperr_get_name(error));
1501 list_remove(&fm->list_node);
1506 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1507 port_run_fast(ofport);
1514 run(struct ofproto *ofproto_)
1516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1517 struct ofport_dpif *ofport;
1518 struct ofbundle *bundle;
1522 complete_operations(ofproto);
1525 if (mbridge_need_revalidate(ofproto->mbridge)) {
1526 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1527 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1528 mac_learning_flush(ofproto->ml);
1529 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1532 /* Do not perform any periodic activity below required by 'ofproto' while
1533 * waiting for flow restore to complete. */
1534 if (ofproto_get_flow_restore_wait()) {
1538 error = run_fast(ofproto_);
1543 if (ofproto->netflow) {
1544 if (netflow_run(ofproto->netflow)) {
1545 send_netflow_active_timeouts(ofproto);
1548 if (ofproto->sflow) {
1549 dpif_sflow_run(ofproto->sflow);
1552 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1555 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1560 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1561 if (mac_learning_run(ofproto->ml)) {
1562 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1564 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1566 /* Check the consistency of a random facet, to aid debugging. */
1567 if (time_msec() >= ofproto->consistency_rl
1568 && !classifier_is_empty(&ofproto->facets)
1569 && !ofproto->backer->need_revalidate) {
1570 struct cls_table *table;
1571 struct cls_rule *cr;
1572 struct facet *facet;
1574 ofproto->consistency_rl = time_msec() + 250;
1576 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1577 struct cls_table, hmap_node);
1578 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1580 facet = CONTAINER_OF(cr, struct facet, cr);
1582 if (!facet_check_consistency(facet)) {
1583 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1591 wait(struct ofproto *ofproto_)
1593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1594 struct ofport_dpif *ofport;
1595 struct ofbundle *bundle;
1597 if (!clogged && !list_is_empty(&ofproto->completions)) {
1598 poll_immediate_wake();
1601 if (ofproto_get_flow_restore_wait()) {
1605 dpif_wait(ofproto->backer->dpif);
1606 dpif_recv_wait(ofproto->backer->dpif);
1607 if (ofproto->sflow) {
1608 dpif_sflow_wait(ofproto->sflow);
1610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1613 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1614 bundle_wait(bundle);
1616 if (ofproto->netflow) {
1617 netflow_wait(ofproto->netflow);
1619 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1620 mac_learning_wait(ofproto->ml);
1621 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1623 if (ofproto->backer->need_revalidate) {
1624 /* Shouldn't happen, but if it does just go around again. */
1625 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1626 poll_immediate_wake();
1631 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1633 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1634 struct cls_cursor cursor;
1635 size_t n_subfacets = 0;
1636 struct facet *facet;
1638 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1640 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1641 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1642 n_subfacets += list_size(&facet->subfacets);
1644 simap_increase(usage, "subfacets", n_subfacets);
1648 flush(struct ofproto *ofproto_)
1650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1651 struct subfacet *subfacet, *next_subfacet;
1652 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1656 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1657 &ofproto->backer->subfacets) {
1658 if (subfacet->facet->ofproto != ofproto) {
1662 if (subfacet->path != SF_NOT_INSTALLED) {
1663 batch[n_batch++] = subfacet;
1664 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1665 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1669 subfacet_destroy(subfacet);
1674 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1679 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1680 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1682 *arp_match_ip = true;
1683 *actions = (OFPUTIL_A_OUTPUT |
1684 OFPUTIL_A_SET_VLAN_VID |
1685 OFPUTIL_A_SET_VLAN_PCP |
1686 OFPUTIL_A_STRIP_VLAN |
1687 OFPUTIL_A_SET_DL_SRC |
1688 OFPUTIL_A_SET_DL_DST |
1689 OFPUTIL_A_SET_NW_SRC |
1690 OFPUTIL_A_SET_NW_DST |
1691 OFPUTIL_A_SET_NW_TOS |
1692 OFPUTIL_A_SET_TP_SRC |
1693 OFPUTIL_A_SET_TP_DST |
1698 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1701 struct dpif_dp_stats s;
1702 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1705 strcpy(ots->name, "classifier");
1707 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1708 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1709 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1710 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1712 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1713 ots->lookup_count = htonll(n_lookup);
1714 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1717 static struct ofport *
1720 struct ofport_dpif *port = xmalloc(sizeof *port);
1725 port_dealloc(struct ofport *port_)
1727 struct ofport_dpif *port = ofport_dpif_cast(port_);
1732 port_construct(struct ofport *port_)
1734 struct ofport_dpif *port = ofport_dpif_cast(port_);
1735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1736 const struct netdev *netdev = port->up.netdev;
1737 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1738 struct dpif_port dpif_port;
1741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1742 port->bundle = NULL;
1745 port->may_enable = true;
1746 port->stp_port = NULL;
1747 port->stp_state = STP_DISABLED;
1748 port->is_tunnel = false;
1752 port->realdev_ofp_port = 0;
1753 port->vlandev_vid = 0;
1754 port->carrier_seq = netdev_get_carrier_resets(netdev);
1756 if (netdev_vport_is_patch(netdev)) {
1757 /* By bailing out here, we don't submit the port to the sFlow module
1758 * to be considered for counter polling export. This is correct
1759 * because the patch port represents an interface that sFlow considers
1760 * to be "internal" to the switch as a whole, and therefore not an
1761 * candidate for counter polling. */
1762 port->odp_port = ODPP_NONE;
1763 ofport_update_peer(port);
1767 error = dpif_port_query_by_name(ofproto->backer->dpif,
1768 netdev_vport_get_dpif_port(netdev, namebuf,
1775 port->odp_port = dpif_port.port_no;
1777 if (netdev_get_tunnel_config(netdev)) {
1778 tnl_port_add(port, port->up.netdev, port->odp_port);
1779 port->is_tunnel = true;
1781 /* Sanity-check that a mapping doesn't already exist. This
1782 * shouldn't happen for non-tunnel ports. */
1783 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1784 VLOG_ERR("port %s already has an OpenFlow port number",
1786 dpif_port_destroy(&dpif_port);
1790 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1791 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1792 hash_odp_port(port->odp_port));
1793 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1795 dpif_port_destroy(&dpif_port);
1797 if (ofproto->sflow) {
1798 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1805 port_destruct(struct ofport *port_)
1807 struct ofport_dpif *port = ofport_dpif_cast(port_);
1808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1809 const char *devname = netdev_get_name(port->up.netdev);
1810 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1811 const char *dp_port_name;
1813 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1814 xlate_ofport_remove(port);
1816 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1818 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1819 /* The underlying device is still there, so delete it. This
1820 * happens when the ofproto is being destroyed, since the caller
1821 * assumes that removal of attached ports will happen as part of
1823 if (!port->is_tunnel) {
1824 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1829 port->peer->peer = NULL;
1833 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1834 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1835 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1836 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1840 sset_find_and_delete(&ofproto->ports, devname);
1841 sset_find_and_delete(&ofproto->ghost_ports, devname);
1842 bundle_remove(port_);
1843 set_cfm(port_, NULL);
1844 set_bfd(port_, NULL);
1845 if (ofproto->sflow) {
1846 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1853 port_modified(struct ofport *port_)
1855 struct ofport_dpif *port = ofport_dpif_cast(port_);
1857 if (port->bundle && port->bundle->bond) {
1858 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1862 cfm_set_netdev(port->cfm, port->up.netdev);
1865 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1867 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1871 ofport_update_peer(port);
1875 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1877 struct ofport_dpif *port = ofport_dpif_cast(port_);
1878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1879 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1881 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1882 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1883 OFPUTIL_PC_NO_PACKET_IN)) {
1884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1886 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1887 bundle_update(port->bundle);
1893 set_sflow(struct ofproto *ofproto_,
1894 const struct ofproto_sflow_options *sflow_options)
1896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1897 struct dpif_sflow *ds = ofproto->sflow;
1899 if (sflow_options) {
1901 struct ofport_dpif *ofport;
1903 ds = ofproto->sflow = dpif_sflow_create();
1904 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1905 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1907 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1909 dpif_sflow_set_options(ds, sflow_options);
1912 dpif_sflow_unref(ds);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914 ofproto->sflow = NULL;
1922 struct ofproto *ofproto_,
1923 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1924 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1925 size_t n_flow_exporters_options)
1927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1928 struct dpif_ipfix *di = ofproto->ipfix;
1930 if (bridge_exporter_options || flow_exporters_options) {
1932 di = ofproto->ipfix = dpif_ipfix_create();
1934 dpif_ipfix_set_options(
1935 di, bridge_exporter_options, flow_exporters_options,
1936 n_flow_exporters_options);
1939 dpif_ipfix_unref(di);
1940 ofproto->ipfix = NULL;
1947 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1949 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 struct ofproto_dpif *ofproto;
1958 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1959 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1960 ofport->cfm = cfm_create(ofport->up.netdev);
1963 if (cfm_configure(ofport->cfm, s)) {
1969 cfm_unref(ofport->cfm);
1975 get_cfm_status(const struct ofport *ofport_,
1976 struct ofproto_cfm_status *status)
1978 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1981 status->faults = cfm_get_fault(ofport->cfm);
1982 status->remote_opstate = cfm_get_opup(ofport->cfm);
1983 status->health = cfm_get_health(ofport->cfm);
1984 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1992 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1995 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2000 if (ofport->bfd != old) {
2001 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2008 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2010 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2013 bfd_get_status(ofport->bfd, smap);
2020 /* Spanning Tree. */
2023 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2025 struct ofproto_dpif *ofproto = ofproto_;
2026 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2027 struct ofport_dpif *ofport;
2029 ofport = stp_port_get_aux(sp);
2031 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2032 ofproto->up.name, port_num);
2034 struct eth_header *eth = pkt->l2;
2036 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2037 if (eth_addr_is_zero(eth->eth_src)) {
2038 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2039 "with unknown MAC", ofproto->up.name, port_num);
2041 send_packet(ofport, pkt);
2047 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2049 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2053 /* Only revalidate flows if the configuration changed. */
2054 if (!s != !ofproto->stp) {
2055 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2059 if (!ofproto->stp) {
2060 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2061 send_bpdu_cb, ofproto);
2062 ofproto->stp_last_tick = time_msec();
2065 stp_set_bridge_id(ofproto->stp, s->system_id);
2066 stp_set_bridge_priority(ofproto->stp, s->priority);
2067 stp_set_hello_time(ofproto->stp, s->hello_time);
2068 stp_set_max_age(ofproto->stp, s->max_age);
2069 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2071 struct ofport *ofport;
2073 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2074 set_stp_port(ofport, NULL);
2077 stp_unref(ofproto->stp);
2078 ofproto->stp = NULL;
2085 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2091 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2092 s->designated_root = stp_get_designated_root(ofproto->stp);
2093 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2102 update_stp_port_state(struct ofport_dpif *ofport)
2104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2105 enum stp_state state;
2107 /* Figure out new state. */
2108 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2112 if (ofport->stp_state != state) {
2113 enum ofputil_port_state of_state;
2116 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2117 netdev_get_name(ofport->up.netdev),
2118 stp_state_name(ofport->stp_state),
2119 stp_state_name(state));
2120 if (stp_learn_in_state(ofport->stp_state)
2121 != stp_learn_in_state(state)) {
2122 /* xxx Learning action flows should also be flushed. */
2123 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2124 mac_learning_flush(ofproto->ml);
2125 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2127 fwd_change = stp_forward_in_state(ofport->stp_state)
2128 != stp_forward_in_state(state);
2130 ofproto->backer->need_revalidate = REV_STP;
2131 ofport->stp_state = state;
2132 ofport->stp_state_entered = time_msec();
2134 if (fwd_change && ofport->bundle) {
2135 bundle_update(ofport->bundle);
2138 /* Update the STP state bits in the OpenFlow port description. */
2139 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2140 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2141 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2142 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2143 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2145 ofproto_port_set_state(&ofport->up, of_state);
2149 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2150 * caller is responsible for assigning STP port numbers and ensuring
2151 * there are no duplicates. */
2153 set_stp_port(struct ofport *ofport_,
2154 const struct ofproto_port_stp_settings *s)
2156 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2158 struct stp_port *sp = ofport->stp_port;
2160 if (!s || !s->enable) {
2162 ofport->stp_port = NULL;
2163 stp_port_disable(sp);
2164 update_stp_port_state(ofport);
2167 } else if (sp && stp_port_no(sp) != s->port_num
2168 && ofport == stp_port_get_aux(sp)) {
2169 /* The port-id changed, so disable the old one if it's not
2170 * already in use by another port. */
2171 stp_port_disable(sp);
2174 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2175 stp_port_enable(sp);
2177 stp_port_set_aux(sp, ofport);
2178 stp_port_set_priority(sp, s->priority);
2179 stp_port_set_path_cost(sp, s->path_cost);
2181 update_stp_port_state(ofport);
2187 get_stp_port_status(struct ofport *ofport_,
2188 struct ofproto_port_stp_status *s)
2190 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2192 struct stp_port *sp = ofport->stp_port;
2194 if (!ofproto->stp || !sp) {
2200 s->port_id = stp_port_get_id(sp);
2201 s->state = stp_port_get_state(sp);
2202 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2203 s->role = stp_port_get_role(sp);
2204 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2210 stp_run(struct ofproto_dpif *ofproto)
2213 long long int now = time_msec();
2214 long long int elapsed = now - ofproto->stp_last_tick;
2215 struct stp_port *sp;
2218 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2219 ofproto->stp_last_tick = now;
2221 while (stp_get_changed_port(ofproto->stp, &sp)) {
2222 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2225 update_stp_port_state(ofport);
2229 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2230 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2231 mac_learning_flush(ofproto->ml);
2232 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2238 stp_wait(struct ofproto_dpif *ofproto)
2241 poll_timer_wait(1000);
2246 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2249 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2252 if (ofport->n_qdscp != n_qdscp
2253 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2254 n_qdscp * sizeof *qdscp))) {
2255 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2256 free(ofport->qdscp);
2257 ofport->qdscp = n_qdscp
2258 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2260 ofport->n_qdscp = n_qdscp;
2268 /* Expires all MAC learning entries associated with 'bundle' and forces its
2269 * ofproto to revalidate every flow.
2271 * Normally MAC learning entries are removed only from the ofproto associated
2272 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2273 * are removed from every ofproto. When patch ports and SLB bonds are in use
2274 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2275 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2276 * with the host from which it migrated. */
2278 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2280 struct ofproto_dpif *ofproto = bundle->ofproto;
2281 struct mac_learning *ml = ofproto->ml;
2282 struct mac_entry *mac, *next_mac;
2284 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2285 ovs_rwlock_wrlock(&ml->rwlock);
2286 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2287 if (mac->port.p == bundle) {
2289 struct ofproto_dpif *o;
2291 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2293 struct mac_entry *e;
2295 ovs_rwlock_wrlock(&o->ml->rwlock);
2296 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2298 mac_learning_expire(o->ml, e);
2300 ovs_rwlock_unlock(&o->ml->rwlock);
2305 mac_learning_expire(ml, mac);
2308 ovs_rwlock_unlock(&ml->rwlock);
2311 static struct ofbundle *
2312 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2314 struct ofbundle *bundle;
2316 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2317 &ofproto->bundles) {
2318 if (bundle->aux == aux) {
2326 bundle_update(struct ofbundle *bundle)
2328 struct ofport_dpif *port;
2330 bundle->floodable = true;
2331 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2332 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2333 || !stp_forward_in_state(port->stp_state)) {
2334 bundle->floodable = false;
2341 bundle_del_port(struct ofport_dpif *port)
2343 struct ofbundle *bundle = port->bundle;
2345 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2347 list_remove(&port->bundle_node);
2348 port->bundle = NULL;
2351 lacp_slave_unregister(bundle->lacp, port);
2354 bond_slave_unregister(bundle->bond, port);
2357 bundle_update(bundle);
2361 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2362 struct lacp_slave_settings *lacp)
2364 struct ofport_dpif *port;
2366 port = get_ofp_port(bundle->ofproto, ofp_port);
2371 if (port->bundle != bundle) {
2372 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2374 bundle_del_port(port);
2377 port->bundle = bundle;
2378 list_push_back(&bundle->ports, &port->bundle_node);
2379 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2380 || !stp_forward_in_state(port->stp_state)) {
2381 bundle->floodable = false;
2385 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 lacp_slave_register(bundle->lacp, port, lacp);
2393 bundle_destroy(struct ofbundle *bundle)
2395 struct ofproto_dpif *ofproto;
2396 struct ofport_dpif *port, *next_port;
2402 ofproto = bundle->ofproto;
2403 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2405 xlate_bundle_remove(bundle);
2407 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2408 bundle_del_port(port);
2411 bundle_flush_macs(bundle, true);
2412 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2414 free(bundle->trunks);
2415 lacp_unref(bundle->lacp);
2416 bond_unref(bundle->bond);
2421 bundle_set(struct ofproto *ofproto_, void *aux,
2422 const struct ofproto_bundle_settings *s)
2424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2425 bool need_flush = false;
2426 struct ofport_dpif *port;
2427 struct ofbundle *bundle;
2428 unsigned long *trunks;
2434 bundle_destroy(bundle_lookup(ofproto, aux));
2438 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2439 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2441 bundle = bundle_lookup(ofproto, aux);
2443 bundle = xmalloc(sizeof *bundle);
2445 bundle->ofproto = ofproto;
2446 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2447 hash_pointer(aux, 0));
2449 bundle->name = NULL;
2451 list_init(&bundle->ports);
2452 bundle->vlan_mode = PORT_VLAN_TRUNK;
2454 bundle->trunks = NULL;
2455 bundle->use_priority_tags = s->use_priority_tags;
2456 bundle->lacp = NULL;
2457 bundle->bond = NULL;
2459 bundle->floodable = true;
2460 mbridge_register_bundle(ofproto->mbridge, bundle);
2463 if (!bundle->name || strcmp(s->name, bundle->name)) {
2465 bundle->name = xstrdup(s->name);
2470 if (!bundle->lacp) {
2471 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2472 bundle->lacp = lacp_create();
2474 lacp_configure(bundle->lacp, s->lacp);
2476 lacp_unref(bundle->lacp);
2477 bundle->lacp = NULL;
2480 /* Update set of ports. */
2482 for (i = 0; i < s->n_slaves; i++) {
2483 if (!bundle_add_port(bundle, s->slaves[i],
2484 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2488 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2489 struct ofport_dpif *next_port;
2491 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2492 for (i = 0; i < s->n_slaves; i++) {
2493 if (s->slaves[i] == port->up.ofp_port) {
2498 bundle_del_port(port);
2502 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2504 if (list_is_empty(&bundle->ports)) {
2505 bundle_destroy(bundle);
2509 /* Set VLAN tagging mode */
2510 if (s->vlan_mode != bundle->vlan_mode
2511 || s->use_priority_tags != bundle->use_priority_tags) {
2512 bundle->vlan_mode = s->vlan_mode;
2513 bundle->use_priority_tags = s->use_priority_tags;
2518 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2519 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2521 if (vlan != bundle->vlan) {
2522 bundle->vlan = vlan;
2526 /* Get trunked VLANs. */
2527 switch (s->vlan_mode) {
2528 case PORT_VLAN_ACCESS:
2532 case PORT_VLAN_TRUNK:
2533 trunks = CONST_CAST(unsigned long *, s->trunks);
2536 case PORT_VLAN_NATIVE_UNTAGGED:
2537 case PORT_VLAN_NATIVE_TAGGED:
2538 if (vlan != 0 && (!s->trunks
2539 || !bitmap_is_set(s->trunks, vlan)
2540 || bitmap_is_set(s->trunks, 0))) {
2541 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2543 trunks = bitmap_clone(s->trunks, 4096);
2545 trunks = bitmap_allocate1(4096);
2547 bitmap_set1(trunks, vlan);
2548 bitmap_set0(trunks, 0);
2550 trunks = CONST_CAST(unsigned long *, s->trunks);
2557 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2558 free(bundle->trunks);
2559 if (trunks == s->trunks) {
2560 bundle->trunks = vlan_bitmap_clone(trunks);
2562 bundle->trunks = trunks;
2567 if (trunks != s->trunks) {
2572 if (!list_is_short(&bundle->ports)) {
2573 bundle->ofproto->has_bonded_bundles = true;
2575 if (bond_reconfigure(bundle->bond, s->bond)) {
2576 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2579 bundle->bond = bond_create(s->bond);
2580 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2583 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2584 bond_slave_register(bundle->bond, port, port->up.netdev);
2587 bond_unref(bundle->bond);
2588 bundle->bond = NULL;
2591 /* If we changed something that would affect MAC learning, un-learn
2592 * everything on this port and force flow revalidation. */
2594 bundle_flush_macs(bundle, false);
2601 bundle_remove(struct ofport *port_)
2603 struct ofport_dpif *port = ofport_dpif_cast(port_);
2604 struct ofbundle *bundle = port->bundle;
2607 bundle_del_port(port);
2608 if (list_is_empty(&bundle->ports)) {
2609 bundle_destroy(bundle);
2610 } else if (list_is_short(&bundle->ports)) {
2611 bond_unref(bundle->bond);
2612 bundle->bond = NULL;
2618 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2620 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2621 struct ofport_dpif *port = port_;
2622 uint8_t ea[ETH_ADDR_LEN];
2625 error = netdev_get_etheraddr(port->up.netdev, ea);
2627 struct ofpbuf packet;
2630 ofpbuf_init(&packet, 0);
2631 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2633 memcpy(packet_pdu, pdu, pdu_size);
2635 send_packet(port, &packet);
2636 ofpbuf_uninit(&packet);
2638 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2639 "%s (%s)", port->bundle->name,
2640 netdev_get_name(port->up.netdev), ovs_strerror(error));
2645 bundle_send_learning_packets(struct ofbundle *bundle)
2647 struct ofproto_dpif *ofproto = bundle->ofproto;
2648 int error, n_packets, n_errors;
2649 struct mac_entry *e;
2651 error = n_packets = n_errors = 0;
2652 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2653 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2654 if (e->port.p != bundle) {
2655 struct ofpbuf *learning_packet;
2656 struct ofport_dpif *port;
2660 /* The assignment to "port" is unnecessary but makes "grep"ing for
2661 * struct ofport_dpif more effective. */
2662 learning_packet = bond_compose_learning_packet(bundle->bond,
2666 ret = send_packet(port, learning_packet);
2667 ofpbuf_delete(learning_packet);
2675 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2678 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2679 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2680 "packets, last error was: %s",
2681 bundle->name, n_errors, n_packets, ovs_strerror(error));
2683 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2684 bundle->name, n_packets);
2689 bundle_run(struct ofbundle *bundle)
2692 lacp_run(bundle->lacp, send_pdu_cb);
2695 struct ofport_dpif *port;
2697 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2698 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2701 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2702 bundle->ofproto->backer->need_revalidate = REV_BOND;
2705 if (bond_should_send_learning_packets(bundle->bond)) {
2706 bundle_send_learning_packets(bundle);
2712 bundle_wait(struct ofbundle *bundle)
2715 lacp_wait(bundle->lacp);
2718 bond_wait(bundle->bond);
2725 mirror_set__(struct ofproto *ofproto_, void *aux,
2726 const struct ofproto_mirror_settings *s)
2728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2729 struct ofbundle **srcs, **dsts;
2734 mirror_destroy(ofproto->mbridge, aux);
2738 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2739 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2741 for (i = 0; i < s->n_srcs; i++) {
2742 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2745 for (i = 0; i < s->n_dsts; i++) {
2746 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2749 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2750 s->n_dsts, s->src_vlans,
2751 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2758 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2759 uint64_t *packets, uint64_t *bytes)
2762 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2767 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2770 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2771 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2772 mac_learning_flush(ofproto->ml);
2774 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2779 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2781 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2782 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2783 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2787 forward_bpdu_changed(struct ofproto *ofproto_)
2789 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2790 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2794 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2798 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2799 mac_learning_set_idle_time(ofproto->ml, idle_time);
2800 mac_learning_set_max_entries(ofproto->ml, max_entries);
2801 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2806 static struct ofport_dpif *
2807 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2809 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2810 return ofport ? ofport_dpif_cast(ofport) : NULL;
2813 static struct ofport_dpif *
2814 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2816 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2817 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2821 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2822 struct ofproto_port *ofproto_port,
2823 struct dpif_port *dpif_port)
2825 ofproto_port->name = dpif_port->name;
2826 ofproto_port->type = dpif_port->type;
2827 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2831 ofport_update_peer(struct ofport_dpif *ofport)
2833 const struct ofproto_dpif *ofproto;
2834 struct dpif_backer *backer;
2835 const char *peer_name;
2837 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2841 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2842 backer->need_revalidate = REV_RECONFIGURE;
2845 ofport->peer->peer = NULL;
2846 ofport->peer = NULL;
2849 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2854 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2855 struct ofport *peer_ofport;
2856 struct ofport_dpif *peer;
2857 const char *peer_peer;
2859 if (ofproto->backer != backer) {
2863 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2868 peer = ofport_dpif_cast(peer_ofport);
2869 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2870 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2872 ofport->peer = peer;
2873 ofport->peer->peer = ofport;
2881 port_run_fast(struct ofport_dpif *ofport)
2883 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2884 struct ofpbuf packet;
2886 ofpbuf_init(&packet, 0);
2887 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2888 send_packet(ofport, &packet);
2889 ofpbuf_uninit(&packet);
2892 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2893 struct ofpbuf packet;
2895 ofpbuf_init(&packet, 0);
2896 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2897 send_packet(ofport, &packet);
2898 ofpbuf_uninit(&packet);
2903 port_run(struct ofport_dpif *ofport)
2905 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2906 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2907 bool enable = netdev_get_carrier(ofport->up.netdev);
2909 ofport->carrier_seq = carrier_seq;
2911 port_run_fast(ofport);
2914 int cfm_opup = cfm_get_opup(ofport->cfm);
2916 cfm_run(ofport->cfm);
2917 enable = enable && !cfm_get_fault(ofport->cfm);
2919 if (cfm_opup >= 0) {
2920 enable = enable && cfm_opup;
2925 bfd_run(ofport->bfd);
2926 enable = enable && bfd_forwarding(ofport->bfd);
2929 if (ofport->bundle) {
2930 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2931 if (carrier_changed) {
2932 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2936 if (ofport->may_enable != enable) {
2937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2938 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2941 ofport->may_enable = enable;
2945 port_wait(struct ofport_dpif *ofport)
2948 cfm_wait(ofport->cfm);
2952 bfd_wait(ofport->bfd);
2957 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2958 struct ofproto_port *ofproto_port)
2960 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2961 struct dpif_port dpif_port;
2964 if (sset_contains(&ofproto->ghost_ports, devname)) {
2965 const char *type = netdev_get_type_from_name(devname);
2967 /* We may be called before ofproto->up.port_by_name is populated with
2968 * the appropriate ofport. For this reason, we must get the name and
2969 * type from the netdev layer directly. */
2971 const struct ofport *ofport;
2973 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2974 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2975 ofproto_port->name = xstrdup(devname);
2976 ofproto_port->type = xstrdup(type);
2982 if (!sset_contains(&ofproto->ports, devname)) {
2985 error = dpif_port_query_by_name(ofproto->backer->dpif,
2986 devname, &dpif_port);
2988 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2994 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2997 const char *devname = netdev_get_name(netdev);
2998 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2999 const char *dp_port_name;
3001 if (netdev_vport_is_patch(netdev)) {
3002 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3006 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3007 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3008 odp_port_t port_no = ODPP_NONE;
3011 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3015 if (netdev_get_tunnel_config(netdev)) {
3016 simap_put(&ofproto->backer->tnl_backers,
3017 dp_port_name, odp_to_u32(port_no));
3021 if (netdev_get_tunnel_config(netdev)) {
3022 sset_add(&ofproto->ghost_ports, devname);
3024 sset_add(&ofproto->ports, devname);
3030 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3033 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3040 sset_find_and_delete(&ofproto->ghost_ports,
3041 netdev_get_name(ofport->up.netdev));
3042 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3043 if (!ofport->is_tunnel) {
3044 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3046 /* The caller is going to close ofport->up.netdev. If this is a
3047 * bonded port, then the bond is using that netdev, so remove it
3048 * from the bond. The client will need to reconfigure everything
3049 * after deleting ports, so then the slave will get re-added. */
3050 bundle_remove(&ofport->up);
3057 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3059 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3064 error = netdev_get_stats(ofport->up.netdev, stats);
3066 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3069 /* ofproto->stats.tx_packets represents packets that we created
3070 * internally and sent to some port (e.g. packets sent with
3071 * send_packet()). Account for them as if they had come from
3072 * OFPP_LOCAL and got forwarded. */
3074 if (stats->rx_packets != UINT64_MAX) {
3075 stats->rx_packets += ofproto->stats.tx_packets;
3078 if (stats->rx_bytes != UINT64_MAX) {
3079 stats->rx_bytes += ofproto->stats.tx_bytes;
3082 /* ofproto->stats.rx_packets represents packets that were received on
3083 * some port and we processed internally and dropped (e.g. STP).
3084 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3086 if (stats->tx_packets != UINT64_MAX) {
3087 stats->tx_packets += ofproto->stats.rx_packets;
3090 if (stats->tx_bytes != UINT64_MAX) {
3091 stats->tx_bytes += ofproto->stats.rx_bytes;
3098 struct port_dump_state {
3103 struct ofproto_port port;
3108 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3110 *statep = xzalloc(sizeof(struct port_dump_state));
3115 port_dump_next(const struct ofproto *ofproto_, void *state_,
3116 struct ofproto_port *port)
3118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3119 struct port_dump_state *state = state_;
3120 const struct sset *sset;
3121 struct sset_node *node;
3123 if (state->has_port) {
3124 ofproto_port_destroy(&state->port);
3125 state->has_port = false;
3127 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3128 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3131 error = port_query_by_name(ofproto_, node->name, &state->port);
3133 *port = state->port;
3134 state->has_port = true;
3136 } else if (error != ENODEV) {
3141 if (!state->ghost) {
3142 state->ghost = true;
3145 return port_dump_next(ofproto_, state_, port);
3152 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3154 struct port_dump_state *state = state_;
3156 if (state->has_port) {
3157 ofproto_port_destroy(&state->port);
3164 port_poll(const struct ofproto *ofproto_, char **devnamep)
3166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3168 if (ofproto->port_poll_errno) {
3169 int error = ofproto->port_poll_errno;
3170 ofproto->port_poll_errno = 0;
3174 if (sset_is_empty(&ofproto->port_poll_set)) {
3178 *devnamep = sset_pop(&ofproto->port_poll_set);
3183 port_poll_wait(const struct ofproto *ofproto_)
3185 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3186 dpif_port_poll_wait(ofproto->backer->dpif);
3190 port_is_lacp_current(const struct ofport *ofport_)
3192 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3193 return (ofport->bundle && ofport->bundle->lacp
3194 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3198 /* Upcall handling. */
3200 /* Flow miss batching.
3202 * Some dpifs implement operations faster when you hand them off in a batch.
3203 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3204 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3205 * more packets, plus possibly installing the flow in the dpif.
3207 * So far we only batch the operations that affect flow setup time the most.
3208 * It's possible to batch more than that, but the benefit might be minimal. */
3210 struct hmap_node hmap_node;
3211 struct ofproto_dpif *ofproto;
3213 enum odp_key_fitness key_fitness;
3214 const struct nlattr *key;
3216 struct list packets;
3217 enum dpif_upcall_type upcall_type;
3220 struct flow_miss_op {
3221 struct dpif_op dpif_op;
3223 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3224 struct xlate_out xout;
3225 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3227 struct ofpbuf mask; /* Flow mask for "put" ops. */
3228 struct odputil_keybuf maskbuf;
3230 /* If this is a "put" op, then a pointer to the subfacet that should
3231 * be marked as uninstalled if the operation fails. */
3232 struct subfacet *subfacet;
3235 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3236 * OpenFlow controller as necessary according to their individual
3237 * configurations. */
3239 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3240 const struct flow *flow)
3242 struct ofputil_packet_in pin;
3244 pin.packet = packet->data;
3245 pin.packet_len = packet->size;
3246 pin.reason = OFPR_NO_MATCH;
3247 pin.controller_id = 0;
3252 pin.send_len = 0; /* not used for flow table misses */
3254 flow_get_metadata(flow, &pin.fmd);
3256 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3259 static struct flow_miss *
3260 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3261 const struct flow *flow, uint32_t hash)
3263 struct flow_miss *miss;
3265 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3266 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3274 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3275 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3276 * 'miss' is associated with a subfacet the caller must also initialize the
3277 * returned op->subfacet, and if anything needs to be freed after processing
3278 * the op, the caller must initialize op->garbage also. */
3280 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3281 struct flow_miss_op *op)
3283 if (miss->flow.in_port.ofp_port
3284 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3285 miss->flow.vlan_tci)) {
3286 /* This packet was received on a VLAN splinter port. We
3287 * added a VLAN to the packet to make the packet resemble
3288 * the flow, but the actions were composed assuming that
3289 * the packet contained no VLAN. So, we must remove the
3290 * VLAN header from the packet before trying to execute the
3292 eth_pop_vlan(packet);
3295 op->subfacet = NULL;
3296 op->xout_garbage = false;
3297 op->dpif_op.type = DPIF_OP_EXECUTE;
3298 op->dpif_op.u.execute.key = miss->key;
3299 op->dpif_op.u.execute.key_len = miss->key_len;
3300 op->dpif_op.u.execute.packet = packet;
3301 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3304 /* Helper for handle_flow_miss_without_facet() and
3305 * handle_flow_miss_with_facet(). */
3307 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3308 const struct flow *flow, bool fail_open)
3312 * Extra-special case for fail-open mode.
3314 * We are in fail-open mode and the packet matched the fail-open
3315 * rule, but we are connected to a controller too. We should send
3316 * the packet up to the controller in the hope that it will try to
3317 * set up a flow and thereby allow us to exit fail-open.
3319 * See the top-level comment in fail-open.c for more information.
3321 send_packet_in_miss(ofproto, packet, flow);
3325 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3326 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3327 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3328 * return value of true). However, for short flows the cost of bookkeeping is
3329 * much higher than the benefits, so when the datapath holds a large number of
3330 * flows we impose some heuristics to decide which flows are likely to be worth
3333 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3335 struct dpif_backer *backer = miss->ofproto->backer;
3338 switch (flow_miss_model) {
3339 case OFPROTO_HANDLE_MISS_AUTO:
3341 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3343 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3347 if (!backer->governor) {
3350 n_subfacets = hmap_count(&backer->subfacets);
3351 if (n_subfacets * 2 <= flow_eviction_threshold) {
3355 backer->governor = governor_create();
3358 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3359 return governor_should_install_flow(backer->governor, hash,
3360 list_size(&miss->packets));
3363 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3364 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3365 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3367 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3368 struct flow_miss *miss,
3369 struct flow_miss_op *ops, size_t *n_ops)
3371 struct ofpbuf *packet;
3373 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3375 COVERAGE_INC(facet_suppress);
3377 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3378 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3381 struct xlate_in xin;
3383 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3384 xlate_actions_for_side_effects(&xin);
3387 if (xout->odp_actions.size) {
3388 struct flow_miss_op *op = &ops[*n_ops];
3389 struct dpif_execute *execute = &op->dpif_op.u.execute;
3391 init_flow_miss_execute_op(miss, packet, op);
3392 xlate_out_copy(&op->xout, xout);
3393 execute->actions = op->xout.odp_actions.data;
3394 execute->actions_len = op->xout.odp_actions.size;
3395 op->xout_garbage = true;
3402 /* Handles 'miss', which matches 'facet'. May add any required datapath
3403 * operations to 'ops', incrementing '*n_ops' for each new op.
3405 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3406 * This is really important only for new facets: if we just called time_msec()
3407 * here, then the new subfacet or its packets could look (occasionally) as
3408 * though it was used some time after the facet was used. That can make a
3409 * one-packet flow look like it has a nonzero duration, which looks odd in
3410 * e.g. NetFlow statistics.
3412 * If non-null, 'stats' will be folded into 'facet'. */
3414 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3415 long long int now, struct dpif_flow_stats *stats,
3416 struct flow_miss_op *ops, size_t *n_ops)
3418 enum subfacet_path want_path;
3419 struct subfacet *subfacet;
3420 struct ofpbuf *packet;
3422 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3424 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3425 struct flow_miss_op *op = &ops[*n_ops];
3427 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3430 if (want_path != SF_FAST_PATH) {
3431 struct rule_dpif *rule;
3432 struct xlate_in xin;
3434 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3435 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3436 xlate_actions_for_side_effects(&xin);
3439 if (facet->xout.odp_actions.size) {
3440 struct dpif_execute *execute = &op->dpif_op.u.execute;
3442 init_flow_miss_execute_op(miss, packet, op);
3443 execute->actions = facet->xout.odp_actions.data,
3444 execute->actions_len = facet->xout.odp_actions.size;
3449 /* Don't install the flow if it's the result of the "userspace"
3450 * action for an already installed facet. This can occur when a
3451 * datapath flow with wildcards has a "userspace" action and flows
3452 * sent to userspace result in a different subfacet, which will then
3453 * be rejected as overlapping by the datapath. */
3454 if (miss->upcall_type == DPIF_UC_ACTION
3455 && !list_is_empty(&facet->subfacets)) {
3457 facet->used = MAX(facet->used, stats->used);
3458 facet->packet_count += stats->n_packets;
3459 facet->byte_count += stats->n_bytes;
3460 facet->tcp_flags |= stats->tcp_flags;
3465 subfacet = subfacet_create(facet, miss, now);
3467 subfacet_update_stats(subfacet, stats);
3470 if (subfacet->path != want_path) {
3471 struct flow_miss_op *op = &ops[(*n_ops)++];
3472 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3474 subfacet->path = want_path;
3476 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3477 if (enable_megaflows) {
3478 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3479 &miss->flow, UINT32_MAX);
3482 op->xout_garbage = false;
3483 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3484 op->subfacet = subfacet;
3485 put->flags = DPIF_FP_CREATE;
3486 put->key = miss->key;
3487 put->key_len = miss->key_len;
3488 put->mask = op->mask.data;
3489 put->mask_len = op->mask.size;
3491 if (want_path == SF_FAST_PATH) {
3492 put->actions = facet->xout.odp_actions.data;
3493 put->actions_len = facet->xout.odp_actions.size;
3495 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3496 op->slow_stub, sizeof op->slow_stub,
3497 &put->actions, &put->actions_len);
3503 /* Handles flow miss 'miss'. May add any required datapath operations
3504 * to 'ops', incrementing '*n_ops' for each new op. */
3506 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3509 struct ofproto_dpif *ofproto = miss->ofproto;
3510 struct dpif_flow_stats stats__;
3511 struct dpif_flow_stats *stats = &stats__;
3512 struct ofpbuf *packet;
3513 struct facet *facet;
3517 memset(stats, 0, sizeof *stats);
3519 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3520 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3521 stats->n_bytes += packet->size;
3525 facet = facet_lookup_valid(ofproto, &miss->flow);
3527 struct flow_wildcards wc;
3528 struct rule_dpif *rule;
3529 struct xlate_out xout;
3530 struct xlate_in xin;
3532 flow_wildcards_init_catchall(&wc);
3533 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3534 rule_credit_stats(rule, stats);
3536 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3538 xin.resubmit_stats = stats;
3539 xin.may_learn = true;
3540 xlate_actions(&xin, &xout);
3541 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3543 /* There does not exist a bijection between 'struct flow' and datapath
3544 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3545 * assumption used throughout the facet and subfacet handling code.
3546 * Since we have to handle these misses in userspace anyway, we simply
3547 * skip facet creation, avoiding the problem altogether. */
3548 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3549 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3550 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3554 facet = facet_create(miss, rule, &xout, stats);
3557 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3560 static struct drop_key *
3561 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3564 struct drop_key *drop_key;
3566 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3567 &backer->drop_keys) {
3568 if (drop_key->key_len == key_len
3569 && !memcmp(drop_key->key, key, key_len)) {
3577 drop_key_clear(struct dpif_backer *backer)
3579 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3580 struct drop_key *drop_key, *next;
3582 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3585 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3587 if (error && !VLOG_DROP_WARN(&rl)) {
3588 struct ds ds = DS_EMPTY_INITIALIZER;
3589 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3590 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3591 ovs_strerror(error), ds_cstr(&ds));
3595 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3596 free(drop_key->key);
3602 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3605 struct dpif_upcall *upcall;
3606 struct flow_miss *miss;
3607 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3608 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3609 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3619 /* Construct the to-do list.
3621 * This just amounts to extracting the flow from each packet and sticking
3622 * the packets that have the same flow in the same "flow_miss" structure so
3623 * that we can process them together. */
3626 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3627 struct flow_miss *miss = &misses[n_misses];
3628 struct flow_miss *existing_miss;
3629 struct ofproto_dpif *ofproto;
3630 odp_port_t odp_in_port;
3635 error = xlate_receive(backer, upcall->packet, upcall->key,
3636 upcall->key_len, &flow, &miss->key_fitness,
3637 &ofproto, &odp_in_port);
3638 if (error == ENODEV) {
3639 struct drop_key *drop_key;
3641 /* Received packet on datapath port for which we couldn't
3642 * associate an ofproto. This can happen if a port is removed
3643 * while traffic is being received. Print a rate-limited message
3644 * in case it happens frequently. Install a drop flow so
3645 * that future packets of the flow are inexpensively dropped
3647 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3648 "%"PRIu32, odp_in_port);
3650 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3653 ret = dpif_flow_put(backer->dpif,
3654 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3655 upcall->key, upcall->key_len,
3656 NULL, 0, NULL, 0, NULL);
3659 drop_key = xmalloc(sizeof *drop_key);
3660 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3661 drop_key->key_len = upcall->key_len;
3663 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3664 hash_bytes(drop_key->key, drop_key->key_len, 0));
3673 ofproto->n_missed++;
3674 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3675 &flow.tunnel, &flow.in_port, &miss->flow);
3677 /* Add other packets to a to-do list. */
3678 hash = flow_hash(&miss->flow, 0);
3679 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3680 if (!existing_miss) {
3681 hmap_insert(&todo, &miss->hmap_node, hash);
3682 miss->ofproto = ofproto;
3683 miss->key = upcall->key;
3684 miss->key_len = upcall->key_len;
3685 miss->upcall_type = upcall->type;
3686 list_init(&miss->packets);
3690 miss = existing_miss;
3692 list_push_back(&miss->packets, &upcall->packet->list_node);
3695 /* Process each element in the to-do list, constructing the set of
3696 * operations to batch. */
3698 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3699 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3701 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3703 /* Execute batch. */
3704 for (i = 0; i < n_ops; i++) {
3705 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3707 dpif_operate(backer->dpif, dpif_ops, n_ops);
3709 for (i = 0; i < n_ops; i++) {
3710 if (dpif_ops[i]->error != 0
3711 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3712 && flow_miss_ops[i].subfacet) {
3713 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3715 COVERAGE_INC(subfacet_install_fail);
3717 /* Zero-out subfacet counters when installation failed, but
3718 * datapath reported hits. This should not happen and
3719 * indicates a bug, since if the datapath flow exists, we
3720 * should not be attempting to create a new subfacet. A
3721 * buggy datapath could trigger this, so just zero out the
3722 * counters and log an error. */
3723 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3724 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3725 "datapath reported hits");
3726 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3729 subfacet->path = SF_NOT_INSTALLED;
3733 if (flow_miss_ops[i].xout_garbage) {
3734 xlate_out_uninit(&flow_miss_ops[i].xout);
3737 hmap_destroy(&todo);
3740 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3742 classify_upcall(const struct dpif_upcall *upcall)
3744 size_t userdata_len;
3745 union user_action_cookie cookie;
3747 /* First look at the upcall type. */
3748 switch (upcall->type) {
3749 case DPIF_UC_ACTION:
3755 case DPIF_N_UC_TYPES:
3757 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3761 /* "action" upcalls need a closer look. */
3762 if (!upcall->userdata) {
3763 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3766 userdata_len = nl_attr_get_size(upcall->userdata);
3767 if (userdata_len < sizeof cookie.type
3768 || userdata_len > sizeof cookie) {
3769 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3773 memset(&cookie, 0, sizeof cookie);
3774 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3775 if (userdata_len == sizeof cookie.sflow
3776 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3777 return SFLOW_UPCALL;
3778 } else if (userdata_len == sizeof cookie.slow_path
3779 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3781 } else if (userdata_len == sizeof cookie.flow_sample
3782 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3783 return FLOW_SAMPLE_UPCALL;
3784 } else if (userdata_len == sizeof cookie.ipfix
3785 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3786 return IPFIX_UPCALL;
3788 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3789 " and size %zu", cookie.type, userdata_len);
3795 handle_sflow_upcall(struct dpif_backer *backer,
3796 const struct dpif_upcall *upcall)
3798 struct ofproto_dpif *ofproto;
3799 union user_action_cookie cookie;
3801 odp_port_t odp_in_port;
3803 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3804 &flow, NULL, &ofproto, &odp_in_port)
3805 || !ofproto->sflow) {
3809 memset(&cookie, 0, sizeof cookie);
3810 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3811 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3812 odp_in_port, &cookie);
3816 handle_flow_sample_upcall(struct dpif_backer *backer,
3817 const struct dpif_upcall *upcall)
3819 struct ofproto_dpif *ofproto;
3820 union user_action_cookie cookie;
3823 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3824 &flow, NULL, &ofproto, NULL)
3825 || !ofproto->ipfix) {
3829 memset(&cookie, 0, sizeof cookie);
3830 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3832 /* The flow reflects exactly the contents of the packet. Sample
3833 * the packet using it. */
3834 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3835 cookie.flow_sample.collector_set_id,
3836 cookie.flow_sample.probability,
3837 cookie.flow_sample.obs_domain_id,
3838 cookie.flow_sample.obs_point_id);
3842 handle_ipfix_upcall(struct dpif_backer *backer,
3843 const struct dpif_upcall *upcall)
3845 struct ofproto_dpif *ofproto;
3848 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3849 &flow, NULL, &ofproto, NULL)
3850 || !ofproto->ipfix) {
3854 /* The flow reflects exactly the contents of the packet. Sample
3855 * the packet using it. */
3856 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3860 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3862 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3863 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3864 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3869 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3872 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3873 struct dpif_upcall *upcall = &misses[n_misses];
3874 struct ofpbuf *buf = &miss_bufs[n_misses];
3877 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3878 sizeof miss_buf_stubs[n_misses]);
3879 error = dpif_recv(backer->dpif, upcall, buf);
3885 switch (classify_upcall(upcall)) {
3887 /* Handle it later. */
3892 handle_sflow_upcall(backer, upcall);
3896 case FLOW_SAMPLE_UPCALL:
3897 handle_flow_sample_upcall(backer, upcall);
3902 handle_ipfix_upcall(backer, upcall);
3912 /* Handle deferred MISS_UPCALL processing. */
3913 handle_miss_upcalls(backer, misses, n_misses);
3914 for (i = 0; i < n_misses; i++) {
3915 ofpbuf_uninit(&miss_bufs[i]);
3921 /* Flow expiration. */
3923 static int subfacet_max_idle(const struct dpif_backer *);
3924 static void update_stats(struct dpif_backer *);
3925 static void rule_expire(struct rule_dpif *);
3926 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3928 /* This function is called periodically by run(). Its job is to collect
3929 * updates for the flows that have been installed into the datapath, most
3930 * importantly when they last were used, and then use that information to
3931 * expire flows that have not been used recently.
3933 * Returns the number of milliseconds after which it should be called again. */
3935 expire(struct dpif_backer *backer)
3937 struct ofproto_dpif *ofproto;
3941 /* Periodically clear out the drop keys in an effort to keep them
3942 * relatively few. */
3943 drop_key_clear(backer);
3945 /* Update stats for each flow in the backer. */
3946 update_stats(backer);
3948 n_subfacets = hmap_count(&backer->subfacets);
3950 struct subfacet *subfacet;
3951 long long int total, now;
3955 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3956 total += now - subfacet->created;
3958 backer->avg_subfacet_life += total / n_subfacets;
3960 backer->avg_subfacet_life /= 2;
3962 backer->avg_n_subfacet += n_subfacets;
3963 backer->avg_n_subfacet /= 2;
3965 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3967 max_idle = subfacet_max_idle(backer);
3968 expire_subfacets(backer, max_idle);
3970 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3971 struct rule *rule, *next_rule;
3973 if (ofproto->backer != backer) {
3977 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3979 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3980 &ofproto->up.expirable) {
3981 rule_expire(rule_dpif_cast(rule));
3984 /* All outstanding data in existing flows has been accounted, so it's a
3985 * good time to do bond rebalancing. */
3986 if (ofproto->has_bonded_bundles) {
3987 struct ofbundle *bundle;
3989 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3991 bond_rebalance(bundle->bond);
3997 return MIN(max_idle, 1000);
4000 /* Updates flow table statistics given that the datapath just reported 'stats'
4001 * as 'subfacet''s statistics. */
4003 update_subfacet_stats(struct subfacet *subfacet,
4004 const struct dpif_flow_stats *stats)
4006 struct facet *facet = subfacet->facet;
4007 struct dpif_flow_stats diff;
4009 diff.tcp_flags = stats->tcp_flags;
4010 diff.used = stats->used;
4012 if (stats->n_packets >= subfacet->dp_packet_count) {
4013 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4015 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4019 if (stats->n_bytes >= subfacet->dp_byte_count) {
4020 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4022 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4026 facet->ofproto->n_hit += diff.n_packets;
4027 subfacet->dp_packet_count = stats->n_packets;
4028 subfacet->dp_byte_count = stats->n_bytes;
4029 subfacet_update_stats(subfacet, &diff);
4031 if (facet->accounted_bytes < facet->byte_count) {
4033 facet_account(facet);
4034 facet->accounted_bytes = facet->byte_count;
4038 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4039 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4041 delete_unexpected_flow(struct dpif_backer *backer,
4042 const struct nlattr *key, size_t key_len)
4044 if (!VLOG_DROP_WARN(&rl)) {
4048 odp_flow_key_format(key, key_len, &s);
4049 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4053 COVERAGE_INC(facet_unexpected);
4054 dpif_flow_del(backer->dpif, key, key_len, NULL);
4057 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4059 * This function also pushes statistics updates to rules which each facet
4060 * resubmits into. Generally these statistics will be accurate. However, if a
4061 * facet changes the rule it resubmits into at some time in between
4062 * update_stats() runs, it is possible that statistics accrued to the
4063 * old rule will be incorrectly attributed to the new rule. This could be
4064 * avoided by calling update_stats() whenever rules are created or
4065 * deleted. However, the performance impact of making so many calls to the
4066 * datapath do not justify the benefit of having perfectly accurate statistics.
4068 * In addition, this function maintains per ofproto flow hit counts. The patch
4069 * port is not treated specially. e.g. A packet ingress from br0 patched into
4070 * br1 will increase the hit count of br0 by 1, however, does not affect
4071 * the hit or miss counts of br1.
4074 update_stats(struct dpif_backer *backer)
4076 const struct dpif_flow_stats *stats;
4077 struct dpif_flow_dump dump;
4078 const struct nlattr *key, *mask;
4079 size_t key_len, mask_len;
4081 dpif_flow_dump_start(&dump, backer->dpif);
4082 while (dpif_flow_dump_next(&dump, &key, &key_len,
4083 &mask, &mask_len, NULL, NULL, &stats)) {
4084 struct subfacet *subfacet;
4087 key_hash = odp_flow_key_hash(key, key_len);
4088 subfacet = subfacet_find(backer, key, key_len, key_hash);
4089 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4091 update_subfacet_stats(subfacet, stats);
4095 /* Stats are updated per-packet. */
4098 case SF_NOT_INSTALLED:
4100 delete_unexpected_flow(backer, key, key_len);
4105 dpif_flow_dump_done(&dump);
4107 update_moving_averages(backer);
4110 /* Calculates and returns the number of milliseconds of idle time after which
4111 * subfacets should expire from the datapath. When a subfacet expires, we fold
4112 * its statistics into its facet, and when a facet's last subfacet expires, we
4113 * fold its statistic into its rule. */
4115 subfacet_max_idle(const struct dpif_backer *backer)
4118 * Idle time histogram.
4120 * Most of the time a switch has a relatively small number of subfacets.
4121 * When this is the case we might as well keep statistics for all of them
4122 * in userspace and to cache them in the kernel datapath for performance as
4125 * As the number of subfacets increases, the memory required to maintain
4126 * statistics about them in userspace and in the kernel becomes
4127 * significant. However, with a large number of subfacets it is likely
4128 * that only a few of them are "heavy hitters" that consume a large amount
4129 * of bandwidth. At this point, only heavy hitters are worth caching in
4130 * the kernel and maintaining in userspaces; other subfacets we can
4133 * The technique used to compute the idle time is to build a histogram with
4134 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4135 * that is installed in the kernel gets dropped in the appropriate bucket.
4136 * After the histogram has been built, we compute the cutoff so that only
4137 * the most-recently-used 1% of subfacets (but at least
4138 * flow_eviction_threshold flows) are kept cached. At least
4139 * the most-recently-used bucket of subfacets is kept, so actually an
4140 * arbitrary number of subfacets can be kept in any given expiration run
4141 * (though the next run will delete most of those unless they receive
4144 * This requires a second pass through the subfacets, in addition to the
4145 * pass made by update_stats(), because the former function never looks at
4146 * uninstallable subfacets.
4148 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4149 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4150 int buckets[N_BUCKETS] = { 0 };
4151 int total, subtotal, bucket;
4152 struct subfacet *subfacet;
4156 total = hmap_count(&backer->subfacets);
4157 if (total <= flow_eviction_threshold) {
4158 return N_BUCKETS * BUCKET_WIDTH;
4161 /* Build histogram. */
4163 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4164 long long int idle = now - subfacet->used;
4165 int bucket = (idle <= 0 ? 0
4166 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4167 : (unsigned int) idle / BUCKET_WIDTH);
4171 /* Find the first bucket whose flows should be expired. */
4172 subtotal = bucket = 0;
4174 subtotal += buckets[bucket++];
4175 } while (bucket < N_BUCKETS &&
4176 subtotal < MAX(flow_eviction_threshold, total / 100));
4178 if (VLOG_IS_DBG_ENABLED()) {
4182 ds_put_cstr(&s, "keep");
4183 for (i = 0; i < N_BUCKETS; i++) {
4185 ds_put_cstr(&s, ", drop");
4188 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4191 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4195 return bucket * BUCKET_WIDTH;
4199 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4201 /* Cutoff time for most flows. */
4202 long long int normal_cutoff = time_msec() - dp_max_idle;
4204 /* We really want to keep flows for special protocols around, so use a more
4205 * conservative cutoff. */
4206 long long int special_cutoff = time_msec() - 10000;
4208 struct subfacet *subfacet, *next_subfacet;
4209 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4213 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4214 &backer->subfacets) {
4215 long long int cutoff;
4217 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4221 if (subfacet->used < cutoff) {
4222 if (subfacet->path != SF_NOT_INSTALLED) {
4223 batch[n_batch++] = subfacet;
4224 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4225 subfacet_destroy_batch(backer, batch, n_batch);
4229 subfacet_destroy(subfacet);
4235 subfacet_destroy_batch(backer, batch, n_batch);
4239 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4240 * then delete it entirely. */
4242 rule_expire(struct rule_dpif *rule)
4247 if (rule->up.pending) {
4248 /* We'll have to expire it later. */
4252 /* Has 'rule' expired? */
4254 if (rule->up.hard_timeout
4255 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4256 reason = OFPRR_HARD_TIMEOUT;
4257 } else if (rule->up.idle_timeout
4258 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4259 reason = OFPRR_IDLE_TIMEOUT;
4264 COVERAGE_INC(ofproto_dpif_expired);
4266 /* Get rid of the rule. */
4267 ofproto_rule_expire(&rule->up, reason);
4272 /* Creates and returns a new facet based on 'miss'.
4274 * The caller must already have determined that no facet with an identical
4275 * 'miss->flow' exists in 'miss->ofproto'.
4277 * 'rule' and 'xout' must have been created based on 'miss'.
4279 * 'facet'' statistics are initialized based on 'stats'.
4281 * The facet will initially have no subfacets. The caller should create (at
4282 * least) one subfacet with subfacet_create(). */
4283 static struct facet *
4284 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4285 struct xlate_out *xout, struct dpif_flow_stats *stats)
4287 struct ofproto_dpif *ofproto = miss->ofproto;
4288 struct facet *facet;
4291 facet = xzalloc(sizeof *facet);
4292 facet->ofproto = miss->ofproto;
4293 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4294 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4295 facet->tcp_flags = stats->tcp_flags;
4296 facet->used = stats->used;
4297 facet->flow = miss->flow;
4298 facet->learn_rl = time_msec() + 500;
4300 list_init(&facet->subfacets);
4301 netflow_flow_init(&facet->nf_flow);
4302 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4304 xlate_out_copy(&facet->xout, xout);
4306 match_init(&match, &facet->flow, &facet->xout.wc);
4307 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4308 classifier_insert(&ofproto->facets, &facet->cr);
4310 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4311 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4317 facet_free(struct facet *facet)
4320 xlate_out_uninit(&facet->xout);
4325 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4326 * 'packet', which arrived on 'in_port'. */
4328 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4329 const struct nlattr *odp_actions, size_t actions_len,
4330 struct ofpbuf *packet)
4332 struct odputil_keybuf keybuf;
4336 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4337 odp_flow_key_from_flow(&key, flow,
4338 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4340 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4341 odp_actions, actions_len, packet);
4345 /* Remove 'facet' from its ofproto and free up the associated memory:
4347 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4348 * rule's statistics, via subfacet_uninstall().
4350 * - Removes 'facet' from its rule and from ofproto->facets.
4353 facet_remove(struct facet *facet)
4355 struct subfacet *subfacet, *next_subfacet;
4357 ovs_assert(!list_is_empty(&facet->subfacets));
4359 /* First uninstall all of the subfacets to get final statistics. */
4360 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4361 subfacet_uninstall(subfacet);
4364 /* Flush the final stats to the rule.
4366 * This might require us to have at least one subfacet around so that we
4367 * can use its actions for accounting in facet_account(), which is why we
4368 * have uninstalled but not yet destroyed the subfacets. */
4369 facet_flush_stats(facet);
4371 /* Now we're really all done so destroy everything. */
4372 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4373 &facet->subfacets) {
4374 subfacet_destroy__(subfacet);
4376 classifier_remove(&facet->ofproto->facets, &facet->cr);
4377 cls_rule_destroy(&facet->cr);
4381 /* Feed information from 'facet' back into the learning table to keep it in
4382 * sync with what is actually flowing through the datapath. */
4384 facet_learn(struct facet *facet)
4386 long long int now = time_msec();
4388 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4392 facet->learn_rl = now + 500;
4394 if (!facet->xout.has_learn
4395 && !facet->xout.has_normal
4396 && (!facet->xout.has_fin_timeout
4397 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4401 facet_push_stats(facet, true);
4405 facet_account(struct facet *facet)
4407 const struct nlattr *a;
4412 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4415 n_bytes = facet->byte_count - facet->accounted_bytes;
4417 /* This loop feeds byte counters to bond_account() for rebalancing to use
4418 * as a basis. We also need to track the actual VLAN on which the packet
4419 * is going to be sent to ensure that it matches the one passed to
4420 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4423 * We use the actions from an arbitrary subfacet because they should all
4424 * be equally valid for our purpose. */
4425 vlan_tci = facet->flow.vlan_tci;
4426 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4427 facet->xout.odp_actions.size) {
4428 const struct ovs_action_push_vlan *vlan;
4429 struct ofport_dpif *port;
4431 switch (nl_attr_type(a)) {
4432 case OVS_ACTION_ATTR_OUTPUT:
4433 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4434 if (port && port->bundle && port->bundle->bond) {
4435 bond_account(port->bundle->bond, &facet->flow,
4436 vlan_tci_to_vid(vlan_tci), n_bytes);
4440 case OVS_ACTION_ATTR_POP_VLAN:
4441 vlan_tci = htons(0);
4444 case OVS_ACTION_ATTR_PUSH_VLAN:
4445 vlan = nl_attr_get(a);
4446 vlan_tci = vlan->vlan_tci;
4452 /* Returns true if the only action for 'facet' is to send to the controller.
4453 * (We don't report NetFlow expiration messages for such facets because they
4454 * are just part of the control logic for the network, not real traffic). */
4456 facet_is_controller_flow(struct facet *facet)
4459 struct ofproto_dpif *ofproto = facet->ofproto;
4460 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4462 const struct ofpact *ofpacts = rule->up.ofpacts;
4463 size_t ofpacts_len = rule->up.ofpacts_len;
4465 if (ofpacts_len > 0 &&
4466 ofpacts->type == OFPACT_CONTROLLER &&
4467 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4474 /* Folds all of 'facet''s statistics into its rule. Also updates the
4475 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4476 * 'facet''s statistics in the datapath should have been zeroed and folded into
4477 * its packet and byte counts before this function is called. */
4479 facet_flush_stats(struct facet *facet)
4481 struct ofproto_dpif *ofproto = facet->ofproto;
4482 struct subfacet *subfacet;
4484 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4485 ovs_assert(!subfacet->dp_byte_count);
4486 ovs_assert(!subfacet->dp_packet_count);
4489 facet_push_stats(facet, false);
4490 if (facet->accounted_bytes < facet->byte_count) {
4491 facet_account(facet);
4492 facet->accounted_bytes = facet->byte_count;
4495 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4496 struct ofexpired expired;
4497 expired.flow = facet->flow;
4498 expired.packet_count = facet->packet_count;
4499 expired.byte_count = facet->byte_count;
4500 expired.used = facet->used;
4501 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4504 /* Reset counters to prevent double counting if 'facet' ever gets
4506 facet_reset_counters(facet);
4508 netflow_flow_clear(&facet->nf_flow);
4509 facet->tcp_flags = 0;
4512 /* Searches 'ofproto''s table of facets for one which would be responsible for
4513 * 'flow'. Returns it if found, otherwise a null pointer.
4515 * The returned facet might need revalidation; use facet_lookup_valid()
4516 * instead if that is important. */
4517 static struct facet *
4518 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4520 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4521 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4524 /* Searches 'ofproto''s table of facets for one capable that covers
4525 * 'flow'. Returns it if found, otherwise a null pointer.
4527 * The returned facet is guaranteed to be valid. */
4528 static struct facet *
4529 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4531 struct facet *facet;
4533 facet = facet_find(ofproto, flow);
4535 && ofproto->backer->need_revalidate
4536 && !facet_revalidate(facet)) {
4544 facet_check_consistency(struct facet *facet)
4546 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4548 struct xlate_out xout;
4549 struct xlate_in xin;
4551 struct rule_dpif *rule;
4554 /* Check the datapath actions for consistency. */
4555 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4556 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4557 xlate_actions(&xin, &xout);
4559 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4560 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4561 && facet->xout.slow == xout.slow
4562 && facet->fail_open == fail_open;
4563 if (!ok && !VLOG_DROP_WARN(&rl)) {
4564 struct ds s = DS_EMPTY_INITIALIZER;
4566 flow_format(&s, &facet->flow);
4567 ds_put_cstr(&s, ": inconsistency in facet");
4569 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4570 ds_put_cstr(&s, " (actions were: ");
4571 format_odp_actions(&s, facet->xout.odp_actions.data,
4572 facet->xout.odp_actions.size);
4573 ds_put_cstr(&s, ") (correct actions: ");
4574 format_odp_actions(&s, xout.odp_actions.data,
4575 xout.odp_actions.size);
4576 ds_put_char(&s, ')');
4579 if (facet->xout.slow != xout.slow) {
4580 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4583 if (facet->fail_open != fail_open) {
4584 ds_put_format(&s, " fail open incorrect. should be %s",
4585 fail_open ? "true" : "false");
4589 xlate_out_uninit(&xout);
4594 /* Re-searches the classifier for 'facet':
4596 * - If the rule found is different from 'facet''s current rule, moves
4597 * 'facet' to the new rule and recompiles its actions.
4599 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4600 * where it is and recompiles its actions anyway.
4602 * - If any of 'facet''s subfacets correspond to a new flow according to
4603 * xlate_receive(), 'facet' is removed.
4605 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4607 facet_revalidate(struct facet *facet)
4609 struct ofproto_dpif *ofproto = facet->ofproto;
4610 struct rule_dpif *new_rule;
4611 struct subfacet *subfacet;
4612 struct flow_wildcards wc;
4613 struct xlate_out xout;
4614 struct xlate_in xin;
4616 COVERAGE_INC(facet_revalidate);
4618 /* Check that child subfacets still correspond to this facet. Tunnel
4619 * configuration changes could cause a subfacet's OpenFlow in_port to
4621 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4622 struct ofproto_dpif *recv_ofproto;
4623 struct flow recv_flow;
4626 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4627 subfacet->key_len, &recv_flow, NULL,
4628 &recv_ofproto, NULL);
4630 || recv_ofproto != ofproto
4631 || facet != facet_find(ofproto, &recv_flow)) {
4632 facet_remove(facet);
4637 flow_wildcards_init_catchall(&wc);
4638 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4640 /* Calculate new datapath actions.
4642 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4643 * emit a NetFlow expiration and, if so, we need to have the old state
4644 * around to properly compose it. */
4645 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4646 xlate_actions(&xin, &xout);
4647 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4649 /* A facet's slow path reason should only change under dramatic
4650 * circumstances. Rather than try to update everything, it's simpler to
4651 * remove the facet and start over.
4653 * More importantly, if a facet's wildcards change, it will be relatively
4654 * difficult to figure out if its subfacets still belong to it, and if not
4655 * which facet they may belong to. Again, to avoid the complexity, we
4656 * simply give up instead. */
4657 if (facet->xout.slow != xout.slow
4658 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4659 facet_remove(facet);
4660 xlate_out_uninit(&xout);
4664 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4665 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4666 if (subfacet->path == SF_FAST_PATH) {
4667 struct dpif_flow_stats stats;
4669 subfacet_install(subfacet, &xout.odp_actions, &stats);
4670 subfacet_update_stats(subfacet, &stats);
4674 facet_flush_stats(facet);
4676 ofpbuf_clear(&facet->xout.odp_actions);
4677 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4678 xout.odp_actions.size);
4681 /* Update 'facet' now that we've taken care of all the old state. */
4682 facet->xout.slow = xout.slow;
4683 facet->xout.has_learn = xout.has_learn;
4684 facet->xout.has_normal = xout.has_normal;
4685 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4686 facet->xout.nf_output_iface = xout.nf_output_iface;
4687 facet->xout.mirrors = xout.mirrors;
4688 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4689 facet->used = MAX(facet->used, new_rule->up.created);
4690 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4692 xlate_out_uninit(&xout);
4697 facet_reset_counters(struct facet *facet)
4699 facet->packet_count = 0;
4700 facet->byte_count = 0;
4701 facet->prev_packet_count = 0;
4702 facet->prev_byte_count = 0;
4703 facet->accounted_bytes = 0;
4707 facet_push_stats(struct facet *facet, bool may_learn)
4709 struct dpif_flow_stats stats;
4711 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4712 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4713 ovs_assert(facet->used >= facet->prev_used);
4715 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4716 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4717 stats.used = facet->used;
4718 stats.tcp_flags = facet->tcp_flags;
4720 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4721 struct ofproto_dpif *ofproto = facet->ofproto;
4722 struct ofport_dpif *in_port;
4723 struct rule_dpif *rule;
4724 struct xlate_in xin;
4726 facet->prev_packet_count = facet->packet_count;
4727 facet->prev_byte_count = facet->byte_count;
4728 facet->prev_used = facet->used;
4730 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4731 if (in_port && in_port->is_tunnel) {
4732 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4735 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4736 rule_credit_stats(rule, &stats);
4737 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4739 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4740 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4741 stats.n_packets, stats.n_bytes);
4743 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4745 xin.resubmit_stats = &stats;
4746 xin.may_learn = may_learn;
4747 xlate_actions_for_side_effects(&xin);
4752 push_all_stats__(bool run_fast)
4754 static long long int rl = LLONG_MIN;
4755 struct ofproto_dpif *ofproto;
4757 if (time_msec() < rl) {
4761 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4762 struct cls_cursor cursor;
4763 struct facet *facet;
4765 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4766 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4767 facet_push_stats(facet, false);
4774 rl = time_msec() + 100;
4778 push_all_stats(void)
4780 push_all_stats__(true);
4784 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4786 ovs_mutex_lock(&rule->stats_mutex);
4787 rule->packet_count += stats->n_packets;
4788 rule->byte_count += stats->n_bytes;
4789 ofproto_rule_update_used(&rule->up, stats->used);
4790 ovs_mutex_unlock(&rule->stats_mutex);
4795 static struct subfacet *
4796 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4797 size_t key_len, uint32_t key_hash)
4799 struct subfacet *subfacet;
4801 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4802 &backer->subfacets) {
4803 if (subfacet->key_len == key_len
4804 && !memcmp(key, subfacet->key, key_len)) {
4812 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4813 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4814 * existing subfacet if there is one, otherwise creates and returns a
4816 static struct subfacet *
4817 subfacet_create(struct facet *facet, struct flow_miss *miss,
4820 struct dpif_backer *backer = miss->ofproto->backer;
4821 enum odp_key_fitness key_fitness = miss->key_fitness;
4822 const struct nlattr *key = miss->key;
4823 size_t key_len = miss->key_len;
4825 struct subfacet *subfacet;
4827 key_hash = odp_flow_key_hash(key, key_len);
4829 if (list_is_empty(&facet->subfacets)) {
4830 subfacet = &facet->one_subfacet;
4832 subfacet = subfacet_find(backer, key, key_len, key_hash);
4834 if (subfacet->facet == facet) {
4838 /* This shouldn't happen. */
4839 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4840 subfacet_destroy(subfacet);
4843 subfacet = xmalloc(sizeof *subfacet);
4846 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4847 list_push_back(&facet->subfacets, &subfacet->list_node);
4848 subfacet->facet = facet;
4849 subfacet->key_fitness = key_fitness;
4850 subfacet->key = xmemdup(key, key_len);
4851 subfacet->key_len = key_len;
4852 subfacet->used = now;
4853 subfacet->created = now;
4854 subfacet->dp_packet_count = 0;
4855 subfacet->dp_byte_count = 0;
4856 subfacet->path = SF_NOT_INSTALLED;
4857 subfacet->backer = backer;
4859 backer->subfacet_add_count++;
4863 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4864 * its facet within 'ofproto', and frees it. */
4866 subfacet_destroy__(struct subfacet *subfacet)
4868 struct facet *facet = subfacet->facet;
4869 struct ofproto_dpif *ofproto = facet->ofproto;
4871 /* Update ofproto stats before uninstall the subfacet. */
4872 ofproto->backer->subfacet_del_count++;
4874 subfacet_uninstall(subfacet);
4875 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4876 list_remove(&subfacet->list_node);
4877 free(subfacet->key);
4878 if (subfacet != &facet->one_subfacet) {
4883 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4884 * last remaining subfacet in its facet destroys the facet too. */
4886 subfacet_destroy(struct subfacet *subfacet)
4888 struct facet *facet = subfacet->facet;
4890 if (list_is_singleton(&facet->subfacets)) {
4891 /* facet_remove() needs at least one subfacet (it will remove it). */
4892 facet_remove(facet);
4894 subfacet_destroy__(subfacet);
4899 subfacet_destroy_batch(struct dpif_backer *backer,
4900 struct subfacet **subfacets, int n)
4902 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4903 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4904 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4907 for (i = 0; i < n; i++) {
4908 ops[i].type = DPIF_OP_FLOW_DEL;
4909 ops[i].u.flow_del.key = subfacets[i]->key;
4910 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4911 ops[i].u.flow_del.stats = &stats[i];
4915 dpif_operate(backer->dpif, opsp, n);
4916 for (i = 0; i < n; i++) {
4917 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4918 subfacets[i]->path = SF_NOT_INSTALLED;
4919 subfacet_destroy(subfacets[i]);
4924 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4925 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4926 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4927 * since 'subfacet' was last updated.
4929 * Returns 0 if successful, otherwise a positive errno value. */
4931 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4932 struct dpif_flow_stats *stats)
4934 struct facet *facet = subfacet->facet;
4935 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4936 const struct nlattr *actions = odp_actions->data;
4937 size_t actions_len = odp_actions->size;
4938 struct odputil_keybuf maskbuf;
4941 uint64_t slow_path_stub[128 / 8];
4942 enum dpif_flow_put_flags flags;
4945 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4948 flags |= DPIF_FP_ZERO_STATS;
4951 if (path == SF_SLOW_PATH) {
4952 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4953 slow_path_stub, sizeof slow_path_stub,
4954 &actions, &actions_len);
4957 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4958 if (enable_megaflows) {
4959 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4960 &facet->flow, UINT32_MAX);
4963 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4964 subfacet->key_len, mask.data, mask.size,
4965 actions, actions_len, stats);
4968 subfacet_reset_dp_stats(subfacet, stats);
4972 COVERAGE_INC(subfacet_install_fail);
4974 subfacet->path = path;
4979 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4981 subfacet_uninstall(struct subfacet *subfacet)
4983 if (subfacet->path != SF_NOT_INSTALLED) {
4984 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4985 struct dpif_flow_stats stats;
4988 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4989 subfacet->key_len, &stats);
4990 subfacet_reset_dp_stats(subfacet, &stats);
4992 subfacet_update_stats(subfacet, &stats);
4994 subfacet->path = SF_NOT_INSTALLED;
4996 ovs_assert(subfacet->dp_packet_count == 0);
4997 ovs_assert(subfacet->dp_byte_count == 0);
5001 /* Resets 'subfacet''s datapath statistics counters. This should be called
5002 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5003 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5004 * was reset in the datapath. 'stats' will be modified to include only
5005 * statistics new since 'subfacet' was last updated. */
5007 subfacet_reset_dp_stats(struct subfacet *subfacet,
5008 struct dpif_flow_stats *stats)
5011 && subfacet->dp_packet_count <= stats->n_packets
5012 && subfacet->dp_byte_count <= stats->n_bytes) {
5013 stats->n_packets -= subfacet->dp_packet_count;
5014 stats->n_bytes -= subfacet->dp_byte_count;
5017 subfacet->dp_packet_count = 0;
5018 subfacet->dp_byte_count = 0;
5021 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5023 * Because of the meaning of a subfacet's counters, it only makes sense to do
5024 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5025 * represents a packet that was sent by hand or if it represents statistics
5026 * that have been cleared out of the datapath. */
5028 subfacet_update_stats(struct subfacet *subfacet,
5029 const struct dpif_flow_stats *stats)
5031 if (stats->n_packets || stats->used > subfacet->used) {
5032 struct facet *facet = subfacet->facet;
5034 subfacet->used = MAX(subfacet->used, stats->used);
5035 facet->used = MAX(facet->used, stats->used);
5036 facet->packet_count += stats->n_packets;
5037 facet->byte_count += stats->n_bytes;
5038 facet->tcp_flags |= stats->tcp_flags;
5044 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5045 * the fields that were relevant as part of the lookup. */
5046 static struct rule_dpif *
5047 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5048 struct flow_wildcards *wc)
5050 struct rule_dpif *rule;
5052 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5057 return rule_dpif_miss_rule(ofproto, flow);
5061 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5062 const struct flow *flow, struct flow_wildcards *wc,
5065 struct cls_rule *cls_rule;
5066 struct classifier *cls;
5069 if (table_id >= N_TABLES) {
5074 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5075 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5078 cls = &ofproto->up.tables[table_id].cls;
5079 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5080 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5081 /* We must pretend that transport ports are unavailable. */
5082 struct flow ofpc_normal_flow = *flow;
5083 ofpc_normal_flow.tp_src = htons(0);
5084 ofpc_normal_flow.tp_dst = htons(0);
5085 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5086 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5087 cls_rule = &ofproto->drop_frags_rule->up.cr;
5089 flow_wildcards_init_exact(wc);
5092 cls_rule = classifier_lookup(cls, flow, wc);
5094 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5098 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5100 struct ofport_dpif *port;
5102 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5104 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5105 flow->in_port.ofp_port);
5106 return ofproto->miss_rule;
5109 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5110 return ofproto->no_packet_in_rule;
5112 return ofproto->miss_rule;
5116 complete_operation(struct rule_dpif *rule)
5118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5120 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5122 struct dpif_completion *c = xmalloc(sizeof *c);
5123 c->op = rule->up.pending;
5124 list_push_back(&ofproto->completions, &c->list_node);
5126 ofoperation_complete(rule->up.pending, 0);
5130 static struct rule *
5133 struct rule_dpif *rule = xmalloc(sizeof *rule);
5138 rule_dealloc(struct rule *rule_)
5140 struct rule_dpif *rule = rule_dpif_cast(rule_);
5145 rule_construct(struct rule *rule_)
5147 struct rule_dpif *rule = rule_dpif_cast(rule_);
5148 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5149 ovs_mutex_lock(&rule->stats_mutex);
5150 rule->packet_count = 0;
5151 rule->byte_count = 0;
5152 ovs_mutex_unlock(&rule->stats_mutex);
5153 complete_operation(rule);
5158 rule_destruct(struct rule *rule_)
5160 struct rule_dpif *rule = rule_dpif_cast(rule_);
5161 complete_operation(rule);
5162 ovs_mutex_destroy(&rule->stats_mutex);
5166 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5168 struct rule_dpif *rule = rule_dpif_cast(rule_);
5170 /* push_all_stats() can handle flow misses which, when using the learn
5171 * action, can cause rules to be added and deleted. This can corrupt our
5172 * caller's datastructures which assume that rule_get_stats() doesn't have
5173 * an impact on the flow table. To be safe, we disable miss handling. */
5174 push_all_stats__(false);
5176 /* Start from historical data for 'rule' itself that are no longer tracked
5177 * in facets. This counts, for example, facets that have expired. */
5178 ovs_mutex_lock(&rule->stats_mutex);
5179 *packets = rule->packet_count;
5180 *bytes = rule->byte_count;
5181 ovs_mutex_unlock(&rule->stats_mutex);
5185 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5186 struct ofpbuf *packet)
5188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5189 struct dpif_flow_stats stats;
5190 struct xlate_out xout;
5191 struct xlate_in xin;
5193 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5194 rule_credit_stats(rule, &stats);
5196 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5197 xin.resubmit_stats = &stats;
5198 xlate_actions(&xin, &xout);
5200 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5201 xout.odp_actions.size, packet);
5203 xlate_out_uninit(&xout);
5207 rule_execute(struct rule *rule, const struct flow *flow,
5208 struct ofpbuf *packet)
5210 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5211 ofpbuf_delete(packet);
5216 rule_modify_actions(struct rule *rule_)
5218 struct rule_dpif *rule = rule_dpif_cast(rule_);
5220 complete_operation(rule);
5223 /* Sends 'packet' out 'ofport'.
5224 * May modify 'packet'.
5225 * Returns 0 if successful, otherwise a positive errno value. */
5227 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5229 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5230 uint64_t odp_actions_stub[1024 / 8];
5231 struct ofpbuf key, odp_actions;
5232 struct dpif_flow_stats stats;
5233 struct odputil_keybuf keybuf;
5234 struct ofpact_output output;
5235 struct xlate_out xout;
5236 struct xlate_in xin;
5238 union flow_in_port in_port_;
5241 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5242 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5244 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5245 in_port_.ofp_port = OFPP_NONE;
5246 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5247 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5249 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5251 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5252 output.port = ofport->up.ofp_port;
5255 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5256 xin.ofpacts_len = sizeof output;
5257 xin.ofpacts = &output.ofpact;
5258 xin.resubmit_stats = &stats;
5259 xlate_actions(&xin, &xout);
5261 error = dpif_execute(ofproto->backer->dpif,
5263 xout.odp_actions.data, xout.odp_actions.size,
5265 xlate_out_uninit(&xout);
5268 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5269 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5270 ovs_strerror(error));
5273 ofproto->stats.tx_packets++;
5274 ofproto->stats.tx_bytes += packet->size;
5278 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5279 * The action will state 'slow' as the reason that the action is in the slow
5280 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5281 * dump-flows" output to see why a flow is in the slow path.)
5283 * The 'stub_size' bytes in 'stub' will be used to store the action.
5284 * 'stub_size' must be large enough for the action.
5286 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5289 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5290 enum slow_path_reason slow,
5291 uint64_t *stub, size_t stub_size,
5292 const struct nlattr **actionsp, size_t *actions_lenp)
5294 union user_action_cookie cookie;
5297 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5298 cookie.slow_path.unused = 0;
5299 cookie.slow_path.reason = slow;
5301 ofpbuf_use_stack(&buf, stub, stub_size);
5302 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5303 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5305 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5307 odp_port_t odp_port;
5310 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5311 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5312 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5314 *actionsp = buf.data;
5315 *actions_lenp = buf.size;
5319 set_frag_handling(struct ofproto *ofproto_,
5320 enum ofp_config_flags frag_handling)
5322 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5323 if (frag_handling != OFPC_FRAG_REASM) {
5324 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5332 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5333 const struct flow *flow,
5334 const struct ofpact *ofpacts, size_t ofpacts_len)
5336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5337 struct odputil_keybuf keybuf;
5338 struct dpif_flow_stats stats;
5339 struct xlate_out xout;
5340 struct xlate_in xin;
5344 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5345 odp_flow_key_from_flow(&key, flow,
5346 ofp_port_to_odp_port(ofproto,
5347 flow->in_port.ofp_port));
5349 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5351 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5352 xin.resubmit_stats = &stats;
5353 xin.ofpacts_len = ofpacts_len;
5354 xin.ofpacts = ofpacts;
5356 xlate_actions(&xin, &xout);
5357 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5358 xout.odp_actions.data, xout.odp_actions.size, packet);
5359 xlate_out_uninit(&xout);
5367 set_netflow(struct ofproto *ofproto_,
5368 const struct netflow_options *netflow_options)
5370 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5372 if (netflow_options) {
5373 if (!ofproto->netflow) {
5374 ofproto->netflow = netflow_create();
5375 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5377 return netflow_set_options(ofproto->netflow, netflow_options);
5378 } else if (ofproto->netflow) {
5379 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5380 netflow_destroy(ofproto->netflow);
5381 ofproto->netflow = NULL;
5388 get_netflow_ids(const struct ofproto *ofproto_,
5389 uint8_t *engine_type, uint8_t *engine_id)
5391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5393 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5397 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5399 if (!facet_is_controller_flow(facet) &&
5400 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5401 struct subfacet *subfacet;
5402 struct ofexpired expired;
5404 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5405 if (subfacet->path == SF_FAST_PATH) {
5406 struct dpif_flow_stats stats;
5408 subfacet_install(subfacet, &facet->xout.odp_actions,
5410 subfacet_update_stats(subfacet, &stats);
5414 expired.flow = facet->flow;
5415 expired.packet_count = facet->packet_count;
5416 expired.byte_count = facet->byte_count;
5417 expired.used = facet->used;
5418 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5423 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5425 struct cls_cursor cursor;
5426 struct facet *facet;
5428 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5429 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5430 send_active_timeout(ofproto, facet);
5434 static struct ofproto_dpif *
5435 ofproto_dpif_lookup(const char *name)
5437 struct ofproto_dpif *ofproto;
5439 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5440 hash_string(name, 0), &all_ofproto_dpifs) {
5441 if (!strcmp(ofproto->up.name, name)) {
5449 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5450 const char *argv[], void *aux OVS_UNUSED)
5452 struct ofproto_dpif *ofproto;
5455 ofproto = ofproto_dpif_lookup(argv[1]);
5457 unixctl_command_reply_error(conn, "no such bridge");
5460 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5461 mac_learning_flush(ofproto->ml);
5462 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5464 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5465 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5466 mac_learning_flush(ofproto->ml);
5467 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5471 unixctl_command_reply(conn, "table successfully flushed");
5474 static struct ofport_dpif *
5475 ofbundle_get_a_port(const struct ofbundle *bundle)
5477 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5482 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5483 const char *argv[], void *aux OVS_UNUSED)
5485 struct ds ds = DS_EMPTY_INITIALIZER;
5486 const struct ofproto_dpif *ofproto;
5487 const struct mac_entry *e;
5489 ofproto = ofproto_dpif_lookup(argv[1]);
5491 unixctl_command_reply_error(conn, "no such bridge");
5495 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5496 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5497 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5498 struct ofbundle *bundle = e->port.p;
5499 char name[OFP_MAX_PORT_NAME_LEN];
5501 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5503 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5504 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5505 mac_entry_age(ofproto->ml, e));
5507 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5508 unixctl_command_reply(conn, ds_cstr(&ds));
5513 struct xlate_out xout;
5514 struct xlate_in xin;
5520 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5522 ds_put_char_multiple(result, '\t', level);
5524 ds_put_cstr(result, "No match\n");
5528 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5529 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5530 cls_rule_format(&rule->up.cr, result);
5531 ds_put_char(result, '\n');
5533 ds_put_char_multiple(result, '\t', level);
5534 ds_put_cstr(result, "OpenFlow ");
5535 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5536 ds_put_char(result, '\n');
5540 trace_format_flow(struct ds *result, int level, const char *title,
5541 struct trace_ctx *trace)
5543 ds_put_char_multiple(result, '\t', level);
5544 ds_put_format(result, "%s: ", title);
5545 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5546 ds_put_cstr(result, "unchanged");
5548 flow_format(result, &trace->xin.flow);
5549 trace->flow = trace->xin.flow;
5551 ds_put_char(result, '\n');
5555 trace_format_regs(struct ds *result, int level, const char *title,
5556 struct trace_ctx *trace)
5560 ds_put_char_multiple(result, '\t', level);
5561 ds_put_format(result, "%s:", title);
5562 for (i = 0; i < FLOW_N_REGS; i++) {
5563 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5565 ds_put_char(result, '\n');
5569 trace_format_odp(struct ds *result, int level, const char *title,
5570 struct trace_ctx *trace)
5572 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5574 ds_put_char_multiple(result, '\t', level);
5575 ds_put_format(result, "%s: ", title);
5576 format_odp_actions(result, odp_actions->data, odp_actions->size);
5577 ds_put_char(result, '\n');
5581 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5583 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5584 struct ds *result = trace->result;
5586 ds_put_char(result, '\n');
5587 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5588 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5589 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5590 trace_format_rule(result, recurse + 1, rule);
5594 trace_report(struct xlate_in *xin, const char *s, int recurse)
5596 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5597 struct ds *result = trace->result;
5599 ds_put_char_multiple(result, '\t', recurse);
5600 ds_put_cstr(result, s);
5601 ds_put_char(result, '\n');
5605 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5606 void *aux OVS_UNUSED)
5608 const struct dpif_backer *backer;
5609 struct ofproto_dpif *ofproto;
5610 struct ofpbuf odp_key, odp_mask;
5611 struct ofpbuf *packet;
5619 ofpbuf_init(&odp_key, 0);
5620 ofpbuf_init(&odp_mask, 0);
5622 /* Handle "-generate" or a hex string as the last argument. */
5623 if (!strcmp(argv[argc - 1], "-generate")) {
5624 packet = ofpbuf_new(0);
5627 const char *error = eth_from_hex(argv[argc - 1], &packet);
5630 } else if (argc == 4) {
5631 /* The 3-argument form must end in "-generate' or a hex string. */
5632 unixctl_command_reply_error(conn, error);
5637 /* Parse the flow and determine whether a datapath or
5638 * bridge is specified. If function odp_flow_key_from_string()
5639 * returns 0, the flow is a odp_flow. If function
5640 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5641 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5642 /* If the odp_flow is the second argument,
5643 * the datapath name is the first argument. */
5645 const char *dp_type;
5646 if (!strncmp(argv[1], "ovs-", 4)) {
5647 dp_type = argv[1] + 4;
5651 backer = shash_find_data(&all_dpif_backers, dp_type);
5653 unixctl_command_reply_error(conn, "Cannot find datapath "
5658 /* No datapath name specified, so there should be only one
5660 struct shash_node *node;
5661 if (shash_count(&all_dpif_backers) != 1) {
5662 unixctl_command_reply_error(conn, "Must specify datapath "
5663 "name, there is more than one type of datapath");
5666 node = shash_first(&all_dpif_backers);
5667 backer = node->data;
5670 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5671 NULL, &ofproto, NULL)) {
5672 unixctl_command_reply_error(conn, "Invalid datapath flow");
5675 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5676 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5678 unixctl_command_reply_error(conn, "Must specify bridge name");
5682 ofproto = ofproto_dpif_lookup(argv[1]);
5684 unixctl_command_reply_error(conn, "Unknown bridge name");
5688 unixctl_command_reply_error(conn, "Bad flow syntax");
5692 /* Generate a packet, if requested. */
5694 if (!packet->size) {
5695 flow_compose(packet, &flow);
5697 union flow_in_port in_port_;
5699 in_port_ = flow.in_port;
5700 ds_put_cstr(&result, "Packet: ");
5701 s = ofp_packet_to_string(packet->data, packet->size);
5702 ds_put_cstr(&result, s);
5705 /* Use the metadata from the flow and the packet argument
5706 * to reconstruct the flow. */
5707 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5712 ofproto_trace(ofproto, &flow, packet, &result);
5713 unixctl_command_reply(conn, ds_cstr(&result));
5716 ds_destroy(&result);
5717 ofpbuf_delete(packet);
5718 ofpbuf_uninit(&odp_key);
5719 ofpbuf_uninit(&odp_mask);
5723 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5724 const struct ofpbuf *packet, struct ds *ds)
5726 struct rule_dpif *rule;
5728 ds_put_cstr(ds, "Flow: ");
5729 flow_format(ds, flow);
5730 ds_put_char(ds, '\n');
5732 rule = rule_dpif_lookup(ofproto, flow, NULL);
5734 trace_format_rule(ds, 0, rule);
5735 if (rule == ofproto->miss_rule) {
5736 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5737 } else if (rule == ofproto->no_packet_in_rule) {
5738 ds_put_cstr(ds, "\nNo match, packets dropped because "
5739 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5740 } else if (rule == ofproto->drop_frags_rule) {
5741 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5742 "and the fragment handling mode is \"drop\".\n");
5746 uint64_t odp_actions_stub[1024 / 8];
5747 struct ofpbuf odp_actions;
5748 struct trace_ctx trace;
5752 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5755 ofpbuf_use_stub(&odp_actions,
5756 odp_actions_stub, sizeof odp_actions_stub);
5757 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5758 trace.xin.resubmit_hook = trace_resubmit;
5759 trace.xin.report_hook = trace_report;
5761 xlate_actions(&trace.xin, &trace.xout);
5763 ds_put_char(ds, '\n');
5764 trace_format_flow(ds, 0, "Final flow", &trace);
5766 match_init(&match, flow, &trace.xout.wc);
5767 ds_put_cstr(ds, "Relevant fields: ");
5768 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5769 ds_put_char(ds, '\n');
5771 ds_put_cstr(ds, "Datapath actions: ");
5772 format_odp_actions(ds, trace.xout.odp_actions.data,
5773 trace.xout.odp_actions.size);
5775 if (trace.xout.slow) {
5776 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5777 "slow path because it:");
5778 switch (trace.xout.slow) {
5780 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5783 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5786 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5789 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5791 case SLOW_CONTROLLER:
5792 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5793 "to the OpenFlow controller.");
5800 xlate_out_uninit(&trace.xout);
5805 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5806 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5809 unixctl_command_reply(conn, NULL);
5813 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5814 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5817 unixctl_command_reply(conn, NULL);
5820 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5821 * 'reply' describing the results. */
5823 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5825 struct cls_cursor cursor;
5826 struct facet *facet;
5830 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5831 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5832 if (!facet_check_consistency(facet)) {
5837 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5841 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5842 ofproto->up.name, errors);
5844 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5849 ofproto_dpif_self_check(struct unixctl_conn *conn,
5850 int argc, const char *argv[], void *aux OVS_UNUSED)
5852 struct ds reply = DS_EMPTY_INITIALIZER;
5853 struct ofproto_dpif *ofproto;
5856 ofproto = ofproto_dpif_lookup(argv[1]);
5858 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5859 "ofproto/list for help)");
5862 ofproto_dpif_self_check__(ofproto, &reply);
5864 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5865 ofproto_dpif_self_check__(ofproto, &reply);
5869 unixctl_command_reply(conn, ds_cstr(&reply));
5873 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5874 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5875 * to destroy 'ofproto_shash' and free the returned value. */
5876 static const struct shash_node **
5877 get_ofprotos(struct shash *ofproto_shash)
5879 const struct ofproto_dpif *ofproto;
5881 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5882 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5883 shash_add_nocopy(ofproto_shash, name, ofproto);
5886 return shash_sort(ofproto_shash);
5890 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5891 const char *argv[] OVS_UNUSED,
5892 void *aux OVS_UNUSED)
5894 struct ds ds = DS_EMPTY_INITIALIZER;
5895 struct shash ofproto_shash;
5896 const struct shash_node **sorted_ofprotos;
5899 shash_init(&ofproto_shash);
5900 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5901 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5902 const struct shash_node *node = sorted_ofprotos[i];
5903 ds_put_format(&ds, "%s\n", node->name);
5906 shash_destroy(&ofproto_shash);
5907 free(sorted_ofprotos);
5909 unixctl_command_reply(conn, ds_cstr(&ds));
5914 show_dp_rates(struct ds *ds, const char *heading,
5915 const struct avg_subfacet_rates *rates)
5917 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5918 heading, rates->add_rate, rates->del_rate);
5922 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5924 const struct shash_node **ofprotos;
5925 struct ofproto_dpif *ofproto;
5926 struct shash ofproto_shash;
5927 uint64_t n_hit, n_missed;
5928 long long int minutes;
5931 n_hit = n_missed = 0;
5932 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5933 if (ofproto->backer == backer) {
5934 n_missed += ofproto->n_missed;
5935 n_hit += ofproto->n_hit;
5939 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5940 dpif_name(backer->dpif), n_hit, n_missed);
5941 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5942 " life span: %lldms\n", hmap_count(&backer->subfacets),
5943 backer->avg_n_subfacet, backer->max_n_subfacet,
5944 backer->avg_subfacet_life);
5946 minutes = (time_msec() - backer->created) / (1000 * 60);
5947 if (minutes >= 60) {
5948 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5950 if (minutes >= 60 * 24) {
5951 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
5953 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
5955 shash_init(&ofproto_shash);
5956 ofprotos = get_ofprotos(&ofproto_shash);
5957 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5958 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5959 const struct shash_node **ports;
5962 if (ofproto->backer != backer) {
5966 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5967 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5969 ports = shash_sort(&ofproto->up.port_by_name);
5970 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5971 const struct shash_node *node = ports[j];
5972 struct ofport *ofport = node->data;
5974 odp_port_t odp_port;
5976 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5979 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5980 if (odp_port != ODPP_NONE) {
5981 ds_put_format(ds, "%"PRIu32":", odp_port);
5983 ds_put_cstr(ds, "none:");
5986 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5989 if (!netdev_get_config(ofport->netdev, &config)) {
5990 const struct smap_node **nodes;
5993 nodes = smap_sort(&config);
5994 for (i = 0; i < smap_count(&config); i++) {
5995 const struct smap_node *node = nodes[i];
5996 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5997 node->key, node->value);
6001 smap_destroy(&config);
6003 ds_put_char(ds, ')');
6004 ds_put_char(ds, '\n');
6008 shash_destroy(&ofproto_shash);
6013 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6014 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6016 struct ds ds = DS_EMPTY_INITIALIZER;
6017 const struct shash_node **backers;
6020 backers = shash_sort(&all_dpif_backers);
6021 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6022 dpif_show_backer(backers[i]->data, &ds);
6026 unixctl_command_reply(conn, ds_cstr(&ds));
6030 /* Dump the megaflow (facet) cache. This is useful to check the
6031 * correctness of flow wildcarding, since the same mechanism is used for
6032 * both xlate caching and kernel wildcarding.
6034 * It's important to note that in the output the flow description uses
6035 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6037 * This command is only needed for advanced debugging, so it's not
6038 * documented in the man page. */
6040 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6041 int argc OVS_UNUSED, const char *argv[],
6042 void *aux OVS_UNUSED)
6044 struct ds ds = DS_EMPTY_INITIALIZER;
6045 const struct ofproto_dpif *ofproto;
6046 long long int now = time_msec();
6047 struct cls_cursor cursor;
6048 struct facet *facet;
6050 ofproto = ofproto_dpif_lookup(argv[1]);
6052 unixctl_command_reply_error(conn, "no such bridge");
6056 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6057 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6058 cls_rule_format(&facet->cr, &ds);
6059 ds_put_cstr(&ds, ", ");
6060 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6061 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6062 ds_put_cstr(&ds, "Datapath actions: ");
6063 if (facet->xout.slow) {
6064 uint64_t slow_path_stub[128 / 8];
6065 const struct nlattr *actions;
6068 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6069 slow_path_stub, sizeof slow_path_stub,
6070 &actions, &actions_len);
6071 format_odp_actions(&ds, actions, actions_len);
6073 format_odp_actions(&ds, facet->xout.odp_actions.data,
6074 facet->xout.odp_actions.size);
6076 ds_put_cstr(&ds, "\n");
6079 ds_chomp(&ds, '\n');
6080 unixctl_command_reply(conn, ds_cstr(&ds));
6084 /* Disable using the megaflows.
6086 * This command is only needed for advanced debugging, so it's not
6087 * documented in the man page. */
6089 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6090 int argc OVS_UNUSED,
6091 const char *argv[] OVS_UNUSED,
6092 void *aux OVS_UNUSED)
6094 struct ofproto_dpif *ofproto;
6096 enable_megaflows = false;
6098 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6099 flush(&ofproto->up);
6102 unixctl_command_reply(conn, "megaflows disabled");
6105 /* Re-enable using megaflows.
6107 * This command is only needed for advanced debugging, so it's not
6108 * documented in the man page. */
6110 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6111 int argc OVS_UNUSED,
6112 const char *argv[] OVS_UNUSED,
6113 void *aux OVS_UNUSED)
6115 struct ofproto_dpif *ofproto;
6117 enable_megaflows = true;
6119 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6120 flush(&ofproto->up);
6123 unixctl_command_reply(conn, "megaflows enabled");
6127 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6128 int argc OVS_UNUSED, const char *argv[],
6129 void *aux OVS_UNUSED)
6131 struct ds ds = DS_EMPTY_INITIALIZER;
6132 const struct ofproto_dpif *ofproto;
6133 struct subfacet *subfacet;
6135 ofproto = ofproto_dpif_lookup(argv[1]);
6137 unixctl_command_reply_error(conn, "no such bridge");
6141 update_stats(ofproto->backer);
6143 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6144 struct facet *facet = subfacet->facet;
6145 struct odputil_keybuf maskbuf;
6148 if (facet->ofproto != ofproto) {
6152 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6153 if (enable_megaflows) {
6154 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6155 &facet->flow, UINT32_MAX);
6158 odp_flow_format(subfacet->key, subfacet->key_len,
6159 mask.data, mask.size, &ds, false);
6161 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6162 subfacet->dp_packet_count, subfacet->dp_byte_count);
6163 if (subfacet->used) {
6164 ds_put_format(&ds, "%.3fs",
6165 (time_msec() - subfacet->used) / 1000.0);
6167 ds_put_format(&ds, "never");
6169 if (subfacet->facet->tcp_flags) {
6170 ds_put_cstr(&ds, ", flags:");
6171 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6174 ds_put_cstr(&ds, ", actions:");
6175 if (facet->xout.slow) {
6176 uint64_t slow_path_stub[128 / 8];
6177 const struct nlattr *actions;
6180 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6181 slow_path_stub, sizeof slow_path_stub,
6182 &actions, &actions_len);
6183 format_odp_actions(&ds, actions, actions_len);
6185 format_odp_actions(&ds, facet->xout.odp_actions.data,
6186 facet->xout.odp_actions.size);
6188 ds_put_char(&ds, '\n');
6191 unixctl_command_reply(conn, ds_cstr(&ds));
6196 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6197 int argc OVS_UNUSED, const char *argv[],
6198 void *aux OVS_UNUSED)
6200 struct ds ds = DS_EMPTY_INITIALIZER;
6201 struct ofproto_dpif *ofproto;
6203 ofproto = ofproto_dpif_lookup(argv[1]);
6205 unixctl_command_reply_error(conn, "no such bridge");
6209 flush(&ofproto->up);
6211 unixctl_command_reply(conn, ds_cstr(&ds));
6216 ofproto_dpif_unixctl_init(void)
6218 static bool registered;
6224 unixctl_command_register(
6226 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6227 1, 3, ofproto_unixctl_trace, NULL);
6228 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6229 ofproto_unixctl_fdb_flush, NULL);
6230 unixctl_command_register("fdb/show", "bridge", 1, 1,
6231 ofproto_unixctl_fdb_show, NULL);
6232 unixctl_command_register("ofproto/clog", "", 0, 0,
6233 ofproto_dpif_clog, NULL);
6234 unixctl_command_register("ofproto/unclog", "", 0, 0,
6235 ofproto_dpif_unclog, NULL);
6236 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6237 ofproto_dpif_self_check, NULL);
6238 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6239 ofproto_unixctl_dpif_dump_dps, NULL);
6240 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6242 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6243 ofproto_unixctl_dpif_dump_flows, NULL);
6244 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6245 ofproto_unixctl_dpif_del_flows, NULL);
6246 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6247 ofproto_unixctl_dpif_dump_megaflows, NULL);
6248 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6249 ofproto_unixctl_dpif_disable_megaflows, NULL);
6250 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6251 ofproto_unixctl_dpif_enable_megaflows, NULL);
6254 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6256 * This is deprecated. It is only for compatibility with broken device drivers
6257 * in old versions of Linux that do not properly support VLANs when VLAN
6258 * devices are not used. When broken device drivers are no longer in
6259 * widespread use, we will delete these interfaces. */
6262 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6265 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6267 if (realdev_ofp_port == ofport->realdev_ofp_port
6268 && vid == ofport->vlandev_vid) {
6272 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6274 if (ofport->realdev_ofp_port) {
6277 if (realdev_ofp_port && ofport->bundle) {
6278 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6279 * themselves be part of a bundle. */
6280 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6283 ofport->realdev_ofp_port = realdev_ofp_port;
6284 ofport->vlandev_vid = vid;
6286 if (realdev_ofp_port) {
6287 vsp_add(ofport, realdev_ofp_port, vid);
6294 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6296 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6300 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6301 OVS_EXCLUDED(ofproto->vsp_mutex)
6305 ovs_mutex_lock(&ofproto->vsp_mutex);
6306 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6307 ovs_mutex_unlock(&ofproto->vsp_mutex);
6312 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6313 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6314 OVS_REQUIRES(ofproto->vsp_mutex)
6316 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6317 int vid = vlan_tci_to_vid(vlan_tci);
6318 const struct vlan_splinter *vsp;
6320 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6321 hash_realdev_vid(realdev_ofp_port, vid),
6322 &ofproto->realdev_vid_map) {
6323 if (vsp->realdev_ofp_port == realdev_ofp_port
6324 && vsp->vid == vid) {
6325 return vsp->vlandev_ofp_port;
6329 return realdev_ofp_port;
6332 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6333 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6334 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6335 * 'vlan_tci' 9, it would return the port number of eth0.9.
6337 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6338 * function just returns its 'realdev_ofp_port' argument. */
6340 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6341 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6342 OVS_EXCLUDED(ofproto->vsp_mutex)
6346 ovs_mutex_lock(&ofproto->vsp_mutex);
6347 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6348 ovs_mutex_unlock(&ofproto->vsp_mutex);
6352 static struct vlan_splinter *
6353 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6355 struct vlan_splinter *vsp;
6357 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6358 hash_ofp_port(vlandev_ofp_port),
6359 &ofproto->vlandev_map) {
6360 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6368 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6369 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6370 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6371 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6372 * eth0 and store 9 in '*vid'.
6374 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6375 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6378 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6379 ofp_port_t vlandev_ofp_port, int *vid)
6380 OVS_REQ_WRLOCK(ofproto->vsp_mutex)
6382 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6383 const struct vlan_splinter *vsp;
6385 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6390 return vsp->realdev_ofp_port;
6396 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6397 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6398 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6399 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6400 * always the case unless VLAN splinters are enabled), returns false without
6401 * making any changes. */
6403 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6404 OVS_EXCLUDED(ofproto->vsp_mutex)
6409 ovs_mutex_lock(&ofproto->vsp_mutex);
6410 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6411 ovs_mutex_unlock(&ofproto->vsp_mutex);
6416 /* Cause the flow to be processed as if it came in on the real device with
6417 * the VLAN device's VLAN ID. */
6418 flow->in_port.ofp_port = realdev;
6419 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6424 vsp_remove(struct ofport_dpif *port)
6426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6427 struct vlan_splinter *vsp;
6429 ovs_mutex_lock(&ofproto->vsp_mutex);
6430 vsp = vlandev_find(ofproto, port->up.ofp_port);
6432 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6433 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6436 port->realdev_ofp_port = 0;
6438 VLOG_ERR("missing vlan device record");
6440 ovs_mutex_unlock(&ofproto->vsp_mutex);
6444 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6448 ovs_mutex_lock(&ofproto->vsp_mutex);
6449 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6450 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6451 == realdev_ofp_port)) {
6452 struct vlan_splinter *vsp;
6454 vsp = xmalloc(sizeof *vsp);
6455 vsp->realdev_ofp_port = realdev_ofp_port;
6456 vsp->vlandev_ofp_port = port->up.ofp_port;
6459 port->realdev_ofp_port = realdev_ofp_port;
6461 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6462 hash_ofp_port(port->up.ofp_port));
6463 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6464 hash_realdev_vid(realdev_ofp_port, vid));
6466 VLOG_ERR("duplicate vlan device record");
6468 ovs_mutex_unlock(&ofproto->vsp_mutex);
6472 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6474 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6475 return ofport ? ofport->odp_port : ODPP_NONE;
6478 struct ofport_dpif *
6479 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6481 struct ofport_dpif *port;
6483 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6484 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6485 &backer->odp_to_ofport_map) {
6486 if (port->odp_port == odp_port) {
6487 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6492 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6497 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6499 struct ofport_dpif *port;
6501 port = odp_port_to_ofport(ofproto->backer, odp_port);
6502 if (port && &ofproto->up == port->up.ofproto) {
6503 return port->up.ofp_port;
6509 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6510 * most heavily weighted element. 'base' designates the rate of decay: after
6511 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6514 exp_mavg(double *avg, int base, double new)
6516 *avg = (*avg * (base - 1) + new) / base;
6520 update_moving_averages(struct dpif_backer *backer)
6522 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6523 long long int minutes = (time_msec() - backer->created) / min_ms;
6526 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6528 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6531 backer->lifetime.add_rate = 0.0;
6532 backer->lifetime.del_rate = 0.0;
6535 /* Update hourly averages on the minute boundaries. */
6536 if (time_msec() - backer->last_minute >= min_ms) {
6537 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6538 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6540 /* Update daily averages on the hour boundaries. */
6541 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6542 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6543 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6546 backer->total_subfacet_add_count += backer->subfacet_add_count;
6547 backer->total_subfacet_del_count += backer->subfacet_del_count;
6548 backer->subfacet_add_count = 0;
6549 backer->subfacet_del_count = 0;
6550 backer->last_minute += min_ms;
6554 const struct ofproto_class ofproto_dpif_class = {
6589 port_is_lacp_current,
6590 NULL, /* rule_choose_table */
6597 rule_modify_actions,
6611 get_stp_port_status,
6618 is_mirror_output_bundle,
6619 forward_bpdu_changed,
6620 set_mac_table_config,
6622 NULL, /* meter_get_features */
6623 NULL, /* meter_set */
6624 NULL, /* meter_get */
6625 NULL, /* meter_del */