2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
74 COVERAGE_DEFINE(flow_mod_overflow);
76 /* Number of implemented OpenFlow tables. */
77 enum { N_TABLES = 255 };
78 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
86 struct flow_wildcards *wc);
88 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
91 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
92 struct ofproto_dpif *ofproto; /* Owning ofproto. */
93 void *aux; /* Key supplied by ofproto's client. */
94 char *name; /* Identifier for log messages. */
97 struct list ports; /* Contains "struct ofport"s. */
98 enum port_vlan_mode vlan_mode; /* VLAN mode */
99 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
100 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
101 * NULL if all VLANs are trunked. */
102 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
103 struct bond *bond; /* Nonnull iff more than one port. */
104 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
107 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
110 static void bundle_remove(struct ofport *);
111 static void bundle_update(struct ofbundle *);
112 static void bundle_destroy(struct ofbundle *);
113 static void bundle_del_port(struct ofport_dpif *);
114 static void bundle_run(struct ofbundle *);
115 static void bundle_wait(struct ofbundle *);
117 static void stp_run(struct ofproto_dpif *ofproto);
118 static void stp_wait(struct ofproto_dpif *ofproto);
119 static int set_stp_port(struct ofport *,
120 const struct ofproto_port_stp_settings *);
122 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
123 enum slow_path_reason,
124 uint64_t *stub, size_t stub_size,
125 const struct nlattr **actionsp,
126 size_t *actions_lenp);
128 /* A subfacet (see "struct subfacet" below) has three possible installation
131 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
132 * case just after the subfacet is created, just before the subfacet is
133 * destroyed, or if the datapath returns an error when we try to install a
136 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
138 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
139 * ofproto_dpif is installed in the datapath.
142 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
143 SF_FAST_PATH, /* Full actions are installed. */
144 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
147 /* A dpif flow and actions associated with a facet.
149 * See also the large comment on struct facet. */
152 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
153 struct list list_node; /* In struct facet's 'facets' list. */
154 struct facet *facet; /* Owning facet. */
155 struct dpif_backer *backer; /* Owning backer. */
157 enum odp_key_fitness key_fitness;
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Time created. */
164 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
165 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
167 enum subfacet_path path; /* Installed in datapath? */
170 #define SUBFACET_DESTROY_MAX_BATCH 50
172 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
174 static struct subfacet *subfacet_find(struct dpif_backer *,
175 const struct nlattr *key, size_t key_len,
177 static void subfacet_destroy(struct subfacet *);
178 static void subfacet_destroy__(struct subfacet *);
179 static void subfacet_destroy_batch(struct dpif_backer *,
180 struct subfacet **, int n);
181 static void subfacet_reset_dp_stats(struct subfacet *,
182 struct dpif_flow_stats *);
183 static void subfacet_update_stats(struct subfacet *,
184 const struct dpif_flow_stats *);
185 static int subfacet_install(struct subfacet *,
186 const struct ofpbuf *odp_actions,
187 struct dpif_flow_stats *);
188 static void subfacet_uninstall(struct subfacet *);
190 /* A unique, non-overlapping instantiation of an OpenFlow flow.
192 * A facet associates a "struct flow", which represents the Open vSwitch
193 * userspace idea of an exact-match flow, with one or more subfacets.
194 * While the facet is created based on an exact-match flow, it is stored
195 * within the ofproto based on the wildcards that could be expressed
196 * based on the flow table and other configuration. (See the 'wc'
197 * description in "struct xlate_out" for more details.)
199 * Each subfacet tracks the datapath's idea of the flow equivalent to
200 * the facet. When the kernel module (or other dpif implementation) and
201 * Open vSwitch userspace agree on the definition of a flow key, there
202 * is exactly one subfacet per facet. If the dpif implementation
203 * supports more-specific flow matching than userspace, however, a facet
204 * can have more than one subfacet. Examples include the dpif
205 * implementation not supporting the same wildcards as userspace or some
206 * distinction in flow that userspace simply doesn't understand.
208 * Flow expiration works in terms of subfacets, so a facet must have at
209 * least one subfacet or it will never expire, leaking memory. */
212 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
213 struct ofproto_dpif *ofproto;
216 struct list subfacets;
217 long long int used; /* Time last used; time created if not used. */
220 struct flow flow; /* Flow of the creating subfacet. */
221 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
225 * - Do include packets and bytes sent "by hand", e.g. with
228 * - Do include packets and bytes that were obtained from the datapath
229 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
230 * DPIF_FP_ZERO_STATS).
232 * - Do not include packets or bytes that can be obtained from the
233 * datapath for any existing subfacet.
235 uint64_t packet_count; /* Number of packets received. */
236 uint64_t byte_count; /* Number of bytes received. */
238 /* Resubmit statistics. */
239 uint64_t prev_packet_count; /* Number of packets from last stats push. */
240 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
241 long long int prev_used; /* Used time from last stats push. */
244 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
245 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
246 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
248 struct xlate_out xout;
249 bool fail_open; /* Facet matched the fail open rule. */
251 /* Storage for a single subfacet, to reduce malloc() time and space
252 * overhead. (A facet always has at least one subfacet and in the common
253 * case has exactly one subfacet. However, 'one_subfacet' may not
254 * always be valid, since it could have been removed after newer
255 * subfacets were pushed onto the 'subfacets' list.) */
256 struct subfacet one_subfacet;
258 long long int learn_rl; /* Rate limiter for facet_learn(). */
261 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
263 struct dpif_flow_stats *);
264 static void facet_remove(struct facet *);
265 static void facet_free(struct facet *);
267 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
268 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
269 const struct flow *);
270 static bool facet_revalidate(struct facet *);
271 static bool facet_check_consistency(struct facet *);
273 static void facet_flush_stats(struct facet *);
275 static void facet_reset_counters(struct facet *);
276 static void facet_push_stats(struct facet *, bool may_learn);
277 static void facet_learn(struct facet *);
278 static void facet_account(struct facet *);
279 static void push_all_stats(void);
281 static bool facet_is_controller_flow(struct facet *);
284 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
288 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
289 struct list bundle_node; /* In struct ofbundle's "ports" list. */
290 struct cfm *cfm; /* Connectivity Fault Management, if any. */
291 struct bfd *bfd; /* BFD, if any. */
292 bool may_enable; /* May be enabled in bonds. */
293 bool is_tunnel; /* This port is a tunnel. */
294 long long int carrier_seq; /* Carrier status changes. */
295 struct ofport_dpif *peer; /* Peer if patch port. */
298 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
299 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
300 long long int stp_state_entered;
302 /* Queue to DSCP mapping. */
303 struct ofproto_port_queue *qdscp;
306 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
308 * This is deprecated. It is only for compatibility with broken device
309 * drivers in old versions of Linux that do not properly support VLANs when
310 * VLAN devices are not used. When broken device drivers are no longer in
311 * widespread use, we will delete these interfaces. */
312 ofp_port_t realdev_ofp_port;
316 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
318 * This is deprecated. It is only for compatibility with broken device drivers
319 * in old versions of Linux that do not properly support VLANs when VLAN
320 * devices are not used. When broken device drivers are no longer in
321 * widespread use, we will delete these interfaces. */
322 struct vlan_splinter {
323 struct hmap_node realdev_vid_node;
324 struct hmap_node vlandev_node;
325 ofp_port_t realdev_ofp_port;
326 ofp_port_t vlandev_ofp_port;
330 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
331 static void vsp_remove(struct ofport_dpif *);
332 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
334 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
337 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
340 static struct ofport_dpif *
341 ofport_dpif_cast(const struct ofport *ofport)
343 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
346 static void port_run(struct ofport_dpif *);
347 static void port_run_fast(struct ofport_dpif *);
348 static void port_wait(struct ofport_dpif *);
349 static int set_bfd(struct ofport *, const struct smap *);
350 static int set_cfm(struct ofport *, const struct cfm_settings *);
351 static void ofport_update_peer(struct ofport_dpif *);
352 static void run_fast_rl(void);
353 static int run_fast(struct ofproto *);
355 struct dpif_completion {
356 struct list list_node;
357 struct ofoperation *op;
360 /* Reasons that we might need to revalidate every facet, and corresponding
363 * A value of 0 means that there is no need to revalidate.
365 * It would be nice to have some cleaner way to integrate with coverage
366 * counters, but with only a few reasons I guess this is good enough for
368 enum revalidate_reason {
369 REV_RECONFIGURE = 1, /* Switch configuration changed. */
370 REV_STP, /* Spanning tree protocol port status change. */
371 REV_BOND, /* Bonding changed. */
372 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
373 REV_FLOW_TABLE, /* Flow table changed. */
374 REV_MAC_LEARNING, /* Mac learning changed. */
375 REV_INCONSISTENCY /* Facet self-check failed. */
377 COVERAGE_DEFINE(rev_reconfigure);
378 COVERAGE_DEFINE(rev_stp);
379 COVERAGE_DEFINE(rev_bond);
380 COVERAGE_DEFINE(rev_port_toggled);
381 COVERAGE_DEFINE(rev_flow_table);
382 COVERAGE_DEFINE(rev_mac_learning);
383 COVERAGE_DEFINE(rev_inconsistency);
385 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
386 * These are datapath flows which have no associated ofproto, if they did we
387 * would use facets. */
389 struct hmap_node hmap_node;
394 struct avg_subfacet_rates {
395 double add_rate; /* Moving average of new flows created per minute. */
396 double del_rate; /* Moving average of flows deleted per minute. */
399 /* All datapaths of a given type share a single dpif backer instance. */
404 struct timer next_expiration;
405 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
407 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
409 /* Facet revalidation flags applying to facets which use this backer. */
410 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
412 struct hmap drop_keys; /* Set of dropped odp keys. */
413 bool recv_set_enable; /* Enables or disables receiving packets. */
415 struct hmap subfacets;
416 struct governor *governor;
418 /* Subfacet statistics.
420 * These keep track of the total number of subfacets added and deleted and
421 * flow life span. They are useful for computing the flow rates stats
422 * exposed via "ovs-appctl dpif/show". The goal is to learn about
423 * traffic patterns in ways that we can use later to improve Open vSwitch
424 * performance in new situations. */
425 long long int created; /* Time when it is created. */
426 unsigned max_n_subfacet; /* Maximum number of flows */
427 unsigned avg_n_subfacet; /* Average number of flows. */
428 long long int avg_subfacet_life; /* Average life span of subfacets. */
430 /* The average number of subfacets... */
431 struct avg_subfacet_rates hourly; /* ...over the last hour. */
432 struct avg_subfacet_rates daily; /* ...over the last day. */
433 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
434 long long int last_minute; /* Last time 'hourly' was updated. */
436 /* Number of subfacets added or deleted since 'last_minute'. */
437 unsigned subfacet_add_count;
438 unsigned subfacet_del_count;
440 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
441 unsigned long long int total_subfacet_add_count;
442 unsigned long long int total_subfacet_del_count;
445 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
446 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
448 static void drop_key_clear(struct dpif_backer *);
449 static struct ofport_dpif *
450 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
451 static void update_moving_averages(struct dpif_backer *backer);
453 struct ofproto_dpif {
454 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
456 struct dpif_backer *backer;
458 /* Special OpenFlow rules. */
459 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
460 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
461 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
464 struct netflow *netflow;
465 struct dpif_sflow *sflow;
466 struct dpif_ipfix *ipfix;
467 struct hmap bundles; /* Contains "struct ofbundle"s. */
468 struct mac_learning *ml;
469 bool has_bonded_bundles;
470 struct mbridge *mbridge;
473 struct classifier facets; /* Contains 'struct facet's. */
474 long long int consistency_rl;
476 /* Support for debugging async flow mods. */
477 struct list completions;
479 struct netdev_stats stats; /* To account packets generated and consumed in
484 long long int stp_last_tick;
486 /* VLAN splinters. */
487 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
488 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
491 struct sset ports; /* Set of standard port names. */
492 struct sset ghost_ports; /* Ports with no datapath port. */
493 struct sset port_poll_set; /* Queued names for port_poll() reply. */
494 int port_poll_errno; /* Last errno for port_poll() reply. */
496 /* Per ofproto's dpif stats. */
501 struct ovs_mutex flow_mod_mutex;
502 struct list flow_mods OVS_GUARDED;
503 size_t n_flow_mods OVS_GUARDED;
506 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
507 * for debugging the asynchronous flow_mod implementation.) */
510 /* By default, flows in the datapath are wildcarded (megaflows). They
511 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
512 static bool enable_megaflows = true;
514 /* All existing ofproto_dpif instances, indexed by ->up.name. */
515 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
517 static void ofproto_dpif_unixctl_init(void);
519 static inline struct ofproto_dpif *
520 ofproto_dpif_cast(const struct ofproto *ofproto)
522 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
523 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
526 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
527 ofp_port_t ofp_port);
528 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
529 const struct ofpbuf *packet, struct ds *);
532 #define FLOW_MISS_MAX_BATCH 50
533 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
535 /* Flow expiration. */
536 static int expire(struct dpif_backer *);
539 static void send_netflow_active_timeouts(struct ofproto_dpif *);
542 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
544 /* Global variables. */
545 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
547 /* Initial mappings of port to bridge mappings. */
548 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
550 /* Executes and takes ownership of 'fm'. */
552 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
553 struct ofputil_flow_mod *fm)
555 ovs_mutex_lock(&ofproto->flow_mod_mutex);
556 if (ofproto->n_flow_mods > 1024) {
557 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
558 COVERAGE_INC(flow_mod_overflow);
564 list_push_back(&ofproto->flow_mods, &fm->list_node);
565 ofproto->n_flow_mods++;
566 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
570 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
571 struct ofputil_packet_in *pin)
573 connmgr_send_packet_in(ofproto->up.connmgr, pin);
576 /* Factory functions. */
579 init(const struct shash *iface_hints)
581 struct shash_node *node;
583 /* Make a local copy, since we don't own 'iface_hints' elements. */
584 SHASH_FOR_EACH(node, iface_hints) {
585 const struct iface_hint *orig_hint = node->data;
586 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
588 new_hint->br_name = xstrdup(orig_hint->br_name);
589 new_hint->br_type = xstrdup(orig_hint->br_type);
590 new_hint->ofp_port = orig_hint->ofp_port;
592 shash_add(&init_ofp_ports, node->name, new_hint);
597 enumerate_types(struct sset *types)
599 dp_enumerate_types(types);
603 enumerate_names(const char *type, struct sset *names)
605 struct ofproto_dpif *ofproto;
608 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
609 if (strcmp(type, ofproto->up.type)) {
612 sset_add(names, ofproto->up.name);
619 del(const char *type, const char *name)
624 error = dpif_open(name, type, &dpif);
626 error = dpif_delete(dpif);
633 port_open_type(const char *datapath_type, const char *port_type)
635 return dpif_port_open_type(datapath_type, port_type);
638 /* Type functions. */
640 static void process_dpif_port_changes(struct dpif_backer *);
641 static void process_dpif_all_ports_changed(struct dpif_backer *);
642 static void process_dpif_port_change(struct dpif_backer *,
643 const char *devname);
644 static void process_dpif_port_error(struct dpif_backer *, int error);
646 static struct ofproto_dpif *
647 lookup_ofproto_dpif_by_port_name(const char *name)
649 struct ofproto_dpif *ofproto;
651 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
652 if (sset_contains(&ofproto->ports, name)) {
661 type_run(const char *type)
663 static long long int push_timer = LLONG_MIN;
664 struct dpif_backer *backer;
666 backer = shash_find_data(&all_dpif_backers, type);
668 /* This is not necessarily a problem, since backers are only
669 * created on demand. */
673 dpif_run(backer->dpif);
675 /* The most natural place to push facet statistics is when they're pulled
676 * from the datapath. However, when there are many flows in the datapath,
677 * this expensive operation can occur so frequently, that it reduces our
678 * ability to quickly set up flows. To reduce the cost, we push statistics
680 if (time_msec() > push_timer) {
681 push_timer = time_msec() + 2000;
685 /* If vswitchd started with other_config:flow_restore_wait set as "true",
686 * and the configuration has now changed to "false", enable receiving
687 * packets from the datapath. */
688 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
691 backer->recv_set_enable = true;
693 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
695 VLOG_ERR("Failed to enable receiving packets in dpif.");
698 dpif_flow_flush(backer->dpif);
699 backer->need_revalidate = REV_RECONFIGURE;
702 if (backer->need_revalidate) {
703 struct ofproto_dpif *ofproto;
704 struct simap_node *node;
705 struct simap tmp_backers;
707 /* Handle tunnel garbage collection. */
708 simap_init(&tmp_backers);
709 simap_swap(&backer->tnl_backers, &tmp_backers);
711 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
712 struct ofport_dpif *iter;
714 if (backer != ofproto->backer) {
718 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
719 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
722 if (!iter->is_tunnel) {
726 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
727 namebuf, sizeof namebuf);
728 node = simap_find(&tmp_backers, dp_port);
730 simap_put(&backer->tnl_backers, dp_port, node->data);
731 simap_delete(&tmp_backers, node);
732 node = simap_find(&backer->tnl_backers, dp_port);
734 node = simap_find(&backer->tnl_backers, dp_port);
736 odp_port_t odp_port = ODPP_NONE;
738 if (!dpif_port_add(backer->dpif, iter->up.netdev,
740 simap_put(&backer->tnl_backers, dp_port,
741 odp_to_u32(odp_port));
742 node = simap_find(&backer->tnl_backers, dp_port);
747 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
748 if (tnl_port_reconfigure(iter, iter->up.netdev,
750 backer->need_revalidate = REV_RECONFIGURE;
755 SIMAP_FOR_EACH (node, &tmp_backers) {
756 dpif_port_del(backer->dpif, u32_to_odp(node->data));
758 simap_destroy(&tmp_backers);
760 switch (backer->need_revalidate) {
761 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
762 case REV_STP: COVERAGE_INC(rev_stp); break;
763 case REV_BOND: COVERAGE_INC(rev_bond); break;
764 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
765 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
766 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
767 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
769 backer->need_revalidate = 0;
771 /* Clear the drop_keys in case we should now be accepting some
772 * formerly dropped flows. */
773 drop_key_clear(backer);
775 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
776 struct facet *facet, *next;
777 struct ofport_dpif *ofport;
778 struct cls_cursor cursor;
779 struct ofbundle *bundle;
781 if (ofproto->backer != backer) {
785 xlate_ofproto_set(ofproto, ofproto->up.name,
786 ofproto->backer->dpif, ofproto->ml,
787 ofproto->stp, ofproto->mbridge,
788 ofproto->sflow, ofproto->ipfix,
789 ofproto->up.frag_handling,
790 ofproto->up.forward_bpdu,
791 connmgr_has_in_band(ofproto->up.connmgr),
792 ofproto->netflow != NULL);
794 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
795 xlate_bundle_set(ofproto, bundle, bundle->name,
796 bundle->vlan_mode, bundle->vlan,
797 bundle->trunks, bundle->use_priority_tags,
798 bundle->bond, bundle->lacp,
802 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
803 int stp_port = ofport->stp_port
804 ? stp_port_no(ofport->stp_port)
806 xlate_ofport_set(ofproto, ofport->bundle, ofport,
807 ofport->up.ofp_port, ofport->odp_port,
808 ofport->up.netdev, ofport->cfm,
809 ofport->bfd, ofport->peer, stp_port,
810 ofport->qdscp, ofport->n_qdscp,
811 ofport->up.pp.config, ofport->is_tunnel,
815 cls_cursor_init(&cursor, &ofproto->facets, NULL);
816 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
817 facet_revalidate(facet);
823 if (!backer->recv_set_enable) {
824 /* Wake up before a max of 1000ms. */
825 timer_set_duration(&backer->next_expiration, 1000);
826 } else if (timer_expired(&backer->next_expiration)) {
827 int delay = expire(backer);
828 timer_set_duration(&backer->next_expiration, delay);
831 process_dpif_port_changes(backer);
833 if (backer->governor) {
836 governor_run(backer->governor);
838 /* If the governor has shrunk to its minimum size and the number of
839 * subfacets has dwindled, then drop the governor entirely.
841 * For hysteresis, the number of subfacets to drop the governor is
842 * smaller than the number needed to trigger its creation. */
843 n_subfacets = hmap_count(&backer->subfacets);
844 if (n_subfacets * 4 < flow_eviction_threshold
845 && governor_is_idle(backer->governor)) {
846 governor_destroy(backer->governor);
847 backer->governor = NULL;
854 /* Check for and handle port changes in 'backer''s dpif. */
856 process_dpif_port_changes(struct dpif_backer *backer)
862 error = dpif_port_poll(backer->dpif, &devname);
868 process_dpif_all_ports_changed(backer);
872 process_dpif_port_change(backer, devname);
877 process_dpif_port_error(backer, error);
884 process_dpif_all_ports_changed(struct dpif_backer *backer)
886 struct ofproto_dpif *ofproto;
887 struct dpif_port dpif_port;
888 struct dpif_port_dump dump;
889 struct sset devnames;
892 sset_init(&devnames);
893 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
894 if (ofproto->backer == backer) {
895 struct ofport *ofport;
897 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
898 sset_add(&devnames, netdev_get_name(ofport->netdev));
902 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
903 sset_add(&devnames, dpif_port.name);
906 SSET_FOR_EACH (devname, &devnames) {
907 process_dpif_port_change(backer, devname);
909 sset_destroy(&devnames);
913 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
915 struct ofproto_dpif *ofproto;
916 struct dpif_port port;
918 /* Don't report on the datapath's device. */
919 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
923 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
924 &all_ofproto_dpifs) {
925 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
930 ofproto = lookup_ofproto_dpif_by_port_name(devname);
931 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
932 /* The port was removed. If we know the datapath,
933 * report it through poll_set(). If we don't, it may be
934 * notifying us of a removal we initiated, so ignore it.
935 * If there's a pending ENOBUFS, let it stand, since
936 * everything will be reevaluated. */
937 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
938 sset_add(&ofproto->port_poll_set, devname);
939 ofproto->port_poll_errno = 0;
941 } else if (!ofproto) {
942 /* The port was added, but we don't know with which
943 * ofproto we should associate it. Delete it. */
944 dpif_port_del(backer->dpif, port.port_no);
946 struct ofport_dpif *ofport;
948 ofport = ofport_dpif_cast(shash_find_data(
949 &ofproto->up.port_by_name, devname));
951 && ofport->odp_port != port.port_no
952 && !odp_port_to_ofport(backer, port.port_no))
954 /* 'ofport''s datapath port number has changed from
955 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
956 * structures to match. */
957 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
958 ofport->odp_port = port.port_no;
959 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
960 hash_odp_port(port.port_no));
961 backer->need_revalidate = REV_RECONFIGURE;
964 dpif_port_destroy(&port);
967 /* Propagate 'error' to all ofprotos based on 'backer'. */
969 process_dpif_port_error(struct dpif_backer *backer, int error)
971 struct ofproto_dpif *ofproto;
973 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
974 if (ofproto->backer == backer) {
975 sset_clear(&ofproto->port_poll_set);
976 ofproto->port_poll_errno = error;
982 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
986 /* If recv_set_enable is false, we should not handle upcalls. */
987 if (!backer->recv_set_enable) {
991 /* Handle one or more batches of upcalls, until there's nothing left to do
992 * or until we do a fixed total amount of work.
994 * We do work in batches because it can be much cheaper to set up a number
995 * of flows and fire off their patches all at once. We do multiple batches
996 * because in some cases handling a packet can cause another packet to be
997 * queued almost immediately as part of the return flow. Both
998 * optimizations can make major improvements on some benchmarks and
999 * presumably for real traffic as well. */
1001 while (work < max_batch) {
1002 int retval = handle_upcalls(backer, max_batch - work);
1013 type_run_fast(const char *type)
1015 struct dpif_backer *backer;
1017 backer = shash_find_data(&all_dpif_backers, type);
1019 /* This is not necessarily a problem, since backers are only
1020 * created on demand. */
1024 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1030 static long long int port_rl = LLONG_MIN;
1031 static unsigned int backer_rl = 0;
1033 if (time_msec() >= port_rl) {
1034 struct ofproto_dpif *ofproto;
1036 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1037 run_fast(&ofproto->up);
1039 port_rl = time_msec() + 200;
1042 /* XXX: We have to be careful not to do too much work in this function. If
1043 * we call dpif_backer_run_fast() too often, or with too large a batch,
1044 * performance improves signifcantly, but at a cost. It's possible for the
1045 * number of flows in the datapath to increase without bound, and for poll
1046 * loops to take 10s of seconds. The correct solution to this problem,
1047 * long term, is to separate flow miss handling into it's own thread so it
1048 * isn't affected by revalidations, and expirations. Until then, this is
1049 * the best we can do. */
1050 if (++backer_rl >= 10) {
1051 struct shash_node *node;
1054 SHASH_FOR_EACH (node, &all_dpif_backers) {
1055 dpif_backer_run_fast(node->data, 1);
1061 type_wait(const char *type)
1063 struct dpif_backer *backer;
1065 backer = shash_find_data(&all_dpif_backers, type);
1067 /* This is not necessarily a problem, since backers are only
1068 * created on demand. */
1072 if (backer->governor) {
1073 governor_wait(backer->governor);
1076 timer_wait(&backer->next_expiration);
1079 /* Basic life-cycle. */
1081 static int add_internal_flows(struct ofproto_dpif *);
1083 static struct ofproto *
1086 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1087 return &ofproto->up;
1091 dealloc(struct ofproto *ofproto_)
1093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1098 close_dpif_backer(struct dpif_backer *backer)
1100 struct shash_node *node;
1102 ovs_assert(backer->refcount > 0);
1104 if (--backer->refcount) {
1108 drop_key_clear(backer);
1109 hmap_destroy(&backer->drop_keys);
1111 simap_destroy(&backer->tnl_backers);
1112 hmap_destroy(&backer->odp_to_ofport_map);
1113 node = shash_find(&all_dpif_backers, backer->type);
1115 shash_delete(&all_dpif_backers, node);
1116 dpif_close(backer->dpif);
1118 ovs_assert(hmap_is_empty(&backer->subfacets));
1119 hmap_destroy(&backer->subfacets);
1120 governor_destroy(backer->governor);
1125 /* Datapath port slated for removal from datapath. */
1126 struct odp_garbage {
1127 struct list list_node;
1128 odp_port_t odp_port;
1132 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1134 struct dpif_backer *backer;
1135 struct dpif_port_dump port_dump;
1136 struct dpif_port port;
1137 struct shash_node *node;
1138 struct list garbage_list;
1139 struct odp_garbage *garbage, *next;
1145 backer = shash_find_data(&all_dpif_backers, type);
1152 backer_name = xasprintf("ovs-%s", type);
1154 /* Remove any existing datapaths, since we assume we're the only
1155 * userspace controlling the datapath. */
1157 dp_enumerate_names(type, &names);
1158 SSET_FOR_EACH(name, &names) {
1159 struct dpif *old_dpif;
1161 /* Don't remove our backer if it exists. */
1162 if (!strcmp(name, backer_name)) {
1166 if (dpif_open(name, type, &old_dpif)) {
1167 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1169 dpif_delete(old_dpif);
1170 dpif_close(old_dpif);
1173 sset_destroy(&names);
1175 backer = xmalloc(sizeof *backer);
1177 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1180 VLOG_ERR("failed to open datapath of type %s: %s", type,
1181 ovs_strerror(error));
1186 backer->type = xstrdup(type);
1187 backer->governor = NULL;
1188 backer->refcount = 1;
1189 hmap_init(&backer->odp_to_ofport_map);
1190 hmap_init(&backer->drop_keys);
1191 hmap_init(&backer->subfacets);
1192 timer_set_duration(&backer->next_expiration, 1000);
1193 backer->need_revalidate = 0;
1194 simap_init(&backer->tnl_backers);
1195 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1198 if (backer->recv_set_enable) {
1199 dpif_flow_flush(backer->dpif);
1202 /* Loop through the ports already on the datapath and remove any
1203 * that we don't need anymore. */
1204 list_init(&garbage_list);
1205 dpif_port_dump_start(&port_dump, backer->dpif);
1206 while (dpif_port_dump_next(&port_dump, &port)) {
1207 node = shash_find(&init_ofp_ports, port.name);
1208 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1209 garbage = xmalloc(sizeof *garbage);
1210 garbage->odp_port = port.port_no;
1211 list_push_front(&garbage_list, &garbage->list_node);
1214 dpif_port_dump_done(&port_dump);
1216 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1217 dpif_port_del(backer->dpif, garbage->odp_port);
1218 list_remove(&garbage->list_node);
1222 shash_add(&all_dpif_backers, type, backer);
1224 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1226 VLOG_ERR("failed to listen on datapath of type %s: %s",
1227 type, ovs_strerror(error));
1228 close_dpif_backer(backer);
1232 backer->max_n_subfacet = 0;
1233 backer->created = time_msec();
1234 backer->last_minute = backer->created;
1235 memset(&backer->hourly, 0, sizeof backer->hourly);
1236 memset(&backer->daily, 0, sizeof backer->daily);
1237 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1238 backer->subfacet_add_count = 0;
1239 backer->subfacet_del_count = 0;
1240 backer->total_subfacet_add_count = 0;
1241 backer->total_subfacet_del_count = 0;
1242 backer->avg_n_subfacet = 0;
1243 backer->avg_subfacet_life = 0;
1249 construct(struct ofproto *ofproto_)
1251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1252 struct shash_node *node, *next;
1253 odp_port_t max_ports;
1256 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1261 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1262 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1263 ofp_to_u16(OFPP_MAX))));
1265 ofproto->netflow = NULL;
1266 ofproto->sflow = NULL;
1267 ofproto->ipfix = NULL;
1268 ofproto->stp = NULL;
1269 hmap_init(&ofproto->bundles);
1270 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1271 ofproto->mbridge = mbridge_create();
1272 ofproto->has_bonded_bundles = false;
1274 classifier_init(&ofproto->facets);
1275 ofproto->consistency_rl = LLONG_MIN;
1277 list_init(&ofproto->completions);
1279 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1280 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1281 list_init(&ofproto->flow_mods);
1282 ofproto->n_flow_mods = 0;
1283 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1285 ofproto_dpif_unixctl_init();
1287 hmap_init(&ofproto->vlandev_map);
1288 hmap_init(&ofproto->realdev_vid_map);
1290 sset_init(&ofproto->ports);
1291 sset_init(&ofproto->ghost_ports);
1292 sset_init(&ofproto->port_poll_set);
1293 ofproto->port_poll_errno = 0;
1295 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1296 struct iface_hint *iface_hint = node->data;
1298 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1299 /* Check if the datapath already has this port. */
1300 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1301 sset_add(&ofproto->ports, node->name);
1304 free(iface_hint->br_name);
1305 free(iface_hint->br_type);
1307 shash_delete(&init_ofp_ports, node);
1311 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1312 hash_string(ofproto->up.name, 0));
1313 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1315 ofproto_init_tables(ofproto_, N_TABLES);
1316 error = add_internal_flows(ofproto);
1317 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1320 ofproto->n_missed = 0;
1326 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1327 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1329 struct ofputil_flow_mod fm;
1332 match_init_catchall(&fm.match);
1334 match_set_reg(&fm.match, 0, id);
1335 fm.new_cookie = htonll(0);
1336 fm.cookie = htonll(0);
1337 fm.cookie_mask = htonll(0);
1338 fm.modify_cookie = false;
1339 fm.table_id = TBL_INTERNAL;
1340 fm.command = OFPFC_ADD;
1341 fm.idle_timeout = 0;
1342 fm.hard_timeout = 0;
1346 fm.ofpacts = ofpacts->data;
1347 fm.ofpacts_len = ofpacts->size;
1349 error = ofproto_flow_mod(&ofproto->up, &fm);
1351 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1352 id, ofperr_to_string(error));
1356 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1358 ovs_assert(*rulep != NULL);
1364 add_internal_flows(struct ofproto_dpif *ofproto)
1366 struct ofpact_controller *controller;
1367 uint64_t ofpacts_stub[128 / 8];
1368 struct ofpbuf ofpacts;
1372 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1375 controller = ofpact_put_CONTROLLER(&ofpacts);
1376 controller->max_len = UINT16_MAX;
1377 controller->controller_id = 0;
1378 controller->reason = OFPR_NO_MATCH;
1379 ofpact_pad(&ofpacts);
1381 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1386 ofpbuf_clear(&ofpacts);
1387 error = add_internal_flow(ofproto, id++, &ofpacts,
1388 &ofproto->no_packet_in_rule);
1393 error = add_internal_flow(ofproto, id++, &ofpacts,
1394 &ofproto->drop_frags_rule);
1399 complete_operations(struct ofproto_dpif *ofproto)
1401 struct dpif_completion *c, *next;
1403 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1404 ofoperation_complete(c->op, 0);
1405 list_remove(&c->list_node);
1411 destruct(struct ofproto *ofproto_)
1413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1414 struct rule_dpif *rule, *next_rule;
1415 struct ofputil_flow_mod *fm, *next_fm;
1416 struct oftable *table;
1418 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1419 xlate_remove_ofproto(ofproto);
1421 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1422 complete_operations(ofproto);
1424 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1425 struct cls_cursor cursor;
1427 cls_cursor_init(&cursor, &table->cls, NULL);
1428 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1429 ofproto_rule_destroy(&rule->up);
1433 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1434 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1435 list_remove(&fm->list_node);
1436 ofproto->n_flow_mods--;
1440 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1441 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1443 mbridge_unref(ofproto->mbridge);
1445 netflow_destroy(ofproto->netflow);
1446 dpif_sflow_unref(ofproto->sflow);
1447 hmap_destroy(&ofproto->bundles);
1448 mac_learning_unref(ofproto->ml);
1450 classifier_destroy(&ofproto->facets);
1452 hmap_destroy(&ofproto->vlandev_map);
1453 hmap_destroy(&ofproto->realdev_vid_map);
1455 sset_destroy(&ofproto->ports);
1456 sset_destroy(&ofproto->ghost_ports);
1457 sset_destroy(&ofproto->port_poll_set);
1459 close_dpif_backer(ofproto->backer);
1463 run_fast(struct ofproto *ofproto_)
1465 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1466 struct ofputil_flow_mod *fm, *next;
1467 struct ofport_dpif *ofport;
1468 struct list flow_mods;
1470 /* Do not perform any periodic activity required by 'ofproto' while
1471 * waiting for flow restore to complete. */
1472 if (ofproto_get_flow_restore_wait()) {
1476 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1477 if (ofproto->n_flow_mods) {
1478 flow_mods = ofproto->flow_mods;
1479 list_moved(&flow_mods);
1480 list_init(&ofproto->flow_mods);
1481 ofproto->n_flow_mods = 0;
1483 list_init(&flow_mods);
1485 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1487 LIST_FOR_EACH_SAFE (fm, next, list_node, &flow_mods) {
1488 int error = ofproto_flow_mod(&ofproto->up, fm);
1489 if (error && !VLOG_DROP_WARN(&rl)) {
1490 VLOG_WARN("learning action failed to modify flow table (%s)",
1491 ofperr_get_name(error));
1494 list_remove(&fm->list_node);
1499 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1500 port_run_fast(ofport);
1507 run(struct ofproto *ofproto_)
1509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1510 struct ofport_dpif *ofport;
1511 struct ofbundle *bundle;
1515 complete_operations(ofproto);
1518 if (mbridge_need_revalidate(ofproto->mbridge)) {
1519 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1520 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1521 mac_learning_flush(ofproto->ml);
1522 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1525 /* Do not perform any periodic activity below required by 'ofproto' while
1526 * waiting for flow restore to complete. */
1527 if (ofproto_get_flow_restore_wait()) {
1531 error = run_fast(ofproto_);
1536 if (ofproto->netflow) {
1537 if (netflow_run(ofproto->netflow)) {
1538 send_netflow_active_timeouts(ofproto);
1541 if (ofproto->sflow) {
1542 dpif_sflow_run(ofproto->sflow);
1545 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1548 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1553 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1554 if (mac_learning_run(ofproto->ml)) {
1555 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1557 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1559 /* Check the consistency of a random facet, to aid debugging. */
1560 if (time_msec() >= ofproto->consistency_rl
1561 && !classifier_is_empty(&ofproto->facets)
1562 && !ofproto->backer->need_revalidate) {
1563 struct cls_table *table;
1564 struct cls_rule *cr;
1565 struct facet *facet;
1567 ofproto->consistency_rl = time_msec() + 250;
1569 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1570 struct cls_table, hmap_node);
1571 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1573 facet = CONTAINER_OF(cr, struct facet, cr);
1575 if (!facet_check_consistency(facet)) {
1576 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1584 wait(struct ofproto *ofproto_)
1586 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1587 struct ofport_dpif *ofport;
1588 struct ofbundle *bundle;
1590 if (!clogged && !list_is_empty(&ofproto->completions)) {
1591 poll_immediate_wake();
1594 if (ofproto_get_flow_restore_wait()) {
1598 dpif_wait(ofproto->backer->dpif);
1599 dpif_recv_wait(ofproto->backer->dpif);
1600 if (ofproto->sflow) {
1601 dpif_sflow_wait(ofproto->sflow);
1603 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1606 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1607 bundle_wait(bundle);
1609 if (ofproto->netflow) {
1610 netflow_wait(ofproto->netflow);
1612 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1613 mac_learning_wait(ofproto->ml);
1614 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1616 if (ofproto->backer->need_revalidate) {
1617 /* Shouldn't happen, but if it does just go around again. */
1618 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1619 poll_immediate_wake();
1624 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1626 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1627 struct cls_cursor cursor;
1628 size_t n_subfacets = 0;
1629 struct facet *facet;
1631 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1633 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1634 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1635 n_subfacets += list_size(&facet->subfacets);
1637 simap_increase(usage, "subfacets", n_subfacets);
1641 flush(struct ofproto *ofproto_)
1643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1644 struct subfacet *subfacet, *next_subfacet;
1645 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1649 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1650 &ofproto->backer->subfacets) {
1651 if (subfacet->facet->ofproto != ofproto) {
1655 if (subfacet->path != SF_NOT_INSTALLED) {
1656 batch[n_batch++] = subfacet;
1657 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1658 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1662 subfacet_destroy(subfacet);
1667 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1672 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1673 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1675 *arp_match_ip = true;
1676 *actions = (OFPUTIL_A_OUTPUT |
1677 OFPUTIL_A_SET_VLAN_VID |
1678 OFPUTIL_A_SET_VLAN_PCP |
1679 OFPUTIL_A_STRIP_VLAN |
1680 OFPUTIL_A_SET_DL_SRC |
1681 OFPUTIL_A_SET_DL_DST |
1682 OFPUTIL_A_SET_NW_SRC |
1683 OFPUTIL_A_SET_NW_DST |
1684 OFPUTIL_A_SET_NW_TOS |
1685 OFPUTIL_A_SET_TP_SRC |
1686 OFPUTIL_A_SET_TP_DST |
1691 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1694 struct dpif_dp_stats s;
1695 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1698 strcpy(ots->name, "classifier");
1700 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1701 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1702 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1703 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1705 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1706 ots->lookup_count = htonll(n_lookup);
1707 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1710 static struct ofport *
1713 struct ofport_dpif *port = xmalloc(sizeof *port);
1718 port_dealloc(struct ofport *port_)
1720 struct ofport_dpif *port = ofport_dpif_cast(port_);
1725 port_construct(struct ofport *port_)
1727 struct ofport_dpif *port = ofport_dpif_cast(port_);
1728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1729 const struct netdev *netdev = port->up.netdev;
1730 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1731 struct dpif_port dpif_port;
1734 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1735 port->bundle = NULL;
1738 port->may_enable = true;
1739 port->stp_port = NULL;
1740 port->stp_state = STP_DISABLED;
1741 port->is_tunnel = false;
1745 port->realdev_ofp_port = 0;
1746 port->vlandev_vid = 0;
1747 port->carrier_seq = netdev_get_carrier_resets(netdev);
1749 if (netdev_vport_is_patch(netdev)) {
1750 /* By bailing out here, we don't submit the port to the sFlow module
1751 * to be considered for counter polling export. This is correct
1752 * because the patch port represents an interface that sFlow considers
1753 * to be "internal" to the switch as a whole, and therefore not an
1754 * candidate for counter polling. */
1755 port->odp_port = ODPP_NONE;
1756 ofport_update_peer(port);
1760 error = dpif_port_query_by_name(ofproto->backer->dpif,
1761 netdev_vport_get_dpif_port(netdev, namebuf,
1768 port->odp_port = dpif_port.port_no;
1770 if (netdev_get_tunnel_config(netdev)) {
1771 tnl_port_add(port, port->up.netdev, port->odp_port);
1772 port->is_tunnel = true;
1774 /* Sanity-check that a mapping doesn't already exist. This
1775 * shouldn't happen for non-tunnel ports. */
1776 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1777 VLOG_ERR("port %s already has an OpenFlow port number",
1779 dpif_port_destroy(&dpif_port);
1783 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1784 hash_odp_port(port->odp_port));
1786 dpif_port_destroy(&dpif_port);
1788 if (ofproto->sflow) {
1789 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1796 port_destruct(struct ofport *port_)
1798 struct ofport_dpif *port = ofport_dpif_cast(port_);
1799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1800 const char *devname = netdev_get_name(port->up.netdev);
1801 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1802 const char *dp_port_name;
1804 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1805 xlate_ofport_remove(port);
1807 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1809 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1810 /* The underlying device is still there, so delete it. This
1811 * happens when the ofproto is being destroyed, since the caller
1812 * assumes that removal of attached ports will happen as part of
1814 if (!port->is_tunnel) {
1815 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1820 port->peer->peer = NULL;
1824 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1825 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1829 sset_find_and_delete(&ofproto->ports, devname);
1830 sset_find_and_delete(&ofproto->ghost_ports, devname);
1831 bundle_remove(port_);
1832 set_cfm(port_, NULL);
1833 set_bfd(port_, NULL);
1834 if (ofproto->sflow) {
1835 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1842 port_modified(struct ofport *port_)
1844 struct ofport_dpif *port = ofport_dpif_cast(port_);
1846 if (port->bundle && port->bundle->bond) {
1847 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1851 cfm_set_netdev(port->cfm, port->up.netdev);
1854 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1856 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1860 ofport_update_peer(port);
1864 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1866 struct ofport_dpif *port = ofport_dpif_cast(port_);
1867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1868 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1870 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1871 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1872 OFPUTIL_PC_NO_PACKET_IN)) {
1873 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1875 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1876 bundle_update(port->bundle);
1882 set_sflow(struct ofproto *ofproto_,
1883 const struct ofproto_sflow_options *sflow_options)
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1886 struct dpif_sflow *ds = ofproto->sflow;
1888 if (sflow_options) {
1890 struct ofport_dpif *ofport;
1892 ds = ofproto->sflow = dpif_sflow_create();
1893 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1894 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1896 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1898 dpif_sflow_set_options(ds, sflow_options);
1901 dpif_sflow_unref(ds);
1902 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1903 ofproto->sflow = NULL;
1911 struct ofproto *ofproto_,
1912 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1913 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1914 size_t n_flow_exporters_options)
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1917 struct dpif_ipfix *di = ofproto->ipfix;
1919 if (bridge_exporter_options || flow_exporters_options) {
1921 di = ofproto->ipfix = dpif_ipfix_create();
1923 dpif_ipfix_set_options(
1924 di, bridge_exporter_options, flow_exporters_options,
1925 n_flow_exporters_options);
1928 dpif_ipfix_unref(di);
1929 ofproto->ipfix = NULL;
1936 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1938 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1945 struct ofproto_dpif *ofproto;
1947 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1948 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949 ofport->cfm = cfm_create(ofport->up.netdev);
1952 if (cfm_configure(ofport->cfm, s)) {
1958 cfm_unref(ofport->cfm);
1964 get_cfm_status(const struct ofport *ofport_,
1965 struct ofproto_cfm_status *status)
1967 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1970 status->faults = cfm_get_fault(ofport->cfm);
1971 status->remote_opstate = cfm_get_opup(ofport->cfm);
1972 status->health = cfm_get_health(ofport->cfm);
1973 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1981 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1988 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1989 if (ofport->bfd != old) {
1990 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1997 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1999 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2002 bfd_get_status(ofport->bfd, smap);
2009 /* Spanning Tree. */
2012 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2014 struct ofproto_dpif *ofproto = ofproto_;
2015 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2016 struct ofport_dpif *ofport;
2018 ofport = stp_port_get_aux(sp);
2020 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2021 ofproto->up.name, port_num);
2023 struct eth_header *eth = pkt->l2;
2025 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2026 if (eth_addr_is_zero(eth->eth_src)) {
2027 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2028 "with unknown MAC", ofproto->up.name, port_num);
2030 send_packet(ofport, pkt);
2036 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2038 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2042 /* Only revalidate flows if the configuration changed. */
2043 if (!s != !ofproto->stp) {
2044 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2048 if (!ofproto->stp) {
2049 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2050 send_bpdu_cb, ofproto);
2051 ofproto->stp_last_tick = time_msec();
2054 stp_set_bridge_id(ofproto->stp, s->system_id);
2055 stp_set_bridge_priority(ofproto->stp, s->priority);
2056 stp_set_hello_time(ofproto->stp, s->hello_time);
2057 stp_set_max_age(ofproto->stp, s->max_age);
2058 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2060 struct ofport *ofport;
2062 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2063 set_stp_port(ofport, NULL);
2066 stp_unref(ofproto->stp);
2067 ofproto->stp = NULL;
2074 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2080 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2081 s->designated_root = stp_get_designated_root(ofproto->stp);
2082 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2091 update_stp_port_state(struct ofport_dpif *ofport)
2093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2094 enum stp_state state;
2096 /* Figure out new state. */
2097 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2101 if (ofport->stp_state != state) {
2102 enum ofputil_port_state of_state;
2105 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2106 netdev_get_name(ofport->up.netdev),
2107 stp_state_name(ofport->stp_state),
2108 stp_state_name(state));
2109 if (stp_learn_in_state(ofport->stp_state)
2110 != stp_learn_in_state(state)) {
2111 /* xxx Learning action flows should also be flushed. */
2112 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2113 mac_learning_flush(ofproto->ml);
2114 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2116 fwd_change = stp_forward_in_state(ofport->stp_state)
2117 != stp_forward_in_state(state);
2119 ofproto->backer->need_revalidate = REV_STP;
2120 ofport->stp_state = state;
2121 ofport->stp_state_entered = time_msec();
2123 if (fwd_change && ofport->bundle) {
2124 bundle_update(ofport->bundle);
2127 /* Update the STP state bits in the OpenFlow port description. */
2128 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2129 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2130 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2131 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2132 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2134 ofproto_port_set_state(&ofport->up, of_state);
2138 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2139 * caller is responsible for assigning STP port numbers and ensuring
2140 * there are no duplicates. */
2142 set_stp_port(struct ofport *ofport_,
2143 const struct ofproto_port_stp_settings *s)
2145 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2147 struct stp_port *sp = ofport->stp_port;
2149 if (!s || !s->enable) {
2151 ofport->stp_port = NULL;
2152 stp_port_disable(sp);
2153 update_stp_port_state(ofport);
2156 } else if (sp && stp_port_no(sp) != s->port_num
2157 && ofport == stp_port_get_aux(sp)) {
2158 /* The port-id changed, so disable the old one if it's not
2159 * already in use by another port. */
2160 stp_port_disable(sp);
2163 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2164 stp_port_enable(sp);
2166 stp_port_set_aux(sp, ofport);
2167 stp_port_set_priority(sp, s->priority);
2168 stp_port_set_path_cost(sp, s->path_cost);
2170 update_stp_port_state(ofport);
2176 get_stp_port_status(struct ofport *ofport_,
2177 struct ofproto_port_stp_status *s)
2179 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2180 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2181 struct stp_port *sp = ofport->stp_port;
2183 if (!ofproto->stp || !sp) {
2189 s->port_id = stp_port_get_id(sp);
2190 s->state = stp_port_get_state(sp);
2191 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2192 s->role = stp_port_get_role(sp);
2193 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2199 stp_run(struct ofproto_dpif *ofproto)
2202 long long int now = time_msec();
2203 long long int elapsed = now - ofproto->stp_last_tick;
2204 struct stp_port *sp;
2207 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2208 ofproto->stp_last_tick = now;
2210 while (stp_get_changed_port(ofproto->stp, &sp)) {
2211 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2214 update_stp_port_state(ofport);
2218 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2219 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2220 mac_learning_flush(ofproto->ml);
2221 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2227 stp_wait(struct ofproto_dpif *ofproto)
2230 poll_timer_wait(1000);
2235 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2238 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2239 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2241 if (ofport->n_qdscp != n_qdscp
2242 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2243 n_qdscp * sizeof *qdscp))) {
2244 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2245 free(ofport->qdscp);
2246 ofport->qdscp = n_qdscp
2247 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2249 ofport->n_qdscp = n_qdscp;
2257 /* Expires all MAC learning entries associated with 'bundle' and forces its
2258 * ofproto to revalidate every flow.
2260 * Normally MAC learning entries are removed only from the ofproto associated
2261 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2262 * are removed from every ofproto. When patch ports and SLB bonds are in use
2263 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2264 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2265 * with the host from which it migrated. */
2267 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2269 struct ofproto_dpif *ofproto = bundle->ofproto;
2270 struct mac_learning *ml = ofproto->ml;
2271 struct mac_entry *mac, *next_mac;
2273 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2274 ovs_rwlock_wrlock(&ml->rwlock);
2275 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2276 if (mac->port.p == bundle) {
2278 struct ofproto_dpif *o;
2280 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2282 struct mac_entry *e;
2284 ovs_rwlock_wrlock(&o->ml->rwlock);
2285 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2287 mac_learning_expire(o->ml, e);
2289 ovs_rwlock_unlock(&o->ml->rwlock);
2294 mac_learning_expire(ml, mac);
2297 ovs_rwlock_unlock(&ml->rwlock);
2300 static struct ofbundle *
2301 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2303 struct ofbundle *bundle;
2305 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2306 &ofproto->bundles) {
2307 if (bundle->aux == aux) {
2315 bundle_update(struct ofbundle *bundle)
2317 struct ofport_dpif *port;
2319 bundle->floodable = true;
2320 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2321 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2322 || !stp_forward_in_state(port->stp_state)) {
2323 bundle->floodable = false;
2330 bundle_del_port(struct ofport_dpif *port)
2332 struct ofbundle *bundle = port->bundle;
2334 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2336 list_remove(&port->bundle_node);
2337 port->bundle = NULL;
2340 lacp_slave_unregister(bundle->lacp, port);
2343 bond_slave_unregister(bundle->bond, port);
2346 bundle_update(bundle);
2350 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2351 struct lacp_slave_settings *lacp)
2353 struct ofport_dpif *port;
2355 port = get_ofp_port(bundle->ofproto, ofp_port);
2360 if (port->bundle != bundle) {
2361 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2363 bundle_del_port(port);
2366 port->bundle = bundle;
2367 list_push_back(&bundle->ports, &port->bundle_node);
2368 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2369 || !stp_forward_in_state(port->stp_state)) {
2370 bundle->floodable = false;
2374 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2375 lacp_slave_register(bundle->lacp, port, lacp);
2382 bundle_destroy(struct ofbundle *bundle)
2384 struct ofproto_dpif *ofproto;
2385 struct ofport_dpif *port, *next_port;
2391 ofproto = bundle->ofproto;
2392 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2394 xlate_bundle_remove(bundle);
2396 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2397 bundle_del_port(port);
2400 bundle_flush_macs(bundle, true);
2401 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2403 free(bundle->trunks);
2404 lacp_unref(bundle->lacp);
2405 bond_unref(bundle->bond);
2410 bundle_set(struct ofproto *ofproto_, void *aux,
2411 const struct ofproto_bundle_settings *s)
2413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2414 bool need_flush = false;
2415 struct ofport_dpif *port;
2416 struct ofbundle *bundle;
2417 unsigned long *trunks;
2423 bundle_destroy(bundle_lookup(ofproto, aux));
2427 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2428 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2430 bundle = bundle_lookup(ofproto, aux);
2432 bundle = xmalloc(sizeof *bundle);
2434 bundle->ofproto = ofproto;
2435 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2436 hash_pointer(aux, 0));
2438 bundle->name = NULL;
2440 list_init(&bundle->ports);
2441 bundle->vlan_mode = PORT_VLAN_TRUNK;
2443 bundle->trunks = NULL;
2444 bundle->use_priority_tags = s->use_priority_tags;
2445 bundle->lacp = NULL;
2446 bundle->bond = NULL;
2448 bundle->floodable = true;
2449 mbridge_register_bundle(ofproto->mbridge, bundle);
2452 if (!bundle->name || strcmp(s->name, bundle->name)) {
2454 bundle->name = xstrdup(s->name);
2459 if (!bundle->lacp) {
2460 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2461 bundle->lacp = lacp_create();
2463 lacp_configure(bundle->lacp, s->lacp);
2465 lacp_unref(bundle->lacp);
2466 bundle->lacp = NULL;
2469 /* Update set of ports. */
2471 for (i = 0; i < s->n_slaves; i++) {
2472 if (!bundle_add_port(bundle, s->slaves[i],
2473 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2477 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2478 struct ofport_dpif *next_port;
2480 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2481 for (i = 0; i < s->n_slaves; i++) {
2482 if (s->slaves[i] == port->up.ofp_port) {
2487 bundle_del_port(port);
2491 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2493 if (list_is_empty(&bundle->ports)) {
2494 bundle_destroy(bundle);
2498 /* Set VLAN tagging mode */
2499 if (s->vlan_mode != bundle->vlan_mode
2500 || s->use_priority_tags != bundle->use_priority_tags) {
2501 bundle->vlan_mode = s->vlan_mode;
2502 bundle->use_priority_tags = s->use_priority_tags;
2507 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2508 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2510 if (vlan != bundle->vlan) {
2511 bundle->vlan = vlan;
2515 /* Get trunked VLANs. */
2516 switch (s->vlan_mode) {
2517 case PORT_VLAN_ACCESS:
2521 case PORT_VLAN_TRUNK:
2522 trunks = CONST_CAST(unsigned long *, s->trunks);
2525 case PORT_VLAN_NATIVE_UNTAGGED:
2526 case PORT_VLAN_NATIVE_TAGGED:
2527 if (vlan != 0 && (!s->trunks
2528 || !bitmap_is_set(s->trunks, vlan)
2529 || bitmap_is_set(s->trunks, 0))) {
2530 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2532 trunks = bitmap_clone(s->trunks, 4096);
2534 trunks = bitmap_allocate1(4096);
2536 bitmap_set1(trunks, vlan);
2537 bitmap_set0(trunks, 0);
2539 trunks = CONST_CAST(unsigned long *, s->trunks);
2546 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2547 free(bundle->trunks);
2548 if (trunks == s->trunks) {
2549 bundle->trunks = vlan_bitmap_clone(trunks);
2551 bundle->trunks = trunks;
2556 if (trunks != s->trunks) {
2561 if (!list_is_short(&bundle->ports)) {
2562 bundle->ofproto->has_bonded_bundles = true;
2564 if (bond_reconfigure(bundle->bond, s->bond)) {
2565 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2568 bundle->bond = bond_create(s->bond);
2569 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2572 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2573 bond_slave_register(bundle->bond, port, port->up.netdev);
2576 bond_unref(bundle->bond);
2577 bundle->bond = NULL;
2580 /* If we changed something that would affect MAC learning, un-learn
2581 * everything on this port and force flow revalidation. */
2583 bundle_flush_macs(bundle, false);
2590 bundle_remove(struct ofport *port_)
2592 struct ofport_dpif *port = ofport_dpif_cast(port_);
2593 struct ofbundle *bundle = port->bundle;
2596 bundle_del_port(port);
2597 if (list_is_empty(&bundle->ports)) {
2598 bundle_destroy(bundle);
2599 } else if (list_is_short(&bundle->ports)) {
2600 bond_unref(bundle->bond);
2601 bundle->bond = NULL;
2607 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2609 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2610 struct ofport_dpif *port = port_;
2611 uint8_t ea[ETH_ADDR_LEN];
2614 error = netdev_get_etheraddr(port->up.netdev, ea);
2616 struct ofpbuf packet;
2619 ofpbuf_init(&packet, 0);
2620 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2622 memcpy(packet_pdu, pdu, pdu_size);
2624 send_packet(port, &packet);
2625 ofpbuf_uninit(&packet);
2627 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2628 "%s (%s)", port->bundle->name,
2629 netdev_get_name(port->up.netdev), ovs_strerror(error));
2634 bundle_send_learning_packets(struct ofbundle *bundle)
2636 struct ofproto_dpif *ofproto = bundle->ofproto;
2637 int error, n_packets, n_errors;
2638 struct mac_entry *e;
2640 error = n_packets = n_errors = 0;
2641 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2642 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2643 if (e->port.p != bundle) {
2644 struct ofpbuf *learning_packet;
2645 struct ofport_dpif *port;
2649 /* The assignment to "port" is unnecessary but makes "grep"ing for
2650 * struct ofport_dpif more effective. */
2651 learning_packet = bond_compose_learning_packet(bundle->bond,
2655 ret = send_packet(port, learning_packet);
2656 ofpbuf_delete(learning_packet);
2664 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2667 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2668 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2669 "packets, last error was: %s",
2670 bundle->name, n_errors, n_packets, ovs_strerror(error));
2672 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2673 bundle->name, n_packets);
2678 bundle_run(struct ofbundle *bundle)
2681 lacp_run(bundle->lacp, send_pdu_cb);
2684 struct ofport_dpif *port;
2686 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2687 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2690 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2691 bundle->ofproto->backer->need_revalidate = REV_BOND;
2694 if (bond_should_send_learning_packets(bundle->bond)) {
2695 bundle_send_learning_packets(bundle);
2701 bundle_wait(struct ofbundle *bundle)
2704 lacp_wait(bundle->lacp);
2707 bond_wait(bundle->bond);
2714 mirror_set__(struct ofproto *ofproto_, void *aux,
2715 const struct ofproto_mirror_settings *s)
2717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2718 struct ofbundle **srcs, **dsts;
2723 mirror_destroy(ofproto->mbridge, aux);
2727 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2728 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2730 for (i = 0; i < s->n_srcs; i++) {
2731 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2734 for (i = 0; i < s->n_dsts; i++) {
2735 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2738 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2739 s->n_dsts, s->src_vlans,
2740 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2747 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2748 uint64_t *packets, uint64_t *bytes)
2751 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2756 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2759 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2760 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2761 mac_learning_flush(ofproto->ml);
2763 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2768 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2771 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2772 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2776 forward_bpdu_changed(struct ofproto *ofproto_)
2778 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2783 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2787 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2788 mac_learning_set_idle_time(ofproto->ml, idle_time);
2789 mac_learning_set_max_entries(ofproto->ml, max_entries);
2790 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2795 static struct ofport_dpif *
2796 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2798 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2799 return ofport ? ofport_dpif_cast(ofport) : NULL;
2802 static struct ofport_dpif *
2803 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2805 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2806 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2810 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2811 struct ofproto_port *ofproto_port,
2812 struct dpif_port *dpif_port)
2814 ofproto_port->name = dpif_port->name;
2815 ofproto_port->type = dpif_port->type;
2816 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2820 ofport_update_peer(struct ofport_dpif *ofport)
2822 const struct ofproto_dpif *ofproto;
2823 struct dpif_backer *backer;
2824 const char *peer_name;
2826 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2830 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2831 backer->need_revalidate = REV_RECONFIGURE;
2834 ofport->peer->peer = NULL;
2835 ofport->peer = NULL;
2838 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2843 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2844 struct ofport *peer_ofport;
2845 struct ofport_dpif *peer;
2846 const char *peer_peer;
2848 if (ofproto->backer != backer) {
2852 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2857 peer = ofport_dpif_cast(peer_ofport);
2858 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2859 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2861 ofport->peer = peer;
2862 ofport->peer->peer = ofport;
2870 port_run_fast(struct ofport_dpif *ofport)
2872 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2873 struct ofpbuf packet;
2875 ofpbuf_init(&packet, 0);
2876 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2877 send_packet(ofport, &packet);
2878 ofpbuf_uninit(&packet);
2881 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2882 struct ofpbuf packet;
2884 ofpbuf_init(&packet, 0);
2885 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2886 send_packet(ofport, &packet);
2887 ofpbuf_uninit(&packet);
2892 port_run(struct ofport_dpif *ofport)
2894 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2895 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2896 bool enable = netdev_get_carrier(ofport->up.netdev);
2898 ofport->carrier_seq = carrier_seq;
2900 port_run_fast(ofport);
2903 int cfm_opup = cfm_get_opup(ofport->cfm);
2905 cfm_run(ofport->cfm);
2906 enable = enable && !cfm_get_fault(ofport->cfm);
2908 if (cfm_opup >= 0) {
2909 enable = enable && cfm_opup;
2914 bfd_run(ofport->bfd);
2915 enable = enable && bfd_forwarding(ofport->bfd);
2918 if (ofport->bundle) {
2919 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2920 if (carrier_changed) {
2921 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2925 if (ofport->may_enable != enable) {
2926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2927 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2930 ofport->may_enable = enable;
2934 port_wait(struct ofport_dpif *ofport)
2937 cfm_wait(ofport->cfm);
2941 bfd_wait(ofport->bfd);
2946 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2947 struct ofproto_port *ofproto_port)
2949 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2950 struct dpif_port dpif_port;
2953 if (sset_contains(&ofproto->ghost_ports, devname)) {
2954 const char *type = netdev_get_type_from_name(devname);
2956 /* We may be called before ofproto->up.port_by_name is populated with
2957 * the appropriate ofport. For this reason, we must get the name and
2958 * type from the netdev layer directly. */
2960 const struct ofport *ofport;
2962 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2963 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2964 ofproto_port->name = xstrdup(devname);
2965 ofproto_port->type = xstrdup(type);
2971 if (!sset_contains(&ofproto->ports, devname)) {
2974 error = dpif_port_query_by_name(ofproto->backer->dpif,
2975 devname, &dpif_port);
2977 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2983 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2986 const char *devname = netdev_get_name(netdev);
2987 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2988 const char *dp_port_name;
2990 if (netdev_vport_is_patch(netdev)) {
2991 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2995 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2996 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2997 odp_port_t port_no = ODPP_NONE;
3000 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3004 if (netdev_get_tunnel_config(netdev)) {
3005 simap_put(&ofproto->backer->tnl_backers,
3006 dp_port_name, odp_to_u32(port_no));
3010 if (netdev_get_tunnel_config(netdev)) {
3011 sset_add(&ofproto->ghost_ports, devname);
3013 sset_add(&ofproto->ports, devname);
3019 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3021 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3022 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3029 sset_find_and_delete(&ofproto->ghost_ports,
3030 netdev_get_name(ofport->up.netdev));
3031 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3032 if (!ofport->is_tunnel) {
3033 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3035 /* The caller is going to close ofport->up.netdev. If this is a
3036 * bonded port, then the bond is using that netdev, so remove it
3037 * from the bond. The client will need to reconfigure everything
3038 * after deleting ports, so then the slave will get re-added. */
3039 bundle_remove(&ofport->up);
3046 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3048 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3053 error = netdev_get_stats(ofport->up.netdev, stats);
3055 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3058 /* ofproto->stats.tx_packets represents packets that we created
3059 * internally and sent to some port (e.g. packets sent with
3060 * send_packet()). Account for them as if they had come from
3061 * OFPP_LOCAL and got forwarded. */
3063 if (stats->rx_packets != UINT64_MAX) {
3064 stats->rx_packets += ofproto->stats.tx_packets;
3067 if (stats->rx_bytes != UINT64_MAX) {
3068 stats->rx_bytes += ofproto->stats.tx_bytes;
3071 /* ofproto->stats.rx_packets represents packets that were received on
3072 * some port and we processed internally and dropped (e.g. STP).
3073 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3075 if (stats->tx_packets != UINT64_MAX) {
3076 stats->tx_packets += ofproto->stats.rx_packets;
3079 if (stats->tx_bytes != UINT64_MAX) {
3080 stats->tx_bytes += ofproto->stats.rx_bytes;
3087 struct port_dump_state {
3092 struct ofproto_port port;
3097 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3099 *statep = xzalloc(sizeof(struct port_dump_state));
3104 port_dump_next(const struct ofproto *ofproto_, void *state_,
3105 struct ofproto_port *port)
3107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3108 struct port_dump_state *state = state_;
3109 const struct sset *sset;
3110 struct sset_node *node;
3112 if (state->has_port) {
3113 ofproto_port_destroy(&state->port);
3114 state->has_port = false;
3116 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3117 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3120 error = port_query_by_name(ofproto_, node->name, &state->port);
3122 *port = state->port;
3123 state->has_port = true;
3125 } else if (error != ENODEV) {
3130 if (!state->ghost) {
3131 state->ghost = true;
3134 return port_dump_next(ofproto_, state_, port);
3141 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3143 struct port_dump_state *state = state_;
3145 if (state->has_port) {
3146 ofproto_port_destroy(&state->port);
3153 port_poll(const struct ofproto *ofproto_, char **devnamep)
3155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3157 if (ofproto->port_poll_errno) {
3158 int error = ofproto->port_poll_errno;
3159 ofproto->port_poll_errno = 0;
3163 if (sset_is_empty(&ofproto->port_poll_set)) {
3167 *devnamep = sset_pop(&ofproto->port_poll_set);
3172 port_poll_wait(const struct ofproto *ofproto_)
3174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3175 dpif_port_poll_wait(ofproto->backer->dpif);
3179 port_is_lacp_current(const struct ofport *ofport_)
3181 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3182 return (ofport->bundle && ofport->bundle->lacp
3183 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3187 /* Upcall handling. */
3189 /* Flow miss batching.
3191 * Some dpifs implement operations faster when you hand them off in a batch.
3192 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3193 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3194 * more packets, plus possibly installing the flow in the dpif.
3196 * So far we only batch the operations that affect flow setup time the most.
3197 * It's possible to batch more than that, but the benefit might be minimal. */
3199 struct hmap_node hmap_node;
3200 struct ofproto_dpif *ofproto;
3202 enum odp_key_fitness key_fitness;
3203 const struct nlattr *key;
3205 struct list packets;
3206 enum dpif_upcall_type upcall_type;
3209 struct flow_miss_op {
3210 struct dpif_op dpif_op;
3212 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3213 struct xlate_out xout;
3214 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3216 struct ofpbuf mask; /* Flow mask for "put" ops. */
3217 struct odputil_keybuf maskbuf;
3219 /* If this is a "put" op, then a pointer to the subfacet that should
3220 * be marked as uninstalled if the operation fails. */
3221 struct subfacet *subfacet;
3224 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3225 * OpenFlow controller as necessary according to their individual
3226 * configurations. */
3228 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3229 const struct flow *flow)
3231 struct ofputil_packet_in pin;
3233 pin.packet = packet->data;
3234 pin.packet_len = packet->size;
3235 pin.reason = OFPR_NO_MATCH;
3236 pin.controller_id = 0;
3241 pin.send_len = 0; /* not used for flow table misses */
3243 flow_get_metadata(flow, &pin.fmd);
3245 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3248 static struct flow_miss *
3249 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3250 const struct flow *flow, uint32_t hash)
3252 struct flow_miss *miss;
3254 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3255 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3263 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3264 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3265 * 'miss' is associated with a subfacet the caller must also initialize the
3266 * returned op->subfacet, and if anything needs to be freed after processing
3267 * the op, the caller must initialize op->garbage also. */
3269 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3270 struct flow_miss_op *op)
3272 if (miss->flow.in_port.ofp_port
3273 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3274 miss->flow.vlan_tci)) {
3275 /* This packet was received on a VLAN splinter port. We
3276 * added a VLAN to the packet to make the packet resemble
3277 * the flow, but the actions were composed assuming that
3278 * the packet contained no VLAN. So, we must remove the
3279 * VLAN header from the packet before trying to execute the
3281 eth_pop_vlan(packet);
3284 op->subfacet = NULL;
3285 op->xout_garbage = false;
3286 op->dpif_op.type = DPIF_OP_EXECUTE;
3287 op->dpif_op.u.execute.key = miss->key;
3288 op->dpif_op.u.execute.key_len = miss->key_len;
3289 op->dpif_op.u.execute.packet = packet;
3290 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3293 /* Helper for handle_flow_miss_without_facet() and
3294 * handle_flow_miss_with_facet(). */
3296 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3297 const struct flow *flow, bool fail_open)
3301 * Extra-special case for fail-open mode.
3303 * We are in fail-open mode and the packet matched the fail-open
3304 * rule, but we are connected to a controller too. We should send
3305 * the packet up to the controller in the hope that it will try to
3306 * set up a flow and thereby allow us to exit fail-open.
3308 * See the top-level comment in fail-open.c for more information.
3310 send_packet_in_miss(ofproto, packet, flow);
3314 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3315 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3316 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3317 * return value of true). However, for short flows the cost of bookkeeping is
3318 * much higher than the benefits, so when the datapath holds a large number of
3319 * flows we impose some heuristics to decide which flows are likely to be worth
3322 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3324 struct dpif_backer *backer = miss->ofproto->backer;
3327 switch (flow_miss_model) {
3328 case OFPROTO_HANDLE_MISS_AUTO:
3330 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3332 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3336 if (!backer->governor) {
3339 n_subfacets = hmap_count(&backer->subfacets);
3340 if (n_subfacets * 2 <= flow_eviction_threshold) {
3344 backer->governor = governor_create();
3347 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3348 return governor_should_install_flow(backer->governor, hash,
3349 list_size(&miss->packets));
3352 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3353 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3354 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3356 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3357 struct flow_miss *miss,
3358 struct flow_miss_op *ops, size_t *n_ops)
3360 struct ofpbuf *packet;
3362 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3364 COVERAGE_INC(facet_suppress);
3366 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3367 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3370 struct xlate_in xin;
3372 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3373 xlate_actions_for_side_effects(&xin);
3376 if (xout->odp_actions.size) {
3377 struct flow_miss_op *op = &ops[*n_ops];
3378 struct dpif_execute *execute = &op->dpif_op.u.execute;
3380 init_flow_miss_execute_op(miss, packet, op);
3381 xlate_out_copy(&op->xout, xout);
3382 execute->actions = op->xout.odp_actions.data;
3383 execute->actions_len = op->xout.odp_actions.size;
3384 op->xout_garbage = true;
3391 /* Handles 'miss', which matches 'facet'. May add any required datapath
3392 * operations to 'ops', incrementing '*n_ops' for each new op.
3394 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3395 * This is really important only for new facets: if we just called time_msec()
3396 * here, then the new subfacet or its packets could look (occasionally) as
3397 * though it was used some time after the facet was used. That can make a
3398 * one-packet flow look like it has a nonzero duration, which looks odd in
3399 * e.g. NetFlow statistics.
3401 * If non-null, 'stats' will be folded into 'facet'. */
3403 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3404 long long int now, struct dpif_flow_stats *stats,
3405 struct flow_miss_op *ops, size_t *n_ops)
3407 enum subfacet_path want_path;
3408 struct subfacet *subfacet;
3409 struct ofpbuf *packet;
3411 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3413 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3414 struct flow_miss_op *op = &ops[*n_ops];
3416 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3419 if (want_path != SF_FAST_PATH) {
3420 struct rule_dpif *rule;
3421 struct xlate_in xin;
3423 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3424 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3425 xlate_actions_for_side_effects(&xin);
3428 if (facet->xout.odp_actions.size) {
3429 struct dpif_execute *execute = &op->dpif_op.u.execute;
3431 init_flow_miss_execute_op(miss, packet, op);
3432 execute->actions = facet->xout.odp_actions.data,
3433 execute->actions_len = facet->xout.odp_actions.size;
3438 /* Don't install the flow if it's the result of the "userspace"
3439 * action for an already installed facet. This can occur when a
3440 * datapath flow with wildcards has a "userspace" action and flows
3441 * sent to userspace result in a different subfacet, which will then
3442 * be rejected as overlapping by the datapath. */
3443 if (miss->upcall_type == DPIF_UC_ACTION
3444 && !list_is_empty(&facet->subfacets)) {
3446 facet->used = MAX(facet->used, stats->used);
3447 facet->packet_count += stats->n_packets;
3448 facet->byte_count += stats->n_bytes;
3449 facet->tcp_flags |= stats->tcp_flags;
3454 subfacet = subfacet_create(facet, miss, now);
3456 subfacet_update_stats(subfacet, stats);
3459 if (subfacet->path != want_path) {
3460 struct flow_miss_op *op = &ops[(*n_ops)++];
3461 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3463 subfacet->path = want_path;
3465 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3466 if (enable_megaflows) {
3467 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3468 &miss->flow, UINT32_MAX);
3471 op->xout_garbage = false;
3472 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3473 op->subfacet = subfacet;
3474 put->flags = DPIF_FP_CREATE;
3475 put->key = miss->key;
3476 put->key_len = miss->key_len;
3477 put->mask = op->mask.data;
3478 put->mask_len = op->mask.size;
3480 if (want_path == SF_FAST_PATH) {
3481 put->actions = facet->xout.odp_actions.data;
3482 put->actions_len = facet->xout.odp_actions.size;
3484 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3485 op->slow_stub, sizeof op->slow_stub,
3486 &put->actions, &put->actions_len);
3492 /* Handles flow miss 'miss'. May add any required datapath operations
3493 * to 'ops', incrementing '*n_ops' for each new op. */
3495 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3498 struct ofproto_dpif *ofproto = miss->ofproto;
3499 struct dpif_flow_stats stats__;
3500 struct dpif_flow_stats *stats = &stats__;
3501 struct ofpbuf *packet;
3502 struct facet *facet;
3506 memset(stats, 0, sizeof *stats);
3508 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3509 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3510 stats->n_bytes += packet->size;
3514 facet = facet_lookup_valid(ofproto, &miss->flow);
3516 struct flow_wildcards wc;
3517 struct rule_dpif *rule;
3518 struct xlate_out xout;
3519 struct xlate_in xin;
3521 flow_wildcards_init_catchall(&wc);
3522 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3523 rule_credit_stats(rule, stats);
3525 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3527 xin.resubmit_stats = stats;
3528 xin.may_learn = true;
3529 xlate_actions(&xin, &xout);
3530 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3532 /* There does not exist a bijection between 'struct flow' and datapath
3533 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3534 * assumption used throughout the facet and subfacet handling code.
3535 * Since we have to handle these misses in userspace anyway, we simply
3536 * skip facet creation, avoiding the problem altogether. */
3537 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3538 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3539 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3543 facet = facet_create(miss, rule, &xout, stats);
3546 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3549 static struct drop_key *
3550 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3553 struct drop_key *drop_key;
3555 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3556 &backer->drop_keys) {
3557 if (drop_key->key_len == key_len
3558 && !memcmp(drop_key->key, key, key_len)) {
3566 drop_key_clear(struct dpif_backer *backer)
3568 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3569 struct drop_key *drop_key, *next;
3571 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3574 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3576 if (error && !VLOG_DROP_WARN(&rl)) {
3577 struct ds ds = DS_EMPTY_INITIALIZER;
3578 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3579 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3580 ovs_strerror(error), ds_cstr(&ds));
3584 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3585 free(drop_key->key);
3590 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3591 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3592 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3593 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3594 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3595 * 'packet' ingressed.
3597 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3598 * 'flow''s in_port to OFPP_NONE.
3600 * This function does post-processing on data returned from
3601 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3602 * of the upcall processing logic. In particular, if the extracted in_port is
3603 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3604 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3605 * a VLAN header onto 'packet' (if it is nonnull).
3607 * Similarly, this function also includes some logic to help with tunnels. It
3608 * may modify 'flow' as necessary to make the tunneling implementation
3609 * transparent to the upcall processing logic.
3611 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3612 * or some other positive errno if there are other problems. */
3614 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3615 const struct nlattr *key, size_t key_len,
3616 struct flow *flow, enum odp_key_fitness *fitnessp,
3617 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3619 const struct ofport_dpif *port;
3620 enum odp_key_fitness fitness;
3623 fitness = odp_flow_key_to_flow(key, key_len, flow);
3624 if (fitness == ODP_FIT_ERROR) {
3630 *odp_in_port = flow->in_port.odp_port;
3633 port = (tnl_port_should_receive(flow)
3634 ? tnl_port_receive(flow)
3635 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3636 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3641 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3642 * it's theoretically possible that we'll receive an ofport belonging to an
3643 * entirely different datapath. In practice, this can't happen because no
3644 * platforms has two separate datapaths which each support tunneling. */
3645 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3647 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3649 /* Make the packet resemble the flow, so that it gets sent to
3650 * an OpenFlow controller properly, so that it looks correct
3651 * for sFlow, and so that flow_extract() will get the correct
3652 * vlan_tci if it is called on 'packet'.
3654 * The allocated space inside 'packet' probably also contains
3655 * 'key', that is, both 'packet' and 'key' are probably part of
3656 * a struct dpif_upcall (see the large comment on that
3657 * structure definition), so pushing data on 'packet' is in
3658 * general not a good idea since it could overwrite 'key' or
3659 * free it as a side effect. However, it's OK in this special
3660 * case because we know that 'packet' is inside a Netlink
3661 * attribute: pushing 4 bytes will just overwrite the 4-byte
3662 * "struct nlattr", which is fine since we don't need that
3663 * header anymore. */
3664 eth_push_vlan(packet, flow->vlan_tci);
3666 /* We can't reproduce 'key' from 'flow'. */
3667 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3672 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3677 *fitnessp = fitness;
3683 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3686 struct dpif_upcall *upcall;
3687 struct flow_miss *miss;
3688 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3689 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3690 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3700 /* Construct the to-do list.
3702 * This just amounts to extracting the flow from each packet and sticking
3703 * the packets that have the same flow in the same "flow_miss" structure so
3704 * that we can process them together. */
3707 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3708 struct flow_miss *miss = &misses[n_misses];
3709 struct flow_miss *existing_miss;
3710 struct ofproto_dpif *ofproto;
3711 odp_port_t odp_in_port;
3716 error = ofproto_receive(backer, upcall->packet, upcall->key,
3717 upcall->key_len, &flow, &miss->key_fitness,
3718 &ofproto, &odp_in_port);
3719 if (error == ENODEV) {
3720 struct drop_key *drop_key;
3722 /* Received packet on datapath port for which we couldn't
3723 * associate an ofproto. This can happen if a port is removed
3724 * while traffic is being received. Print a rate-limited message
3725 * in case it happens frequently. Install a drop flow so
3726 * that future packets of the flow are inexpensively dropped
3728 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3729 "%"PRIu32, odp_in_port);
3731 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3734 ret = dpif_flow_put(backer->dpif,
3735 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3736 upcall->key, upcall->key_len,
3737 NULL, 0, NULL, 0, NULL);
3740 drop_key = xmalloc(sizeof *drop_key);
3741 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3742 drop_key->key_len = upcall->key_len;
3744 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3745 hash_bytes(drop_key->key, drop_key->key_len, 0));
3754 ofproto->n_missed++;
3755 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3756 &flow.tunnel, &flow.in_port, &miss->flow);
3758 /* Add other packets to a to-do list. */
3759 hash = flow_hash(&miss->flow, 0);
3760 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3761 if (!existing_miss) {
3762 hmap_insert(&todo, &miss->hmap_node, hash);
3763 miss->ofproto = ofproto;
3764 miss->key = upcall->key;
3765 miss->key_len = upcall->key_len;
3766 miss->upcall_type = upcall->type;
3767 list_init(&miss->packets);
3771 miss = existing_miss;
3773 list_push_back(&miss->packets, &upcall->packet->list_node);
3776 /* Process each element in the to-do list, constructing the set of
3777 * operations to batch. */
3779 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3780 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3782 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3784 /* Execute batch. */
3785 for (i = 0; i < n_ops; i++) {
3786 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3788 dpif_operate(backer->dpif, dpif_ops, n_ops);
3790 for (i = 0; i < n_ops; i++) {
3791 if (dpif_ops[i]->error != 0
3792 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3793 && flow_miss_ops[i].subfacet) {
3794 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3796 COVERAGE_INC(subfacet_install_fail);
3798 /* Zero-out subfacet counters when installation failed, but
3799 * datapath reported hits. This should not happen and
3800 * indicates a bug, since if the datapath flow exists, we
3801 * should not be attempting to create a new subfacet. A
3802 * buggy datapath could trigger this, so just zero out the
3803 * counters and log an error. */
3804 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3805 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3806 "datapath reported hits");
3807 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3810 subfacet->path = SF_NOT_INSTALLED;
3814 if (flow_miss_ops[i].xout_garbage) {
3815 xlate_out_uninit(&flow_miss_ops[i].xout);
3818 hmap_destroy(&todo);
3821 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3823 classify_upcall(const struct dpif_upcall *upcall)
3825 size_t userdata_len;
3826 union user_action_cookie cookie;
3828 /* First look at the upcall type. */
3829 switch (upcall->type) {
3830 case DPIF_UC_ACTION:
3836 case DPIF_N_UC_TYPES:
3838 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3842 /* "action" upcalls need a closer look. */
3843 if (!upcall->userdata) {
3844 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3847 userdata_len = nl_attr_get_size(upcall->userdata);
3848 if (userdata_len < sizeof cookie.type
3849 || userdata_len > sizeof cookie) {
3850 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3854 memset(&cookie, 0, sizeof cookie);
3855 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3856 if (userdata_len == sizeof cookie.sflow
3857 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3858 return SFLOW_UPCALL;
3859 } else if (userdata_len == sizeof cookie.slow_path
3860 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3862 } else if (userdata_len == sizeof cookie.flow_sample
3863 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3864 return FLOW_SAMPLE_UPCALL;
3865 } else if (userdata_len == sizeof cookie.ipfix
3866 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3867 return IPFIX_UPCALL;
3869 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3870 " and size %zu", cookie.type, userdata_len);
3876 handle_sflow_upcall(struct dpif_backer *backer,
3877 const struct dpif_upcall *upcall)
3879 struct ofproto_dpif *ofproto;
3880 union user_action_cookie cookie;
3882 odp_port_t odp_in_port;
3884 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3885 &flow, NULL, &ofproto, &odp_in_port)
3886 || !ofproto->sflow) {
3890 memset(&cookie, 0, sizeof cookie);
3891 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3892 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3893 odp_in_port, &cookie);
3897 handle_flow_sample_upcall(struct dpif_backer *backer,
3898 const struct dpif_upcall *upcall)
3900 struct ofproto_dpif *ofproto;
3901 union user_action_cookie cookie;
3904 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3905 &flow, NULL, &ofproto, NULL)
3906 || !ofproto->ipfix) {
3910 memset(&cookie, 0, sizeof cookie);
3911 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3913 /* The flow reflects exactly the contents of the packet. Sample
3914 * the packet using it. */
3915 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3916 cookie.flow_sample.collector_set_id,
3917 cookie.flow_sample.probability,
3918 cookie.flow_sample.obs_domain_id,
3919 cookie.flow_sample.obs_point_id);
3923 handle_ipfix_upcall(struct dpif_backer *backer,
3924 const struct dpif_upcall *upcall)
3926 struct ofproto_dpif *ofproto;
3929 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3930 &flow, NULL, &ofproto, NULL)
3931 || !ofproto->ipfix) {
3935 /* The flow reflects exactly the contents of the packet. Sample
3936 * the packet using it. */
3937 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3941 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3943 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3944 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3945 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3950 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3953 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3954 struct dpif_upcall *upcall = &misses[n_misses];
3955 struct ofpbuf *buf = &miss_bufs[n_misses];
3958 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3959 sizeof miss_buf_stubs[n_misses]);
3960 error = dpif_recv(backer->dpif, upcall, buf);
3966 switch (classify_upcall(upcall)) {
3968 /* Handle it later. */
3973 handle_sflow_upcall(backer, upcall);
3977 case FLOW_SAMPLE_UPCALL:
3978 handle_flow_sample_upcall(backer, upcall);
3983 handle_ipfix_upcall(backer, upcall);
3993 /* Handle deferred MISS_UPCALL processing. */
3994 handle_miss_upcalls(backer, misses, n_misses);
3995 for (i = 0; i < n_misses; i++) {
3996 ofpbuf_uninit(&miss_bufs[i]);
4002 /* Flow expiration. */
4004 static int subfacet_max_idle(const struct dpif_backer *);
4005 static void update_stats(struct dpif_backer *);
4006 static void rule_expire(struct rule_dpif *);
4007 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4009 /* This function is called periodically by run(). Its job is to collect
4010 * updates for the flows that have been installed into the datapath, most
4011 * importantly when they last were used, and then use that information to
4012 * expire flows that have not been used recently.
4014 * Returns the number of milliseconds after which it should be called again. */
4016 expire(struct dpif_backer *backer)
4018 struct ofproto_dpif *ofproto;
4022 /* Periodically clear out the drop keys in an effort to keep them
4023 * relatively few. */
4024 drop_key_clear(backer);
4026 /* Update stats for each flow in the backer. */
4027 update_stats(backer);
4029 n_subfacets = hmap_count(&backer->subfacets);
4031 struct subfacet *subfacet;
4032 long long int total, now;
4036 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4037 total += now - subfacet->created;
4039 backer->avg_subfacet_life += total / n_subfacets;
4041 backer->avg_subfacet_life /= 2;
4043 backer->avg_n_subfacet += n_subfacets;
4044 backer->avg_n_subfacet /= 2;
4046 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4048 max_idle = subfacet_max_idle(backer);
4049 expire_subfacets(backer, max_idle);
4051 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4052 struct rule *rule, *next_rule;
4054 if (ofproto->backer != backer) {
4058 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4060 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4061 &ofproto->up.expirable) {
4062 rule_expire(rule_dpif_cast(rule));
4065 /* All outstanding data in existing flows has been accounted, so it's a
4066 * good time to do bond rebalancing. */
4067 if (ofproto->has_bonded_bundles) {
4068 struct ofbundle *bundle;
4070 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4072 bond_rebalance(bundle->bond);
4078 return MIN(max_idle, 1000);
4081 /* Updates flow table statistics given that the datapath just reported 'stats'
4082 * as 'subfacet''s statistics. */
4084 update_subfacet_stats(struct subfacet *subfacet,
4085 const struct dpif_flow_stats *stats)
4087 struct facet *facet = subfacet->facet;
4088 struct dpif_flow_stats diff;
4090 diff.tcp_flags = stats->tcp_flags;
4091 diff.used = stats->used;
4093 if (stats->n_packets >= subfacet->dp_packet_count) {
4094 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4096 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4100 if (stats->n_bytes >= subfacet->dp_byte_count) {
4101 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4103 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4107 facet->ofproto->n_hit += diff.n_packets;
4108 subfacet->dp_packet_count = stats->n_packets;
4109 subfacet->dp_byte_count = stats->n_bytes;
4110 subfacet_update_stats(subfacet, &diff);
4112 if (facet->accounted_bytes < facet->byte_count) {
4114 facet_account(facet);
4115 facet->accounted_bytes = facet->byte_count;
4119 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4120 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4122 delete_unexpected_flow(struct dpif_backer *backer,
4123 const struct nlattr *key, size_t key_len)
4125 if (!VLOG_DROP_WARN(&rl)) {
4129 odp_flow_key_format(key, key_len, &s);
4130 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4134 COVERAGE_INC(facet_unexpected);
4135 dpif_flow_del(backer->dpif, key, key_len, NULL);
4138 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4140 * This function also pushes statistics updates to rules which each facet
4141 * resubmits into. Generally these statistics will be accurate. However, if a
4142 * facet changes the rule it resubmits into at some time in between
4143 * update_stats() runs, it is possible that statistics accrued to the
4144 * old rule will be incorrectly attributed to the new rule. This could be
4145 * avoided by calling update_stats() whenever rules are created or
4146 * deleted. However, the performance impact of making so many calls to the
4147 * datapath do not justify the benefit of having perfectly accurate statistics.
4149 * In addition, this function maintains per ofproto flow hit counts. The patch
4150 * port is not treated specially. e.g. A packet ingress from br0 patched into
4151 * br1 will increase the hit count of br0 by 1, however, does not affect
4152 * the hit or miss counts of br1.
4155 update_stats(struct dpif_backer *backer)
4157 const struct dpif_flow_stats *stats;
4158 struct dpif_flow_dump dump;
4159 const struct nlattr *key, *mask;
4160 size_t key_len, mask_len;
4162 dpif_flow_dump_start(&dump, backer->dpif);
4163 while (dpif_flow_dump_next(&dump, &key, &key_len,
4164 &mask, &mask_len, NULL, NULL, &stats)) {
4165 struct subfacet *subfacet;
4168 key_hash = odp_flow_key_hash(key, key_len);
4169 subfacet = subfacet_find(backer, key, key_len, key_hash);
4170 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4172 update_subfacet_stats(subfacet, stats);
4176 /* Stats are updated per-packet. */
4179 case SF_NOT_INSTALLED:
4181 delete_unexpected_flow(backer, key, key_len);
4186 dpif_flow_dump_done(&dump);
4188 update_moving_averages(backer);
4191 /* Calculates and returns the number of milliseconds of idle time after which
4192 * subfacets should expire from the datapath. When a subfacet expires, we fold
4193 * its statistics into its facet, and when a facet's last subfacet expires, we
4194 * fold its statistic into its rule. */
4196 subfacet_max_idle(const struct dpif_backer *backer)
4199 * Idle time histogram.
4201 * Most of the time a switch has a relatively small number of subfacets.
4202 * When this is the case we might as well keep statistics for all of them
4203 * in userspace and to cache them in the kernel datapath for performance as
4206 * As the number of subfacets increases, the memory required to maintain
4207 * statistics about them in userspace and in the kernel becomes
4208 * significant. However, with a large number of subfacets it is likely
4209 * that only a few of them are "heavy hitters" that consume a large amount
4210 * of bandwidth. At this point, only heavy hitters are worth caching in
4211 * the kernel and maintaining in userspaces; other subfacets we can
4214 * The technique used to compute the idle time is to build a histogram with
4215 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4216 * that is installed in the kernel gets dropped in the appropriate bucket.
4217 * After the histogram has been built, we compute the cutoff so that only
4218 * the most-recently-used 1% of subfacets (but at least
4219 * flow_eviction_threshold flows) are kept cached. At least
4220 * the most-recently-used bucket of subfacets is kept, so actually an
4221 * arbitrary number of subfacets can be kept in any given expiration run
4222 * (though the next run will delete most of those unless they receive
4225 * This requires a second pass through the subfacets, in addition to the
4226 * pass made by update_stats(), because the former function never looks at
4227 * uninstallable subfacets.
4229 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4230 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4231 int buckets[N_BUCKETS] = { 0 };
4232 int total, subtotal, bucket;
4233 struct subfacet *subfacet;
4237 total = hmap_count(&backer->subfacets);
4238 if (total <= flow_eviction_threshold) {
4239 return N_BUCKETS * BUCKET_WIDTH;
4242 /* Build histogram. */
4244 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4245 long long int idle = now - subfacet->used;
4246 int bucket = (idle <= 0 ? 0
4247 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4248 : (unsigned int) idle / BUCKET_WIDTH);
4252 /* Find the first bucket whose flows should be expired. */
4253 subtotal = bucket = 0;
4255 subtotal += buckets[bucket++];
4256 } while (bucket < N_BUCKETS &&
4257 subtotal < MAX(flow_eviction_threshold, total / 100));
4259 if (VLOG_IS_DBG_ENABLED()) {
4263 ds_put_cstr(&s, "keep");
4264 for (i = 0; i < N_BUCKETS; i++) {
4266 ds_put_cstr(&s, ", drop");
4269 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4272 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4276 return bucket * BUCKET_WIDTH;
4280 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4282 /* Cutoff time for most flows. */
4283 long long int normal_cutoff = time_msec() - dp_max_idle;
4285 /* We really want to keep flows for special protocols around, so use a more
4286 * conservative cutoff. */
4287 long long int special_cutoff = time_msec() - 10000;
4289 struct subfacet *subfacet, *next_subfacet;
4290 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4294 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4295 &backer->subfacets) {
4296 long long int cutoff;
4298 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4302 if (subfacet->used < cutoff) {
4303 if (subfacet->path != SF_NOT_INSTALLED) {
4304 batch[n_batch++] = subfacet;
4305 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4306 subfacet_destroy_batch(backer, batch, n_batch);
4310 subfacet_destroy(subfacet);
4316 subfacet_destroy_batch(backer, batch, n_batch);
4320 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4321 * then delete it entirely. */
4323 rule_expire(struct rule_dpif *rule)
4328 if (rule->up.pending) {
4329 /* We'll have to expire it later. */
4333 /* Has 'rule' expired? */
4335 if (rule->up.hard_timeout
4336 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4337 reason = OFPRR_HARD_TIMEOUT;
4338 } else if (rule->up.idle_timeout
4339 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4340 reason = OFPRR_IDLE_TIMEOUT;
4345 COVERAGE_INC(ofproto_dpif_expired);
4347 /* Get rid of the rule. */
4348 ofproto_rule_expire(&rule->up, reason);
4353 /* Creates and returns a new facet based on 'miss'.
4355 * The caller must already have determined that no facet with an identical
4356 * 'miss->flow' exists in 'miss->ofproto'.
4358 * 'rule' and 'xout' must have been created based on 'miss'.
4360 * 'facet'' statistics are initialized based on 'stats'.
4362 * The facet will initially have no subfacets. The caller should create (at
4363 * least) one subfacet with subfacet_create(). */
4364 static struct facet *
4365 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4366 struct xlate_out *xout, struct dpif_flow_stats *stats)
4368 struct ofproto_dpif *ofproto = miss->ofproto;
4369 struct facet *facet;
4372 facet = xzalloc(sizeof *facet);
4373 facet->ofproto = miss->ofproto;
4374 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4375 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4376 facet->tcp_flags = stats->tcp_flags;
4377 facet->used = stats->used;
4378 facet->flow = miss->flow;
4379 facet->learn_rl = time_msec() + 500;
4381 list_init(&facet->subfacets);
4382 netflow_flow_init(&facet->nf_flow);
4383 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4385 xlate_out_copy(&facet->xout, xout);
4387 match_init(&match, &facet->flow, &facet->xout.wc);
4388 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4389 classifier_insert(&ofproto->facets, &facet->cr);
4391 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4392 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4398 facet_free(struct facet *facet)
4401 xlate_out_uninit(&facet->xout);
4406 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4407 * 'packet', which arrived on 'in_port'. */
4409 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4410 const struct nlattr *odp_actions, size_t actions_len,
4411 struct ofpbuf *packet)
4413 struct odputil_keybuf keybuf;
4417 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4418 odp_flow_key_from_flow(&key, flow,
4419 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4421 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4422 odp_actions, actions_len, packet);
4426 /* Remove 'facet' from its ofproto and free up the associated memory:
4428 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4429 * rule's statistics, via subfacet_uninstall().
4431 * - Removes 'facet' from its rule and from ofproto->facets.
4434 facet_remove(struct facet *facet)
4436 struct subfacet *subfacet, *next_subfacet;
4438 ovs_assert(!list_is_empty(&facet->subfacets));
4440 /* First uninstall all of the subfacets to get final statistics. */
4441 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4442 subfacet_uninstall(subfacet);
4445 /* Flush the final stats to the rule.
4447 * This might require us to have at least one subfacet around so that we
4448 * can use its actions for accounting in facet_account(), which is why we
4449 * have uninstalled but not yet destroyed the subfacets. */
4450 facet_flush_stats(facet);
4452 /* Now we're really all done so destroy everything. */
4453 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4454 &facet->subfacets) {
4455 subfacet_destroy__(subfacet);
4457 classifier_remove(&facet->ofproto->facets, &facet->cr);
4458 cls_rule_destroy(&facet->cr);
4462 /* Feed information from 'facet' back into the learning table to keep it in
4463 * sync with what is actually flowing through the datapath. */
4465 facet_learn(struct facet *facet)
4467 long long int now = time_msec();
4469 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4473 facet->learn_rl = now + 500;
4475 if (!facet->xout.has_learn
4476 && !facet->xout.has_normal
4477 && (!facet->xout.has_fin_timeout
4478 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4482 facet_push_stats(facet, true);
4486 facet_account(struct facet *facet)
4488 const struct nlattr *a;
4493 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4496 n_bytes = facet->byte_count - facet->accounted_bytes;
4498 /* This loop feeds byte counters to bond_account() for rebalancing to use
4499 * as a basis. We also need to track the actual VLAN on which the packet
4500 * is going to be sent to ensure that it matches the one passed to
4501 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4504 * We use the actions from an arbitrary subfacet because they should all
4505 * be equally valid for our purpose. */
4506 vlan_tci = facet->flow.vlan_tci;
4507 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4508 facet->xout.odp_actions.size) {
4509 const struct ovs_action_push_vlan *vlan;
4510 struct ofport_dpif *port;
4512 switch (nl_attr_type(a)) {
4513 case OVS_ACTION_ATTR_OUTPUT:
4514 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4515 if (port && port->bundle && port->bundle->bond) {
4516 bond_account(port->bundle->bond, &facet->flow,
4517 vlan_tci_to_vid(vlan_tci), n_bytes);
4521 case OVS_ACTION_ATTR_POP_VLAN:
4522 vlan_tci = htons(0);
4525 case OVS_ACTION_ATTR_PUSH_VLAN:
4526 vlan = nl_attr_get(a);
4527 vlan_tci = vlan->vlan_tci;
4533 /* Returns true if the only action for 'facet' is to send to the controller.
4534 * (We don't report NetFlow expiration messages for such facets because they
4535 * are just part of the control logic for the network, not real traffic). */
4537 facet_is_controller_flow(struct facet *facet)
4540 struct ofproto_dpif *ofproto = facet->ofproto;
4541 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4543 const struct ofpact *ofpacts = rule->up.ofpacts;
4544 size_t ofpacts_len = rule->up.ofpacts_len;
4546 if (ofpacts_len > 0 &&
4547 ofpacts->type == OFPACT_CONTROLLER &&
4548 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4555 /* Folds all of 'facet''s statistics into its rule. Also updates the
4556 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4557 * 'facet''s statistics in the datapath should have been zeroed and folded into
4558 * its packet and byte counts before this function is called. */
4560 facet_flush_stats(struct facet *facet)
4562 struct ofproto_dpif *ofproto = facet->ofproto;
4563 struct subfacet *subfacet;
4565 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4566 ovs_assert(!subfacet->dp_byte_count);
4567 ovs_assert(!subfacet->dp_packet_count);
4570 facet_push_stats(facet, false);
4571 if (facet->accounted_bytes < facet->byte_count) {
4572 facet_account(facet);
4573 facet->accounted_bytes = facet->byte_count;
4576 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4577 struct ofexpired expired;
4578 expired.flow = facet->flow;
4579 expired.packet_count = facet->packet_count;
4580 expired.byte_count = facet->byte_count;
4581 expired.used = facet->used;
4582 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4585 /* Reset counters to prevent double counting if 'facet' ever gets
4587 facet_reset_counters(facet);
4589 netflow_flow_clear(&facet->nf_flow);
4590 facet->tcp_flags = 0;
4593 /* Searches 'ofproto''s table of facets for one which would be responsible for
4594 * 'flow'. Returns it if found, otherwise a null pointer.
4596 * The returned facet might need revalidation; use facet_lookup_valid()
4597 * instead if that is important. */
4598 static struct facet *
4599 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4601 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4602 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4605 /* Searches 'ofproto''s table of facets for one capable that covers
4606 * 'flow'. Returns it if found, otherwise a null pointer.
4608 * The returned facet is guaranteed to be valid. */
4609 static struct facet *
4610 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4612 struct facet *facet;
4614 facet = facet_find(ofproto, flow);
4616 && ofproto->backer->need_revalidate
4617 && !facet_revalidate(facet)) {
4625 facet_check_consistency(struct facet *facet)
4627 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4629 struct xlate_out xout;
4630 struct xlate_in xin;
4632 struct rule_dpif *rule;
4635 /* Check the datapath actions for consistency. */
4636 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4637 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4638 xlate_actions(&xin, &xout);
4640 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4641 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4642 && facet->xout.slow == xout.slow
4643 && facet->fail_open == fail_open;
4644 if (!ok && !VLOG_DROP_WARN(&rl)) {
4645 struct ds s = DS_EMPTY_INITIALIZER;
4647 flow_format(&s, &facet->flow);
4648 ds_put_cstr(&s, ": inconsistency in facet");
4650 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4651 ds_put_cstr(&s, " (actions were: ");
4652 format_odp_actions(&s, facet->xout.odp_actions.data,
4653 facet->xout.odp_actions.size);
4654 ds_put_cstr(&s, ") (correct actions: ");
4655 format_odp_actions(&s, xout.odp_actions.data,
4656 xout.odp_actions.size);
4657 ds_put_char(&s, ')');
4660 if (facet->xout.slow != xout.slow) {
4661 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4664 if (facet->fail_open != fail_open) {
4665 ds_put_format(&s, " fail open incorrect. should be %s",
4666 fail_open ? "true" : "false");
4670 xlate_out_uninit(&xout);
4675 /* Re-searches the classifier for 'facet':
4677 * - If the rule found is different from 'facet''s current rule, moves
4678 * 'facet' to the new rule and recompiles its actions.
4680 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4681 * where it is and recompiles its actions anyway.
4683 * - If any of 'facet''s subfacets correspond to a new flow according to
4684 * ofproto_receive(), 'facet' is removed.
4686 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4688 facet_revalidate(struct facet *facet)
4690 struct ofproto_dpif *ofproto = facet->ofproto;
4691 struct rule_dpif *new_rule;
4692 struct subfacet *subfacet;
4693 struct flow_wildcards wc;
4694 struct xlate_out xout;
4695 struct xlate_in xin;
4697 COVERAGE_INC(facet_revalidate);
4699 /* Check that child subfacets still correspond to this facet. Tunnel
4700 * configuration changes could cause a subfacet's OpenFlow in_port to
4702 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4703 struct ofproto_dpif *recv_ofproto;
4704 struct flow recv_flow;
4707 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4708 subfacet->key_len, &recv_flow, NULL,
4709 &recv_ofproto, NULL);
4711 || recv_ofproto != ofproto
4712 || facet != facet_find(ofproto, &recv_flow)) {
4713 facet_remove(facet);
4718 flow_wildcards_init_catchall(&wc);
4719 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4721 /* Calculate new datapath actions.
4723 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4724 * emit a NetFlow expiration and, if so, we need to have the old state
4725 * around to properly compose it. */
4726 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4727 xlate_actions(&xin, &xout);
4728 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4730 /* A facet's slow path reason should only change under dramatic
4731 * circumstances. Rather than try to update everything, it's simpler to
4732 * remove the facet and start over.
4734 * More importantly, if a facet's wildcards change, it will be relatively
4735 * difficult to figure out if its subfacets still belong to it, and if not
4736 * which facet they may belong to. Again, to avoid the complexity, we
4737 * simply give up instead. */
4738 if (facet->xout.slow != xout.slow
4739 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4740 facet_remove(facet);
4741 xlate_out_uninit(&xout);
4745 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4746 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4747 if (subfacet->path == SF_FAST_PATH) {
4748 struct dpif_flow_stats stats;
4750 subfacet_install(subfacet, &xout.odp_actions, &stats);
4751 subfacet_update_stats(subfacet, &stats);
4755 facet_flush_stats(facet);
4757 ofpbuf_clear(&facet->xout.odp_actions);
4758 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4759 xout.odp_actions.size);
4762 /* Update 'facet' now that we've taken care of all the old state. */
4763 facet->xout.slow = xout.slow;
4764 facet->xout.has_learn = xout.has_learn;
4765 facet->xout.has_normal = xout.has_normal;
4766 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4767 facet->xout.nf_output_iface = xout.nf_output_iface;
4768 facet->xout.mirrors = xout.mirrors;
4769 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4770 facet->used = MAX(facet->used, new_rule->up.created);
4771 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4773 xlate_out_uninit(&xout);
4778 facet_reset_counters(struct facet *facet)
4780 facet->packet_count = 0;
4781 facet->byte_count = 0;
4782 facet->prev_packet_count = 0;
4783 facet->prev_byte_count = 0;
4784 facet->accounted_bytes = 0;
4788 facet_push_stats(struct facet *facet, bool may_learn)
4790 struct dpif_flow_stats stats;
4792 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4793 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4794 ovs_assert(facet->used >= facet->prev_used);
4796 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4797 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4798 stats.used = facet->used;
4799 stats.tcp_flags = facet->tcp_flags;
4801 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4802 struct ofproto_dpif *ofproto = facet->ofproto;
4803 struct ofport_dpif *in_port;
4804 struct rule_dpif *rule;
4805 struct xlate_in xin;
4807 facet->prev_packet_count = facet->packet_count;
4808 facet->prev_byte_count = facet->byte_count;
4809 facet->prev_used = facet->used;
4811 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4812 if (in_port && in_port->is_tunnel) {
4813 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4816 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4817 rule_credit_stats(rule, &stats);
4818 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4820 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4821 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4822 stats.n_packets, stats.n_bytes);
4824 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4826 xin.resubmit_stats = &stats;
4827 xin.may_learn = may_learn;
4828 xlate_actions_for_side_effects(&xin);
4833 push_all_stats__(bool run_fast)
4835 static long long int rl = LLONG_MIN;
4836 struct ofproto_dpif *ofproto;
4838 if (time_msec() < rl) {
4842 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4843 struct cls_cursor cursor;
4844 struct facet *facet;
4846 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4847 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4848 facet_push_stats(facet, false);
4855 rl = time_msec() + 100;
4859 push_all_stats(void)
4861 push_all_stats__(true);
4865 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4867 ovs_mutex_lock(&rule->stats_mutex);
4868 rule->packet_count += stats->n_packets;
4869 rule->byte_count += stats->n_bytes;
4870 ofproto_rule_update_used(&rule->up, stats->used);
4871 ovs_mutex_unlock(&rule->stats_mutex);
4876 static struct subfacet *
4877 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4878 size_t key_len, uint32_t key_hash)
4880 struct subfacet *subfacet;
4882 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4883 &backer->subfacets) {
4884 if (subfacet->key_len == key_len
4885 && !memcmp(key, subfacet->key, key_len)) {
4893 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4894 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4895 * existing subfacet if there is one, otherwise creates and returns a
4897 static struct subfacet *
4898 subfacet_create(struct facet *facet, struct flow_miss *miss,
4901 struct dpif_backer *backer = miss->ofproto->backer;
4902 enum odp_key_fitness key_fitness = miss->key_fitness;
4903 const struct nlattr *key = miss->key;
4904 size_t key_len = miss->key_len;
4906 struct subfacet *subfacet;
4908 key_hash = odp_flow_key_hash(key, key_len);
4910 if (list_is_empty(&facet->subfacets)) {
4911 subfacet = &facet->one_subfacet;
4913 subfacet = subfacet_find(backer, key, key_len, key_hash);
4915 if (subfacet->facet == facet) {
4919 /* This shouldn't happen. */
4920 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4921 subfacet_destroy(subfacet);
4924 subfacet = xmalloc(sizeof *subfacet);
4927 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4928 list_push_back(&facet->subfacets, &subfacet->list_node);
4929 subfacet->facet = facet;
4930 subfacet->key_fitness = key_fitness;
4931 subfacet->key = xmemdup(key, key_len);
4932 subfacet->key_len = key_len;
4933 subfacet->used = now;
4934 subfacet->created = now;
4935 subfacet->dp_packet_count = 0;
4936 subfacet->dp_byte_count = 0;
4937 subfacet->path = SF_NOT_INSTALLED;
4938 subfacet->backer = backer;
4940 backer->subfacet_add_count++;
4944 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4945 * its facet within 'ofproto', and frees it. */
4947 subfacet_destroy__(struct subfacet *subfacet)
4949 struct facet *facet = subfacet->facet;
4950 struct ofproto_dpif *ofproto = facet->ofproto;
4952 /* Update ofproto stats before uninstall the subfacet. */
4953 ofproto->backer->subfacet_del_count++;
4955 subfacet_uninstall(subfacet);
4956 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4957 list_remove(&subfacet->list_node);
4958 free(subfacet->key);
4959 if (subfacet != &facet->one_subfacet) {
4964 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4965 * last remaining subfacet in its facet destroys the facet too. */
4967 subfacet_destroy(struct subfacet *subfacet)
4969 struct facet *facet = subfacet->facet;
4971 if (list_is_singleton(&facet->subfacets)) {
4972 /* facet_remove() needs at least one subfacet (it will remove it). */
4973 facet_remove(facet);
4975 subfacet_destroy__(subfacet);
4980 subfacet_destroy_batch(struct dpif_backer *backer,
4981 struct subfacet **subfacets, int n)
4983 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4984 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4985 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4988 for (i = 0; i < n; i++) {
4989 ops[i].type = DPIF_OP_FLOW_DEL;
4990 ops[i].u.flow_del.key = subfacets[i]->key;
4991 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4992 ops[i].u.flow_del.stats = &stats[i];
4996 dpif_operate(backer->dpif, opsp, n);
4997 for (i = 0; i < n; i++) {
4998 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4999 subfacets[i]->path = SF_NOT_INSTALLED;
5000 subfacet_destroy(subfacets[i]);
5005 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5006 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5007 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5008 * since 'subfacet' was last updated.
5010 * Returns 0 if successful, otherwise a positive errno value. */
5012 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5013 struct dpif_flow_stats *stats)
5015 struct facet *facet = subfacet->facet;
5016 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5017 const struct nlattr *actions = odp_actions->data;
5018 size_t actions_len = odp_actions->size;
5019 struct odputil_keybuf maskbuf;
5022 uint64_t slow_path_stub[128 / 8];
5023 enum dpif_flow_put_flags flags;
5026 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5029 flags |= DPIF_FP_ZERO_STATS;
5032 if (path == SF_SLOW_PATH) {
5033 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5034 slow_path_stub, sizeof slow_path_stub,
5035 &actions, &actions_len);
5038 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5039 if (enable_megaflows) {
5040 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5041 &facet->flow, UINT32_MAX);
5044 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5045 subfacet->key_len, mask.data, mask.size,
5046 actions, actions_len, stats);
5049 subfacet_reset_dp_stats(subfacet, stats);
5053 COVERAGE_INC(subfacet_install_fail);
5055 subfacet->path = path;
5060 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5062 subfacet_uninstall(struct subfacet *subfacet)
5064 if (subfacet->path != SF_NOT_INSTALLED) {
5065 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5066 struct dpif_flow_stats stats;
5069 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5070 subfacet->key_len, &stats);
5071 subfacet_reset_dp_stats(subfacet, &stats);
5073 subfacet_update_stats(subfacet, &stats);
5075 subfacet->path = SF_NOT_INSTALLED;
5077 ovs_assert(subfacet->dp_packet_count == 0);
5078 ovs_assert(subfacet->dp_byte_count == 0);
5082 /* Resets 'subfacet''s datapath statistics counters. This should be called
5083 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5084 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5085 * was reset in the datapath. 'stats' will be modified to include only
5086 * statistics new since 'subfacet' was last updated. */
5088 subfacet_reset_dp_stats(struct subfacet *subfacet,
5089 struct dpif_flow_stats *stats)
5092 && subfacet->dp_packet_count <= stats->n_packets
5093 && subfacet->dp_byte_count <= stats->n_bytes) {
5094 stats->n_packets -= subfacet->dp_packet_count;
5095 stats->n_bytes -= subfacet->dp_byte_count;
5098 subfacet->dp_packet_count = 0;
5099 subfacet->dp_byte_count = 0;
5102 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5104 * Because of the meaning of a subfacet's counters, it only makes sense to do
5105 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5106 * represents a packet that was sent by hand or if it represents statistics
5107 * that have been cleared out of the datapath. */
5109 subfacet_update_stats(struct subfacet *subfacet,
5110 const struct dpif_flow_stats *stats)
5112 if (stats->n_packets || stats->used > subfacet->used) {
5113 struct facet *facet = subfacet->facet;
5115 subfacet->used = MAX(subfacet->used, stats->used);
5116 facet->used = MAX(facet->used, stats->used);
5117 facet->packet_count += stats->n_packets;
5118 facet->byte_count += stats->n_bytes;
5119 facet->tcp_flags |= stats->tcp_flags;
5125 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5126 * the fields that were relevant as part of the lookup. */
5127 static struct rule_dpif *
5128 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5129 struct flow_wildcards *wc)
5131 struct rule_dpif *rule;
5133 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5138 return rule_dpif_miss_rule(ofproto, flow);
5142 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5143 const struct flow *flow, struct flow_wildcards *wc,
5146 struct cls_rule *cls_rule;
5147 struct classifier *cls;
5150 if (table_id >= N_TABLES) {
5155 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5156 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5159 cls = &ofproto->up.tables[table_id].cls;
5160 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5161 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5162 /* We must pretend that transport ports are unavailable. */
5163 struct flow ofpc_normal_flow = *flow;
5164 ofpc_normal_flow.tp_src = htons(0);
5165 ofpc_normal_flow.tp_dst = htons(0);
5166 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5167 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5168 cls_rule = &ofproto->drop_frags_rule->up.cr;
5170 flow_wildcards_init_exact(wc);
5173 cls_rule = classifier_lookup(cls, flow, wc);
5175 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5179 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5181 struct ofport_dpif *port;
5183 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5185 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5186 flow->in_port.ofp_port);
5187 return ofproto->miss_rule;
5190 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5191 return ofproto->no_packet_in_rule;
5193 return ofproto->miss_rule;
5197 complete_operation(struct rule_dpif *rule)
5199 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5201 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5203 struct dpif_completion *c = xmalloc(sizeof *c);
5204 c->op = rule->up.pending;
5205 list_push_back(&ofproto->completions, &c->list_node);
5207 ofoperation_complete(rule->up.pending, 0);
5211 static struct rule *
5214 struct rule_dpif *rule = xmalloc(sizeof *rule);
5219 rule_dealloc(struct rule *rule_)
5221 struct rule_dpif *rule = rule_dpif_cast(rule_);
5226 rule_construct(struct rule *rule_)
5228 struct rule_dpif *rule = rule_dpif_cast(rule_);
5229 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5230 ovs_mutex_lock(&rule->stats_mutex);
5231 rule->packet_count = 0;
5232 rule->byte_count = 0;
5233 ovs_mutex_unlock(&rule->stats_mutex);
5234 complete_operation(rule);
5239 rule_destruct(struct rule *rule_)
5241 struct rule_dpif *rule = rule_dpif_cast(rule_);
5242 complete_operation(rule);
5243 ovs_mutex_destroy(&rule->stats_mutex);
5247 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5249 struct rule_dpif *rule = rule_dpif_cast(rule_);
5251 /* push_all_stats() can handle flow misses which, when using the learn
5252 * action, can cause rules to be added and deleted. This can corrupt our
5253 * caller's datastructures which assume that rule_get_stats() doesn't have
5254 * an impact on the flow table. To be safe, we disable miss handling. */
5255 push_all_stats__(false);
5257 /* Start from historical data for 'rule' itself that are no longer tracked
5258 * in facets. This counts, for example, facets that have expired. */
5259 ovs_mutex_lock(&rule->stats_mutex);
5260 *packets = rule->packet_count;
5261 *bytes = rule->byte_count;
5262 ovs_mutex_unlock(&rule->stats_mutex);
5266 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5267 struct ofpbuf *packet)
5269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5270 struct dpif_flow_stats stats;
5271 struct xlate_out xout;
5272 struct xlate_in xin;
5274 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5275 rule_credit_stats(rule, &stats);
5277 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5278 xin.resubmit_stats = &stats;
5279 xlate_actions(&xin, &xout);
5281 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5282 xout.odp_actions.size, packet);
5284 xlate_out_uninit(&xout);
5288 rule_execute(struct rule *rule, const struct flow *flow,
5289 struct ofpbuf *packet)
5291 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5292 ofpbuf_delete(packet);
5297 rule_modify_actions(struct rule *rule_)
5299 struct rule_dpif *rule = rule_dpif_cast(rule_);
5301 complete_operation(rule);
5304 /* Sends 'packet' out 'ofport'.
5305 * May modify 'packet'.
5306 * Returns 0 if successful, otherwise a positive errno value. */
5308 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5310 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5311 uint64_t odp_actions_stub[1024 / 8];
5312 struct ofpbuf key, odp_actions;
5313 struct dpif_flow_stats stats;
5314 struct odputil_keybuf keybuf;
5315 struct ofpact_output output;
5316 struct xlate_out xout;
5317 struct xlate_in xin;
5319 union flow_in_port in_port_;
5322 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5323 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5325 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5326 in_port_.ofp_port = OFPP_NONE;
5327 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5328 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5330 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5332 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5333 output.port = ofport->up.ofp_port;
5336 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5337 xin.ofpacts_len = sizeof output;
5338 xin.ofpacts = &output.ofpact;
5339 xin.resubmit_stats = &stats;
5340 xlate_actions(&xin, &xout);
5342 error = dpif_execute(ofproto->backer->dpif,
5344 xout.odp_actions.data, xout.odp_actions.size,
5346 xlate_out_uninit(&xout);
5349 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5350 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5351 ovs_strerror(error));
5354 ofproto->stats.tx_packets++;
5355 ofproto->stats.tx_bytes += packet->size;
5359 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5360 * The action will state 'slow' as the reason that the action is in the slow
5361 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5362 * dump-flows" output to see why a flow is in the slow path.)
5364 * The 'stub_size' bytes in 'stub' will be used to store the action.
5365 * 'stub_size' must be large enough for the action.
5367 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5370 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5371 enum slow_path_reason slow,
5372 uint64_t *stub, size_t stub_size,
5373 const struct nlattr **actionsp, size_t *actions_lenp)
5375 union user_action_cookie cookie;
5378 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5379 cookie.slow_path.unused = 0;
5380 cookie.slow_path.reason = slow;
5382 ofpbuf_use_stack(&buf, stub, stub_size);
5383 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5384 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5386 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5388 odp_port_t odp_port;
5391 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5392 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5393 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5395 *actionsp = buf.data;
5396 *actions_lenp = buf.size;
5400 set_frag_handling(struct ofproto *ofproto_,
5401 enum ofp_config_flags frag_handling)
5403 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5404 if (frag_handling != OFPC_FRAG_REASM) {
5405 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5413 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5414 const struct flow *flow,
5415 const struct ofpact *ofpacts, size_t ofpacts_len)
5417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5418 struct odputil_keybuf keybuf;
5419 struct dpif_flow_stats stats;
5420 struct xlate_out xout;
5421 struct xlate_in xin;
5425 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5426 odp_flow_key_from_flow(&key, flow,
5427 ofp_port_to_odp_port(ofproto,
5428 flow->in_port.ofp_port));
5430 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5432 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5433 xin.resubmit_stats = &stats;
5434 xin.ofpacts_len = ofpacts_len;
5435 xin.ofpacts = ofpacts;
5437 xlate_actions(&xin, &xout);
5438 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5439 xout.odp_actions.data, xout.odp_actions.size, packet);
5440 xlate_out_uninit(&xout);
5448 set_netflow(struct ofproto *ofproto_,
5449 const struct netflow_options *netflow_options)
5451 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5453 if (netflow_options) {
5454 if (!ofproto->netflow) {
5455 ofproto->netflow = netflow_create();
5456 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5458 return netflow_set_options(ofproto->netflow, netflow_options);
5459 } else if (ofproto->netflow) {
5460 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5461 netflow_destroy(ofproto->netflow);
5462 ofproto->netflow = NULL;
5469 get_netflow_ids(const struct ofproto *ofproto_,
5470 uint8_t *engine_type, uint8_t *engine_id)
5472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5474 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5478 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5480 if (!facet_is_controller_flow(facet) &&
5481 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5482 struct subfacet *subfacet;
5483 struct ofexpired expired;
5485 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5486 if (subfacet->path == SF_FAST_PATH) {
5487 struct dpif_flow_stats stats;
5489 subfacet_install(subfacet, &facet->xout.odp_actions,
5491 subfacet_update_stats(subfacet, &stats);
5495 expired.flow = facet->flow;
5496 expired.packet_count = facet->packet_count;
5497 expired.byte_count = facet->byte_count;
5498 expired.used = facet->used;
5499 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5504 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5506 struct cls_cursor cursor;
5507 struct facet *facet;
5509 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5510 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5511 send_active_timeout(ofproto, facet);
5515 static struct ofproto_dpif *
5516 ofproto_dpif_lookup(const char *name)
5518 struct ofproto_dpif *ofproto;
5520 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5521 hash_string(name, 0), &all_ofproto_dpifs) {
5522 if (!strcmp(ofproto->up.name, name)) {
5530 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5531 const char *argv[], void *aux OVS_UNUSED)
5533 struct ofproto_dpif *ofproto;
5536 ofproto = ofproto_dpif_lookup(argv[1]);
5538 unixctl_command_reply_error(conn, "no such bridge");
5541 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5542 mac_learning_flush(ofproto->ml);
5543 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5545 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5546 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5547 mac_learning_flush(ofproto->ml);
5548 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5552 unixctl_command_reply(conn, "table successfully flushed");
5555 static struct ofport_dpif *
5556 ofbundle_get_a_port(const struct ofbundle *bundle)
5558 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5563 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5564 const char *argv[], void *aux OVS_UNUSED)
5566 struct ds ds = DS_EMPTY_INITIALIZER;
5567 const struct ofproto_dpif *ofproto;
5568 const struct mac_entry *e;
5570 ofproto = ofproto_dpif_lookup(argv[1]);
5572 unixctl_command_reply_error(conn, "no such bridge");
5576 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5577 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5578 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5579 struct ofbundle *bundle = e->port.p;
5580 char name[OFP_MAX_PORT_NAME_LEN];
5582 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5584 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5585 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5586 mac_entry_age(ofproto->ml, e));
5588 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5589 unixctl_command_reply(conn, ds_cstr(&ds));
5594 struct xlate_out xout;
5595 struct xlate_in xin;
5601 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5603 ds_put_char_multiple(result, '\t', level);
5605 ds_put_cstr(result, "No match\n");
5609 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5610 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5611 cls_rule_format(&rule->up.cr, result);
5612 ds_put_char(result, '\n');
5614 ds_put_char_multiple(result, '\t', level);
5615 ds_put_cstr(result, "OpenFlow ");
5616 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5617 ds_put_char(result, '\n');
5621 trace_format_flow(struct ds *result, int level, const char *title,
5622 struct trace_ctx *trace)
5624 ds_put_char_multiple(result, '\t', level);
5625 ds_put_format(result, "%s: ", title);
5626 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5627 ds_put_cstr(result, "unchanged");
5629 flow_format(result, &trace->xin.flow);
5630 trace->flow = trace->xin.flow;
5632 ds_put_char(result, '\n');
5636 trace_format_regs(struct ds *result, int level, const char *title,
5637 struct trace_ctx *trace)
5641 ds_put_char_multiple(result, '\t', level);
5642 ds_put_format(result, "%s:", title);
5643 for (i = 0; i < FLOW_N_REGS; i++) {
5644 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5646 ds_put_char(result, '\n');
5650 trace_format_odp(struct ds *result, int level, const char *title,
5651 struct trace_ctx *trace)
5653 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5655 ds_put_char_multiple(result, '\t', level);
5656 ds_put_format(result, "%s: ", title);
5657 format_odp_actions(result, odp_actions->data, odp_actions->size);
5658 ds_put_char(result, '\n');
5662 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5664 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5665 struct ds *result = trace->result;
5667 ds_put_char(result, '\n');
5668 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5669 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5670 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5671 trace_format_rule(result, recurse + 1, rule);
5675 trace_report(struct xlate_in *xin, const char *s, int recurse)
5677 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5678 struct ds *result = trace->result;
5680 ds_put_char_multiple(result, '\t', recurse);
5681 ds_put_cstr(result, s);
5682 ds_put_char(result, '\n');
5686 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5687 void *aux OVS_UNUSED)
5689 const struct dpif_backer *backer;
5690 struct ofproto_dpif *ofproto;
5691 struct ofpbuf odp_key, odp_mask;
5692 struct ofpbuf *packet;
5700 ofpbuf_init(&odp_key, 0);
5701 ofpbuf_init(&odp_mask, 0);
5703 /* Handle "-generate" or a hex string as the last argument. */
5704 if (!strcmp(argv[argc - 1], "-generate")) {
5705 packet = ofpbuf_new(0);
5708 const char *error = eth_from_hex(argv[argc - 1], &packet);
5711 } else if (argc == 4) {
5712 /* The 3-argument form must end in "-generate' or a hex string. */
5713 unixctl_command_reply_error(conn, error);
5718 /* Parse the flow and determine whether a datapath or
5719 * bridge is specified. If function odp_flow_key_from_string()
5720 * returns 0, the flow is a odp_flow. If function
5721 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5722 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5723 /* If the odp_flow is the second argument,
5724 * the datapath name is the first argument. */
5726 const char *dp_type;
5727 if (!strncmp(argv[1], "ovs-", 4)) {
5728 dp_type = argv[1] + 4;
5732 backer = shash_find_data(&all_dpif_backers, dp_type);
5734 unixctl_command_reply_error(conn, "Cannot find datapath "
5739 /* No datapath name specified, so there should be only one
5741 struct shash_node *node;
5742 if (shash_count(&all_dpif_backers) != 1) {
5743 unixctl_command_reply_error(conn, "Must specify datapath "
5744 "name, there is more than one type of datapath");
5747 node = shash_first(&all_dpif_backers);
5748 backer = node->data;
5751 /* Extract the ofproto_dpif object from the ofproto_receive()
5753 if (ofproto_receive(backer, NULL, odp_key.data,
5754 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5755 unixctl_command_reply_error(conn, "Invalid datapath flow");
5758 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5759 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5761 unixctl_command_reply_error(conn, "Must specify bridge name");
5765 ofproto = ofproto_dpif_lookup(argv[1]);
5767 unixctl_command_reply_error(conn, "Unknown bridge name");
5771 unixctl_command_reply_error(conn, "Bad flow syntax");
5775 /* Generate a packet, if requested. */
5777 if (!packet->size) {
5778 flow_compose(packet, &flow);
5780 union flow_in_port in_port_;
5782 in_port_ = flow.in_port;
5783 ds_put_cstr(&result, "Packet: ");
5784 s = ofp_packet_to_string(packet->data, packet->size);
5785 ds_put_cstr(&result, s);
5788 /* Use the metadata from the flow and the packet argument
5789 * to reconstruct the flow. */
5790 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5795 ofproto_trace(ofproto, &flow, packet, &result);
5796 unixctl_command_reply(conn, ds_cstr(&result));
5799 ds_destroy(&result);
5800 ofpbuf_delete(packet);
5801 ofpbuf_uninit(&odp_key);
5802 ofpbuf_uninit(&odp_mask);
5806 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5807 const struct ofpbuf *packet, struct ds *ds)
5809 struct rule_dpif *rule;
5811 ds_put_cstr(ds, "Flow: ");
5812 flow_format(ds, flow);
5813 ds_put_char(ds, '\n');
5815 rule = rule_dpif_lookup(ofproto, flow, NULL);
5817 trace_format_rule(ds, 0, rule);
5818 if (rule == ofproto->miss_rule) {
5819 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5820 } else if (rule == ofproto->no_packet_in_rule) {
5821 ds_put_cstr(ds, "\nNo match, packets dropped because "
5822 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5823 } else if (rule == ofproto->drop_frags_rule) {
5824 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5825 "and the fragment handling mode is \"drop\".\n");
5829 uint64_t odp_actions_stub[1024 / 8];
5830 struct ofpbuf odp_actions;
5831 struct trace_ctx trace;
5835 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5838 ofpbuf_use_stub(&odp_actions,
5839 odp_actions_stub, sizeof odp_actions_stub);
5840 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5841 trace.xin.resubmit_hook = trace_resubmit;
5842 trace.xin.report_hook = trace_report;
5844 xlate_actions(&trace.xin, &trace.xout);
5846 ds_put_char(ds, '\n');
5847 trace_format_flow(ds, 0, "Final flow", &trace);
5849 match_init(&match, flow, &trace.xout.wc);
5850 ds_put_cstr(ds, "Relevant fields: ");
5851 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5852 ds_put_char(ds, '\n');
5854 ds_put_cstr(ds, "Datapath actions: ");
5855 format_odp_actions(ds, trace.xout.odp_actions.data,
5856 trace.xout.odp_actions.size);
5858 if (trace.xout.slow) {
5859 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5860 "slow path because it:");
5861 switch (trace.xout.slow) {
5863 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5866 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5869 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5872 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5874 case SLOW_CONTROLLER:
5875 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5876 "to the OpenFlow controller.");
5883 xlate_out_uninit(&trace.xout);
5888 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5889 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5892 unixctl_command_reply(conn, NULL);
5896 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5897 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5900 unixctl_command_reply(conn, NULL);
5903 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5904 * 'reply' describing the results. */
5906 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5908 struct cls_cursor cursor;
5909 struct facet *facet;
5913 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5914 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5915 if (!facet_check_consistency(facet)) {
5920 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5924 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5925 ofproto->up.name, errors);
5927 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5932 ofproto_dpif_self_check(struct unixctl_conn *conn,
5933 int argc, const char *argv[], void *aux OVS_UNUSED)
5935 struct ds reply = DS_EMPTY_INITIALIZER;
5936 struct ofproto_dpif *ofproto;
5939 ofproto = ofproto_dpif_lookup(argv[1]);
5941 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5942 "ofproto/list for help)");
5945 ofproto_dpif_self_check__(ofproto, &reply);
5947 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5948 ofproto_dpif_self_check__(ofproto, &reply);
5952 unixctl_command_reply(conn, ds_cstr(&reply));
5956 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5957 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5958 * to destroy 'ofproto_shash' and free the returned value. */
5959 static const struct shash_node **
5960 get_ofprotos(struct shash *ofproto_shash)
5962 const struct ofproto_dpif *ofproto;
5964 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5965 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5966 shash_add_nocopy(ofproto_shash, name, ofproto);
5969 return shash_sort(ofproto_shash);
5973 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5974 const char *argv[] OVS_UNUSED,
5975 void *aux OVS_UNUSED)
5977 struct ds ds = DS_EMPTY_INITIALIZER;
5978 struct shash ofproto_shash;
5979 const struct shash_node **sorted_ofprotos;
5982 shash_init(&ofproto_shash);
5983 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5984 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5985 const struct shash_node *node = sorted_ofprotos[i];
5986 ds_put_format(&ds, "%s\n", node->name);
5989 shash_destroy(&ofproto_shash);
5990 free(sorted_ofprotos);
5992 unixctl_command_reply(conn, ds_cstr(&ds));
5997 show_dp_rates(struct ds *ds, const char *heading,
5998 const struct avg_subfacet_rates *rates)
6000 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6001 heading, rates->add_rate, rates->del_rate);
6005 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6007 const struct shash_node **ofprotos;
6008 struct ofproto_dpif *ofproto;
6009 struct shash ofproto_shash;
6010 uint64_t n_hit, n_missed;
6011 long long int minutes;
6014 n_hit = n_missed = 0;
6015 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6016 if (ofproto->backer == backer) {
6017 n_missed += ofproto->n_missed;
6018 n_hit += ofproto->n_hit;
6022 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6023 dpif_name(backer->dpif), n_hit, n_missed);
6024 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6025 " life span: %lldms\n", hmap_count(&backer->subfacets),
6026 backer->avg_n_subfacet, backer->max_n_subfacet,
6027 backer->avg_subfacet_life);
6029 minutes = (time_msec() - backer->created) / (1000 * 60);
6030 if (minutes >= 60) {
6031 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6033 if (minutes >= 60 * 24) {
6034 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6036 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6038 shash_init(&ofproto_shash);
6039 ofprotos = get_ofprotos(&ofproto_shash);
6040 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6041 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6042 const struct shash_node **ports;
6045 if (ofproto->backer != backer) {
6049 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6050 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6052 ports = shash_sort(&ofproto->up.port_by_name);
6053 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6054 const struct shash_node *node = ports[j];
6055 struct ofport *ofport = node->data;
6057 odp_port_t odp_port;
6059 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6062 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6063 if (odp_port != ODPP_NONE) {
6064 ds_put_format(ds, "%"PRIu32":", odp_port);
6066 ds_put_cstr(ds, "none:");
6069 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6072 if (!netdev_get_config(ofport->netdev, &config)) {
6073 const struct smap_node **nodes;
6076 nodes = smap_sort(&config);
6077 for (i = 0; i < smap_count(&config); i++) {
6078 const struct smap_node *node = nodes[i];
6079 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6080 node->key, node->value);
6084 smap_destroy(&config);
6086 ds_put_char(ds, ')');
6087 ds_put_char(ds, '\n');
6091 shash_destroy(&ofproto_shash);
6096 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6097 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6099 struct ds ds = DS_EMPTY_INITIALIZER;
6100 const struct shash_node **backers;
6103 backers = shash_sort(&all_dpif_backers);
6104 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6105 dpif_show_backer(backers[i]->data, &ds);
6109 unixctl_command_reply(conn, ds_cstr(&ds));
6113 /* Dump the megaflow (facet) cache. This is useful to check the
6114 * correctness of flow wildcarding, since the same mechanism is used for
6115 * both xlate caching and kernel wildcarding.
6117 * It's important to note that in the output the flow description uses
6118 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6120 * This command is only needed for advanced debugging, so it's not
6121 * documented in the man page. */
6123 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6124 int argc OVS_UNUSED, const char *argv[],
6125 void *aux OVS_UNUSED)
6127 struct ds ds = DS_EMPTY_INITIALIZER;
6128 const struct ofproto_dpif *ofproto;
6129 long long int now = time_msec();
6130 struct cls_cursor cursor;
6131 struct facet *facet;
6133 ofproto = ofproto_dpif_lookup(argv[1]);
6135 unixctl_command_reply_error(conn, "no such bridge");
6139 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6140 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6141 cls_rule_format(&facet->cr, &ds);
6142 ds_put_cstr(&ds, ", ");
6143 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6144 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6145 ds_put_cstr(&ds, "Datapath actions: ");
6146 if (facet->xout.slow) {
6147 uint64_t slow_path_stub[128 / 8];
6148 const struct nlattr *actions;
6151 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6152 slow_path_stub, sizeof slow_path_stub,
6153 &actions, &actions_len);
6154 format_odp_actions(&ds, actions, actions_len);
6156 format_odp_actions(&ds, facet->xout.odp_actions.data,
6157 facet->xout.odp_actions.size);
6159 ds_put_cstr(&ds, "\n");
6162 ds_chomp(&ds, '\n');
6163 unixctl_command_reply(conn, ds_cstr(&ds));
6167 /* Disable using the megaflows.
6169 * This command is only needed for advanced debugging, so it's not
6170 * documented in the man page. */
6172 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6173 int argc OVS_UNUSED,
6174 const char *argv[] OVS_UNUSED,
6175 void *aux OVS_UNUSED)
6177 struct ofproto_dpif *ofproto;
6179 enable_megaflows = false;
6181 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6182 flush(&ofproto->up);
6185 unixctl_command_reply(conn, "megaflows disabled");
6188 /* Re-enable using megaflows.
6190 * This command is only needed for advanced debugging, so it's not
6191 * documented in the man page. */
6193 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6194 int argc OVS_UNUSED,
6195 const char *argv[] OVS_UNUSED,
6196 void *aux OVS_UNUSED)
6198 struct ofproto_dpif *ofproto;
6200 enable_megaflows = true;
6202 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6203 flush(&ofproto->up);
6206 unixctl_command_reply(conn, "megaflows enabled");
6210 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6211 int argc OVS_UNUSED, const char *argv[],
6212 void *aux OVS_UNUSED)
6214 struct ds ds = DS_EMPTY_INITIALIZER;
6215 const struct ofproto_dpif *ofproto;
6216 struct subfacet *subfacet;
6218 ofproto = ofproto_dpif_lookup(argv[1]);
6220 unixctl_command_reply_error(conn, "no such bridge");
6224 update_stats(ofproto->backer);
6226 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6227 struct facet *facet = subfacet->facet;
6228 struct odputil_keybuf maskbuf;
6231 if (facet->ofproto != ofproto) {
6235 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6236 if (enable_megaflows) {
6237 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6238 &facet->flow, UINT32_MAX);
6241 odp_flow_format(subfacet->key, subfacet->key_len,
6242 mask.data, mask.size, &ds, false);
6244 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6245 subfacet->dp_packet_count, subfacet->dp_byte_count);
6246 if (subfacet->used) {
6247 ds_put_format(&ds, "%.3fs",
6248 (time_msec() - subfacet->used) / 1000.0);
6250 ds_put_format(&ds, "never");
6252 if (subfacet->facet->tcp_flags) {
6253 ds_put_cstr(&ds, ", flags:");
6254 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6257 ds_put_cstr(&ds, ", actions:");
6258 if (facet->xout.slow) {
6259 uint64_t slow_path_stub[128 / 8];
6260 const struct nlattr *actions;
6263 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6264 slow_path_stub, sizeof slow_path_stub,
6265 &actions, &actions_len);
6266 format_odp_actions(&ds, actions, actions_len);
6268 format_odp_actions(&ds, facet->xout.odp_actions.data,
6269 facet->xout.odp_actions.size);
6271 ds_put_char(&ds, '\n');
6274 unixctl_command_reply(conn, ds_cstr(&ds));
6279 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6280 int argc OVS_UNUSED, const char *argv[],
6281 void *aux OVS_UNUSED)
6283 struct ds ds = DS_EMPTY_INITIALIZER;
6284 struct ofproto_dpif *ofproto;
6286 ofproto = ofproto_dpif_lookup(argv[1]);
6288 unixctl_command_reply_error(conn, "no such bridge");
6292 flush(&ofproto->up);
6294 unixctl_command_reply(conn, ds_cstr(&ds));
6299 ofproto_dpif_unixctl_init(void)
6301 static bool registered;
6307 unixctl_command_register(
6309 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6310 1, 3, ofproto_unixctl_trace, NULL);
6311 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6312 ofproto_unixctl_fdb_flush, NULL);
6313 unixctl_command_register("fdb/show", "bridge", 1, 1,
6314 ofproto_unixctl_fdb_show, NULL);
6315 unixctl_command_register("ofproto/clog", "", 0, 0,
6316 ofproto_dpif_clog, NULL);
6317 unixctl_command_register("ofproto/unclog", "", 0, 0,
6318 ofproto_dpif_unclog, NULL);
6319 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6320 ofproto_dpif_self_check, NULL);
6321 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6322 ofproto_unixctl_dpif_dump_dps, NULL);
6323 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6325 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6326 ofproto_unixctl_dpif_dump_flows, NULL);
6327 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6328 ofproto_unixctl_dpif_del_flows, NULL);
6329 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6330 ofproto_unixctl_dpif_dump_megaflows, NULL);
6331 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6332 ofproto_unixctl_dpif_disable_megaflows, NULL);
6333 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6334 ofproto_unixctl_dpif_enable_megaflows, NULL);
6337 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6339 * This is deprecated. It is only for compatibility with broken device drivers
6340 * in old versions of Linux that do not properly support VLANs when VLAN
6341 * devices are not used. When broken device drivers are no longer in
6342 * widespread use, we will delete these interfaces. */
6345 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6348 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6350 if (realdev_ofp_port == ofport->realdev_ofp_port
6351 && vid == ofport->vlandev_vid) {
6355 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6357 if (ofport->realdev_ofp_port) {
6360 if (realdev_ofp_port && ofport->bundle) {
6361 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6362 * themselves be part of a bundle. */
6363 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6366 ofport->realdev_ofp_port = realdev_ofp_port;
6367 ofport->vlandev_vid = vid;
6369 if (realdev_ofp_port) {
6370 vsp_add(ofport, realdev_ofp_port, vid);
6377 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6379 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6383 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6385 return !hmap_is_empty(&ofproto->realdev_vid_map);
6388 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6389 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6390 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6391 * 'vlan_tci' 9, it would return the port number of eth0.9.
6393 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6394 * function just returns its 'realdev_ofp_port' argument. */
6396 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6397 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6399 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6400 int vid = vlan_tci_to_vid(vlan_tci);
6401 const struct vlan_splinter *vsp;
6403 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6404 hash_realdev_vid(realdev_ofp_port, vid),
6405 &ofproto->realdev_vid_map) {
6406 if (vsp->realdev_ofp_port == realdev_ofp_port
6407 && vsp->vid == vid) {
6408 return vsp->vlandev_ofp_port;
6412 return realdev_ofp_port;
6415 static struct vlan_splinter *
6416 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6418 struct vlan_splinter *vsp;
6420 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6421 hash_ofp_port(vlandev_ofp_port),
6422 &ofproto->vlandev_map) {
6423 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6431 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6432 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6433 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6434 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6435 * eth0 and store 9 in '*vid'.
6437 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6438 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6441 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6442 ofp_port_t vlandev_ofp_port, int *vid)
6444 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6445 const struct vlan_splinter *vsp;
6447 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6452 return vsp->realdev_ofp_port;
6458 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6459 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6460 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6461 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6462 * always the case unless VLAN splinters are enabled), returns false without
6463 * making any changes. */
6465 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6470 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6475 /* Cause the flow to be processed as if it came in on the real device with
6476 * the VLAN device's VLAN ID. */
6477 flow->in_port.ofp_port = realdev;
6478 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6483 vsp_remove(struct ofport_dpif *port)
6485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6486 struct vlan_splinter *vsp;
6488 vsp = vlandev_find(ofproto, port->up.ofp_port);
6490 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6491 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6494 port->realdev_ofp_port = 0;
6496 VLOG_ERR("missing vlan device record");
6501 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6505 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6506 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6507 == realdev_ofp_port)) {
6508 struct vlan_splinter *vsp;
6510 vsp = xmalloc(sizeof *vsp);
6511 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6512 hash_ofp_port(port->up.ofp_port));
6513 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6514 hash_realdev_vid(realdev_ofp_port, vid));
6515 vsp->realdev_ofp_port = realdev_ofp_port;
6516 vsp->vlandev_ofp_port = port->up.ofp_port;
6519 port->realdev_ofp_port = realdev_ofp_port;
6521 VLOG_ERR("duplicate vlan device record");
6526 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6528 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6529 return ofport ? ofport->odp_port : ODPP_NONE;
6532 static struct ofport_dpif *
6533 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6535 struct ofport_dpif *port;
6537 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6538 &backer->odp_to_ofport_map) {
6539 if (port->odp_port == odp_port) {
6548 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6550 struct ofport_dpif *port;
6552 port = odp_port_to_ofport(ofproto->backer, odp_port);
6553 if (port && &ofproto->up == port->up.ofproto) {
6554 return port->up.ofp_port;
6560 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6561 * most heavily weighted element. 'base' designates the rate of decay: after
6562 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6565 exp_mavg(double *avg, int base, double new)
6567 *avg = (*avg * (base - 1) + new) / base;
6571 update_moving_averages(struct dpif_backer *backer)
6573 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6574 long long int minutes = (time_msec() - backer->created) / min_ms;
6577 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6579 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6582 backer->lifetime.add_rate = 0.0;
6583 backer->lifetime.del_rate = 0.0;
6586 /* Update hourly averages on the minute boundaries. */
6587 if (time_msec() - backer->last_minute >= min_ms) {
6588 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6589 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6591 /* Update daily averages on the hour boundaries. */
6592 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6593 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6594 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6597 backer->total_subfacet_add_count += backer->subfacet_add_count;
6598 backer->total_subfacet_del_count += backer->subfacet_del_count;
6599 backer->subfacet_add_count = 0;
6600 backer->subfacet_del_count = 0;
6601 backer->last_minute += min_ms;
6605 const struct ofproto_class ofproto_dpif_class = {
6640 port_is_lacp_current,
6641 NULL, /* rule_choose_table */
6648 rule_modify_actions,
6662 get_stp_port_status,
6669 is_mirror_output_bundle,
6670 forward_bpdu_changed,
6671 set_mac_table_config,
6673 NULL, /* meter_get_features */
6674 NULL, /* meter_set */
6675 NULL, /* meter_get */
6676 NULL, /* meter_del */