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 bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
331 static void vsp_remove(struct ofport_dpif *);
332 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
334 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
337 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
340 static struct ofport_dpif *
341 ofport_dpif_cast(const struct ofport *ofport)
343 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
346 static void port_run(struct ofport_dpif *);
347 static void port_run_fast(struct ofport_dpif *);
348 static void port_wait(struct ofport_dpif *);
349 static int set_bfd(struct ofport *, const struct smap *);
350 static int set_cfm(struct ofport *, const struct cfm_settings *);
351 static void ofport_update_peer(struct ofport_dpif *);
352 static void run_fast_rl(void);
353 static int run_fast(struct ofproto *);
355 struct dpif_completion {
356 struct list list_node;
357 struct ofoperation *op;
360 /* Reasons that we might need to revalidate every facet, and corresponding
363 * A value of 0 means that there is no need to revalidate.
365 * It would be nice to have some cleaner way to integrate with coverage
366 * counters, but with only a few reasons I guess this is good enough for
368 enum revalidate_reason {
369 REV_RECONFIGURE = 1, /* Switch configuration changed. */
370 REV_STP, /* Spanning tree protocol port status change. */
371 REV_BOND, /* Bonding changed. */
372 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
373 REV_FLOW_TABLE, /* Flow table changed. */
374 REV_MAC_LEARNING, /* Mac learning changed. */
375 REV_INCONSISTENCY /* Facet self-check failed. */
377 COVERAGE_DEFINE(rev_reconfigure);
378 COVERAGE_DEFINE(rev_stp);
379 COVERAGE_DEFINE(rev_bond);
380 COVERAGE_DEFINE(rev_port_toggled);
381 COVERAGE_DEFINE(rev_flow_table);
382 COVERAGE_DEFINE(rev_mac_learning);
383 COVERAGE_DEFINE(rev_inconsistency);
385 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
386 * These are datapath flows which have no associated ofproto, if they did we
387 * would use facets. */
389 struct hmap_node hmap_node;
394 struct avg_subfacet_rates {
395 double add_rate; /* Moving average of new flows created per minute. */
396 double del_rate; /* Moving average of flows deleted per minute. */
399 /* All datapaths of a given type share a single dpif backer instance. */
404 struct timer next_expiration;
405 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
407 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
409 /* Facet revalidation flags applying to facets which use this backer. */
410 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
412 struct hmap drop_keys; /* Set of dropped odp keys. */
413 bool recv_set_enable; /* Enables or disables receiving packets. */
415 struct hmap subfacets;
416 struct governor *governor;
418 /* Subfacet statistics.
420 * These keep track of the total number of subfacets added and deleted and
421 * flow life span. They are useful for computing the flow rates stats
422 * exposed via "ovs-appctl dpif/show". The goal is to learn about
423 * traffic patterns in ways that we can use later to improve Open vSwitch
424 * performance in new situations. */
425 long long int created; /* Time when it is created. */
426 unsigned max_n_subfacet; /* Maximum number of flows */
427 unsigned avg_n_subfacet; /* Average number of flows. */
428 long long int avg_subfacet_life; /* Average life span of subfacets. */
430 /* The average number of subfacets... */
431 struct avg_subfacet_rates hourly; /* ...over the last hour. */
432 struct avg_subfacet_rates daily; /* ...over the last day. */
433 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
434 long long int last_minute; /* Last time 'hourly' was updated. */
436 /* Number of subfacets added or deleted since 'last_minute'. */
437 unsigned subfacet_add_count;
438 unsigned subfacet_del_count;
440 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
441 unsigned long long int total_subfacet_add_count;
442 unsigned long long int total_subfacet_del_count;
445 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
446 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
448 static void drop_key_clear(struct dpif_backer *);
449 static struct ofport_dpif *
450 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
451 static void update_moving_averages(struct dpif_backer *backer);
453 struct ofproto_dpif {
454 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
456 struct dpif_backer *backer;
458 /* Special OpenFlow rules. */
459 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
460 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
461 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
464 struct netflow *netflow;
465 struct dpif_sflow *sflow;
466 struct dpif_ipfix *ipfix;
467 struct hmap bundles; /* Contains "struct ofbundle"s. */
468 struct mac_learning *ml;
469 bool has_bonded_bundles;
470 struct mbridge *mbridge;
473 struct classifier facets; /* Contains 'struct facet's. */
474 long long int consistency_rl;
476 /* Support for debugging async flow mods. */
477 struct list completions;
479 struct netdev_stats stats; /* To account packets generated and consumed in
484 long long int stp_last_tick;
486 /* VLAN splinters. */
487 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
488 struct hmap vlandev_map; /* 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, ofproto->ml,
786 ofproto->stp, ofproto->mbridge,
787 ofproto->sflow, ofproto->ipfix,
788 ofproto->up.frag_handling,
789 ofproto->up.forward_bpdu,
790 connmgr_has_in_band(ofproto->up.connmgr),
791 ofproto->netflow != NULL);
793 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
794 xlate_bundle_set(ofproto, bundle, bundle->name,
795 bundle->vlan_mode, bundle->vlan,
796 bundle->trunks, bundle->use_priority_tags,
797 bundle->bond, bundle->lacp,
801 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
802 int stp_port = ofport->stp_port
803 ? stp_port_no(ofport->stp_port)
805 xlate_ofport_set(ofproto, ofport->bundle, ofport,
806 ofport->up.ofp_port, ofport->odp_port,
807 ofport->up.netdev, ofport->cfm,
808 ofport->bfd, ofport->peer, stp_port,
809 ofport->qdscp, ofport->n_qdscp,
810 ofport->up.pp.config, ofport->is_tunnel,
814 cls_cursor_init(&cursor, &ofproto->facets, NULL);
815 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
816 facet_revalidate(facet);
822 if (!backer->recv_set_enable) {
823 /* Wake up before a max of 1000ms. */
824 timer_set_duration(&backer->next_expiration, 1000);
825 } else if (timer_expired(&backer->next_expiration)) {
826 int delay = expire(backer);
827 timer_set_duration(&backer->next_expiration, delay);
830 process_dpif_port_changes(backer);
832 if (backer->governor) {
835 governor_run(backer->governor);
837 /* If the governor has shrunk to its minimum size and the number of
838 * subfacets has dwindled, then drop the governor entirely.
840 * For hysteresis, the number of subfacets to drop the governor is
841 * smaller than the number needed to trigger its creation. */
842 n_subfacets = hmap_count(&backer->subfacets);
843 if (n_subfacets * 4 < flow_eviction_threshold
844 && governor_is_idle(backer->governor)) {
845 governor_destroy(backer->governor);
846 backer->governor = NULL;
853 /* Check for and handle port changes in 'backer''s dpif. */
855 process_dpif_port_changes(struct dpif_backer *backer)
861 error = dpif_port_poll(backer->dpif, &devname);
867 process_dpif_all_ports_changed(backer);
871 process_dpif_port_change(backer, devname);
876 process_dpif_port_error(backer, error);
883 process_dpif_all_ports_changed(struct dpif_backer *backer)
885 struct ofproto_dpif *ofproto;
886 struct dpif_port dpif_port;
887 struct dpif_port_dump dump;
888 struct sset devnames;
891 sset_init(&devnames);
892 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
893 if (ofproto->backer == backer) {
894 struct ofport *ofport;
896 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
897 sset_add(&devnames, netdev_get_name(ofport->netdev));
901 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
902 sset_add(&devnames, dpif_port.name);
905 SSET_FOR_EACH (devname, &devnames) {
906 process_dpif_port_change(backer, devname);
908 sset_destroy(&devnames);
912 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
914 struct ofproto_dpif *ofproto;
915 struct dpif_port port;
917 /* Don't report on the datapath's device. */
918 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
922 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
923 &all_ofproto_dpifs) {
924 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
929 ofproto = lookup_ofproto_dpif_by_port_name(devname);
930 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
931 /* The port was removed. If we know the datapath,
932 * report it through poll_set(). If we don't, it may be
933 * notifying us of a removal we initiated, so ignore it.
934 * If there's a pending ENOBUFS, let it stand, since
935 * everything will be reevaluated. */
936 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
937 sset_add(&ofproto->port_poll_set, devname);
938 ofproto->port_poll_errno = 0;
940 } else if (!ofproto) {
941 /* The port was added, but we don't know with which
942 * ofproto we should associate it. Delete it. */
943 dpif_port_del(backer->dpif, port.port_no);
945 struct ofport_dpif *ofport;
947 ofport = ofport_dpif_cast(shash_find_data(
948 &ofproto->up.port_by_name, devname));
950 && ofport->odp_port != port.port_no
951 && !odp_port_to_ofport(backer, port.port_no))
953 /* 'ofport''s datapath port number has changed from
954 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
955 * structures to match. */
956 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
957 ofport->odp_port = port.port_no;
958 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
959 hash_odp_port(port.port_no));
960 backer->need_revalidate = REV_RECONFIGURE;
963 dpif_port_destroy(&port);
966 /* Propagate 'error' to all ofprotos based on 'backer'. */
968 process_dpif_port_error(struct dpif_backer *backer, int error)
970 struct ofproto_dpif *ofproto;
972 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
973 if (ofproto->backer == backer) {
974 sset_clear(&ofproto->port_poll_set);
975 ofproto->port_poll_errno = error;
981 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
985 /* If recv_set_enable is false, we should not handle upcalls. */
986 if (!backer->recv_set_enable) {
990 /* Handle one or more batches of upcalls, until there's nothing left to do
991 * or until we do a fixed total amount of work.
993 * We do work in batches because it can be much cheaper to set up a number
994 * of flows and fire off their patches all at once. We do multiple batches
995 * because in some cases handling a packet can cause another packet to be
996 * queued almost immediately as part of the return flow. Both
997 * optimizations can make major improvements on some benchmarks and
998 * presumably for real traffic as well. */
1000 while (work < max_batch) {
1001 int retval = handle_upcalls(backer, max_batch - work);
1012 type_run_fast(const char *type)
1014 struct dpif_backer *backer;
1016 backer = shash_find_data(&all_dpif_backers, type);
1018 /* This is not necessarily a problem, since backers are only
1019 * created on demand. */
1023 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1029 static long long int port_rl = LLONG_MIN;
1030 static unsigned int backer_rl = 0;
1032 if (time_msec() >= port_rl) {
1033 struct ofproto_dpif *ofproto;
1035 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1036 run_fast(&ofproto->up);
1038 port_rl = time_msec() + 200;
1041 /* XXX: We have to be careful not to do too much work in this function. If
1042 * we call dpif_backer_run_fast() too often, or with too large a batch,
1043 * performance improves signifcantly, but at a cost. It's possible for the
1044 * number of flows in the datapath to increase without bound, and for poll
1045 * loops to take 10s of seconds. The correct solution to this problem,
1046 * long term, is to separate flow miss handling into it's own thread so it
1047 * isn't affected by revalidations, and expirations. Until then, this is
1048 * the best we can do. */
1049 if (++backer_rl >= 10) {
1050 struct shash_node *node;
1053 SHASH_FOR_EACH (node, &all_dpif_backers) {
1054 dpif_backer_run_fast(node->data, 1);
1060 type_wait(const char *type)
1062 struct dpif_backer *backer;
1064 backer = shash_find_data(&all_dpif_backers, type);
1066 /* This is not necessarily a problem, since backers are only
1067 * created on demand. */
1071 if (backer->governor) {
1072 governor_wait(backer->governor);
1075 timer_wait(&backer->next_expiration);
1078 /* Basic life-cycle. */
1080 static int add_internal_flows(struct ofproto_dpif *);
1082 static struct ofproto *
1085 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1086 return &ofproto->up;
1090 dealloc(struct ofproto *ofproto_)
1092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1097 close_dpif_backer(struct dpif_backer *backer)
1099 struct shash_node *node;
1101 ovs_assert(backer->refcount > 0);
1103 if (--backer->refcount) {
1107 drop_key_clear(backer);
1108 hmap_destroy(&backer->drop_keys);
1110 simap_destroy(&backer->tnl_backers);
1111 hmap_destroy(&backer->odp_to_ofport_map);
1112 node = shash_find(&all_dpif_backers, backer->type);
1114 shash_delete(&all_dpif_backers, node);
1115 dpif_close(backer->dpif);
1117 ovs_assert(hmap_is_empty(&backer->subfacets));
1118 hmap_destroy(&backer->subfacets);
1119 governor_destroy(backer->governor);
1124 /* Datapath port slated for removal from datapath. */
1125 struct odp_garbage {
1126 struct list list_node;
1127 odp_port_t odp_port;
1131 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1133 struct dpif_backer *backer;
1134 struct dpif_port_dump port_dump;
1135 struct dpif_port port;
1136 struct shash_node *node;
1137 struct list garbage_list;
1138 struct odp_garbage *garbage, *next;
1144 backer = shash_find_data(&all_dpif_backers, type);
1151 backer_name = xasprintf("ovs-%s", type);
1153 /* Remove any existing datapaths, since we assume we're the only
1154 * userspace controlling the datapath. */
1156 dp_enumerate_names(type, &names);
1157 SSET_FOR_EACH(name, &names) {
1158 struct dpif *old_dpif;
1160 /* Don't remove our backer if it exists. */
1161 if (!strcmp(name, backer_name)) {
1165 if (dpif_open(name, type, &old_dpif)) {
1166 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1168 dpif_delete(old_dpif);
1169 dpif_close(old_dpif);
1172 sset_destroy(&names);
1174 backer = xmalloc(sizeof *backer);
1176 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1179 VLOG_ERR("failed to open datapath of type %s: %s", type,
1180 ovs_strerror(error));
1185 backer->type = xstrdup(type);
1186 backer->governor = NULL;
1187 backer->refcount = 1;
1188 hmap_init(&backer->odp_to_ofport_map);
1189 hmap_init(&backer->drop_keys);
1190 hmap_init(&backer->subfacets);
1191 timer_set_duration(&backer->next_expiration, 1000);
1192 backer->need_revalidate = 0;
1193 simap_init(&backer->tnl_backers);
1194 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1197 if (backer->recv_set_enable) {
1198 dpif_flow_flush(backer->dpif);
1201 /* Loop through the ports already on the datapath and remove any
1202 * that we don't need anymore. */
1203 list_init(&garbage_list);
1204 dpif_port_dump_start(&port_dump, backer->dpif);
1205 while (dpif_port_dump_next(&port_dump, &port)) {
1206 node = shash_find(&init_ofp_ports, port.name);
1207 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1208 garbage = xmalloc(sizeof *garbage);
1209 garbage->odp_port = port.port_no;
1210 list_push_front(&garbage_list, &garbage->list_node);
1213 dpif_port_dump_done(&port_dump);
1215 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1216 dpif_port_del(backer->dpif, garbage->odp_port);
1217 list_remove(&garbage->list_node);
1221 shash_add(&all_dpif_backers, type, backer);
1223 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1225 VLOG_ERR("failed to listen on datapath of type %s: %s",
1226 type, ovs_strerror(error));
1227 close_dpif_backer(backer);
1231 backer->max_n_subfacet = 0;
1232 backer->created = time_msec();
1233 backer->last_minute = backer->created;
1234 memset(&backer->hourly, 0, sizeof backer->hourly);
1235 memset(&backer->daily, 0, sizeof backer->daily);
1236 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1237 backer->subfacet_add_count = 0;
1238 backer->subfacet_del_count = 0;
1239 backer->total_subfacet_add_count = 0;
1240 backer->total_subfacet_del_count = 0;
1241 backer->avg_n_subfacet = 0;
1242 backer->avg_subfacet_life = 0;
1248 construct(struct ofproto *ofproto_)
1250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1251 struct shash_node *node, *next;
1252 odp_port_t max_ports;
1255 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1260 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1261 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1262 ofp_to_u16(OFPP_MAX))));
1264 ofproto->netflow = NULL;
1265 ofproto->sflow = NULL;
1266 ofproto->ipfix = NULL;
1267 ofproto->stp = NULL;
1268 hmap_init(&ofproto->bundles);
1269 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1270 ofproto->mbridge = mbridge_create();
1271 ofproto->has_bonded_bundles = false;
1273 classifier_init(&ofproto->facets);
1274 ofproto->consistency_rl = LLONG_MIN;
1276 list_init(&ofproto->completions);
1278 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1279 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1280 list_init(&ofproto->flow_mods);
1281 ofproto->n_flow_mods = 0;
1282 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1284 ofproto_dpif_unixctl_init();
1286 hmap_init(&ofproto->vlandev_map);
1287 hmap_init(&ofproto->realdev_vid_map);
1289 sset_init(&ofproto->ports);
1290 sset_init(&ofproto->ghost_ports);
1291 sset_init(&ofproto->port_poll_set);
1292 ofproto->port_poll_errno = 0;
1294 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1295 struct iface_hint *iface_hint = node->data;
1297 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1298 /* Check if the datapath already has this port. */
1299 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1300 sset_add(&ofproto->ports, node->name);
1303 free(iface_hint->br_name);
1304 free(iface_hint->br_type);
1306 shash_delete(&init_ofp_ports, node);
1310 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1311 hash_string(ofproto->up.name, 0));
1312 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1314 ofproto_init_tables(ofproto_, N_TABLES);
1315 error = add_internal_flows(ofproto);
1316 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1319 ofproto->n_missed = 0;
1325 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1326 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1328 struct ofputil_flow_mod fm;
1331 match_init_catchall(&fm.match);
1333 match_set_reg(&fm.match, 0, id);
1334 fm.new_cookie = htonll(0);
1335 fm.cookie = htonll(0);
1336 fm.cookie_mask = htonll(0);
1337 fm.modify_cookie = false;
1338 fm.table_id = TBL_INTERNAL;
1339 fm.command = OFPFC_ADD;
1340 fm.idle_timeout = 0;
1341 fm.hard_timeout = 0;
1345 fm.ofpacts = ofpacts->data;
1346 fm.ofpacts_len = ofpacts->size;
1348 error = ofproto_flow_mod(&ofproto->up, &fm);
1350 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1351 id, ofperr_to_string(error));
1355 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1357 ovs_assert(*rulep != NULL);
1363 add_internal_flows(struct ofproto_dpif *ofproto)
1365 struct ofpact_controller *controller;
1366 uint64_t ofpacts_stub[128 / 8];
1367 struct ofpbuf ofpacts;
1371 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1374 controller = ofpact_put_CONTROLLER(&ofpacts);
1375 controller->max_len = UINT16_MAX;
1376 controller->controller_id = 0;
1377 controller->reason = OFPR_NO_MATCH;
1378 ofpact_pad(&ofpacts);
1380 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1385 ofpbuf_clear(&ofpacts);
1386 error = add_internal_flow(ofproto, id++, &ofpacts,
1387 &ofproto->no_packet_in_rule);
1392 error = add_internal_flow(ofproto, id++, &ofpacts,
1393 &ofproto->drop_frags_rule);
1398 complete_operations(struct ofproto_dpif *ofproto)
1400 struct dpif_completion *c, *next;
1402 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1403 ofoperation_complete(c->op, 0);
1404 list_remove(&c->list_node);
1410 destruct(struct ofproto *ofproto_)
1412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1413 struct rule_dpif *rule, *next_rule;
1414 struct ofputil_flow_mod *fm, *next_fm;
1415 struct oftable *table;
1417 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1418 xlate_remove_ofproto(ofproto);
1420 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1421 complete_operations(ofproto);
1423 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1424 struct cls_cursor cursor;
1426 cls_cursor_init(&cursor, &table->cls, NULL);
1427 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1428 ofproto_rule_destroy(&rule->up);
1432 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1433 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1434 list_remove(&fm->list_node);
1435 ofproto->n_flow_mods--;
1439 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1440 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1442 mbridge_unref(ofproto->mbridge);
1444 netflow_destroy(ofproto->netflow);
1445 dpif_sflow_unref(ofproto->sflow);
1446 hmap_destroy(&ofproto->bundles);
1447 mac_learning_unref(ofproto->ml);
1449 classifier_destroy(&ofproto->facets);
1451 hmap_destroy(&ofproto->vlandev_map);
1452 hmap_destroy(&ofproto->realdev_vid_map);
1454 sset_destroy(&ofproto->ports);
1455 sset_destroy(&ofproto->ghost_ports);
1456 sset_destroy(&ofproto->port_poll_set);
1458 close_dpif_backer(ofproto->backer);
1462 run_fast(struct ofproto *ofproto_)
1464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1465 struct ofputil_flow_mod *fm, *next;
1466 struct ofport_dpif *ofport;
1467 struct list flow_mods;
1469 /* Do not perform any periodic activity required by 'ofproto' while
1470 * waiting for flow restore to complete. */
1471 if (ofproto_get_flow_restore_wait()) {
1475 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1476 if (ofproto->n_flow_mods) {
1477 flow_mods = ofproto->flow_mods;
1478 list_moved(&flow_mods);
1479 list_init(&ofproto->flow_mods);
1480 ofproto->n_flow_mods = 0;
1482 list_init(&flow_mods);
1484 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1486 LIST_FOR_EACH_SAFE (fm, next, list_node, &flow_mods) {
1487 int error = ofproto_flow_mod(&ofproto->up, fm);
1488 if (error && !VLOG_DROP_WARN(&rl)) {
1489 VLOG_WARN("learning action failed to modify flow table (%s)",
1490 ofperr_get_name(error));
1493 list_remove(&fm->list_node);
1498 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1499 port_run_fast(ofport);
1506 run(struct ofproto *ofproto_)
1508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1509 struct ofport_dpif *ofport;
1510 struct ofbundle *bundle;
1514 complete_operations(ofproto);
1517 if (mbridge_need_revalidate(ofproto->mbridge)) {
1518 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1519 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1520 mac_learning_flush(ofproto->ml);
1521 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1524 /* Do not perform any periodic activity below required by 'ofproto' while
1525 * waiting for flow restore to complete. */
1526 if (ofproto_get_flow_restore_wait()) {
1530 error = run_fast(ofproto_);
1535 if (ofproto->netflow) {
1536 if (netflow_run(ofproto->netflow)) {
1537 send_netflow_active_timeouts(ofproto);
1540 if (ofproto->sflow) {
1541 dpif_sflow_run(ofproto->sflow);
1544 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1547 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1552 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1553 if (mac_learning_run(ofproto->ml)) {
1554 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1556 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1558 /* Check the consistency of a random facet, to aid debugging. */
1559 if (time_msec() >= ofproto->consistency_rl
1560 && !classifier_is_empty(&ofproto->facets)
1561 && !ofproto->backer->need_revalidate) {
1562 struct cls_table *table;
1563 struct cls_rule *cr;
1564 struct facet *facet;
1566 ofproto->consistency_rl = time_msec() + 250;
1568 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1569 struct cls_table, hmap_node);
1570 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1572 facet = CONTAINER_OF(cr, struct facet, cr);
1574 if (!facet_check_consistency(facet)) {
1575 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1583 wait(struct ofproto *ofproto_)
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586 struct ofport_dpif *ofport;
1587 struct ofbundle *bundle;
1589 if (!clogged && !list_is_empty(&ofproto->completions)) {
1590 poll_immediate_wake();
1593 if (ofproto_get_flow_restore_wait()) {
1597 dpif_wait(ofproto->backer->dpif);
1598 dpif_recv_wait(ofproto->backer->dpif);
1599 if (ofproto->sflow) {
1600 dpif_sflow_wait(ofproto->sflow);
1602 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1605 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1606 bundle_wait(bundle);
1608 if (ofproto->netflow) {
1609 netflow_wait(ofproto->netflow);
1611 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1612 mac_learning_wait(ofproto->ml);
1613 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1615 if (ofproto->backer->need_revalidate) {
1616 /* Shouldn't happen, but if it does just go around again. */
1617 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1618 poll_immediate_wake();
1623 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1625 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1626 struct cls_cursor cursor;
1627 size_t n_subfacets = 0;
1628 struct facet *facet;
1630 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1632 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1633 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1634 n_subfacets += list_size(&facet->subfacets);
1636 simap_increase(usage, "subfacets", n_subfacets);
1640 flush(struct ofproto *ofproto_)
1642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1643 struct subfacet *subfacet, *next_subfacet;
1644 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1648 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1649 &ofproto->backer->subfacets) {
1650 if (subfacet->facet->ofproto != ofproto) {
1654 if (subfacet->path != SF_NOT_INSTALLED) {
1655 batch[n_batch++] = subfacet;
1656 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1657 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1661 subfacet_destroy(subfacet);
1666 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1671 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1672 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1674 *arp_match_ip = true;
1675 *actions = (OFPUTIL_A_OUTPUT |
1676 OFPUTIL_A_SET_VLAN_VID |
1677 OFPUTIL_A_SET_VLAN_PCP |
1678 OFPUTIL_A_STRIP_VLAN |
1679 OFPUTIL_A_SET_DL_SRC |
1680 OFPUTIL_A_SET_DL_DST |
1681 OFPUTIL_A_SET_NW_SRC |
1682 OFPUTIL_A_SET_NW_DST |
1683 OFPUTIL_A_SET_NW_TOS |
1684 OFPUTIL_A_SET_TP_SRC |
1685 OFPUTIL_A_SET_TP_DST |
1690 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1693 struct dpif_dp_stats s;
1694 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1697 strcpy(ots->name, "classifier");
1699 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1700 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1701 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1702 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1704 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1705 ots->lookup_count = htonll(n_lookup);
1706 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1709 static struct ofport *
1712 struct ofport_dpif *port = xmalloc(sizeof *port);
1717 port_dealloc(struct ofport *port_)
1719 struct ofport_dpif *port = ofport_dpif_cast(port_);
1724 port_construct(struct ofport *port_)
1726 struct ofport_dpif *port = ofport_dpif_cast(port_);
1727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1728 const struct netdev *netdev = port->up.netdev;
1729 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1730 struct dpif_port dpif_port;
1733 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1734 port->bundle = NULL;
1737 port->may_enable = true;
1738 port->stp_port = NULL;
1739 port->stp_state = STP_DISABLED;
1740 port->is_tunnel = false;
1744 port->realdev_ofp_port = 0;
1745 port->vlandev_vid = 0;
1746 port->carrier_seq = netdev_get_carrier_resets(netdev);
1748 if (netdev_vport_is_patch(netdev)) {
1749 /* By bailing out here, we don't submit the port to the sFlow module
1750 * to be considered for counter polling export. This is correct
1751 * because the patch port represents an interface that sFlow considers
1752 * to be "internal" to the switch as a whole, and therefore not an
1753 * candidate for counter polling. */
1754 port->odp_port = ODPP_NONE;
1755 ofport_update_peer(port);
1759 error = dpif_port_query_by_name(ofproto->backer->dpif,
1760 netdev_vport_get_dpif_port(netdev, namebuf,
1767 port->odp_port = dpif_port.port_no;
1769 if (netdev_get_tunnel_config(netdev)) {
1770 tnl_port_add(port, port->up.netdev, port->odp_port);
1771 port->is_tunnel = true;
1773 /* Sanity-check that a mapping doesn't already exist. This
1774 * shouldn't happen for non-tunnel ports. */
1775 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1776 VLOG_ERR("port %s already has an OpenFlow port number",
1778 dpif_port_destroy(&dpif_port);
1782 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1783 hash_odp_port(port->odp_port));
1785 dpif_port_destroy(&dpif_port);
1787 if (ofproto->sflow) {
1788 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1795 port_destruct(struct ofport *port_)
1797 struct ofport_dpif *port = ofport_dpif_cast(port_);
1798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1799 const char *devname = netdev_get_name(port->up.netdev);
1800 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1801 const char *dp_port_name;
1803 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1804 xlate_ofport_remove(port);
1806 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1808 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1809 /* The underlying device is still there, so delete it. This
1810 * happens when the ofproto is being destroyed, since the caller
1811 * assumes that removal of attached ports will happen as part of
1813 if (!port->is_tunnel) {
1814 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1819 port->peer->peer = NULL;
1823 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1824 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1828 sset_find_and_delete(&ofproto->ports, devname);
1829 sset_find_and_delete(&ofproto->ghost_ports, devname);
1830 bundle_remove(port_);
1831 set_cfm(port_, NULL);
1832 set_bfd(port_, NULL);
1833 if (ofproto->sflow) {
1834 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1841 port_modified(struct ofport *port_)
1843 struct ofport_dpif *port = ofport_dpif_cast(port_);
1845 if (port->bundle && port->bundle->bond) {
1846 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1850 cfm_set_netdev(port->cfm, port->up.netdev);
1853 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1855 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1859 ofport_update_peer(port);
1863 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1865 struct ofport_dpif *port = ofport_dpif_cast(port_);
1866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1867 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1869 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1870 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1871 OFPUTIL_PC_NO_PACKET_IN)) {
1872 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1874 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1875 bundle_update(port->bundle);
1881 set_sflow(struct ofproto *ofproto_,
1882 const struct ofproto_sflow_options *sflow_options)
1884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1885 struct dpif_sflow *ds = ofproto->sflow;
1887 if (sflow_options) {
1889 struct ofport_dpif *ofport;
1891 ds = ofproto->sflow = dpif_sflow_create();
1892 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1893 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1895 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1897 dpif_sflow_set_options(ds, sflow_options);
1900 dpif_sflow_unref(ds);
1901 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1902 ofproto->sflow = NULL;
1910 struct ofproto *ofproto_,
1911 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1912 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1913 size_t n_flow_exporters_options)
1915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1916 struct dpif_ipfix *di = ofproto->ipfix;
1918 if (bridge_exporter_options || flow_exporters_options) {
1920 di = ofproto->ipfix = dpif_ipfix_create();
1922 dpif_ipfix_set_options(
1923 di, bridge_exporter_options, flow_exporters_options,
1924 n_flow_exporters_options);
1927 dpif_ipfix_unref(di);
1928 ofproto->ipfix = NULL;
1935 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1937 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1944 struct ofproto_dpif *ofproto;
1946 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1947 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1948 ofport->cfm = cfm_create(ofport->up.netdev);
1951 if (cfm_configure(ofport->cfm, s)) {
1957 cfm_unref(ofport->cfm);
1963 get_cfm_status(const struct ofport *ofport_,
1964 struct ofproto_cfm_status *status)
1966 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1969 status->faults = cfm_get_fault(ofport->cfm);
1970 status->remote_opstate = cfm_get_opup(ofport->cfm);
1971 status->health = cfm_get_health(ofport->cfm);
1972 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1980 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1983 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1988 if (ofport->bfd != old) {
1989 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1996 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1998 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2001 bfd_get_status(ofport->bfd, smap);
2008 /* Spanning Tree. */
2011 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2013 struct ofproto_dpif *ofproto = ofproto_;
2014 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2015 struct ofport_dpif *ofport;
2017 ofport = stp_port_get_aux(sp);
2019 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2020 ofproto->up.name, port_num);
2022 struct eth_header *eth = pkt->l2;
2024 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2025 if (eth_addr_is_zero(eth->eth_src)) {
2026 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2027 "with unknown MAC", ofproto->up.name, port_num);
2029 send_packet(ofport, pkt);
2035 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2037 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2041 /* Only revalidate flows if the configuration changed. */
2042 if (!s != !ofproto->stp) {
2043 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2047 if (!ofproto->stp) {
2048 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2049 send_bpdu_cb, ofproto);
2050 ofproto->stp_last_tick = time_msec();
2053 stp_set_bridge_id(ofproto->stp, s->system_id);
2054 stp_set_bridge_priority(ofproto->stp, s->priority);
2055 stp_set_hello_time(ofproto->stp, s->hello_time);
2056 stp_set_max_age(ofproto->stp, s->max_age);
2057 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2059 struct ofport *ofport;
2061 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2062 set_stp_port(ofport, NULL);
2065 stp_unref(ofproto->stp);
2066 ofproto->stp = NULL;
2073 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2079 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2080 s->designated_root = stp_get_designated_root(ofproto->stp);
2081 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2090 update_stp_port_state(struct ofport_dpif *ofport)
2092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2093 enum stp_state state;
2095 /* Figure out new state. */
2096 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2100 if (ofport->stp_state != state) {
2101 enum ofputil_port_state of_state;
2104 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2105 netdev_get_name(ofport->up.netdev),
2106 stp_state_name(ofport->stp_state),
2107 stp_state_name(state));
2108 if (stp_learn_in_state(ofport->stp_state)
2109 != stp_learn_in_state(state)) {
2110 /* xxx Learning action flows should also be flushed. */
2111 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2112 mac_learning_flush(ofproto->ml);
2113 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2115 fwd_change = stp_forward_in_state(ofport->stp_state)
2116 != stp_forward_in_state(state);
2118 ofproto->backer->need_revalidate = REV_STP;
2119 ofport->stp_state = state;
2120 ofport->stp_state_entered = time_msec();
2122 if (fwd_change && ofport->bundle) {
2123 bundle_update(ofport->bundle);
2126 /* Update the STP state bits in the OpenFlow port description. */
2127 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2128 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2129 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2130 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2131 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2133 ofproto_port_set_state(&ofport->up, of_state);
2137 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2138 * caller is responsible for assigning STP port numbers and ensuring
2139 * there are no duplicates. */
2141 set_stp_port(struct ofport *ofport_,
2142 const struct ofproto_port_stp_settings *s)
2144 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2146 struct stp_port *sp = ofport->stp_port;
2148 if (!s || !s->enable) {
2150 ofport->stp_port = NULL;
2151 stp_port_disable(sp);
2152 update_stp_port_state(ofport);
2155 } else if (sp && stp_port_no(sp) != s->port_num
2156 && ofport == stp_port_get_aux(sp)) {
2157 /* The port-id changed, so disable the old one if it's not
2158 * already in use by another port. */
2159 stp_port_disable(sp);
2162 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2163 stp_port_enable(sp);
2165 stp_port_set_aux(sp, ofport);
2166 stp_port_set_priority(sp, s->priority);
2167 stp_port_set_path_cost(sp, s->path_cost);
2169 update_stp_port_state(ofport);
2175 get_stp_port_status(struct ofport *ofport_,
2176 struct ofproto_port_stp_status *s)
2178 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2180 struct stp_port *sp = ofport->stp_port;
2182 if (!ofproto->stp || !sp) {
2188 s->port_id = stp_port_get_id(sp);
2189 s->state = stp_port_get_state(sp);
2190 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2191 s->role = stp_port_get_role(sp);
2192 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2198 stp_run(struct ofproto_dpif *ofproto)
2201 long long int now = time_msec();
2202 long long int elapsed = now - ofproto->stp_last_tick;
2203 struct stp_port *sp;
2206 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2207 ofproto->stp_last_tick = now;
2209 while (stp_get_changed_port(ofproto->stp, &sp)) {
2210 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2213 update_stp_port_state(ofport);
2217 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2218 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2219 mac_learning_flush(ofproto->ml);
2220 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2226 stp_wait(struct ofproto_dpif *ofproto)
2229 poll_timer_wait(1000);
2234 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
2235 uint32_t queue_id, uint32_t *priority)
2237 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
2241 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2244 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2247 if (ofport->n_qdscp != n_qdscp
2248 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2249 n_qdscp * sizeof *qdscp))) {
2250 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2251 free(ofport->qdscp);
2252 ofport->qdscp = n_qdscp
2253 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2255 ofport->n_qdscp = n_qdscp;
2263 /* Expires all MAC learning entries associated with 'bundle' and forces its
2264 * ofproto to revalidate every flow.
2266 * Normally MAC learning entries are removed only from the ofproto associated
2267 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2268 * are removed from every ofproto. When patch ports and SLB bonds are in use
2269 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2270 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2271 * with the host from which it migrated. */
2273 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2275 struct ofproto_dpif *ofproto = bundle->ofproto;
2276 struct mac_learning *ml = ofproto->ml;
2277 struct mac_entry *mac, *next_mac;
2279 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2280 ovs_rwlock_wrlock(&ml->rwlock);
2281 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2282 if (mac->port.p == bundle) {
2284 struct ofproto_dpif *o;
2286 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2288 struct mac_entry *e;
2290 ovs_rwlock_wrlock(&o->ml->rwlock);
2291 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2293 mac_learning_expire(o->ml, e);
2295 ovs_rwlock_unlock(&o->ml->rwlock);
2300 mac_learning_expire(ml, mac);
2303 ovs_rwlock_unlock(&ml->rwlock);
2306 static struct ofbundle *
2307 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2309 struct ofbundle *bundle;
2311 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2312 &ofproto->bundles) {
2313 if (bundle->aux == aux) {
2321 bundle_update(struct ofbundle *bundle)
2323 struct ofport_dpif *port;
2325 bundle->floodable = true;
2326 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2327 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2328 || !stp_forward_in_state(port->stp_state)) {
2329 bundle->floodable = false;
2336 bundle_del_port(struct ofport_dpif *port)
2338 struct ofbundle *bundle = port->bundle;
2340 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2342 list_remove(&port->bundle_node);
2343 port->bundle = NULL;
2346 lacp_slave_unregister(bundle->lacp, port);
2349 bond_slave_unregister(bundle->bond, port);
2352 bundle_update(bundle);
2356 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2357 struct lacp_slave_settings *lacp)
2359 struct ofport_dpif *port;
2361 port = get_ofp_port(bundle->ofproto, ofp_port);
2366 if (port->bundle != bundle) {
2367 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369 bundle_del_port(port);
2372 port->bundle = bundle;
2373 list_push_back(&bundle->ports, &port->bundle_node);
2374 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2375 || !stp_forward_in_state(port->stp_state)) {
2376 bundle->floodable = false;
2380 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2381 lacp_slave_register(bundle->lacp, port, lacp);
2388 bundle_destroy(struct ofbundle *bundle)
2390 struct ofproto_dpif *ofproto;
2391 struct ofport_dpif *port, *next_port;
2397 ofproto = bundle->ofproto;
2398 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2400 xlate_bundle_remove(bundle);
2402 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2403 bundle_del_port(port);
2406 bundle_flush_macs(bundle, true);
2407 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2409 free(bundle->trunks);
2410 lacp_unref(bundle->lacp);
2411 bond_unref(bundle->bond);
2416 bundle_set(struct ofproto *ofproto_, void *aux,
2417 const struct ofproto_bundle_settings *s)
2419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2420 bool need_flush = false;
2421 struct ofport_dpif *port;
2422 struct ofbundle *bundle;
2423 unsigned long *trunks;
2429 bundle_destroy(bundle_lookup(ofproto, aux));
2433 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2434 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2436 bundle = bundle_lookup(ofproto, aux);
2438 bundle = xmalloc(sizeof *bundle);
2440 bundle->ofproto = ofproto;
2441 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2442 hash_pointer(aux, 0));
2444 bundle->name = NULL;
2446 list_init(&bundle->ports);
2447 bundle->vlan_mode = PORT_VLAN_TRUNK;
2449 bundle->trunks = NULL;
2450 bundle->use_priority_tags = s->use_priority_tags;
2451 bundle->lacp = NULL;
2452 bundle->bond = NULL;
2454 bundle->floodable = true;
2455 mbridge_register_bundle(ofproto->mbridge, bundle);
2458 if (!bundle->name || strcmp(s->name, bundle->name)) {
2460 bundle->name = xstrdup(s->name);
2465 if (!bundle->lacp) {
2466 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2467 bundle->lacp = lacp_create();
2469 lacp_configure(bundle->lacp, s->lacp);
2471 lacp_unref(bundle->lacp);
2472 bundle->lacp = NULL;
2475 /* Update set of ports. */
2477 for (i = 0; i < s->n_slaves; i++) {
2478 if (!bundle_add_port(bundle, s->slaves[i],
2479 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2483 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2484 struct ofport_dpif *next_port;
2486 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2487 for (i = 0; i < s->n_slaves; i++) {
2488 if (s->slaves[i] == port->up.ofp_port) {
2493 bundle_del_port(port);
2497 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2499 if (list_is_empty(&bundle->ports)) {
2500 bundle_destroy(bundle);
2504 /* Set VLAN tagging mode */
2505 if (s->vlan_mode != bundle->vlan_mode
2506 || s->use_priority_tags != bundle->use_priority_tags) {
2507 bundle->vlan_mode = s->vlan_mode;
2508 bundle->use_priority_tags = s->use_priority_tags;
2513 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2514 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2516 if (vlan != bundle->vlan) {
2517 bundle->vlan = vlan;
2521 /* Get trunked VLANs. */
2522 switch (s->vlan_mode) {
2523 case PORT_VLAN_ACCESS:
2527 case PORT_VLAN_TRUNK:
2528 trunks = CONST_CAST(unsigned long *, s->trunks);
2531 case PORT_VLAN_NATIVE_UNTAGGED:
2532 case PORT_VLAN_NATIVE_TAGGED:
2533 if (vlan != 0 && (!s->trunks
2534 || !bitmap_is_set(s->trunks, vlan)
2535 || bitmap_is_set(s->trunks, 0))) {
2536 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2538 trunks = bitmap_clone(s->trunks, 4096);
2540 trunks = bitmap_allocate1(4096);
2542 bitmap_set1(trunks, vlan);
2543 bitmap_set0(trunks, 0);
2545 trunks = CONST_CAST(unsigned long *, s->trunks);
2552 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2553 free(bundle->trunks);
2554 if (trunks == s->trunks) {
2555 bundle->trunks = vlan_bitmap_clone(trunks);
2557 bundle->trunks = trunks;
2562 if (trunks != s->trunks) {
2567 if (!list_is_short(&bundle->ports)) {
2568 bundle->ofproto->has_bonded_bundles = true;
2570 if (bond_reconfigure(bundle->bond, s->bond)) {
2571 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2574 bundle->bond = bond_create(s->bond);
2575 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2578 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2579 bond_slave_register(bundle->bond, port, port->up.netdev);
2582 bond_unref(bundle->bond);
2583 bundle->bond = NULL;
2586 /* If we changed something that would affect MAC learning, un-learn
2587 * everything on this port and force flow revalidation. */
2589 bundle_flush_macs(bundle, false);
2596 bundle_remove(struct ofport *port_)
2598 struct ofport_dpif *port = ofport_dpif_cast(port_);
2599 struct ofbundle *bundle = port->bundle;
2602 bundle_del_port(port);
2603 if (list_is_empty(&bundle->ports)) {
2604 bundle_destroy(bundle);
2605 } else if (list_is_short(&bundle->ports)) {
2606 bond_unref(bundle->bond);
2607 bundle->bond = NULL;
2613 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2615 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2616 struct ofport_dpif *port = port_;
2617 uint8_t ea[ETH_ADDR_LEN];
2620 error = netdev_get_etheraddr(port->up.netdev, ea);
2622 struct ofpbuf packet;
2625 ofpbuf_init(&packet, 0);
2626 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2628 memcpy(packet_pdu, pdu, pdu_size);
2630 send_packet(port, &packet);
2631 ofpbuf_uninit(&packet);
2633 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2634 "%s (%s)", port->bundle->name,
2635 netdev_get_name(port->up.netdev), ovs_strerror(error));
2640 bundle_send_learning_packets(struct ofbundle *bundle)
2642 struct ofproto_dpif *ofproto = bundle->ofproto;
2643 int error, n_packets, n_errors;
2644 struct mac_entry *e;
2646 error = n_packets = n_errors = 0;
2647 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2648 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2649 if (e->port.p != bundle) {
2650 struct ofpbuf *learning_packet;
2651 struct ofport_dpif *port;
2655 /* The assignment to "port" is unnecessary but makes "grep"ing for
2656 * struct ofport_dpif more effective. */
2657 learning_packet = bond_compose_learning_packet(bundle->bond,
2661 ret = send_packet(port, learning_packet);
2662 ofpbuf_delete(learning_packet);
2670 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2674 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2675 "packets, last error was: %s",
2676 bundle->name, n_errors, n_packets, ovs_strerror(error));
2678 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2679 bundle->name, n_packets);
2684 bundle_run(struct ofbundle *bundle)
2687 lacp_run(bundle->lacp, send_pdu_cb);
2690 struct ofport_dpif *port;
2692 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2693 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2696 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2697 bundle->ofproto->backer->need_revalidate = REV_BOND;
2700 if (bond_should_send_learning_packets(bundle->bond)) {
2701 bundle_send_learning_packets(bundle);
2707 bundle_wait(struct ofbundle *bundle)
2710 lacp_wait(bundle->lacp);
2713 bond_wait(bundle->bond);
2720 mirror_set__(struct ofproto *ofproto_, void *aux,
2721 const struct ofproto_mirror_settings *s)
2723 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2724 struct ofbundle **srcs, **dsts;
2729 mirror_destroy(ofproto->mbridge, aux);
2733 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2734 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2736 for (i = 0; i < s->n_srcs; i++) {
2737 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2740 for (i = 0; i < s->n_dsts; i++) {
2741 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2744 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2745 s->n_dsts, s->src_vlans,
2746 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2753 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2754 uint64_t *packets, uint64_t *bytes)
2757 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2762 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2765 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2766 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2767 mac_learning_flush(ofproto->ml);
2769 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2774 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2776 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2777 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2778 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2782 forward_bpdu_changed(struct ofproto *ofproto_)
2784 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2785 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2789 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2793 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2794 mac_learning_set_idle_time(ofproto->ml, idle_time);
2795 mac_learning_set_max_entries(ofproto->ml, max_entries);
2796 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2801 static struct ofport_dpif *
2802 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2804 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2805 return ofport ? ofport_dpif_cast(ofport) : NULL;
2808 static struct ofport_dpif *
2809 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2811 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2812 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2816 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2817 struct ofproto_port *ofproto_port,
2818 struct dpif_port *dpif_port)
2820 ofproto_port->name = dpif_port->name;
2821 ofproto_port->type = dpif_port->type;
2822 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2826 ofport_update_peer(struct ofport_dpif *ofport)
2828 const struct ofproto_dpif *ofproto;
2829 struct dpif_backer *backer;
2830 const char *peer_name;
2832 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2836 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2837 backer->need_revalidate = REV_RECONFIGURE;
2840 ofport->peer->peer = NULL;
2841 ofport->peer = NULL;
2844 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2849 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2850 struct ofport *peer_ofport;
2851 struct ofport_dpif *peer;
2852 const char *peer_peer;
2854 if (ofproto->backer != backer) {
2858 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2863 peer = ofport_dpif_cast(peer_ofport);
2864 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2865 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2867 ofport->peer = peer;
2868 ofport->peer->peer = ofport;
2876 port_run_fast(struct ofport_dpif *ofport)
2878 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2879 struct ofpbuf packet;
2881 ofpbuf_init(&packet, 0);
2882 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2883 send_packet(ofport, &packet);
2884 ofpbuf_uninit(&packet);
2887 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2888 struct ofpbuf packet;
2890 ofpbuf_init(&packet, 0);
2891 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2892 send_packet(ofport, &packet);
2893 ofpbuf_uninit(&packet);
2898 port_run(struct ofport_dpif *ofport)
2900 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2901 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2902 bool enable = netdev_get_carrier(ofport->up.netdev);
2904 ofport->carrier_seq = carrier_seq;
2906 port_run_fast(ofport);
2909 int cfm_opup = cfm_get_opup(ofport->cfm);
2911 cfm_run(ofport->cfm);
2912 enable = enable && !cfm_get_fault(ofport->cfm);
2914 if (cfm_opup >= 0) {
2915 enable = enable && cfm_opup;
2920 bfd_run(ofport->bfd);
2921 enable = enable && bfd_forwarding(ofport->bfd);
2924 if (ofport->bundle) {
2925 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2926 if (carrier_changed) {
2927 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2931 if (ofport->may_enable != enable) {
2932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2933 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2936 ofport->may_enable = enable;
2940 port_wait(struct ofport_dpif *ofport)
2943 cfm_wait(ofport->cfm);
2947 bfd_wait(ofport->bfd);
2952 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2953 struct ofproto_port *ofproto_port)
2955 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2956 struct dpif_port dpif_port;
2959 if (sset_contains(&ofproto->ghost_ports, devname)) {
2960 const char *type = netdev_get_type_from_name(devname);
2962 /* We may be called before ofproto->up.port_by_name is populated with
2963 * the appropriate ofport. For this reason, we must get the name and
2964 * type from the netdev layer directly. */
2966 const struct ofport *ofport;
2968 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2969 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2970 ofproto_port->name = xstrdup(devname);
2971 ofproto_port->type = xstrdup(type);
2977 if (!sset_contains(&ofproto->ports, devname)) {
2980 error = dpif_port_query_by_name(ofproto->backer->dpif,
2981 devname, &dpif_port);
2983 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2989 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2992 const char *devname = netdev_get_name(netdev);
2993 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2994 const char *dp_port_name;
2996 if (netdev_vport_is_patch(netdev)) {
2997 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3001 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3002 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3003 odp_port_t port_no = ODPP_NONE;
3006 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3010 if (netdev_get_tunnel_config(netdev)) {
3011 simap_put(&ofproto->backer->tnl_backers,
3012 dp_port_name, odp_to_u32(port_no));
3016 if (netdev_get_tunnel_config(netdev)) {
3017 sset_add(&ofproto->ghost_ports, devname);
3019 sset_add(&ofproto->ports, devname);
3025 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3028 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3035 sset_find_and_delete(&ofproto->ghost_ports,
3036 netdev_get_name(ofport->up.netdev));
3037 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3038 if (!ofport->is_tunnel) {
3039 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3041 /* The caller is going to close ofport->up.netdev. If this is a
3042 * bonded port, then the bond is using that netdev, so remove it
3043 * from the bond. The client will need to reconfigure everything
3044 * after deleting ports, so then the slave will get re-added. */
3045 bundle_remove(&ofport->up);
3052 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3054 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3059 error = netdev_get_stats(ofport->up.netdev, stats);
3061 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3064 /* ofproto->stats.tx_packets represents packets that we created
3065 * internally and sent to some port (e.g. packets sent with
3066 * send_packet()). Account for them as if they had come from
3067 * OFPP_LOCAL and got forwarded. */
3069 if (stats->rx_packets != UINT64_MAX) {
3070 stats->rx_packets += ofproto->stats.tx_packets;
3073 if (stats->rx_bytes != UINT64_MAX) {
3074 stats->rx_bytes += ofproto->stats.tx_bytes;
3077 /* ofproto->stats.rx_packets represents packets that were received on
3078 * some port and we processed internally and dropped (e.g. STP).
3079 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3081 if (stats->tx_packets != UINT64_MAX) {
3082 stats->tx_packets += ofproto->stats.rx_packets;
3085 if (stats->tx_bytes != UINT64_MAX) {
3086 stats->tx_bytes += ofproto->stats.rx_bytes;
3093 struct port_dump_state {
3098 struct ofproto_port port;
3103 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3105 *statep = xzalloc(sizeof(struct port_dump_state));
3110 port_dump_next(const struct ofproto *ofproto_, void *state_,
3111 struct ofproto_port *port)
3113 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3114 struct port_dump_state *state = state_;
3115 const struct sset *sset;
3116 struct sset_node *node;
3118 if (state->has_port) {
3119 ofproto_port_destroy(&state->port);
3120 state->has_port = false;
3122 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3123 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3126 error = port_query_by_name(ofproto_, node->name, &state->port);
3128 *port = state->port;
3129 state->has_port = true;
3131 } else if (error != ENODEV) {
3136 if (!state->ghost) {
3137 state->ghost = true;
3140 return port_dump_next(ofproto_, state_, port);
3147 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3149 struct port_dump_state *state = state_;
3151 if (state->has_port) {
3152 ofproto_port_destroy(&state->port);
3159 port_poll(const struct ofproto *ofproto_, char **devnamep)
3161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3163 if (ofproto->port_poll_errno) {
3164 int error = ofproto->port_poll_errno;
3165 ofproto->port_poll_errno = 0;
3169 if (sset_is_empty(&ofproto->port_poll_set)) {
3173 *devnamep = sset_pop(&ofproto->port_poll_set);
3178 port_poll_wait(const struct ofproto *ofproto_)
3180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3181 dpif_port_poll_wait(ofproto->backer->dpif);
3185 port_is_lacp_current(const struct ofport *ofport_)
3187 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3188 return (ofport->bundle && ofport->bundle->lacp
3189 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3193 /* Upcall handling. */
3195 /* Flow miss batching.
3197 * Some dpifs implement operations faster when you hand them off in a batch.
3198 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3199 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3200 * more packets, plus possibly installing the flow in the dpif.
3202 * So far we only batch the operations that affect flow setup time the most.
3203 * It's possible to batch more than that, but the benefit might be minimal. */
3205 struct hmap_node hmap_node;
3206 struct ofproto_dpif *ofproto;
3208 enum odp_key_fitness key_fitness;
3209 const struct nlattr *key;
3211 struct list packets;
3212 enum dpif_upcall_type upcall_type;
3215 struct flow_miss_op {
3216 struct dpif_op dpif_op;
3218 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3219 struct xlate_out xout;
3220 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3222 struct ofpbuf mask; /* Flow mask for "put" ops. */
3223 struct odputil_keybuf maskbuf;
3225 /* If this is a "put" op, then a pointer to the subfacet that should
3226 * be marked as uninstalled if the operation fails. */
3227 struct subfacet *subfacet;
3230 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3231 * OpenFlow controller as necessary according to their individual
3232 * configurations. */
3234 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3235 const struct flow *flow)
3237 struct ofputil_packet_in pin;
3239 pin.packet = packet->data;
3240 pin.packet_len = packet->size;
3241 pin.reason = OFPR_NO_MATCH;
3242 pin.controller_id = 0;
3247 pin.send_len = 0; /* not used for flow table misses */
3249 flow_get_metadata(flow, &pin.fmd);
3251 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3254 static struct flow_miss *
3255 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3256 const struct flow *flow, uint32_t hash)
3258 struct flow_miss *miss;
3260 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3261 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3269 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3270 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3271 * 'miss' is associated with a subfacet the caller must also initialize the
3272 * returned op->subfacet, and if anything needs to be freed after processing
3273 * the op, the caller must initialize op->garbage also. */
3275 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3276 struct flow_miss_op *op)
3278 if (miss->flow.in_port.ofp_port
3279 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3280 miss->flow.vlan_tci)) {
3281 /* This packet was received on a VLAN splinter port. We
3282 * added a VLAN to the packet to make the packet resemble
3283 * the flow, but the actions were composed assuming that
3284 * the packet contained no VLAN. So, we must remove the
3285 * VLAN header from the packet before trying to execute the
3287 eth_pop_vlan(packet);
3290 op->subfacet = NULL;
3291 op->xout_garbage = false;
3292 op->dpif_op.type = DPIF_OP_EXECUTE;
3293 op->dpif_op.u.execute.key = miss->key;
3294 op->dpif_op.u.execute.key_len = miss->key_len;
3295 op->dpif_op.u.execute.packet = packet;
3296 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3299 /* Helper for handle_flow_miss_without_facet() and
3300 * handle_flow_miss_with_facet(). */
3302 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3303 const struct flow *flow, bool fail_open)
3307 * Extra-special case for fail-open mode.
3309 * We are in fail-open mode and the packet matched the fail-open
3310 * rule, but we are connected to a controller too. We should send
3311 * the packet up to the controller in the hope that it will try to
3312 * set up a flow and thereby allow us to exit fail-open.
3314 * See the top-level comment in fail-open.c for more information.
3316 send_packet_in_miss(ofproto, packet, flow);
3320 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3321 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3322 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3323 * return value of true). However, for short flows the cost of bookkeeping is
3324 * much higher than the benefits, so when the datapath holds a large number of
3325 * flows we impose some heuristics to decide which flows are likely to be worth
3328 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3330 struct dpif_backer *backer = miss->ofproto->backer;
3333 switch (flow_miss_model) {
3334 case OFPROTO_HANDLE_MISS_AUTO:
3336 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3338 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3342 if (!backer->governor) {
3345 n_subfacets = hmap_count(&backer->subfacets);
3346 if (n_subfacets * 2 <= flow_eviction_threshold) {
3350 backer->governor = governor_create();
3353 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3354 return governor_should_install_flow(backer->governor, hash,
3355 list_size(&miss->packets));
3358 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3359 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3360 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3362 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3363 struct flow_miss *miss,
3364 struct flow_miss_op *ops, size_t *n_ops)
3366 struct ofpbuf *packet;
3368 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3370 COVERAGE_INC(facet_suppress);
3372 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3373 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3376 struct xlate_in xin;
3378 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3379 xlate_actions_for_side_effects(&xin);
3382 if (xout->odp_actions.size) {
3383 struct flow_miss_op *op = &ops[*n_ops];
3384 struct dpif_execute *execute = &op->dpif_op.u.execute;
3386 init_flow_miss_execute_op(miss, packet, op);
3387 xlate_out_copy(&op->xout, xout);
3388 execute->actions = op->xout.odp_actions.data;
3389 execute->actions_len = op->xout.odp_actions.size;
3390 op->xout_garbage = true;
3397 /* Handles 'miss', which matches 'facet'. May add any required datapath
3398 * operations to 'ops', incrementing '*n_ops' for each new op.
3400 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3401 * This is really important only for new facets: if we just called time_msec()
3402 * here, then the new subfacet or its packets could look (occasionally) as
3403 * though it was used some time after the facet was used. That can make a
3404 * one-packet flow look like it has a nonzero duration, which looks odd in
3405 * e.g. NetFlow statistics.
3407 * If non-null, 'stats' will be folded into 'facet'. */
3409 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3410 long long int now, struct dpif_flow_stats *stats,
3411 struct flow_miss_op *ops, size_t *n_ops)
3413 enum subfacet_path want_path;
3414 struct subfacet *subfacet;
3415 struct ofpbuf *packet;
3417 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3419 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3420 struct flow_miss_op *op = &ops[*n_ops];
3422 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3425 if (want_path != SF_FAST_PATH) {
3426 struct rule_dpif *rule;
3427 struct xlate_in xin;
3429 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3430 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3431 xlate_actions_for_side_effects(&xin);
3434 if (facet->xout.odp_actions.size) {
3435 struct dpif_execute *execute = &op->dpif_op.u.execute;
3437 init_flow_miss_execute_op(miss, packet, op);
3438 execute->actions = facet->xout.odp_actions.data,
3439 execute->actions_len = facet->xout.odp_actions.size;
3444 /* Don't install the flow if it's the result of the "userspace"
3445 * action for an already installed facet. This can occur when a
3446 * datapath flow with wildcards has a "userspace" action and flows
3447 * sent to userspace result in a different subfacet, which will then
3448 * be rejected as overlapping by the datapath. */
3449 if (miss->upcall_type == DPIF_UC_ACTION
3450 && !list_is_empty(&facet->subfacets)) {
3452 facet->used = MAX(facet->used, stats->used);
3453 facet->packet_count += stats->n_packets;
3454 facet->byte_count += stats->n_bytes;
3455 facet->tcp_flags |= stats->tcp_flags;
3460 subfacet = subfacet_create(facet, miss, now);
3462 subfacet_update_stats(subfacet, stats);
3465 if (subfacet->path != want_path) {
3466 struct flow_miss_op *op = &ops[(*n_ops)++];
3467 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3469 subfacet->path = want_path;
3471 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3472 if (enable_megaflows) {
3473 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3474 &miss->flow, UINT32_MAX);
3477 op->xout_garbage = false;
3478 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3479 op->subfacet = subfacet;
3480 put->flags = DPIF_FP_CREATE;
3481 put->key = miss->key;
3482 put->key_len = miss->key_len;
3483 put->mask = op->mask.data;
3484 put->mask_len = op->mask.size;
3486 if (want_path == SF_FAST_PATH) {
3487 put->actions = facet->xout.odp_actions.data;
3488 put->actions_len = facet->xout.odp_actions.size;
3490 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3491 op->slow_stub, sizeof op->slow_stub,
3492 &put->actions, &put->actions_len);
3498 /* Handles flow miss 'miss'. May add any required datapath operations
3499 * to 'ops', incrementing '*n_ops' for each new op. */
3501 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3504 struct ofproto_dpif *ofproto = miss->ofproto;
3505 struct dpif_flow_stats stats__;
3506 struct dpif_flow_stats *stats = &stats__;
3507 struct ofpbuf *packet;
3508 struct facet *facet;
3512 memset(stats, 0, sizeof *stats);
3514 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3515 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3516 stats->n_bytes += packet->size;
3520 facet = facet_lookup_valid(ofproto, &miss->flow);
3522 struct flow_wildcards wc;
3523 struct rule_dpif *rule;
3524 struct xlate_out xout;
3525 struct xlate_in xin;
3527 flow_wildcards_init_catchall(&wc);
3528 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3529 rule_credit_stats(rule, stats);
3531 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3533 xin.resubmit_stats = stats;
3534 xin.may_learn = true;
3535 xlate_actions(&xin, &xout);
3536 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3538 /* There does not exist a bijection between 'struct flow' and datapath
3539 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3540 * assumption used throughout the facet and subfacet handling code.
3541 * Since we have to handle these misses in userspace anyway, we simply
3542 * skip facet creation, avoiding the problem altogether. */
3543 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3544 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3545 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3549 facet = facet_create(miss, rule, &xout, stats);
3552 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3555 static struct drop_key *
3556 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3559 struct drop_key *drop_key;
3561 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3562 &backer->drop_keys) {
3563 if (drop_key->key_len == key_len
3564 && !memcmp(drop_key->key, key, key_len)) {
3572 drop_key_clear(struct dpif_backer *backer)
3574 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3575 struct drop_key *drop_key, *next;
3577 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3580 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3582 if (error && !VLOG_DROP_WARN(&rl)) {
3583 struct ds ds = DS_EMPTY_INITIALIZER;
3584 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3585 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3586 ovs_strerror(error), ds_cstr(&ds));
3590 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3591 free(drop_key->key);
3596 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3597 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3598 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3599 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3600 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3601 * 'packet' ingressed.
3603 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3604 * 'flow''s in_port to OFPP_NONE.
3606 * This function does post-processing on data returned from
3607 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3608 * of the upcall processing logic. In particular, if the extracted in_port is
3609 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3610 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3611 * a VLAN header onto 'packet' (if it is nonnull).
3613 * Similarly, this function also includes some logic to help with tunnels. It
3614 * may modify 'flow' as necessary to make the tunneling implementation
3615 * transparent to the upcall processing logic.
3617 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3618 * or some other positive errno if there are other problems. */
3620 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3621 const struct nlattr *key, size_t key_len,
3622 struct flow *flow, enum odp_key_fitness *fitnessp,
3623 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3625 const struct ofport_dpif *port;
3626 enum odp_key_fitness fitness;
3629 fitness = odp_flow_key_to_flow(key, key_len, flow);
3630 if (fitness == ODP_FIT_ERROR) {
3636 *odp_in_port = flow->in_port.odp_port;
3639 port = (tnl_port_should_receive(flow)
3640 ? tnl_port_receive(flow)
3641 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3642 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3647 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3648 * it's theoretically possible that we'll receive an ofport belonging to an
3649 * entirely different datapath. In practice, this can't happen because no
3650 * platforms has two separate datapaths which each support tunneling. */
3651 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3653 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3655 /* Make the packet resemble the flow, so that it gets sent to
3656 * an OpenFlow controller properly, so that it looks correct
3657 * for sFlow, and so that flow_extract() will get the correct
3658 * vlan_tci if it is called on 'packet'.
3660 * The allocated space inside 'packet' probably also contains
3661 * 'key', that is, both 'packet' and 'key' are probably part of
3662 * a struct dpif_upcall (see the large comment on that
3663 * structure definition), so pushing data on 'packet' is in
3664 * general not a good idea since it could overwrite 'key' or
3665 * free it as a side effect. However, it's OK in this special
3666 * case because we know that 'packet' is inside a Netlink
3667 * attribute: pushing 4 bytes will just overwrite the 4-byte
3668 * "struct nlattr", which is fine since we don't need that
3669 * header anymore. */
3670 eth_push_vlan(packet, flow->vlan_tci);
3672 /* We can't reproduce 'key' from 'flow'. */
3673 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3678 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3683 *fitnessp = fitness;
3689 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3692 struct dpif_upcall *upcall;
3693 struct flow_miss *miss;
3694 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3695 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3696 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3706 /* Construct the to-do list.
3708 * This just amounts to extracting the flow from each packet and sticking
3709 * the packets that have the same flow in the same "flow_miss" structure so
3710 * that we can process them together. */
3713 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3714 struct flow_miss *miss = &misses[n_misses];
3715 struct flow_miss *existing_miss;
3716 struct ofproto_dpif *ofproto;
3717 odp_port_t odp_in_port;
3722 error = ofproto_receive(backer, upcall->packet, upcall->key,
3723 upcall->key_len, &flow, &miss->key_fitness,
3724 &ofproto, &odp_in_port);
3725 if (error == ENODEV) {
3726 struct drop_key *drop_key;
3728 /* Received packet on datapath port for which we couldn't
3729 * associate an ofproto. This can happen if a port is removed
3730 * while traffic is being received. Print a rate-limited message
3731 * in case it happens frequently. Install a drop flow so
3732 * that future packets of the flow are inexpensively dropped
3734 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3735 "%"PRIu32, odp_in_port);
3737 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3740 ret = dpif_flow_put(backer->dpif,
3741 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3742 upcall->key, upcall->key_len,
3743 NULL, 0, NULL, 0, NULL);
3746 drop_key = xmalloc(sizeof *drop_key);
3747 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3748 drop_key->key_len = upcall->key_len;
3750 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3751 hash_bytes(drop_key->key, drop_key->key_len, 0));
3760 ofproto->n_missed++;
3761 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3762 &flow.tunnel, &flow.in_port, &miss->flow);
3764 /* Add other packets to a to-do list. */
3765 hash = flow_hash(&miss->flow, 0);
3766 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3767 if (!existing_miss) {
3768 hmap_insert(&todo, &miss->hmap_node, hash);
3769 miss->ofproto = ofproto;
3770 miss->key = upcall->key;
3771 miss->key_len = upcall->key_len;
3772 miss->upcall_type = upcall->type;
3773 list_init(&miss->packets);
3777 miss = existing_miss;
3779 list_push_back(&miss->packets, &upcall->packet->list_node);
3782 /* Process each element in the to-do list, constructing the set of
3783 * operations to batch. */
3785 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3786 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3788 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3790 /* Execute batch. */
3791 for (i = 0; i < n_ops; i++) {
3792 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3794 dpif_operate(backer->dpif, dpif_ops, n_ops);
3796 for (i = 0; i < n_ops; i++) {
3797 if (dpif_ops[i]->error != 0
3798 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3799 && flow_miss_ops[i].subfacet) {
3800 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3802 COVERAGE_INC(subfacet_install_fail);
3804 /* Zero-out subfacet counters when installation failed, but
3805 * datapath reported hits. This should not happen and
3806 * indicates a bug, since if the datapath flow exists, we
3807 * should not be attempting to create a new subfacet. A
3808 * buggy datapath could trigger this, so just zero out the
3809 * counters and log an error. */
3810 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3811 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3812 "datapath reported hits");
3813 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3816 subfacet->path = SF_NOT_INSTALLED;
3820 if (flow_miss_ops[i].xout_garbage) {
3821 xlate_out_uninit(&flow_miss_ops[i].xout);
3824 hmap_destroy(&todo);
3827 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3829 classify_upcall(const struct dpif_upcall *upcall)
3831 size_t userdata_len;
3832 union user_action_cookie cookie;
3834 /* First look at the upcall type. */
3835 switch (upcall->type) {
3836 case DPIF_UC_ACTION:
3842 case DPIF_N_UC_TYPES:
3844 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3848 /* "action" upcalls need a closer look. */
3849 if (!upcall->userdata) {
3850 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3853 userdata_len = nl_attr_get_size(upcall->userdata);
3854 if (userdata_len < sizeof cookie.type
3855 || userdata_len > sizeof cookie) {
3856 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3860 memset(&cookie, 0, sizeof cookie);
3861 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3862 if (userdata_len == sizeof cookie.sflow
3863 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3864 return SFLOW_UPCALL;
3865 } else if (userdata_len == sizeof cookie.slow_path
3866 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3868 } else if (userdata_len == sizeof cookie.flow_sample
3869 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3870 return FLOW_SAMPLE_UPCALL;
3871 } else if (userdata_len == sizeof cookie.ipfix
3872 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3873 return IPFIX_UPCALL;
3875 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3876 " and size %zu", cookie.type, userdata_len);
3882 handle_sflow_upcall(struct dpif_backer *backer,
3883 const struct dpif_upcall *upcall)
3885 struct ofproto_dpif *ofproto;
3886 union user_action_cookie cookie;
3888 odp_port_t odp_in_port;
3890 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3891 &flow, NULL, &ofproto, &odp_in_port)
3892 || !ofproto->sflow) {
3896 memset(&cookie, 0, sizeof cookie);
3897 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3898 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3899 odp_in_port, &cookie);
3903 handle_flow_sample_upcall(struct dpif_backer *backer,
3904 const struct dpif_upcall *upcall)
3906 struct ofproto_dpif *ofproto;
3907 union user_action_cookie cookie;
3910 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3911 &flow, NULL, &ofproto, NULL)
3912 || !ofproto->ipfix) {
3916 memset(&cookie, 0, sizeof cookie);
3917 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3919 /* The flow reflects exactly the contents of the packet. Sample
3920 * the packet using it. */
3921 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3922 cookie.flow_sample.collector_set_id,
3923 cookie.flow_sample.probability,
3924 cookie.flow_sample.obs_domain_id,
3925 cookie.flow_sample.obs_point_id);
3929 handle_ipfix_upcall(struct dpif_backer *backer,
3930 const struct dpif_upcall *upcall)
3932 struct ofproto_dpif *ofproto;
3935 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3936 &flow, NULL, &ofproto, NULL)
3937 || !ofproto->ipfix) {
3941 /* The flow reflects exactly the contents of the packet. Sample
3942 * the packet using it. */
3943 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3947 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3949 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3950 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3951 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3956 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3959 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3960 struct dpif_upcall *upcall = &misses[n_misses];
3961 struct ofpbuf *buf = &miss_bufs[n_misses];
3964 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3965 sizeof miss_buf_stubs[n_misses]);
3966 error = dpif_recv(backer->dpif, upcall, buf);
3972 switch (classify_upcall(upcall)) {
3974 /* Handle it later. */
3979 handle_sflow_upcall(backer, upcall);
3983 case FLOW_SAMPLE_UPCALL:
3984 handle_flow_sample_upcall(backer, upcall);
3989 handle_ipfix_upcall(backer, upcall);
3999 /* Handle deferred MISS_UPCALL processing. */
4000 handle_miss_upcalls(backer, misses, n_misses);
4001 for (i = 0; i < n_misses; i++) {
4002 ofpbuf_uninit(&miss_bufs[i]);
4008 /* Flow expiration. */
4010 static int subfacet_max_idle(const struct dpif_backer *);
4011 static void update_stats(struct dpif_backer *);
4012 static void rule_expire(struct rule_dpif *);
4013 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4015 /* This function is called periodically by run(). Its job is to collect
4016 * updates for the flows that have been installed into the datapath, most
4017 * importantly when they last were used, and then use that information to
4018 * expire flows that have not been used recently.
4020 * Returns the number of milliseconds after which it should be called again. */
4022 expire(struct dpif_backer *backer)
4024 struct ofproto_dpif *ofproto;
4028 /* Periodically clear out the drop keys in an effort to keep them
4029 * relatively few. */
4030 drop_key_clear(backer);
4032 /* Update stats for each flow in the backer. */
4033 update_stats(backer);
4035 n_subfacets = hmap_count(&backer->subfacets);
4037 struct subfacet *subfacet;
4038 long long int total, now;
4042 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4043 total += now - subfacet->created;
4045 backer->avg_subfacet_life += total / n_subfacets;
4047 backer->avg_subfacet_life /= 2;
4049 backer->avg_n_subfacet += n_subfacets;
4050 backer->avg_n_subfacet /= 2;
4052 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4054 max_idle = subfacet_max_idle(backer);
4055 expire_subfacets(backer, max_idle);
4057 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4058 struct rule *rule, *next_rule;
4060 if (ofproto->backer != backer) {
4064 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4066 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4067 &ofproto->up.expirable) {
4068 rule_expire(rule_dpif_cast(rule));
4071 /* All outstanding data in existing flows has been accounted, so it's a
4072 * good time to do bond rebalancing. */
4073 if (ofproto->has_bonded_bundles) {
4074 struct ofbundle *bundle;
4076 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4078 bond_rebalance(bundle->bond);
4084 return MIN(max_idle, 1000);
4087 /* Updates flow table statistics given that the datapath just reported 'stats'
4088 * as 'subfacet''s statistics. */
4090 update_subfacet_stats(struct subfacet *subfacet,
4091 const struct dpif_flow_stats *stats)
4093 struct facet *facet = subfacet->facet;
4094 struct dpif_flow_stats diff;
4096 diff.tcp_flags = stats->tcp_flags;
4097 diff.used = stats->used;
4099 if (stats->n_packets >= subfacet->dp_packet_count) {
4100 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4102 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4106 if (stats->n_bytes >= subfacet->dp_byte_count) {
4107 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4109 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4113 facet->ofproto->n_hit += diff.n_packets;
4114 subfacet->dp_packet_count = stats->n_packets;
4115 subfacet->dp_byte_count = stats->n_bytes;
4116 subfacet_update_stats(subfacet, &diff);
4118 if (facet->accounted_bytes < facet->byte_count) {
4120 facet_account(facet);
4121 facet->accounted_bytes = facet->byte_count;
4125 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4126 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4128 delete_unexpected_flow(struct dpif_backer *backer,
4129 const struct nlattr *key, size_t key_len)
4131 if (!VLOG_DROP_WARN(&rl)) {
4135 odp_flow_key_format(key, key_len, &s);
4136 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4140 COVERAGE_INC(facet_unexpected);
4141 dpif_flow_del(backer->dpif, key, key_len, NULL);
4144 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4146 * This function also pushes statistics updates to rules which each facet
4147 * resubmits into. Generally these statistics will be accurate. However, if a
4148 * facet changes the rule it resubmits into at some time in between
4149 * update_stats() runs, it is possible that statistics accrued to the
4150 * old rule will be incorrectly attributed to the new rule. This could be
4151 * avoided by calling update_stats() whenever rules are created or
4152 * deleted. However, the performance impact of making so many calls to the
4153 * datapath do not justify the benefit of having perfectly accurate statistics.
4155 * In addition, this function maintains per ofproto flow hit counts. The patch
4156 * port is not treated specially. e.g. A packet ingress from br0 patched into
4157 * br1 will increase the hit count of br0 by 1, however, does not affect
4158 * the hit or miss counts of br1.
4161 update_stats(struct dpif_backer *backer)
4163 const struct dpif_flow_stats *stats;
4164 struct dpif_flow_dump dump;
4165 const struct nlattr *key, *mask;
4166 size_t key_len, mask_len;
4168 dpif_flow_dump_start(&dump, backer->dpif);
4169 while (dpif_flow_dump_next(&dump, &key, &key_len,
4170 &mask, &mask_len, NULL, NULL, &stats)) {
4171 struct subfacet *subfacet;
4174 key_hash = odp_flow_key_hash(key, key_len);
4175 subfacet = subfacet_find(backer, key, key_len, key_hash);
4176 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4178 update_subfacet_stats(subfacet, stats);
4182 /* Stats are updated per-packet. */
4185 case SF_NOT_INSTALLED:
4187 delete_unexpected_flow(backer, key, key_len);
4192 dpif_flow_dump_done(&dump);
4194 update_moving_averages(backer);
4197 /* Calculates and returns the number of milliseconds of idle time after which
4198 * subfacets should expire from the datapath. When a subfacet expires, we fold
4199 * its statistics into its facet, and when a facet's last subfacet expires, we
4200 * fold its statistic into its rule. */
4202 subfacet_max_idle(const struct dpif_backer *backer)
4205 * Idle time histogram.
4207 * Most of the time a switch has a relatively small number of subfacets.
4208 * When this is the case we might as well keep statistics for all of them
4209 * in userspace and to cache them in the kernel datapath for performance as
4212 * As the number of subfacets increases, the memory required to maintain
4213 * statistics about them in userspace and in the kernel becomes
4214 * significant. However, with a large number of subfacets it is likely
4215 * that only a few of them are "heavy hitters" that consume a large amount
4216 * of bandwidth. At this point, only heavy hitters are worth caching in
4217 * the kernel and maintaining in userspaces; other subfacets we can
4220 * The technique used to compute the idle time is to build a histogram with
4221 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4222 * that is installed in the kernel gets dropped in the appropriate bucket.
4223 * After the histogram has been built, we compute the cutoff so that only
4224 * the most-recently-used 1% of subfacets (but at least
4225 * flow_eviction_threshold flows) are kept cached. At least
4226 * the most-recently-used bucket of subfacets is kept, so actually an
4227 * arbitrary number of subfacets can be kept in any given expiration run
4228 * (though the next run will delete most of those unless they receive
4231 * This requires a second pass through the subfacets, in addition to the
4232 * pass made by update_stats(), because the former function never looks at
4233 * uninstallable subfacets.
4235 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4236 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4237 int buckets[N_BUCKETS] = { 0 };
4238 int total, subtotal, bucket;
4239 struct subfacet *subfacet;
4243 total = hmap_count(&backer->subfacets);
4244 if (total <= flow_eviction_threshold) {
4245 return N_BUCKETS * BUCKET_WIDTH;
4248 /* Build histogram. */
4250 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4251 long long int idle = now - subfacet->used;
4252 int bucket = (idle <= 0 ? 0
4253 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4254 : (unsigned int) idle / BUCKET_WIDTH);
4258 /* Find the first bucket whose flows should be expired. */
4259 subtotal = bucket = 0;
4261 subtotal += buckets[bucket++];
4262 } while (bucket < N_BUCKETS &&
4263 subtotal < MAX(flow_eviction_threshold, total / 100));
4265 if (VLOG_IS_DBG_ENABLED()) {
4269 ds_put_cstr(&s, "keep");
4270 for (i = 0; i < N_BUCKETS; i++) {
4272 ds_put_cstr(&s, ", drop");
4275 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4278 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4282 return bucket * BUCKET_WIDTH;
4286 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4288 /* Cutoff time for most flows. */
4289 long long int normal_cutoff = time_msec() - dp_max_idle;
4291 /* We really want to keep flows for special protocols around, so use a more
4292 * conservative cutoff. */
4293 long long int special_cutoff = time_msec() - 10000;
4295 struct subfacet *subfacet, *next_subfacet;
4296 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4300 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4301 &backer->subfacets) {
4302 long long int cutoff;
4304 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4308 if (subfacet->used < cutoff) {
4309 if (subfacet->path != SF_NOT_INSTALLED) {
4310 batch[n_batch++] = subfacet;
4311 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4312 subfacet_destroy_batch(backer, batch, n_batch);
4316 subfacet_destroy(subfacet);
4322 subfacet_destroy_batch(backer, batch, n_batch);
4326 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4327 * then delete it entirely. */
4329 rule_expire(struct rule_dpif *rule)
4334 if (rule->up.pending) {
4335 /* We'll have to expire it later. */
4339 /* Has 'rule' expired? */
4341 if (rule->up.hard_timeout
4342 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4343 reason = OFPRR_HARD_TIMEOUT;
4344 } else if (rule->up.idle_timeout
4345 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4346 reason = OFPRR_IDLE_TIMEOUT;
4351 COVERAGE_INC(ofproto_dpif_expired);
4353 /* Get rid of the rule. */
4354 ofproto_rule_expire(&rule->up, reason);
4359 /* Creates and returns a new facet based on 'miss'.
4361 * The caller must already have determined that no facet with an identical
4362 * 'miss->flow' exists in 'miss->ofproto'.
4364 * 'rule' and 'xout' must have been created based on 'miss'.
4366 * 'facet'' statistics are initialized based on 'stats'.
4368 * The facet will initially have no subfacets. The caller should create (at
4369 * least) one subfacet with subfacet_create(). */
4370 static struct facet *
4371 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4372 struct xlate_out *xout, struct dpif_flow_stats *stats)
4374 struct ofproto_dpif *ofproto = miss->ofproto;
4375 struct facet *facet;
4378 facet = xzalloc(sizeof *facet);
4379 facet->ofproto = miss->ofproto;
4380 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4381 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4382 facet->tcp_flags = stats->tcp_flags;
4383 facet->used = stats->used;
4384 facet->flow = miss->flow;
4385 facet->learn_rl = time_msec() + 500;
4387 list_init(&facet->subfacets);
4388 netflow_flow_init(&facet->nf_flow);
4389 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4391 xlate_out_copy(&facet->xout, xout);
4393 match_init(&match, &facet->flow, &facet->xout.wc);
4394 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4395 classifier_insert(&ofproto->facets, &facet->cr);
4397 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4398 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4404 facet_free(struct facet *facet)
4407 xlate_out_uninit(&facet->xout);
4412 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4413 * 'packet', which arrived on 'in_port'. */
4415 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4416 const struct nlattr *odp_actions, size_t actions_len,
4417 struct ofpbuf *packet)
4419 struct odputil_keybuf keybuf;
4423 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4424 odp_flow_key_from_flow(&key, flow,
4425 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4427 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4428 odp_actions, actions_len, packet);
4432 /* Remove 'facet' from its ofproto and free up the associated memory:
4434 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4435 * rule's statistics, via subfacet_uninstall().
4437 * - Removes 'facet' from its rule and from ofproto->facets.
4440 facet_remove(struct facet *facet)
4442 struct subfacet *subfacet, *next_subfacet;
4444 ovs_assert(!list_is_empty(&facet->subfacets));
4446 /* First uninstall all of the subfacets to get final statistics. */
4447 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4448 subfacet_uninstall(subfacet);
4451 /* Flush the final stats to the rule.
4453 * This might require us to have at least one subfacet around so that we
4454 * can use its actions for accounting in facet_account(), which is why we
4455 * have uninstalled but not yet destroyed the subfacets. */
4456 facet_flush_stats(facet);
4458 /* Now we're really all done so destroy everything. */
4459 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4460 &facet->subfacets) {
4461 subfacet_destroy__(subfacet);
4463 classifier_remove(&facet->ofproto->facets, &facet->cr);
4464 cls_rule_destroy(&facet->cr);
4468 /* Feed information from 'facet' back into the learning table to keep it in
4469 * sync with what is actually flowing through the datapath. */
4471 facet_learn(struct facet *facet)
4473 long long int now = time_msec();
4475 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4479 facet->learn_rl = now + 500;
4481 if (!facet->xout.has_learn
4482 && !facet->xout.has_normal
4483 && (!facet->xout.has_fin_timeout
4484 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4488 facet_push_stats(facet, true);
4492 facet_account(struct facet *facet)
4494 const struct nlattr *a;
4499 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4502 n_bytes = facet->byte_count - facet->accounted_bytes;
4504 /* This loop feeds byte counters to bond_account() for rebalancing to use
4505 * as a basis. We also need to track the actual VLAN on which the packet
4506 * is going to be sent to ensure that it matches the one passed to
4507 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4510 * We use the actions from an arbitrary subfacet because they should all
4511 * be equally valid for our purpose. */
4512 vlan_tci = facet->flow.vlan_tci;
4513 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4514 facet->xout.odp_actions.size) {
4515 const struct ovs_action_push_vlan *vlan;
4516 struct ofport_dpif *port;
4518 switch (nl_attr_type(a)) {
4519 case OVS_ACTION_ATTR_OUTPUT:
4520 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4521 if (port && port->bundle && port->bundle->bond) {
4522 bond_account(port->bundle->bond, &facet->flow,
4523 vlan_tci_to_vid(vlan_tci), n_bytes);
4527 case OVS_ACTION_ATTR_POP_VLAN:
4528 vlan_tci = htons(0);
4531 case OVS_ACTION_ATTR_PUSH_VLAN:
4532 vlan = nl_attr_get(a);
4533 vlan_tci = vlan->vlan_tci;
4539 /* Returns true if the only action for 'facet' is to send to the controller.
4540 * (We don't report NetFlow expiration messages for such facets because they
4541 * are just part of the control logic for the network, not real traffic). */
4543 facet_is_controller_flow(struct facet *facet)
4546 struct ofproto_dpif *ofproto = facet->ofproto;
4547 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4549 const struct ofpact *ofpacts = rule->up.ofpacts;
4550 size_t ofpacts_len = rule->up.ofpacts_len;
4552 if (ofpacts_len > 0 &&
4553 ofpacts->type == OFPACT_CONTROLLER &&
4554 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4561 /* Folds all of 'facet''s statistics into its rule. Also updates the
4562 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4563 * 'facet''s statistics in the datapath should have been zeroed and folded into
4564 * its packet and byte counts before this function is called. */
4566 facet_flush_stats(struct facet *facet)
4568 struct ofproto_dpif *ofproto = facet->ofproto;
4569 struct subfacet *subfacet;
4571 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4572 ovs_assert(!subfacet->dp_byte_count);
4573 ovs_assert(!subfacet->dp_packet_count);
4576 facet_push_stats(facet, false);
4577 if (facet->accounted_bytes < facet->byte_count) {
4578 facet_account(facet);
4579 facet->accounted_bytes = facet->byte_count;
4582 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4583 struct ofexpired expired;
4584 expired.flow = facet->flow;
4585 expired.packet_count = facet->packet_count;
4586 expired.byte_count = facet->byte_count;
4587 expired.used = facet->used;
4588 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4591 /* Reset counters to prevent double counting if 'facet' ever gets
4593 facet_reset_counters(facet);
4595 netflow_flow_clear(&facet->nf_flow);
4596 facet->tcp_flags = 0;
4599 /* Searches 'ofproto''s table of facets for one which would be responsible for
4600 * 'flow'. Returns it if found, otherwise a null pointer.
4602 * The returned facet might need revalidation; use facet_lookup_valid()
4603 * instead if that is important. */
4604 static struct facet *
4605 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4607 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4608 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4611 /* Searches 'ofproto''s table of facets for one capable that covers
4612 * 'flow'. Returns it if found, otherwise a null pointer.
4614 * The returned facet is guaranteed to be valid. */
4615 static struct facet *
4616 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4618 struct facet *facet;
4620 facet = facet_find(ofproto, flow);
4622 && ofproto->backer->need_revalidate
4623 && !facet_revalidate(facet)) {
4631 facet_check_consistency(struct facet *facet)
4633 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4635 struct xlate_out xout;
4636 struct xlate_in xin;
4638 struct rule_dpif *rule;
4641 /* Check the datapath actions for consistency. */
4642 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4643 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4644 xlate_actions(&xin, &xout);
4646 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4647 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4648 && facet->xout.slow == xout.slow
4649 && facet->fail_open == fail_open;
4650 if (!ok && !VLOG_DROP_WARN(&rl)) {
4651 struct ds s = DS_EMPTY_INITIALIZER;
4653 flow_format(&s, &facet->flow);
4654 ds_put_cstr(&s, ": inconsistency in facet");
4656 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4657 ds_put_cstr(&s, " (actions were: ");
4658 format_odp_actions(&s, facet->xout.odp_actions.data,
4659 facet->xout.odp_actions.size);
4660 ds_put_cstr(&s, ") (correct actions: ");
4661 format_odp_actions(&s, xout.odp_actions.data,
4662 xout.odp_actions.size);
4663 ds_put_char(&s, ')');
4666 if (facet->xout.slow != xout.slow) {
4667 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4670 if (facet->fail_open != fail_open) {
4671 ds_put_format(&s, " fail open incorrect. should be %s",
4672 fail_open ? "true" : "false");
4676 xlate_out_uninit(&xout);
4681 /* Re-searches the classifier for 'facet':
4683 * - If the rule found is different from 'facet''s current rule, moves
4684 * 'facet' to the new rule and recompiles its actions.
4686 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4687 * where it is and recompiles its actions anyway.
4689 * - If any of 'facet''s subfacets correspond to a new flow according to
4690 * ofproto_receive(), 'facet' is removed.
4692 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4694 facet_revalidate(struct facet *facet)
4696 struct ofproto_dpif *ofproto = facet->ofproto;
4697 struct rule_dpif *new_rule;
4698 struct subfacet *subfacet;
4699 struct flow_wildcards wc;
4700 struct xlate_out xout;
4701 struct xlate_in xin;
4703 COVERAGE_INC(facet_revalidate);
4705 /* Check that child subfacets still correspond to this facet. Tunnel
4706 * configuration changes could cause a subfacet's OpenFlow in_port to
4708 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4709 struct ofproto_dpif *recv_ofproto;
4710 struct flow recv_flow;
4713 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4714 subfacet->key_len, &recv_flow, NULL,
4715 &recv_ofproto, NULL);
4717 || recv_ofproto != ofproto
4718 || facet != facet_find(ofproto, &recv_flow)) {
4719 facet_remove(facet);
4724 flow_wildcards_init_catchall(&wc);
4725 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4727 /* Calculate new datapath actions.
4729 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4730 * emit a NetFlow expiration and, if so, we need to have the old state
4731 * around to properly compose it. */
4732 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4733 xlate_actions(&xin, &xout);
4734 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4736 /* A facet's slow path reason should only change under dramatic
4737 * circumstances. Rather than try to update everything, it's simpler to
4738 * remove the facet and start over.
4740 * More importantly, if a facet's wildcards change, it will be relatively
4741 * difficult to figure out if its subfacets still belong to it, and if not
4742 * which facet they may belong to. Again, to avoid the complexity, we
4743 * simply give up instead. */
4744 if (facet->xout.slow != xout.slow
4745 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4746 facet_remove(facet);
4747 xlate_out_uninit(&xout);
4751 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4752 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4753 if (subfacet->path == SF_FAST_PATH) {
4754 struct dpif_flow_stats stats;
4756 subfacet_install(subfacet, &xout.odp_actions, &stats);
4757 subfacet_update_stats(subfacet, &stats);
4761 facet_flush_stats(facet);
4763 ofpbuf_clear(&facet->xout.odp_actions);
4764 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4765 xout.odp_actions.size);
4768 /* Update 'facet' now that we've taken care of all the old state. */
4769 facet->xout.slow = xout.slow;
4770 facet->xout.has_learn = xout.has_learn;
4771 facet->xout.has_normal = xout.has_normal;
4772 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4773 facet->xout.nf_output_iface = xout.nf_output_iface;
4774 facet->xout.mirrors = xout.mirrors;
4775 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4776 facet->used = MAX(facet->used, new_rule->up.created);
4777 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4779 xlate_out_uninit(&xout);
4784 facet_reset_counters(struct facet *facet)
4786 facet->packet_count = 0;
4787 facet->byte_count = 0;
4788 facet->prev_packet_count = 0;
4789 facet->prev_byte_count = 0;
4790 facet->accounted_bytes = 0;
4794 facet_push_stats(struct facet *facet, bool may_learn)
4796 struct dpif_flow_stats stats;
4798 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4799 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4800 ovs_assert(facet->used >= facet->prev_used);
4802 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4803 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4804 stats.used = facet->used;
4805 stats.tcp_flags = facet->tcp_flags;
4807 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4808 struct ofproto_dpif *ofproto = facet->ofproto;
4809 struct ofport_dpif *in_port;
4810 struct rule_dpif *rule;
4811 struct xlate_in xin;
4813 facet->prev_packet_count = facet->packet_count;
4814 facet->prev_byte_count = facet->byte_count;
4815 facet->prev_used = facet->used;
4817 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4818 if (in_port && in_port->is_tunnel) {
4819 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4822 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4823 rule_credit_stats(rule, &stats);
4824 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4826 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4827 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4828 stats.n_packets, stats.n_bytes);
4830 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4832 xin.resubmit_stats = &stats;
4833 xin.may_learn = may_learn;
4834 xlate_actions_for_side_effects(&xin);
4839 push_all_stats__(bool run_fast)
4841 static long long int rl = LLONG_MIN;
4842 struct ofproto_dpif *ofproto;
4844 if (time_msec() < rl) {
4848 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4849 struct cls_cursor cursor;
4850 struct facet *facet;
4852 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4853 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4854 facet_push_stats(facet, false);
4861 rl = time_msec() + 100;
4865 push_all_stats(void)
4867 push_all_stats__(true);
4871 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4873 rule->packet_count += stats->n_packets;
4874 rule->byte_count += stats->n_bytes;
4875 ofproto_rule_update_used(&rule->up, stats->used);
4880 static struct subfacet *
4881 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4882 size_t key_len, uint32_t key_hash)
4884 struct subfacet *subfacet;
4886 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4887 &backer->subfacets) {
4888 if (subfacet->key_len == key_len
4889 && !memcmp(key, subfacet->key, key_len)) {
4897 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4898 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4899 * existing subfacet if there is one, otherwise creates and returns a
4901 static struct subfacet *
4902 subfacet_create(struct facet *facet, struct flow_miss *miss,
4905 struct dpif_backer *backer = miss->ofproto->backer;
4906 enum odp_key_fitness key_fitness = miss->key_fitness;
4907 const struct nlattr *key = miss->key;
4908 size_t key_len = miss->key_len;
4910 struct subfacet *subfacet;
4912 key_hash = odp_flow_key_hash(key, key_len);
4914 if (list_is_empty(&facet->subfacets)) {
4915 subfacet = &facet->one_subfacet;
4917 subfacet = subfacet_find(backer, key, key_len, key_hash);
4919 if (subfacet->facet == facet) {
4923 /* This shouldn't happen. */
4924 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4925 subfacet_destroy(subfacet);
4928 subfacet = xmalloc(sizeof *subfacet);
4931 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4932 list_push_back(&facet->subfacets, &subfacet->list_node);
4933 subfacet->facet = facet;
4934 subfacet->key_fitness = key_fitness;
4935 subfacet->key = xmemdup(key, key_len);
4936 subfacet->key_len = key_len;
4937 subfacet->used = now;
4938 subfacet->created = now;
4939 subfacet->dp_packet_count = 0;
4940 subfacet->dp_byte_count = 0;
4941 subfacet->path = SF_NOT_INSTALLED;
4942 subfacet->backer = backer;
4944 backer->subfacet_add_count++;
4948 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4949 * its facet within 'ofproto', and frees it. */
4951 subfacet_destroy__(struct subfacet *subfacet)
4953 struct facet *facet = subfacet->facet;
4954 struct ofproto_dpif *ofproto = facet->ofproto;
4956 /* Update ofproto stats before uninstall the subfacet. */
4957 ofproto->backer->subfacet_del_count++;
4959 subfacet_uninstall(subfacet);
4960 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4961 list_remove(&subfacet->list_node);
4962 free(subfacet->key);
4963 if (subfacet != &facet->one_subfacet) {
4968 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4969 * last remaining subfacet in its facet destroys the facet too. */
4971 subfacet_destroy(struct subfacet *subfacet)
4973 struct facet *facet = subfacet->facet;
4975 if (list_is_singleton(&facet->subfacets)) {
4976 /* facet_remove() needs at least one subfacet (it will remove it). */
4977 facet_remove(facet);
4979 subfacet_destroy__(subfacet);
4984 subfacet_destroy_batch(struct dpif_backer *backer,
4985 struct subfacet **subfacets, int n)
4987 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4988 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4989 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4992 for (i = 0; i < n; i++) {
4993 ops[i].type = DPIF_OP_FLOW_DEL;
4994 ops[i].u.flow_del.key = subfacets[i]->key;
4995 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4996 ops[i].u.flow_del.stats = &stats[i];
5000 dpif_operate(backer->dpif, opsp, n);
5001 for (i = 0; i < n; i++) {
5002 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5003 subfacets[i]->path = SF_NOT_INSTALLED;
5004 subfacet_destroy(subfacets[i]);
5009 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5010 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5011 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5012 * since 'subfacet' was last updated.
5014 * Returns 0 if successful, otherwise a positive errno value. */
5016 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5017 struct dpif_flow_stats *stats)
5019 struct facet *facet = subfacet->facet;
5020 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5021 const struct nlattr *actions = odp_actions->data;
5022 size_t actions_len = odp_actions->size;
5023 struct odputil_keybuf maskbuf;
5026 uint64_t slow_path_stub[128 / 8];
5027 enum dpif_flow_put_flags flags;
5030 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5033 flags |= DPIF_FP_ZERO_STATS;
5036 if (path == SF_SLOW_PATH) {
5037 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5038 slow_path_stub, sizeof slow_path_stub,
5039 &actions, &actions_len);
5042 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5043 if (enable_megaflows) {
5044 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5045 &facet->flow, UINT32_MAX);
5048 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5049 subfacet->key_len, mask.data, mask.size,
5050 actions, actions_len, stats);
5053 subfacet_reset_dp_stats(subfacet, stats);
5057 COVERAGE_INC(subfacet_install_fail);
5059 subfacet->path = path;
5064 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5066 subfacet_uninstall(struct subfacet *subfacet)
5068 if (subfacet->path != SF_NOT_INSTALLED) {
5069 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5070 struct dpif_flow_stats stats;
5073 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5074 subfacet->key_len, &stats);
5075 subfacet_reset_dp_stats(subfacet, &stats);
5077 subfacet_update_stats(subfacet, &stats);
5079 subfacet->path = SF_NOT_INSTALLED;
5081 ovs_assert(subfacet->dp_packet_count == 0);
5082 ovs_assert(subfacet->dp_byte_count == 0);
5086 /* Resets 'subfacet''s datapath statistics counters. This should be called
5087 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5088 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5089 * was reset in the datapath. 'stats' will be modified to include only
5090 * statistics new since 'subfacet' was last updated. */
5092 subfacet_reset_dp_stats(struct subfacet *subfacet,
5093 struct dpif_flow_stats *stats)
5096 && subfacet->dp_packet_count <= stats->n_packets
5097 && subfacet->dp_byte_count <= stats->n_bytes) {
5098 stats->n_packets -= subfacet->dp_packet_count;
5099 stats->n_bytes -= subfacet->dp_byte_count;
5102 subfacet->dp_packet_count = 0;
5103 subfacet->dp_byte_count = 0;
5106 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5108 * Because of the meaning of a subfacet's counters, it only makes sense to do
5109 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5110 * represents a packet that was sent by hand or if it represents statistics
5111 * that have been cleared out of the datapath. */
5113 subfacet_update_stats(struct subfacet *subfacet,
5114 const struct dpif_flow_stats *stats)
5116 if (stats->n_packets || stats->used > subfacet->used) {
5117 struct facet *facet = subfacet->facet;
5119 subfacet->used = MAX(subfacet->used, stats->used);
5120 facet->used = MAX(facet->used, stats->used);
5121 facet->packet_count += stats->n_packets;
5122 facet->byte_count += stats->n_bytes;
5123 facet->tcp_flags |= stats->tcp_flags;
5129 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5130 * the fields that were relevant as part of the lookup. */
5131 static struct rule_dpif *
5132 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5133 struct flow_wildcards *wc)
5135 struct rule_dpif *rule;
5137 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5142 return rule_dpif_miss_rule(ofproto, flow);
5146 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5147 const struct flow *flow, struct flow_wildcards *wc,
5150 struct cls_rule *cls_rule;
5151 struct classifier *cls;
5154 if (table_id >= N_TABLES) {
5159 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5160 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5163 cls = &ofproto->up.tables[table_id].cls;
5164 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5165 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5166 /* We must pretend that transport ports are unavailable. */
5167 struct flow ofpc_normal_flow = *flow;
5168 ofpc_normal_flow.tp_src = htons(0);
5169 ofpc_normal_flow.tp_dst = htons(0);
5170 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5171 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5172 cls_rule = &ofproto->drop_frags_rule->up.cr;
5174 flow_wildcards_init_exact(wc);
5177 cls_rule = classifier_lookup(cls, flow, wc);
5179 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5183 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5185 struct ofport_dpif *port;
5187 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5189 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5190 flow->in_port.ofp_port);
5191 return ofproto->miss_rule;
5194 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5195 return ofproto->no_packet_in_rule;
5197 return ofproto->miss_rule;
5201 complete_operation(struct rule_dpif *rule)
5203 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5205 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5207 struct dpif_completion *c = xmalloc(sizeof *c);
5208 c->op = rule->up.pending;
5209 list_push_back(&ofproto->completions, &c->list_node);
5211 ofoperation_complete(rule->up.pending, 0);
5215 static struct rule *
5218 struct rule_dpif *rule = xmalloc(sizeof *rule);
5223 rule_dealloc(struct rule *rule_)
5225 struct rule_dpif *rule = rule_dpif_cast(rule_);
5230 rule_construct(struct rule *rule_)
5232 struct rule_dpif *rule = rule_dpif_cast(rule_);
5233 rule->packet_count = 0;
5234 rule->byte_count = 0;
5235 complete_operation(rule);
5240 rule_destruct(struct rule *rule)
5242 complete_operation(rule_dpif_cast(rule));
5246 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5248 struct rule_dpif *rule = rule_dpif_cast(rule_);
5250 /* push_all_stats() can handle flow misses which, when using the learn
5251 * action, can cause rules to be added and deleted. This can corrupt our
5252 * caller's datastructures which assume that rule_get_stats() doesn't have
5253 * an impact on the flow table. To be safe, we disable miss handling. */
5254 push_all_stats__(false);
5256 /* Start from historical data for 'rule' itself that are no longer tracked
5257 * in facets. This counts, for example, facets that have expired. */
5258 *packets = rule->packet_count;
5259 *bytes = rule->byte_count;
5263 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5264 struct ofpbuf *packet)
5266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5267 struct dpif_flow_stats stats;
5268 struct xlate_out xout;
5269 struct xlate_in xin;
5271 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5272 rule_credit_stats(rule, &stats);
5274 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5275 xin.resubmit_stats = &stats;
5276 xlate_actions(&xin, &xout);
5278 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5279 xout.odp_actions.size, packet);
5281 xlate_out_uninit(&xout);
5285 rule_execute(struct rule *rule, const struct flow *flow,
5286 struct ofpbuf *packet)
5288 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5289 ofpbuf_delete(packet);
5294 rule_modify_actions(struct rule *rule_)
5296 struct rule_dpif *rule = rule_dpif_cast(rule_);
5298 complete_operation(rule);
5301 /* Sends 'packet' out 'ofport'.
5302 * May modify 'packet'.
5303 * Returns 0 if successful, otherwise a positive errno value. */
5305 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5308 uint64_t odp_actions_stub[1024 / 8];
5309 struct ofpbuf key, odp_actions;
5310 struct dpif_flow_stats stats;
5311 struct odputil_keybuf keybuf;
5312 struct ofpact_output output;
5313 struct xlate_out xout;
5314 struct xlate_in xin;
5316 union flow_in_port in_port_;
5319 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5320 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5322 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5323 in_port_.ofp_port = OFPP_NONE;
5324 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5325 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5327 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5329 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5330 output.port = ofport->up.ofp_port;
5333 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5334 xin.ofpacts_len = sizeof output;
5335 xin.ofpacts = &output.ofpact;
5336 xin.resubmit_stats = &stats;
5337 xlate_actions(&xin, &xout);
5339 error = dpif_execute(ofproto->backer->dpif,
5341 xout.odp_actions.data, xout.odp_actions.size,
5343 xlate_out_uninit(&xout);
5346 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5347 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5348 ovs_strerror(error));
5351 ofproto->stats.tx_packets++;
5352 ofproto->stats.tx_bytes += packet->size;
5356 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5357 * The action will state 'slow' as the reason that the action is in the slow
5358 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5359 * dump-flows" output to see why a flow is in the slow path.)
5361 * The 'stub_size' bytes in 'stub' will be used to store the action.
5362 * 'stub_size' must be large enough for the action.
5364 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5367 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5368 enum slow_path_reason slow,
5369 uint64_t *stub, size_t stub_size,
5370 const struct nlattr **actionsp, size_t *actions_lenp)
5372 union user_action_cookie cookie;
5375 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5376 cookie.slow_path.unused = 0;
5377 cookie.slow_path.reason = slow;
5379 ofpbuf_use_stack(&buf, stub, stub_size);
5380 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5381 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5383 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5385 put_userspace_action(ofproto, &buf, flow, &cookie,
5386 sizeof cookie.slow_path);
5388 *actionsp = buf.data;
5389 *actions_lenp = buf.size;
5393 put_userspace_action(const struct ofproto_dpif *ofproto,
5394 struct ofpbuf *odp_actions,
5395 const struct flow *flow,
5396 const union user_action_cookie *cookie,
5397 const size_t cookie_size)
5401 pid = dpif_port_get_pid(ofproto->backer->dpif,
5402 ofp_port_to_odp_port(ofproto,
5403 flow->in_port.ofp_port));
5405 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5409 set_frag_handling(struct ofproto *ofproto_,
5410 enum ofp_config_flags frag_handling)
5412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5413 if (frag_handling != OFPC_FRAG_REASM) {
5414 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5422 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5423 const struct flow *flow,
5424 const struct ofpact *ofpacts, size_t ofpacts_len)
5426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5427 struct odputil_keybuf keybuf;
5428 struct dpif_flow_stats stats;
5429 struct xlate_out xout;
5430 struct xlate_in xin;
5434 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5435 odp_flow_key_from_flow(&key, flow,
5436 ofp_port_to_odp_port(ofproto,
5437 flow->in_port.ofp_port));
5439 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5441 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5442 xin.resubmit_stats = &stats;
5443 xin.ofpacts_len = ofpacts_len;
5444 xin.ofpacts = ofpacts;
5446 xlate_actions(&xin, &xout);
5447 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5448 xout.odp_actions.data, xout.odp_actions.size, packet);
5449 xlate_out_uninit(&xout);
5457 set_netflow(struct ofproto *ofproto_,
5458 const struct netflow_options *netflow_options)
5460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5462 if (netflow_options) {
5463 if (!ofproto->netflow) {
5464 ofproto->netflow = netflow_create();
5465 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5467 return netflow_set_options(ofproto->netflow, netflow_options);
5468 } else if (ofproto->netflow) {
5469 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5470 netflow_destroy(ofproto->netflow);
5471 ofproto->netflow = NULL;
5478 get_netflow_ids(const struct ofproto *ofproto_,
5479 uint8_t *engine_type, uint8_t *engine_id)
5481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5483 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5487 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5489 if (!facet_is_controller_flow(facet) &&
5490 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5491 struct subfacet *subfacet;
5492 struct ofexpired expired;
5494 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5495 if (subfacet->path == SF_FAST_PATH) {
5496 struct dpif_flow_stats stats;
5498 subfacet_install(subfacet, &facet->xout.odp_actions,
5500 subfacet_update_stats(subfacet, &stats);
5504 expired.flow = facet->flow;
5505 expired.packet_count = facet->packet_count;
5506 expired.byte_count = facet->byte_count;
5507 expired.used = facet->used;
5508 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5513 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5515 struct cls_cursor cursor;
5516 struct facet *facet;
5518 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5519 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5520 send_active_timeout(ofproto, facet);
5524 static struct ofproto_dpif *
5525 ofproto_dpif_lookup(const char *name)
5527 struct ofproto_dpif *ofproto;
5529 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5530 hash_string(name, 0), &all_ofproto_dpifs) {
5531 if (!strcmp(ofproto->up.name, name)) {
5539 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5540 const char *argv[], void *aux OVS_UNUSED)
5542 struct ofproto_dpif *ofproto;
5545 ofproto = ofproto_dpif_lookup(argv[1]);
5547 unixctl_command_reply_error(conn, "no such bridge");
5550 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5551 mac_learning_flush(ofproto->ml);
5552 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5554 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5555 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5556 mac_learning_flush(ofproto->ml);
5557 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5561 unixctl_command_reply(conn, "table successfully flushed");
5564 static struct ofport_dpif *
5565 ofbundle_get_a_port(const struct ofbundle *bundle)
5567 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5572 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5573 const char *argv[], void *aux OVS_UNUSED)
5575 struct ds ds = DS_EMPTY_INITIALIZER;
5576 const struct ofproto_dpif *ofproto;
5577 const struct mac_entry *e;
5579 ofproto = ofproto_dpif_lookup(argv[1]);
5581 unixctl_command_reply_error(conn, "no such bridge");
5585 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5586 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5587 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5588 struct ofbundle *bundle = e->port.p;
5589 char name[OFP_MAX_PORT_NAME_LEN];
5591 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5593 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5594 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5595 mac_entry_age(ofproto->ml, e));
5597 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5598 unixctl_command_reply(conn, ds_cstr(&ds));
5603 struct xlate_out xout;
5604 struct xlate_in xin;
5610 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5612 ds_put_char_multiple(result, '\t', level);
5614 ds_put_cstr(result, "No match\n");
5618 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5619 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5620 cls_rule_format(&rule->up.cr, result);
5621 ds_put_char(result, '\n');
5623 ds_put_char_multiple(result, '\t', level);
5624 ds_put_cstr(result, "OpenFlow ");
5625 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5626 ds_put_char(result, '\n');
5630 trace_format_flow(struct ds *result, int level, const char *title,
5631 struct trace_ctx *trace)
5633 ds_put_char_multiple(result, '\t', level);
5634 ds_put_format(result, "%s: ", title);
5635 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5636 ds_put_cstr(result, "unchanged");
5638 flow_format(result, &trace->xin.flow);
5639 trace->flow = trace->xin.flow;
5641 ds_put_char(result, '\n');
5645 trace_format_regs(struct ds *result, int level, const char *title,
5646 struct trace_ctx *trace)
5650 ds_put_char_multiple(result, '\t', level);
5651 ds_put_format(result, "%s:", title);
5652 for (i = 0; i < FLOW_N_REGS; i++) {
5653 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5655 ds_put_char(result, '\n');
5659 trace_format_odp(struct ds *result, int level, const char *title,
5660 struct trace_ctx *trace)
5662 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5664 ds_put_char_multiple(result, '\t', level);
5665 ds_put_format(result, "%s: ", title);
5666 format_odp_actions(result, odp_actions->data, odp_actions->size);
5667 ds_put_char(result, '\n');
5671 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5673 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5674 struct ds *result = trace->result;
5676 ds_put_char(result, '\n');
5677 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5678 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5679 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5680 trace_format_rule(result, recurse + 1, rule);
5684 trace_report(struct xlate_in *xin, const char *s, int recurse)
5686 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5687 struct ds *result = trace->result;
5689 ds_put_char_multiple(result, '\t', recurse);
5690 ds_put_cstr(result, s);
5691 ds_put_char(result, '\n');
5695 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5696 void *aux OVS_UNUSED)
5698 const struct dpif_backer *backer;
5699 struct ofproto_dpif *ofproto;
5700 struct ofpbuf odp_key, odp_mask;
5701 struct ofpbuf *packet;
5709 ofpbuf_init(&odp_key, 0);
5710 ofpbuf_init(&odp_mask, 0);
5712 /* Handle "-generate" or a hex string as the last argument. */
5713 if (!strcmp(argv[argc - 1], "-generate")) {
5714 packet = ofpbuf_new(0);
5717 const char *error = eth_from_hex(argv[argc - 1], &packet);
5720 } else if (argc == 4) {
5721 /* The 3-argument form must end in "-generate' or a hex string. */
5722 unixctl_command_reply_error(conn, error);
5727 /* Parse the flow and determine whether a datapath or
5728 * bridge is specified. If function odp_flow_key_from_string()
5729 * returns 0, the flow is a odp_flow. If function
5730 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5731 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5732 /* If the odp_flow is the second argument,
5733 * the datapath name is the first argument. */
5735 const char *dp_type;
5736 if (!strncmp(argv[1], "ovs-", 4)) {
5737 dp_type = argv[1] + 4;
5741 backer = shash_find_data(&all_dpif_backers, dp_type);
5743 unixctl_command_reply_error(conn, "Cannot find datapath "
5748 /* No datapath name specified, so there should be only one
5750 struct shash_node *node;
5751 if (shash_count(&all_dpif_backers) != 1) {
5752 unixctl_command_reply_error(conn, "Must specify datapath "
5753 "name, there is more than one type of datapath");
5756 node = shash_first(&all_dpif_backers);
5757 backer = node->data;
5760 /* Extract the ofproto_dpif object from the ofproto_receive()
5762 if (ofproto_receive(backer, NULL, odp_key.data,
5763 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5764 unixctl_command_reply_error(conn, "Invalid datapath flow");
5767 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5768 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5770 unixctl_command_reply_error(conn, "Must specify bridge name");
5774 ofproto = ofproto_dpif_lookup(argv[1]);
5776 unixctl_command_reply_error(conn, "Unknown bridge name");
5780 unixctl_command_reply_error(conn, "Bad flow syntax");
5784 /* Generate a packet, if requested. */
5786 if (!packet->size) {
5787 flow_compose(packet, &flow);
5789 union flow_in_port in_port_;
5791 in_port_ = flow.in_port;
5792 ds_put_cstr(&result, "Packet: ");
5793 s = ofp_packet_to_string(packet->data, packet->size);
5794 ds_put_cstr(&result, s);
5797 /* Use the metadata from the flow and the packet argument
5798 * to reconstruct the flow. */
5799 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5804 ofproto_trace(ofproto, &flow, packet, &result);
5805 unixctl_command_reply(conn, ds_cstr(&result));
5808 ds_destroy(&result);
5809 ofpbuf_delete(packet);
5810 ofpbuf_uninit(&odp_key);
5811 ofpbuf_uninit(&odp_mask);
5815 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5816 const struct ofpbuf *packet, struct ds *ds)
5818 struct rule_dpif *rule;
5820 ds_put_cstr(ds, "Flow: ");
5821 flow_format(ds, flow);
5822 ds_put_char(ds, '\n');
5824 rule = rule_dpif_lookup(ofproto, flow, NULL);
5826 trace_format_rule(ds, 0, rule);
5827 if (rule == ofproto->miss_rule) {
5828 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5829 } else if (rule == ofproto->no_packet_in_rule) {
5830 ds_put_cstr(ds, "\nNo match, packets dropped because "
5831 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5832 } else if (rule == ofproto->drop_frags_rule) {
5833 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5834 "and the fragment handling mode is \"drop\".\n");
5838 uint64_t odp_actions_stub[1024 / 8];
5839 struct ofpbuf odp_actions;
5840 struct trace_ctx trace;
5844 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5847 ofpbuf_use_stub(&odp_actions,
5848 odp_actions_stub, sizeof odp_actions_stub);
5849 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5850 trace.xin.resubmit_hook = trace_resubmit;
5851 trace.xin.report_hook = trace_report;
5853 xlate_actions(&trace.xin, &trace.xout);
5855 ds_put_char(ds, '\n');
5856 trace_format_flow(ds, 0, "Final flow", &trace);
5858 match_init(&match, flow, &trace.xout.wc);
5859 ds_put_cstr(ds, "Relevant fields: ");
5860 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5861 ds_put_char(ds, '\n');
5863 ds_put_cstr(ds, "Datapath actions: ");
5864 format_odp_actions(ds, trace.xout.odp_actions.data,
5865 trace.xout.odp_actions.size);
5867 if (trace.xout.slow) {
5868 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5869 "slow path because it:");
5870 switch (trace.xout.slow) {
5872 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5875 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5878 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5881 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5883 case SLOW_CONTROLLER:
5884 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5885 "to the OpenFlow controller.");
5892 xlate_out_uninit(&trace.xout);
5897 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5898 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5901 unixctl_command_reply(conn, NULL);
5905 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5906 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5909 unixctl_command_reply(conn, NULL);
5912 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5913 * 'reply' describing the results. */
5915 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5917 struct cls_cursor cursor;
5918 struct facet *facet;
5922 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5923 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5924 if (!facet_check_consistency(facet)) {
5929 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5933 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5934 ofproto->up.name, errors);
5936 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5941 ofproto_dpif_self_check(struct unixctl_conn *conn,
5942 int argc, const char *argv[], void *aux OVS_UNUSED)
5944 struct ds reply = DS_EMPTY_INITIALIZER;
5945 struct ofproto_dpif *ofproto;
5948 ofproto = ofproto_dpif_lookup(argv[1]);
5950 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5951 "ofproto/list for help)");
5954 ofproto_dpif_self_check__(ofproto, &reply);
5956 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5957 ofproto_dpif_self_check__(ofproto, &reply);
5961 unixctl_command_reply(conn, ds_cstr(&reply));
5965 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5966 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5967 * to destroy 'ofproto_shash' and free the returned value. */
5968 static const struct shash_node **
5969 get_ofprotos(struct shash *ofproto_shash)
5971 const struct ofproto_dpif *ofproto;
5973 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5974 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5975 shash_add_nocopy(ofproto_shash, name, ofproto);
5978 return shash_sort(ofproto_shash);
5982 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5983 const char *argv[] OVS_UNUSED,
5984 void *aux OVS_UNUSED)
5986 struct ds ds = DS_EMPTY_INITIALIZER;
5987 struct shash ofproto_shash;
5988 const struct shash_node **sorted_ofprotos;
5991 shash_init(&ofproto_shash);
5992 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5993 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5994 const struct shash_node *node = sorted_ofprotos[i];
5995 ds_put_format(&ds, "%s\n", node->name);
5998 shash_destroy(&ofproto_shash);
5999 free(sorted_ofprotos);
6001 unixctl_command_reply(conn, ds_cstr(&ds));
6006 show_dp_rates(struct ds *ds, const char *heading,
6007 const struct avg_subfacet_rates *rates)
6009 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6010 heading, rates->add_rate, rates->del_rate);
6014 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6016 const struct shash_node **ofprotos;
6017 struct ofproto_dpif *ofproto;
6018 struct shash ofproto_shash;
6019 uint64_t n_hit, n_missed;
6020 long long int minutes;
6023 n_hit = n_missed = 0;
6024 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6025 if (ofproto->backer == backer) {
6026 n_missed += ofproto->n_missed;
6027 n_hit += ofproto->n_hit;
6031 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6032 dpif_name(backer->dpif), n_hit, n_missed);
6033 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6034 " life span: %lldms\n", hmap_count(&backer->subfacets),
6035 backer->avg_n_subfacet, backer->max_n_subfacet,
6036 backer->avg_subfacet_life);
6038 minutes = (time_msec() - backer->created) / (1000 * 60);
6039 if (minutes >= 60) {
6040 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6042 if (minutes >= 60 * 24) {
6043 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6045 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6047 shash_init(&ofproto_shash);
6048 ofprotos = get_ofprotos(&ofproto_shash);
6049 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6050 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6051 const struct shash_node **ports;
6054 if (ofproto->backer != backer) {
6058 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6059 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6061 ports = shash_sort(&ofproto->up.port_by_name);
6062 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6063 const struct shash_node *node = ports[j];
6064 struct ofport *ofport = node->data;
6066 odp_port_t odp_port;
6068 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6071 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6072 if (odp_port != ODPP_NONE) {
6073 ds_put_format(ds, "%"PRIu32":", odp_port);
6075 ds_put_cstr(ds, "none:");
6078 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6081 if (!netdev_get_config(ofport->netdev, &config)) {
6082 const struct smap_node **nodes;
6085 nodes = smap_sort(&config);
6086 for (i = 0; i < smap_count(&config); i++) {
6087 const struct smap_node *node = nodes[i];
6088 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6089 node->key, node->value);
6093 smap_destroy(&config);
6095 ds_put_char(ds, ')');
6096 ds_put_char(ds, '\n');
6100 shash_destroy(&ofproto_shash);
6105 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6106 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6108 struct ds ds = DS_EMPTY_INITIALIZER;
6109 const struct shash_node **backers;
6112 backers = shash_sort(&all_dpif_backers);
6113 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6114 dpif_show_backer(backers[i]->data, &ds);
6118 unixctl_command_reply(conn, ds_cstr(&ds));
6122 /* Dump the megaflow (facet) cache. This is useful to check the
6123 * correctness of flow wildcarding, since the same mechanism is used for
6124 * both xlate caching and kernel wildcarding.
6126 * It's important to note that in the output the flow description uses
6127 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6129 * This command is only needed for advanced debugging, so it's not
6130 * documented in the man page. */
6132 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6133 int argc OVS_UNUSED, const char *argv[],
6134 void *aux OVS_UNUSED)
6136 struct ds ds = DS_EMPTY_INITIALIZER;
6137 const struct ofproto_dpif *ofproto;
6138 long long int now = time_msec();
6139 struct cls_cursor cursor;
6140 struct facet *facet;
6142 ofproto = ofproto_dpif_lookup(argv[1]);
6144 unixctl_command_reply_error(conn, "no such bridge");
6148 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6149 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6150 cls_rule_format(&facet->cr, &ds);
6151 ds_put_cstr(&ds, ", ");
6152 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6153 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6154 ds_put_cstr(&ds, "Datapath actions: ");
6155 if (facet->xout.slow) {
6156 uint64_t slow_path_stub[128 / 8];
6157 const struct nlattr *actions;
6160 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6161 slow_path_stub, sizeof slow_path_stub,
6162 &actions, &actions_len);
6163 format_odp_actions(&ds, actions, actions_len);
6165 format_odp_actions(&ds, facet->xout.odp_actions.data,
6166 facet->xout.odp_actions.size);
6168 ds_put_cstr(&ds, "\n");
6171 ds_chomp(&ds, '\n');
6172 unixctl_command_reply(conn, ds_cstr(&ds));
6176 /* Disable using the megaflows.
6178 * This command is only needed for advanced debugging, so it's not
6179 * documented in the man page. */
6181 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6182 int argc OVS_UNUSED,
6183 const char *argv[] OVS_UNUSED,
6184 void *aux OVS_UNUSED)
6186 struct ofproto_dpif *ofproto;
6188 enable_megaflows = false;
6190 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6191 flush(&ofproto->up);
6194 unixctl_command_reply(conn, "megaflows disabled");
6197 /* Re-enable using megaflows.
6199 * This command is only needed for advanced debugging, so it's not
6200 * documented in the man page. */
6202 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6203 int argc OVS_UNUSED,
6204 const char *argv[] OVS_UNUSED,
6205 void *aux OVS_UNUSED)
6207 struct ofproto_dpif *ofproto;
6209 enable_megaflows = true;
6211 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6212 flush(&ofproto->up);
6215 unixctl_command_reply(conn, "megaflows enabled");
6219 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6220 int argc OVS_UNUSED, const char *argv[],
6221 void *aux OVS_UNUSED)
6223 struct ds ds = DS_EMPTY_INITIALIZER;
6224 const struct ofproto_dpif *ofproto;
6225 struct subfacet *subfacet;
6227 ofproto = ofproto_dpif_lookup(argv[1]);
6229 unixctl_command_reply_error(conn, "no such bridge");
6233 update_stats(ofproto->backer);
6235 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6236 struct facet *facet = subfacet->facet;
6237 struct odputil_keybuf maskbuf;
6240 if (facet->ofproto != ofproto) {
6244 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6245 if (enable_megaflows) {
6246 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6247 &facet->flow, UINT32_MAX);
6250 odp_flow_format(subfacet->key, subfacet->key_len,
6251 mask.data, mask.size, &ds, false);
6253 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6254 subfacet->dp_packet_count, subfacet->dp_byte_count);
6255 if (subfacet->used) {
6256 ds_put_format(&ds, "%.3fs",
6257 (time_msec() - subfacet->used) / 1000.0);
6259 ds_put_format(&ds, "never");
6261 if (subfacet->facet->tcp_flags) {
6262 ds_put_cstr(&ds, ", flags:");
6263 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6266 ds_put_cstr(&ds, ", actions:");
6267 if (facet->xout.slow) {
6268 uint64_t slow_path_stub[128 / 8];
6269 const struct nlattr *actions;
6272 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6273 slow_path_stub, sizeof slow_path_stub,
6274 &actions, &actions_len);
6275 format_odp_actions(&ds, actions, actions_len);
6277 format_odp_actions(&ds, facet->xout.odp_actions.data,
6278 facet->xout.odp_actions.size);
6280 ds_put_char(&ds, '\n');
6283 unixctl_command_reply(conn, ds_cstr(&ds));
6288 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6289 int argc OVS_UNUSED, const char *argv[],
6290 void *aux OVS_UNUSED)
6292 struct ds ds = DS_EMPTY_INITIALIZER;
6293 struct ofproto_dpif *ofproto;
6295 ofproto = ofproto_dpif_lookup(argv[1]);
6297 unixctl_command_reply_error(conn, "no such bridge");
6301 flush(&ofproto->up);
6303 unixctl_command_reply(conn, ds_cstr(&ds));
6308 ofproto_dpif_unixctl_init(void)
6310 static bool registered;
6316 unixctl_command_register(
6318 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6319 1, 3, ofproto_unixctl_trace, NULL);
6320 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6321 ofproto_unixctl_fdb_flush, NULL);
6322 unixctl_command_register("fdb/show", "bridge", 1, 1,
6323 ofproto_unixctl_fdb_show, NULL);
6324 unixctl_command_register("ofproto/clog", "", 0, 0,
6325 ofproto_dpif_clog, NULL);
6326 unixctl_command_register("ofproto/unclog", "", 0, 0,
6327 ofproto_dpif_unclog, NULL);
6328 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6329 ofproto_dpif_self_check, NULL);
6330 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6331 ofproto_unixctl_dpif_dump_dps, NULL);
6332 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6334 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6335 ofproto_unixctl_dpif_dump_flows, NULL);
6336 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6337 ofproto_unixctl_dpif_del_flows, NULL);
6338 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6339 ofproto_unixctl_dpif_dump_megaflows, NULL);
6340 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6341 ofproto_unixctl_dpif_disable_megaflows, NULL);
6342 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6343 ofproto_unixctl_dpif_enable_megaflows, NULL);
6346 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6348 * This is deprecated. It is only for compatibility with broken device drivers
6349 * in old versions of Linux that do not properly support VLANs when VLAN
6350 * devices are not used. When broken device drivers are no longer in
6351 * widespread use, we will delete these interfaces. */
6354 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6356 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6357 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6359 if (realdev_ofp_port == ofport->realdev_ofp_port
6360 && vid == ofport->vlandev_vid) {
6364 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6366 if (ofport->realdev_ofp_port) {
6369 if (realdev_ofp_port && ofport->bundle) {
6370 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6371 * themselves be part of a bundle. */
6372 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6375 ofport->realdev_ofp_port = realdev_ofp_port;
6376 ofport->vlandev_vid = vid;
6378 if (realdev_ofp_port) {
6379 vsp_add(ofport, realdev_ofp_port, vid);
6386 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6388 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6392 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6394 return !hmap_is_empty(&ofproto->realdev_vid_map);
6397 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6398 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6399 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6400 * 'vlan_tci' 9, it would return the port number of eth0.9.
6402 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6403 * function just returns its 'realdev_ofp_port' argument. */
6405 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6406 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6408 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6409 int vid = vlan_tci_to_vid(vlan_tci);
6410 const struct vlan_splinter *vsp;
6412 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6413 hash_realdev_vid(realdev_ofp_port, vid),
6414 &ofproto->realdev_vid_map) {
6415 if (vsp->realdev_ofp_port == realdev_ofp_port
6416 && vsp->vid == vid) {
6417 return vsp->vlandev_ofp_port;
6421 return realdev_ofp_port;
6424 static struct vlan_splinter *
6425 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6427 struct vlan_splinter *vsp;
6429 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6430 hash_ofp_port(vlandev_ofp_port),
6431 &ofproto->vlandev_map) {
6432 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6440 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6441 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6442 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6443 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6444 * eth0 and store 9 in '*vid'.
6446 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6447 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6450 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6451 ofp_port_t vlandev_ofp_port, int *vid)
6453 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6454 const struct vlan_splinter *vsp;
6456 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6461 return vsp->realdev_ofp_port;
6467 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6468 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6469 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6470 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6471 * always the case unless VLAN splinters are enabled), returns false without
6472 * making any changes. */
6474 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6479 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6484 /* Cause the flow to be processed as if it came in on the real device with
6485 * the VLAN device's VLAN ID. */
6486 flow->in_port.ofp_port = realdev;
6487 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6492 vsp_remove(struct ofport_dpif *port)
6494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6495 struct vlan_splinter *vsp;
6497 vsp = vlandev_find(ofproto, port->up.ofp_port);
6499 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6500 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6503 port->realdev_ofp_port = 0;
6505 VLOG_ERR("missing vlan device record");
6510 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6512 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6514 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6515 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6516 == realdev_ofp_port)) {
6517 struct vlan_splinter *vsp;
6519 vsp = xmalloc(sizeof *vsp);
6520 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6521 hash_ofp_port(port->up.ofp_port));
6522 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6523 hash_realdev_vid(realdev_ofp_port, vid));
6524 vsp->realdev_ofp_port = realdev_ofp_port;
6525 vsp->vlandev_ofp_port = port->up.ofp_port;
6528 port->realdev_ofp_port = realdev_ofp_port;
6530 VLOG_ERR("duplicate vlan device record");
6535 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6537 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6538 return ofport ? ofport->odp_port : ODPP_NONE;
6541 static struct ofport_dpif *
6542 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6544 struct ofport_dpif *port;
6546 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6547 &backer->odp_to_ofport_map) {
6548 if (port->odp_port == odp_port) {
6557 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6559 struct ofport_dpif *port;
6561 port = odp_port_to_ofport(ofproto->backer, odp_port);
6562 if (port && &ofproto->up == port->up.ofproto) {
6563 return port->up.ofp_port;
6569 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6570 * most heavily weighted element. 'base' designates the rate of decay: after
6571 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6574 exp_mavg(double *avg, int base, double new)
6576 *avg = (*avg * (base - 1) + new) / base;
6580 update_moving_averages(struct dpif_backer *backer)
6582 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6583 long long int minutes = (time_msec() - backer->created) / min_ms;
6586 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6588 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6591 backer->lifetime.add_rate = 0.0;
6592 backer->lifetime.del_rate = 0.0;
6595 /* Update hourly averages on the minute boundaries. */
6596 if (time_msec() - backer->last_minute >= min_ms) {
6597 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6598 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6600 /* Update daily averages on the hour boundaries. */
6601 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6602 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6603 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6606 backer->total_subfacet_add_count += backer->subfacet_add_count;
6607 backer->total_subfacet_del_count += backer->subfacet_del_count;
6608 backer->subfacet_add_count = 0;
6609 backer->subfacet_del_count = 0;
6610 backer->last_minute += min_ms;
6614 const struct ofproto_class ofproto_dpif_class = {
6649 port_is_lacp_current,
6650 NULL, /* rule_choose_table */
6657 rule_modify_actions,
6671 get_stp_port_status,
6678 is_mirror_output_bundle,
6679 forward_bpdu_changed,
6680 set_mac_table_config,
6682 NULL, /* meter_get_features */
6683 NULL, /* meter_set */
6684 NULL, /* meter_get */
6685 NULL, /* meter_del */