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 struct ofproto_dpif *
642 lookup_ofproto_dpif_by_port_name(const char *name)
644 struct ofproto_dpif *ofproto;
646 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
647 if (sset_contains(&ofproto->ports, name)) {
656 type_run(const char *type)
658 static long long int push_timer = LLONG_MIN;
659 struct dpif_backer *backer;
663 backer = shash_find_data(&all_dpif_backers, type);
665 /* This is not necessarily a problem, since backers are only
666 * created on demand. */
670 dpif_run(backer->dpif);
672 /* The most natural place to push facet statistics is when they're pulled
673 * from the datapath. However, when there are many flows in the datapath,
674 * this expensive operation can occur so frequently, that it reduces our
675 * ability to quickly set up flows. To reduce the cost, we push statistics
677 if (time_msec() > push_timer) {
678 push_timer = time_msec() + 2000;
682 /* If vswitchd started with other_config:flow_restore_wait set as "true",
683 * and the configuration has now changed to "false", enable receiving
684 * packets from the datapath. */
685 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
686 backer->recv_set_enable = true;
688 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
690 VLOG_ERR("Failed to enable receiving packets in dpif.");
693 dpif_flow_flush(backer->dpif);
694 backer->need_revalidate = REV_RECONFIGURE;
697 if (backer->need_revalidate
698 || !tag_set_is_empty(&backer->revalidate_set)) {
699 struct tag_set revalidate_set = backer->revalidate_set;
700 bool need_revalidate = backer->need_revalidate;
701 struct ofproto_dpif *ofproto;
702 struct simap_node *node;
703 struct simap tmp_backers;
705 /* Handle tunnel garbage collection. */
706 simap_init(&tmp_backers);
707 simap_swap(&backer->tnl_backers, &tmp_backers);
709 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
710 struct ofport_dpif *iter;
712 if (backer != ofproto->backer) {
716 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
717 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
720 if (!iter->is_tunnel) {
724 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
725 namebuf, sizeof namebuf);
726 node = simap_find(&tmp_backers, dp_port);
728 simap_put(&backer->tnl_backers, dp_port, node->data);
729 simap_delete(&tmp_backers, node);
730 node = simap_find(&backer->tnl_backers, dp_port);
732 node = simap_find(&backer->tnl_backers, dp_port);
734 odp_port_t odp_port = ODPP_NONE;
736 if (!dpif_port_add(backer->dpif, iter->up.netdev,
738 simap_put(&backer->tnl_backers, dp_port,
739 odp_to_u32(odp_port));
740 node = simap_find(&backer->tnl_backers, dp_port);
745 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
746 if (tnl_port_reconfigure(iter, iter->up.netdev,
748 backer->need_revalidate = REV_RECONFIGURE;
753 SIMAP_FOR_EACH (node, &tmp_backers) {
754 dpif_port_del(backer->dpif, u32_to_odp(node->data));
756 simap_destroy(&tmp_backers);
758 switch (backer->need_revalidate) {
759 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
760 case REV_STP: COVERAGE_INC(rev_stp); break;
761 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
762 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
763 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
766 if (backer->need_revalidate) {
767 /* Clear the drop_keys in case we should now be accepting some
768 * formerly dropped flows. */
769 drop_key_clear(backer);
772 /* Clear the revalidation flags. */
773 tag_set_init(&backer->revalidate_set);
774 backer->need_revalidate = 0;
776 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
777 struct facet *facet, *next;
778 struct cls_cursor cursor;
780 if (ofproto->backer != backer) {
784 if (need_revalidate) {
785 struct ofport_dpif *ofport;
786 struct ofbundle *bundle;
788 xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
789 ofproto->mbridge, ofproto->sflow,
790 ofproto->ipfix, ofproto->up.frag_handling,
791 ofproto->up.forward_bpdu,
792 connmgr_has_in_band(ofproto->up.connmgr),
793 ofproto->netflow != NULL,
794 ofproto->stp != NULL);
796 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
797 xlate_bundle_set(ofproto, bundle, bundle->name,
798 bundle->vlan_mode, bundle->vlan,
799 bundle->trunks, bundle->use_priority_tags,
800 bundle->bond, bundle->lacp,
804 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
805 xlate_ofport_set(ofproto, ofport->bundle, ofport,
806 ofport->up.ofp_port, ofport->odp_port,
807 ofport->up.netdev, ofport->cfm,
808 ofport->bfd, ofport->peer,
809 ofport->up.pp.config, ofport->stp_state,
810 ofport->is_tunnel, ofport->may_enable);
814 cls_cursor_init(&cursor, &ofproto->facets, NULL);
815 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
817 || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
818 facet_revalidate(facet);
825 if (!backer->recv_set_enable) {
826 /* Wake up before a max of 1000ms. */
827 timer_set_duration(&backer->next_expiration, 1000);
828 } else if (timer_expired(&backer->next_expiration)) {
829 int delay = expire(backer);
830 timer_set_duration(&backer->next_expiration, delay);
833 /* Check for port changes in the dpif. */
834 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
835 struct ofproto_dpif *ofproto;
836 struct dpif_port port;
838 /* Don't report on the datapath's device. */
839 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
843 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
844 &all_ofproto_dpifs) {
845 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
850 ofproto = lookup_ofproto_dpif_by_port_name(devname);
851 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
852 /* The port was removed. If we know the datapath,
853 * report it through poll_set(). If we don't, it may be
854 * notifying us of a removal we initiated, so ignore it.
855 * If there's a pending ENOBUFS, let it stand, since
856 * everything will be reevaluated. */
857 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
858 sset_add(&ofproto->port_poll_set, devname);
859 ofproto->port_poll_errno = 0;
861 } else if (!ofproto) {
862 /* The port was added, but we don't know with which
863 * ofproto we should associate it. Delete it. */
864 dpif_port_del(backer->dpif, port.port_no);
866 dpif_port_destroy(&port);
872 if (error != EAGAIN) {
873 struct ofproto_dpif *ofproto;
875 /* There was some sort of error, so propagate it to all
876 * ofprotos that use this backer. */
877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
878 &all_ofproto_dpifs) {
879 if (ofproto->backer == backer) {
880 sset_clear(&ofproto->port_poll_set);
881 ofproto->port_poll_errno = error;
886 if (backer->governor) {
889 governor_run(backer->governor);
891 /* If the governor has shrunk to its minimum size and the number of
892 * subfacets has dwindled, then drop the governor entirely.
894 * For hysteresis, the number of subfacets to drop the governor is
895 * smaller than the number needed to trigger its creation. */
896 n_subfacets = hmap_count(&backer->subfacets);
897 if (n_subfacets * 4 < flow_eviction_threshold
898 && governor_is_idle(backer->governor)) {
899 governor_destroy(backer->governor);
900 backer->governor = NULL;
908 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
912 /* If recv_set_enable is false, we should not handle upcalls. */
913 if (!backer->recv_set_enable) {
917 /* Handle one or more batches of upcalls, until there's nothing left to do
918 * or until we do a fixed total amount of work.
920 * We do work in batches because it can be much cheaper to set up a number
921 * of flows and fire off their patches all at once. We do multiple batches
922 * because in some cases handling a packet can cause another packet to be
923 * queued almost immediately as part of the return flow. Both
924 * optimizations can make major improvements on some benchmarks and
925 * presumably for real traffic as well. */
927 while (work < max_batch) {
928 int retval = handle_upcalls(backer, max_batch - work);
939 type_run_fast(const char *type)
941 struct dpif_backer *backer;
943 backer = shash_find_data(&all_dpif_backers, type);
945 /* This is not necessarily a problem, since backers are only
946 * created on demand. */
950 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
956 static long long int port_rl = LLONG_MIN;
957 static unsigned int backer_rl = 0;
959 if (time_msec() >= port_rl) {
960 struct ofproto_dpif *ofproto;
961 struct ofport_dpif *ofport;
963 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
965 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
966 port_run_fast(ofport);
969 port_rl = time_msec() + 200;
972 /* XXX: We have to be careful not to do too much work in this function. If
973 * we call dpif_backer_run_fast() too often, or with too large a batch,
974 * performance improves signifcantly, but at a cost. It's possible for the
975 * number of flows in the datapath to increase without bound, and for poll
976 * loops to take 10s of seconds. The correct solution to this problem,
977 * long term, is to separate flow miss handling into it's own thread so it
978 * isn't affected by revalidations, and expirations. Until then, this is
979 * the best we can do. */
980 if (++backer_rl >= 10) {
981 struct shash_node *node;
984 SHASH_FOR_EACH (node, &all_dpif_backers) {
985 dpif_backer_run_fast(node->data, 1);
991 type_wait(const char *type)
993 struct dpif_backer *backer;
995 backer = shash_find_data(&all_dpif_backers, type);
997 /* This is not necessarily a problem, since backers are only
998 * created on demand. */
1002 if (backer->governor) {
1003 governor_wait(backer->governor);
1006 timer_wait(&backer->next_expiration);
1009 /* Basic life-cycle. */
1011 static int add_internal_flows(struct ofproto_dpif *);
1013 static struct ofproto *
1016 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1017 return &ofproto->up;
1021 dealloc(struct ofproto *ofproto_)
1023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1028 close_dpif_backer(struct dpif_backer *backer)
1030 struct shash_node *node;
1032 ovs_assert(backer->refcount > 0);
1034 if (--backer->refcount) {
1038 drop_key_clear(backer);
1039 hmap_destroy(&backer->drop_keys);
1041 simap_destroy(&backer->tnl_backers);
1042 hmap_destroy(&backer->odp_to_ofport_map);
1043 node = shash_find(&all_dpif_backers, backer->type);
1045 shash_delete(&all_dpif_backers, node);
1046 dpif_close(backer->dpif);
1048 ovs_assert(hmap_is_empty(&backer->subfacets));
1049 hmap_destroy(&backer->subfacets);
1050 governor_destroy(backer->governor);
1055 /* Datapath port slated for removal from datapath. */
1056 struct odp_garbage {
1057 struct list list_node;
1058 odp_port_t odp_port;
1062 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1064 struct dpif_backer *backer;
1065 struct dpif_port_dump port_dump;
1066 struct dpif_port port;
1067 struct shash_node *node;
1068 struct list garbage_list;
1069 struct odp_garbage *garbage, *next;
1075 backer = shash_find_data(&all_dpif_backers, type);
1082 backer_name = xasprintf("ovs-%s", type);
1084 /* Remove any existing datapaths, since we assume we're the only
1085 * userspace controlling the datapath. */
1087 dp_enumerate_names(type, &names);
1088 SSET_FOR_EACH(name, &names) {
1089 struct dpif *old_dpif;
1091 /* Don't remove our backer if it exists. */
1092 if (!strcmp(name, backer_name)) {
1096 if (dpif_open(name, type, &old_dpif)) {
1097 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1099 dpif_delete(old_dpif);
1100 dpif_close(old_dpif);
1103 sset_destroy(&names);
1105 backer = xmalloc(sizeof *backer);
1107 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1110 VLOG_ERR("failed to open datapath of type %s: %s", type,
1111 ovs_strerror(error));
1116 backer->type = xstrdup(type);
1117 backer->governor = NULL;
1118 backer->refcount = 1;
1119 hmap_init(&backer->odp_to_ofport_map);
1120 hmap_init(&backer->drop_keys);
1121 hmap_init(&backer->subfacets);
1122 timer_set_duration(&backer->next_expiration, 1000);
1123 backer->need_revalidate = 0;
1124 simap_init(&backer->tnl_backers);
1125 tag_set_init(&backer->revalidate_set);
1126 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1129 if (backer->recv_set_enable) {
1130 dpif_flow_flush(backer->dpif);
1133 /* Loop through the ports already on the datapath and remove any
1134 * that we don't need anymore. */
1135 list_init(&garbage_list);
1136 dpif_port_dump_start(&port_dump, backer->dpif);
1137 while (dpif_port_dump_next(&port_dump, &port)) {
1138 node = shash_find(&init_ofp_ports, port.name);
1139 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1140 garbage = xmalloc(sizeof *garbage);
1141 garbage->odp_port = port.port_no;
1142 list_push_front(&garbage_list, &garbage->list_node);
1145 dpif_port_dump_done(&port_dump);
1147 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1148 dpif_port_del(backer->dpif, garbage->odp_port);
1149 list_remove(&garbage->list_node);
1153 shash_add(&all_dpif_backers, type, backer);
1155 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1157 VLOG_ERR("failed to listen on datapath of type %s: %s",
1158 type, ovs_strerror(error));
1159 close_dpif_backer(backer);
1163 backer->max_n_subfacet = 0;
1164 backer->created = time_msec();
1165 backer->last_minute = backer->created;
1166 memset(&backer->hourly, 0, sizeof backer->hourly);
1167 memset(&backer->daily, 0, sizeof backer->daily);
1168 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1169 backer->subfacet_add_count = 0;
1170 backer->subfacet_del_count = 0;
1171 backer->total_subfacet_add_count = 0;
1172 backer->total_subfacet_del_count = 0;
1173 backer->avg_n_subfacet = 0;
1174 backer->avg_subfacet_life = 0;
1180 construct(struct ofproto *ofproto_)
1182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1183 struct shash_node *node, *next;
1184 odp_port_t max_ports;
1188 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1193 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1194 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1195 ofp_to_u16(OFPP_MAX))));
1197 ofproto->netflow = NULL;
1198 ofproto->sflow = NULL;
1199 ofproto->ipfix = NULL;
1200 ofproto->stp = NULL;
1201 hmap_init(&ofproto->bundles);
1202 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1203 ofproto->mbridge = mbridge_create();
1204 ofproto->has_bonded_bundles = false;
1206 classifier_init(&ofproto->facets);
1207 ofproto->consistency_rl = LLONG_MIN;
1209 for (i = 0; i < N_TABLES; i++) {
1210 struct table_dpif *table = &ofproto->tables[i];
1212 table->catchall_table = NULL;
1213 table->other_table = NULL;
1214 table->basis = random_uint32();
1217 list_init(&ofproto->completions);
1219 ofproto_dpif_unixctl_init();
1221 hmap_init(&ofproto->vlandev_map);
1222 hmap_init(&ofproto->realdev_vid_map);
1224 sset_init(&ofproto->ports);
1225 sset_init(&ofproto->ghost_ports);
1226 sset_init(&ofproto->port_poll_set);
1227 ofproto->port_poll_errno = 0;
1229 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1230 struct iface_hint *iface_hint = node->data;
1232 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1233 /* Check if the datapath already has this port. */
1234 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1235 sset_add(&ofproto->ports, node->name);
1238 free(iface_hint->br_name);
1239 free(iface_hint->br_type);
1241 shash_delete(&init_ofp_ports, node);
1245 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1246 hash_string(ofproto->up.name, 0));
1247 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1249 ofproto_init_tables(ofproto_, N_TABLES);
1250 error = add_internal_flows(ofproto);
1251 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1254 ofproto->n_missed = 0;
1260 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1261 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1263 struct ofputil_flow_mod fm;
1266 match_init_catchall(&fm.match);
1268 match_set_reg(&fm.match, 0, id);
1269 fm.new_cookie = htonll(0);
1270 fm.cookie = htonll(0);
1271 fm.cookie_mask = htonll(0);
1272 fm.modify_cookie = false;
1273 fm.table_id = TBL_INTERNAL;
1274 fm.command = OFPFC_ADD;
1275 fm.idle_timeout = 0;
1276 fm.hard_timeout = 0;
1280 fm.ofpacts = ofpacts->data;
1281 fm.ofpacts_len = ofpacts->size;
1283 error = ofproto_flow_mod(&ofproto->up, &fm);
1285 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1286 id, ofperr_to_string(error));
1290 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1292 ovs_assert(*rulep != NULL);
1298 add_internal_flows(struct ofproto_dpif *ofproto)
1300 struct ofpact_controller *controller;
1301 uint64_t ofpacts_stub[128 / 8];
1302 struct ofpbuf ofpacts;
1306 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1309 controller = ofpact_put_CONTROLLER(&ofpacts);
1310 controller->max_len = UINT16_MAX;
1311 controller->controller_id = 0;
1312 controller->reason = OFPR_NO_MATCH;
1313 ofpact_pad(&ofpacts);
1315 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1320 ofpbuf_clear(&ofpacts);
1321 error = add_internal_flow(ofproto, id++, &ofpacts,
1322 &ofproto->no_packet_in_rule);
1327 error = add_internal_flow(ofproto, id++, &ofpacts,
1328 &ofproto->drop_frags_rule);
1333 complete_operations(struct ofproto_dpif *ofproto)
1335 struct dpif_completion *c, *next;
1337 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1338 ofoperation_complete(c->op, 0);
1339 list_remove(&c->list_node);
1345 destruct(struct ofproto *ofproto_)
1347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1348 struct rule_dpif *rule, *next_rule;
1349 struct oftable *table;
1351 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1352 xlate_remove_ofproto(ofproto);
1354 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1355 complete_operations(ofproto);
1357 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1358 struct cls_cursor cursor;
1360 cls_cursor_init(&cursor, &table->cls, NULL);
1361 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1362 ofproto_rule_destroy(&rule->up);
1366 mbridge_unref(ofproto->mbridge);
1368 netflow_destroy(ofproto->netflow);
1369 dpif_sflow_unref(ofproto->sflow);
1370 hmap_destroy(&ofproto->bundles);
1371 mac_learning_unref(ofproto->ml);
1373 classifier_destroy(&ofproto->facets);
1375 hmap_destroy(&ofproto->vlandev_map);
1376 hmap_destroy(&ofproto->realdev_vid_map);
1378 sset_destroy(&ofproto->ports);
1379 sset_destroy(&ofproto->ghost_ports);
1380 sset_destroy(&ofproto->port_poll_set);
1382 close_dpif_backer(ofproto->backer);
1386 run_fast(struct ofproto *ofproto_)
1388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1389 struct ofport_dpif *ofport;
1391 /* Do not perform any periodic activity required by 'ofproto' while
1392 * waiting for flow restore to complete. */
1393 if (ofproto_get_flow_restore_wait()) {
1397 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1398 port_run_fast(ofport);
1405 run(struct ofproto *ofproto_)
1407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1408 struct ofport_dpif *ofport;
1409 struct ofbundle *bundle;
1413 complete_operations(ofproto);
1416 if (mbridge_need_revalidate(ofproto->mbridge)) {
1417 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1418 mac_learning_flush(ofproto->ml, NULL);
1421 /* Do not perform any periodic activity below required by 'ofproto' while
1422 * waiting for flow restore to complete. */
1423 if (ofproto_get_flow_restore_wait()) {
1427 error = run_fast(ofproto_);
1432 if (ofproto->netflow) {
1433 if (netflow_run(ofproto->netflow)) {
1434 send_netflow_active_timeouts(ofproto);
1437 if (ofproto->sflow) {
1438 dpif_sflow_run(ofproto->sflow);
1441 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1444 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1449 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1451 /* Check the consistency of a random facet, to aid debugging. */
1452 if (time_msec() >= ofproto->consistency_rl
1453 && !classifier_is_empty(&ofproto->facets)
1454 && !ofproto->backer->need_revalidate) {
1455 struct cls_table *table;
1456 struct cls_rule *cr;
1457 struct facet *facet;
1459 ofproto->consistency_rl = time_msec() + 250;
1461 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1462 struct cls_table, hmap_node);
1463 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1465 facet = CONTAINER_OF(cr, struct facet, cr);
1467 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1468 facet->xout.tags)) {
1469 if (!facet_check_consistency(facet)) {
1470 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1479 wait(struct ofproto *ofproto_)
1481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1482 struct ofport_dpif *ofport;
1483 struct ofbundle *bundle;
1485 if (!clogged && !list_is_empty(&ofproto->completions)) {
1486 poll_immediate_wake();
1489 if (ofproto_get_flow_restore_wait()) {
1493 dpif_wait(ofproto->backer->dpif);
1494 dpif_recv_wait(ofproto->backer->dpif);
1495 if (ofproto->sflow) {
1496 dpif_sflow_wait(ofproto->sflow);
1498 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1499 poll_immediate_wake();
1501 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1504 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1505 bundle_wait(bundle);
1507 if (ofproto->netflow) {
1508 netflow_wait(ofproto->netflow);
1510 mac_learning_wait(ofproto->ml);
1512 if (ofproto->backer->need_revalidate) {
1513 /* Shouldn't happen, but if it does just go around again. */
1514 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1515 poll_immediate_wake();
1520 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1522 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1523 struct cls_cursor cursor;
1524 size_t n_subfacets = 0;
1525 struct facet *facet;
1527 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1529 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1530 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1531 n_subfacets += list_size(&facet->subfacets);
1533 simap_increase(usage, "subfacets", n_subfacets);
1537 flush(struct ofproto *ofproto_)
1539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1540 struct subfacet *subfacet, *next_subfacet;
1541 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1545 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1546 &ofproto->backer->subfacets) {
1547 if (subfacet->facet->ofproto != ofproto) {
1551 if (subfacet->path != SF_NOT_INSTALLED) {
1552 batch[n_batch++] = subfacet;
1553 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1554 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1558 subfacet_destroy(subfacet);
1563 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1568 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1569 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1571 *arp_match_ip = true;
1572 *actions = (OFPUTIL_A_OUTPUT |
1573 OFPUTIL_A_SET_VLAN_VID |
1574 OFPUTIL_A_SET_VLAN_PCP |
1575 OFPUTIL_A_STRIP_VLAN |
1576 OFPUTIL_A_SET_DL_SRC |
1577 OFPUTIL_A_SET_DL_DST |
1578 OFPUTIL_A_SET_NW_SRC |
1579 OFPUTIL_A_SET_NW_DST |
1580 OFPUTIL_A_SET_NW_TOS |
1581 OFPUTIL_A_SET_TP_SRC |
1582 OFPUTIL_A_SET_TP_DST |
1587 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1590 struct dpif_dp_stats s;
1591 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1594 strcpy(ots->name, "classifier");
1596 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1597 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1598 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1599 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1601 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1602 ots->lookup_count = htonll(n_lookup);
1603 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1606 static struct ofport *
1609 struct ofport_dpif *port = xmalloc(sizeof *port);
1614 port_dealloc(struct ofport *port_)
1616 struct ofport_dpif *port = ofport_dpif_cast(port_);
1621 port_construct(struct ofport *port_)
1623 struct ofport_dpif *port = ofport_dpif_cast(port_);
1624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1625 const struct netdev *netdev = port->up.netdev;
1626 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1627 struct dpif_port dpif_port;
1630 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1631 port->bundle = NULL;
1634 port->tag = tag_create_random();
1635 port->may_enable = true;
1636 port->stp_port = NULL;
1637 port->stp_state = STP_DISABLED;
1638 port->is_tunnel = false;
1640 hmap_init(&port->priorities);
1641 port->realdev_ofp_port = 0;
1642 port->vlandev_vid = 0;
1643 port->carrier_seq = netdev_get_carrier_resets(netdev);
1645 if (netdev_vport_is_patch(netdev)) {
1646 /* By bailing out here, we don't submit the port to the sFlow module
1647 * to be considered for counter polling export. This is correct
1648 * because the patch port represents an interface that sFlow considers
1649 * to be "internal" to the switch as a whole, and therefore not an
1650 * candidate for counter polling. */
1651 port->odp_port = ODPP_NONE;
1652 ofport_update_peer(port);
1656 error = dpif_port_query_by_name(ofproto->backer->dpif,
1657 netdev_vport_get_dpif_port(netdev, namebuf,
1664 port->odp_port = dpif_port.port_no;
1666 if (netdev_get_tunnel_config(netdev)) {
1667 tnl_port_add(port, port->up.netdev, port->odp_port);
1668 port->is_tunnel = true;
1670 /* Sanity-check that a mapping doesn't already exist. This
1671 * shouldn't happen for non-tunnel ports. */
1672 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1673 VLOG_ERR("port %s already has an OpenFlow port number",
1675 dpif_port_destroy(&dpif_port);
1679 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1680 hash_odp_port(port->odp_port));
1682 dpif_port_destroy(&dpif_port);
1684 if (ofproto->sflow) {
1685 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1692 port_destruct(struct ofport *port_)
1694 struct ofport_dpif *port = ofport_dpif_cast(port_);
1695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1696 const char *devname = netdev_get_name(port->up.netdev);
1697 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1698 const char *dp_port_name;
1700 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1701 xlate_ofport_remove(port);
1703 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1705 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1706 /* The underlying device is still there, so delete it. This
1707 * happens when the ofproto is being destroyed, since the caller
1708 * assumes that removal of attached ports will happen as part of
1710 if (!port->is_tunnel) {
1711 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1716 port->peer->peer = NULL;
1720 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1721 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1725 sset_find_and_delete(&ofproto->ports, devname);
1726 sset_find_and_delete(&ofproto->ghost_ports, devname);
1727 bundle_remove(port_);
1728 set_cfm(port_, NULL);
1729 set_bfd(port_, NULL);
1730 if (ofproto->sflow) {
1731 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1734 ofport_clear_priorities(port);
1735 hmap_destroy(&port->priorities);
1739 port_modified(struct ofport *port_)
1741 struct ofport_dpif *port = ofport_dpif_cast(port_);
1743 if (port->bundle && port->bundle->bond) {
1744 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1748 cfm_set_netdev(port->cfm, port->up.netdev);
1751 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1753 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1757 ofport_update_peer(port);
1761 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1763 struct ofport_dpif *port = ofport_dpif_cast(port_);
1764 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1765 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1767 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1768 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1769 OFPUTIL_PC_NO_PACKET_IN)) {
1770 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1772 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1773 bundle_update(port->bundle);
1779 set_sflow(struct ofproto *ofproto_,
1780 const struct ofproto_sflow_options *sflow_options)
1782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1783 struct dpif_sflow *ds = ofproto->sflow;
1785 if (sflow_options) {
1787 struct ofport_dpif *ofport;
1789 ds = ofproto->sflow = dpif_sflow_create();
1790 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1791 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1793 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1795 dpif_sflow_set_options(ds, sflow_options);
1798 dpif_sflow_unref(ds);
1799 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1800 ofproto->sflow = NULL;
1808 struct ofproto *ofproto_,
1809 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1810 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1811 size_t n_flow_exporters_options)
1813 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1814 struct dpif_ipfix *di = ofproto->ipfix;
1816 if (bridge_exporter_options || flow_exporters_options) {
1818 di = ofproto->ipfix = dpif_ipfix_create();
1820 dpif_ipfix_set_options(
1821 di, bridge_exporter_options, flow_exporters_options,
1822 n_flow_exporters_options);
1825 dpif_ipfix_unref(di);
1826 ofproto->ipfix = NULL;
1833 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1835 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1842 struct ofproto_dpif *ofproto;
1844 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1845 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1846 ofport->cfm = cfm_create(ofport->up.netdev);
1849 if (cfm_configure(ofport->cfm, s)) {
1855 cfm_unref(ofport->cfm);
1861 get_cfm_status(const struct ofport *ofport_,
1862 struct ofproto_cfm_status *status)
1864 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1867 status->faults = cfm_get_fault(ofport->cfm);
1868 status->remote_opstate = cfm_get_opup(ofport->cfm);
1869 status->health = cfm_get_health(ofport->cfm);
1870 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1878 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1881 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1885 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1886 if (ofport->bfd != old) {
1887 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1896 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1899 bfd_get_status(ofport->bfd, smap);
1906 /* Spanning Tree. */
1909 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1911 struct ofproto_dpif *ofproto = ofproto_;
1912 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1913 struct ofport_dpif *ofport;
1915 ofport = stp_port_get_aux(sp);
1917 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1918 ofproto->up.name, port_num);
1920 struct eth_header *eth = pkt->l2;
1922 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1923 if (eth_addr_is_zero(eth->eth_src)) {
1924 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1925 "with unknown MAC", ofproto->up.name, port_num);
1927 send_packet(ofport, pkt);
1933 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1935 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1939 /* Only revalidate flows if the configuration changed. */
1940 if (!s != !ofproto->stp) {
1941 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1945 if (!ofproto->stp) {
1946 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1947 send_bpdu_cb, ofproto);
1948 ofproto->stp_last_tick = time_msec();
1951 stp_set_bridge_id(ofproto->stp, s->system_id);
1952 stp_set_bridge_priority(ofproto->stp, s->priority);
1953 stp_set_hello_time(ofproto->stp, s->hello_time);
1954 stp_set_max_age(ofproto->stp, s->max_age);
1955 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1957 struct ofport *ofport;
1959 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1960 set_stp_port(ofport, NULL);
1963 stp_destroy(ofproto->stp);
1964 ofproto->stp = NULL;
1971 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1977 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1978 s->designated_root = stp_get_designated_root(ofproto->stp);
1979 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1988 update_stp_port_state(struct ofport_dpif *ofport)
1990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1991 enum stp_state state;
1993 /* Figure out new state. */
1994 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1998 if (ofport->stp_state != state) {
1999 enum ofputil_port_state of_state;
2002 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2003 netdev_get_name(ofport->up.netdev),
2004 stp_state_name(ofport->stp_state),
2005 stp_state_name(state));
2006 if (stp_learn_in_state(ofport->stp_state)
2007 != stp_learn_in_state(state)) {
2008 /* xxx Learning action flows should also be flushed. */
2009 mac_learning_flush(ofproto->ml,
2010 &ofproto->backer->revalidate_set);
2012 fwd_change = stp_forward_in_state(ofport->stp_state)
2013 != stp_forward_in_state(state);
2015 ofproto->backer->need_revalidate = REV_STP;
2016 ofport->stp_state = state;
2017 ofport->stp_state_entered = time_msec();
2019 if (fwd_change && ofport->bundle) {
2020 bundle_update(ofport->bundle);
2023 /* Update the STP state bits in the OpenFlow port description. */
2024 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2025 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2026 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2027 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2028 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2030 ofproto_port_set_state(&ofport->up, of_state);
2034 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2035 * caller is responsible for assigning STP port numbers and ensuring
2036 * there are no duplicates. */
2038 set_stp_port(struct ofport *ofport_,
2039 const struct ofproto_port_stp_settings *s)
2041 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2043 struct stp_port *sp = ofport->stp_port;
2045 if (!s || !s->enable) {
2047 ofport->stp_port = NULL;
2048 stp_port_disable(sp);
2049 update_stp_port_state(ofport);
2052 } else if (sp && stp_port_no(sp) != s->port_num
2053 && ofport == stp_port_get_aux(sp)) {
2054 /* The port-id changed, so disable the old one if it's not
2055 * already in use by another port. */
2056 stp_port_disable(sp);
2059 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2060 stp_port_enable(sp);
2062 stp_port_set_aux(sp, ofport);
2063 stp_port_set_priority(sp, s->priority);
2064 stp_port_set_path_cost(sp, s->path_cost);
2066 update_stp_port_state(ofport);
2072 get_stp_port_status(struct ofport *ofport_,
2073 struct ofproto_port_stp_status *s)
2075 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2077 struct stp_port *sp = ofport->stp_port;
2079 if (!ofproto->stp || !sp) {
2085 s->port_id = stp_port_get_id(sp);
2086 s->state = stp_port_get_state(sp);
2087 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2088 s->role = stp_port_get_role(sp);
2089 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2095 stp_run(struct ofproto_dpif *ofproto)
2098 long long int now = time_msec();
2099 long long int elapsed = now - ofproto->stp_last_tick;
2100 struct stp_port *sp;
2103 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2104 ofproto->stp_last_tick = now;
2106 while (stp_get_changed_port(ofproto->stp, &sp)) {
2107 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2110 update_stp_port_state(ofport);
2114 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2115 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2121 stp_wait(struct ofproto_dpif *ofproto)
2124 poll_timer_wait(1000);
2128 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2129 * were used to make the determination.*/
2131 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2133 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2134 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2138 stp_process_packet(const struct ofport_dpif *ofport,
2139 const struct ofpbuf *packet)
2141 struct ofpbuf payload = *packet;
2142 struct eth_header *eth = payload.data;
2143 struct stp_port *sp = ofport->stp_port;
2145 /* Sink packets on ports that have STP disabled when the bridge has
2147 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2151 /* Trim off padding on payload. */
2152 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2153 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2156 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2157 stp_received_bpdu(sp, payload.data, payload.size);
2162 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
2163 uint32_t queue_id, uint32_t *priority)
2165 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
2168 static struct priority_to_dscp *
2169 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2171 struct priority_to_dscp *pdscp;
2174 hash = hash_int(priority, 0);
2175 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2176 if (pdscp->priority == priority) {
2184 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2185 uint32_t priority, uint8_t *dscp)
2187 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2188 *dscp = pdscp ? pdscp->dscp : 0;
2189 return pdscp != NULL;
2193 ofport_clear_priorities(struct ofport_dpif *ofport)
2195 struct priority_to_dscp *pdscp, *next;
2197 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2198 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2204 set_queues(struct ofport *ofport_,
2205 const struct ofproto_port_queue *qdscp_list,
2208 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2210 struct hmap new = HMAP_INITIALIZER(&new);
2213 for (i = 0; i < n_qdscp; i++) {
2214 struct priority_to_dscp *pdscp;
2218 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2219 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2224 pdscp = get_priority(ofport, priority);
2226 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2228 pdscp = xmalloc(sizeof *pdscp);
2229 pdscp->priority = priority;
2231 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2234 if (pdscp->dscp != dscp) {
2236 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2239 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2242 if (!hmap_is_empty(&ofport->priorities)) {
2243 ofport_clear_priorities(ofport);
2244 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2247 hmap_swap(&new, &ofport->priorities);
2255 /* Expires all MAC learning entries associated with 'bundle' and forces its
2256 * ofproto to revalidate every flow.
2258 * Normally MAC learning entries are removed only from the ofproto associated
2259 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2260 * are removed from every ofproto. When patch ports and SLB bonds are in use
2261 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2262 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2263 * with the host from which it migrated. */
2265 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2267 struct ofproto_dpif *ofproto = bundle->ofproto;
2268 struct mac_learning *ml = ofproto->ml;
2269 struct mac_entry *mac, *next_mac;
2271 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2272 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2273 if (mac->port.p == bundle) {
2275 struct ofproto_dpif *o;
2277 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2279 struct mac_entry *e;
2281 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2284 mac_learning_expire(o->ml, e);
2290 mac_learning_expire(ml, mac);
2295 static struct ofbundle *
2296 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2298 struct ofbundle *bundle;
2300 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2301 &ofproto->bundles) {
2302 if (bundle->aux == aux) {
2310 bundle_update(struct ofbundle *bundle)
2312 struct ofport_dpif *port;
2314 bundle->floodable = true;
2315 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2316 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2317 || !stp_forward_in_state(port->stp_state)) {
2318 bundle->floodable = false;
2325 bundle_del_port(struct ofport_dpif *port)
2327 struct ofbundle *bundle = port->bundle;
2329 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2331 list_remove(&port->bundle_node);
2332 port->bundle = NULL;
2335 lacp_slave_unregister(bundle->lacp, port);
2338 bond_slave_unregister(bundle->bond, port);
2341 bundle_update(bundle);
2345 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2346 struct lacp_slave_settings *lacp)
2348 struct ofport_dpif *port;
2350 port = get_ofp_port(bundle->ofproto, ofp_port);
2355 if (port->bundle != bundle) {
2356 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2358 bundle_del_port(port);
2361 port->bundle = bundle;
2362 list_push_back(&bundle->ports, &port->bundle_node);
2363 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2364 || !stp_forward_in_state(port->stp_state)) {
2365 bundle->floodable = false;
2369 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370 lacp_slave_register(bundle->lacp, port, lacp);
2377 bundle_destroy(struct ofbundle *bundle)
2379 struct ofproto_dpif *ofproto;
2380 struct ofport_dpif *port, *next_port;
2386 ofproto = bundle->ofproto;
2387 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2389 xlate_bundle_remove(bundle);
2391 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2392 bundle_del_port(port);
2395 bundle_flush_macs(bundle, true);
2396 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2398 free(bundle->trunks);
2399 lacp_unref(bundle->lacp);
2400 bond_unref(bundle->bond);
2405 bundle_set(struct ofproto *ofproto_, void *aux,
2406 const struct ofproto_bundle_settings *s)
2408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2409 bool need_flush = false;
2410 struct ofport_dpif *port;
2411 struct ofbundle *bundle;
2412 unsigned long *trunks;
2418 bundle_destroy(bundle_lookup(ofproto, aux));
2422 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2423 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2425 bundle = bundle_lookup(ofproto, aux);
2427 bundle = xmalloc(sizeof *bundle);
2429 bundle->ofproto = ofproto;
2430 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2431 hash_pointer(aux, 0));
2433 bundle->name = NULL;
2435 list_init(&bundle->ports);
2436 bundle->vlan_mode = PORT_VLAN_TRUNK;
2438 bundle->trunks = NULL;
2439 bundle->use_priority_tags = s->use_priority_tags;
2440 bundle->lacp = NULL;
2441 bundle->bond = NULL;
2443 bundle->floodable = true;
2444 mbridge_register_bundle(ofproto->mbridge, bundle);
2447 if (!bundle->name || strcmp(s->name, bundle->name)) {
2449 bundle->name = xstrdup(s->name);
2454 if (!bundle->lacp) {
2455 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2456 bundle->lacp = lacp_create();
2458 lacp_configure(bundle->lacp, s->lacp);
2460 lacp_unref(bundle->lacp);
2461 bundle->lacp = NULL;
2464 /* Update set of ports. */
2466 for (i = 0; i < s->n_slaves; i++) {
2467 if (!bundle_add_port(bundle, s->slaves[i],
2468 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2472 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2473 struct ofport_dpif *next_port;
2475 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2476 for (i = 0; i < s->n_slaves; i++) {
2477 if (s->slaves[i] == port->up.ofp_port) {
2482 bundle_del_port(port);
2486 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2488 if (list_is_empty(&bundle->ports)) {
2489 bundle_destroy(bundle);
2493 /* Set VLAN tagging mode */
2494 if (s->vlan_mode != bundle->vlan_mode
2495 || s->use_priority_tags != bundle->use_priority_tags) {
2496 bundle->vlan_mode = s->vlan_mode;
2497 bundle->use_priority_tags = s->use_priority_tags;
2502 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2503 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2505 if (vlan != bundle->vlan) {
2506 bundle->vlan = vlan;
2510 /* Get trunked VLANs. */
2511 switch (s->vlan_mode) {
2512 case PORT_VLAN_ACCESS:
2516 case PORT_VLAN_TRUNK:
2517 trunks = CONST_CAST(unsigned long *, s->trunks);
2520 case PORT_VLAN_NATIVE_UNTAGGED:
2521 case PORT_VLAN_NATIVE_TAGGED:
2522 if (vlan != 0 && (!s->trunks
2523 || !bitmap_is_set(s->trunks, vlan)
2524 || bitmap_is_set(s->trunks, 0))) {
2525 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2527 trunks = bitmap_clone(s->trunks, 4096);
2529 trunks = bitmap_allocate1(4096);
2531 bitmap_set1(trunks, vlan);
2532 bitmap_set0(trunks, 0);
2534 trunks = CONST_CAST(unsigned long *, s->trunks);
2541 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2542 free(bundle->trunks);
2543 if (trunks == s->trunks) {
2544 bundle->trunks = vlan_bitmap_clone(trunks);
2546 bundle->trunks = trunks;
2551 if (trunks != s->trunks) {
2556 if (!list_is_short(&bundle->ports)) {
2557 bundle->ofproto->has_bonded_bundles = true;
2559 if (bond_reconfigure(bundle->bond, s->bond)) {
2560 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2563 bundle->bond = bond_create(s->bond);
2564 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2567 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2568 bond_slave_register(bundle->bond, port, port->up.netdev);
2571 bond_unref(bundle->bond);
2572 bundle->bond = NULL;
2575 /* If we changed something that would affect MAC learning, un-learn
2576 * everything on this port and force flow revalidation. */
2578 bundle_flush_macs(bundle, false);
2585 bundle_remove(struct ofport *port_)
2587 struct ofport_dpif *port = ofport_dpif_cast(port_);
2588 struct ofbundle *bundle = port->bundle;
2591 bundle_del_port(port);
2592 if (list_is_empty(&bundle->ports)) {
2593 bundle_destroy(bundle);
2594 } else if (list_is_short(&bundle->ports)) {
2595 bond_unref(bundle->bond);
2596 bundle->bond = NULL;
2602 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2604 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2605 struct ofport_dpif *port = port_;
2606 uint8_t ea[ETH_ADDR_LEN];
2609 error = netdev_get_etheraddr(port->up.netdev, ea);
2611 struct ofpbuf packet;
2614 ofpbuf_init(&packet, 0);
2615 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2617 memcpy(packet_pdu, pdu, pdu_size);
2619 send_packet(port, &packet);
2620 ofpbuf_uninit(&packet);
2622 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2623 "%s (%s)", port->bundle->name,
2624 netdev_get_name(port->up.netdev), ovs_strerror(error));
2629 bundle_send_learning_packets(struct ofbundle *bundle)
2631 struct ofproto_dpif *ofproto = bundle->ofproto;
2632 int error, n_packets, n_errors;
2633 struct mac_entry *e;
2635 error = n_packets = n_errors = 0;
2636 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2637 if (e->port.p != bundle) {
2638 struct ofpbuf *learning_packet;
2639 struct ofport_dpif *port;
2643 /* The assignment to "port" is unnecessary but makes "grep"ing for
2644 * struct ofport_dpif more effective. */
2645 learning_packet = bond_compose_learning_packet(bundle->bond,
2649 ret = send_packet(port, learning_packet);
2650 ofpbuf_delete(learning_packet);
2660 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2661 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2662 "packets, last error was: %s",
2663 bundle->name, n_errors, n_packets, ovs_strerror(error));
2665 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2666 bundle->name, n_packets);
2671 bundle_run(struct ofbundle *bundle)
2674 lacp_run(bundle->lacp, send_pdu_cb);
2677 struct ofport_dpif *port;
2679 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2680 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2683 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2684 lacp_status(bundle->lacp));
2685 if (bond_should_send_learning_packets(bundle->bond)) {
2686 bundle_send_learning_packets(bundle);
2692 bundle_wait(struct ofbundle *bundle)
2695 lacp_wait(bundle->lacp);
2698 bond_wait(bundle->bond);
2705 mirror_set__(struct ofproto *ofproto_, void *aux,
2706 const struct ofproto_mirror_settings *s)
2708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2709 struct ofbundle **srcs, **dsts;
2714 mirror_destroy(ofproto->mbridge, aux);
2718 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2719 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2721 for (i = 0; i < s->n_srcs; i++) {
2722 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2725 for (i = 0; i < s->n_dsts; i++) {
2726 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2729 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2730 s->n_dsts, s->src_vlans,
2731 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2738 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2739 uint64_t *packets, uint64_t *bytes)
2742 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2747 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2750 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2751 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2757 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2759 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2760 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2761 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2765 forward_bpdu_changed(struct ofproto *ofproto_)
2767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2768 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2772 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2776 mac_learning_set_idle_time(ofproto->ml, idle_time);
2777 mac_learning_set_max_entries(ofproto->ml, max_entries);
2782 static struct ofport_dpif *
2783 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2785 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2786 return ofport ? ofport_dpif_cast(ofport) : NULL;
2789 static struct ofport_dpif *
2790 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2792 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2793 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2797 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2798 struct ofproto_port *ofproto_port,
2799 struct dpif_port *dpif_port)
2801 ofproto_port->name = dpif_port->name;
2802 ofproto_port->type = dpif_port->type;
2803 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2807 ofport_update_peer(struct ofport_dpif *ofport)
2809 const struct ofproto_dpif *ofproto;
2810 struct dpif_backer *backer;
2811 const char *peer_name;
2813 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2817 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2818 backer->need_revalidate = REV_RECONFIGURE;
2821 ofport->peer->peer = NULL;
2822 ofport->peer = NULL;
2825 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2830 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2831 struct ofport *peer_ofport;
2832 struct ofport_dpif *peer;
2833 const char *peer_peer;
2835 if (ofproto->backer != backer) {
2839 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2844 peer = ofport_dpif_cast(peer_ofport);
2845 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2846 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2848 ofport->peer = peer;
2849 ofport->peer->peer = ofport;
2857 port_run_fast(struct ofport_dpif *ofport)
2859 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2860 struct ofpbuf packet;
2862 ofpbuf_init(&packet, 0);
2863 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2864 send_packet(ofport, &packet);
2865 ofpbuf_uninit(&packet);
2868 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2869 struct ofpbuf packet;
2871 ofpbuf_init(&packet, 0);
2872 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2873 send_packet(ofport, &packet);
2874 ofpbuf_uninit(&packet);
2879 port_run(struct ofport_dpif *ofport)
2881 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2882 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2883 bool enable = netdev_get_carrier(ofport->up.netdev);
2885 ofport->carrier_seq = carrier_seq;
2887 port_run_fast(ofport);
2890 int cfm_opup = cfm_get_opup(ofport->cfm);
2892 cfm_run(ofport->cfm);
2893 enable = enable && !cfm_get_fault(ofport->cfm);
2895 if (cfm_opup >= 0) {
2896 enable = enable && cfm_opup;
2901 bfd_run(ofport->bfd);
2902 enable = enable && bfd_forwarding(ofport->bfd);
2905 if (ofport->bundle) {
2906 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2907 if (carrier_changed) {
2908 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2912 if (ofport->may_enable != enable) {
2913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2914 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2917 ofport->may_enable = enable;
2921 port_wait(struct ofport_dpif *ofport)
2924 cfm_wait(ofport->cfm);
2928 bfd_wait(ofport->bfd);
2933 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2934 struct ofproto_port *ofproto_port)
2936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2937 struct dpif_port dpif_port;
2940 if (sset_contains(&ofproto->ghost_ports, devname)) {
2941 const char *type = netdev_get_type_from_name(devname);
2943 /* We may be called before ofproto->up.port_by_name is populated with
2944 * the appropriate ofport. For this reason, we must get the name and
2945 * type from the netdev layer directly. */
2947 const struct ofport *ofport;
2949 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2950 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2951 ofproto_port->name = xstrdup(devname);
2952 ofproto_port->type = xstrdup(type);
2958 if (!sset_contains(&ofproto->ports, devname)) {
2961 error = dpif_port_query_by_name(ofproto->backer->dpif,
2962 devname, &dpif_port);
2964 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2970 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2973 const char *devname = netdev_get_name(netdev);
2974 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2975 const char *dp_port_name;
2977 if (netdev_vport_is_patch(netdev)) {
2978 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2982 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2983 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2984 odp_port_t port_no = ODPP_NONE;
2987 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2991 if (netdev_get_tunnel_config(netdev)) {
2992 simap_put(&ofproto->backer->tnl_backers,
2993 dp_port_name, odp_to_u32(port_no));
2997 if (netdev_get_tunnel_config(netdev)) {
2998 sset_add(&ofproto->ghost_ports, devname);
3000 sset_add(&ofproto->ports, devname);
3006 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3009 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3016 sset_find_and_delete(&ofproto->ghost_ports,
3017 netdev_get_name(ofport->up.netdev));
3018 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3019 if (!ofport->is_tunnel) {
3020 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3022 /* The caller is going to close ofport->up.netdev. If this is a
3023 * bonded port, then the bond is using that netdev, so remove it
3024 * from the bond. The client will need to reconfigure everything
3025 * after deleting ports, so then the slave will get re-added. */
3026 bundle_remove(&ofport->up);
3033 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3035 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3040 error = netdev_get_stats(ofport->up.netdev, stats);
3042 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3045 /* ofproto->stats.tx_packets represents packets that we created
3046 * internally and sent to some port (e.g. packets sent with
3047 * send_packet()). Account for them as if they had come from
3048 * OFPP_LOCAL and got forwarded. */
3050 if (stats->rx_packets != UINT64_MAX) {
3051 stats->rx_packets += ofproto->stats.tx_packets;
3054 if (stats->rx_bytes != UINT64_MAX) {
3055 stats->rx_bytes += ofproto->stats.tx_bytes;
3058 /* ofproto->stats.rx_packets represents packets that were received on
3059 * some port and we processed internally and dropped (e.g. STP).
3060 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3062 if (stats->tx_packets != UINT64_MAX) {
3063 stats->tx_packets += ofproto->stats.rx_packets;
3066 if (stats->tx_bytes != UINT64_MAX) {
3067 stats->tx_bytes += ofproto->stats.rx_bytes;
3074 struct port_dump_state {
3079 struct ofproto_port port;
3084 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3086 *statep = xzalloc(sizeof(struct port_dump_state));
3091 port_dump_next(const struct ofproto *ofproto_, void *state_,
3092 struct ofproto_port *port)
3094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095 struct port_dump_state *state = state_;
3096 const struct sset *sset;
3097 struct sset_node *node;
3099 if (state->has_port) {
3100 ofproto_port_destroy(&state->port);
3101 state->has_port = false;
3103 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3104 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3107 error = port_query_by_name(ofproto_, node->name, &state->port);
3109 *port = state->port;
3110 state->has_port = true;
3112 } else if (error != ENODEV) {
3117 if (!state->ghost) {
3118 state->ghost = true;
3121 return port_dump_next(ofproto_, state_, port);
3128 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3130 struct port_dump_state *state = state_;
3132 if (state->has_port) {
3133 ofproto_port_destroy(&state->port);
3140 port_poll(const struct ofproto *ofproto_, char **devnamep)
3142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3144 if (ofproto->port_poll_errno) {
3145 int error = ofproto->port_poll_errno;
3146 ofproto->port_poll_errno = 0;
3150 if (sset_is_empty(&ofproto->port_poll_set)) {
3154 *devnamep = sset_pop(&ofproto->port_poll_set);
3159 port_poll_wait(const struct ofproto *ofproto_)
3161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3162 dpif_port_poll_wait(ofproto->backer->dpif);
3166 port_is_lacp_current(const struct ofport *ofport_)
3168 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3169 return (ofport->bundle && ofport->bundle->lacp
3170 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3174 /* Upcall handling. */
3176 /* Flow miss batching.
3178 * Some dpifs implement operations faster when you hand them off in a batch.
3179 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3180 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3181 * more packets, plus possibly installing the flow in the dpif.
3183 * So far we only batch the operations that affect flow setup time the most.
3184 * It's possible to batch more than that, but the benefit might be minimal. */
3186 struct hmap_node hmap_node;
3187 struct ofproto_dpif *ofproto;
3189 enum odp_key_fitness key_fitness;
3190 const struct nlattr *key;
3192 struct list packets;
3193 enum dpif_upcall_type upcall_type;
3196 struct flow_miss_op {
3197 struct dpif_op dpif_op;
3199 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3200 struct xlate_out xout;
3201 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3203 struct ofpbuf mask; /* Flow mask for "put" ops. */
3204 struct odputil_keybuf maskbuf;
3206 /* If this is a "put" op, then a pointer to the subfacet that should
3207 * be marked as uninstalled if the operation fails. */
3208 struct subfacet *subfacet;
3211 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3212 * OpenFlow controller as necessary according to their individual
3213 * configurations. */
3215 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3216 const struct flow *flow)
3218 struct ofputil_packet_in pin;
3220 pin.packet = packet->data;
3221 pin.packet_len = packet->size;
3222 pin.reason = OFPR_NO_MATCH;
3223 pin.controller_id = 0;
3228 pin.send_len = 0; /* not used for flow table misses */
3230 flow_get_metadata(flow, &pin.fmd);
3232 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3235 static struct flow_miss *
3236 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3237 const struct flow *flow, uint32_t hash)
3239 struct flow_miss *miss;
3241 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3242 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3250 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3251 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3252 * 'miss' is associated with a subfacet the caller must also initialize the
3253 * returned op->subfacet, and if anything needs to be freed after processing
3254 * the op, the caller must initialize op->garbage also. */
3256 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3257 struct flow_miss_op *op)
3259 if (miss->flow.in_port.ofp_port
3260 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3261 miss->flow.vlan_tci)) {
3262 /* This packet was received on a VLAN splinter port. We
3263 * added a VLAN to the packet to make the packet resemble
3264 * the flow, but the actions were composed assuming that
3265 * the packet contained no VLAN. So, we must remove the
3266 * VLAN header from the packet before trying to execute the
3268 eth_pop_vlan(packet);
3271 op->subfacet = NULL;
3272 op->xout_garbage = false;
3273 op->dpif_op.type = DPIF_OP_EXECUTE;
3274 op->dpif_op.u.execute.key = miss->key;
3275 op->dpif_op.u.execute.key_len = miss->key_len;
3276 op->dpif_op.u.execute.packet = packet;
3277 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3280 /* Helper for handle_flow_miss_without_facet() and
3281 * handle_flow_miss_with_facet(). */
3283 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3284 const struct flow *flow, bool fail_open)
3288 * Extra-special case for fail-open mode.
3290 * We are in fail-open mode and the packet matched the fail-open
3291 * rule, but we are connected to a controller too. We should send
3292 * the packet up to the controller in the hope that it will try to
3293 * set up a flow and thereby allow us to exit fail-open.
3295 * See the top-level comment in fail-open.c for more information.
3297 send_packet_in_miss(ofproto, packet, flow);
3301 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3302 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3303 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3304 * return value of true). However, for short flows the cost of bookkeeping is
3305 * much higher than the benefits, so when the datapath holds a large number of
3306 * flows we impose some heuristics to decide which flows are likely to be worth
3309 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3311 struct dpif_backer *backer = miss->ofproto->backer;
3314 switch (flow_miss_model) {
3315 case OFPROTO_HANDLE_MISS_AUTO:
3317 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3319 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3323 if (!backer->governor) {
3326 n_subfacets = hmap_count(&backer->subfacets);
3327 if (n_subfacets * 2 <= flow_eviction_threshold) {
3331 backer->governor = governor_create();
3334 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3335 return governor_should_install_flow(backer->governor, hash,
3336 list_size(&miss->packets));
3339 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3340 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3341 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3343 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3344 struct flow_miss *miss,
3345 struct flow_miss_op *ops, size_t *n_ops)
3347 struct ofpbuf *packet;
3349 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3351 COVERAGE_INC(facet_suppress);
3353 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3354 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3357 struct xlate_in xin;
3359 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3360 xlate_actions_for_side_effects(&xin);
3363 if (xout->odp_actions.size) {
3364 struct flow_miss_op *op = &ops[*n_ops];
3365 struct dpif_execute *execute = &op->dpif_op.u.execute;
3367 init_flow_miss_execute_op(miss, packet, op);
3368 xlate_out_copy(&op->xout, xout);
3369 execute->actions = op->xout.odp_actions.data;
3370 execute->actions_len = op->xout.odp_actions.size;
3371 op->xout_garbage = true;
3378 /* Handles 'miss', which matches 'facet'. May add any required datapath
3379 * operations to 'ops', incrementing '*n_ops' for each new op.
3381 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3382 * This is really important only for new facets: if we just called time_msec()
3383 * here, then the new subfacet or its packets could look (occasionally) as
3384 * though it was used some time after the facet was used. That can make a
3385 * one-packet flow look like it has a nonzero duration, which looks odd in
3386 * e.g. NetFlow statistics.
3388 * If non-null, 'stats' will be folded into 'facet'. */
3390 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3391 long long int now, struct dpif_flow_stats *stats,
3392 struct flow_miss_op *ops, size_t *n_ops)
3394 enum subfacet_path want_path;
3395 struct subfacet *subfacet;
3396 struct ofpbuf *packet;
3398 subfacet = subfacet_create(facet, miss, now);
3399 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3401 subfacet_update_stats(subfacet, stats);
3404 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3405 struct flow_miss_op *op = &ops[*n_ops];
3407 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3410 if (want_path != SF_FAST_PATH) {
3411 struct rule_dpif *rule;
3412 struct xlate_in xin;
3414 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3415 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3416 xlate_actions_for_side_effects(&xin);
3419 if (facet->xout.odp_actions.size) {
3420 struct dpif_execute *execute = &op->dpif_op.u.execute;
3422 init_flow_miss_execute_op(miss, packet, op);
3423 execute->actions = facet->xout.odp_actions.data,
3424 execute->actions_len = facet->xout.odp_actions.size;
3429 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3430 struct flow_miss_op *op = &ops[(*n_ops)++];
3431 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3433 subfacet->path = want_path;
3435 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3436 if (enable_megaflows) {
3437 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3438 &miss->flow, UINT32_MAX);
3441 op->xout_garbage = false;
3442 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3443 op->subfacet = subfacet;
3444 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3445 put->key = miss->key;
3446 put->key_len = miss->key_len;
3447 put->mask = op->mask.data;
3448 put->mask_len = op->mask.size;
3450 if (want_path == SF_FAST_PATH) {
3451 put->actions = facet->xout.odp_actions.data;
3452 put->actions_len = facet->xout.odp_actions.size;
3454 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3455 op->slow_stub, sizeof op->slow_stub,
3456 &put->actions, &put->actions_len);
3462 /* Handles flow miss 'miss'. May add any required datapath operations
3463 * to 'ops', incrementing '*n_ops' for each new op. */
3465 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3468 struct ofproto_dpif *ofproto = miss->ofproto;
3469 struct dpif_flow_stats stats__;
3470 struct dpif_flow_stats *stats = &stats__;
3471 struct ofpbuf *packet;
3472 struct facet *facet;
3476 memset(stats, 0, sizeof *stats);
3478 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3479 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3480 stats->n_bytes += packet->size;
3484 facet = facet_lookup_valid(ofproto, &miss->flow);
3486 struct flow_wildcards wc;
3487 struct rule_dpif *rule;
3488 struct xlate_out xout;
3489 struct xlate_in xin;
3491 flow_wildcards_init_catchall(&wc);
3492 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3493 rule_credit_stats(rule, stats);
3495 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3497 xin.resubmit_stats = stats;
3498 xin.may_learn = true;
3499 xlate_actions(&xin, &xout);
3500 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3502 /* There does not exist a bijection between 'struct flow' and datapath
3503 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3504 * assumption used throughout the facet and subfacet handling code.
3505 * Since we have to handle these misses in userspace anyway, we simply
3506 * skip facet creation, avoiding the problem altogether. */
3507 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3508 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3509 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3513 facet = facet_create(miss, rule, &xout, stats);
3516 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3519 static struct drop_key *
3520 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3523 struct drop_key *drop_key;
3525 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3526 &backer->drop_keys) {
3527 if (drop_key->key_len == key_len
3528 && !memcmp(drop_key->key, key, key_len)) {
3536 drop_key_clear(struct dpif_backer *backer)
3538 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3539 struct drop_key *drop_key, *next;
3541 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3544 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3546 if (error && !VLOG_DROP_WARN(&rl)) {
3547 struct ds ds = DS_EMPTY_INITIALIZER;
3548 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3549 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3550 ovs_strerror(error), ds_cstr(&ds));
3554 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3555 free(drop_key->key);
3560 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3561 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3562 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3563 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3564 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3565 * 'packet' ingressed.
3567 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3568 * 'flow''s in_port to OFPP_NONE.
3570 * This function does post-processing on data returned from
3571 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3572 * of the upcall processing logic. In particular, if the extracted in_port is
3573 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3574 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3575 * a VLAN header onto 'packet' (if it is nonnull).
3577 * Similarly, this function also includes some logic to help with tunnels. It
3578 * may modify 'flow' as necessary to make the tunneling implementation
3579 * transparent to the upcall processing logic.
3581 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3582 * or some other positive errno if there are other problems. */
3584 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3585 const struct nlattr *key, size_t key_len,
3586 struct flow *flow, enum odp_key_fitness *fitnessp,
3587 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3589 const struct ofport_dpif *port;
3590 enum odp_key_fitness fitness;
3593 fitness = odp_flow_key_to_flow(key, key_len, flow);
3594 if (fitness == ODP_FIT_ERROR) {
3600 *odp_in_port = flow->in_port.odp_port;
3603 port = (tnl_port_should_receive(flow)
3604 ? tnl_port_receive(flow)
3605 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3606 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3611 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3612 * it's theoretically possible that we'll receive an ofport belonging to an
3613 * entirely different datapath. In practice, this can't happen because no
3614 * platforms has two separate datapaths which each support tunneling. */
3615 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3617 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3619 /* Make the packet resemble the flow, so that it gets sent to
3620 * an OpenFlow controller properly, so that it looks correct
3621 * for sFlow, and so that flow_extract() will get the correct
3622 * vlan_tci if it is called on 'packet'.
3624 * The allocated space inside 'packet' probably also contains
3625 * 'key', that is, both 'packet' and 'key' are probably part of
3626 * a struct dpif_upcall (see the large comment on that
3627 * structure definition), so pushing data on 'packet' is in
3628 * general not a good idea since it could overwrite 'key' or
3629 * free it as a side effect. However, it's OK in this special
3630 * case because we know that 'packet' is inside a Netlink
3631 * attribute: pushing 4 bytes will just overwrite the 4-byte
3632 * "struct nlattr", which is fine since we don't need that
3633 * header anymore. */
3634 eth_push_vlan(packet, flow->vlan_tci);
3636 /* We can't reproduce 'key' from 'flow'. */
3637 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3642 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3647 *fitnessp = fitness;
3653 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3656 struct dpif_upcall *upcall;
3657 struct flow_miss *miss;
3658 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3659 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3660 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3670 /* Construct the to-do list.
3672 * This just amounts to extracting the flow from each packet and sticking
3673 * the packets that have the same flow in the same "flow_miss" structure so
3674 * that we can process them together. */
3677 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3678 struct flow_miss *miss = &misses[n_misses];
3679 struct flow_miss *existing_miss;
3680 struct ofproto_dpif *ofproto;
3681 odp_port_t odp_in_port;
3686 error = ofproto_receive(backer, upcall->packet, upcall->key,
3687 upcall->key_len, &flow, &miss->key_fitness,
3688 &ofproto, &odp_in_port);
3689 if (error == ENODEV) {
3690 struct drop_key *drop_key;
3692 /* Received packet on datapath port for which we couldn't
3693 * associate an ofproto. This can happen if a port is removed
3694 * while traffic is being received. Print a rate-limited message
3695 * in case it happens frequently. Install a drop flow so
3696 * that future packets of the flow are inexpensively dropped
3698 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3699 "%"PRIu32, odp_in_port);
3701 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3703 drop_key = xmalloc(sizeof *drop_key);
3704 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3705 drop_key->key_len = upcall->key_len;
3707 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3708 hash_bytes(drop_key->key, drop_key->key_len, 0));
3709 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3710 drop_key->key, drop_key->key_len,
3711 NULL, 0, NULL, 0, NULL);
3719 ofproto->n_missed++;
3720 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3721 &flow.tunnel, &flow.in_port, &miss->flow);
3723 /* Add other packets to a to-do list. */
3724 hash = flow_hash(&miss->flow, 0);
3725 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3726 if (!existing_miss) {
3727 hmap_insert(&todo, &miss->hmap_node, hash);
3728 miss->ofproto = ofproto;
3729 miss->key = upcall->key;
3730 miss->key_len = upcall->key_len;
3731 miss->upcall_type = upcall->type;
3732 list_init(&miss->packets);
3736 miss = existing_miss;
3738 list_push_back(&miss->packets, &upcall->packet->list_node);
3741 /* Process each element in the to-do list, constructing the set of
3742 * operations to batch. */
3744 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3745 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3747 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3749 /* Execute batch. */
3750 for (i = 0; i < n_ops; i++) {
3751 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3753 dpif_operate(backer->dpif, dpif_ops, n_ops);
3755 for (i = 0; i < n_ops; i++) {
3756 if (dpif_ops[i]->error != 0
3757 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3758 && flow_miss_ops[i].subfacet) {
3759 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3761 COVERAGE_INC(subfacet_install_fail);
3763 subfacet->path = SF_NOT_INSTALLED;
3767 if (flow_miss_ops[i].xout_garbage) {
3768 xlate_out_uninit(&flow_miss_ops[i].xout);
3771 hmap_destroy(&todo);
3774 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3776 classify_upcall(const struct dpif_upcall *upcall)
3778 size_t userdata_len;
3779 union user_action_cookie cookie;
3781 /* First look at the upcall type. */
3782 switch (upcall->type) {
3783 case DPIF_UC_ACTION:
3789 case DPIF_N_UC_TYPES:
3791 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3795 /* "action" upcalls need a closer look. */
3796 if (!upcall->userdata) {
3797 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3800 userdata_len = nl_attr_get_size(upcall->userdata);
3801 if (userdata_len < sizeof cookie.type
3802 || userdata_len > sizeof cookie) {
3803 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3807 memset(&cookie, 0, sizeof cookie);
3808 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3809 if (userdata_len == sizeof cookie.sflow
3810 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3811 return SFLOW_UPCALL;
3812 } else if (userdata_len == sizeof cookie.slow_path
3813 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3815 } else if (userdata_len == sizeof cookie.flow_sample
3816 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3817 return FLOW_SAMPLE_UPCALL;
3818 } else if (userdata_len == sizeof cookie.ipfix
3819 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3820 return IPFIX_UPCALL;
3822 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3823 " and size %zu", cookie.type, userdata_len);
3829 handle_sflow_upcall(struct dpif_backer *backer,
3830 const struct dpif_upcall *upcall)
3832 struct ofproto_dpif *ofproto;
3833 union user_action_cookie cookie;
3835 odp_port_t odp_in_port;
3837 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3838 &flow, NULL, &ofproto, &odp_in_port)
3839 || !ofproto->sflow) {
3843 memset(&cookie, 0, sizeof cookie);
3844 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3845 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3846 odp_in_port, &cookie);
3850 handle_flow_sample_upcall(struct dpif_backer *backer,
3851 const struct dpif_upcall *upcall)
3853 struct ofproto_dpif *ofproto;
3854 union user_action_cookie cookie;
3857 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3858 &flow, NULL, &ofproto, NULL)
3859 || !ofproto->ipfix) {
3863 memset(&cookie, 0, sizeof cookie);
3864 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3866 /* The flow reflects exactly the contents of the packet. Sample
3867 * the packet using it. */
3868 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3869 cookie.flow_sample.collector_set_id,
3870 cookie.flow_sample.probability,
3871 cookie.flow_sample.obs_domain_id,
3872 cookie.flow_sample.obs_point_id);
3876 handle_ipfix_upcall(struct dpif_backer *backer,
3877 const struct dpif_upcall *upcall)
3879 struct ofproto_dpif *ofproto;
3882 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3883 &flow, NULL, &ofproto, NULL)
3884 || !ofproto->ipfix) {
3888 /* The flow reflects exactly the contents of the packet. Sample
3889 * the packet using it. */
3890 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3894 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3896 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3897 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3898 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3903 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3906 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3907 struct dpif_upcall *upcall = &misses[n_misses];
3908 struct ofpbuf *buf = &miss_bufs[n_misses];
3911 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3912 sizeof miss_buf_stubs[n_misses]);
3913 error = dpif_recv(backer->dpif, upcall, buf);
3919 switch (classify_upcall(upcall)) {
3921 /* Handle it later. */
3926 handle_sflow_upcall(backer, upcall);
3930 case FLOW_SAMPLE_UPCALL:
3931 handle_flow_sample_upcall(backer, upcall);
3936 handle_ipfix_upcall(backer, upcall);
3946 /* Handle deferred MISS_UPCALL processing. */
3947 handle_miss_upcalls(backer, misses, n_misses);
3948 for (i = 0; i < n_misses; i++) {
3949 ofpbuf_uninit(&miss_bufs[i]);
3955 /* Flow expiration. */
3957 static int subfacet_max_idle(const struct dpif_backer *);
3958 static void update_stats(struct dpif_backer *);
3959 static void rule_expire(struct rule_dpif *);
3960 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3962 /* This function is called periodically by run(). Its job is to collect
3963 * updates for the flows that have been installed into the datapath, most
3964 * importantly when they last were used, and then use that information to
3965 * expire flows that have not been used recently.
3967 * Returns the number of milliseconds after which it should be called again. */
3969 expire(struct dpif_backer *backer)
3971 struct ofproto_dpif *ofproto;
3975 /* Periodically clear out the drop keys in an effort to keep them
3976 * relatively few. */
3977 drop_key_clear(backer);
3979 /* Update stats for each flow in the backer. */
3980 update_stats(backer);
3982 n_subfacets = hmap_count(&backer->subfacets);
3984 struct subfacet *subfacet;
3985 long long int total, now;
3989 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3990 total += now - subfacet->created;
3992 backer->avg_subfacet_life += total / n_subfacets;
3994 backer->avg_subfacet_life /= 2;
3996 backer->avg_n_subfacet += n_subfacets;
3997 backer->avg_n_subfacet /= 2;
3999 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4001 max_idle = subfacet_max_idle(backer);
4002 expire_subfacets(backer, max_idle);
4004 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4005 struct rule *rule, *next_rule;
4007 if (ofproto->backer != backer) {
4011 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4013 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4014 &ofproto->up.expirable) {
4015 rule_expire(rule_dpif_cast(rule));
4018 /* All outstanding data in existing flows has been accounted, so it's a
4019 * good time to do bond rebalancing. */
4020 if (ofproto->has_bonded_bundles) {
4021 struct ofbundle *bundle;
4023 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4025 bond_rebalance(bundle->bond, &backer->revalidate_set);
4031 return MIN(max_idle, 1000);
4034 /* Updates flow table statistics given that the datapath just reported 'stats'
4035 * as 'subfacet''s statistics. */
4037 update_subfacet_stats(struct subfacet *subfacet,
4038 const struct dpif_flow_stats *stats)
4040 struct facet *facet = subfacet->facet;
4041 struct dpif_flow_stats diff;
4043 diff.tcp_flags = stats->tcp_flags;
4044 diff.used = stats->used;
4046 if (stats->n_packets >= subfacet->dp_packet_count) {
4047 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4049 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4053 if (stats->n_bytes >= subfacet->dp_byte_count) {
4054 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4056 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4060 facet->ofproto->n_hit += diff.n_packets;
4061 subfacet->dp_packet_count = stats->n_packets;
4062 subfacet->dp_byte_count = stats->n_bytes;
4063 subfacet_update_stats(subfacet, &diff);
4065 if (facet->accounted_bytes < facet->byte_count) {
4067 facet_account(facet);
4068 facet->accounted_bytes = facet->byte_count;
4072 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4073 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4075 delete_unexpected_flow(struct dpif_backer *backer,
4076 const struct nlattr *key, size_t key_len)
4078 if (!VLOG_DROP_WARN(&rl)) {
4082 odp_flow_key_format(key, key_len, &s);
4083 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4087 COVERAGE_INC(facet_unexpected);
4088 dpif_flow_del(backer->dpif, key, key_len, NULL);
4091 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4093 * This function also pushes statistics updates to rules which each facet
4094 * resubmits into. Generally these statistics will be accurate. However, if a
4095 * facet changes the rule it resubmits into at some time in between
4096 * update_stats() runs, it is possible that statistics accrued to the
4097 * old rule will be incorrectly attributed to the new rule. This could be
4098 * avoided by calling update_stats() whenever rules are created or
4099 * deleted. However, the performance impact of making so many calls to the
4100 * datapath do not justify the benefit of having perfectly accurate statistics.
4102 * In addition, this function maintains per ofproto flow hit counts. The patch
4103 * port is not treated specially. e.g. A packet ingress from br0 patched into
4104 * br1 will increase the hit count of br0 by 1, however, does not affect
4105 * the hit or miss counts of br1.
4108 update_stats(struct dpif_backer *backer)
4110 const struct dpif_flow_stats *stats;
4111 struct dpif_flow_dump dump;
4112 const struct nlattr *key, *mask;
4113 size_t key_len, mask_len;
4115 dpif_flow_dump_start(&dump, backer->dpif);
4116 while (dpif_flow_dump_next(&dump, &key, &key_len,
4117 &mask, &mask_len, NULL, NULL, &stats)) {
4118 struct subfacet *subfacet;
4121 key_hash = odp_flow_key_hash(key, key_len);
4122 subfacet = subfacet_find(backer, key, key_len, key_hash);
4123 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4125 update_subfacet_stats(subfacet, stats);
4129 /* Stats are updated per-packet. */
4132 case SF_NOT_INSTALLED:
4134 delete_unexpected_flow(backer, key, key_len);
4139 dpif_flow_dump_done(&dump);
4141 update_moving_averages(backer);
4144 /* Calculates and returns the number of milliseconds of idle time after which
4145 * subfacets should expire from the datapath. When a subfacet expires, we fold
4146 * its statistics into its facet, and when a facet's last subfacet expires, we
4147 * fold its statistic into its rule. */
4149 subfacet_max_idle(const struct dpif_backer *backer)
4152 * Idle time histogram.
4154 * Most of the time a switch has a relatively small number of subfacets.
4155 * When this is the case we might as well keep statistics for all of them
4156 * in userspace and to cache them in the kernel datapath for performance as
4159 * As the number of subfacets increases, the memory required to maintain
4160 * statistics about them in userspace and in the kernel becomes
4161 * significant. However, with a large number of subfacets it is likely
4162 * that only a few of them are "heavy hitters" that consume a large amount
4163 * of bandwidth. At this point, only heavy hitters are worth caching in
4164 * the kernel and maintaining in userspaces; other subfacets we can
4167 * The technique used to compute the idle time is to build a histogram with
4168 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4169 * that is installed in the kernel gets dropped in the appropriate bucket.
4170 * After the histogram has been built, we compute the cutoff so that only
4171 * the most-recently-used 1% of subfacets (but at least
4172 * flow_eviction_threshold flows) are kept cached. At least
4173 * the most-recently-used bucket of subfacets is kept, so actually an
4174 * arbitrary number of subfacets can be kept in any given expiration run
4175 * (though the next run will delete most of those unless they receive
4178 * This requires a second pass through the subfacets, in addition to the
4179 * pass made by update_stats(), because the former function never looks at
4180 * uninstallable subfacets.
4182 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4183 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4184 int buckets[N_BUCKETS] = { 0 };
4185 int total, subtotal, bucket;
4186 struct subfacet *subfacet;
4190 total = hmap_count(&backer->subfacets);
4191 if (total <= flow_eviction_threshold) {
4192 return N_BUCKETS * BUCKET_WIDTH;
4195 /* Build histogram. */
4197 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4198 long long int idle = now - subfacet->used;
4199 int bucket = (idle <= 0 ? 0
4200 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4201 : (unsigned int) idle / BUCKET_WIDTH);
4205 /* Find the first bucket whose flows should be expired. */
4206 subtotal = bucket = 0;
4208 subtotal += buckets[bucket++];
4209 } while (bucket < N_BUCKETS &&
4210 subtotal < MAX(flow_eviction_threshold, total / 100));
4212 if (VLOG_IS_DBG_ENABLED()) {
4216 ds_put_cstr(&s, "keep");
4217 for (i = 0; i < N_BUCKETS; i++) {
4219 ds_put_cstr(&s, ", drop");
4222 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4225 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4229 return bucket * BUCKET_WIDTH;
4233 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4235 /* Cutoff time for most flows. */
4236 long long int normal_cutoff = time_msec() - dp_max_idle;
4238 /* We really want to keep flows for special protocols around, so use a more
4239 * conservative cutoff. */
4240 long long int special_cutoff = time_msec() - 10000;
4242 struct subfacet *subfacet, *next_subfacet;
4243 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4247 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4248 &backer->subfacets) {
4249 long long int cutoff;
4251 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4255 if (subfacet->used < cutoff) {
4256 if (subfacet->path != SF_NOT_INSTALLED) {
4257 batch[n_batch++] = subfacet;
4258 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4259 subfacet_destroy_batch(backer, batch, n_batch);
4263 subfacet_destroy(subfacet);
4269 subfacet_destroy_batch(backer, batch, n_batch);
4273 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4274 * then delete it entirely. */
4276 rule_expire(struct rule_dpif *rule)
4281 if (rule->up.pending) {
4282 /* We'll have to expire it later. */
4286 /* Has 'rule' expired? */
4288 if (rule->up.hard_timeout
4289 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4290 reason = OFPRR_HARD_TIMEOUT;
4291 } else if (rule->up.idle_timeout
4292 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4293 reason = OFPRR_IDLE_TIMEOUT;
4298 COVERAGE_INC(ofproto_dpif_expired);
4300 /* Get rid of the rule. */
4301 ofproto_rule_expire(&rule->up, reason);
4306 /* Creates and returns a new facet based on 'miss'.
4308 * The caller must already have determined that no facet with an identical
4309 * 'miss->flow' exists in 'miss->ofproto'.
4311 * 'rule' and 'xout' must have been created based on 'miss'.
4313 * 'facet'' statistics are initialized based on 'stats'.
4315 * The facet will initially have no subfacets. The caller should create (at
4316 * least) one subfacet with subfacet_create(). */
4317 static struct facet *
4318 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4319 struct xlate_out *xout, struct dpif_flow_stats *stats)
4321 struct ofproto_dpif *ofproto = miss->ofproto;
4322 struct facet *facet;
4325 facet = xzalloc(sizeof *facet);
4326 facet->ofproto = miss->ofproto;
4327 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4328 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4329 facet->tcp_flags = stats->tcp_flags;
4330 facet->used = stats->used;
4331 facet->flow = miss->flow;
4332 facet->learn_rl = time_msec() + 500;
4334 list_init(&facet->subfacets);
4335 netflow_flow_init(&facet->nf_flow);
4336 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4338 xlate_out_copy(&facet->xout, xout);
4340 match_init(&match, &facet->flow, &facet->xout.wc);
4341 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4342 classifier_insert(&ofproto->facets, &facet->cr);
4344 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4345 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4351 facet_free(struct facet *facet)
4354 xlate_out_uninit(&facet->xout);
4359 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4360 * 'packet', which arrived on 'in_port'. */
4362 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4363 const struct nlattr *odp_actions, size_t actions_len,
4364 struct ofpbuf *packet)
4366 struct odputil_keybuf keybuf;
4370 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4371 odp_flow_key_from_flow(&key, flow,
4372 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4374 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4375 odp_actions, actions_len, packet);
4379 /* Remove 'facet' from its ofproto and free up the associated memory:
4381 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4382 * rule's statistics, via subfacet_uninstall().
4384 * - Removes 'facet' from its rule and from ofproto->facets.
4387 facet_remove(struct facet *facet)
4389 struct subfacet *subfacet, *next_subfacet;
4391 ovs_assert(!list_is_empty(&facet->subfacets));
4393 /* First uninstall all of the subfacets to get final statistics. */
4394 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4395 subfacet_uninstall(subfacet);
4398 /* Flush the final stats to the rule.
4400 * This might require us to have at least one subfacet around so that we
4401 * can use its actions for accounting in facet_account(), which is why we
4402 * have uninstalled but not yet destroyed the subfacets. */
4403 facet_flush_stats(facet);
4405 /* Now we're really all done so destroy everything. */
4406 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4407 &facet->subfacets) {
4408 subfacet_destroy__(subfacet);
4410 classifier_remove(&facet->ofproto->facets, &facet->cr);
4411 cls_rule_destroy(&facet->cr);
4415 /* Feed information from 'facet' back into the learning table to keep it in
4416 * sync with what is actually flowing through the datapath. */
4418 facet_learn(struct facet *facet)
4420 long long int now = time_msec();
4422 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4426 facet->learn_rl = now + 500;
4428 if (!facet->xout.has_learn
4429 && !facet->xout.has_normal
4430 && (!facet->xout.has_fin_timeout
4431 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4435 facet_push_stats(facet, true);
4439 facet_account(struct facet *facet)
4441 const struct nlattr *a;
4446 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4449 n_bytes = facet->byte_count - facet->accounted_bytes;
4451 /* This loop feeds byte counters to bond_account() for rebalancing to use
4452 * as a basis. We also need to track the actual VLAN on which the packet
4453 * is going to be sent to ensure that it matches the one passed to
4454 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4457 * We use the actions from an arbitrary subfacet because they should all
4458 * be equally valid for our purpose. */
4459 vlan_tci = facet->flow.vlan_tci;
4460 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4461 facet->xout.odp_actions.size) {
4462 const struct ovs_action_push_vlan *vlan;
4463 struct ofport_dpif *port;
4465 switch (nl_attr_type(a)) {
4466 case OVS_ACTION_ATTR_OUTPUT:
4467 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4468 if (port && port->bundle && port->bundle->bond) {
4469 bond_account(port->bundle->bond, &facet->flow,
4470 vlan_tci_to_vid(vlan_tci), n_bytes);
4474 case OVS_ACTION_ATTR_POP_VLAN:
4475 vlan_tci = htons(0);
4478 case OVS_ACTION_ATTR_PUSH_VLAN:
4479 vlan = nl_attr_get(a);
4480 vlan_tci = vlan->vlan_tci;
4486 /* Returns true if the only action for 'facet' is to send to the controller.
4487 * (We don't report NetFlow expiration messages for such facets because they
4488 * are just part of the control logic for the network, not real traffic). */
4490 facet_is_controller_flow(struct facet *facet)
4493 struct ofproto_dpif *ofproto = facet->ofproto;
4494 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4496 const struct ofpact *ofpacts = rule->up.ofpacts;
4497 size_t ofpacts_len = rule->up.ofpacts_len;
4499 if (ofpacts_len > 0 &&
4500 ofpacts->type == OFPACT_CONTROLLER &&
4501 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4508 /* Folds all of 'facet''s statistics into its rule. Also updates the
4509 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4510 * 'facet''s statistics in the datapath should have been zeroed and folded into
4511 * its packet and byte counts before this function is called. */
4513 facet_flush_stats(struct facet *facet)
4515 struct ofproto_dpif *ofproto = facet->ofproto;
4516 struct subfacet *subfacet;
4518 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4519 ovs_assert(!subfacet->dp_byte_count);
4520 ovs_assert(!subfacet->dp_packet_count);
4523 facet_push_stats(facet, false);
4524 if (facet->accounted_bytes < facet->byte_count) {
4525 facet_account(facet);
4526 facet->accounted_bytes = facet->byte_count;
4529 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4530 struct ofexpired expired;
4531 expired.flow = facet->flow;
4532 expired.packet_count = facet->packet_count;
4533 expired.byte_count = facet->byte_count;
4534 expired.used = facet->used;
4535 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4538 /* Reset counters to prevent double counting if 'facet' ever gets
4540 facet_reset_counters(facet);
4542 netflow_flow_clear(&facet->nf_flow);
4543 facet->tcp_flags = 0;
4546 /* Searches 'ofproto''s table of facets for one which would be responsible for
4547 * 'flow'. Returns it if found, otherwise a null pointer.
4549 * The returned facet might need revalidation; use facet_lookup_valid()
4550 * instead if that is important. */
4551 static struct facet *
4552 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4554 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4555 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4558 /* Searches 'ofproto''s table of facets for one capable that covers
4559 * 'flow'. Returns it if found, otherwise a null pointer.
4561 * The returned facet is guaranteed to be valid. */
4562 static struct facet *
4563 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4565 struct facet *facet;
4567 facet = facet_find(ofproto, flow);
4569 && (ofproto->backer->need_revalidate
4570 || tag_set_intersects(&ofproto->backer->revalidate_set,
4572 && !facet_revalidate(facet)) {
4580 facet_check_consistency(struct facet *facet)
4582 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4584 struct xlate_out xout;
4585 struct xlate_in xin;
4587 struct rule_dpif *rule;
4590 /* Check the datapath actions for consistency. */
4591 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4592 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4593 xlate_actions(&xin, &xout);
4595 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4596 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4597 && facet->xout.slow == xout.slow
4598 && facet->fail_open == fail_open;
4599 if (!ok && !VLOG_DROP_WARN(&rl)) {
4600 struct ds s = DS_EMPTY_INITIALIZER;
4602 flow_format(&s, &facet->flow);
4603 ds_put_cstr(&s, ": inconsistency in facet");
4605 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4606 ds_put_cstr(&s, " (actions were: ");
4607 format_odp_actions(&s, facet->xout.odp_actions.data,
4608 facet->xout.odp_actions.size);
4609 ds_put_cstr(&s, ") (correct actions: ");
4610 format_odp_actions(&s, xout.odp_actions.data,
4611 xout.odp_actions.size);
4612 ds_put_char(&s, ')');
4615 if (facet->xout.slow != xout.slow) {
4616 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4619 if (facet->fail_open != fail_open) {
4620 ds_put_format(&s, " fail open incorrect. should be %s",
4621 fail_open ? "true" : "false");
4625 xlate_out_uninit(&xout);
4630 /* Re-searches the classifier for 'facet':
4632 * - If the rule found is different from 'facet''s current rule, moves
4633 * 'facet' to the new rule and recompiles its actions.
4635 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4636 * where it is and recompiles its actions anyway.
4638 * - If any of 'facet''s subfacets correspond to a new flow according to
4639 * ofproto_receive(), 'facet' is removed.
4641 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4643 facet_revalidate(struct facet *facet)
4645 struct ofproto_dpif *ofproto = facet->ofproto;
4646 struct rule_dpif *new_rule;
4647 struct subfacet *subfacet;
4648 struct flow_wildcards wc;
4649 struct xlate_out xout;
4650 struct xlate_in xin;
4652 COVERAGE_INC(facet_revalidate);
4654 /* Check that child subfacets still correspond to this facet. Tunnel
4655 * configuration changes could cause a subfacet's OpenFlow in_port to
4657 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4658 struct ofproto_dpif *recv_ofproto;
4659 struct flow recv_flow;
4662 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4663 subfacet->key_len, &recv_flow, NULL,
4664 &recv_ofproto, NULL);
4666 || recv_ofproto != ofproto
4667 || facet != facet_find(ofproto, &recv_flow)) {
4668 facet_remove(facet);
4673 flow_wildcards_init_catchall(&wc);
4674 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4676 /* Calculate new datapath actions.
4678 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4679 * emit a NetFlow expiration and, if so, we need to have the old state
4680 * around to properly compose it. */
4681 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4682 xlate_actions(&xin, &xout);
4683 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4685 /* A facet's slow path reason should only change under dramatic
4686 * circumstances. Rather than try to update everything, it's simpler to
4687 * remove the facet and start over.
4689 * More importantly, if a facet's wildcards change, it will be relatively
4690 * difficult to figure out if its subfacets still belong to it, and if not
4691 * which facet they may belong to. Again, to avoid the complexity, we
4692 * simply give up instead. */
4693 if (facet->xout.slow != xout.slow
4694 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4695 facet_remove(facet);
4696 xlate_out_uninit(&xout);
4700 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4701 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4702 if (subfacet->path == SF_FAST_PATH) {
4703 struct dpif_flow_stats stats;
4705 subfacet_install(subfacet, &xout.odp_actions, &stats);
4706 subfacet_update_stats(subfacet, &stats);
4710 facet_flush_stats(facet);
4712 ofpbuf_clear(&facet->xout.odp_actions);
4713 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4714 xout.odp_actions.size);
4717 /* Update 'facet' now that we've taken care of all the old state. */
4718 facet->xout.tags = xout.tags;
4719 facet->xout.slow = xout.slow;
4720 facet->xout.has_learn = xout.has_learn;
4721 facet->xout.has_normal = xout.has_normal;
4722 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4723 facet->xout.nf_output_iface = xout.nf_output_iface;
4724 facet->xout.mirrors = xout.mirrors;
4725 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4726 facet->used = MAX(facet->used, new_rule->up.created);
4727 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4729 xlate_out_uninit(&xout);
4734 facet_reset_counters(struct facet *facet)
4736 facet->packet_count = 0;
4737 facet->byte_count = 0;
4738 facet->prev_packet_count = 0;
4739 facet->prev_byte_count = 0;
4740 facet->accounted_bytes = 0;
4744 facet_push_stats(struct facet *facet, bool may_learn)
4746 struct dpif_flow_stats stats;
4748 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4749 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4750 ovs_assert(facet->used >= facet->prev_used);
4752 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4753 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4754 stats.used = facet->used;
4755 stats.tcp_flags = facet->tcp_flags;
4757 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4758 struct ofproto_dpif *ofproto = facet->ofproto;
4759 struct ofport_dpif *in_port;
4760 struct rule_dpif *rule;
4761 struct xlate_in xin;
4763 facet->prev_packet_count = facet->packet_count;
4764 facet->prev_byte_count = facet->byte_count;
4765 facet->prev_used = facet->used;
4767 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4768 if (in_port && in_port->is_tunnel) {
4769 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4772 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4773 rule_credit_stats(rule, &stats);
4774 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4776 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4777 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4778 stats.n_packets, stats.n_bytes);
4780 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4782 xin.resubmit_stats = &stats;
4783 xin.may_learn = may_learn;
4784 xlate_actions_for_side_effects(&xin);
4789 push_all_stats__(bool run_fast)
4791 static long long int rl = LLONG_MIN;
4792 struct ofproto_dpif *ofproto;
4794 if (time_msec() < rl) {
4798 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4799 struct cls_cursor cursor;
4800 struct facet *facet;
4802 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4803 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4804 facet_push_stats(facet, false);
4811 rl = time_msec() + 100;
4815 push_all_stats(void)
4817 push_all_stats__(true);
4821 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4823 rule->packet_count += stats->n_packets;
4824 rule->byte_count += stats->n_bytes;
4825 ofproto_rule_update_used(&rule->up, stats->used);
4830 static struct subfacet *
4831 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4832 size_t key_len, uint32_t key_hash)
4834 struct subfacet *subfacet;
4836 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4837 &backer->subfacets) {
4838 if (subfacet->key_len == key_len
4839 && !memcmp(key, subfacet->key, key_len)) {
4847 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4848 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4849 * existing subfacet if there is one, otherwise creates and returns a
4851 static struct subfacet *
4852 subfacet_create(struct facet *facet, struct flow_miss *miss,
4855 struct dpif_backer *backer = miss->ofproto->backer;
4856 enum odp_key_fitness key_fitness = miss->key_fitness;
4857 const struct nlattr *key = miss->key;
4858 size_t key_len = miss->key_len;
4860 struct subfacet *subfacet;
4862 key_hash = odp_flow_key_hash(key, key_len);
4864 if (list_is_empty(&facet->subfacets)) {
4865 subfacet = &facet->one_subfacet;
4867 subfacet = subfacet_find(backer, key, key_len, key_hash);
4869 if (subfacet->facet == facet) {
4873 /* This shouldn't happen. */
4874 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4875 subfacet_destroy(subfacet);
4878 subfacet = xmalloc(sizeof *subfacet);
4881 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4882 list_push_back(&facet->subfacets, &subfacet->list_node);
4883 subfacet->facet = facet;
4884 subfacet->key_fitness = key_fitness;
4885 subfacet->key = xmemdup(key, key_len);
4886 subfacet->key_len = key_len;
4887 subfacet->used = now;
4888 subfacet->created = now;
4889 subfacet->dp_packet_count = 0;
4890 subfacet->dp_byte_count = 0;
4891 subfacet->path = SF_NOT_INSTALLED;
4892 subfacet->backer = backer;
4894 backer->subfacet_add_count++;
4898 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4899 * its facet within 'ofproto', and frees it. */
4901 subfacet_destroy__(struct subfacet *subfacet)
4903 struct facet *facet = subfacet->facet;
4904 struct ofproto_dpif *ofproto = facet->ofproto;
4906 /* Update ofproto stats before uninstall the subfacet. */
4907 ofproto->backer->subfacet_del_count++;
4909 subfacet_uninstall(subfacet);
4910 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4911 list_remove(&subfacet->list_node);
4912 free(subfacet->key);
4913 if (subfacet != &facet->one_subfacet) {
4918 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4919 * last remaining subfacet in its facet destroys the facet too. */
4921 subfacet_destroy(struct subfacet *subfacet)
4923 struct facet *facet = subfacet->facet;
4925 if (list_is_singleton(&facet->subfacets)) {
4926 /* facet_remove() needs at least one subfacet (it will remove it). */
4927 facet_remove(facet);
4929 subfacet_destroy__(subfacet);
4934 subfacet_destroy_batch(struct dpif_backer *backer,
4935 struct subfacet **subfacets, int n)
4937 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4938 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4939 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4942 for (i = 0; i < n; i++) {
4943 ops[i].type = DPIF_OP_FLOW_DEL;
4944 ops[i].u.flow_del.key = subfacets[i]->key;
4945 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4946 ops[i].u.flow_del.stats = &stats[i];
4950 dpif_operate(backer->dpif, opsp, n);
4951 for (i = 0; i < n; i++) {
4952 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4953 subfacets[i]->path = SF_NOT_INSTALLED;
4954 subfacet_destroy(subfacets[i]);
4959 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4960 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4961 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4962 * since 'subfacet' was last updated.
4964 * Returns 0 if successful, otherwise a positive errno value. */
4966 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4967 struct dpif_flow_stats *stats)
4969 struct facet *facet = subfacet->facet;
4970 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4971 const struct nlattr *actions = odp_actions->data;
4972 size_t actions_len = odp_actions->size;
4973 struct odputil_keybuf maskbuf;
4976 uint64_t slow_path_stub[128 / 8];
4977 enum dpif_flow_put_flags flags;
4980 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4982 flags |= DPIF_FP_ZERO_STATS;
4985 if (path == SF_SLOW_PATH) {
4986 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4987 slow_path_stub, sizeof slow_path_stub,
4988 &actions, &actions_len);
4991 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4992 if (enable_megaflows) {
4993 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4994 &facet->flow, UINT32_MAX);
4997 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4998 subfacet->key_len, mask.data, mask.size,
4999 actions, actions_len, stats);
5002 subfacet_reset_dp_stats(subfacet, stats);
5006 COVERAGE_INC(subfacet_install_fail);
5008 subfacet->path = path;
5013 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5015 subfacet_uninstall(struct subfacet *subfacet)
5017 if (subfacet->path != SF_NOT_INSTALLED) {
5018 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5019 struct dpif_flow_stats stats;
5022 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5023 subfacet->key_len, &stats);
5024 subfacet_reset_dp_stats(subfacet, &stats);
5026 subfacet_update_stats(subfacet, &stats);
5028 subfacet->path = SF_NOT_INSTALLED;
5030 ovs_assert(subfacet->dp_packet_count == 0);
5031 ovs_assert(subfacet->dp_byte_count == 0);
5035 /* Resets 'subfacet''s datapath statistics counters. This should be called
5036 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5037 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5038 * was reset in the datapath. 'stats' will be modified to include only
5039 * statistics new since 'subfacet' was last updated. */
5041 subfacet_reset_dp_stats(struct subfacet *subfacet,
5042 struct dpif_flow_stats *stats)
5045 && subfacet->dp_packet_count <= stats->n_packets
5046 && subfacet->dp_byte_count <= stats->n_bytes) {
5047 stats->n_packets -= subfacet->dp_packet_count;
5048 stats->n_bytes -= subfacet->dp_byte_count;
5051 subfacet->dp_packet_count = 0;
5052 subfacet->dp_byte_count = 0;
5055 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5057 * Because of the meaning of a subfacet's counters, it only makes sense to do
5058 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5059 * represents a packet that was sent by hand or if it represents statistics
5060 * that have been cleared out of the datapath. */
5062 subfacet_update_stats(struct subfacet *subfacet,
5063 const struct dpif_flow_stats *stats)
5065 if (stats->n_packets || stats->used > subfacet->used) {
5066 struct facet *facet = subfacet->facet;
5068 subfacet->used = MAX(subfacet->used, stats->used);
5069 facet->used = MAX(facet->used, stats->used);
5070 facet->packet_count += stats->n_packets;
5071 facet->byte_count += stats->n_bytes;
5072 facet->tcp_flags |= stats->tcp_flags;
5078 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5079 * the fields that were relevant as part of the lookup. */
5080 static struct rule_dpif *
5081 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5082 struct flow_wildcards *wc)
5084 struct rule_dpif *rule;
5086 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5091 return rule_dpif_miss_rule(ofproto, flow);
5095 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5096 const struct flow *flow, struct flow_wildcards *wc,
5099 struct cls_rule *cls_rule;
5100 struct classifier *cls;
5103 if (table_id >= N_TABLES) {
5108 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5109 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5112 cls = &ofproto->up.tables[table_id].cls;
5113 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5114 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5115 /* We must pretend that transport ports are unavailable. */
5116 struct flow ofpc_normal_flow = *flow;
5117 ofpc_normal_flow.tp_src = htons(0);
5118 ofpc_normal_flow.tp_dst = htons(0);
5119 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5120 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5121 cls_rule = &ofproto->drop_frags_rule->up.cr;
5123 flow_wildcards_init_exact(wc);
5126 cls_rule = classifier_lookup(cls, flow, wc);
5128 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5132 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5134 struct ofport_dpif *port;
5136 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5138 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5139 flow->in_port.ofp_port);
5140 return ofproto->miss_rule;
5143 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5144 return ofproto->no_packet_in_rule;
5146 return ofproto->miss_rule;
5150 complete_operation(struct rule_dpif *rule)
5152 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5154 rule_invalidate(rule);
5156 struct dpif_completion *c = xmalloc(sizeof *c);
5157 c->op = rule->up.pending;
5158 list_push_back(&ofproto->completions, &c->list_node);
5160 ofoperation_complete(rule->up.pending, 0);
5164 static struct rule *
5167 struct rule_dpif *rule = xmalloc(sizeof *rule);
5172 rule_dealloc(struct rule *rule_)
5174 struct rule_dpif *rule = rule_dpif_cast(rule_);
5179 rule_construct(struct rule *rule_)
5181 struct rule_dpif *rule = rule_dpif_cast(rule_);
5182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5183 struct rule_dpif *victim;
5186 rule->packet_count = 0;
5187 rule->byte_count = 0;
5189 table_id = rule->up.table_id;
5190 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5192 rule->tag = victim->tag;
5193 } else if (table_id == 0) {
5198 miniflow_expand(&rule->up.cr.match.flow, &flow);
5199 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5200 ofproto->tables[table_id].basis);
5203 complete_operation(rule);
5208 rule_destruct(struct rule *rule)
5210 complete_operation(rule_dpif_cast(rule));
5214 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5216 struct rule_dpif *rule = rule_dpif_cast(rule_);
5218 /* push_all_stats() can handle flow misses which, when using the learn
5219 * action, can cause rules to be added and deleted. This can corrupt our
5220 * caller's datastructures which assume that rule_get_stats() doesn't have
5221 * an impact on the flow table. To be safe, we disable miss handling. */
5222 push_all_stats__(false);
5224 /* Start from historical data for 'rule' itself that are no longer tracked
5225 * in facets. This counts, for example, facets that have expired. */
5226 *packets = rule->packet_count;
5227 *bytes = rule->byte_count;
5231 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5232 struct ofpbuf *packet)
5234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5235 struct dpif_flow_stats stats;
5236 struct xlate_out xout;
5237 struct xlate_in xin;
5239 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5240 rule_credit_stats(rule, &stats);
5242 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5243 xin.resubmit_stats = &stats;
5244 xlate_actions(&xin, &xout);
5246 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5247 xout.odp_actions.size, packet);
5249 xlate_out_uninit(&xout);
5253 rule_execute(struct rule *rule, const struct flow *flow,
5254 struct ofpbuf *packet)
5256 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5257 ofpbuf_delete(packet);
5262 rule_modify_actions(struct rule *rule_)
5264 struct rule_dpif *rule = rule_dpif_cast(rule_);
5266 complete_operation(rule);
5269 /* Sends 'packet' out 'ofport'.
5270 * May modify 'packet'.
5271 * Returns 0 if successful, otherwise a positive errno value. */
5273 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5276 uint64_t odp_actions_stub[1024 / 8];
5277 struct ofpbuf key, odp_actions;
5278 struct dpif_flow_stats stats;
5279 struct odputil_keybuf keybuf;
5280 struct ofpact_output output;
5281 struct xlate_out xout;
5282 struct xlate_in xin;
5284 union flow_in_port in_port_;
5287 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5288 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5290 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5291 in_port_.ofp_port = OFPP_NONE;
5292 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5293 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5295 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5297 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5298 output.port = ofport->up.ofp_port;
5301 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5302 xin.ofpacts_len = sizeof output;
5303 xin.ofpacts = &output.ofpact;
5304 xin.resubmit_stats = &stats;
5305 xlate_actions(&xin, &xout);
5307 error = dpif_execute(ofproto->backer->dpif,
5309 xout.odp_actions.data, xout.odp_actions.size,
5311 xlate_out_uninit(&xout);
5314 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5315 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5316 ovs_strerror(error));
5319 ofproto->stats.tx_packets++;
5320 ofproto->stats.tx_bytes += packet->size;
5324 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5325 * The action will state 'slow' as the reason that the action is in the slow
5326 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5327 * dump-flows" output to see why a flow is in the slow path.)
5329 * The 'stub_size' bytes in 'stub' will be used to store the action.
5330 * 'stub_size' must be large enough for the action.
5332 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5335 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5336 enum slow_path_reason slow,
5337 uint64_t *stub, size_t stub_size,
5338 const struct nlattr **actionsp, size_t *actions_lenp)
5340 union user_action_cookie cookie;
5343 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5344 cookie.slow_path.unused = 0;
5345 cookie.slow_path.reason = slow;
5347 ofpbuf_use_stack(&buf, stub, stub_size);
5348 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5349 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5351 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5353 put_userspace_action(ofproto, &buf, flow, &cookie,
5354 sizeof cookie.slow_path);
5356 *actionsp = buf.data;
5357 *actions_lenp = buf.size;
5361 put_userspace_action(const struct ofproto_dpif *ofproto,
5362 struct ofpbuf *odp_actions,
5363 const struct flow *flow,
5364 const union user_action_cookie *cookie,
5365 const size_t cookie_size)
5369 pid = dpif_port_get_pid(ofproto->backer->dpif,
5370 ofp_port_to_odp_port(ofproto,
5371 flow->in_port.ofp_port));
5373 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5377 calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
5378 uint8_t table_id, struct rule_dpif *rule)
5380 if (table_id > 0 && table_id < N_TABLES) {
5381 struct table_dpif *table = &ofproto->tables[table_id];
5382 if (table->other_table) {
5383 return (rule && rule->tag
5385 : rule_calculate_tag(flow, &table->other_table->mask,
5393 /* Optimized flow revalidation.
5395 * It's a difficult problem, in general, to tell which facets need to have
5396 * their actions recalculated whenever the OpenFlow flow table changes. We
5397 * don't try to solve that general problem: for most kinds of OpenFlow flow
5398 * table changes, we recalculate the actions for every facet. This is
5399 * relatively expensive, but it's good enough if the OpenFlow flow table
5400 * doesn't change very often.
5402 * However, we can expect one particular kind of OpenFlow flow table change to
5403 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
5404 * of CPU on revalidating every facet whenever MAC learning modifies the flow
5405 * table, we add a special case that applies to flow tables in which every rule
5406 * has the same form (that is, the same wildcards), except that the table is
5407 * also allowed to have a single "catch-all" flow that matches all packets. We
5408 * optimize this case by tagging all of the facets that resubmit into the table
5409 * and invalidating the same tag whenever a flow changes in that table. The
5410 * end result is that we revalidate just the facets that need it (and sometimes
5411 * a few more, but not all of the facets or even all of the facets that
5412 * resubmit to the table modified by MAC learning). */
5414 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
5415 * into an OpenFlow table with the given 'basis'. */
5417 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
5420 if (minimask_is_catchall(mask)) {
5423 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
5424 return tag_create_deterministic(hash);
5428 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
5429 * taggability of that table.
5431 * This function must be called after *each* change to a flow table. If you
5432 * skip calling it on some changes then the pointer comparisons at the end can
5433 * be invalid if you get unlucky. For example, if a flow removal causes a
5434 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5435 * different wildcards to be created with the same address, then this function
5436 * will incorrectly skip revalidation. */
5438 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5440 struct table_dpif *table = &ofproto->tables[table_id];
5441 const struct oftable *oftable = &ofproto->up.tables[table_id];
5442 struct cls_table *catchall, *other;
5443 struct cls_table *t;
5445 catchall = other = NULL;
5447 switch (hmap_count(&oftable->cls.tables)) {
5449 /* We could tag this OpenFlow table but it would make the logic a
5450 * little harder and it's a corner case that doesn't seem worth it
5456 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
5457 if (cls_table_is_catchall(t)) {
5459 } else if (!other) {
5462 /* Indicate that we can't tag this by setting both tables to
5463 * NULL. (We know that 'catchall' is already NULL.) */
5470 /* Can't tag this table. */
5474 if (table->catchall_table != catchall || table->other_table != other) {
5475 table->catchall_table = catchall;
5476 table->other_table = other;
5477 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5481 /* Given 'rule' that has changed in some way (either it is a rule being
5482 * inserted, a rule being deleted, or a rule whose actions are being
5483 * modified), marks facets for revalidation to ensure that packets will be
5484 * forwarded correctly according to the new state of the flow table.
5486 * This function must be called after *each* change to a flow table. See
5487 * the comment on table_update_taggable() for more information. */
5489 rule_invalidate(const struct rule_dpif *rule)
5491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5493 table_update_taggable(ofproto, rule->up.table_id);
5495 if (!ofproto->backer->need_revalidate) {
5496 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5498 if (table->other_table && rule->tag) {
5499 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
5501 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5507 set_frag_handling(struct ofproto *ofproto_,
5508 enum ofp_config_flags frag_handling)
5510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5511 if (frag_handling != OFPC_FRAG_REASM) {
5512 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5520 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5521 const struct flow *flow,
5522 const struct ofpact *ofpacts, size_t ofpacts_len)
5524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5525 struct odputil_keybuf keybuf;
5526 struct dpif_flow_stats stats;
5527 struct xlate_out xout;
5528 struct xlate_in xin;
5532 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5533 odp_flow_key_from_flow(&key, flow,
5534 ofp_port_to_odp_port(ofproto,
5535 flow->in_port.ofp_port));
5537 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5539 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5540 xin.resubmit_stats = &stats;
5541 xin.ofpacts_len = ofpacts_len;
5542 xin.ofpacts = ofpacts;
5544 xlate_actions(&xin, &xout);
5545 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5546 xout.odp_actions.data, xout.odp_actions.size, packet);
5547 xlate_out_uninit(&xout);
5555 set_netflow(struct ofproto *ofproto_,
5556 const struct netflow_options *netflow_options)
5558 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5560 if (netflow_options) {
5561 if (!ofproto->netflow) {
5562 ofproto->netflow = netflow_create();
5563 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5565 return netflow_set_options(ofproto->netflow, netflow_options);
5566 } else if (ofproto->netflow) {
5567 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5568 netflow_destroy(ofproto->netflow);
5569 ofproto->netflow = NULL;
5576 get_netflow_ids(const struct ofproto *ofproto_,
5577 uint8_t *engine_type, uint8_t *engine_id)
5579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5581 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5585 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5587 if (!facet_is_controller_flow(facet) &&
5588 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5589 struct subfacet *subfacet;
5590 struct ofexpired expired;
5592 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5593 if (subfacet->path == SF_FAST_PATH) {
5594 struct dpif_flow_stats stats;
5596 subfacet_install(subfacet, &facet->xout.odp_actions,
5598 subfacet_update_stats(subfacet, &stats);
5602 expired.flow = facet->flow;
5603 expired.packet_count = facet->packet_count;
5604 expired.byte_count = facet->byte_count;
5605 expired.used = facet->used;
5606 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5611 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5613 struct cls_cursor cursor;
5614 struct facet *facet;
5616 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5617 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5618 send_active_timeout(ofproto, facet);
5622 static struct ofproto_dpif *
5623 ofproto_dpif_lookup(const char *name)
5625 struct ofproto_dpif *ofproto;
5627 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5628 hash_string(name, 0), &all_ofproto_dpifs) {
5629 if (!strcmp(ofproto->up.name, name)) {
5637 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5638 const char *argv[], void *aux OVS_UNUSED)
5640 struct ofproto_dpif *ofproto;
5643 ofproto = ofproto_dpif_lookup(argv[1]);
5645 unixctl_command_reply_error(conn, "no such bridge");
5648 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5650 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5651 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
5655 unixctl_command_reply(conn, "table successfully flushed");
5658 static struct ofport_dpif *
5659 ofbundle_get_a_port(const struct ofbundle *bundle)
5661 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5666 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5667 const char *argv[], void *aux OVS_UNUSED)
5669 struct ds ds = DS_EMPTY_INITIALIZER;
5670 const struct ofproto_dpif *ofproto;
5671 const struct mac_entry *e;
5673 ofproto = ofproto_dpif_lookup(argv[1]);
5675 unixctl_command_reply_error(conn, "no such bridge");
5679 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5680 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5681 struct ofbundle *bundle = e->port.p;
5682 char name[OFP_MAX_PORT_NAME_LEN];
5684 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5686 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5687 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5688 mac_entry_age(ofproto->ml, e));
5690 unixctl_command_reply(conn, ds_cstr(&ds));
5695 struct xlate_out xout;
5696 struct xlate_in xin;
5702 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5704 ds_put_char_multiple(result, '\t', level);
5706 ds_put_cstr(result, "No match\n");
5710 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5711 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5712 cls_rule_format(&rule->up.cr, result);
5713 ds_put_char(result, '\n');
5715 ds_put_char_multiple(result, '\t', level);
5716 ds_put_cstr(result, "OpenFlow ");
5717 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5718 ds_put_char(result, '\n');
5722 trace_format_flow(struct ds *result, int level, const char *title,
5723 struct trace_ctx *trace)
5725 ds_put_char_multiple(result, '\t', level);
5726 ds_put_format(result, "%s: ", title);
5727 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5728 ds_put_cstr(result, "unchanged");
5730 flow_format(result, &trace->xin.flow);
5731 trace->flow = trace->xin.flow;
5733 ds_put_char(result, '\n');
5737 trace_format_regs(struct ds *result, int level, const char *title,
5738 struct trace_ctx *trace)
5742 ds_put_char_multiple(result, '\t', level);
5743 ds_put_format(result, "%s:", title);
5744 for (i = 0; i < FLOW_N_REGS; i++) {
5745 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5747 ds_put_char(result, '\n');
5751 trace_format_odp(struct ds *result, int level, const char *title,
5752 struct trace_ctx *trace)
5754 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5756 ds_put_char_multiple(result, '\t', level);
5757 ds_put_format(result, "%s: ", title);
5758 format_odp_actions(result, odp_actions->data, odp_actions->size);
5759 ds_put_char(result, '\n');
5763 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5765 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5766 struct ds *result = trace->result;
5768 ds_put_char(result, '\n');
5769 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5770 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5771 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5772 trace_format_rule(result, recurse + 1, rule);
5776 trace_report(struct xlate_in *xin, const char *s, int recurse)
5778 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5779 struct ds *result = trace->result;
5781 ds_put_char_multiple(result, '\t', recurse);
5782 ds_put_cstr(result, s);
5783 ds_put_char(result, '\n');
5787 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5788 void *aux OVS_UNUSED)
5790 const struct dpif_backer *backer;
5791 struct ofproto_dpif *ofproto;
5792 struct ofpbuf odp_key, odp_mask;
5793 struct ofpbuf *packet;
5801 ofpbuf_init(&odp_key, 0);
5802 ofpbuf_init(&odp_mask, 0);
5804 /* Handle "-generate" or a hex string as the last argument. */
5805 if (!strcmp(argv[argc - 1], "-generate")) {
5806 packet = ofpbuf_new(0);
5809 const char *error = eth_from_hex(argv[argc - 1], &packet);
5812 } else if (argc == 4) {
5813 /* The 3-argument form must end in "-generate' or a hex string. */
5814 unixctl_command_reply_error(conn, error);
5819 /* Parse the flow and determine whether a datapath or
5820 * bridge is specified. If function odp_flow_key_from_string()
5821 * returns 0, the flow is a odp_flow. If function
5822 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5823 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5824 /* If the odp_flow is the second argument,
5825 * the datapath name is the first argument. */
5827 const char *dp_type;
5828 if (!strncmp(argv[1], "ovs-", 4)) {
5829 dp_type = argv[1] + 4;
5833 backer = shash_find_data(&all_dpif_backers, dp_type);
5835 unixctl_command_reply_error(conn, "Cannot find datapath "
5840 /* No datapath name specified, so there should be only one
5842 struct shash_node *node;
5843 if (shash_count(&all_dpif_backers) != 1) {
5844 unixctl_command_reply_error(conn, "Must specify datapath "
5845 "name, there is more than one type of datapath");
5848 node = shash_first(&all_dpif_backers);
5849 backer = node->data;
5852 /* Extract the ofproto_dpif object from the ofproto_receive()
5854 if (ofproto_receive(backer, NULL, odp_key.data,
5855 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5856 unixctl_command_reply_error(conn, "Invalid datapath flow");
5859 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5860 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5862 unixctl_command_reply_error(conn, "Must specify bridge name");
5866 ofproto = ofproto_dpif_lookup(argv[1]);
5868 unixctl_command_reply_error(conn, "Unknown bridge name");
5872 unixctl_command_reply_error(conn, "Bad flow syntax");
5876 /* Generate a packet, if requested. */
5878 if (!packet->size) {
5879 flow_compose(packet, &flow);
5881 union flow_in_port in_port_;
5883 in_port_ = flow.in_port;
5884 ds_put_cstr(&result, "Packet: ");
5885 s = ofp_packet_to_string(packet->data, packet->size);
5886 ds_put_cstr(&result, s);
5889 /* Use the metadata from the flow and the packet argument
5890 * to reconstruct the flow. */
5891 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5896 ofproto_trace(ofproto, &flow, packet, &result);
5897 unixctl_command_reply(conn, ds_cstr(&result));
5900 ds_destroy(&result);
5901 ofpbuf_delete(packet);
5902 ofpbuf_uninit(&odp_key);
5903 ofpbuf_uninit(&odp_mask);
5907 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5908 const struct ofpbuf *packet, struct ds *ds)
5910 struct rule_dpif *rule;
5912 ds_put_cstr(ds, "Flow: ");
5913 flow_format(ds, flow);
5914 ds_put_char(ds, '\n');
5916 rule = rule_dpif_lookup(ofproto, flow, NULL);
5918 trace_format_rule(ds, 0, rule);
5919 if (rule == ofproto->miss_rule) {
5920 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5921 } else if (rule == ofproto->no_packet_in_rule) {
5922 ds_put_cstr(ds, "\nNo match, packets dropped because "
5923 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5924 } else if (rule == ofproto->drop_frags_rule) {
5925 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5926 "and the fragment handling mode is \"drop\".\n");
5930 uint64_t odp_actions_stub[1024 / 8];
5931 struct ofpbuf odp_actions;
5932 struct trace_ctx trace;
5936 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5939 ofpbuf_use_stub(&odp_actions,
5940 odp_actions_stub, sizeof odp_actions_stub);
5941 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5942 trace.xin.resubmit_hook = trace_resubmit;
5943 trace.xin.report_hook = trace_report;
5945 xlate_actions(&trace.xin, &trace.xout);
5947 ds_put_char(ds, '\n');
5948 trace_format_flow(ds, 0, "Final flow", &trace);
5950 match_init(&match, flow, &trace.xout.wc);
5951 ds_put_cstr(ds, "Relevant fields: ");
5952 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5953 ds_put_char(ds, '\n');
5955 ds_put_cstr(ds, "Datapath actions: ");
5956 format_odp_actions(ds, trace.xout.odp_actions.data,
5957 trace.xout.odp_actions.size);
5959 if (trace.xout.slow) {
5960 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5961 "slow path because it:");
5962 switch (trace.xout.slow) {
5964 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5967 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5970 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5973 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5975 case SLOW_CONTROLLER:
5976 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5977 "to the OpenFlow controller.");
5984 xlate_out_uninit(&trace.xout);
5989 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5990 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5993 unixctl_command_reply(conn, NULL);
5997 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5998 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6001 unixctl_command_reply(conn, NULL);
6004 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6005 * 'reply' describing the results. */
6007 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6009 struct cls_cursor cursor;
6010 struct facet *facet;
6014 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6015 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6016 if (!facet_check_consistency(facet)) {
6021 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
6025 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6026 ofproto->up.name, errors);
6028 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6033 ofproto_dpif_self_check(struct unixctl_conn *conn,
6034 int argc, const char *argv[], void *aux OVS_UNUSED)
6036 struct ds reply = DS_EMPTY_INITIALIZER;
6037 struct ofproto_dpif *ofproto;
6040 ofproto = ofproto_dpif_lookup(argv[1]);
6042 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6043 "ofproto/list for help)");
6046 ofproto_dpif_self_check__(ofproto, &reply);
6048 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6049 ofproto_dpif_self_check__(ofproto, &reply);
6053 unixctl_command_reply(conn, ds_cstr(&reply));
6057 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6058 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6059 * to destroy 'ofproto_shash' and free the returned value. */
6060 static const struct shash_node **
6061 get_ofprotos(struct shash *ofproto_shash)
6063 const struct ofproto_dpif *ofproto;
6065 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6066 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6067 shash_add_nocopy(ofproto_shash, name, ofproto);
6070 return shash_sort(ofproto_shash);
6074 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6075 const char *argv[] OVS_UNUSED,
6076 void *aux OVS_UNUSED)
6078 struct ds ds = DS_EMPTY_INITIALIZER;
6079 struct shash ofproto_shash;
6080 const struct shash_node **sorted_ofprotos;
6083 shash_init(&ofproto_shash);
6084 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6085 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6086 const struct shash_node *node = sorted_ofprotos[i];
6087 ds_put_format(&ds, "%s\n", node->name);
6090 shash_destroy(&ofproto_shash);
6091 free(sorted_ofprotos);
6093 unixctl_command_reply(conn, ds_cstr(&ds));
6098 show_dp_rates(struct ds *ds, const char *heading,
6099 const struct avg_subfacet_rates *rates)
6101 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6102 heading, rates->add_rate, rates->del_rate);
6106 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6108 const struct shash_node **ofprotos;
6109 struct ofproto_dpif *ofproto;
6110 struct shash ofproto_shash;
6111 uint64_t n_hit, n_missed;
6112 long long int minutes;
6115 n_hit = n_missed = 0;
6116 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6117 if (ofproto->backer == backer) {
6118 n_missed += ofproto->n_missed;
6119 n_hit += ofproto->n_hit;
6123 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6124 dpif_name(backer->dpif), n_hit, n_missed);
6125 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6126 " life span: %lldms\n", hmap_count(&backer->subfacets),
6127 backer->avg_n_subfacet, backer->max_n_subfacet,
6128 backer->avg_subfacet_life);
6130 minutes = (time_msec() - backer->created) / (1000 * 60);
6131 if (minutes >= 60) {
6132 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6134 if (minutes >= 60 * 24) {
6135 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6137 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6139 shash_init(&ofproto_shash);
6140 ofprotos = get_ofprotos(&ofproto_shash);
6141 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6142 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6143 const struct shash_node **ports;
6146 if (ofproto->backer != backer) {
6150 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6151 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6153 ports = shash_sort(&ofproto->up.port_by_name);
6154 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6155 const struct shash_node *node = ports[j];
6156 struct ofport *ofport = node->data;
6158 odp_port_t odp_port;
6160 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6163 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6164 if (odp_port != ODPP_NONE) {
6165 ds_put_format(ds, "%"PRIu32":", odp_port);
6167 ds_put_cstr(ds, "none:");
6170 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6173 if (!netdev_get_config(ofport->netdev, &config)) {
6174 const struct smap_node **nodes;
6177 nodes = smap_sort(&config);
6178 for (i = 0; i < smap_count(&config); i++) {
6179 const struct smap_node *node = nodes[i];
6180 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6181 node->key, node->value);
6185 smap_destroy(&config);
6187 ds_put_char(ds, ')');
6188 ds_put_char(ds, '\n');
6192 shash_destroy(&ofproto_shash);
6197 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6198 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6200 struct ds ds = DS_EMPTY_INITIALIZER;
6201 const struct shash_node **backers;
6204 backers = shash_sort(&all_dpif_backers);
6205 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6206 dpif_show_backer(backers[i]->data, &ds);
6210 unixctl_command_reply(conn, ds_cstr(&ds));
6214 /* Dump the megaflow (facet) cache. This is useful to check the
6215 * correctness of flow wildcarding, since the same mechanism is used for
6216 * both xlate caching and kernel wildcarding.
6218 * It's important to note that in the output the flow description uses
6219 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6221 * This command is only needed for advanced debugging, so it's not
6222 * documented in the man page. */
6224 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6225 int argc OVS_UNUSED, const char *argv[],
6226 void *aux OVS_UNUSED)
6228 struct ds ds = DS_EMPTY_INITIALIZER;
6229 const struct ofproto_dpif *ofproto;
6230 long long int now = time_msec();
6231 struct cls_cursor cursor;
6232 struct facet *facet;
6234 ofproto = ofproto_dpif_lookup(argv[1]);
6236 unixctl_command_reply_error(conn, "no such bridge");
6240 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6241 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6242 cls_rule_format(&facet->cr, &ds);
6243 ds_put_cstr(&ds, ", ");
6244 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6245 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6246 ds_put_cstr(&ds, "Datapath actions: ");
6247 if (facet->xout.slow) {
6248 uint64_t slow_path_stub[128 / 8];
6249 const struct nlattr *actions;
6252 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6253 slow_path_stub, sizeof slow_path_stub,
6254 &actions, &actions_len);
6255 format_odp_actions(&ds, actions, actions_len);
6257 format_odp_actions(&ds, facet->xout.odp_actions.data,
6258 facet->xout.odp_actions.size);
6260 ds_put_cstr(&ds, "\n");
6263 ds_chomp(&ds, '\n');
6264 unixctl_command_reply(conn, ds_cstr(&ds));
6268 /* Disable using the megaflows.
6270 * This command is only needed for advanced debugging, so it's not
6271 * documented in the man page. */
6273 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6274 int argc OVS_UNUSED,
6275 const char *argv[] OVS_UNUSED,
6276 void *aux OVS_UNUSED)
6278 struct ofproto_dpif *ofproto;
6280 enable_megaflows = false;
6282 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6283 flush(&ofproto->up);
6286 unixctl_command_reply(conn, "megaflows disabled");
6289 /* Re-enable using megaflows.
6291 * This command is only needed for advanced debugging, so it's not
6292 * documented in the man page. */
6294 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6295 int argc OVS_UNUSED,
6296 const char *argv[] OVS_UNUSED,
6297 void *aux OVS_UNUSED)
6299 struct ofproto_dpif *ofproto;
6301 enable_megaflows = true;
6303 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6304 flush(&ofproto->up);
6307 unixctl_command_reply(conn, "megaflows enabled");
6311 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6312 int argc OVS_UNUSED, const char *argv[],
6313 void *aux OVS_UNUSED)
6315 struct ds ds = DS_EMPTY_INITIALIZER;
6316 const struct ofproto_dpif *ofproto;
6317 struct subfacet *subfacet;
6319 ofproto = ofproto_dpif_lookup(argv[1]);
6321 unixctl_command_reply_error(conn, "no such bridge");
6325 update_stats(ofproto->backer);
6327 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6328 struct facet *facet = subfacet->facet;
6330 if (facet->ofproto != ofproto) {
6334 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
6336 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6337 subfacet->dp_packet_count, subfacet->dp_byte_count);
6338 if (subfacet->used) {
6339 ds_put_format(&ds, "%.3fs",
6340 (time_msec() - subfacet->used) / 1000.0);
6342 ds_put_format(&ds, "never");
6344 if (subfacet->facet->tcp_flags) {
6345 ds_put_cstr(&ds, ", flags:");
6346 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6349 ds_put_cstr(&ds, ", actions:");
6350 if (facet->xout.slow) {
6351 uint64_t slow_path_stub[128 / 8];
6352 const struct nlattr *actions;
6355 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6356 slow_path_stub, sizeof slow_path_stub,
6357 &actions, &actions_len);
6358 format_odp_actions(&ds, actions, actions_len);
6360 format_odp_actions(&ds, facet->xout.odp_actions.data,
6361 facet->xout.odp_actions.size);
6363 ds_put_char(&ds, '\n');
6366 unixctl_command_reply(conn, ds_cstr(&ds));
6371 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6372 int argc OVS_UNUSED, const char *argv[],
6373 void *aux OVS_UNUSED)
6375 struct ds ds = DS_EMPTY_INITIALIZER;
6376 struct ofproto_dpif *ofproto;
6378 ofproto = ofproto_dpif_lookup(argv[1]);
6380 unixctl_command_reply_error(conn, "no such bridge");
6384 flush(&ofproto->up);
6386 unixctl_command_reply(conn, ds_cstr(&ds));
6391 ofproto_dpif_unixctl_init(void)
6393 static bool registered;
6399 unixctl_command_register(
6401 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6402 1, 3, ofproto_unixctl_trace, NULL);
6403 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6404 ofproto_unixctl_fdb_flush, NULL);
6405 unixctl_command_register("fdb/show", "bridge", 1, 1,
6406 ofproto_unixctl_fdb_show, NULL);
6407 unixctl_command_register("ofproto/clog", "", 0, 0,
6408 ofproto_dpif_clog, NULL);
6409 unixctl_command_register("ofproto/unclog", "", 0, 0,
6410 ofproto_dpif_unclog, NULL);
6411 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6412 ofproto_dpif_self_check, NULL);
6413 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6414 ofproto_unixctl_dpif_dump_dps, NULL);
6415 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6417 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6418 ofproto_unixctl_dpif_dump_flows, NULL);
6419 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6420 ofproto_unixctl_dpif_del_flows, NULL);
6421 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6422 ofproto_unixctl_dpif_dump_megaflows, NULL);
6423 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6424 ofproto_unixctl_dpif_disable_megaflows, NULL);
6425 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6426 ofproto_unixctl_dpif_enable_megaflows, NULL);
6429 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6431 * This is deprecated. It is only for compatibility with broken device drivers
6432 * in old versions of Linux that do not properly support VLANs when VLAN
6433 * devices are not used. When broken device drivers are no longer in
6434 * widespread use, we will delete these interfaces. */
6437 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6439 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6440 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6442 if (realdev_ofp_port == ofport->realdev_ofp_port
6443 && vid == ofport->vlandev_vid) {
6447 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6449 if (ofport->realdev_ofp_port) {
6452 if (realdev_ofp_port && ofport->bundle) {
6453 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6454 * themselves be part of a bundle. */
6455 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6458 ofport->realdev_ofp_port = realdev_ofp_port;
6459 ofport->vlandev_vid = vid;
6461 if (realdev_ofp_port) {
6462 vsp_add(ofport, realdev_ofp_port, vid);
6469 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6471 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6475 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6477 return !hmap_is_empty(&ofproto->realdev_vid_map);
6480 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6481 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6482 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6483 * 'vlan_tci' 9, it would return the port number of eth0.9.
6485 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6486 * function just returns its 'realdev_ofp_port' argument. */
6488 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6489 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6491 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6492 int vid = vlan_tci_to_vid(vlan_tci);
6493 const struct vlan_splinter *vsp;
6495 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6496 hash_realdev_vid(realdev_ofp_port, vid),
6497 &ofproto->realdev_vid_map) {
6498 if (vsp->realdev_ofp_port == realdev_ofp_port
6499 && vsp->vid == vid) {
6500 return vsp->vlandev_ofp_port;
6504 return realdev_ofp_port;
6507 static struct vlan_splinter *
6508 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6510 struct vlan_splinter *vsp;
6512 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6513 hash_ofp_port(vlandev_ofp_port),
6514 &ofproto->vlandev_map) {
6515 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6523 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6524 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6525 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6526 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6527 * eth0 and store 9 in '*vid'.
6529 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6530 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6533 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6534 ofp_port_t vlandev_ofp_port, int *vid)
6536 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6537 const struct vlan_splinter *vsp;
6539 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6544 return vsp->realdev_ofp_port;
6550 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6551 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6552 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6553 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6554 * always the case unless VLAN splinters are enabled), returns false without
6555 * making any changes. */
6557 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6562 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6567 /* Cause the flow to be processed as if it came in on the real device with
6568 * the VLAN device's VLAN ID. */
6569 flow->in_port.ofp_port = realdev;
6570 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6575 vsp_remove(struct ofport_dpif *port)
6577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6578 struct vlan_splinter *vsp;
6580 vsp = vlandev_find(ofproto, port->up.ofp_port);
6582 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6583 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6586 port->realdev_ofp_port = 0;
6588 VLOG_ERR("missing vlan device record");
6593 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6597 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6598 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6599 == realdev_ofp_port)) {
6600 struct vlan_splinter *vsp;
6602 vsp = xmalloc(sizeof *vsp);
6603 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6604 hash_ofp_port(port->up.ofp_port));
6605 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6606 hash_realdev_vid(realdev_ofp_port, vid));
6607 vsp->realdev_ofp_port = realdev_ofp_port;
6608 vsp->vlandev_ofp_port = port->up.ofp_port;
6611 port->realdev_ofp_port = realdev_ofp_port;
6613 VLOG_ERR("duplicate vlan device record");
6618 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6620 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6621 return ofport ? ofport->odp_port : ODPP_NONE;
6624 static struct ofport_dpif *
6625 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6627 struct ofport_dpif *port;
6629 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6630 &backer->odp_to_ofport_map) {
6631 if (port->odp_port == odp_port) {
6640 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6642 struct ofport_dpif *port;
6644 port = odp_port_to_ofport(ofproto->backer, odp_port);
6645 if (port && &ofproto->up == port->up.ofproto) {
6646 return port->up.ofp_port;
6652 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6653 * most heavily weighted element. 'base' designates the rate of decay: after
6654 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6657 exp_mavg(double *avg, int base, double new)
6659 *avg = (*avg * (base - 1) + new) / base;
6663 update_moving_averages(struct dpif_backer *backer)
6665 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6666 long long int minutes = (time_msec() - backer->created) / min_ms;
6669 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6671 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6674 backer->lifetime.add_rate = 0.0;
6675 backer->lifetime.del_rate = 0.0;
6678 /* Update hourly averages on the minute boundaries. */
6679 if (time_msec() - backer->last_minute >= min_ms) {
6680 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6681 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6683 /* Update daily averages on the hour boundaries. */
6684 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6685 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6686 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6689 backer->total_subfacet_add_count += backer->subfacet_add_count;
6690 backer->total_subfacet_del_count += backer->subfacet_del_count;
6691 backer->subfacet_add_count = 0;
6692 backer->subfacet_del_count = 0;
6693 backer->last_minute += min_ms;
6697 const struct ofproto_class ofproto_dpif_class = {
6732 port_is_lacp_current,
6733 NULL, /* rule_choose_table */
6740 rule_modify_actions,
6754 get_stp_port_status,
6761 is_mirror_output_bundle,
6762 forward_bpdu_changed,
6763 set_mac_table_config,
6765 NULL, /* meter_get_features */
6766 NULL, /* meter_set */
6767 NULL, /* meter_get */
6768 NULL, /* meter_del */