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);
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
83 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
85 struct flow_wildcards *wc);
87 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
88 static void rule_invalidate(const struct rule_dpif *);
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 tag_type tag; /* Tag associated with this port. */
293 bool may_enable; /* May be enabled in bonds. */
294 bool is_tunnel; /* This port is a tunnel. */
295 long long int carrier_seq; /* Carrier status changes. */
296 struct ofport_dpif *peer; /* Peer if patch port. */
299 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
300 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
301 long long int stp_state_entered;
303 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
305 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
307 * This is deprecated. It is only for compatibility with broken device
308 * drivers in old versions of Linux that do not properly support VLANs when
309 * VLAN devices are not used. When broken device drivers are no longer in
310 * widespread use, we will delete these interfaces. */
311 ofp_port_t realdev_ofp_port;
315 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
316 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
317 * traffic egressing the 'ofport' with that priority should be marked with. */
318 struct priority_to_dscp {
319 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
320 uint32_t priority; /* Priority of this queue (see struct flow). */
322 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
325 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
327 * This is deprecated. It is only for compatibility with broken device drivers
328 * in old versions of Linux that do not properly support VLANs when VLAN
329 * devices are not used. When broken device drivers are no longer in
330 * widespread use, we will delete these interfaces. */
331 struct vlan_splinter {
332 struct hmap_node realdev_vid_node;
333 struct hmap_node vlandev_node;
334 ofp_port_t realdev_ofp_port;
335 ofp_port_t vlandev_ofp_port;
339 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
340 static void vsp_remove(struct ofport_dpif *);
341 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
343 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
346 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
349 static struct ofport_dpif *
350 ofport_dpif_cast(const struct ofport *ofport)
352 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
355 static void port_run(struct ofport_dpif *);
356 static void port_run_fast(struct ofport_dpif *);
357 static void port_wait(struct ofport_dpif *);
358 static int set_bfd(struct ofport *, const struct smap *);
359 static int set_cfm(struct ofport *, const struct cfm_settings *);
360 static void ofport_clear_priorities(struct ofport_dpif *);
361 static void ofport_update_peer(struct ofport_dpif *);
362 static void run_fast_rl(void);
364 struct dpif_completion {
365 struct list list_node;
366 struct ofoperation *op;
369 /* Extra information about a classifier table.
370 * Currently used just for optimized flow revalidation. */
372 /* If either of these is nonnull, then this table has a form that allows
373 * flows to be tagged to avoid revalidating most flows for the most common
374 * kinds of flow table changes. */
375 struct cls_table *catchall_table; /* Table that wildcards all fields. */
376 struct cls_table *other_table; /* Table with any other wildcard set. */
377 uint32_t basis; /* Keeps each table's tags separate. */
380 /* Reasons that we might need to revalidate every facet, and corresponding
383 * A value of 0 means that there is no need to revalidate.
385 * It would be nice to have some cleaner way to integrate with coverage
386 * counters, but with only a few reasons I guess this is good enough for
388 enum revalidate_reason {
389 REV_RECONFIGURE = 1, /* Switch configuration changed. */
390 REV_STP, /* Spanning tree protocol port status change. */
391 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
392 REV_FLOW_TABLE, /* Flow table changed. */
393 REV_INCONSISTENCY /* Facet self-check failed. */
395 COVERAGE_DEFINE(rev_reconfigure);
396 COVERAGE_DEFINE(rev_stp);
397 COVERAGE_DEFINE(rev_port_toggled);
398 COVERAGE_DEFINE(rev_flow_table);
399 COVERAGE_DEFINE(rev_inconsistency);
401 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
402 * These are datapath flows which have no associated ofproto, if they did we
403 * would use facets. */
405 struct hmap_node hmap_node;
410 struct avg_subfacet_rates {
411 double add_rate; /* Moving average of new flows created per minute. */
412 double del_rate; /* Moving average of flows deleted per minute. */
415 /* All datapaths of a given type share a single dpif backer instance. */
420 struct timer next_expiration;
421 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
423 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
425 /* Facet revalidation flags applying to facets which use this backer. */
426 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
427 struct tag_set revalidate_set; /* Revalidate only matching facets. */
429 struct hmap drop_keys; /* Set of dropped odp keys. */
430 bool recv_set_enable; /* Enables or disables receiving packets. */
432 struct hmap subfacets;
433 struct governor *governor;
435 /* Subfacet statistics.
437 * These keep track of the total number of subfacets added and deleted and
438 * flow life span. They are useful for computing the flow rates stats
439 * exposed via "ovs-appctl dpif/show". The goal is to learn about
440 * traffic patterns in ways that we can use later to improve Open vSwitch
441 * performance in new situations. */
442 long long int created; /* Time when it is created. */
443 unsigned max_n_subfacet; /* Maximum number of flows */
444 unsigned avg_n_subfacet; /* Average number of flows. */
445 long long int avg_subfacet_life; /* Average life span of subfacets. */
447 /* The average number of subfacets... */
448 struct avg_subfacet_rates hourly; /* ...over the last hour. */
449 struct avg_subfacet_rates daily; /* ...over the last day. */
450 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
451 long long int last_minute; /* Last time 'hourly' was updated. */
453 /* Number of subfacets added or deleted since 'last_minute'. */
454 unsigned subfacet_add_count;
455 unsigned subfacet_del_count;
457 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
458 unsigned long long int total_subfacet_add_count;
459 unsigned long long int total_subfacet_del_count;
462 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
463 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
465 static void drop_key_clear(struct dpif_backer *);
466 static struct ofport_dpif *
467 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
468 static void update_moving_averages(struct dpif_backer *backer);
470 struct ofproto_dpif {
471 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
473 struct dpif_backer *backer;
475 /* Special OpenFlow rules. */
476 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
477 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
478 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
481 struct netflow *netflow;
482 struct dpif_sflow *sflow;
483 struct dpif_ipfix *ipfix;
484 struct hmap bundles; /* Contains "struct ofbundle"s. */
485 struct mac_learning *ml;
486 bool has_bonded_bundles;
487 struct mbridge *mbridge;
490 struct classifier facets; /* Contains 'struct facet's. */
491 long long int consistency_rl;
494 struct table_dpif tables[N_TABLES];
496 /* Support for debugging async flow mods. */
497 struct list completions;
499 struct netdev_stats stats; /* To account packets generated and consumed in
504 long long int stp_last_tick;
506 /* VLAN splinters. */
507 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
508 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
511 struct sset ports; /* Set of standard port names. */
512 struct sset ghost_ports; /* Ports with no datapath port. */
513 struct sset port_poll_set; /* Queued names for port_poll() reply. */
514 int port_poll_errno; /* Last errno for port_poll() reply. */
516 /* Per ofproto's dpif stats. */
521 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
522 * for debugging the asynchronous flow_mod implementation.) */
525 /* By default, flows in the datapath are wildcarded (megaflows). They
526 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
527 static bool enable_megaflows = true;
529 /* All existing ofproto_dpif instances, indexed by ->up.name. */
530 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
532 static void ofproto_dpif_unixctl_init(void);
534 static inline struct ofproto_dpif *
535 ofproto_dpif_cast(const struct ofproto *ofproto)
537 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
538 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
541 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
542 ofp_port_t ofp_port);
545 #define FLOW_MISS_MAX_BATCH 50
546 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
548 /* Flow expiration. */
549 static int expire(struct dpif_backer *);
552 static void send_netflow_active_timeouts(struct ofproto_dpif *);
555 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
557 /* Global variables. */
558 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
560 /* Initial mappings of port to bridge mappings. */
561 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
564 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
565 struct ofputil_flow_mod *fm)
567 return ofproto_flow_mod(&ofproto->up, fm);
571 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
572 struct ofputil_packet_in *pin)
574 connmgr_send_packet_in(ofproto->up.connmgr, pin);
577 /* Factory functions. */
580 init(const struct shash *iface_hints)
582 struct shash_node *node;
584 /* Make a local copy, since we don't own 'iface_hints' elements. */
585 SHASH_FOR_EACH(node, iface_hints) {
586 const struct iface_hint *orig_hint = node->data;
587 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
589 new_hint->br_name = xstrdup(orig_hint->br_name);
590 new_hint->br_type = xstrdup(orig_hint->br_type);
591 new_hint->ofp_port = orig_hint->ofp_port;
593 shash_add(&init_ofp_ports, node->name, new_hint);
598 enumerate_types(struct sset *types)
600 dp_enumerate_types(types);
604 enumerate_names(const char *type, struct sset *names)
606 struct ofproto_dpif *ofproto;
609 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
610 if (strcmp(type, ofproto->up.type)) {
613 sset_add(names, ofproto->up.name);
620 del(const char *type, const char *name)
625 error = dpif_open(name, type, &dpif);
627 error = dpif_delete(dpif);
634 port_open_type(const char *datapath_type, const char *port_type)
636 return dpif_port_open_type(datapath_type, port_type);
639 /* Type functions. */
641 static void process_dpif_port_changes(struct dpif_backer *);
642 static void process_dpif_all_ports_changed(struct dpif_backer *);
643 static void process_dpif_port_change(struct dpif_backer *,
644 const char *devname);
645 static void process_dpif_port_error(struct dpif_backer *, int error);
647 static struct ofproto_dpif *
648 lookup_ofproto_dpif_by_port_name(const char *name)
650 struct ofproto_dpif *ofproto;
652 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
653 if (sset_contains(&ofproto->ports, name)) {
662 type_run(const char *type)
664 static long long int push_timer = LLONG_MIN;
665 struct dpif_backer *backer;
667 backer = shash_find_data(&all_dpif_backers, type);
669 /* This is not necessarily a problem, since backers are only
670 * created on demand. */
674 dpif_run(backer->dpif);
676 /* The most natural place to push facet statistics is when they're pulled
677 * from the datapath. However, when there are many flows in the datapath,
678 * this expensive operation can occur so frequently, that it reduces our
679 * ability to quickly set up flows. To reduce the cost, we push statistics
681 if (time_msec() > push_timer) {
682 push_timer = time_msec() + 2000;
686 /* If vswitchd started with other_config:flow_restore_wait set as "true",
687 * and the configuration has now changed to "false", enable receiving
688 * packets from the datapath. */
689 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
692 backer->recv_set_enable = true;
694 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
696 VLOG_ERR("Failed to enable receiving packets in dpif.");
699 dpif_flow_flush(backer->dpif);
700 backer->need_revalidate = REV_RECONFIGURE;
703 if (backer->need_revalidate
704 || !tag_set_is_empty(&backer->revalidate_set)) {
705 struct tag_set revalidate_set = backer->revalidate_set;
706 bool need_revalidate = backer->need_revalidate;
707 struct ofproto_dpif *ofproto;
708 struct simap_node *node;
709 struct simap tmp_backers;
711 /* Handle tunnel garbage collection. */
712 simap_init(&tmp_backers);
713 simap_swap(&backer->tnl_backers, &tmp_backers);
715 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
716 struct ofport_dpif *iter;
718 if (backer != ofproto->backer) {
722 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
723 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
726 if (!iter->is_tunnel) {
730 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
731 namebuf, sizeof namebuf);
732 node = simap_find(&tmp_backers, dp_port);
734 simap_put(&backer->tnl_backers, dp_port, node->data);
735 simap_delete(&tmp_backers, node);
736 node = simap_find(&backer->tnl_backers, dp_port);
738 node = simap_find(&backer->tnl_backers, dp_port);
740 odp_port_t odp_port = ODPP_NONE;
742 if (!dpif_port_add(backer->dpif, iter->up.netdev,
744 simap_put(&backer->tnl_backers, dp_port,
745 odp_to_u32(odp_port));
746 node = simap_find(&backer->tnl_backers, dp_port);
751 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
752 if (tnl_port_reconfigure(iter, iter->up.netdev,
754 backer->need_revalidate = REV_RECONFIGURE;
759 SIMAP_FOR_EACH (node, &tmp_backers) {
760 dpif_port_del(backer->dpif, u32_to_odp(node->data));
762 simap_destroy(&tmp_backers);
764 switch (backer->need_revalidate) {
765 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
766 case REV_STP: COVERAGE_INC(rev_stp); break;
767 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
768 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
769 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
772 if (backer->need_revalidate) {
773 /* Clear the drop_keys in case we should now be accepting some
774 * formerly dropped flows. */
775 drop_key_clear(backer);
778 /* Clear the revalidation flags. */
779 tag_set_init(&backer->revalidate_set);
780 backer->need_revalidate = 0;
782 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
783 struct facet *facet, *next;
784 struct cls_cursor cursor;
786 if (ofproto->backer != backer) {
790 if (need_revalidate) {
791 struct ofport_dpif *ofport;
792 struct ofbundle *bundle;
794 xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
795 ofproto->mbridge, ofproto->sflow,
796 ofproto->ipfix, ofproto->up.frag_handling,
797 ofproto->up.forward_bpdu,
798 connmgr_has_in_band(ofproto->up.connmgr),
799 ofproto->netflow != NULL,
800 ofproto->stp != NULL);
802 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
803 xlate_bundle_set(ofproto, bundle, bundle->name,
804 bundle->vlan_mode, bundle->vlan,
805 bundle->trunks, bundle->use_priority_tags,
806 bundle->bond, bundle->lacp,
810 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
811 xlate_ofport_set(ofproto, ofport->bundle, ofport,
812 ofport->up.ofp_port, ofport->odp_port,
813 ofport->up.netdev, ofport->cfm,
814 ofport->bfd, ofport->peer,
815 ofport->up.pp.config, ofport->stp_state,
816 ofport->is_tunnel, ofport->may_enable);
820 cls_cursor_init(&cursor, &ofproto->facets, NULL);
821 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
823 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
824 facet_revalidate(facet);
831 if (!backer->recv_set_enable) {
832 /* Wake up before a max of 1000ms. */
833 timer_set_duration(&backer->next_expiration, 1000);
834 } else if (timer_expired(&backer->next_expiration)) {
835 int delay = expire(backer);
836 timer_set_duration(&backer->next_expiration, delay);
839 process_dpif_port_changes(backer);
841 if (backer->governor) {
844 governor_run(backer->governor);
846 /* If the governor has shrunk to its minimum size and the number of
847 * subfacets has dwindled, then drop the governor entirely.
849 * For hysteresis, the number of subfacets to drop the governor is
850 * smaller than the number needed to trigger its creation. */
851 n_subfacets = hmap_count(&backer->subfacets);
852 if (n_subfacets * 4 < flow_eviction_threshold
853 && governor_is_idle(backer->governor)) {
854 governor_destroy(backer->governor);
855 backer->governor = NULL;
862 /* Check for and handle port changes in 'backer''s dpif. */
864 process_dpif_port_changes(struct dpif_backer *backer)
870 error = dpif_port_poll(backer->dpif, &devname);
876 process_dpif_all_ports_changed(backer);
880 process_dpif_port_change(backer, devname);
885 process_dpif_port_error(backer, error);
892 process_dpif_all_ports_changed(struct dpif_backer *backer)
894 struct ofproto_dpif *ofproto;
895 struct dpif_port dpif_port;
896 struct dpif_port_dump dump;
897 struct sset devnames;
900 sset_init(&devnames);
901 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
902 if (ofproto->backer == backer) {
903 struct ofport *ofport;
905 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
906 sset_add(&devnames, netdev_get_name(ofport->netdev));
910 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
911 sset_add(&devnames, dpif_port.name);
914 SSET_FOR_EACH (devname, &devnames) {
915 process_dpif_port_change(backer, devname);
917 sset_destroy(&devnames);
921 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
923 struct ofproto_dpif *ofproto;
924 struct dpif_port port;
926 /* Don't report on the datapath's device. */
927 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
931 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
932 &all_ofproto_dpifs) {
933 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
938 ofproto = lookup_ofproto_dpif_by_port_name(devname);
939 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
940 /* The port was removed. If we know the datapath,
941 * report it through poll_set(). If we don't, it may be
942 * notifying us of a removal we initiated, so ignore it.
943 * If there's a pending ENOBUFS, let it stand, since
944 * everything will be reevaluated. */
945 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
946 sset_add(&ofproto->port_poll_set, devname);
947 ofproto->port_poll_errno = 0;
949 } else if (!ofproto) {
950 /* The port was added, but we don't know with which
951 * ofproto we should associate it. Delete it. */
952 dpif_port_del(backer->dpif, port.port_no);
954 dpif_port_destroy(&port);
957 /* Propagate 'error' to all ofprotos based on 'backer'. */
959 process_dpif_port_error(struct dpif_backer *backer, int error)
961 struct ofproto_dpif *ofproto;
963 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
964 if (ofproto->backer == backer) {
965 sset_clear(&ofproto->port_poll_set);
966 ofproto->port_poll_errno = error;
972 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
976 /* If recv_set_enable is false, we should not handle upcalls. */
977 if (!backer->recv_set_enable) {
981 /* Handle one or more batches of upcalls, until there's nothing left to do
982 * or until we do a fixed total amount of work.
984 * We do work in batches because it can be much cheaper to set up a number
985 * of flows and fire off their patches all at once. We do multiple batches
986 * because in some cases handling a packet can cause another packet to be
987 * queued almost immediately as part of the return flow. Both
988 * optimizations can make major improvements on some benchmarks and
989 * presumably for real traffic as well. */
991 while (work < max_batch) {
992 int retval = handle_upcalls(backer, max_batch - work);
1003 type_run_fast(const char *type)
1005 struct dpif_backer *backer;
1007 backer = shash_find_data(&all_dpif_backers, type);
1009 /* This is not necessarily a problem, since backers are only
1010 * created on demand. */
1014 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1020 static long long int port_rl = LLONG_MIN;
1021 static unsigned int backer_rl = 0;
1023 if (time_msec() >= port_rl) {
1024 struct ofproto_dpif *ofproto;
1025 struct ofport_dpif *ofport;
1027 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1029 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1030 port_run_fast(ofport);
1033 port_rl = time_msec() + 200;
1036 /* XXX: We have to be careful not to do too much work in this function. If
1037 * we call dpif_backer_run_fast() too often, or with too large a batch,
1038 * performance improves signifcantly, but at a cost. It's possible for the
1039 * number of flows in the datapath to increase without bound, and for poll
1040 * loops to take 10s of seconds. The correct solution to this problem,
1041 * long term, is to separate flow miss handling into it's own thread so it
1042 * isn't affected by revalidations, and expirations. Until then, this is
1043 * the best we can do. */
1044 if (++backer_rl >= 10) {
1045 struct shash_node *node;
1048 SHASH_FOR_EACH (node, &all_dpif_backers) {
1049 dpif_backer_run_fast(node->data, 1);
1055 type_wait(const char *type)
1057 struct dpif_backer *backer;
1059 backer = shash_find_data(&all_dpif_backers, type);
1061 /* This is not necessarily a problem, since backers are only
1062 * created on demand. */
1066 if (backer->governor) {
1067 governor_wait(backer->governor);
1070 timer_wait(&backer->next_expiration);
1073 /* Basic life-cycle. */
1075 static int add_internal_flows(struct ofproto_dpif *);
1077 static struct ofproto *
1080 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1081 return &ofproto->up;
1085 dealloc(struct ofproto *ofproto_)
1087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1092 close_dpif_backer(struct dpif_backer *backer)
1094 struct shash_node *node;
1096 ovs_assert(backer->refcount > 0);
1098 if (--backer->refcount) {
1102 drop_key_clear(backer);
1103 hmap_destroy(&backer->drop_keys);
1105 simap_destroy(&backer->tnl_backers);
1106 hmap_destroy(&backer->odp_to_ofport_map);
1107 node = shash_find(&all_dpif_backers, backer->type);
1109 shash_delete(&all_dpif_backers, node);
1110 dpif_close(backer->dpif);
1112 ovs_assert(hmap_is_empty(&backer->subfacets));
1113 hmap_destroy(&backer->subfacets);
1114 governor_destroy(backer->governor);
1119 /* Datapath port slated for removal from datapath. */
1120 struct odp_garbage {
1121 struct list list_node;
1122 odp_port_t odp_port;
1126 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1128 struct dpif_backer *backer;
1129 struct dpif_port_dump port_dump;
1130 struct dpif_port port;
1131 struct shash_node *node;
1132 struct list garbage_list;
1133 struct odp_garbage *garbage, *next;
1139 backer = shash_find_data(&all_dpif_backers, type);
1146 backer_name = xasprintf("ovs-%s", type);
1148 /* Remove any existing datapaths, since we assume we're the only
1149 * userspace controlling the datapath. */
1151 dp_enumerate_names(type, &names);
1152 SSET_FOR_EACH(name, &names) {
1153 struct dpif *old_dpif;
1155 /* Don't remove our backer if it exists. */
1156 if (!strcmp(name, backer_name)) {
1160 if (dpif_open(name, type, &old_dpif)) {
1161 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1163 dpif_delete(old_dpif);
1164 dpif_close(old_dpif);
1167 sset_destroy(&names);
1169 backer = xmalloc(sizeof *backer);
1171 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1174 VLOG_ERR("failed to open datapath of type %s: %s", type,
1175 ovs_strerror(error));
1180 backer->type = xstrdup(type);
1181 backer->governor = NULL;
1182 backer->refcount = 1;
1183 hmap_init(&backer->odp_to_ofport_map);
1184 hmap_init(&backer->drop_keys);
1185 hmap_init(&backer->subfacets);
1186 timer_set_duration(&backer->next_expiration, 1000);
1187 backer->need_revalidate = 0;
1188 simap_init(&backer->tnl_backers);
1189 tag_set_init(&backer->revalidate_set);
1190 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1193 if (backer->recv_set_enable) {
1194 dpif_flow_flush(backer->dpif);
1197 /* Loop through the ports already on the datapath and remove any
1198 * that we don't need anymore. */
1199 list_init(&garbage_list);
1200 dpif_port_dump_start(&port_dump, backer->dpif);
1201 while (dpif_port_dump_next(&port_dump, &port)) {
1202 node = shash_find(&init_ofp_ports, port.name);
1203 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204 garbage = xmalloc(sizeof *garbage);
1205 garbage->odp_port = port.port_no;
1206 list_push_front(&garbage_list, &garbage->list_node);
1209 dpif_port_dump_done(&port_dump);
1211 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212 dpif_port_del(backer->dpif, garbage->odp_port);
1213 list_remove(&garbage->list_node);
1217 shash_add(&all_dpif_backers, type, backer);
1219 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1221 VLOG_ERR("failed to listen on datapath of type %s: %s",
1222 type, ovs_strerror(error));
1223 close_dpif_backer(backer);
1227 backer->max_n_subfacet = 0;
1228 backer->created = time_msec();
1229 backer->last_minute = backer->created;
1230 memset(&backer->hourly, 0, sizeof backer->hourly);
1231 memset(&backer->daily, 0, sizeof backer->daily);
1232 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1233 backer->subfacet_add_count = 0;
1234 backer->subfacet_del_count = 0;
1235 backer->total_subfacet_add_count = 0;
1236 backer->total_subfacet_del_count = 0;
1237 backer->avg_n_subfacet = 0;
1238 backer->avg_subfacet_life = 0;
1244 construct(struct ofproto *ofproto_)
1246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1247 struct shash_node *node, *next;
1248 odp_port_t max_ports;
1252 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1257 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1258 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1259 ofp_to_u16(OFPP_MAX))));
1261 ofproto->netflow = NULL;
1262 ofproto->sflow = NULL;
1263 ofproto->ipfix = NULL;
1264 ofproto->stp = NULL;
1265 hmap_init(&ofproto->bundles);
1266 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1267 ofproto->mbridge = mbridge_create();
1268 ofproto->has_bonded_bundles = false;
1270 classifier_init(&ofproto->facets);
1271 ofproto->consistency_rl = LLONG_MIN;
1273 for (i = 0; i < N_TABLES; i++) {
1274 struct table_dpif *table = &ofproto->tables[i];
1276 table->catchall_table = NULL;
1277 table->other_table = NULL;
1278 table->basis = random_uint32();
1281 list_init(&ofproto->completions);
1283 ofproto_dpif_unixctl_init();
1285 hmap_init(&ofproto->vlandev_map);
1286 hmap_init(&ofproto->realdev_vid_map);
1288 sset_init(&ofproto->ports);
1289 sset_init(&ofproto->ghost_ports);
1290 sset_init(&ofproto->port_poll_set);
1291 ofproto->port_poll_errno = 0;
1293 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1294 struct iface_hint *iface_hint = node->data;
1296 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1297 /* Check if the datapath already has this port. */
1298 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1299 sset_add(&ofproto->ports, node->name);
1302 free(iface_hint->br_name);
1303 free(iface_hint->br_type);
1305 shash_delete(&init_ofp_ports, node);
1309 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1310 hash_string(ofproto->up.name, 0));
1311 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1313 ofproto_init_tables(ofproto_, N_TABLES);
1314 error = add_internal_flows(ofproto);
1315 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1318 ofproto->n_missed = 0;
1324 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1325 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1327 struct ofputil_flow_mod fm;
1330 match_init_catchall(&fm.match);
1332 match_set_reg(&fm.match, 0, id);
1333 fm.new_cookie = htonll(0);
1334 fm.cookie = htonll(0);
1335 fm.cookie_mask = htonll(0);
1336 fm.modify_cookie = false;
1337 fm.table_id = TBL_INTERNAL;
1338 fm.command = OFPFC_ADD;
1339 fm.idle_timeout = 0;
1340 fm.hard_timeout = 0;
1344 fm.ofpacts = ofpacts->data;
1345 fm.ofpacts_len = ofpacts->size;
1347 error = ofproto_flow_mod(&ofproto->up, &fm);
1349 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1350 id, ofperr_to_string(error));
1354 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1356 ovs_assert(*rulep != NULL);
1362 add_internal_flows(struct ofproto_dpif *ofproto)
1364 struct ofpact_controller *controller;
1365 uint64_t ofpacts_stub[128 / 8];
1366 struct ofpbuf ofpacts;
1370 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1373 controller = ofpact_put_CONTROLLER(&ofpacts);
1374 controller->max_len = UINT16_MAX;
1375 controller->controller_id = 0;
1376 controller->reason = OFPR_NO_MATCH;
1377 ofpact_pad(&ofpacts);
1379 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1384 ofpbuf_clear(&ofpacts);
1385 error = add_internal_flow(ofproto, id++, &ofpacts,
1386 &ofproto->no_packet_in_rule);
1391 error = add_internal_flow(ofproto, id++, &ofpacts,
1392 &ofproto->drop_frags_rule);
1397 complete_operations(struct ofproto_dpif *ofproto)
1399 struct dpif_completion *c, *next;
1401 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1402 ofoperation_complete(c->op, 0);
1403 list_remove(&c->list_node);
1409 destruct(struct ofproto *ofproto_)
1411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1412 struct rule_dpif *rule, *next_rule;
1413 struct oftable *table;
1415 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1416 xlate_remove_ofproto(ofproto);
1418 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1419 complete_operations(ofproto);
1421 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1422 struct cls_cursor cursor;
1424 cls_cursor_init(&cursor, &table->cls, NULL);
1425 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1426 ofproto_rule_destroy(&rule->up);
1430 mbridge_unref(ofproto->mbridge);
1432 netflow_destroy(ofproto->netflow);
1433 dpif_sflow_unref(ofproto->sflow);
1434 hmap_destroy(&ofproto->bundles);
1435 mac_learning_unref(ofproto->ml);
1437 classifier_destroy(&ofproto->facets);
1439 hmap_destroy(&ofproto->vlandev_map);
1440 hmap_destroy(&ofproto->realdev_vid_map);
1442 sset_destroy(&ofproto->ports);
1443 sset_destroy(&ofproto->ghost_ports);
1444 sset_destroy(&ofproto->port_poll_set);
1446 close_dpif_backer(ofproto->backer);
1450 run_fast(struct ofproto *ofproto_)
1452 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1453 struct ofport_dpif *ofport;
1455 /* Do not perform any periodic activity required by 'ofproto' while
1456 * waiting for flow restore to complete. */
1457 if (ofproto_get_flow_restore_wait()) {
1461 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1462 port_run_fast(ofport);
1469 run(struct ofproto *ofproto_)
1471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1472 struct ofport_dpif *ofport;
1473 struct ofbundle *bundle;
1477 complete_operations(ofproto);
1480 if (mbridge_need_revalidate(ofproto->mbridge)) {
1481 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1482 mac_learning_flush(ofproto->ml, NULL);
1485 /* Do not perform any periodic activity below required by 'ofproto' while
1486 * waiting for flow restore to complete. */
1487 if (ofproto_get_flow_restore_wait()) {
1491 error = run_fast(ofproto_);
1496 if (ofproto->netflow) {
1497 if (netflow_run(ofproto->netflow)) {
1498 send_netflow_active_timeouts(ofproto);
1501 if (ofproto->sflow) {
1502 dpif_sflow_run(ofproto->sflow);
1505 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1508 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1513 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1515 /* Check the consistency of a random facet, to aid debugging. */
1516 if (time_msec() >= ofproto->consistency_rl
1517 && !classifier_is_empty(&ofproto->facets)
1518 && !ofproto->backer->need_revalidate) {
1519 struct cls_table *table;
1520 struct cls_rule *cr;
1521 struct facet *facet;
1523 ofproto->consistency_rl = time_msec() + 250;
1525 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1526 struct cls_table, hmap_node);
1527 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1529 facet = CONTAINER_OF(cr, struct facet, cr);
1531 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1532 facet->xout.tags)) {
1533 if (!facet_check_consistency(facet)) {
1534 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1543 wait(struct ofproto *ofproto_)
1545 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1546 struct ofport_dpif *ofport;
1547 struct ofbundle *bundle;
1549 if (!clogged && !list_is_empty(&ofproto->completions)) {
1550 poll_immediate_wake();
1553 if (ofproto_get_flow_restore_wait()) {
1557 dpif_wait(ofproto->backer->dpif);
1558 dpif_recv_wait(ofproto->backer->dpif);
1559 if (ofproto->sflow) {
1560 dpif_sflow_wait(ofproto->sflow);
1562 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1563 poll_immediate_wake();
1565 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1568 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1569 bundle_wait(bundle);
1571 if (ofproto->netflow) {
1572 netflow_wait(ofproto->netflow);
1574 mac_learning_wait(ofproto->ml);
1576 if (ofproto->backer->need_revalidate) {
1577 /* Shouldn't happen, but if it does just go around again. */
1578 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1579 poll_immediate_wake();
1584 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1586 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1587 struct cls_cursor cursor;
1588 size_t n_subfacets = 0;
1589 struct facet *facet;
1591 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1593 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1594 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1595 n_subfacets += list_size(&facet->subfacets);
1597 simap_increase(usage, "subfacets", n_subfacets);
1601 flush(struct ofproto *ofproto_)
1603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1604 struct subfacet *subfacet, *next_subfacet;
1605 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1609 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1610 &ofproto->backer->subfacets) {
1611 if (subfacet->facet->ofproto != ofproto) {
1615 if (subfacet->path != SF_NOT_INSTALLED) {
1616 batch[n_batch++] = subfacet;
1617 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1618 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1622 subfacet_destroy(subfacet);
1627 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1632 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1633 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1635 *arp_match_ip = true;
1636 *actions = (OFPUTIL_A_OUTPUT |
1637 OFPUTIL_A_SET_VLAN_VID |
1638 OFPUTIL_A_SET_VLAN_PCP |
1639 OFPUTIL_A_STRIP_VLAN |
1640 OFPUTIL_A_SET_DL_SRC |
1641 OFPUTIL_A_SET_DL_DST |
1642 OFPUTIL_A_SET_NW_SRC |
1643 OFPUTIL_A_SET_NW_DST |
1644 OFPUTIL_A_SET_NW_TOS |
1645 OFPUTIL_A_SET_TP_SRC |
1646 OFPUTIL_A_SET_TP_DST |
1651 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1653 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1654 struct dpif_dp_stats s;
1655 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1658 strcpy(ots->name, "classifier");
1660 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1661 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1662 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1663 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1665 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1666 ots->lookup_count = htonll(n_lookup);
1667 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1670 static struct ofport *
1673 struct ofport_dpif *port = xmalloc(sizeof *port);
1678 port_dealloc(struct ofport *port_)
1680 struct ofport_dpif *port = ofport_dpif_cast(port_);
1685 port_construct(struct ofport *port_)
1687 struct ofport_dpif *port = ofport_dpif_cast(port_);
1688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1689 const struct netdev *netdev = port->up.netdev;
1690 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1691 struct dpif_port dpif_port;
1694 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1695 port->bundle = NULL;
1698 port->tag = tag_create_random();
1699 port->may_enable = true;
1700 port->stp_port = NULL;
1701 port->stp_state = STP_DISABLED;
1702 port->is_tunnel = false;
1704 hmap_init(&port->priorities);
1705 port->realdev_ofp_port = 0;
1706 port->vlandev_vid = 0;
1707 port->carrier_seq = netdev_get_carrier_resets(netdev);
1709 if (netdev_vport_is_patch(netdev)) {
1710 /* By bailing out here, we don't submit the port to the sFlow module
1711 * to be considered for counter polling export. This is correct
1712 * because the patch port represents an interface that sFlow considers
1713 * to be "internal" to the switch as a whole, and therefore not an
1714 * candidate for counter polling. */
1715 port->odp_port = ODPP_NONE;
1716 ofport_update_peer(port);
1720 error = dpif_port_query_by_name(ofproto->backer->dpif,
1721 netdev_vport_get_dpif_port(netdev, namebuf,
1728 port->odp_port = dpif_port.port_no;
1730 if (netdev_get_tunnel_config(netdev)) {
1731 tnl_port_add(port, port->up.netdev, port->odp_port);
1732 port->is_tunnel = true;
1734 /* Sanity-check that a mapping doesn't already exist. This
1735 * shouldn't happen for non-tunnel ports. */
1736 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1737 VLOG_ERR("port %s already has an OpenFlow port number",
1739 dpif_port_destroy(&dpif_port);
1743 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1744 hash_odp_port(port->odp_port));
1746 dpif_port_destroy(&dpif_port);
1748 if (ofproto->sflow) {
1749 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1756 port_destruct(struct ofport *port_)
1758 struct ofport_dpif *port = ofport_dpif_cast(port_);
1759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1760 const char *devname = netdev_get_name(port->up.netdev);
1761 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1762 const char *dp_port_name;
1764 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1765 xlate_ofport_remove(port);
1767 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1769 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1770 /* The underlying device is still there, so delete it. This
1771 * happens when the ofproto is being destroyed, since the caller
1772 * assumes that removal of attached ports will happen as part of
1774 if (!port->is_tunnel) {
1775 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1780 port->peer->peer = NULL;
1784 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1785 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1789 sset_find_and_delete(&ofproto->ports, devname);
1790 sset_find_and_delete(&ofproto->ghost_ports, devname);
1791 bundle_remove(port_);
1792 set_cfm(port_, NULL);
1793 set_bfd(port_, NULL);
1794 if (ofproto->sflow) {
1795 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1798 ofport_clear_priorities(port);
1799 hmap_destroy(&port->priorities);
1803 port_modified(struct ofport *port_)
1805 struct ofport_dpif *port = ofport_dpif_cast(port_);
1807 if (port->bundle && port->bundle->bond) {
1808 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1812 cfm_set_netdev(port->cfm, port->up.netdev);
1815 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1817 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1821 ofport_update_peer(port);
1825 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1827 struct ofport_dpif *port = ofport_dpif_cast(port_);
1828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1829 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1831 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1832 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1833 OFPUTIL_PC_NO_PACKET_IN)) {
1834 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1837 bundle_update(port->bundle);
1843 set_sflow(struct ofproto *ofproto_,
1844 const struct ofproto_sflow_options *sflow_options)
1846 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1847 struct dpif_sflow *ds = ofproto->sflow;
1849 if (sflow_options) {
1851 struct ofport_dpif *ofport;
1853 ds = ofproto->sflow = dpif_sflow_create();
1854 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1855 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1859 dpif_sflow_set_options(ds, sflow_options);
1862 dpif_sflow_unref(ds);
1863 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1864 ofproto->sflow = NULL;
1872 struct ofproto *ofproto_,
1873 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1874 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1875 size_t n_flow_exporters_options)
1877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1878 struct dpif_ipfix *di = ofproto->ipfix;
1880 if (bridge_exporter_options || flow_exporters_options) {
1882 di = ofproto->ipfix = dpif_ipfix_create();
1884 dpif_ipfix_set_options(
1885 di, bridge_exporter_options, flow_exporters_options,
1886 n_flow_exporters_options);
1889 dpif_ipfix_unref(di);
1890 ofproto->ipfix = NULL;
1897 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1899 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1906 struct ofproto_dpif *ofproto;
1908 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1909 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1910 ofport->cfm = cfm_create(ofport->up.netdev);
1913 if (cfm_configure(ofport->cfm, s)) {
1919 cfm_unref(ofport->cfm);
1925 get_cfm_status(const struct ofport *ofport_,
1926 struct ofproto_cfm_status *status)
1928 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1931 status->faults = cfm_get_fault(ofport->cfm);
1932 status->remote_opstate = cfm_get_opup(ofport->cfm);
1933 status->health = cfm_get_health(ofport->cfm);
1934 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1942 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1945 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1949 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1950 if (ofport->bfd != old) {
1951 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1960 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1963 bfd_get_status(ofport->bfd, smap);
1970 /* Spanning Tree. */
1973 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1975 struct ofproto_dpif *ofproto = ofproto_;
1976 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1977 struct ofport_dpif *ofport;
1979 ofport = stp_port_get_aux(sp);
1981 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1982 ofproto->up.name, port_num);
1984 struct eth_header *eth = pkt->l2;
1986 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1987 if (eth_addr_is_zero(eth->eth_src)) {
1988 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1989 "with unknown MAC", ofproto->up.name, port_num);
1991 send_packet(ofport, pkt);
1997 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1999 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2003 /* Only revalidate flows if the configuration changed. */
2004 if (!s != !ofproto->stp) {
2005 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2009 if (!ofproto->stp) {
2010 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2011 send_bpdu_cb, ofproto);
2012 ofproto->stp_last_tick = time_msec();
2015 stp_set_bridge_id(ofproto->stp, s->system_id);
2016 stp_set_bridge_priority(ofproto->stp, s->priority);
2017 stp_set_hello_time(ofproto->stp, s->hello_time);
2018 stp_set_max_age(ofproto->stp, s->max_age);
2019 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2021 struct ofport *ofport;
2023 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2024 set_stp_port(ofport, NULL);
2027 stp_destroy(ofproto->stp);
2028 ofproto->stp = NULL;
2035 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2041 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2042 s->designated_root = stp_get_designated_root(ofproto->stp);
2043 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2052 update_stp_port_state(struct ofport_dpif *ofport)
2054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2055 enum stp_state state;
2057 /* Figure out new state. */
2058 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2062 if (ofport->stp_state != state) {
2063 enum ofputil_port_state of_state;
2066 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2067 netdev_get_name(ofport->up.netdev),
2068 stp_state_name(ofport->stp_state),
2069 stp_state_name(state));
2070 if (stp_learn_in_state(ofport->stp_state)
2071 != stp_learn_in_state(state)) {
2072 /* xxx Learning action flows should also be flushed. */
2073 mac_learning_flush(ofproto->ml,
2074 &ofproto->backer->revalidate_set);
2076 fwd_change = stp_forward_in_state(ofport->stp_state)
2077 != stp_forward_in_state(state);
2079 ofproto->backer->need_revalidate = REV_STP;
2080 ofport->stp_state = state;
2081 ofport->stp_state_entered = time_msec();
2083 if (fwd_change && ofport->bundle) {
2084 bundle_update(ofport->bundle);
2087 /* Update the STP state bits in the OpenFlow port description. */
2088 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2089 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2090 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2091 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2092 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2094 ofproto_port_set_state(&ofport->up, of_state);
2098 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2099 * caller is responsible for assigning STP port numbers and ensuring
2100 * there are no duplicates. */
2102 set_stp_port(struct ofport *ofport_,
2103 const struct ofproto_port_stp_settings *s)
2105 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2106 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2107 struct stp_port *sp = ofport->stp_port;
2109 if (!s || !s->enable) {
2111 ofport->stp_port = NULL;
2112 stp_port_disable(sp);
2113 update_stp_port_state(ofport);
2116 } else if (sp && stp_port_no(sp) != s->port_num
2117 && ofport == stp_port_get_aux(sp)) {
2118 /* The port-id changed, so disable the old one if it's not
2119 * already in use by another port. */
2120 stp_port_disable(sp);
2123 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2124 stp_port_enable(sp);
2126 stp_port_set_aux(sp, ofport);
2127 stp_port_set_priority(sp, s->priority);
2128 stp_port_set_path_cost(sp, s->path_cost);
2130 update_stp_port_state(ofport);
2136 get_stp_port_status(struct ofport *ofport_,
2137 struct ofproto_port_stp_status *s)
2139 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2141 struct stp_port *sp = ofport->stp_port;
2143 if (!ofproto->stp || !sp) {
2149 s->port_id = stp_port_get_id(sp);
2150 s->state = stp_port_get_state(sp);
2151 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2152 s->role = stp_port_get_role(sp);
2153 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2159 stp_run(struct ofproto_dpif *ofproto)
2162 long long int now = time_msec();
2163 long long int elapsed = now - ofproto->stp_last_tick;
2164 struct stp_port *sp;
2167 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2168 ofproto->stp_last_tick = now;
2170 while (stp_get_changed_port(ofproto->stp, &sp)) {
2171 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2174 update_stp_port_state(ofport);
2178 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2179 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2185 stp_wait(struct ofproto_dpif *ofproto)
2188 poll_timer_wait(1000);
2192 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2193 * were used to make the determination.*/
2195 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2197 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2198 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2202 stp_process_packet(const struct ofport_dpif *ofport,
2203 const struct ofpbuf *packet)
2205 struct ofpbuf payload = *packet;
2206 struct eth_header *eth = payload.data;
2207 struct stp_port *sp = ofport->stp_port;
2209 /* Sink packets on ports that have STP disabled when the bridge has
2211 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2215 /* Trim off padding on payload. */
2216 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2217 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2220 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2221 stp_received_bpdu(sp, payload.data, payload.size);
2226 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
2227 uint32_t queue_id, uint32_t *priority)
2229 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
2232 static struct priority_to_dscp *
2233 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2235 struct priority_to_dscp *pdscp;
2238 hash = hash_int(priority, 0);
2239 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2240 if (pdscp->priority == priority) {
2248 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2249 uint32_t priority, uint8_t *dscp)
2251 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2252 *dscp = pdscp ? pdscp->dscp : 0;
2253 return pdscp != NULL;
2257 ofport_clear_priorities(struct ofport_dpif *ofport)
2259 struct priority_to_dscp *pdscp, *next;
2261 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2262 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2268 set_queues(struct ofport *ofport_,
2269 const struct ofproto_port_queue *qdscp_list,
2272 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2273 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2274 struct hmap new = HMAP_INITIALIZER(&new);
2277 for (i = 0; i < n_qdscp; i++) {
2278 struct priority_to_dscp *pdscp;
2282 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2283 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2288 pdscp = get_priority(ofport, priority);
2290 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2292 pdscp = xmalloc(sizeof *pdscp);
2293 pdscp->priority = priority;
2295 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2298 if (pdscp->dscp != dscp) {
2300 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2303 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2306 if (!hmap_is_empty(&ofport->priorities)) {
2307 ofport_clear_priorities(ofport);
2308 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2311 hmap_swap(&new, &ofport->priorities);
2319 /* Expires all MAC learning entries associated with 'bundle' and forces its
2320 * ofproto to revalidate every flow.
2322 * Normally MAC learning entries are removed only from the ofproto associated
2323 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2324 * are removed from every ofproto. When patch ports and SLB bonds are in use
2325 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2326 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2327 * with the host from which it migrated. */
2329 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2331 struct ofproto_dpif *ofproto = bundle->ofproto;
2332 struct mac_learning *ml = ofproto->ml;
2333 struct mac_entry *mac, *next_mac;
2335 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2336 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2337 if (mac->port.p == bundle) {
2339 struct ofproto_dpif *o;
2341 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2343 struct mac_entry *e;
2345 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2348 mac_learning_expire(o->ml, e);
2354 mac_learning_expire(ml, mac);
2359 static struct ofbundle *
2360 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2362 struct ofbundle *bundle;
2364 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2365 &ofproto->bundles) {
2366 if (bundle->aux == aux) {
2374 bundle_update(struct ofbundle *bundle)
2376 struct ofport_dpif *port;
2378 bundle->floodable = true;
2379 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2380 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2381 || !stp_forward_in_state(port->stp_state)) {
2382 bundle->floodable = false;
2389 bundle_del_port(struct ofport_dpif *port)
2391 struct ofbundle *bundle = port->bundle;
2393 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2395 list_remove(&port->bundle_node);
2396 port->bundle = NULL;
2399 lacp_slave_unregister(bundle->lacp, port);
2402 bond_slave_unregister(bundle->bond, port);
2405 bundle_update(bundle);
2409 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2410 struct lacp_slave_settings *lacp)
2412 struct ofport_dpif *port;
2414 port = get_ofp_port(bundle->ofproto, ofp_port);
2419 if (port->bundle != bundle) {
2420 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2422 bundle_del_port(port);
2425 port->bundle = bundle;
2426 list_push_back(&bundle->ports, &port->bundle_node);
2427 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2428 || !stp_forward_in_state(port->stp_state)) {
2429 bundle->floodable = false;
2433 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2434 lacp_slave_register(bundle->lacp, port, lacp);
2441 bundle_destroy(struct ofbundle *bundle)
2443 struct ofproto_dpif *ofproto;
2444 struct ofport_dpif *port, *next_port;
2450 ofproto = bundle->ofproto;
2451 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2453 xlate_bundle_remove(bundle);
2455 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2456 bundle_del_port(port);
2459 bundle_flush_macs(bundle, true);
2460 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2462 free(bundle->trunks);
2463 lacp_unref(bundle->lacp);
2464 bond_unref(bundle->bond);
2469 bundle_set(struct ofproto *ofproto_, void *aux,
2470 const struct ofproto_bundle_settings *s)
2472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2473 bool need_flush = false;
2474 struct ofport_dpif *port;
2475 struct ofbundle *bundle;
2476 unsigned long *trunks;
2482 bundle_destroy(bundle_lookup(ofproto, aux));
2486 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2487 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2489 bundle = bundle_lookup(ofproto, aux);
2491 bundle = xmalloc(sizeof *bundle);
2493 bundle->ofproto = ofproto;
2494 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2495 hash_pointer(aux, 0));
2497 bundle->name = NULL;
2499 list_init(&bundle->ports);
2500 bundle->vlan_mode = PORT_VLAN_TRUNK;
2502 bundle->trunks = NULL;
2503 bundle->use_priority_tags = s->use_priority_tags;
2504 bundle->lacp = NULL;
2505 bundle->bond = NULL;
2507 bundle->floodable = true;
2508 mbridge_register_bundle(ofproto->mbridge, bundle);
2511 if (!bundle->name || strcmp(s->name, bundle->name)) {
2513 bundle->name = xstrdup(s->name);
2518 if (!bundle->lacp) {
2519 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2520 bundle->lacp = lacp_create();
2522 lacp_configure(bundle->lacp, s->lacp);
2524 lacp_unref(bundle->lacp);
2525 bundle->lacp = NULL;
2528 /* Update set of ports. */
2530 for (i = 0; i < s->n_slaves; i++) {
2531 if (!bundle_add_port(bundle, s->slaves[i],
2532 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2536 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2537 struct ofport_dpif *next_port;
2539 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2540 for (i = 0; i < s->n_slaves; i++) {
2541 if (s->slaves[i] == port->up.ofp_port) {
2546 bundle_del_port(port);
2550 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2552 if (list_is_empty(&bundle->ports)) {
2553 bundle_destroy(bundle);
2557 /* Set VLAN tagging mode */
2558 if (s->vlan_mode != bundle->vlan_mode
2559 || s->use_priority_tags != bundle->use_priority_tags) {
2560 bundle->vlan_mode = s->vlan_mode;
2561 bundle->use_priority_tags = s->use_priority_tags;
2566 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2567 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2569 if (vlan != bundle->vlan) {
2570 bundle->vlan = vlan;
2574 /* Get trunked VLANs. */
2575 switch (s->vlan_mode) {
2576 case PORT_VLAN_ACCESS:
2580 case PORT_VLAN_TRUNK:
2581 trunks = CONST_CAST(unsigned long *, s->trunks);
2584 case PORT_VLAN_NATIVE_UNTAGGED:
2585 case PORT_VLAN_NATIVE_TAGGED:
2586 if (vlan != 0 && (!s->trunks
2587 || !bitmap_is_set(s->trunks, vlan)
2588 || bitmap_is_set(s->trunks, 0))) {
2589 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2591 trunks = bitmap_clone(s->trunks, 4096);
2593 trunks = bitmap_allocate1(4096);
2595 bitmap_set1(trunks, vlan);
2596 bitmap_set0(trunks, 0);
2598 trunks = CONST_CAST(unsigned long *, s->trunks);
2605 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2606 free(bundle->trunks);
2607 if (trunks == s->trunks) {
2608 bundle->trunks = vlan_bitmap_clone(trunks);
2610 bundle->trunks = trunks;
2615 if (trunks != s->trunks) {
2620 if (!list_is_short(&bundle->ports)) {
2621 bundle->ofproto->has_bonded_bundles = true;
2623 if (bond_reconfigure(bundle->bond, s->bond)) {
2624 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2627 bundle->bond = bond_create(s->bond);
2628 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2631 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2632 bond_slave_register(bundle->bond, port, port->up.netdev);
2635 bond_unref(bundle->bond);
2636 bundle->bond = NULL;
2639 /* If we changed something that would affect MAC learning, un-learn
2640 * everything on this port and force flow revalidation. */
2642 bundle_flush_macs(bundle, false);
2649 bundle_remove(struct ofport *port_)
2651 struct ofport_dpif *port = ofport_dpif_cast(port_);
2652 struct ofbundle *bundle = port->bundle;
2655 bundle_del_port(port);
2656 if (list_is_empty(&bundle->ports)) {
2657 bundle_destroy(bundle);
2658 } else if (list_is_short(&bundle->ports)) {
2659 bond_unref(bundle->bond);
2660 bundle->bond = NULL;
2666 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2668 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2669 struct ofport_dpif *port = port_;
2670 uint8_t ea[ETH_ADDR_LEN];
2673 error = netdev_get_etheraddr(port->up.netdev, ea);
2675 struct ofpbuf packet;
2678 ofpbuf_init(&packet, 0);
2679 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2681 memcpy(packet_pdu, pdu, pdu_size);
2683 send_packet(port, &packet);
2684 ofpbuf_uninit(&packet);
2686 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2687 "%s (%s)", port->bundle->name,
2688 netdev_get_name(port->up.netdev), ovs_strerror(error));
2693 bundle_send_learning_packets(struct ofbundle *bundle)
2695 struct ofproto_dpif *ofproto = bundle->ofproto;
2696 int error, n_packets, n_errors;
2697 struct mac_entry *e;
2699 error = n_packets = n_errors = 0;
2700 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2701 if (e->port.p != bundle) {
2702 struct ofpbuf *learning_packet;
2703 struct ofport_dpif *port;
2707 /* The assignment to "port" is unnecessary but makes "grep"ing for
2708 * struct ofport_dpif more effective. */
2709 learning_packet = bond_compose_learning_packet(bundle->bond,
2713 ret = send_packet(port, learning_packet);
2714 ofpbuf_delete(learning_packet);
2724 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2725 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2726 "packets, last error was: %s",
2727 bundle->name, n_errors, n_packets, ovs_strerror(error));
2729 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2730 bundle->name, n_packets);
2735 bundle_run(struct ofbundle *bundle)
2738 lacp_run(bundle->lacp, send_pdu_cb);
2741 struct ofport_dpif *port;
2743 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2744 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2747 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2748 lacp_status(bundle->lacp));
2749 if (bond_should_send_learning_packets(bundle->bond)) {
2750 bundle_send_learning_packets(bundle);
2756 bundle_wait(struct ofbundle *bundle)
2759 lacp_wait(bundle->lacp);
2762 bond_wait(bundle->bond);
2769 mirror_set__(struct ofproto *ofproto_, void *aux,
2770 const struct ofproto_mirror_settings *s)
2772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2773 struct ofbundle **srcs, **dsts;
2778 mirror_destroy(ofproto->mbridge, aux);
2782 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2783 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2785 for (i = 0; i < s->n_srcs; i++) {
2786 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2789 for (i = 0; i < s->n_dsts; i++) {
2790 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2793 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2794 s->n_dsts, s->src_vlans,
2795 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2802 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2803 uint64_t *packets, uint64_t *bytes)
2806 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2811 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2813 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2814 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2815 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2821 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2823 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2824 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2825 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2829 forward_bpdu_changed(struct ofproto *ofproto_)
2831 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2832 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2836 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840 mac_learning_set_idle_time(ofproto->ml, idle_time);
2841 mac_learning_set_max_entries(ofproto->ml, max_entries);
2846 static struct ofport_dpif *
2847 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2849 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2850 return ofport ? ofport_dpif_cast(ofport) : NULL;
2853 static struct ofport_dpif *
2854 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2856 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2857 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2861 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2862 struct ofproto_port *ofproto_port,
2863 struct dpif_port *dpif_port)
2865 ofproto_port->name = dpif_port->name;
2866 ofproto_port->type = dpif_port->type;
2867 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2871 ofport_update_peer(struct ofport_dpif *ofport)
2873 const struct ofproto_dpif *ofproto;
2874 struct dpif_backer *backer;
2875 const char *peer_name;
2877 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2881 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2882 backer->need_revalidate = REV_RECONFIGURE;
2885 ofport->peer->peer = NULL;
2886 ofport->peer = NULL;
2889 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2894 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2895 struct ofport *peer_ofport;
2896 struct ofport_dpif *peer;
2897 const char *peer_peer;
2899 if (ofproto->backer != backer) {
2903 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2908 peer = ofport_dpif_cast(peer_ofport);
2909 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2910 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2912 ofport->peer = peer;
2913 ofport->peer->peer = ofport;
2921 port_run_fast(struct ofport_dpif *ofport)
2923 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2924 struct ofpbuf packet;
2926 ofpbuf_init(&packet, 0);
2927 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2928 send_packet(ofport, &packet);
2929 ofpbuf_uninit(&packet);
2932 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2933 struct ofpbuf packet;
2935 ofpbuf_init(&packet, 0);
2936 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2937 send_packet(ofport, &packet);
2938 ofpbuf_uninit(&packet);
2943 port_run(struct ofport_dpif *ofport)
2945 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2946 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2947 bool enable = netdev_get_carrier(ofport->up.netdev);
2949 ofport->carrier_seq = carrier_seq;
2951 port_run_fast(ofport);
2954 int cfm_opup = cfm_get_opup(ofport->cfm);
2956 cfm_run(ofport->cfm);
2957 enable = enable && !cfm_get_fault(ofport->cfm);
2959 if (cfm_opup >= 0) {
2960 enable = enable && cfm_opup;
2965 bfd_run(ofport->bfd);
2966 enable = enable && bfd_forwarding(ofport->bfd);
2969 if (ofport->bundle) {
2970 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2971 if (carrier_changed) {
2972 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2976 if (ofport->may_enable != enable) {
2977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2978 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2981 ofport->may_enable = enable;
2985 port_wait(struct ofport_dpif *ofport)
2988 cfm_wait(ofport->cfm);
2992 bfd_wait(ofport->bfd);
2997 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2998 struct ofproto_port *ofproto_port)
3000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3001 struct dpif_port dpif_port;
3004 if (sset_contains(&ofproto->ghost_ports, devname)) {
3005 const char *type = netdev_get_type_from_name(devname);
3007 /* We may be called before ofproto->up.port_by_name is populated with
3008 * the appropriate ofport. For this reason, we must get the name and
3009 * type from the netdev layer directly. */
3011 const struct ofport *ofport;
3013 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3014 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3015 ofproto_port->name = xstrdup(devname);
3016 ofproto_port->type = xstrdup(type);
3022 if (!sset_contains(&ofproto->ports, devname)) {
3025 error = dpif_port_query_by_name(ofproto->backer->dpif,
3026 devname, &dpif_port);
3028 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3034 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3037 const char *devname = netdev_get_name(netdev);
3038 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3039 const char *dp_port_name;
3041 if (netdev_vport_is_patch(netdev)) {
3042 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3046 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3047 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3048 odp_port_t port_no = ODPP_NONE;
3051 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3055 if (netdev_get_tunnel_config(netdev)) {
3056 simap_put(&ofproto->backer->tnl_backers,
3057 dp_port_name, odp_to_u32(port_no));
3061 if (netdev_get_tunnel_config(netdev)) {
3062 sset_add(&ofproto->ghost_ports, devname);
3064 sset_add(&ofproto->ports, devname);
3070 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3073 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3080 sset_find_and_delete(&ofproto->ghost_ports,
3081 netdev_get_name(ofport->up.netdev));
3082 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3083 if (!ofport->is_tunnel) {
3084 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3086 /* The caller is going to close ofport->up.netdev. If this is a
3087 * bonded port, then the bond is using that netdev, so remove it
3088 * from the bond. The client will need to reconfigure everything
3089 * after deleting ports, so then the slave will get re-added. */
3090 bundle_remove(&ofport->up);
3097 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3099 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3104 error = netdev_get_stats(ofport->up.netdev, stats);
3106 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3109 /* ofproto->stats.tx_packets represents packets that we created
3110 * internally and sent to some port (e.g. packets sent with
3111 * send_packet()). Account for them as if they had come from
3112 * OFPP_LOCAL and got forwarded. */
3114 if (stats->rx_packets != UINT64_MAX) {
3115 stats->rx_packets += ofproto->stats.tx_packets;
3118 if (stats->rx_bytes != UINT64_MAX) {
3119 stats->rx_bytes += ofproto->stats.tx_bytes;
3122 /* ofproto->stats.rx_packets represents packets that were received on
3123 * some port and we processed internally and dropped (e.g. STP).
3124 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3126 if (stats->tx_packets != UINT64_MAX) {
3127 stats->tx_packets += ofproto->stats.rx_packets;
3130 if (stats->tx_bytes != UINT64_MAX) {
3131 stats->tx_bytes += ofproto->stats.rx_bytes;
3138 struct port_dump_state {
3143 struct ofproto_port port;
3148 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3150 *statep = xzalloc(sizeof(struct port_dump_state));
3155 port_dump_next(const struct ofproto *ofproto_, void *state_,
3156 struct ofproto_port *port)
3158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3159 struct port_dump_state *state = state_;
3160 const struct sset *sset;
3161 struct sset_node *node;
3163 if (state->has_port) {
3164 ofproto_port_destroy(&state->port);
3165 state->has_port = false;
3167 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3168 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3171 error = port_query_by_name(ofproto_, node->name, &state->port);
3173 *port = state->port;
3174 state->has_port = true;
3176 } else if (error != ENODEV) {
3181 if (!state->ghost) {
3182 state->ghost = true;
3185 return port_dump_next(ofproto_, state_, port);
3192 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3194 struct port_dump_state *state = state_;
3196 if (state->has_port) {
3197 ofproto_port_destroy(&state->port);
3204 port_poll(const struct ofproto *ofproto_, char **devnamep)
3206 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3208 if (ofproto->port_poll_errno) {
3209 int error = ofproto->port_poll_errno;
3210 ofproto->port_poll_errno = 0;
3214 if (sset_is_empty(&ofproto->port_poll_set)) {
3218 *devnamep = sset_pop(&ofproto->port_poll_set);
3223 port_poll_wait(const struct ofproto *ofproto_)
3225 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3226 dpif_port_poll_wait(ofproto->backer->dpif);
3230 port_is_lacp_current(const struct ofport *ofport_)
3232 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3233 return (ofport->bundle && ofport->bundle->lacp
3234 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3238 /* Upcall handling. */
3240 /* Flow miss batching.
3242 * Some dpifs implement operations faster when you hand them off in a batch.
3243 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3244 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3245 * more packets, plus possibly installing the flow in the dpif.
3247 * So far we only batch the operations that affect flow setup time the most.
3248 * It's possible to batch more than that, but the benefit might be minimal. */
3250 struct hmap_node hmap_node;
3251 struct ofproto_dpif *ofproto;
3253 enum odp_key_fitness key_fitness;
3254 const struct nlattr *key;
3256 struct list packets;
3257 enum dpif_upcall_type upcall_type;
3260 struct flow_miss_op {
3261 struct dpif_op dpif_op;
3263 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3264 struct xlate_out xout;
3265 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3267 struct ofpbuf mask; /* Flow mask for "put" ops. */
3268 struct odputil_keybuf maskbuf;
3270 /* If this is a "put" op, then a pointer to the subfacet that should
3271 * be marked as uninstalled if the operation fails. */
3272 struct subfacet *subfacet;
3275 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3276 * OpenFlow controller as necessary according to their individual
3277 * configurations. */
3279 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3280 const struct flow *flow)
3282 struct ofputil_packet_in pin;
3284 pin.packet = packet->data;
3285 pin.packet_len = packet->size;
3286 pin.reason = OFPR_NO_MATCH;
3287 pin.controller_id = 0;
3292 pin.send_len = 0; /* not used for flow table misses */
3294 flow_get_metadata(flow, &pin.fmd);
3296 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3299 static struct flow_miss *
3300 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3301 const struct flow *flow, uint32_t hash)
3303 struct flow_miss *miss;
3305 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3306 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3314 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3315 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3316 * 'miss' is associated with a subfacet the caller must also initialize the
3317 * returned op->subfacet, and if anything needs to be freed after processing
3318 * the op, the caller must initialize op->garbage also. */
3320 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3321 struct flow_miss_op *op)
3323 if (miss->flow.in_port.ofp_port
3324 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3325 miss->flow.vlan_tci)) {
3326 /* This packet was received on a VLAN splinter port. We
3327 * added a VLAN to the packet to make the packet resemble
3328 * the flow, but the actions were composed assuming that
3329 * the packet contained no VLAN. So, we must remove the
3330 * VLAN header from the packet before trying to execute the
3332 eth_pop_vlan(packet);
3335 op->subfacet = NULL;
3336 op->xout_garbage = false;
3337 op->dpif_op.type = DPIF_OP_EXECUTE;
3338 op->dpif_op.u.execute.key = miss->key;
3339 op->dpif_op.u.execute.key_len = miss->key_len;
3340 op->dpif_op.u.execute.packet = packet;
3341 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3344 /* Helper for handle_flow_miss_without_facet() and
3345 * handle_flow_miss_with_facet(). */
3347 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3348 const struct flow *flow, bool fail_open)
3352 * Extra-special case for fail-open mode.
3354 * We are in fail-open mode and the packet matched the fail-open
3355 * rule, but we are connected to a controller too. We should send
3356 * the packet up to the controller in the hope that it will try to
3357 * set up a flow and thereby allow us to exit fail-open.
3359 * See the top-level comment in fail-open.c for more information.
3361 send_packet_in_miss(ofproto, packet, flow);
3365 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3366 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3367 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3368 * return value of true). However, for short flows the cost of bookkeeping is
3369 * much higher than the benefits, so when the datapath holds a large number of
3370 * flows we impose some heuristics to decide which flows are likely to be worth
3373 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3375 struct dpif_backer *backer = miss->ofproto->backer;
3378 switch (flow_miss_model) {
3379 case OFPROTO_HANDLE_MISS_AUTO:
3381 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3383 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3387 if (!backer->governor) {
3390 n_subfacets = hmap_count(&backer->subfacets);
3391 if (n_subfacets * 2 <= flow_eviction_threshold) {
3395 backer->governor = governor_create();
3398 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3399 return governor_should_install_flow(backer->governor, hash,
3400 list_size(&miss->packets));
3403 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3404 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3405 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3407 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3408 struct flow_miss *miss,
3409 struct flow_miss_op *ops, size_t *n_ops)
3411 struct ofpbuf *packet;
3413 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3415 COVERAGE_INC(facet_suppress);
3417 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3418 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3421 struct xlate_in xin;
3423 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3424 xlate_actions_for_side_effects(&xin);
3427 if (xout->odp_actions.size) {
3428 struct flow_miss_op *op = &ops[*n_ops];
3429 struct dpif_execute *execute = &op->dpif_op.u.execute;
3431 init_flow_miss_execute_op(miss, packet, op);
3432 xlate_out_copy(&op->xout, xout);
3433 execute->actions = op->xout.odp_actions.data;
3434 execute->actions_len = op->xout.odp_actions.size;
3435 op->xout_garbage = true;
3442 /* Handles 'miss', which matches 'facet'. May add any required datapath
3443 * operations to 'ops', incrementing '*n_ops' for each new op.
3445 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3446 * This is really important only for new facets: if we just called time_msec()
3447 * here, then the new subfacet or its packets could look (occasionally) as
3448 * though it was used some time after the facet was used. That can make a
3449 * one-packet flow look like it has a nonzero duration, which looks odd in
3450 * e.g. NetFlow statistics.
3452 * If non-null, 'stats' will be folded into 'facet'. */
3454 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3455 long long int now, struct dpif_flow_stats *stats,
3456 struct flow_miss_op *ops, size_t *n_ops)
3458 enum subfacet_path want_path;
3459 struct subfacet *subfacet;
3460 struct ofpbuf *packet;
3462 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3464 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3465 struct flow_miss_op *op = &ops[*n_ops];
3467 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3470 if (want_path != SF_FAST_PATH) {
3471 struct rule_dpif *rule;
3472 struct xlate_in xin;
3474 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3475 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3476 xlate_actions_for_side_effects(&xin);
3479 if (facet->xout.odp_actions.size) {
3480 struct dpif_execute *execute = &op->dpif_op.u.execute;
3482 init_flow_miss_execute_op(miss, packet, op);
3483 execute->actions = facet->xout.odp_actions.data,
3484 execute->actions_len = facet->xout.odp_actions.size;
3489 /* Don't install the flow if it's the result of the "userspace"
3490 * action for an already installed facet. This can occur when a
3491 * datapath flow with wildcards has a "userspace" action and flows
3492 * sent to userspace result in a different subfacet, which will then
3493 * be rejected as overlapping by the datapath. */
3494 if (miss->upcall_type == DPIF_UC_ACTION
3495 && !list_is_empty(&facet->subfacets)) {
3497 facet->used = MAX(facet->used, stats->used);
3498 facet->packet_count += stats->n_packets;
3499 facet->byte_count += stats->n_bytes;
3500 facet->tcp_flags |= stats->tcp_flags;
3505 subfacet = subfacet_create(facet, miss, now);
3507 subfacet_update_stats(subfacet, stats);
3510 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3511 struct flow_miss_op *op = &ops[(*n_ops)++];
3512 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3514 subfacet->path = want_path;
3516 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3517 if (enable_megaflows) {
3518 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3519 &miss->flow, UINT32_MAX);
3522 op->xout_garbage = false;
3523 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3524 op->subfacet = subfacet;
3525 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3526 put->key = miss->key;
3527 put->key_len = miss->key_len;
3528 put->mask = op->mask.data;
3529 put->mask_len = op->mask.size;
3531 if (want_path == SF_FAST_PATH) {
3532 put->actions = facet->xout.odp_actions.data;
3533 put->actions_len = facet->xout.odp_actions.size;
3535 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3536 op->slow_stub, sizeof op->slow_stub,
3537 &put->actions, &put->actions_len);
3543 /* Handles flow miss 'miss'. May add any required datapath operations
3544 * to 'ops', incrementing '*n_ops' for each new op. */
3546 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3549 struct ofproto_dpif *ofproto = miss->ofproto;
3550 struct dpif_flow_stats stats__;
3551 struct dpif_flow_stats *stats = &stats__;
3552 struct ofpbuf *packet;
3553 struct facet *facet;
3557 memset(stats, 0, sizeof *stats);
3559 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3560 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3561 stats->n_bytes += packet->size;
3565 facet = facet_lookup_valid(ofproto, &miss->flow);
3567 struct flow_wildcards wc;
3568 struct rule_dpif *rule;
3569 struct xlate_out xout;
3570 struct xlate_in xin;
3572 flow_wildcards_init_catchall(&wc);
3573 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3574 rule_credit_stats(rule, stats);
3576 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3578 xin.resubmit_stats = stats;
3579 xin.may_learn = true;
3580 xlate_actions(&xin, &xout);
3581 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3583 /* There does not exist a bijection between 'struct flow' and datapath
3584 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3585 * assumption used throughout the facet and subfacet handling code.
3586 * Since we have to handle these misses in userspace anyway, we simply
3587 * skip facet creation, avoiding the problem altogether. */
3588 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3589 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3590 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3594 facet = facet_create(miss, rule, &xout, stats);
3597 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3600 static struct drop_key *
3601 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3604 struct drop_key *drop_key;
3606 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3607 &backer->drop_keys) {
3608 if (drop_key->key_len == key_len
3609 && !memcmp(drop_key->key, key, key_len)) {
3617 drop_key_clear(struct dpif_backer *backer)
3619 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3620 struct drop_key *drop_key, *next;
3622 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3625 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3627 if (error && !VLOG_DROP_WARN(&rl)) {
3628 struct ds ds = DS_EMPTY_INITIALIZER;
3629 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3630 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3631 ovs_strerror(error), ds_cstr(&ds));
3635 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3636 free(drop_key->key);
3641 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3642 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3643 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3644 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3645 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3646 * 'packet' ingressed.
3648 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3649 * 'flow''s in_port to OFPP_NONE.
3651 * This function does post-processing on data returned from
3652 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3653 * of the upcall processing logic. In particular, if the extracted in_port is
3654 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3655 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3656 * a VLAN header onto 'packet' (if it is nonnull).
3658 * Similarly, this function also includes some logic to help with tunnels. It
3659 * may modify 'flow' as necessary to make the tunneling implementation
3660 * transparent to the upcall processing logic.
3662 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3663 * or some other positive errno if there are other problems. */
3665 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3666 const struct nlattr *key, size_t key_len,
3667 struct flow *flow, enum odp_key_fitness *fitnessp,
3668 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3670 const struct ofport_dpif *port;
3671 enum odp_key_fitness fitness;
3674 fitness = odp_flow_key_to_flow(key, key_len, flow);
3675 if (fitness == ODP_FIT_ERROR) {
3681 *odp_in_port = flow->in_port.odp_port;
3684 port = (tnl_port_should_receive(flow)
3685 ? tnl_port_receive(flow)
3686 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3687 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3692 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3693 * it's theoretically possible that we'll receive an ofport belonging to an
3694 * entirely different datapath. In practice, this can't happen because no
3695 * platforms has two separate datapaths which each support tunneling. */
3696 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3698 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3700 /* Make the packet resemble the flow, so that it gets sent to
3701 * an OpenFlow controller properly, so that it looks correct
3702 * for sFlow, and so that flow_extract() will get the correct
3703 * vlan_tci if it is called on 'packet'.
3705 * The allocated space inside 'packet' probably also contains
3706 * 'key', that is, both 'packet' and 'key' are probably part of
3707 * a struct dpif_upcall (see the large comment on that
3708 * structure definition), so pushing data on 'packet' is in
3709 * general not a good idea since it could overwrite 'key' or
3710 * free it as a side effect. However, it's OK in this special
3711 * case because we know that 'packet' is inside a Netlink
3712 * attribute: pushing 4 bytes will just overwrite the 4-byte
3713 * "struct nlattr", which is fine since we don't need that
3714 * header anymore. */
3715 eth_push_vlan(packet, flow->vlan_tci);
3717 /* We can't reproduce 'key' from 'flow'. */
3718 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3723 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3728 *fitnessp = fitness;
3734 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3737 struct dpif_upcall *upcall;
3738 struct flow_miss *miss;
3739 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3740 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3741 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3751 /* Construct the to-do list.
3753 * This just amounts to extracting the flow from each packet and sticking
3754 * the packets that have the same flow in the same "flow_miss" structure so
3755 * that we can process them together. */
3758 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3759 struct flow_miss *miss = &misses[n_misses];
3760 struct flow_miss *existing_miss;
3761 struct ofproto_dpif *ofproto;
3762 odp_port_t odp_in_port;
3767 error = ofproto_receive(backer, upcall->packet, upcall->key,
3768 upcall->key_len, &flow, &miss->key_fitness,
3769 &ofproto, &odp_in_port);
3770 if (error == ENODEV) {
3771 struct drop_key *drop_key;
3773 /* Received packet on datapath port for which we couldn't
3774 * associate an ofproto. This can happen if a port is removed
3775 * while traffic is being received. Print a rate-limited message
3776 * in case it happens frequently. Install a drop flow so
3777 * that future packets of the flow are inexpensively dropped
3779 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3780 "%"PRIu32, odp_in_port);
3782 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3784 drop_key = xmalloc(sizeof *drop_key);
3785 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3786 drop_key->key_len = upcall->key_len;
3788 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3789 hash_bytes(drop_key->key, drop_key->key_len, 0));
3790 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3791 drop_key->key, drop_key->key_len,
3792 NULL, 0, NULL, 0, NULL);
3800 ofproto->n_missed++;
3801 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3802 &flow.tunnel, &flow.in_port, &miss->flow);
3804 /* Add other packets to a to-do list. */
3805 hash = flow_hash(&miss->flow, 0);
3806 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3807 if (!existing_miss) {
3808 hmap_insert(&todo, &miss->hmap_node, hash);
3809 miss->ofproto = ofproto;
3810 miss->key = upcall->key;
3811 miss->key_len = upcall->key_len;
3812 miss->upcall_type = upcall->type;
3813 list_init(&miss->packets);
3817 miss = existing_miss;
3819 list_push_back(&miss->packets, &upcall->packet->list_node);
3822 /* Process each element in the to-do list, constructing the set of
3823 * operations to batch. */
3825 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3826 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3828 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3830 /* Execute batch. */
3831 for (i = 0; i < n_ops; i++) {
3832 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3834 dpif_operate(backer->dpif, dpif_ops, n_ops);
3836 for (i = 0; i < n_ops; i++) {
3837 if (dpif_ops[i]->error != 0
3838 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3839 && flow_miss_ops[i].subfacet) {
3840 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3842 COVERAGE_INC(subfacet_install_fail);
3844 /* Zero-out subfacet counters when installation failed, but
3845 * datapath reported hits. This should not happen and
3846 * indicates a bug, since if the datapath flow exists, we
3847 * should not be attempting to create a new subfacet. A
3848 * buggy datapath could trigger this, so just zero out the
3849 * counters and log an error. */
3850 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3851 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3852 "datapath reported hits");
3853 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3856 subfacet->path = SF_NOT_INSTALLED;
3860 if (flow_miss_ops[i].xout_garbage) {
3861 xlate_out_uninit(&flow_miss_ops[i].xout);
3864 hmap_destroy(&todo);
3867 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3869 classify_upcall(const struct dpif_upcall *upcall)
3871 size_t userdata_len;
3872 union user_action_cookie cookie;
3874 /* First look at the upcall type. */
3875 switch (upcall->type) {
3876 case DPIF_UC_ACTION:
3882 case DPIF_N_UC_TYPES:
3884 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3888 /* "action" upcalls need a closer look. */
3889 if (!upcall->userdata) {
3890 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3893 userdata_len = nl_attr_get_size(upcall->userdata);
3894 if (userdata_len < sizeof cookie.type
3895 || userdata_len > sizeof cookie) {
3896 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3900 memset(&cookie, 0, sizeof cookie);
3901 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3902 if (userdata_len == sizeof cookie.sflow
3903 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3904 return SFLOW_UPCALL;
3905 } else if (userdata_len == sizeof cookie.slow_path
3906 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3908 } else if (userdata_len == sizeof cookie.flow_sample
3909 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3910 return FLOW_SAMPLE_UPCALL;
3911 } else if (userdata_len == sizeof cookie.ipfix
3912 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3913 return IPFIX_UPCALL;
3915 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3916 " and size %zu", cookie.type, userdata_len);
3922 handle_sflow_upcall(struct dpif_backer *backer,
3923 const struct dpif_upcall *upcall)
3925 struct ofproto_dpif *ofproto;
3926 union user_action_cookie cookie;
3928 odp_port_t odp_in_port;
3930 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3931 &flow, NULL, &ofproto, &odp_in_port)
3932 || !ofproto->sflow) {
3936 memset(&cookie, 0, sizeof cookie);
3937 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3938 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3939 odp_in_port, &cookie);
3943 handle_flow_sample_upcall(struct dpif_backer *backer,
3944 const struct dpif_upcall *upcall)
3946 struct ofproto_dpif *ofproto;
3947 union user_action_cookie cookie;
3950 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3951 &flow, NULL, &ofproto, NULL)
3952 || !ofproto->ipfix) {
3956 memset(&cookie, 0, sizeof cookie);
3957 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3959 /* The flow reflects exactly the contents of the packet. Sample
3960 * the packet using it. */
3961 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3962 cookie.flow_sample.collector_set_id,
3963 cookie.flow_sample.probability,
3964 cookie.flow_sample.obs_domain_id,
3965 cookie.flow_sample.obs_point_id);
3969 handle_ipfix_upcall(struct dpif_backer *backer,
3970 const struct dpif_upcall *upcall)
3972 struct ofproto_dpif *ofproto;
3975 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3976 &flow, NULL, &ofproto, NULL)
3977 || !ofproto->ipfix) {
3981 /* The flow reflects exactly the contents of the packet. Sample
3982 * the packet using it. */
3983 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3987 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3989 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3990 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3991 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3996 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3999 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4000 struct dpif_upcall *upcall = &misses[n_misses];
4001 struct ofpbuf *buf = &miss_bufs[n_misses];
4004 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4005 sizeof miss_buf_stubs[n_misses]);
4006 error = dpif_recv(backer->dpif, upcall, buf);
4012 switch (classify_upcall(upcall)) {
4014 /* Handle it later. */
4019 handle_sflow_upcall(backer, upcall);
4023 case FLOW_SAMPLE_UPCALL:
4024 handle_flow_sample_upcall(backer, upcall);
4029 handle_ipfix_upcall(backer, upcall);
4039 /* Handle deferred MISS_UPCALL processing. */
4040 handle_miss_upcalls(backer, misses, n_misses);
4041 for (i = 0; i < n_misses; i++) {
4042 ofpbuf_uninit(&miss_bufs[i]);
4048 /* Flow expiration. */
4050 static int subfacet_max_idle(const struct dpif_backer *);
4051 static void update_stats(struct dpif_backer *);
4052 static void rule_expire(struct rule_dpif *);
4053 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4055 /* This function is called periodically by run(). Its job is to collect
4056 * updates for the flows that have been installed into the datapath, most
4057 * importantly when they last were used, and then use that information to
4058 * expire flows that have not been used recently.
4060 * Returns the number of milliseconds after which it should be called again. */
4062 expire(struct dpif_backer *backer)
4064 struct ofproto_dpif *ofproto;
4068 /* Periodically clear out the drop keys in an effort to keep them
4069 * relatively few. */
4070 drop_key_clear(backer);
4072 /* Update stats for each flow in the backer. */
4073 update_stats(backer);
4075 n_subfacets = hmap_count(&backer->subfacets);
4077 struct subfacet *subfacet;
4078 long long int total, now;
4082 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4083 total += now - subfacet->created;
4085 backer->avg_subfacet_life += total / n_subfacets;
4087 backer->avg_subfacet_life /= 2;
4089 backer->avg_n_subfacet += n_subfacets;
4090 backer->avg_n_subfacet /= 2;
4092 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4094 max_idle = subfacet_max_idle(backer);
4095 expire_subfacets(backer, max_idle);
4097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4098 struct rule *rule, *next_rule;
4100 if (ofproto->backer != backer) {
4104 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4106 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4107 &ofproto->up.expirable) {
4108 rule_expire(rule_dpif_cast(rule));
4111 /* All outstanding data in existing flows has been accounted, so it's a
4112 * good time to do bond rebalancing. */
4113 if (ofproto->has_bonded_bundles) {
4114 struct ofbundle *bundle;
4116 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4118 bond_rebalance(bundle->bond, &backer->revalidate_set);
4124 return MIN(max_idle, 1000);
4127 /* Updates flow table statistics given that the datapath just reported 'stats'
4128 * as 'subfacet''s statistics. */
4130 update_subfacet_stats(struct subfacet *subfacet,
4131 const struct dpif_flow_stats *stats)
4133 struct facet *facet = subfacet->facet;
4134 struct dpif_flow_stats diff;
4136 diff.tcp_flags = stats->tcp_flags;
4137 diff.used = stats->used;
4139 if (stats->n_packets >= subfacet->dp_packet_count) {
4140 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4142 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4146 if (stats->n_bytes >= subfacet->dp_byte_count) {
4147 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4149 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4153 facet->ofproto->n_hit += diff.n_packets;
4154 subfacet->dp_packet_count = stats->n_packets;
4155 subfacet->dp_byte_count = stats->n_bytes;
4156 subfacet_update_stats(subfacet, &diff);
4158 if (facet->accounted_bytes < facet->byte_count) {
4160 facet_account(facet);
4161 facet->accounted_bytes = facet->byte_count;
4165 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4166 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4168 delete_unexpected_flow(struct dpif_backer *backer,
4169 const struct nlattr *key, size_t key_len)
4171 if (!VLOG_DROP_WARN(&rl)) {
4175 odp_flow_key_format(key, key_len, &s);
4176 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4180 COVERAGE_INC(facet_unexpected);
4181 dpif_flow_del(backer->dpif, key, key_len, NULL);
4184 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4186 * This function also pushes statistics updates to rules which each facet
4187 * resubmits into. Generally these statistics will be accurate. However, if a
4188 * facet changes the rule it resubmits into at some time in between
4189 * update_stats() runs, it is possible that statistics accrued to the
4190 * old rule will be incorrectly attributed to the new rule. This could be
4191 * avoided by calling update_stats() whenever rules are created or
4192 * deleted. However, the performance impact of making so many calls to the
4193 * datapath do not justify the benefit of having perfectly accurate statistics.
4195 * In addition, this function maintains per ofproto flow hit counts. The patch
4196 * port is not treated specially. e.g. A packet ingress from br0 patched into
4197 * br1 will increase the hit count of br0 by 1, however, does not affect
4198 * the hit or miss counts of br1.
4201 update_stats(struct dpif_backer *backer)
4203 const struct dpif_flow_stats *stats;
4204 struct dpif_flow_dump dump;
4205 const struct nlattr *key, *mask;
4206 size_t key_len, mask_len;
4208 dpif_flow_dump_start(&dump, backer->dpif);
4209 while (dpif_flow_dump_next(&dump, &key, &key_len,
4210 &mask, &mask_len, NULL, NULL, &stats)) {
4211 struct subfacet *subfacet;
4214 key_hash = odp_flow_key_hash(key, key_len);
4215 subfacet = subfacet_find(backer, key, key_len, key_hash);
4216 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4218 update_subfacet_stats(subfacet, stats);
4222 /* Stats are updated per-packet. */
4225 case SF_NOT_INSTALLED:
4227 delete_unexpected_flow(backer, key, key_len);
4232 dpif_flow_dump_done(&dump);
4234 update_moving_averages(backer);
4237 /* Calculates and returns the number of milliseconds of idle time after which
4238 * subfacets should expire from the datapath. When a subfacet expires, we fold
4239 * its statistics into its facet, and when a facet's last subfacet expires, we
4240 * fold its statistic into its rule. */
4242 subfacet_max_idle(const struct dpif_backer *backer)
4245 * Idle time histogram.
4247 * Most of the time a switch has a relatively small number of subfacets.
4248 * When this is the case we might as well keep statistics for all of them
4249 * in userspace and to cache them in the kernel datapath for performance as
4252 * As the number of subfacets increases, the memory required to maintain
4253 * statistics about them in userspace and in the kernel becomes
4254 * significant. However, with a large number of subfacets it is likely
4255 * that only a few of them are "heavy hitters" that consume a large amount
4256 * of bandwidth. At this point, only heavy hitters are worth caching in
4257 * the kernel and maintaining in userspaces; other subfacets we can
4260 * The technique used to compute the idle time is to build a histogram with
4261 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4262 * that is installed in the kernel gets dropped in the appropriate bucket.
4263 * After the histogram has been built, we compute the cutoff so that only
4264 * the most-recently-used 1% of subfacets (but at least
4265 * flow_eviction_threshold flows) are kept cached. At least
4266 * the most-recently-used bucket of subfacets is kept, so actually an
4267 * arbitrary number of subfacets can be kept in any given expiration run
4268 * (though the next run will delete most of those unless they receive
4271 * This requires a second pass through the subfacets, in addition to the
4272 * pass made by update_stats(), because the former function never looks at
4273 * uninstallable subfacets.
4275 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4276 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4277 int buckets[N_BUCKETS] = { 0 };
4278 int total, subtotal, bucket;
4279 struct subfacet *subfacet;
4283 total = hmap_count(&backer->subfacets);
4284 if (total <= flow_eviction_threshold) {
4285 return N_BUCKETS * BUCKET_WIDTH;
4288 /* Build histogram. */
4290 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4291 long long int idle = now - subfacet->used;
4292 int bucket = (idle <= 0 ? 0
4293 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4294 : (unsigned int) idle / BUCKET_WIDTH);
4298 /* Find the first bucket whose flows should be expired. */
4299 subtotal = bucket = 0;
4301 subtotal += buckets[bucket++];
4302 } while (bucket < N_BUCKETS &&
4303 subtotal < MAX(flow_eviction_threshold, total / 100));
4305 if (VLOG_IS_DBG_ENABLED()) {
4309 ds_put_cstr(&s, "keep");
4310 for (i = 0; i < N_BUCKETS; i++) {
4312 ds_put_cstr(&s, ", drop");
4315 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4318 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4322 return bucket * BUCKET_WIDTH;
4326 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4328 /* Cutoff time for most flows. */
4329 long long int normal_cutoff = time_msec() - dp_max_idle;
4331 /* We really want to keep flows for special protocols around, so use a more
4332 * conservative cutoff. */
4333 long long int special_cutoff = time_msec() - 10000;
4335 struct subfacet *subfacet, *next_subfacet;
4336 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4340 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4341 &backer->subfacets) {
4342 long long int cutoff;
4344 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4348 if (subfacet->used < cutoff) {
4349 if (subfacet->path != SF_NOT_INSTALLED) {
4350 batch[n_batch++] = subfacet;
4351 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4352 subfacet_destroy_batch(backer, batch, n_batch);
4356 subfacet_destroy(subfacet);
4362 subfacet_destroy_batch(backer, batch, n_batch);
4366 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4367 * then delete it entirely. */
4369 rule_expire(struct rule_dpif *rule)
4374 if (rule->up.pending) {
4375 /* We'll have to expire it later. */
4379 /* Has 'rule' expired? */
4381 if (rule->up.hard_timeout
4382 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4383 reason = OFPRR_HARD_TIMEOUT;
4384 } else if (rule->up.idle_timeout
4385 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4386 reason = OFPRR_IDLE_TIMEOUT;
4391 COVERAGE_INC(ofproto_dpif_expired);
4393 /* Get rid of the rule. */
4394 ofproto_rule_expire(&rule->up, reason);
4399 /* Creates and returns a new facet based on 'miss'.
4401 * The caller must already have determined that no facet with an identical
4402 * 'miss->flow' exists in 'miss->ofproto'.
4404 * 'rule' and 'xout' must have been created based on 'miss'.
4406 * 'facet'' statistics are initialized based on 'stats'.
4408 * The facet will initially have no subfacets. The caller should create (at
4409 * least) one subfacet with subfacet_create(). */
4410 static struct facet *
4411 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4412 struct xlate_out *xout, struct dpif_flow_stats *stats)
4414 struct ofproto_dpif *ofproto = miss->ofproto;
4415 struct facet *facet;
4418 facet = xzalloc(sizeof *facet);
4419 facet->ofproto = miss->ofproto;
4420 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4421 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4422 facet->tcp_flags = stats->tcp_flags;
4423 facet->used = stats->used;
4424 facet->flow = miss->flow;
4425 facet->learn_rl = time_msec() + 500;
4427 list_init(&facet->subfacets);
4428 netflow_flow_init(&facet->nf_flow);
4429 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4431 xlate_out_copy(&facet->xout, xout);
4433 match_init(&match, &facet->flow, &facet->xout.wc);
4434 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4435 classifier_insert(&ofproto->facets, &facet->cr);
4437 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4438 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4444 facet_free(struct facet *facet)
4447 xlate_out_uninit(&facet->xout);
4452 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4453 * 'packet', which arrived on 'in_port'. */
4455 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4456 const struct nlattr *odp_actions, size_t actions_len,
4457 struct ofpbuf *packet)
4459 struct odputil_keybuf keybuf;
4463 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4464 odp_flow_key_from_flow(&key, flow,
4465 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4467 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4468 odp_actions, actions_len, packet);
4472 /* Remove 'facet' from its ofproto and free up the associated memory:
4474 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4475 * rule's statistics, via subfacet_uninstall().
4477 * - Removes 'facet' from its rule and from ofproto->facets.
4480 facet_remove(struct facet *facet)
4482 struct subfacet *subfacet, *next_subfacet;
4484 ovs_assert(!list_is_empty(&facet->subfacets));
4486 /* First uninstall all of the subfacets to get final statistics. */
4487 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4488 subfacet_uninstall(subfacet);
4491 /* Flush the final stats to the rule.
4493 * This might require us to have at least one subfacet around so that we
4494 * can use its actions for accounting in facet_account(), which is why we
4495 * have uninstalled but not yet destroyed the subfacets. */
4496 facet_flush_stats(facet);
4498 /* Now we're really all done so destroy everything. */
4499 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4500 &facet->subfacets) {
4501 subfacet_destroy__(subfacet);
4503 classifier_remove(&facet->ofproto->facets, &facet->cr);
4504 cls_rule_destroy(&facet->cr);
4508 /* Feed information from 'facet' back into the learning table to keep it in
4509 * sync with what is actually flowing through the datapath. */
4511 facet_learn(struct facet *facet)
4513 long long int now = time_msec();
4515 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4519 facet->learn_rl = now + 500;
4521 if (!facet->xout.has_learn
4522 && !facet->xout.has_normal
4523 && (!facet->xout.has_fin_timeout
4524 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4528 facet_push_stats(facet, true);
4532 facet_account(struct facet *facet)
4534 const struct nlattr *a;
4539 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4542 n_bytes = facet->byte_count - facet->accounted_bytes;
4544 /* This loop feeds byte counters to bond_account() for rebalancing to use
4545 * as a basis. We also need to track the actual VLAN on which the packet
4546 * is going to be sent to ensure that it matches the one passed to
4547 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4550 * We use the actions from an arbitrary subfacet because they should all
4551 * be equally valid for our purpose. */
4552 vlan_tci = facet->flow.vlan_tci;
4553 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4554 facet->xout.odp_actions.size) {
4555 const struct ovs_action_push_vlan *vlan;
4556 struct ofport_dpif *port;
4558 switch (nl_attr_type(a)) {
4559 case OVS_ACTION_ATTR_OUTPUT:
4560 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4561 if (port && port->bundle && port->bundle->bond) {
4562 bond_account(port->bundle->bond, &facet->flow,
4563 vlan_tci_to_vid(vlan_tci), n_bytes);
4567 case OVS_ACTION_ATTR_POP_VLAN:
4568 vlan_tci = htons(0);
4571 case OVS_ACTION_ATTR_PUSH_VLAN:
4572 vlan = nl_attr_get(a);
4573 vlan_tci = vlan->vlan_tci;
4579 /* Returns true if the only action for 'facet' is to send to the controller.
4580 * (We don't report NetFlow expiration messages for such facets because they
4581 * are just part of the control logic for the network, not real traffic). */
4583 facet_is_controller_flow(struct facet *facet)
4586 struct ofproto_dpif *ofproto = facet->ofproto;
4587 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4589 const struct ofpact *ofpacts = rule->up.ofpacts;
4590 size_t ofpacts_len = rule->up.ofpacts_len;
4592 if (ofpacts_len > 0 &&
4593 ofpacts->type == OFPACT_CONTROLLER &&
4594 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4601 /* Folds all of 'facet''s statistics into its rule. Also updates the
4602 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4603 * 'facet''s statistics in the datapath should have been zeroed and folded into
4604 * its packet and byte counts before this function is called. */
4606 facet_flush_stats(struct facet *facet)
4608 struct ofproto_dpif *ofproto = facet->ofproto;
4609 struct subfacet *subfacet;
4611 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4612 ovs_assert(!subfacet->dp_byte_count);
4613 ovs_assert(!subfacet->dp_packet_count);
4616 facet_push_stats(facet, false);
4617 if (facet->accounted_bytes < facet->byte_count) {
4618 facet_account(facet);
4619 facet->accounted_bytes = facet->byte_count;
4622 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4623 struct ofexpired expired;
4624 expired.flow = facet->flow;
4625 expired.packet_count = facet->packet_count;
4626 expired.byte_count = facet->byte_count;
4627 expired.used = facet->used;
4628 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4631 /* Reset counters to prevent double counting if 'facet' ever gets
4633 facet_reset_counters(facet);
4635 netflow_flow_clear(&facet->nf_flow);
4636 facet->tcp_flags = 0;
4639 /* Searches 'ofproto''s table of facets for one which would be responsible for
4640 * 'flow'. Returns it if found, otherwise a null pointer.
4642 * The returned facet might need revalidation; use facet_lookup_valid()
4643 * instead if that is important. */
4644 static struct facet *
4645 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4647 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4648 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4651 /* Searches 'ofproto''s table of facets for one capable that covers
4652 * 'flow'. Returns it if found, otherwise a null pointer.
4654 * The returned facet is guaranteed to be valid. */
4655 static struct facet *
4656 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4658 struct facet *facet;
4660 facet = facet_find(ofproto, flow);
4662 && (ofproto->backer->need_revalidate
4663 || tag_set_intersects(&ofproto->backer->revalidate_set,
4665 && !facet_revalidate(facet)) {
4673 facet_check_consistency(struct facet *facet)
4675 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4677 struct xlate_out xout;
4678 struct xlate_in xin;
4680 struct rule_dpif *rule;
4683 /* Check the datapath actions for consistency. */
4684 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4685 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4686 xlate_actions(&xin, &xout);
4688 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4689 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4690 && facet->xout.slow == xout.slow
4691 && facet->fail_open == fail_open;
4692 if (!ok && !VLOG_DROP_WARN(&rl)) {
4693 struct ds s = DS_EMPTY_INITIALIZER;
4695 flow_format(&s, &facet->flow);
4696 ds_put_cstr(&s, ": inconsistency in facet");
4698 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4699 ds_put_cstr(&s, " (actions were: ");
4700 format_odp_actions(&s, facet->xout.odp_actions.data,
4701 facet->xout.odp_actions.size);
4702 ds_put_cstr(&s, ") (correct actions: ");
4703 format_odp_actions(&s, xout.odp_actions.data,
4704 xout.odp_actions.size);
4705 ds_put_char(&s, ')');
4708 if (facet->xout.slow != xout.slow) {
4709 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4712 if (facet->fail_open != fail_open) {
4713 ds_put_format(&s, " fail open incorrect. should be %s",
4714 fail_open ? "true" : "false");
4718 xlate_out_uninit(&xout);
4723 /* Re-searches the classifier for 'facet':
4725 * - If the rule found is different from 'facet''s current rule, moves
4726 * 'facet' to the new rule and recompiles its actions.
4728 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4729 * where it is and recompiles its actions anyway.
4731 * - If any of 'facet''s subfacets correspond to a new flow according to
4732 * ofproto_receive(), 'facet' is removed.
4734 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4736 facet_revalidate(struct facet *facet)
4738 struct ofproto_dpif *ofproto = facet->ofproto;
4739 struct rule_dpif *new_rule;
4740 struct subfacet *subfacet;
4741 struct flow_wildcards wc;
4742 struct xlate_out xout;
4743 struct xlate_in xin;
4745 COVERAGE_INC(facet_revalidate);
4747 /* Check that child subfacets still correspond to this facet. Tunnel
4748 * configuration changes could cause a subfacet's OpenFlow in_port to
4750 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4751 struct ofproto_dpif *recv_ofproto;
4752 struct flow recv_flow;
4755 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4756 subfacet->key_len, &recv_flow, NULL,
4757 &recv_ofproto, NULL);
4759 || recv_ofproto != ofproto
4760 || facet != facet_find(ofproto, &recv_flow)) {
4761 facet_remove(facet);
4766 flow_wildcards_init_catchall(&wc);
4767 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4769 /* Calculate new datapath actions.
4771 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4772 * emit a NetFlow expiration and, if so, we need to have the old state
4773 * around to properly compose it. */
4774 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4775 xlate_actions(&xin, &xout);
4776 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4778 /* A facet's slow path reason should only change under dramatic
4779 * circumstances. Rather than try to update everything, it's simpler to
4780 * remove the facet and start over.
4782 * More importantly, if a facet's wildcards change, it will be relatively
4783 * difficult to figure out if its subfacets still belong to it, and if not
4784 * which facet they may belong to. Again, to avoid the complexity, we
4785 * simply give up instead. */
4786 if (facet->xout.slow != xout.slow
4787 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4788 facet_remove(facet);
4789 xlate_out_uninit(&xout);
4793 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4794 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4795 if (subfacet->path == SF_FAST_PATH) {
4796 struct dpif_flow_stats stats;
4798 subfacet_install(subfacet, &xout.odp_actions, &stats);
4799 subfacet_update_stats(subfacet, &stats);
4803 facet_flush_stats(facet);
4805 ofpbuf_clear(&facet->xout.odp_actions);
4806 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4807 xout.odp_actions.size);
4810 /* Update 'facet' now that we've taken care of all the old state. */
4811 facet->xout.tags = xout.tags;
4812 facet->xout.slow = xout.slow;
4813 facet->xout.has_learn = xout.has_learn;
4814 facet->xout.has_normal = xout.has_normal;
4815 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4816 facet->xout.nf_output_iface = xout.nf_output_iface;
4817 facet->xout.mirrors = xout.mirrors;
4818 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4819 facet->used = MAX(facet->used, new_rule->up.created);
4820 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4822 xlate_out_uninit(&xout);
4827 facet_reset_counters(struct facet *facet)
4829 facet->packet_count = 0;
4830 facet->byte_count = 0;
4831 facet->prev_packet_count = 0;
4832 facet->prev_byte_count = 0;
4833 facet->accounted_bytes = 0;
4837 facet_push_stats(struct facet *facet, bool may_learn)
4839 struct dpif_flow_stats stats;
4841 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4842 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4843 ovs_assert(facet->used >= facet->prev_used);
4845 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4846 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4847 stats.used = facet->used;
4848 stats.tcp_flags = facet->tcp_flags;
4850 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4851 struct ofproto_dpif *ofproto = facet->ofproto;
4852 struct ofport_dpif *in_port;
4853 struct rule_dpif *rule;
4854 struct xlate_in xin;
4856 facet->prev_packet_count = facet->packet_count;
4857 facet->prev_byte_count = facet->byte_count;
4858 facet->prev_used = facet->used;
4860 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4861 if (in_port && in_port->is_tunnel) {
4862 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4865 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4866 rule_credit_stats(rule, &stats);
4867 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4869 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4870 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4871 stats.n_packets, stats.n_bytes);
4873 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4875 xin.resubmit_stats = &stats;
4876 xin.may_learn = may_learn;
4877 xlate_actions_for_side_effects(&xin);
4882 push_all_stats__(bool run_fast)
4884 static long long int rl = LLONG_MIN;
4885 struct ofproto_dpif *ofproto;
4887 if (time_msec() < rl) {
4891 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4892 struct cls_cursor cursor;
4893 struct facet *facet;
4895 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4896 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4897 facet_push_stats(facet, false);
4904 rl = time_msec() + 100;
4908 push_all_stats(void)
4910 push_all_stats__(true);
4914 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4916 rule->packet_count += stats->n_packets;
4917 rule->byte_count += stats->n_bytes;
4918 ofproto_rule_update_used(&rule->up, stats->used);
4923 static struct subfacet *
4924 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4925 size_t key_len, uint32_t key_hash)
4927 struct subfacet *subfacet;
4929 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4930 &backer->subfacets) {
4931 if (subfacet->key_len == key_len
4932 && !memcmp(key, subfacet->key, key_len)) {
4940 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4941 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4942 * existing subfacet if there is one, otherwise creates and returns a
4944 static struct subfacet *
4945 subfacet_create(struct facet *facet, struct flow_miss *miss,
4948 struct dpif_backer *backer = miss->ofproto->backer;
4949 enum odp_key_fitness key_fitness = miss->key_fitness;
4950 const struct nlattr *key = miss->key;
4951 size_t key_len = miss->key_len;
4953 struct subfacet *subfacet;
4955 key_hash = odp_flow_key_hash(key, key_len);
4957 if (list_is_empty(&facet->subfacets)) {
4958 subfacet = &facet->one_subfacet;
4960 subfacet = subfacet_find(backer, key, key_len, key_hash);
4962 if (subfacet->facet == facet) {
4966 /* This shouldn't happen. */
4967 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4968 subfacet_destroy(subfacet);
4971 subfacet = xmalloc(sizeof *subfacet);
4974 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4975 list_push_back(&facet->subfacets, &subfacet->list_node);
4976 subfacet->facet = facet;
4977 subfacet->key_fitness = key_fitness;
4978 subfacet->key = xmemdup(key, key_len);
4979 subfacet->key_len = key_len;
4980 subfacet->used = now;
4981 subfacet->created = now;
4982 subfacet->dp_packet_count = 0;
4983 subfacet->dp_byte_count = 0;
4984 subfacet->path = SF_NOT_INSTALLED;
4985 subfacet->backer = backer;
4987 backer->subfacet_add_count++;
4991 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4992 * its facet within 'ofproto', and frees it. */
4994 subfacet_destroy__(struct subfacet *subfacet)
4996 struct facet *facet = subfacet->facet;
4997 struct ofproto_dpif *ofproto = facet->ofproto;
4999 /* Update ofproto stats before uninstall the subfacet. */
5000 ofproto->backer->subfacet_del_count++;
5002 subfacet_uninstall(subfacet);
5003 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
5004 list_remove(&subfacet->list_node);
5005 free(subfacet->key);
5006 if (subfacet != &facet->one_subfacet) {
5011 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5012 * last remaining subfacet in its facet destroys the facet too. */
5014 subfacet_destroy(struct subfacet *subfacet)
5016 struct facet *facet = subfacet->facet;
5018 if (list_is_singleton(&facet->subfacets)) {
5019 /* facet_remove() needs at least one subfacet (it will remove it). */
5020 facet_remove(facet);
5022 subfacet_destroy__(subfacet);
5027 subfacet_destroy_batch(struct dpif_backer *backer,
5028 struct subfacet **subfacets, int n)
5030 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5031 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5032 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5035 for (i = 0; i < n; i++) {
5036 ops[i].type = DPIF_OP_FLOW_DEL;
5037 ops[i].u.flow_del.key = subfacets[i]->key;
5038 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5039 ops[i].u.flow_del.stats = &stats[i];
5043 dpif_operate(backer->dpif, opsp, n);
5044 for (i = 0; i < n; i++) {
5045 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5046 subfacets[i]->path = SF_NOT_INSTALLED;
5047 subfacet_destroy(subfacets[i]);
5052 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5053 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5054 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5055 * since 'subfacet' was last updated.
5057 * Returns 0 if successful, otherwise a positive errno value. */
5059 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5060 struct dpif_flow_stats *stats)
5062 struct facet *facet = subfacet->facet;
5063 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5064 const struct nlattr *actions = odp_actions->data;
5065 size_t actions_len = odp_actions->size;
5066 struct odputil_keybuf maskbuf;
5069 uint64_t slow_path_stub[128 / 8];
5070 enum dpif_flow_put_flags flags;
5073 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5075 flags |= DPIF_FP_ZERO_STATS;
5078 if (path == SF_SLOW_PATH) {
5079 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5080 slow_path_stub, sizeof slow_path_stub,
5081 &actions, &actions_len);
5084 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5085 if (enable_megaflows) {
5086 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5087 &facet->flow, UINT32_MAX);
5090 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5091 subfacet->key_len, mask.data, mask.size,
5092 actions, actions_len, stats);
5095 subfacet_reset_dp_stats(subfacet, stats);
5099 COVERAGE_INC(subfacet_install_fail);
5101 subfacet->path = path;
5106 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5108 subfacet_uninstall(struct subfacet *subfacet)
5110 if (subfacet->path != SF_NOT_INSTALLED) {
5111 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5112 struct dpif_flow_stats stats;
5115 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5116 subfacet->key_len, &stats);
5117 subfacet_reset_dp_stats(subfacet, &stats);
5119 subfacet_update_stats(subfacet, &stats);
5121 subfacet->path = SF_NOT_INSTALLED;
5123 ovs_assert(subfacet->dp_packet_count == 0);
5124 ovs_assert(subfacet->dp_byte_count == 0);
5128 /* Resets 'subfacet''s datapath statistics counters. This should be called
5129 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5130 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5131 * was reset in the datapath. 'stats' will be modified to include only
5132 * statistics new since 'subfacet' was last updated. */
5134 subfacet_reset_dp_stats(struct subfacet *subfacet,
5135 struct dpif_flow_stats *stats)
5138 && subfacet->dp_packet_count <= stats->n_packets
5139 && subfacet->dp_byte_count <= stats->n_bytes) {
5140 stats->n_packets -= subfacet->dp_packet_count;
5141 stats->n_bytes -= subfacet->dp_byte_count;
5144 subfacet->dp_packet_count = 0;
5145 subfacet->dp_byte_count = 0;
5148 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5150 * Because of the meaning of a subfacet's counters, it only makes sense to do
5151 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5152 * represents a packet that was sent by hand or if it represents statistics
5153 * that have been cleared out of the datapath. */
5155 subfacet_update_stats(struct subfacet *subfacet,
5156 const struct dpif_flow_stats *stats)
5158 if (stats->n_packets || stats->used > subfacet->used) {
5159 struct facet *facet = subfacet->facet;
5161 subfacet->used = MAX(subfacet->used, stats->used);
5162 facet->used = MAX(facet->used, stats->used);
5163 facet->packet_count += stats->n_packets;
5164 facet->byte_count += stats->n_bytes;
5165 facet->tcp_flags |= stats->tcp_flags;
5171 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5172 * the fields that were relevant as part of the lookup. */
5173 static struct rule_dpif *
5174 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5175 struct flow_wildcards *wc)
5177 struct rule_dpif *rule;
5179 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5184 return rule_dpif_miss_rule(ofproto, flow);
5188 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5189 const struct flow *flow, struct flow_wildcards *wc,
5192 struct cls_rule *cls_rule;
5193 struct classifier *cls;
5196 if (table_id >= N_TABLES) {
5201 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5202 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5205 cls = &ofproto->up.tables[table_id].cls;
5206 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5207 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5208 /* We must pretend that transport ports are unavailable. */
5209 struct flow ofpc_normal_flow = *flow;
5210 ofpc_normal_flow.tp_src = htons(0);
5211 ofpc_normal_flow.tp_dst = htons(0);
5212 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5213 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5214 cls_rule = &ofproto->drop_frags_rule->up.cr;
5216 flow_wildcards_init_exact(wc);
5219 cls_rule = classifier_lookup(cls, flow, wc);
5221 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5225 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5227 struct ofport_dpif *port;
5229 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5231 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5232 flow->in_port.ofp_port);
5233 return ofproto->miss_rule;
5236 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5237 return ofproto->no_packet_in_rule;
5239 return ofproto->miss_rule;
5243 complete_operation(struct rule_dpif *rule)
5245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5247 rule_invalidate(rule);
5249 struct dpif_completion *c = xmalloc(sizeof *c);
5250 c->op = rule->up.pending;
5251 list_push_back(&ofproto->completions, &c->list_node);
5253 ofoperation_complete(rule->up.pending, 0);
5257 static struct rule *
5260 struct rule_dpif *rule = xmalloc(sizeof *rule);
5265 rule_dealloc(struct rule *rule_)
5267 struct rule_dpif *rule = rule_dpif_cast(rule_);
5272 rule_construct(struct rule *rule_)
5274 struct rule_dpif *rule = rule_dpif_cast(rule_);
5275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5276 struct rule_dpif *victim;
5279 rule->packet_count = 0;
5280 rule->byte_count = 0;
5282 table_id = rule->up.table_id;
5283 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5285 rule->tag = victim->tag;
5286 } else if (table_id == 0) {
5291 miniflow_expand(&rule->up.cr.match.flow, &flow);
5292 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5293 ofproto->tables[table_id].basis);
5296 complete_operation(rule);
5301 rule_destruct(struct rule *rule)
5303 complete_operation(rule_dpif_cast(rule));
5307 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5309 struct rule_dpif *rule = rule_dpif_cast(rule_);
5311 /* push_all_stats() can handle flow misses which, when using the learn
5312 * action, can cause rules to be added and deleted. This can corrupt our
5313 * caller's datastructures which assume that rule_get_stats() doesn't have
5314 * an impact on the flow table. To be safe, we disable miss handling. */
5315 push_all_stats__(false);
5317 /* Start from historical data for 'rule' itself that are no longer tracked
5318 * in facets. This counts, for example, facets that have expired. */
5319 *packets = rule->packet_count;
5320 *bytes = rule->byte_count;
5324 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5325 struct ofpbuf *packet)
5327 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5328 struct dpif_flow_stats stats;
5329 struct xlate_out xout;
5330 struct xlate_in xin;
5332 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5333 rule_credit_stats(rule, &stats);
5335 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5336 xin.resubmit_stats = &stats;
5337 xlate_actions(&xin, &xout);
5339 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5340 xout.odp_actions.size, packet);
5342 xlate_out_uninit(&xout);
5346 rule_execute(struct rule *rule, const struct flow *flow,
5347 struct ofpbuf *packet)
5349 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5350 ofpbuf_delete(packet);
5355 rule_modify_actions(struct rule *rule_)
5357 struct rule_dpif *rule = rule_dpif_cast(rule_);
5359 complete_operation(rule);
5362 /* Sends 'packet' out 'ofport'.
5363 * May modify 'packet'.
5364 * Returns 0 if successful, otherwise a positive errno value. */
5366 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5369 uint64_t odp_actions_stub[1024 / 8];
5370 struct ofpbuf key, odp_actions;
5371 struct dpif_flow_stats stats;
5372 struct odputil_keybuf keybuf;
5373 struct ofpact_output output;
5374 struct xlate_out xout;
5375 struct xlate_in xin;
5377 union flow_in_port in_port_;
5380 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5381 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5383 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5384 in_port_.ofp_port = OFPP_NONE;
5385 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5386 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5388 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5390 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5391 output.port = ofport->up.ofp_port;
5394 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5395 xin.ofpacts_len = sizeof output;
5396 xin.ofpacts = &output.ofpact;
5397 xin.resubmit_stats = &stats;
5398 xlate_actions(&xin, &xout);
5400 error = dpif_execute(ofproto->backer->dpif,
5402 xout.odp_actions.data, xout.odp_actions.size,
5404 xlate_out_uninit(&xout);
5407 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5408 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5409 ovs_strerror(error));
5412 ofproto->stats.tx_packets++;
5413 ofproto->stats.tx_bytes += packet->size;
5417 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5418 * The action will state 'slow' as the reason that the action is in the slow
5419 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5420 * dump-flows" output to see why a flow is in the slow path.)
5422 * The 'stub_size' bytes in 'stub' will be used to store the action.
5423 * 'stub_size' must be large enough for the action.
5425 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5428 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5429 enum slow_path_reason slow,
5430 uint64_t *stub, size_t stub_size,
5431 const struct nlattr **actionsp, size_t *actions_lenp)
5433 union user_action_cookie cookie;
5436 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5437 cookie.slow_path.unused = 0;
5438 cookie.slow_path.reason = slow;
5440 ofpbuf_use_stack(&buf, stub, stub_size);
5441 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5442 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5444 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5446 put_userspace_action(ofproto, &buf, flow, &cookie,
5447 sizeof cookie.slow_path);
5449 *actionsp = buf.data;
5450 *actions_lenp = buf.size;
5454 put_userspace_action(const struct ofproto_dpif *ofproto,
5455 struct ofpbuf *odp_actions,
5456 const struct flow *flow,
5457 const union user_action_cookie *cookie,
5458 const size_t cookie_size)
5462 pid = dpif_port_get_pid(ofproto->backer->dpif,
5463 ofp_port_to_odp_port(ofproto,
5464 flow->in_port.ofp_port));
5466 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5470 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5471 uint8_t table_id, struct rule_dpif *rule)
5473 if (table_id > 0 && table_id < N_TABLES) {
5474 struct table_dpif *table = &ofproto->tables[table_id];
5475 if (table->other_table) {
5476 return (rule && rule->tag
5478 : rule_calculate_tag(flow, &table->other_table->mask,
5486 /* Optimized flow revalidation.
5488 * It's a difficult problem, in general, to tell which facets need to have
5489 * their actions recalculated whenever the OpenFlow flow table changes. We
5490 * don't try to solve that general problem: for most kinds of OpenFlow flow
5491 * table changes, we recalculate the actions for every facet. This is
5492 * relatively expensive, but it's good enough if the OpenFlow flow table
5493 * doesn't change very often.
5495 * However, we can expect one particular kind of OpenFlow flow table change to
5496 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5497 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5498 * table, we add a special case that applies to flow tables in which every rule
5499 * has the same form (that is, the same wildcards), except that the table is
5500 * also allowed to have a single "catch-all" flow that matches all packets. We
5501 * optimize this case by tagging all of the facets that resubmit into the table
5502 * and invalidating the same tag whenever a flow changes in that table. The
5503 * end result is that we revalidate just the facets that need it (and sometimes
5504 * a few more, but not all of the facets or even all of the facets that
5505 * resubmit to the table modified by MAC learning). */
5507 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5508 * into an OpenFlow table with the given 'basis'. */
5510 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5513 if (minimask_is_catchall(mask)) {
5516 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5517 return tag_create_deterministic(hash);
5521 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5522 * taggability of that table.
5524 * This function must be called after *each* change to a flow table. If you
5525 * skip calling it on some changes then the pointer comparisons at the end can
5526 * be invalid if you get unlucky. For example, if a flow removal causes a
5527 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5528 * different wildcards to be created with the same address, then this function
5529 * will incorrectly skip revalidation. */
5531 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5533 struct table_dpif *table = &ofproto->tables[table_id];
5534 const struct oftable *oftable = &ofproto->up.tables[table_id];
5535 struct cls_table *catchall, *other;
5536 struct cls_table *t;
5538 catchall = other = NULL;
5540 switch (hmap_count(&oftable->cls.tables)) {
5542 /* We could tag this OpenFlow table but it would make the logic a
5543 * little harder and it's a corner case that doesn't seem worth it
5549 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5550 if (cls_table_is_catchall(t)) {
5552 } else if (!other) {
5555 /* Indicate that we can't tag this by setting both tables to
5556 * NULL. (We know that 'catchall' is already NULL.) */
5563 /* Can't tag this table. */
5567 if (table->catchall_table != catchall || table->other_table != other) {
5568 table->catchall_table = catchall;
5569 table->other_table = other;
5570 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5574 /* Given 'rule' that has changed in some way (either it is a rule being
5575 * inserted, a rule being deleted, or a rule whose actions are being
5576 * modified), marks facets for revalidation to ensure that packets will be
5577 * forwarded correctly according to the new state of the flow table.
5579 * This function must be called after *each* change to a flow table. See
5580 * the comment on table_update_taggable() for more information. */
5582 rule_invalidate(const struct rule_dpif *rule)
5584 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5586 table_update_taggable(ofproto, rule->up.table_id);
5588 if (!ofproto->backer->need_revalidate) {
5589 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5591 if (table->other_table && rule->tag) {
5592 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5594 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5600 set_frag_handling(struct ofproto *ofproto_,
5601 enum ofp_config_flags frag_handling)
5603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5604 if (frag_handling != OFPC_FRAG_REASM) {
5605 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5613 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5614 const struct flow *flow,
5615 const struct ofpact *ofpacts, size_t ofpacts_len)
5617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5618 struct odputil_keybuf keybuf;
5619 struct dpif_flow_stats stats;
5620 struct xlate_out xout;
5621 struct xlate_in xin;
5625 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5626 odp_flow_key_from_flow(&key, flow,
5627 ofp_port_to_odp_port(ofproto,
5628 flow->in_port.ofp_port));
5630 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5632 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5633 xin.resubmit_stats = &stats;
5634 xin.ofpacts_len = ofpacts_len;
5635 xin.ofpacts = ofpacts;
5637 xlate_actions(&xin, &xout);
5638 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5639 xout.odp_actions.data, xout.odp_actions.size, packet);
5640 xlate_out_uninit(&xout);
5648 set_netflow(struct ofproto *ofproto_,
5649 const struct netflow_options *netflow_options)
5651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5653 if (netflow_options) {
5654 if (!ofproto->netflow) {
5655 ofproto->netflow = netflow_create();
5656 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5658 return netflow_set_options(ofproto->netflow, netflow_options);
5659 } else if (ofproto->netflow) {
5660 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5661 netflow_destroy(ofproto->netflow);
5662 ofproto->netflow = NULL;
5669 get_netflow_ids(const struct ofproto *ofproto_,
5670 uint8_t *engine_type, uint8_t *engine_id)
5672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5674 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5678 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5680 if (!facet_is_controller_flow(facet) &&
5681 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5682 struct subfacet *subfacet;
5683 struct ofexpired expired;
5685 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5686 if (subfacet->path == SF_FAST_PATH) {
5687 struct dpif_flow_stats stats;
5689 subfacet_install(subfacet, &facet->xout.odp_actions,
5691 subfacet_update_stats(subfacet, &stats);
5695 expired.flow = facet->flow;
5696 expired.packet_count = facet->packet_count;
5697 expired.byte_count = facet->byte_count;
5698 expired.used = facet->used;
5699 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5704 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5706 struct cls_cursor cursor;
5707 struct facet *facet;
5709 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5710 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5711 send_active_timeout(ofproto, facet);
5715 static struct ofproto_dpif *
5716 ofproto_dpif_lookup(const char *name)
5718 struct ofproto_dpif *ofproto;
5720 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5721 hash_string(name, 0), &all_ofproto_dpifs) {
5722 if (!strcmp(ofproto->up.name, name)) {
5730 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5731 const char *argv[], void *aux OVS_UNUSED)
5733 struct ofproto_dpif *ofproto;
5736 ofproto = ofproto_dpif_lookup(argv[1]);
5738 unixctl_command_reply_error(conn, "no such bridge");
5741 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5743 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5744 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5748 unixctl_command_reply(conn, "table successfully flushed");
5751 static struct ofport_dpif *
5752 ofbundle_get_a_port(const struct ofbundle *bundle)
5754 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5759 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5760 const char *argv[], void *aux OVS_UNUSED)
5762 struct ds ds = DS_EMPTY_INITIALIZER;
5763 const struct ofproto_dpif *ofproto;
5764 const struct mac_entry *e;
5766 ofproto = ofproto_dpif_lookup(argv[1]);
5768 unixctl_command_reply_error(conn, "no such bridge");
5772 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5773 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5774 struct ofbundle *bundle = e->port.p;
5775 char name[OFP_MAX_PORT_NAME_LEN];
5777 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5779 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5780 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5781 mac_entry_age(ofproto->ml, e));
5783 unixctl_command_reply(conn, ds_cstr(&ds));
5788 struct xlate_out xout;
5789 struct xlate_in xin;
5795 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5797 ds_put_char_multiple(result, '\t', level);
5799 ds_put_cstr(result, "No match\n");
5803 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5804 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5805 cls_rule_format(&rule->up.cr, result);
5806 ds_put_char(result, '\n');
5808 ds_put_char_multiple(result, '\t', level);
5809 ds_put_cstr(result, "OpenFlow ");
5810 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5811 ds_put_char(result, '\n');
5815 trace_format_flow(struct ds *result, int level, const char *title,
5816 struct trace_ctx *trace)
5818 ds_put_char_multiple(result, '\t', level);
5819 ds_put_format(result, "%s: ", title);
5820 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5821 ds_put_cstr(result, "unchanged");
5823 flow_format(result, &trace->xin.flow);
5824 trace->flow = trace->xin.flow;
5826 ds_put_char(result, '\n');
5830 trace_format_regs(struct ds *result, int level, const char *title,
5831 struct trace_ctx *trace)
5835 ds_put_char_multiple(result, '\t', level);
5836 ds_put_format(result, "%s:", title);
5837 for (i = 0; i < FLOW_N_REGS; i++) {
5838 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5840 ds_put_char(result, '\n');
5844 trace_format_odp(struct ds *result, int level, const char *title,
5845 struct trace_ctx *trace)
5847 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5849 ds_put_char_multiple(result, '\t', level);
5850 ds_put_format(result, "%s: ", title);
5851 format_odp_actions(result, odp_actions->data, odp_actions->size);
5852 ds_put_char(result, '\n');
5856 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5858 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5859 struct ds *result = trace->result;
5861 ds_put_char(result, '\n');
5862 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5863 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5864 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5865 trace_format_rule(result, recurse + 1, rule);
5869 trace_report(struct xlate_in *xin, const char *s, int recurse)
5871 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5872 struct ds *result = trace->result;
5874 ds_put_char_multiple(result, '\t', recurse);
5875 ds_put_cstr(result, s);
5876 ds_put_char(result, '\n');
5880 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5881 void *aux OVS_UNUSED)
5883 const struct dpif_backer *backer;
5884 struct ofproto_dpif *ofproto;
5885 struct ofpbuf odp_key, odp_mask;
5886 struct ofpbuf *packet;
5894 ofpbuf_init(&odp_key, 0);
5895 ofpbuf_init(&odp_mask, 0);
5897 /* Handle "-generate" or a hex string as the last argument. */
5898 if (!strcmp(argv[argc - 1], "-generate")) {
5899 packet = ofpbuf_new(0);
5902 const char *error = eth_from_hex(argv[argc - 1], &packet);
5905 } else if (argc == 4) {
5906 /* The 3-argument form must end in "-generate' or a hex string. */
5907 unixctl_command_reply_error(conn, error);
5912 /* Parse the flow and determine whether a datapath or
5913 * bridge is specified. If function odp_flow_key_from_string()
5914 * returns 0, the flow is a odp_flow. If function
5915 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5916 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5917 /* If the odp_flow is the second argument,
5918 * the datapath name is the first argument. */
5920 const char *dp_type;
5921 if (!strncmp(argv[1], "ovs-", 4)) {
5922 dp_type = argv[1] + 4;
5926 backer = shash_find_data(&all_dpif_backers, dp_type);
5928 unixctl_command_reply_error(conn, "Cannot find datapath "
5933 /* No datapath name specified, so there should be only one
5935 struct shash_node *node;
5936 if (shash_count(&all_dpif_backers) != 1) {
5937 unixctl_command_reply_error(conn, "Must specify datapath "
5938 "name, there is more than one type of datapath");
5941 node = shash_first(&all_dpif_backers);
5942 backer = node->data;
5945 /* Extract the ofproto_dpif object from the ofproto_receive()
5947 if (ofproto_receive(backer, NULL, odp_key.data,
5948 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5949 unixctl_command_reply_error(conn, "Invalid datapath flow");
5952 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5953 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5955 unixctl_command_reply_error(conn, "Must specify bridge name");
5959 ofproto = ofproto_dpif_lookup(argv[1]);
5961 unixctl_command_reply_error(conn, "Unknown bridge name");
5965 unixctl_command_reply_error(conn, "Bad flow syntax");
5969 /* Generate a packet, if requested. */
5971 if (!packet->size) {
5972 flow_compose(packet, &flow);
5974 union flow_in_port in_port_;
5976 in_port_ = flow.in_port;
5977 ds_put_cstr(&result, "Packet: ");
5978 s = ofp_packet_to_string(packet->data, packet->size);
5979 ds_put_cstr(&result, s);
5982 /* Use the metadata from the flow and the packet argument
5983 * to reconstruct the flow. */
5984 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5989 ofproto_trace(ofproto, &flow, packet, &result);
5990 unixctl_command_reply(conn, ds_cstr(&result));
5993 ds_destroy(&result);
5994 ofpbuf_delete(packet);
5995 ofpbuf_uninit(&odp_key);
5996 ofpbuf_uninit(&odp_mask);
6000 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6001 const struct ofpbuf *packet, struct ds *ds)
6003 struct rule_dpif *rule;
6005 ds_put_cstr(ds, "Flow: ");
6006 flow_format(ds, flow);
6007 ds_put_char(ds, '\n');
6009 rule = rule_dpif_lookup(ofproto, flow, NULL);
6011 trace_format_rule(ds, 0, rule);
6012 if (rule == ofproto->miss_rule) {
6013 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6014 } else if (rule == ofproto->no_packet_in_rule) {
6015 ds_put_cstr(ds, "\nNo match, packets dropped because "
6016 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6017 } else if (rule == ofproto->drop_frags_rule) {
6018 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
6019 "and the fragment handling mode is \"drop\".\n");
6023 uint64_t odp_actions_stub[1024 / 8];
6024 struct ofpbuf odp_actions;
6025 struct trace_ctx trace;
6029 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6032 ofpbuf_use_stub(&odp_actions,
6033 odp_actions_stub, sizeof odp_actions_stub);
6034 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
6035 trace.xin.resubmit_hook = trace_resubmit;
6036 trace.xin.report_hook = trace_report;
6038 xlate_actions(&trace.xin, &trace.xout);
6040 ds_put_char(ds, '\n');
6041 trace_format_flow(ds, 0, "Final flow", &trace);
6043 match_init(&match, flow, &trace.xout.wc);
6044 ds_put_cstr(ds, "Relevant fields: ");
6045 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
6046 ds_put_char(ds, '\n');
6048 ds_put_cstr(ds, "Datapath actions: ");
6049 format_odp_actions(ds, trace.xout.odp_actions.data,
6050 trace.xout.odp_actions.size);
6052 if (trace.xout.slow) {
6053 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6054 "slow path because it:");
6055 switch (trace.xout.slow) {
6057 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6060 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6063 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6066 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
6068 case SLOW_CONTROLLER:
6069 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6070 "to the OpenFlow controller.");
6077 xlate_out_uninit(&trace.xout);
6082 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6083 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6086 unixctl_command_reply(conn, NULL);
6090 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6091 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6094 unixctl_command_reply(conn, NULL);
6097 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6098 * 'reply' describing the results. */
6100 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6102 struct cls_cursor cursor;
6103 struct facet *facet;
6107 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6108 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6109 if (!facet_check_consistency(facet)) {
6114 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6118 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6119 ofproto->up.name, errors);
6121 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6126 ofproto_dpif_self_check(struct unixctl_conn *conn,
6127 int argc, const char *argv[], void *aux OVS_UNUSED)
6129 struct ds reply = DS_EMPTY_INITIALIZER;
6130 struct ofproto_dpif *ofproto;
6133 ofproto = ofproto_dpif_lookup(argv[1]);
6135 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6136 "ofproto/list for help)");
6139 ofproto_dpif_self_check__(ofproto, &reply);
6141 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6142 ofproto_dpif_self_check__(ofproto, &reply);
6146 unixctl_command_reply(conn, ds_cstr(&reply));
6150 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6151 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6152 * to destroy 'ofproto_shash' and free the returned value. */
6153 static const struct shash_node **
6154 get_ofprotos(struct shash *ofproto_shash)
6156 const struct ofproto_dpif *ofproto;
6158 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6159 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6160 shash_add_nocopy(ofproto_shash, name, ofproto);
6163 return shash_sort(ofproto_shash);
6167 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6168 const char *argv[] OVS_UNUSED,
6169 void *aux OVS_UNUSED)
6171 struct ds ds = DS_EMPTY_INITIALIZER;
6172 struct shash ofproto_shash;
6173 const struct shash_node **sorted_ofprotos;
6176 shash_init(&ofproto_shash);
6177 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6178 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6179 const struct shash_node *node = sorted_ofprotos[i];
6180 ds_put_format(&ds, "%s\n", node->name);
6183 shash_destroy(&ofproto_shash);
6184 free(sorted_ofprotos);
6186 unixctl_command_reply(conn, ds_cstr(&ds));
6191 show_dp_rates(struct ds *ds, const char *heading,
6192 const struct avg_subfacet_rates *rates)
6194 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6195 heading, rates->add_rate, rates->del_rate);
6199 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6201 const struct shash_node **ofprotos;
6202 struct ofproto_dpif *ofproto;
6203 struct shash ofproto_shash;
6204 uint64_t n_hit, n_missed;
6205 long long int minutes;
6208 n_hit = n_missed = 0;
6209 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6210 if (ofproto->backer == backer) {
6211 n_missed += ofproto->n_missed;
6212 n_hit += ofproto->n_hit;
6216 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6217 dpif_name(backer->dpif), n_hit, n_missed);
6218 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6219 " life span: %lldms\n", hmap_count(&backer->subfacets),
6220 backer->avg_n_subfacet, backer->max_n_subfacet,
6221 backer->avg_subfacet_life);
6223 minutes = (time_msec() - backer->created) / (1000 * 60);
6224 if (minutes >= 60) {
6225 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6227 if (minutes >= 60 * 24) {
6228 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6230 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6232 shash_init(&ofproto_shash);
6233 ofprotos = get_ofprotos(&ofproto_shash);
6234 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6235 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6236 const struct shash_node **ports;
6239 if (ofproto->backer != backer) {
6243 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6244 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6246 ports = shash_sort(&ofproto->up.port_by_name);
6247 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6248 const struct shash_node *node = ports[j];
6249 struct ofport *ofport = node->data;
6251 odp_port_t odp_port;
6253 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6256 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6257 if (odp_port != ODPP_NONE) {
6258 ds_put_format(ds, "%"PRIu32":", odp_port);
6260 ds_put_cstr(ds, "none:");
6263 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6266 if (!netdev_get_config(ofport->netdev, &config)) {
6267 const struct smap_node **nodes;
6270 nodes = smap_sort(&config);
6271 for (i = 0; i < smap_count(&config); i++) {
6272 const struct smap_node *node = nodes[i];
6273 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6274 node->key, node->value);
6278 smap_destroy(&config);
6280 ds_put_char(ds, ')');
6281 ds_put_char(ds, '\n');
6285 shash_destroy(&ofproto_shash);
6290 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6291 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6293 struct ds ds = DS_EMPTY_INITIALIZER;
6294 const struct shash_node **backers;
6297 backers = shash_sort(&all_dpif_backers);
6298 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6299 dpif_show_backer(backers[i]->data, &ds);
6303 unixctl_command_reply(conn, ds_cstr(&ds));
6307 /* Dump the megaflow (facet) cache. This is useful to check the
6308 * correctness of flow wildcarding, since the same mechanism is used for
6309 * both xlate caching and kernel wildcarding.
6311 * It's important to note that in the output the flow description uses
6312 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6314 * This command is only needed for advanced debugging, so it's not
6315 * documented in the man page. */
6317 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6318 int argc OVS_UNUSED, const char *argv[],
6319 void *aux OVS_UNUSED)
6321 struct ds ds = DS_EMPTY_INITIALIZER;
6322 const struct ofproto_dpif *ofproto;
6323 long long int now = time_msec();
6324 struct cls_cursor cursor;
6325 struct facet *facet;
6327 ofproto = ofproto_dpif_lookup(argv[1]);
6329 unixctl_command_reply_error(conn, "no such bridge");
6333 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6334 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6335 cls_rule_format(&facet->cr, &ds);
6336 ds_put_cstr(&ds, ", ");
6337 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6338 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6339 ds_put_cstr(&ds, "Datapath actions: ");
6340 if (facet->xout.slow) {
6341 uint64_t slow_path_stub[128 / 8];
6342 const struct nlattr *actions;
6345 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6346 slow_path_stub, sizeof slow_path_stub,
6347 &actions, &actions_len);
6348 format_odp_actions(&ds, actions, actions_len);
6350 format_odp_actions(&ds, facet->xout.odp_actions.data,
6351 facet->xout.odp_actions.size);
6353 ds_put_cstr(&ds, "\n");
6356 ds_chomp(&ds, '\n');
6357 unixctl_command_reply(conn, ds_cstr(&ds));
6361 /* Disable using the megaflows.
6363 * This command is only needed for advanced debugging, so it's not
6364 * documented in the man page. */
6366 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6367 int argc OVS_UNUSED,
6368 const char *argv[] OVS_UNUSED,
6369 void *aux OVS_UNUSED)
6371 struct ofproto_dpif *ofproto;
6373 enable_megaflows = false;
6375 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6376 flush(&ofproto->up);
6379 unixctl_command_reply(conn, "megaflows disabled");
6382 /* Re-enable using megaflows.
6384 * This command is only needed for advanced debugging, so it's not
6385 * documented in the man page. */
6387 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6388 int argc OVS_UNUSED,
6389 const char *argv[] OVS_UNUSED,
6390 void *aux OVS_UNUSED)
6392 struct ofproto_dpif *ofproto;
6394 enable_megaflows = true;
6396 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6397 flush(&ofproto->up);
6400 unixctl_command_reply(conn, "megaflows enabled");
6404 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6405 int argc OVS_UNUSED, const char *argv[],
6406 void *aux OVS_UNUSED)
6408 struct ds ds = DS_EMPTY_INITIALIZER;
6409 const struct ofproto_dpif *ofproto;
6410 struct subfacet *subfacet;
6412 ofproto = ofproto_dpif_lookup(argv[1]);
6414 unixctl_command_reply_error(conn, "no such bridge");
6418 update_stats(ofproto->backer);
6420 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6421 struct facet *facet = subfacet->facet;
6422 struct odputil_keybuf maskbuf;
6425 if (facet->ofproto != ofproto) {
6429 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6430 if (enable_megaflows) {
6431 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6432 &facet->flow, UINT32_MAX);
6435 odp_flow_format(subfacet->key, subfacet->key_len,
6436 mask.data, mask.size, &ds);
6438 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6439 subfacet->dp_packet_count, subfacet->dp_byte_count);
6440 if (subfacet->used) {
6441 ds_put_format(&ds, "%.3fs",
6442 (time_msec() - subfacet->used) / 1000.0);
6444 ds_put_format(&ds, "never");
6446 if (subfacet->facet->tcp_flags) {
6447 ds_put_cstr(&ds, ", flags:");
6448 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6451 ds_put_cstr(&ds, ", actions:");
6452 if (facet->xout.slow) {
6453 uint64_t slow_path_stub[128 / 8];
6454 const struct nlattr *actions;
6457 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6458 slow_path_stub, sizeof slow_path_stub,
6459 &actions, &actions_len);
6460 format_odp_actions(&ds, actions, actions_len);
6462 format_odp_actions(&ds, facet->xout.odp_actions.data,
6463 facet->xout.odp_actions.size);
6465 ds_put_char(&ds, '\n');
6468 unixctl_command_reply(conn, ds_cstr(&ds));
6473 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6474 int argc OVS_UNUSED, const char *argv[],
6475 void *aux OVS_UNUSED)
6477 struct ds ds = DS_EMPTY_INITIALIZER;
6478 struct ofproto_dpif *ofproto;
6480 ofproto = ofproto_dpif_lookup(argv[1]);
6482 unixctl_command_reply_error(conn, "no such bridge");
6486 flush(&ofproto->up);
6488 unixctl_command_reply(conn, ds_cstr(&ds));
6493 ofproto_dpif_unixctl_init(void)
6495 static bool registered;
6501 unixctl_command_register(
6503 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6504 1, 3, ofproto_unixctl_trace, NULL);
6505 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6506 ofproto_unixctl_fdb_flush, NULL);
6507 unixctl_command_register("fdb/show", "bridge", 1, 1,
6508 ofproto_unixctl_fdb_show, NULL);
6509 unixctl_command_register("ofproto/clog", "", 0, 0,
6510 ofproto_dpif_clog, NULL);
6511 unixctl_command_register("ofproto/unclog", "", 0, 0,
6512 ofproto_dpif_unclog, NULL);
6513 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6514 ofproto_dpif_self_check, NULL);
6515 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6516 ofproto_unixctl_dpif_dump_dps, NULL);
6517 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6519 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6520 ofproto_unixctl_dpif_dump_flows, NULL);
6521 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6522 ofproto_unixctl_dpif_del_flows, NULL);
6523 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6524 ofproto_unixctl_dpif_dump_megaflows, NULL);
6525 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6526 ofproto_unixctl_dpif_disable_megaflows, NULL);
6527 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6528 ofproto_unixctl_dpif_enable_megaflows, NULL);
6531 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6533 * This is deprecated. It is only for compatibility with broken device drivers
6534 * in old versions of Linux that do not properly support VLANs when VLAN
6535 * devices are not used. When broken device drivers are no longer in
6536 * widespread use, we will delete these interfaces. */
6539 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6542 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6544 if (realdev_ofp_port == ofport->realdev_ofp_port
6545 && vid == ofport->vlandev_vid) {
6549 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6551 if (ofport->realdev_ofp_port) {
6554 if (realdev_ofp_port && ofport->bundle) {
6555 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6556 * themselves be part of a bundle. */
6557 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6560 ofport->realdev_ofp_port = realdev_ofp_port;
6561 ofport->vlandev_vid = vid;
6563 if (realdev_ofp_port) {
6564 vsp_add(ofport, realdev_ofp_port, vid);
6571 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6573 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6577 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6579 return !hmap_is_empty(&ofproto->realdev_vid_map);
6582 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6583 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6584 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6585 * 'vlan_tci' 9, it would return the port number of eth0.9.
6587 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6588 * function just returns its 'realdev_ofp_port' argument. */
6590 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6591 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6593 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6594 int vid = vlan_tci_to_vid(vlan_tci);
6595 const struct vlan_splinter *vsp;
6597 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6598 hash_realdev_vid(realdev_ofp_port, vid),
6599 &ofproto->realdev_vid_map) {
6600 if (vsp->realdev_ofp_port == realdev_ofp_port
6601 && vsp->vid == vid) {
6602 return vsp->vlandev_ofp_port;
6606 return realdev_ofp_port;
6609 static struct vlan_splinter *
6610 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6612 struct vlan_splinter *vsp;
6614 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6615 hash_ofp_port(vlandev_ofp_port),
6616 &ofproto->vlandev_map) {
6617 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6625 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6626 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6627 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6628 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6629 * eth0 and store 9 in '*vid'.
6631 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6632 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6635 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6636 ofp_port_t vlandev_ofp_port, int *vid)
6638 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6639 const struct vlan_splinter *vsp;
6641 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6646 return vsp->realdev_ofp_port;
6652 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6653 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6654 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6655 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6656 * always the case unless VLAN splinters are enabled), returns false without
6657 * making any changes. */
6659 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6664 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6669 /* Cause the flow to be processed as if it came in on the real device with
6670 * the VLAN device's VLAN ID. */
6671 flow->in_port.ofp_port = realdev;
6672 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6677 vsp_remove(struct ofport_dpif *port)
6679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6680 struct vlan_splinter *vsp;
6682 vsp = vlandev_find(ofproto, port->up.ofp_port);
6684 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6685 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6688 port->realdev_ofp_port = 0;
6690 VLOG_ERR("missing vlan device record");
6695 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6699 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6700 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6701 == realdev_ofp_port)) {
6702 struct vlan_splinter *vsp;
6704 vsp = xmalloc(sizeof *vsp);
6705 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6706 hash_ofp_port(port->up.ofp_port));
6707 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6708 hash_realdev_vid(realdev_ofp_port, vid));
6709 vsp->realdev_ofp_port = realdev_ofp_port;
6710 vsp->vlandev_ofp_port = port->up.ofp_port;
6713 port->realdev_ofp_port = realdev_ofp_port;
6715 VLOG_ERR("duplicate vlan device record");
6720 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6722 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6723 return ofport ? ofport->odp_port : ODPP_NONE;
6726 static struct ofport_dpif *
6727 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6729 struct ofport_dpif *port;
6731 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6732 &backer->odp_to_ofport_map) {
6733 if (port->odp_port == odp_port) {
6742 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6744 struct ofport_dpif *port;
6746 port = odp_port_to_ofport(ofproto->backer, odp_port);
6747 if (port && &ofproto->up == port->up.ofproto) {
6748 return port->up.ofp_port;
6754 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6755 * most heavily weighted element. 'base' designates the rate of decay: after
6756 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6759 exp_mavg(double *avg, int base, double new)
6761 *avg = (*avg * (base - 1) + new) / base;
6765 update_moving_averages(struct dpif_backer *backer)
6767 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6768 long long int minutes = (time_msec() - backer->created) / min_ms;
6771 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6773 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6776 backer->lifetime.add_rate = 0.0;
6777 backer->lifetime.del_rate = 0.0;
6780 /* Update hourly averages on the minute boundaries. */
6781 if (time_msec() - backer->last_minute >= min_ms) {
6782 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6783 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6785 /* Update daily averages on the hour boundaries. */
6786 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6787 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6788 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6791 backer->total_subfacet_add_count += backer->subfacet_add_count;
6792 backer->total_subfacet_del_count += backer->subfacet_del_count;
6793 backer->subfacet_add_count = 0;
6794 backer->subfacet_del_count = 0;
6795 backer->last_minute += min_ms;
6799 const struct ofproto_class ofproto_dpif_class = {
6834 port_is_lacp_current,
6835 NULL, /* rule_choose_table */
6842 rule_modify_actions,
6856 get_stp_port_status,
6863 is_mirror_output_bundle,
6864 forward_bpdu_changed,
6865 set_mac_table_config,
6867 NULL, /* meter_get_features */
6868 NULL, /* meter_set */
6869 NULL, /* meter_get */
6870 NULL, /* meter_del */