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(packet_in_overflow);
75 COVERAGE_DEFINE(flow_mod_overflow);
77 /* Number of implemented OpenFlow tables. */
78 enum { N_TABLES = 255 };
79 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
80 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
85 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
88 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
89 struct ofproto_dpif *ofproto; /* Owning ofproto. */
90 void *aux; /* Key supplied by ofproto's client. */
91 char *name; /* Identifier for log messages. */
94 struct list ports; /* Contains "struct ofport"s. */
95 enum port_vlan_mode vlan_mode; /* VLAN mode */
96 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
97 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
98 * NULL if all VLANs are trunked. */
99 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
100 struct bond *bond; /* Nonnull iff more than one port. */
101 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
104 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
107 static void bundle_remove(struct ofport *);
108 static void bundle_update(struct ofbundle *);
109 static void bundle_destroy(struct ofbundle *);
110 static void bundle_del_port(struct ofport_dpif *);
111 static void bundle_run(struct ofbundle *);
112 static void bundle_wait(struct ofbundle *);
114 static void stp_run(struct ofproto_dpif *ofproto);
115 static void stp_wait(struct ofproto_dpif *ofproto);
116 static int set_stp_port(struct ofport *,
117 const struct ofproto_port_stp_settings *);
119 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
120 enum slow_path_reason,
121 uint64_t *stub, size_t stub_size,
122 const struct nlattr **actionsp,
123 size_t *actions_lenp);
125 /* A subfacet (see "struct subfacet" below) has three possible installation
128 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
129 * case just after the subfacet is created, just before the subfacet is
130 * destroyed, or if the datapath returns an error when we try to install a
133 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
135 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
136 * ofproto_dpif is installed in the datapath.
139 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
140 SF_FAST_PATH, /* Full actions are installed. */
141 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
144 /* A dpif flow and actions associated with a facet.
146 * See also the large comment on struct facet. */
149 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
150 struct list list_node; /* In struct facet's 'facets' list. */
151 struct facet *facet; /* Owning facet. */
152 struct dpif_backer *backer; /* Owning backer. */
154 enum odp_key_fitness key_fitness;
158 long long int used; /* Time last used; time created if not used. */
159 long long int created; /* Time created. */
161 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
162 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
164 enum subfacet_path path; /* Installed in datapath? */
167 #define SUBFACET_DESTROY_MAX_BATCH 50
169 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
171 static struct subfacet *subfacet_find(struct dpif_backer *,
172 const struct nlattr *key, size_t key_len,
174 static void subfacet_destroy(struct subfacet *);
175 static void subfacet_destroy__(struct subfacet *);
176 static void subfacet_destroy_batch(struct dpif_backer *,
177 struct subfacet **, int n);
178 static void subfacet_reset_dp_stats(struct subfacet *,
179 struct dpif_flow_stats *);
180 static void subfacet_update_stats(struct subfacet *,
181 const struct dpif_flow_stats *);
182 static int subfacet_install(struct subfacet *,
183 const struct ofpbuf *odp_actions,
184 struct dpif_flow_stats *);
185 static void subfacet_uninstall(struct subfacet *);
187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
189 * A facet associates a "struct flow", which represents the Open vSwitch
190 * userspace idea of an exact-match flow, with one or more subfacets.
191 * While the facet is created based on an exact-match flow, it is stored
192 * within the ofproto based on the wildcards that could be expressed
193 * based on the flow table and other configuration. (See the 'wc'
194 * description in "struct xlate_out" for more details.)
196 * Each subfacet tracks the datapath's idea of the flow equivalent to
197 * the facet. When the kernel module (or other dpif implementation) and
198 * Open vSwitch userspace agree on the definition of a flow key, there
199 * is exactly one subfacet per facet. If the dpif implementation
200 * supports more-specific flow matching than userspace, however, a facet
201 * can have more than one subfacet. Examples include the dpif
202 * implementation not supporting the same wildcards as userspace or some
203 * distinction in flow that userspace simply doesn't understand.
205 * Flow expiration works in terms of subfacets, so a facet must have at
206 * least one subfacet or it will never expire, leaking memory. */
209 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
210 struct ofproto_dpif *ofproto;
213 struct list subfacets;
214 long long int used; /* Time last used; time created if not used. */
217 struct flow flow; /* Flow of the creating subfacet. */
218 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
222 * - Do include packets and bytes sent "by hand", e.g. with
225 * - Do include packets and bytes that were obtained from the datapath
226 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
227 * DPIF_FP_ZERO_STATS).
229 * - Do not include packets or bytes that can be obtained from the
230 * datapath for any existing subfacet.
232 uint64_t packet_count; /* Number of packets received. */
233 uint64_t byte_count; /* Number of bytes received. */
235 /* Resubmit statistics. */
236 uint64_t prev_packet_count; /* Number of packets from last stats push. */
237 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
238 long long int prev_used; /* Used time from last stats push. */
241 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
242 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
243 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
245 struct xlate_out xout;
246 bool fail_open; /* Facet matched the fail open rule. */
248 /* Storage for a single subfacet, to reduce malloc() time and space
249 * overhead. (A facet always has at least one subfacet and in the common
250 * case has exactly one subfacet. However, 'one_subfacet' may not
251 * always be valid, since it could have been removed after newer
252 * subfacets were pushed onto the 'subfacets' list.) */
253 struct subfacet one_subfacet;
255 long long int learn_rl; /* Rate limiter for facet_learn(). */
258 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
260 struct dpif_flow_stats *);
261 static void facet_remove(struct facet *);
262 static void facet_free(struct facet *);
264 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
265 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
266 const struct flow *);
267 static bool facet_revalidate(struct facet *);
268 static bool facet_check_consistency(struct facet *);
270 static void facet_flush_stats(struct facet *);
272 static void facet_reset_counters(struct facet *);
273 static void facet_push_stats(struct facet *, bool may_learn);
274 static void facet_learn(struct facet *);
275 static void facet_account(struct facet *);
276 static void push_all_stats(void);
278 static bool facet_is_controller_flow(struct facet *);
281 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
285 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
286 struct list bundle_node; /* In struct ofbundle's "ports" list. */
287 struct cfm *cfm; /* Connectivity Fault Management, if any. */
288 struct bfd *bfd; /* BFD, if any. */
289 bool may_enable; /* May be enabled in bonds. */
290 bool is_tunnel; /* This port is a tunnel. */
291 long long int carrier_seq; /* Carrier status changes. */
292 struct ofport_dpif *peer; /* Peer if patch port. */
295 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
296 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
297 long long int stp_state_entered;
299 /* Queue to DSCP mapping. */
300 struct ofproto_port_queue *qdscp;
303 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
305 * This is deprecated. It is only for compatibility with broken device
306 * drivers in old versions of Linux that do not properly support VLANs when
307 * VLAN devices are not used. When broken device drivers are no longer in
308 * widespread use, we will delete these interfaces. */
309 ofp_port_t realdev_ofp_port;
313 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
315 * This is deprecated. It is only for compatibility with broken device drivers
316 * in old versions of Linux that do not properly support VLANs when VLAN
317 * devices are not used. When broken device drivers are no longer in
318 * widespread use, we will delete these interfaces. */
319 struct vlan_splinter {
320 struct hmap_node realdev_vid_node;
321 struct hmap_node vlandev_node;
322 ofp_port_t realdev_ofp_port;
323 ofp_port_t vlandev_ofp_port;
327 static void vsp_remove(struct ofport_dpif *);
328 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
330 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
333 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
336 static struct ofport_dpif *
337 ofport_dpif_cast(const struct ofport *ofport)
339 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
342 static void port_run(struct ofport_dpif *);
343 static void port_run_fast(struct ofport_dpif *);
344 static void port_wait(struct ofport_dpif *);
345 static int set_bfd(struct ofport *, const struct smap *);
346 static int set_cfm(struct ofport *, const struct cfm_settings *);
347 static void ofport_update_peer(struct ofport_dpif *);
348 static void run_fast_rl(void);
349 static int run_fast(struct ofproto *);
351 struct dpif_completion {
352 struct list list_node;
353 struct ofoperation *op;
356 /* Reasons that we might need to revalidate every facet, and corresponding
359 * A value of 0 means that there is no need to revalidate.
361 * It would be nice to have some cleaner way to integrate with coverage
362 * counters, but with only a few reasons I guess this is good enough for
364 enum revalidate_reason {
365 REV_RECONFIGURE = 1, /* Switch configuration changed. */
366 REV_STP, /* Spanning tree protocol port status change. */
367 REV_BOND, /* Bonding changed. */
368 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
369 REV_FLOW_TABLE, /* Flow table changed. */
370 REV_MAC_LEARNING, /* Mac learning changed. */
371 REV_INCONSISTENCY /* Facet self-check failed. */
373 COVERAGE_DEFINE(rev_reconfigure);
374 COVERAGE_DEFINE(rev_stp);
375 COVERAGE_DEFINE(rev_bond);
376 COVERAGE_DEFINE(rev_port_toggled);
377 COVERAGE_DEFINE(rev_flow_table);
378 COVERAGE_DEFINE(rev_mac_learning);
379 COVERAGE_DEFINE(rev_inconsistency);
381 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
382 * These are datapath flows which have no associated ofproto, if they did we
383 * would use facets. */
385 struct hmap_node hmap_node;
390 struct avg_subfacet_rates {
391 double add_rate; /* Moving average of new flows created per minute. */
392 double del_rate; /* Moving average of flows deleted per minute. */
395 /* All datapaths of a given type share a single dpif backer instance. */
400 struct timer next_expiration;
402 struct ovs_rwlock odp_to_ofport_lock;
403 struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
405 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
407 /* Facet revalidation flags applying to facets which use this backer. */
408 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
410 struct hmap drop_keys; /* Set of dropped odp keys. */
411 bool recv_set_enable; /* Enables or disables receiving packets. */
413 struct hmap subfacets;
414 struct governor *governor;
416 /* Subfacet statistics.
418 * These keep track of the total number of subfacets added and deleted and
419 * flow life span. They are useful for computing the flow rates stats
420 * exposed via "ovs-appctl dpif/show". The goal is to learn about
421 * traffic patterns in ways that we can use later to improve Open vSwitch
422 * performance in new situations. */
423 long long int created; /* Time when it is created. */
424 unsigned max_n_subfacet; /* Maximum number of flows */
425 unsigned avg_n_subfacet; /* Average number of flows. */
426 long long int avg_subfacet_life; /* Average life span of subfacets. */
428 /* The average number of subfacets... */
429 struct avg_subfacet_rates hourly; /* ...over the last hour. */
430 struct avg_subfacet_rates daily; /* ...over the last day. */
431 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
432 long long int last_minute; /* Last time 'hourly' was updated. */
434 /* Number of subfacets added or deleted since 'last_minute'. */
435 unsigned subfacet_add_count;
436 unsigned subfacet_del_count;
438 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
439 unsigned long long int total_subfacet_add_count;
440 unsigned long long int total_subfacet_del_count;
443 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
444 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
446 static void drop_key_clear(struct dpif_backer *);
447 static void update_moving_averages(struct dpif_backer *backer);
449 struct ofproto_dpif {
450 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
452 struct dpif_backer *backer;
454 /* Special OpenFlow rules. */
455 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
456 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
457 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
460 struct netflow *netflow;
461 struct dpif_sflow *sflow;
462 struct dpif_ipfix *ipfix;
463 struct hmap bundles; /* Contains "struct ofbundle"s. */
464 struct mac_learning *ml;
465 bool has_bonded_bundles;
466 struct mbridge *mbridge;
469 struct classifier facets; /* Contains 'struct facet's. */
470 long long int consistency_rl;
472 /* Support for debugging async flow mods. */
473 struct list completions;
475 struct netdev_stats stats; /* To account packets generated and consumed in
480 long long int stp_last_tick;
482 /* VLAN splinters. */
483 struct ovs_mutex vsp_mutex;
484 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
485 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
488 struct sset ports; /* Set of standard port names. */
489 struct sset ghost_ports; /* Ports with no datapath port. */
490 struct sset port_poll_set; /* Queued names for port_poll() reply. */
491 int port_poll_errno; /* Last errno for port_poll() reply. */
493 /* Per ofproto's dpif stats. */
498 struct ovs_mutex flow_mod_mutex;
499 struct list flow_mods OVS_GUARDED;
500 size_t n_flow_mods OVS_GUARDED;
502 struct ovs_mutex pin_mutex;
503 struct list pins OVS_GUARDED;
504 size_t n_pins OVS_GUARDED;
507 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
508 * for debugging the asynchronous flow_mod implementation.) */
511 /* By default, flows in the datapath are wildcarded (megaflows). They
512 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
513 static bool enable_megaflows = true;
515 /* All existing ofproto_dpif instances, indexed by ->up.name. */
516 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
518 static void ofproto_dpif_unixctl_init(void);
520 static inline struct ofproto_dpif *
521 ofproto_dpif_cast(const struct ofproto *ofproto)
523 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
524 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
527 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
528 ofp_port_t ofp_port);
529 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
530 const struct ofpbuf *packet, struct ds *);
533 #define FLOW_MISS_MAX_BATCH 50
534 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
536 /* Flow expiration. */
537 static int expire(struct dpif_backer *);
540 static void send_netflow_active_timeouts(struct ofproto_dpif *);
543 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
545 /* Global variables. */
546 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
548 /* Initial mappings of port to bridge mappings. */
549 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
551 /* Executes and takes ownership of 'fm'. */
553 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
554 struct ofputil_flow_mod *fm)
556 ovs_mutex_lock(&ofproto->flow_mod_mutex);
557 if (ofproto->n_flow_mods > 1024) {
558 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
559 COVERAGE_INC(flow_mod_overflow);
565 list_push_back(&ofproto->flow_mods, &fm->list_node);
566 ofproto->n_flow_mods++;
567 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
571 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
572 struct ofputil_packet_in *pin)
574 ovs_mutex_lock(&ofproto->pin_mutex);
575 if (ofproto->n_pins > 1024) {
576 ovs_mutex_unlock(&ofproto->pin_mutex);
577 COVERAGE_INC(packet_in_overflow);
578 free(CONST_CAST(void *, pin->packet));
583 list_push_back(&ofproto->pins, &pin->list_node);
585 ovs_mutex_unlock(&ofproto->pin_mutex);
588 /* Factory functions. */
591 init(const struct shash *iface_hints)
593 struct shash_node *node;
595 /* Make a local copy, since we don't own 'iface_hints' elements. */
596 SHASH_FOR_EACH(node, iface_hints) {
597 const struct iface_hint *orig_hint = node->data;
598 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
600 new_hint->br_name = xstrdup(orig_hint->br_name);
601 new_hint->br_type = xstrdup(orig_hint->br_type);
602 new_hint->ofp_port = orig_hint->ofp_port;
604 shash_add(&init_ofp_ports, node->name, new_hint);
609 enumerate_types(struct sset *types)
611 dp_enumerate_types(types);
615 enumerate_names(const char *type, struct sset *names)
617 struct ofproto_dpif *ofproto;
620 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
621 if (strcmp(type, ofproto->up.type)) {
624 sset_add(names, ofproto->up.name);
631 del(const char *type, const char *name)
636 error = dpif_open(name, type, &dpif);
638 error = dpif_delete(dpif);
645 port_open_type(const char *datapath_type, const char *port_type)
647 return dpif_port_open_type(datapath_type, port_type);
650 /* Type functions. */
652 static void process_dpif_port_changes(struct dpif_backer *);
653 static void process_dpif_all_ports_changed(struct dpif_backer *);
654 static void process_dpif_port_change(struct dpif_backer *,
655 const char *devname);
656 static void process_dpif_port_error(struct dpif_backer *, int error);
658 static struct ofproto_dpif *
659 lookup_ofproto_dpif_by_port_name(const char *name)
661 struct ofproto_dpif *ofproto;
663 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
664 if (sset_contains(&ofproto->ports, name)) {
673 type_run(const char *type)
675 static long long int push_timer = LLONG_MIN;
676 struct dpif_backer *backer;
678 backer = shash_find_data(&all_dpif_backers, type);
680 /* This is not necessarily a problem, since backers are only
681 * created on demand. */
685 dpif_run(backer->dpif);
687 /* The most natural place to push facet statistics is when they're pulled
688 * from the datapath. However, when there are many flows in the datapath,
689 * this expensive operation can occur so frequently, that it reduces our
690 * ability to quickly set up flows. To reduce the cost, we push statistics
692 if (time_msec() > push_timer) {
693 push_timer = time_msec() + 2000;
697 /* If vswitchd started with other_config:flow_restore_wait set as "true",
698 * and the configuration has now changed to "false", enable receiving
699 * packets from the datapath. */
700 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
703 backer->recv_set_enable = true;
705 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
707 VLOG_ERR("Failed to enable receiving packets in dpif.");
710 dpif_flow_flush(backer->dpif);
711 backer->need_revalidate = REV_RECONFIGURE;
714 if (backer->need_revalidate) {
715 struct ofproto_dpif *ofproto;
716 struct simap_node *node;
717 struct simap tmp_backers;
719 /* Handle tunnel garbage collection. */
720 simap_init(&tmp_backers);
721 simap_swap(&backer->tnl_backers, &tmp_backers);
723 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
724 struct ofport_dpif *iter;
726 if (backer != ofproto->backer) {
730 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
731 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
734 if (!iter->is_tunnel) {
738 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
739 namebuf, sizeof namebuf);
740 node = simap_find(&tmp_backers, dp_port);
742 simap_put(&backer->tnl_backers, dp_port, node->data);
743 simap_delete(&tmp_backers, node);
744 node = simap_find(&backer->tnl_backers, dp_port);
746 node = simap_find(&backer->tnl_backers, dp_port);
748 odp_port_t odp_port = ODPP_NONE;
750 if (!dpif_port_add(backer->dpif, iter->up.netdev,
752 simap_put(&backer->tnl_backers, dp_port,
753 odp_to_u32(odp_port));
754 node = simap_find(&backer->tnl_backers, dp_port);
759 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
760 if (tnl_port_reconfigure(iter, iter->up.netdev,
762 backer->need_revalidate = REV_RECONFIGURE;
767 SIMAP_FOR_EACH (node, &tmp_backers) {
768 dpif_port_del(backer->dpif, u32_to_odp(node->data));
770 simap_destroy(&tmp_backers);
772 switch (backer->need_revalidate) {
773 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
774 case REV_STP: COVERAGE_INC(rev_stp); break;
775 case REV_BOND: COVERAGE_INC(rev_bond); break;
776 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
777 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
778 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
779 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
781 backer->need_revalidate = 0;
783 /* Clear the drop_keys in case we should now be accepting some
784 * formerly dropped flows. */
785 drop_key_clear(backer);
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 struct facet *facet, *next;
789 struct ofport_dpif *ofport;
790 struct cls_cursor cursor;
791 struct ofbundle *bundle;
793 if (ofproto->backer != backer) {
797 ovs_rwlock_wrlock(&xlate_rwlock);
798 xlate_ofproto_set(ofproto, ofproto->up.name,
799 ofproto->backer->dpif, ofproto->miss_rule,
800 ofproto->no_packet_in_rule, ofproto->ml,
801 ofproto->stp, ofproto->mbridge,
802 ofproto->sflow, ofproto->ipfix,
803 ofproto->up.frag_handling,
804 ofproto->up.forward_bpdu,
805 connmgr_has_in_band(ofproto->up.connmgr),
806 ofproto->netflow != NULL);
808 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
809 xlate_bundle_set(ofproto, bundle, bundle->name,
810 bundle->vlan_mode, bundle->vlan,
811 bundle->trunks, bundle->use_priority_tags,
812 bundle->bond, bundle->lacp,
816 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
817 int stp_port = ofport->stp_port
818 ? stp_port_no(ofport->stp_port)
820 xlate_ofport_set(ofproto, ofport->bundle, ofport,
821 ofport->up.ofp_port, ofport->odp_port,
822 ofport->up.netdev, ofport->cfm,
823 ofport->bfd, ofport->peer, stp_port,
824 ofport->qdscp, ofport->n_qdscp,
825 ofport->up.pp.config, ofport->is_tunnel,
828 ovs_rwlock_unlock(&xlate_rwlock);
830 /* Only ofproto-dpif cares about the facet classifier so we just
831 * lock cls_cursor_init() to appease the thread safety analysis. */
832 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
833 cls_cursor_init(&cursor, &ofproto->facets, NULL);
834 ovs_rwlock_unlock(&ofproto->facets.rwlock);
835 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
836 facet_revalidate(facet);
842 if (!backer->recv_set_enable) {
843 /* Wake up before a max of 1000ms. */
844 timer_set_duration(&backer->next_expiration, 1000);
845 } else if (timer_expired(&backer->next_expiration)) {
846 int delay = expire(backer);
847 timer_set_duration(&backer->next_expiration, delay);
850 process_dpif_port_changes(backer);
852 if (backer->governor) {
855 governor_run(backer->governor);
857 /* If the governor has shrunk to its minimum size and the number of
858 * subfacets has dwindled, then drop the governor entirely.
860 * For hysteresis, the number of subfacets to drop the governor is
861 * smaller than the number needed to trigger its creation. */
862 n_subfacets = hmap_count(&backer->subfacets);
863 if (n_subfacets * 4 < flow_eviction_threshold
864 && governor_is_idle(backer->governor)) {
865 governor_destroy(backer->governor);
866 backer->governor = NULL;
873 /* Check for and handle port changes in 'backer''s dpif. */
875 process_dpif_port_changes(struct dpif_backer *backer)
881 error = dpif_port_poll(backer->dpif, &devname);
887 process_dpif_all_ports_changed(backer);
891 process_dpif_port_change(backer, devname);
896 process_dpif_port_error(backer, error);
903 process_dpif_all_ports_changed(struct dpif_backer *backer)
905 struct ofproto_dpif *ofproto;
906 struct dpif_port dpif_port;
907 struct dpif_port_dump dump;
908 struct sset devnames;
911 sset_init(&devnames);
912 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
913 if (ofproto->backer == backer) {
914 struct ofport *ofport;
916 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
917 sset_add(&devnames, netdev_get_name(ofport->netdev));
921 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
922 sset_add(&devnames, dpif_port.name);
925 SSET_FOR_EACH (devname, &devnames) {
926 process_dpif_port_change(backer, devname);
928 sset_destroy(&devnames);
932 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
934 struct ofproto_dpif *ofproto;
935 struct dpif_port port;
937 /* Don't report on the datapath's device. */
938 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
942 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
943 &all_ofproto_dpifs) {
944 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
949 ofproto = lookup_ofproto_dpif_by_port_name(devname);
950 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
951 /* The port was removed. If we know the datapath,
952 * report it through poll_set(). If we don't, it may be
953 * notifying us of a removal we initiated, so ignore it.
954 * If there's a pending ENOBUFS, let it stand, since
955 * everything will be reevaluated. */
956 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
957 sset_add(&ofproto->port_poll_set, devname);
958 ofproto->port_poll_errno = 0;
960 } else if (!ofproto) {
961 /* The port was added, but we don't know with which
962 * ofproto we should associate it. Delete it. */
963 dpif_port_del(backer->dpif, port.port_no);
965 struct ofport_dpif *ofport;
967 ofport = ofport_dpif_cast(shash_find_data(
968 &ofproto->up.port_by_name, devname));
970 && ofport->odp_port != port.port_no
971 && !odp_port_to_ofport(backer, port.port_no))
973 /* 'ofport''s datapath port number has changed from
974 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
975 * structures to match. */
976 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
977 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
978 ofport->odp_port = port.port_no;
979 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
980 hash_odp_port(port.port_no));
981 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
982 backer->need_revalidate = REV_RECONFIGURE;
985 dpif_port_destroy(&port);
988 /* Propagate 'error' to all ofprotos based on 'backer'. */
990 process_dpif_port_error(struct dpif_backer *backer, int error)
992 struct ofproto_dpif *ofproto;
994 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
995 if (ofproto->backer == backer) {
996 sset_clear(&ofproto->port_poll_set);
997 ofproto->port_poll_errno = error;
1003 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1007 /* If recv_set_enable is false, we should not handle upcalls. */
1008 if (!backer->recv_set_enable) {
1012 /* Handle one or more batches of upcalls, until there's nothing left to do
1013 * or until we do a fixed total amount of work.
1015 * We do work in batches because it can be much cheaper to set up a number
1016 * of flows and fire off their patches all at once. We do multiple batches
1017 * because in some cases handling a packet can cause another packet to be
1018 * queued almost immediately as part of the return flow. Both
1019 * optimizations can make major improvements on some benchmarks and
1020 * presumably for real traffic as well. */
1022 while (work < max_batch) {
1023 int retval = handle_upcalls(backer, max_batch - work);
1034 type_run_fast(const char *type)
1036 struct dpif_backer *backer;
1038 backer = shash_find_data(&all_dpif_backers, type);
1040 /* This is not necessarily a problem, since backers are only
1041 * created on demand. */
1045 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1051 static long long int port_rl = LLONG_MIN;
1052 static unsigned int backer_rl = 0;
1054 if (time_msec() >= port_rl) {
1055 struct ofproto_dpif *ofproto;
1057 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1058 run_fast(&ofproto->up);
1060 port_rl = time_msec() + 200;
1063 /* XXX: We have to be careful not to do too much work in this function. If
1064 * we call dpif_backer_run_fast() too often, or with too large a batch,
1065 * performance improves signifcantly, but at a cost. It's possible for the
1066 * number of flows in the datapath to increase without bound, and for poll
1067 * loops to take 10s of seconds. The correct solution to this problem,
1068 * long term, is to separate flow miss handling into it's own thread so it
1069 * isn't affected by revalidations, and expirations. Until then, this is
1070 * the best we can do. */
1071 if (++backer_rl >= 10) {
1072 struct shash_node *node;
1075 SHASH_FOR_EACH (node, &all_dpif_backers) {
1076 dpif_backer_run_fast(node->data, 1);
1082 type_wait(const char *type)
1084 struct dpif_backer *backer;
1086 backer = shash_find_data(&all_dpif_backers, type);
1088 /* This is not necessarily a problem, since backers are only
1089 * created on demand. */
1093 if (backer->governor) {
1094 governor_wait(backer->governor);
1097 timer_wait(&backer->next_expiration);
1100 /* Basic life-cycle. */
1102 static int add_internal_flows(struct ofproto_dpif *);
1104 static struct ofproto *
1107 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1108 return &ofproto->up;
1112 dealloc(struct ofproto *ofproto_)
1114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1119 close_dpif_backer(struct dpif_backer *backer)
1121 struct shash_node *node;
1123 ovs_assert(backer->refcount > 0);
1125 if (--backer->refcount) {
1129 drop_key_clear(backer);
1130 hmap_destroy(&backer->drop_keys);
1132 simap_destroy(&backer->tnl_backers);
1133 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1134 hmap_destroy(&backer->odp_to_ofport_map);
1135 node = shash_find(&all_dpif_backers, backer->type);
1137 shash_delete(&all_dpif_backers, node);
1138 dpif_close(backer->dpif);
1140 ovs_assert(hmap_is_empty(&backer->subfacets));
1141 hmap_destroy(&backer->subfacets);
1142 governor_destroy(backer->governor);
1147 /* Datapath port slated for removal from datapath. */
1148 struct odp_garbage {
1149 struct list list_node;
1150 odp_port_t odp_port;
1154 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1156 struct dpif_backer *backer;
1157 struct dpif_port_dump port_dump;
1158 struct dpif_port port;
1159 struct shash_node *node;
1160 struct list garbage_list;
1161 struct odp_garbage *garbage, *next;
1167 backer = shash_find_data(&all_dpif_backers, type);
1174 backer_name = xasprintf("ovs-%s", type);
1176 /* Remove any existing datapaths, since we assume we're the only
1177 * userspace controlling the datapath. */
1179 dp_enumerate_names(type, &names);
1180 SSET_FOR_EACH(name, &names) {
1181 struct dpif *old_dpif;
1183 /* Don't remove our backer if it exists. */
1184 if (!strcmp(name, backer_name)) {
1188 if (dpif_open(name, type, &old_dpif)) {
1189 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1191 dpif_delete(old_dpif);
1192 dpif_close(old_dpif);
1195 sset_destroy(&names);
1197 backer = xmalloc(sizeof *backer);
1199 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1202 VLOG_ERR("failed to open datapath of type %s: %s", type,
1203 ovs_strerror(error));
1208 backer->type = xstrdup(type);
1209 backer->governor = NULL;
1210 backer->refcount = 1;
1211 hmap_init(&backer->odp_to_ofport_map);
1212 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1213 hmap_init(&backer->drop_keys);
1214 hmap_init(&backer->subfacets);
1215 timer_set_duration(&backer->next_expiration, 1000);
1216 backer->need_revalidate = 0;
1217 simap_init(&backer->tnl_backers);
1218 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1221 if (backer->recv_set_enable) {
1222 dpif_flow_flush(backer->dpif);
1225 /* Loop through the ports already on the datapath and remove any
1226 * that we don't need anymore. */
1227 list_init(&garbage_list);
1228 dpif_port_dump_start(&port_dump, backer->dpif);
1229 while (dpif_port_dump_next(&port_dump, &port)) {
1230 node = shash_find(&init_ofp_ports, port.name);
1231 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1232 garbage = xmalloc(sizeof *garbage);
1233 garbage->odp_port = port.port_no;
1234 list_push_front(&garbage_list, &garbage->list_node);
1237 dpif_port_dump_done(&port_dump);
1239 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1240 dpif_port_del(backer->dpif, garbage->odp_port);
1241 list_remove(&garbage->list_node);
1245 shash_add(&all_dpif_backers, type, backer);
1247 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1249 VLOG_ERR("failed to listen on datapath of type %s: %s",
1250 type, ovs_strerror(error));
1251 close_dpif_backer(backer);
1255 backer->max_n_subfacet = 0;
1256 backer->created = time_msec();
1257 backer->last_minute = backer->created;
1258 memset(&backer->hourly, 0, sizeof backer->hourly);
1259 memset(&backer->daily, 0, sizeof backer->daily);
1260 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1261 backer->subfacet_add_count = 0;
1262 backer->subfacet_del_count = 0;
1263 backer->total_subfacet_add_count = 0;
1264 backer->total_subfacet_del_count = 0;
1265 backer->avg_n_subfacet = 0;
1266 backer->avg_subfacet_life = 0;
1272 construct(struct ofproto *ofproto_)
1274 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1275 struct shash_node *node, *next;
1276 odp_port_t max_ports;
1279 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1284 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1285 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1286 ofp_to_u16(OFPP_MAX))));
1288 ofproto->netflow = NULL;
1289 ofproto->sflow = NULL;
1290 ofproto->ipfix = NULL;
1291 ofproto->stp = NULL;
1292 hmap_init(&ofproto->bundles);
1293 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1294 ofproto->mbridge = mbridge_create();
1295 ofproto->has_bonded_bundles = false;
1296 ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1298 classifier_init(&ofproto->facets);
1299 ofproto->consistency_rl = LLONG_MIN;
1301 list_init(&ofproto->completions);
1303 ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1304 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1305 list_init(&ofproto->flow_mods);
1306 ofproto->n_flow_mods = 0;
1307 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1309 ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL);
1310 ovs_mutex_lock(&ofproto->pin_mutex);
1311 list_init(&ofproto->pins);
1312 ofproto->n_pins = 0;
1313 ovs_mutex_unlock(&ofproto->pin_mutex);
1315 ofproto_dpif_unixctl_init();
1317 hmap_init(&ofproto->vlandev_map);
1318 hmap_init(&ofproto->realdev_vid_map);
1320 sset_init(&ofproto->ports);
1321 sset_init(&ofproto->ghost_ports);
1322 sset_init(&ofproto->port_poll_set);
1323 ofproto->port_poll_errno = 0;
1325 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1326 struct iface_hint *iface_hint = node->data;
1328 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1329 /* Check if the datapath already has this port. */
1330 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1331 sset_add(&ofproto->ports, node->name);
1334 free(iface_hint->br_name);
1335 free(iface_hint->br_type);
1337 shash_delete(&init_ofp_ports, node);
1341 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1342 hash_string(ofproto->up.name, 0));
1343 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1345 ofproto_init_tables(ofproto_, N_TABLES);
1346 error = add_internal_flows(ofproto);
1347 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1350 ofproto->n_missed = 0;
1356 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1357 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1359 struct ofputil_flow_mod fm;
1362 match_init_catchall(&fm.match);
1364 match_set_reg(&fm.match, 0, id);
1365 fm.new_cookie = htonll(0);
1366 fm.cookie = htonll(0);
1367 fm.cookie_mask = htonll(0);
1368 fm.modify_cookie = false;
1369 fm.table_id = TBL_INTERNAL;
1370 fm.command = OFPFC_ADD;
1371 fm.idle_timeout = 0;
1372 fm.hard_timeout = 0;
1376 fm.ofpacts = ofpacts->data;
1377 fm.ofpacts_len = ofpacts->size;
1379 error = ofproto_flow_mod(&ofproto->up, &fm);
1381 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1382 id, ofperr_to_string(error));
1386 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1388 ovs_rwlock_unlock(&(*rulep)->up.evict);
1397 add_internal_flows(struct ofproto_dpif *ofproto)
1399 struct ofpact_controller *controller;
1400 uint64_t ofpacts_stub[128 / 8];
1401 struct ofpbuf ofpacts;
1405 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1408 controller = ofpact_put_CONTROLLER(&ofpacts);
1409 controller->max_len = UINT16_MAX;
1410 controller->controller_id = 0;
1411 controller->reason = OFPR_NO_MATCH;
1412 ofpact_pad(&ofpacts);
1414 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1419 ofpbuf_clear(&ofpacts);
1420 error = add_internal_flow(ofproto, id++, &ofpacts,
1421 &ofproto->no_packet_in_rule);
1426 error = add_internal_flow(ofproto, id++, &ofpacts,
1427 &ofproto->drop_frags_rule);
1432 complete_operations(struct ofproto_dpif *ofproto)
1434 struct dpif_completion *c, *next;
1436 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1437 ofoperation_complete(c->op, 0);
1438 list_remove(&c->list_node);
1444 destruct(struct ofproto *ofproto_)
1446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1447 struct rule_dpif *rule, *next_rule;
1448 struct ofputil_flow_mod *pin, *next_pin;
1449 struct ofputil_flow_mod *fm, *next_fm;
1450 struct oftable *table;
1452 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1453 ovs_rwlock_wrlock(&xlate_rwlock);
1454 xlate_remove_ofproto(ofproto);
1455 ovs_rwlock_unlock(&xlate_rwlock);
1457 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1458 complete_operations(ofproto);
1460 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1461 struct cls_cursor cursor;
1463 ovs_rwlock_wrlock(&table->cls.rwlock);
1464 cls_cursor_init(&cursor, &table->cls, NULL);
1465 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1466 ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up);
1468 ovs_rwlock_unlock(&table->cls.rwlock);
1471 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1472 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1473 list_remove(&fm->list_node);
1474 ofproto->n_flow_mods--;
1478 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1479 ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1481 ovs_mutex_lock(&ofproto->pin_mutex);
1482 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1483 list_remove(&pin->list_node);
1488 ovs_mutex_unlock(&ofproto->pin_mutex);
1489 ovs_mutex_destroy(&ofproto->pin_mutex);
1491 mbridge_unref(ofproto->mbridge);
1493 netflow_destroy(ofproto->netflow);
1494 dpif_sflow_unref(ofproto->sflow);
1495 hmap_destroy(&ofproto->bundles);
1496 mac_learning_unref(ofproto->ml);
1498 classifier_destroy(&ofproto->facets);
1500 hmap_destroy(&ofproto->vlandev_map);
1501 hmap_destroy(&ofproto->realdev_vid_map);
1503 sset_destroy(&ofproto->ports);
1504 sset_destroy(&ofproto->ghost_ports);
1505 sset_destroy(&ofproto->port_poll_set);
1507 ovs_mutex_destroy(&ofproto->vsp_mutex);
1509 close_dpif_backer(ofproto->backer);
1513 run_fast(struct ofproto *ofproto_)
1515 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1516 struct ofputil_packet_in *pin, *next_pin;
1517 struct ofputil_flow_mod *fm, *next_fm;
1518 struct list flow_mods, pins;
1519 struct ofport_dpif *ofport;
1521 /* Do not perform any periodic activity required by 'ofproto' while
1522 * waiting for flow restore to complete. */
1523 if (ofproto_get_flow_restore_wait()) {
1527 ovs_mutex_lock(&ofproto->flow_mod_mutex);
1528 if (ofproto->n_flow_mods) {
1529 flow_mods = ofproto->flow_mods;
1530 list_moved(&flow_mods);
1531 list_init(&ofproto->flow_mods);
1532 ofproto->n_flow_mods = 0;
1534 list_init(&flow_mods);
1536 ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1538 LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1539 int error = ofproto_flow_mod(&ofproto->up, fm);
1540 if (error && !VLOG_DROP_WARN(&rl)) {
1541 VLOG_WARN("learning action failed to modify flow table (%s)",
1542 ofperr_get_name(error));
1545 list_remove(&fm->list_node);
1550 ovs_mutex_lock(&ofproto->pin_mutex);
1551 if (ofproto->n_pins) {
1552 pins = ofproto->pins;
1554 list_init(&ofproto->pins);
1555 ofproto->n_pins = 0;
1559 ovs_mutex_unlock(&ofproto->pin_mutex);
1561 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1562 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1563 list_remove(&pin->list_node);
1564 free(CONST_CAST(void *, pin->packet));
1568 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1569 port_run_fast(ofport);
1576 run(struct ofproto *ofproto_)
1578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1579 struct ofport_dpif *ofport;
1580 struct ofbundle *bundle;
1584 complete_operations(ofproto);
1587 if (mbridge_need_revalidate(ofproto->mbridge)) {
1588 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1589 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1590 mac_learning_flush(ofproto->ml);
1591 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1594 /* Do not perform any periodic activity below required by 'ofproto' while
1595 * waiting for flow restore to complete. */
1596 if (ofproto_get_flow_restore_wait()) {
1600 error = run_fast(ofproto_);
1605 if (ofproto->netflow) {
1606 if (netflow_run(ofproto->netflow)) {
1607 send_netflow_active_timeouts(ofproto);
1610 if (ofproto->sflow) {
1611 dpif_sflow_run(ofproto->sflow);
1614 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1617 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1622 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1623 if (mac_learning_run(ofproto->ml)) {
1624 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1626 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1628 /* Check the consistency of a random facet, to aid debugging. */
1629 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1630 if (time_msec() >= ofproto->consistency_rl
1631 && !classifier_is_empty(&ofproto->facets)
1632 && !ofproto->backer->need_revalidate) {
1633 struct cls_table *table;
1634 struct cls_rule *cr;
1635 struct facet *facet;
1637 ofproto->consistency_rl = time_msec() + 250;
1639 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1640 struct cls_table, hmap_node);
1641 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1643 facet = CONTAINER_OF(cr, struct facet, cr);
1645 if (!facet_check_consistency(facet)) {
1646 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1649 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1655 wait(struct ofproto *ofproto_)
1657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1658 struct ofport_dpif *ofport;
1659 struct ofbundle *bundle;
1661 if (!clogged && !list_is_empty(&ofproto->completions)) {
1662 poll_immediate_wake();
1665 if (ofproto_get_flow_restore_wait()) {
1669 dpif_wait(ofproto->backer->dpif);
1670 dpif_recv_wait(ofproto->backer->dpif);
1671 if (ofproto->sflow) {
1672 dpif_sflow_wait(ofproto->sflow);
1674 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1677 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1678 bundle_wait(bundle);
1680 if (ofproto->netflow) {
1681 netflow_wait(ofproto->netflow);
1683 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1684 mac_learning_wait(ofproto->ml);
1685 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1687 if (ofproto->backer->need_revalidate) {
1688 /* Shouldn't happen, but if it does just go around again. */
1689 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1690 poll_immediate_wake();
1695 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1697 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1698 struct cls_cursor cursor;
1699 size_t n_subfacets = 0;
1700 struct facet *facet;
1702 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1703 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1704 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1706 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1707 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1708 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1709 n_subfacets += list_size(&facet->subfacets);
1711 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1712 simap_increase(usage, "subfacets", n_subfacets);
1716 flush(struct ofproto *ofproto_)
1718 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1719 struct subfacet *subfacet, *next_subfacet;
1720 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1724 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1725 &ofproto->backer->subfacets) {
1726 if (subfacet->facet->ofproto != ofproto) {
1730 if (subfacet->path != SF_NOT_INSTALLED) {
1731 batch[n_batch++] = subfacet;
1732 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1733 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1737 subfacet_destroy(subfacet);
1742 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1747 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1748 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1750 *arp_match_ip = true;
1751 *actions = (OFPUTIL_A_OUTPUT |
1752 OFPUTIL_A_SET_VLAN_VID |
1753 OFPUTIL_A_SET_VLAN_PCP |
1754 OFPUTIL_A_STRIP_VLAN |
1755 OFPUTIL_A_SET_DL_SRC |
1756 OFPUTIL_A_SET_DL_DST |
1757 OFPUTIL_A_SET_NW_SRC |
1758 OFPUTIL_A_SET_NW_DST |
1759 OFPUTIL_A_SET_NW_TOS |
1760 OFPUTIL_A_SET_TP_SRC |
1761 OFPUTIL_A_SET_TP_DST |
1766 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1769 struct dpif_dp_stats s;
1770 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1773 strcpy(ots->name, "classifier");
1775 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1776 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1777 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1778 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1780 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1781 ots->lookup_count = htonll(n_lookup);
1782 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1785 static struct ofport *
1788 struct ofport_dpif *port = xmalloc(sizeof *port);
1793 port_dealloc(struct ofport *port_)
1795 struct ofport_dpif *port = ofport_dpif_cast(port_);
1800 port_construct(struct ofport *port_)
1802 struct ofport_dpif *port = ofport_dpif_cast(port_);
1803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1804 const struct netdev *netdev = port->up.netdev;
1805 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1806 struct dpif_port dpif_port;
1809 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1810 port->bundle = NULL;
1813 port->may_enable = true;
1814 port->stp_port = NULL;
1815 port->stp_state = STP_DISABLED;
1816 port->is_tunnel = false;
1820 port->realdev_ofp_port = 0;
1821 port->vlandev_vid = 0;
1822 port->carrier_seq = netdev_get_carrier_resets(netdev);
1824 if (netdev_vport_is_patch(netdev)) {
1825 /* By bailing out here, we don't submit the port to the sFlow module
1826 * to be considered for counter polling export. This is correct
1827 * because the patch port represents an interface that sFlow considers
1828 * to be "internal" to the switch as a whole, and therefore not an
1829 * candidate for counter polling. */
1830 port->odp_port = ODPP_NONE;
1831 ofport_update_peer(port);
1835 error = dpif_port_query_by_name(ofproto->backer->dpif,
1836 netdev_vport_get_dpif_port(netdev, namebuf,
1843 port->odp_port = dpif_port.port_no;
1845 if (netdev_get_tunnel_config(netdev)) {
1846 tnl_port_add(port, port->up.netdev, port->odp_port);
1847 port->is_tunnel = true;
1849 /* Sanity-check that a mapping doesn't already exist. This
1850 * shouldn't happen for non-tunnel ports. */
1851 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1852 VLOG_ERR("port %s already has an OpenFlow port number",
1854 dpif_port_destroy(&dpif_port);
1858 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1859 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1860 hash_odp_port(port->odp_port));
1861 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1863 dpif_port_destroy(&dpif_port);
1865 if (ofproto->sflow) {
1866 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1873 port_destruct(struct ofport *port_)
1875 struct ofport_dpif *port = ofport_dpif_cast(port_);
1876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1877 const char *devname = netdev_get_name(port->up.netdev);
1878 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1879 const char *dp_port_name;
1881 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 ovs_rwlock_wrlock(&xlate_rwlock);
1883 xlate_ofport_remove(port);
1884 ovs_rwlock_unlock(&xlate_rwlock);
1886 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1888 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1889 /* The underlying device is still there, so delete it. This
1890 * happens when the ofproto is being destroyed, since the caller
1891 * assumes that removal of attached ports will happen as part of
1893 if (!port->is_tunnel) {
1894 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1899 port->peer->peer = NULL;
1903 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1904 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1905 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1906 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1910 sset_find_and_delete(&ofproto->ports, devname);
1911 sset_find_and_delete(&ofproto->ghost_ports, devname);
1912 bundle_remove(port_);
1913 set_cfm(port_, NULL);
1914 set_bfd(port_, NULL);
1915 if (ofproto->sflow) {
1916 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1923 port_modified(struct ofport *port_)
1925 struct ofport_dpif *port = ofport_dpif_cast(port_);
1927 if (port->bundle && port->bundle->bond) {
1928 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1932 cfm_set_netdev(port->cfm, port->up.netdev);
1935 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1937 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1941 ofport_update_peer(port);
1945 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1947 struct ofport_dpif *port = ofport_dpif_cast(port_);
1948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1949 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1951 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1952 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1953 OFPUTIL_PC_NO_PACKET_IN)) {
1954 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1956 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1957 bundle_update(port->bundle);
1963 set_sflow(struct ofproto *ofproto_,
1964 const struct ofproto_sflow_options *sflow_options)
1966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1967 struct dpif_sflow *ds = ofproto->sflow;
1969 if (sflow_options) {
1971 struct ofport_dpif *ofport;
1973 ds = ofproto->sflow = dpif_sflow_create();
1974 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1975 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1977 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1979 dpif_sflow_set_options(ds, sflow_options);
1982 dpif_sflow_unref(ds);
1983 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1984 ofproto->sflow = NULL;
1992 struct ofproto *ofproto_,
1993 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1994 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1995 size_t n_flow_exporters_options)
1997 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1998 struct dpif_ipfix *di = ofproto->ipfix;
2000 if (bridge_exporter_options || flow_exporters_options) {
2002 di = ofproto->ipfix = dpif_ipfix_create();
2004 dpif_ipfix_set_options(
2005 di, bridge_exporter_options, flow_exporters_options,
2006 n_flow_exporters_options);
2009 dpif_ipfix_unref(di);
2010 ofproto->ipfix = NULL;
2017 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2019 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2026 struct ofproto_dpif *ofproto;
2028 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2029 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2030 ofport->cfm = cfm_create(ofport->up.netdev);
2033 if (cfm_configure(ofport->cfm, s)) {
2039 cfm_unref(ofport->cfm);
2045 get_cfm_status(const struct ofport *ofport_,
2046 struct ofproto_cfm_status *status)
2048 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2051 status->faults = cfm_get_fault(ofport->cfm);
2052 status->remote_opstate = cfm_get_opup(ofport->cfm);
2053 status->health = cfm_get_health(ofport->cfm);
2054 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2062 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2065 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2069 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2070 if (ofport->bfd != old) {
2071 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2078 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2080 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2083 bfd_get_status(ofport->bfd, smap);
2090 /* Spanning Tree. */
2093 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2095 struct ofproto_dpif *ofproto = ofproto_;
2096 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2097 struct ofport_dpif *ofport;
2099 ofport = stp_port_get_aux(sp);
2101 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2102 ofproto->up.name, port_num);
2104 struct eth_header *eth = pkt->l2;
2106 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2107 if (eth_addr_is_zero(eth->eth_src)) {
2108 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2109 "with unknown MAC", ofproto->up.name, port_num);
2111 send_packet(ofport, pkt);
2117 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2119 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2121 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2123 /* Only revalidate flows if the configuration changed. */
2124 if (!s != !ofproto->stp) {
2125 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2129 if (!ofproto->stp) {
2130 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2131 send_bpdu_cb, ofproto);
2132 ofproto->stp_last_tick = time_msec();
2135 stp_set_bridge_id(ofproto->stp, s->system_id);
2136 stp_set_bridge_priority(ofproto->stp, s->priority);
2137 stp_set_hello_time(ofproto->stp, s->hello_time);
2138 stp_set_max_age(ofproto->stp, s->max_age);
2139 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2141 struct ofport *ofport;
2143 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2144 set_stp_port(ofport, NULL);
2147 stp_unref(ofproto->stp);
2148 ofproto->stp = NULL;
2155 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2161 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2162 s->designated_root = stp_get_designated_root(ofproto->stp);
2163 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2172 update_stp_port_state(struct ofport_dpif *ofport)
2174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2175 enum stp_state state;
2177 /* Figure out new state. */
2178 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2182 if (ofport->stp_state != state) {
2183 enum ofputil_port_state of_state;
2186 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2187 netdev_get_name(ofport->up.netdev),
2188 stp_state_name(ofport->stp_state),
2189 stp_state_name(state));
2190 if (stp_learn_in_state(ofport->stp_state)
2191 != stp_learn_in_state(state)) {
2192 /* xxx Learning action flows should also be flushed. */
2193 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2194 mac_learning_flush(ofproto->ml);
2195 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2197 fwd_change = stp_forward_in_state(ofport->stp_state)
2198 != stp_forward_in_state(state);
2200 ofproto->backer->need_revalidate = REV_STP;
2201 ofport->stp_state = state;
2202 ofport->stp_state_entered = time_msec();
2204 if (fwd_change && ofport->bundle) {
2205 bundle_update(ofport->bundle);
2208 /* Update the STP state bits in the OpenFlow port description. */
2209 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2210 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2211 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2212 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2213 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2215 ofproto_port_set_state(&ofport->up, of_state);
2219 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2220 * caller is responsible for assigning STP port numbers and ensuring
2221 * there are no duplicates. */
2223 set_stp_port(struct ofport *ofport_,
2224 const struct ofproto_port_stp_settings *s)
2226 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2228 struct stp_port *sp = ofport->stp_port;
2230 if (!s || !s->enable) {
2232 ofport->stp_port = NULL;
2233 stp_port_disable(sp);
2234 update_stp_port_state(ofport);
2237 } else if (sp && stp_port_no(sp) != s->port_num
2238 && ofport == stp_port_get_aux(sp)) {
2239 /* The port-id changed, so disable the old one if it's not
2240 * already in use by another port. */
2241 stp_port_disable(sp);
2244 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2245 stp_port_enable(sp);
2247 stp_port_set_aux(sp, ofport);
2248 stp_port_set_priority(sp, s->priority);
2249 stp_port_set_path_cost(sp, s->path_cost);
2251 update_stp_port_state(ofport);
2257 get_stp_port_status(struct ofport *ofport_,
2258 struct ofproto_port_stp_status *s)
2260 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2262 struct stp_port *sp = ofport->stp_port;
2264 if (!ofproto->stp || !sp) {
2270 s->port_id = stp_port_get_id(sp);
2271 s->state = stp_port_get_state(sp);
2272 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2273 s->role = stp_port_get_role(sp);
2274 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2280 stp_run(struct ofproto_dpif *ofproto)
2283 long long int now = time_msec();
2284 long long int elapsed = now - ofproto->stp_last_tick;
2285 struct stp_port *sp;
2288 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2289 ofproto->stp_last_tick = now;
2291 while (stp_get_changed_port(ofproto->stp, &sp)) {
2292 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2295 update_stp_port_state(ofport);
2299 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2300 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2301 mac_learning_flush(ofproto->ml);
2302 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2308 stp_wait(struct ofproto_dpif *ofproto)
2311 poll_timer_wait(1000);
2316 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2319 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2320 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2322 if (ofport->n_qdscp != n_qdscp
2323 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2324 n_qdscp * sizeof *qdscp))) {
2325 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2326 free(ofport->qdscp);
2327 ofport->qdscp = n_qdscp
2328 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2330 ofport->n_qdscp = n_qdscp;
2338 /* Expires all MAC learning entries associated with 'bundle' and forces its
2339 * ofproto to revalidate every flow.
2341 * Normally MAC learning entries are removed only from the ofproto associated
2342 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2343 * are removed from every ofproto. When patch ports and SLB bonds are in use
2344 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2345 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2346 * with the host from which it migrated. */
2348 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2350 struct ofproto_dpif *ofproto = bundle->ofproto;
2351 struct mac_learning *ml = ofproto->ml;
2352 struct mac_entry *mac, *next_mac;
2354 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2355 ovs_rwlock_wrlock(&ml->rwlock);
2356 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2357 if (mac->port.p == bundle) {
2359 struct ofproto_dpif *o;
2361 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2363 struct mac_entry *e;
2365 ovs_rwlock_wrlock(&o->ml->rwlock);
2366 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2368 mac_learning_expire(o->ml, e);
2370 ovs_rwlock_unlock(&o->ml->rwlock);
2375 mac_learning_expire(ml, mac);
2378 ovs_rwlock_unlock(&ml->rwlock);
2381 static struct ofbundle *
2382 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2384 struct ofbundle *bundle;
2386 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2387 &ofproto->bundles) {
2388 if (bundle->aux == aux) {
2396 bundle_update(struct ofbundle *bundle)
2398 struct ofport_dpif *port;
2400 bundle->floodable = true;
2401 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2402 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2403 || !stp_forward_in_state(port->stp_state)) {
2404 bundle->floodable = false;
2411 bundle_del_port(struct ofport_dpif *port)
2413 struct ofbundle *bundle = port->bundle;
2415 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2417 list_remove(&port->bundle_node);
2418 port->bundle = NULL;
2421 lacp_slave_unregister(bundle->lacp, port);
2424 bond_slave_unregister(bundle->bond, port);
2427 bundle_update(bundle);
2431 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2432 struct lacp_slave_settings *lacp)
2434 struct ofport_dpif *port;
2436 port = get_ofp_port(bundle->ofproto, ofp_port);
2441 if (port->bundle != bundle) {
2442 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2444 bundle_del_port(port);
2447 port->bundle = bundle;
2448 list_push_back(&bundle->ports, &port->bundle_node);
2449 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2450 || !stp_forward_in_state(port->stp_state)) {
2451 bundle->floodable = false;
2455 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2456 lacp_slave_register(bundle->lacp, port, lacp);
2463 bundle_destroy(struct ofbundle *bundle)
2465 struct ofproto_dpif *ofproto;
2466 struct ofport_dpif *port, *next_port;
2472 ofproto = bundle->ofproto;
2473 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2475 ovs_rwlock_wrlock(&xlate_rwlock);
2476 xlate_bundle_remove(bundle);
2477 ovs_rwlock_unlock(&xlate_rwlock);
2479 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2480 bundle_del_port(port);
2483 bundle_flush_macs(bundle, true);
2484 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2486 free(bundle->trunks);
2487 lacp_unref(bundle->lacp);
2488 bond_unref(bundle->bond);
2493 bundle_set(struct ofproto *ofproto_, void *aux,
2494 const struct ofproto_bundle_settings *s)
2496 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2497 bool need_flush = false;
2498 struct ofport_dpif *port;
2499 struct ofbundle *bundle;
2500 unsigned long *trunks;
2506 bundle_destroy(bundle_lookup(ofproto, aux));
2510 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2511 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2513 bundle = bundle_lookup(ofproto, aux);
2515 bundle = xmalloc(sizeof *bundle);
2517 bundle->ofproto = ofproto;
2518 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2519 hash_pointer(aux, 0));
2521 bundle->name = NULL;
2523 list_init(&bundle->ports);
2524 bundle->vlan_mode = PORT_VLAN_TRUNK;
2526 bundle->trunks = NULL;
2527 bundle->use_priority_tags = s->use_priority_tags;
2528 bundle->lacp = NULL;
2529 bundle->bond = NULL;
2531 bundle->floodable = true;
2532 mbridge_register_bundle(ofproto->mbridge, bundle);
2535 if (!bundle->name || strcmp(s->name, bundle->name)) {
2537 bundle->name = xstrdup(s->name);
2542 if (!bundle->lacp) {
2543 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2544 bundle->lacp = lacp_create();
2546 lacp_configure(bundle->lacp, s->lacp);
2548 lacp_unref(bundle->lacp);
2549 bundle->lacp = NULL;
2552 /* Update set of ports. */
2554 for (i = 0; i < s->n_slaves; i++) {
2555 if (!bundle_add_port(bundle, s->slaves[i],
2556 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2560 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2561 struct ofport_dpif *next_port;
2563 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2564 for (i = 0; i < s->n_slaves; i++) {
2565 if (s->slaves[i] == port->up.ofp_port) {
2570 bundle_del_port(port);
2574 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2576 if (list_is_empty(&bundle->ports)) {
2577 bundle_destroy(bundle);
2581 /* Set VLAN tagging mode */
2582 if (s->vlan_mode != bundle->vlan_mode
2583 || s->use_priority_tags != bundle->use_priority_tags) {
2584 bundle->vlan_mode = s->vlan_mode;
2585 bundle->use_priority_tags = s->use_priority_tags;
2590 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2591 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2593 if (vlan != bundle->vlan) {
2594 bundle->vlan = vlan;
2598 /* Get trunked VLANs. */
2599 switch (s->vlan_mode) {
2600 case PORT_VLAN_ACCESS:
2604 case PORT_VLAN_TRUNK:
2605 trunks = CONST_CAST(unsigned long *, s->trunks);
2608 case PORT_VLAN_NATIVE_UNTAGGED:
2609 case PORT_VLAN_NATIVE_TAGGED:
2610 if (vlan != 0 && (!s->trunks
2611 || !bitmap_is_set(s->trunks, vlan)
2612 || bitmap_is_set(s->trunks, 0))) {
2613 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2615 trunks = bitmap_clone(s->trunks, 4096);
2617 trunks = bitmap_allocate1(4096);
2619 bitmap_set1(trunks, vlan);
2620 bitmap_set0(trunks, 0);
2622 trunks = CONST_CAST(unsigned long *, s->trunks);
2629 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2630 free(bundle->trunks);
2631 if (trunks == s->trunks) {
2632 bundle->trunks = vlan_bitmap_clone(trunks);
2634 bundle->trunks = trunks;
2639 if (trunks != s->trunks) {
2644 if (!list_is_short(&bundle->ports)) {
2645 bundle->ofproto->has_bonded_bundles = true;
2647 if (bond_reconfigure(bundle->bond, s->bond)) {
2648 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2651 bundle->bond = bond_create(s->bond);
2652 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2655 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2656 bond_slave_register(bundle->bond, port, port->up.netdev);
2659 bond_unref(bundle->bond);
2660 bundle->bond = NULL;
2663 /* If we changed something that would affect MAC learning, un-learn
2664 * everything on this port and force flow revalidation. */
2666 bundle_flush_macs(bundle, false);
2673 bundle_remove(struct ofport *port_)
2675 struct ofport_dpif *port = ofport_dpif_cast(port_);
2676 struct ofbundle *bundle = port->bundle;
2679 bundle_del_port(port);
2680 if (list_is_empty(&bundle->ports)) {
2681 bundle_destroy(bundle);
2682 } else if (list_is_short(&bundle->ports)) {
2683 bond_unref(bundle->bond);
2684 bundle->bond = NULL;
2690 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2693 struct ofport_dpif *port = port_;
2694 uint8_t ea[ETH_ADDR_LEN];
2697 error = netdev_get_etheraddr(port->up.netdev, ea);
2699 struct ofpbuf packet;
2702 ofpbuf_init(&packet, 0);
2703 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2705 memcpy(packet_pdu, pdu, pdu_size);
2707 send_packet(port, &packet);
2708 ofpbuf_uninit(&packet);
2710 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2711 "%s (%s)", port->bundle->name,
2712 netdev_get_name(port->up.netdev), ovs_strerror(error));
2717 bundle_send_learning_packets(struct ofbundle *bundle)
2719 struct ofproto_dpif *ofproto = bundle->ofproto;
2720 int error, n_packets, n_errors;
2721 struct mac_entry *e;
2723 error = n_packets = n_errors = 0;
2724 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2725 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2726 if (e->port.p != bundle) {
2727 struct ofpbuf *learning_packet;
2728 struct ofport_dpif *port;
2732 /* The assignment to "port" is unnecessary but makes "grep"ing for
2733 * struct ofport_dpif more effective. */
2734 learning_packet = bond_compose_learning_packet(bundle->bond,
2738 ret = send_packet(port, learning_packet);
2739 ofpbuf_delete(learning_packet);
2747 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2750 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2751 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2752 "packets, last error was: %s",
2753 bundle->name, n_errors, n_packets, ovs_strerror(error));
2755 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2756 bundle->name, n_packets);
2761 bundle_run(struct ofbundle *bundle)
2764 lacp_run(bundle->lacp, send_pdu_cb);
2767 struct ofport_dpif *port;
2769 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2770 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2773 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2774 bundle->ofproto->backer->need_revalidate = REV_BOND;
2777 if (bond_should_send_learning_packets(bundle->bond)) {
2778 bundle_send_learning_packets(bundle);
2784 bundle_wait(struct ofbundle *bundle)
2787 lacp_wait(bundle->lacp);
2790 bond_wait(bundle->bond);
2797 mirror_set__(struct ofproto *ofproto_, void *aux,
2798 const struct ofproto_mirror_settings *s)
2800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801 struct ofbundle **srcs, **dsts;
2806 mirror_destroy(ofproto->mbridge, aux);
2810 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2811 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2813 for (i = 0; i < s->n_srcs; i++) {
2814 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2817 for (i = 0; i < s->n_dsts; i++) {
2818 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2821 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2822 s->n_dsts, s->src_vlans,
2823 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2830 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2831 uint64_t *packets, uint64_t *bytes)
2834 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2839 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2842 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2843 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2844 mac_learning_flush(ofproto->ml);
2846 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2851 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2855 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2859 forward_bpdu_changed(struct ofproto *ofproto_)
2861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2862 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2866 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2870 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2871 mac_learning_set_idle_time(ofproto->ml, idle_time);
2872 mac_learning_set_max_entries(ofproto->ml, max_entries);
2873 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2878 static struct ofport_dpif *
2879 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2881 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2882 return ofport ? ofport_dpif_cast(ofport) : NULL;
2885 static struct ofport_dpif *
2886 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2888 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2889 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2893 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2894 struct ofproto_port *ofproto_port,
2895 struct dpif_port *dpif_port)
2897 ofproto_port->name = dpif_port->name;
2898 ofproto_port->type = dpif_port->type;
2899 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2903 ofport_update_peer(struct ofport_dpif *ofport)
2905 const struct ofproto_dpif *ofproto;
2906 struct dpif_backer *backer;
2909 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2913 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2914 backer->need_revalidate = REV_RECONFIGURE;
2917 ofport->peer->peer = NULL;
2918 ofport->peer = NULL;
2921 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2926 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2927 struct ofport *peer_ofport;
2928 struct ofport_dpif *peer;
2931 if (ofproto->backer != backer) {
2935 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2940 peer = ofport_dpif_cast(peer_ofport);
2941 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2942 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2944 ofport->peer = peer;
2945 ofport->peer->peer = ofport;
2955 port_run_fast(struct ofport_dpif *ofport)
2957 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2958 struct ofpbuf packet;
2960 ofpbuf_init(&packet, 0);
2961 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2962 send_packet(ofport, &packet);
2963 ofpbuf_uninit(&packet);
2966 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2967 struct ofpbuf packet;
2969 ofpbuf_init(&packet, 0);
2970 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2971 send_packet(ofport, &packet);
2972 ofpbuf_uninit(&packet);
2977 port_run(struct ofport_dpif *ofport)
2979 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2980 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2981 bool enable = netdev_get_carrier(ofport->up.netdev);
2983 ofport->carrier_seq = carrier_seq;
2985 port_run_fast(ofport);
2988 int cfm_opup = cfm_get_opup(ofport->cfm);
2990 cfm_run(ofport->cfm);
2991 enable = enable && !cfm_get_fault(ofport->cfm);
2993 if (cfm_opup >= 0) {
2994 enable = enable && cfm_opup;
2999 bfd_run(ofport->bfd);
3000 enable = enable && bfd_forwarding(ofport->bfd);
3003 if (ofport->bundle) {
3004 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3005 if (carrier_changed) {
3006 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3010 if (ofport->may_enable != enable) {
3011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3012 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3015 ofport->may_enable = enable;
3019 port_wait(struct ofport_dpif *ofport)
3022 cfm_wait(ofport->cfm);
3026 bfd_wait(ofport->bfd);
3031 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3032 struct ofproto_port *ofproto_port)
3034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3035 struct dpif_port dpif_port;
3038 if (sset_contains(&ofproto->ghost_ports, devname)) {
3039 const char *type = netdev_get_type_from_name(devname);
3041 /* We may be called before ofproto->up.port_by_name is populated with
3042 * the appropriate ofport. For this reason, we must get the name and
3043 * type from the netdev layer directly. */
3045 const struct ofport *ofport;
3047 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3048 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3049 ofproto_port->name = xstrdup(devname);
3050 ofproto_port->type = xstrdup(type);
3056 if (!sset_contains(&ofproto->ports, devname)) {
3059 error = dpif_port_query_by_name(ofproto->backer->dpif,
3060 devname, &dpif_port);
3062 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3068 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3071 const char *devname = netdev_get_name(netdev);
3072 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3073 const char *dp_port_name;
3075 if (netdev_vport_is_patch(netdev)) {
3076 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3080 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3081 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3082 odp_port_t port_no = ODPP_NONE;
3085 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3089 if (netdev_get_tunnel_config(netdev)) {
3090 simap_put(&ofproto->backer->tnl_backers,
3091 dp_port_name, odp_to_u32(port_no));
3095 if (netdev_get_tunnel_config(netdev)) {
3096 sset_add(&ofproto->ghost_ports, devname);
3098 sset_add(&ofproto->ports, devname);
3104 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3106 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3107 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3114 sset_find_and_delete(&ofproto->ghost_ports,
3115 netdev_get_name(ofport->up.netdev));
3116 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3117 if (!ofport->is_tunnel) {
3118 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3120 /* The caller is going to close ofport->up.netdev. If this is a
3121 * bonded port, then the bond is using that netdev, so remove it
3122 * from the bond. The client will need to reconfigure everything
3123 * after deleting ports, so then the slave will get re-added. */
3124 bundle_remove(&ofport->up);
3131 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3133 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3138 error = netdev_get_stats(ofport->up.netdev, stats);
3140 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3143 /* ofproto->stats.tx_packets represents packets that we created
3144 * internally and sent to some port (e.g. packets sent with
3145 * send_packet()). Account for them as if they had come from
3146 * OFPP_LOCAL and got forwarded. */
3148 if (stats->rx_packets != UINT64_MAX) {
3149 stats->rx_packets += ofproto->stats.tx_packets;
3152 if (stats->rx_bytes != UINT64_MAX) {
3153 stats->rx_bytes += ofproto->stats.tx_bytes;
3156 /* ofproto->stats.rx_packets represents packets that were received on
3157 * some port and we processed internally and dropped (e.g. STP).
3158 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3160 if (stats->tx_packets != UINT64_MAX) {
3161 stats->tx_packets += ofproto->stats.rx_packets;
3164 if (stats->tx_bytes != UINT64_MAX) {
3165 stats->tx_bytes += ofproto->stats.rx_bytes;
3172 struct port_dump_state {
3177 struct ofproto_port port;
3182 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3184 *statep = xzalloc(sizeof(struct port_dump_state));
3189 port_dump_next(const struct ofproto *ofproto_, void *state_,
3190 struct ofproto_port *port)
3192 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3193 struct port_dump_state *state = state_;
3194 const struct sset *sset;
3195 struct sset_node *node;
3197 if (state->has_port) {
3198 ofproto_port_destroy(&state->port);
3199 state->has_port = false;
3201 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3202 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3205 error = port_query_by_name(ofproto_, node->name, &state->port);
3207 *port = state->port;
3208 state->has_port = true;
3210 } else if (error != ENODEV) {
3215 if (!state->ghost) {
3216 state->ghost = true;
3219 return port_dump_next(ofproto_, state_, port);
3226 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3228 struct port_dump_state *state = state_;
3230 if (state->has_port) {
3231 ofproto_port_destroy(&state->port);
3238 port_poll(const struct ofproto *ofproto_, char **devnamep)
3240 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3242 if (ofproto->port_poll_errno) {
3243 int error = ofproto->port_poll_errno;
3244 ofproto->port_poll_errno = 0;
3248 if (sset_is_empty(&ofproto->port_poll_set)) {
3252 *devnamep = sset_pop(&ofproto->port_poll_set);
3257 port_poll_wait(const struct ofproto *ofproto_)
3259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3260 dpif_port_poll_wait(ofproto->backer->dpif);
3264 port_is_lacp_current(const struct ofport *ofport_)
3266 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3267 return (ofport->bundle && ofport->bundle->lacp
3268 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3272 /* Upcall handling. */
3274 /* Flow miss batching.
3276 * Some dpifs implement operations faster when you hand them off in a batch.
3277 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3278 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3279 * more packets, plus possibly installing the flow in the dpif.
3281 * So far we only batch the operations that affect flow setup time the most.
3282 * It's possible to batch more than that, but the benefit might be minimal. */
3284 struct hmap_node hmap_node;
3285 struct ofproto_dpif *ofproto;
3287 enum odp_key_fitness key_fitness;
3288 const struct nlattr *key;
3290 struct list packets;
3291 enum dpif_upcall_type upcall_type;
3294 struct flow_miss_op {
3295 struct dpif_op dpif_op;
3297 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3298 struct xlate_out xout;
3299 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3301 struct ofpbuf mask; /* Flow mask for "put" ops. */
3302 struct odputil_keybuf maskbuf;
3304 /* If this is a "put" op, then a pointer to the subfacet that should
3305 * be marked as uninstalled if the operation fails. */
3306 struct subfacet *subfacet;
3309 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3310 * OpenFlow controller as necessary according to their individual
3311 * configurations. */
3313 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3314 const struct flow *flow)
3316 struct ofputil_packet_in pin;
3318 pin.packet = packet->data;
3319 pin.packet_len = packet->size;
3320 pin.reason = OFPR_NO_MATCH;
3321 pin.controller_id = 0;
3326 pin.send_len = 0; /* not used for flow table misses */
3328 flow_get_metadata(flow, &pin.fmd);
3330 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3333 static struct flow_miss *
3334 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3335 const struct flow *flow, uint32_t hash)
3337 struct flow_miss *miss;
3339 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3340 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3348 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3349 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3350 * 'miss' is associated with a subfacet the caller must also initialize the
3351 * returned op->subfacet, and if anything needs to be freed after processing
3352 * the op, the caller must initialize op->garbage also. */
3354 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3355 struct flow_miss_op *op)
3357 if (miss->flow.in_port.ofp_port
3358 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3359 miss->flow.vlan_tci)) {
3360 /* This packet was received on a VLAN splinter port. We
3361 * added a VLAN to the packet to make the packet resemble
3362 * the flow, but the actions were composed assuming that
3363 * the packet contained no VLAN. So, we must remove the
3364 * VLAN header from the packet before trying to execute the
3366 eth_pop_vlan(packet);
3369 op->subfacet = NULL;
3370 op->xout_garbage = false;
3371 op->dpif_op.type = DPIF_OP_EXECUTE;
3372 op->dpif_op.u.execute.key = miss->key;
3373 op->dpif_op.u.execute.key_len = miss->key_len;
3374 op->dpif_op.u.execute.packet = packet;
3375 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3378 /* Helper for handle_flow_miss_without_facet() and
3379 * handle_flow_miss_with_facet(). */
3381 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3382 const struct flow *flow, bool fail_open)
3386 * Extra-special case for fail-open mode.
3388 * We are in fail-open mode and the packet matched the fail-open
3389 * rule, but we are connected to a controller too. We should send
3390 * the packet up to the controller in the hope that it will try to
3391 * set up a flow and thereby allow us to exit fail-open.
3393 * See the top-level comment in fail-open.c for more information.
3395 send_packet_in_miss(ofproto, packet, flow);
3399 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3400 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3401 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3402 * return value of true). However, for short flows the cost of bookkeeping is
3403 * much higher than the benefits, so when the datapath holds a large number of
3404 * flows we impose some heuristics to decide which flows are likely to be worth
3407 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3409 struct dpif_backer *backer = miss->ofproto->backer;
3412 switch (flow_miss_model) {
3413 case OFPROTO_HANDLE_MISS_AUTO:
3415 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3417 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3421 if (!backer->governor) {
3424 n_subfacets = hmap_count(&backer->subfacets);
3425 if (n_subfacets * 2 <= flow_eviction_threshold) {
3429 backer->governor = governor_create();
3432 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3433 return governor_should_install_flow(backer->governor, hash,
3434 list_size(&miss->packets));
3437 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3438 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3439 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3441 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3442 struct flow_miss *miss,
3443 struct flow_miss_op *ops, size_t *n_ops)
3445 struct ofpbuf *packet;
3447 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3449 COVERAGE_INC(facet_suppress);
3451 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3452 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3455 struct xlate_in xin;
3457 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3458 xlate_actions_for_side_effects(&xin);
3461 if (xout->odp_actions.size) {
3462 struct flow_miss_op *op = &ops[*n_ops];
3463 struct dpif_execute *execute = &op->dpif_op.u.execute;
3465 init_flow_miss_execute_op(miss, packet, op);
3466 xlate_out_copy(&op->xout, xout);
3467 execute->actions = op->xout.odp_actions.data;
3468 execute->actions_len = op->xout.odp_actions.size;
3469 op->xout_garbage = true;
3476 /* Handles 'miss', which matches 'facet'. May add any required datapath
3477 * operations to 'ops', incrementing '*n_ops' for each new op.
3479 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3480 * This is really important only for new facets: if we just called time_msec()
3481 * here, then the new subfacet or its packets could look (occasionally) as
3482 * though it was used some time after the facet was used. That can make a
3483 * one-packet flow look like it has a nonzero duration, which looks odd in
3484 * e.g. NetFlow statistics.
3486 * If non-null, 'stats' will be folded into 'facet'. */
3488 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3489 long long int now, struct dpif_flow_stats *stats,
3490 struct flow_miss_op *ops, size_t *n_ops)
3492 enum subfacet_path want_path;
3493 struct subfacet *subfacet;
3494 struct ofpbuf *packet;
3496 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3498 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3499 struct flow_miss_op *op = &ops[*n_ops];
3501 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3504 if (want_path != SF_FAST_PATH) {
3505 struct rule_dpif *rule;
3506 struct xlate_in xin;
3508 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
3509 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3510 xlate_actions_for_side_effects(&xin);
3514 if (facet->xout.odp_actions.size) {
3515 struct dpif_execute *execute = &op->dpif_op.u.execute;
3517 init_flow_miss_execute_op(miss, packet, op);
3518 execute->actions = facet->xout.odp_actions.data,
3519 execute->actions_len = facet->xout.odp_actions.size;
3524 /* Don't install the flow if it's the result of the "userspace"
3525 * action for an already installed facet. This can occur when a
3526 * datapath flow with wildcards has a "userspace" action and flows
3527 * sent to userspace result in a different subfacet, which will then
3528 * be rejected as overlapping by the datapath. */
3529 if (miss->upcall_type == DPIF_UC_ACTION
3530 && !list_is_empty(&facet->subfacets)) {
3532 facet->used = MAX(facet->used, stats->used);
3533 facet->packet_count += stats->n_packets;
3534 facet->byte_count += stats->n_bytes;
3535 facet->tcp_flags |= stats->tcp_flags;
3540 subfacet = subfacet_create(facet, miss, now);
3542 subfacet_update_stats(subfacet, stats);
3545 if (subfacet->path != want_path) {
3546 struct flow_miss_op *op = &ops[(*n_ops)++];
3547 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3549 subfacet->path = want_path;
3551 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3552 if (enable_megaflows) {
3553 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3554 &miss->flow, UINT32_MAX);
3557 op->xout_garbage = false;
3558 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3559 op->subfacet = subfacet;
3560 put->flags = DPIF_FP_CREATE;
3561 put->key = miss->key;
3562 put->key_len = miss->key_len;
3563 put->mask = op->mask.data;
3564 put->mask_len = op->mask.size;
3566 if (want_path == SF_FAST_PATH) {
3567 put->actions = facet->xout.odp_actions.data;
3568 put->actions_len = facet->xout.odp_actions.size;
3570 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3571 op->slow_stub, sizeof op->slow_stub,
3572 &put->actions, &put->actions_len);
3578 /* Handles flow miss 'miss'. May add any required datapath operations
3579 * to 'ops', incrementing '*n_ops' for each new op. */
3581 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3584 struct ofproto_dpif *ofproto = miss->ofproto;
3585 struct dpif_flow_stats stats__;
3586 struct dpif_flow_stats *stats = &stats__;
3587 struct ofpbuf *packet;
3588 struct facet *facet;
3592 memset(stats, 0, sizeof *stats);
3594 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3595 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3596 stats->n_bytes += packet->size;
3600 facet = facet_lookup_valid(ofproto, &miss->flow);
3602 struct flow_wildcards wc;
3603 struct rule_dpif *rule;
3604 struct xlate_out xout;
3605 struct xlate_in xin;
3607 flow_wildcards_init_catchall(&wc);
3608 rule_dpif_lookup(ofproto, &miss->flow, &wc, &rule);
3609 rule_credit_stats(rule, stats);
3611 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3613 xin.resubmit_stats = stats;
3614 xin.may_learn = true;
3615 xlate_actions(&xin, &xout);
3616 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3618 /* There does not exist a bijection between 'struct flow' and datapath
3619 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3620 * assumption used throughout the facet and subfacet handling code.
3621 * Since we have to handle these misses in userspace anyway, we simply
3622 * skip facet creation, avoiding the problem altogether. */
3623 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3624 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3625 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3630 facet = facet_create(miss, rule, &xout, stats);
3634 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3637 static struct drop_key *
3638 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3641 struct drop_key *drop_key;
3643 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3644 &backer->drop_keys) {
3645 if (drop_key->key_len == key_len
3646 && !memcmp(drop_key->key, key, key_len)) {
3654 drop_key_clear(struct dpif_backer *backer)
3656 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3657 struct drop_key *drop_key, *next;
3659 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3662 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3664 if (error && !VLOG_DROP_WARN(&rl)) {
3665 struct ds ds = DS_EMPTY_INITIALIZER;
3666 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3667 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3668 ovs_strerror(error), ds_cstr(&ds));
3672 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3673 free(drop_key->key);
3679 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3682 struct dpif_upcall *upcall;
3683 struct flow_miss *miss;
3684 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3685 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3686 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3696 /* Construct the to-do list.
3698 * This just amounts to extracting the flow from each packet and sticking
3699 * the packets that have the same flow in the same "flow_miss" structure so
3700 * that we can process them together. */
3703 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3704 struct flow_miss *miss = &misses[n_misses];
3705 struct flow_miss *existing_miss;
3706 struct ofproto_dpif *ofproto;
3707 odp_port_t odp_in_port;
3712 error = xlate_receive(backer, upcall->packet, upcall->key,
3713 upcall->key_len, &flow, &miss->key_fitness,
3714 &ofproto, &odp_in_port);
3715 if (error == ENODEV) {
3716 struct drop_key *drop_key;
3718 /* Received packet on datapath port for which we couldn't
3719 * associate an ofproto. This can happen if a port is removed
3720 * while traffic is being received. Print a rate-limited message
3721 * in case it happens frequently. Install a drop flow so
3722 * that future packets of the flow are inexpensively dropped
3724 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3725 "%"PRIu32, odp_in_port);
3727 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3730 ret = dpif_flow_put(backer->dpif,
3731 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3732 upcall->key, upcall->key_len,
3733 NULL, 0, NULL, 0, NULL);
3736 drop_key = xmalloc(sizeof *drop_key);
3737 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3738 drop_key->key_len = upcall->key_len;
3740 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3741 hash_bytes(drop_key->key, drop_key->key_len, 0));
3750 ofproto->n_missed++;
3751 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3752 &flow.tunnel, &flow.in_port, &miss->flow);
3754 /* Add other packets to a to-do list. */
3755 hash = flow_hash(&miss->flow, 0);
3756 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3757 if (!existing_miss) {
3758 hmap_insert(&todo, &miss->hmap_node, hash);
3759 miss->ofproto = ofproto;
3760 miss->key = upcall->key;
3761 miss->key_len = upcall->key_len;
3762 miss->upcall_type = upcall->type;
3763 list_init(&miss->packets);
3767 miss = existing_miss;
3769 list_push_back(&miss->packets, &upcall->packet->list_node);
3772 /* Process each element in the to-do list, constructing the set of
3773 * operations to batch. */
3775 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3776 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3778 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3780 /* Execute batch. */
3781 for (i = 0; i < n_ops; i++) {
3782 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3784 dpif_operate(backer->dpif, dpif_ops, n_ops);
3786 for (i = 0; i < n_ops; i++) {
3787 if (dpif_ops[i]->error != 0
3788 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3789 && flow_miss_ops[i].subfacet) {
3790 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3792 COVERAGE_INC(subfacet_install_fail);
3794 /* Zero-out subfacet counters when installation failed, but
3795 * datapath reported hits. This should not happen and
3796 * indicates a bug, since if the datapath flow exists, we
3797 * should not be attempting to create a new subfacet. A
3798 * buggy datapath could trigger this, so just zero out the
3799 * counters and log an error. */
3800 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3801 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3802 "datapath reported hits");
3803 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3806 subfacet->path = SF_NOT_INSTALLED;
3810 if (flow_miss_ops[i].xout_garbage) {
3811 xlate_out_uninit(&flow_miss_ops[i].xout);
3814 hmap_destroy(&todo);
3817 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3819 classify_upcall(const struct dpif_upcall *upcall)
3821 size_t userdata_len;
3822 union user_action_cookie cookie;
3824 /* First look at the upcall type. */
3825 switch (upcall->type) {
3826 case DPIF_UC_ACTION:
3832 case DPIF_N_UC_TYPES:
3834 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3838 /* "action" upcalls need a closer look. */
3839 if (!upcall->userdata) {
3840 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3843 userdata_len = nl_attr_get_size(upcall->userdata);
3844 if (userdata_len < sizeof cookie.type
3845 || userdata_len > sizeof cookie) {
3846 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3850 memset(&cookie, 0, sizeof cookie);
3851 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3852 if (userdata_len == sizeof cookie.sflow
3853 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3854 return SFLOW_UPCALL;
3855 } else if (userdata_len == sizeof cookie.slow_path
3856 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3858 } else if (userdata_len == sizeof cookie.flow_sample
3859 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3860 return FLOW_SAMPLE_UPCALL;
3861 } else if (userdata_len == sizeof cookie.ipfix
3862 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3863 return IPFIX_UPCALL;
3865 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3866 " and size %zu", cookie.type, userdata_len);
3872 handle_sflow_upcall(struct dpif_backer *backer,
3873 const struct dpif_upcall *upcall)
3875 struct ofproto_dpif *ofproto;
3876 union user_action_cookie cookie;
3878 odp_port_t odp_in_port;
3880 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3881 &flow, NULL, &ofproto, &odp_in_port)
3882 || !ofproto->sflow) {
3886 memset(&cookie, 0, sizeof cookie);
3887 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3888 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3889 odp_in_port, &cookie);
3893 handle_flow_sample_upcall(struct dpif_backer *backer,
3894 const struct dpif_upcall *upcall)
3896 struct ofproto_dpif *ofproto;
3897 union user_action_cookie cookie;
3900 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3901 &flow, NULL, &ofproto, NULL)
3902 || !ofproto->ipfix) {
3906 memset(&cookie, 0, sizeof cookie);
3907 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3909 /* The flow reflects exactly the contents of the packet. Sample
3910 * the packet using it. */
3911 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3912 cookie.flow_sample.collector_set_id,
3913 cookie.flow_sample.probability,
3914 cookie.flow_sample.obs_domain_id,
3915 cookie.flow_sample.obs_point_id);
3919 handle_ipfix_upcall(struct dpif_backer *backer,
3920 const struct dpif_upcall *upcall)
3922 struct ofproto_dpif *ofproto;
3925 if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3926 &flow, NULL, &ofproto, NULL)
3927 || !ofproto->ipfix) {
3931 /* The flow reflects exactly the contents of the packet. Sample
3932 * the packet using it. */
3933 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3937 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3939 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3940 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3941 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3946 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3949 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3950 struct dpif_upcall *upcall = &misses[n_misses];
3951 struct ofpbuf *buf = &miss_bufs[n_misses];
3954 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3955 sizeof miss_buf_stubs[n_misses]);
3956 error = dpif_recv(backer->dpif, upcall, buf);
3962 switch (classify_upcall(upcall)) {
3964 /* Handle it later. */
3969 handle_sflow_upcall(backer, upcall);
3973 case FLOW_SAMPLE_UPCALL:
3974 handle_flow_sample_upcall(backer, upcall);
3979 handle_ipfix_upcall(backer, upcall);
3989 /* Handle deferred MISS_UPCALL processing. */
3990 handle_miss_upcalls(backer, misses, n_misses);
3991 for (i = 0; i < n_misses; i++) {
3992 ofpbuf_uninit(&miss_bufs[i]);
3998 /* Flow expiration. */
4000 static int subfacet_max_idle(const struct dpif_backer *);
4001 static void update_stats(struct dpif_backer *);
4002 static void rule_expire(struct rule_dpif *);
4003 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4005 /* This function is called periodically by run(). Its job is to collect
4006 * updates for the flows that have been installed into the datapath, most
4007 * importantly when they last were used, and then use that information to
4008 * expire flows that have not been used recently.
4010 * Returns the number of milliseconds after which it should be called again. */
4012 expire(struct dpif_backer *backer)
4014 struct ofproto_dpif *ofproto;
4018 /* Periodically clear out the drop keys in an effort to keep them
4019 * relatively few. */
4020 drop_key_clear(backer);
4022 /* Update stats for each flow in the backer. */
4023 update_stats(backer);
4025 n_subfacets = hmap_count(&backer->subfacets);
4027 struct subfacet *subfacet;
4028 long long int total, now;
4032 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4033 total += now - subfacet->created;
4035 backer->avg_subfacet_life += total / n_subfacets;
4037 backer->avg_subfacet_life /= 2;
4039 backer->avg_n_subfacet += n_subfacets;
4040 backer->avg_n_subfacet /= 2;
4042 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4044 max_idle = subfacet_max_idle(backer);
4045 expire_subfacets(backer, max_idle);
4047 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4048 struct rule *rule, *next_rule;
4050 if (ofproto->backer != backer) {
4054 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4056 ovs_mutex_lock(&ofproto->up.expirable_mutex);
4057 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4058 &ofproto->up.expirable) {
4059 rule_expire(rule_dpif_cast(rule));
4061 ovs_mutex_unlock(&ofproto->up.expirable_mutex);
4063 /* All outstanding data in existing flows has been accounted, so it's a
4064 * good time to do bond rebalancing. */
4065 if (ofproto->has_bonded_bundles) {
4066 struct ofbundle *bundle;
4068 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4070 bond_rebalance(bundle->bond);
4076 return MIN(max_idle, 1000);
4079 /* Updates flow table statistics given that the datapath just reported 'stats'
4080 * as 'subfacet''s statistics. */
4082 update_subfacet_stats(struct subfacet *subfacet,
4083 const struct dpif_flow_stats *stats)
4085 struct facet *facet = subfacet->facet;
4086 struct dpif_flow_stats diff;
4088 diff.tcp_flags = stats->tcp_flags;
4089 diff.used = stats->used;
4091 if (stats->n_packets >= subfacet->dp_packet_count) {
4092 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4094 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4098 if (stats->n_bytes >= subfacet->dp_byte_count) {
4099 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4101 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4105 facet->ofproto->n_hit += diff.n_packets;
4106 subfacet->dp_packet_count = stats->n_packets;
4107 subfacet->dp_byte_count = stats->n_bytes;
4108 subfacet_update_stats(subfacet, &diff);
4110 if (facet->accounted_bytes < facet->byte_count) {
4112 facet_account(facet);
4113 facet->accounted_bytes = facet->byte_count;
4117 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4118 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4120 delete_unexpected_flow(struct dpif_backer *backer,
4121 const struct nlattr *key, size_t key_len)
4123 if (!VLOG_DROP_WARN(&rl)) {
4127 odp_flow_key_format(key, key_len, &s);
4128 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4132 COVERAGE_INC(facet_unexpected);
4133 dpif_flow_del(backer->dpif, key, key_len, NULL);
4136 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4138 * This function also pushes statistics updates to rules which each facet
4139 * resubmits into. Generally these statistics will be accurate. However, if a
4140 * facet changes the rule it resubmits into at some time in between
4141 * update_stats() runs, it is possible that statistics accrued to the
4142 * old rule will be incorrectly attributed to the new rule. This could be
4143 * avoided by calling update_stats() whenever rules are created or
4144 * deleted. However, the performance impact of making so many calls to the
4145 * datapath do not justify the benefit of having perfectly accurate statistics.
4147 * In addition, this function maintains per ofproto flow hit counts. The patch
4148 * port is not treated specially. e.g. A packet ingress from br0 patched into
4149 * br1 will increase the hit count of br0 by 1, however, does not affect
4150 * the hit or miss counts of br1.
4153 update_stats(struct dpif_backer *backer)
4155 const struct dpif_flow_stats *stats;
4156 struct dpif_flow_dump dump;
4157 const struct nlattr *key, *mask;
4158 size_t key_len, mask_len;
4160 dpif_flow_dump_start(&dump, backer->dpif);
4161 while (dpif_flow_dump_next(&dump, &key, &key_len,
4162 &mask, &mask_len, NULL, NULL, &stats)) {
4163 struct subfacet *subfacet;
4166 key_hash = odp_flow_key_hash(key, key_len);
4167 subfacet = subfacet_find(backer, key, key_len, key_hash);
4168 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4170 update_subfacet_stats(subfacet, stats);
4174 /* Stats are updated per-packet. */
4177 case SF_NOT_INSTALLED:
4179 delete_unexpected_flow(backer, key, key_len);
4184 dpif_flow_dump_done(&dump);
4186 update_moving_averages(backer);
4189 /* Calculates and returns the number of milliseconds of idle time after which
4190 * subfacets should expire from the datapath. When a subfacet expires, we fold
4191 * its statistics into its facet, and when a facet's last subfacet expires, we
4192 * fold its statistic into its rule. */
4194 subfacet_max_idle(const struct dpif_backer *backer)
4197 * Idle time histogram.
4199 * Most of the time a switch has a relatively small number of subfacets.
4200 * When this is the case we might as well keep statistics for all of them
4201 * in userspace and to cache them in the kernel datapath for performance as
4204 * As the number of subfacets increases, the memory required to maintain
4205 * statistics about them in userspace and in the kernel becomes
4206 * significant. However, with a large number of subfacets it is likely
4207 * that only a few of them are "heavy hitters" that consume a large amount
4208 * of bandwidth. At this point, only heavy hitters are worth caching in
4209 * the kernel and maintaining in userspaces; other subfacets we can
4212 * The technique used to compute the idle time is to build a histogram with
4213 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4214 * that is installed in the kernel gets dropped in the appropriate bucket.
4215 * After the histogram has been built, we compute the cutoff so that only
4216 * the most-recently-used 1% of subfacets (but at least
4217 * flow_eviction_threshold flows) are kept cached. At least
4218 * the most-recently-used bucket of subfacets is kept, so actually an
4219 * arbitrary number of subfacets can be kept in any given expiration run
4220 * (though the next run will delete most of those unless they receive
4223 * This requires a second pass through the subfacets, in addition to the
4224 * pass made by update_stats(), because the former function never looks at
4225 * uninstallable subfacets.
4227 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4228 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4229 int buckets[N_BUCKETS] = { 0 };
4230 int total, subtotal, bucket;
4231 struct subfacet *subfacet;
4235 total = hmap_count(&backer->subfacets);
4236 if (total <= flow_eviction_threshold) {
4237 return N_BUCKETS * BUCKET_WIDTH;
4240 /* Build histogram. */
4242 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4243 long long int idle = now - subfacet->used;
4244 int bucket = (idle <= 0 ? 0
4245 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4246 : (unsigned int) idle / BUCKET_WIDTH);
4250 /* Find the first bucket whose flows should be expired. */
4251 subtotal = bucket = 0;
4253 subtotal += buckets[bucket++];
4254 } while (bucket < N_BUCKETS &&
4255 subtotal < MAX(flow_eviction_threshold, total / 100));
4257 if (VLOG_IS_DBG_ENABLED()) {
4261 ds_put_cstr(&s, "keep");
4262 for (i = 0; i < N_BUCKETS; i++) {
4264 ds_put_cstr(&s, ", drop");
4267 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4270 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4274 return bucket * BUCKET_WIDTH;
4278 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4280 /* Cutoff time for most flows. */
4281 long long int normal_cutoff = time_msec() - dp_max_idle;
4283 /* We really want to keep flows for special protocols around, so use a more
4284 * conservative cutoff. */
4285 long long int special_cutoff = time_msec() - 10000;
4287 struct subfacet *subfacet, *next_subfacet;
4288 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4292 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4293 &backer->subfacets) {
4294 long long int cutoff;
4296 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4300 if (subfacet->used < cutoff) {
4301 if (subfacet->path != SF_NOT_INSTALLED) {
4302 batch[n_batch++] = subfacet;
4303 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4304 subfacet_destroy_batch(backer, batch, n_batch);
4308 subfacet_destroy(subfacet);
4314 subfacet_destroy_batch(backer, batch, n_batch);
4318 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4319 * then delete it entirely. */
4321 rule_expire(struct rule_dpif *rule)
4323 uint16_t idle_timeout, hard_timeout;
4327 if (rule->up.pending) {
4328 /* We'll have to expire it later. */
4332 ovs_mutex_lock(&rule->up.timeout_mutex);
4333 hard_timeout = rule->up.hard_timeout;
4334 idle_timeout = rule->up.idle_timeout;
4335 ovs_mutex_unlock(&rule->up.timeout_mutex);
4337 /* Has 'rule' expired? */
4339 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
4340 reason = OFPRR_HARD_TIMEOUT;
4341 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
4342 reason = OFPRR_IDLE_TIMEOUT;
4347 if (!ovs_rwlock_trywrlock(&rule->up.evict)) {
4348 COVERAGE_INC(ofproto_dpif_expired);
4350 /* Get rid of the rule. */
4351 ofproto_rule_expire(&rule->up, reason);
4357 /* Creates and returns a new facet based on 'miss'.
4359 * The caller must already have determined that no facet with an identical
4360 * 'miss->flow' exists in 'miss->ofproto'.
4362 * 'rule' and 'xout' must have been created based on 'miss'.
4364 * 'facet'' statistics are initialized based on 'stats'.
4366 * The facet will initially have no subfacets. The caller should create (at
4367 * least) one subfacet with subfacet_create(). */
4368 static struct facet *
4369 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4370 struct xlate_out *xout, struct dpif_flow_stats *stats)
4372 struct ofproto_dpif *ofproto = miss->ofproto;
4373 struct facet *facet;
4376 facet = xzalloc(sizeof *facet);
4377 facet->ofproto = miss->ofproto;
4378 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4379 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4380 facet->tcp_flags = stats->tcp_flags;
4381 facet->used = stats->used;
4382 facet->flow = miss->flow;
4383 facet->learn_rl = time_msec() + 500;
4385 list_init(&facet->subfacets);
4386 netflow_flow_init(&facet->nf_flow);
4387 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4389 xlate_out_copy(&facet->xout, xout);
4391 match_init(&match, &facet->flow, &facet->xout.wc);
4392 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4393 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4394 classifier_insert(&ofproto->facets, &facet->cr);
4395 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4397 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4398 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4404 facet_free(struct facet *facet)
4407 xlate_out_uninit(&facet->xout);
4412 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4413 * 'packet', which arrived on 'in_port'. */
4415 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4416 const struct nlattr *odp_actions, size_t actions_len,
4417 struct ofpbuf *packet)
4419 struct odputil_keybuf keybuf;
4423 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4424 odp_flow_key_from_flow(&key, flow,
4425 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4427 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4428 odp_actions, actions_len, packet);
4432 /* Remove 'facet' from its ofproto and free up the associated memory:
4434 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4435 * rule's statistics, via subfacet_uninstall().
4437 * - Removes 'facet' from its rule and from ofproto->facets.
4440 facet_remove(struct facet *facet)
4442 struct subfacet *subfacet, *next_subfacet;
4444 ovs_assert(!list_is_empty(&facet->subfacets));
4446 /* First uninstall all of the subfacets to get final statistics. */
4447 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4448 subfacet_uninstall(subfacet);
4451 /* Flush the final stats to the rule.
4453 * This might require us to have at least one subfacet around so that we
4454 * can use its actions for accounting in facet_account(), which is why we
4455 * have uninstalled but not yet destroyed the subfacets. */
4456 facet_flush_stats(facet);
4458 /* Now we're really all done so destroy everything. */
4459 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4460 &facet->subfacets) {
4461 subfacet_destroy__(subfacet);
4463 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4464 classifier_remove(&facet->ofproto->facets, &facet->cr);
4465 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4466 cls_rule_destroy(&facet->cr);
4470 /* Feed information from 'facet' back into the learning table to keep it in
4471 * sync with what is actually flowing through the datapath. */
4473 facet_learn(struct facet *facet)
4475 long long int now = time_msec();
4477 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4481 facet->learn_rl = now + 500;
4483 if (!facet->xout.has_learn
4484 && !facet->xout.has_normal
4485 && (!facet->xout.has_fin_timeout
4486 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4490 facet_push_stats(facet, true);
4494 facet_account(struct facet *facet)
4496 const struct nlattr *a;
4501 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4504 n_bytes = facet->byte_count - facet->accounted_bytes;
4506 /* This loop feeds byte counters to bond_account() for rebalancing to use
4507 * as a basis. We also need to track the actual VLAN on which the packet
4508 * is going to be sent to ensure that it matches the one passed to
4509 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4512 * We use the actions from an arbitrary subfacet because they should all
4513 * be equally valid for our purpose. */
4514 vlan_tci = facet->flow.vlan_tci;
4515 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4516 facet->xout.odp_actions.size) {
4517 const struct ovs_action_push_vlan *vlan;
4518 struct ofport_dpif *port;
4520 switch (nl_attr_type(a)) {
4521 case OVS_ACTION_ATTR_OUTPUT:
4522 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4523 if (port && port->bundle && port->bundle->bond) {
4524 bond_account(port->bundle->bond, &facet->flow,
4525 vlan_tci_to_vid(vlan_tci), n_bytes);
4529 case OVS_ACTION_ATTR_POP_VLAN:
4530 vlan_tci = htons(0);
4533 case OVS_ACTION_ATTR_PUSH_VLAN:
4534 vlan = nl_attr_get(a);
4535 vlan_tci = vlan->vlan_tci;
4541 /* Returns true if the only action for 'facet' is to send to the controller.
4542 * (We don't report NetFlow expiration messages for such facets because they
4543 * are just part of the control logic for the network, not real traffic). */
4545 facet_is_controller_flow(struct facet *facet)
4548 struct ofproto_dpif *ofproto = facet->ofproto;
4549 const struct ofpact *ofpacts;
4550 struct rule_dpif *rule;
4554 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4555 ofpacts_len = rule->up.ofpacts_len;
4556 ofpacts = rule->up.ofpacts;
4557 is_controller = ofpacts_len > 0
4558 && ofpacts->type == OFPACT_CONTROLLER
4559 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4561 return is_controller;
4566 /* Folds all of 'facet''s statistics into its rule. Also updates the
4567 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4568 * 'facet''s statistics in the datapath should have been zeroed and folded into
4569 * its packet and byte counts before this function is called. */
4571 facet_flush_stats(struct facet *facet)
4573 struct ofproto_dpif *ofproto = facet->ofproto;
4574 struct subfacet *subfacet;
4576 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4577 ovs_assert(!subfacet->dp_byte_count);
4578 ovs_assert(!subfacet->dp_packet_count);
4581 facet_push_stats(facet, false);
4582 if (facet->accounted_bytes < facet->byte_count) {
4583 facet_account(facet);
4584 facet->accounted_bytes = facet->byte_count;
4587 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4588 struct ofexpired expired;
4589 expired.flow = facet->flow;
4590 expired.packet_count = facet->packet_count;
4591 expired.byte_count = facet->byte_count;
4592 expired.used = facet->used;
4593 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4596 /* Reset counters to prevent double counting if 'facet' ever gets
4598 facet_reset_counters(facet);
4600 netflow_flow_clear(&facet->nf_flow);
4601 facet->tcp_flags = 0;
4604 /* Searches 'ofproto''s table of facets for one which would be responsible for
4605 * 'flow'. Returns it if found, otherwise a null pointer.
4607 * The returned facet might need revalidation; use facet_lookup_valid()
4608 * instead if that is important. */
4609 static struct facet *
4610 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4612 struct cls_rule *cr;
4614 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4615 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4616 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4617 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4620 /* Searches 'ofproto''s table of facets for one capable that covers
4621 * 'flow'. Returns it if found, otherwise a null pointer.
4623 * The returned facet is guaranteed to be valid. */
4624 static struct facet *
4625 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4627 struct facet *facet;
4629 facet = facet_find(ofproto, flow);
4631 && ofproto->backer->need_revalidate
4632 && !facet_revalidate(facet)) {
4640 facet_check_consistency(struct facet *facet)
4642 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4644 struct xlate_out xout;
4645 struct xlate_in xin;
4647 struct rule_dpif *rule;
4650 /* Check the datapath actions for consistency. */
4651 rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4652 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4653 xlate_actions(&xin, &xout);
4656 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4657 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4658 && facet->xout.slow == xout.slow
4659 && facet->fail_open == fail_open;
4660 if (!ok && !VLOG_DROP_WARN(&rl)) {
4661 struct ds s = DS_EMPTY_INITIALIZER;
4663 flow_format(&s, &facet->flow);
4664 ds_put_cstr(&s, ": inconsistency in facet");
4666 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4667 ds_put_cstr(&s, " (actions were: ");
4668 format_odp_actions(&s, facet->xout.odp_actions.data,
4669 facet->xout.odp_actions.size);
4670 ds_put_cstr(&s, ") (correct actions: ");
4671 format_odp_actions(&s, xout.odp_actions.data,
4672 xout.odp_actions.size);
4673 ds_put_char(&s, ')');
4676 if (facet->xout.slow != xout.slow) {
4677 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4680 if (facet->fail_open != fail_open) {
4681 ds_put_format(&s, " fail open incorrect. should be %s",
4682 fail_open ? "true" : "false");
4686 xlate_out_uninit(&xout);
4691 /* Re-searches the classifier for 'facet':
4693 * - If the rule found is different from 'facet''s current rule, moves
4694 * 'facet' to the new rule and recompiles its actions.
4696 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4697 * where it is and recompiles its actions anyway.
4699 * - If any of 'facet''s subfacets correspond to a new flow according to
4700 * xlate_receive(), 'facet' is removed.
4702 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4704 facet_revalidate(struct facet *facet)
4706 struct ofproto_dpif *ofproto = facet->ofproto;
4707 struct rule_dpif *new_rule;
4708 struct subfacet *subfacet;
4709 struct flow_wildcards wc;
4710 struct xlate_out xout;
4711 struct xlate_in xin;
4713 COVERAGE_INC(facet_revalidate);
4715 /* Check that child subfacets still correspond to this facet. Tunnel
4716 * configuration changes could cause a subfacet's OpenFlow in_port to
4718 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4719 struct ofproto_dpif *recv_ofproto;
4720 struct flow recv_flow;
4723 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4724 subfacet->key_len, &recv_flow, NULL,
4725 &recv_ofproto, NULL);
4727 || recv_ofproto != ofproto
4728 || facet != facet_find(ofproto, &recv_flow)) {
4729 facet_remove(facet);
4734 flow_wildcards_init_catchall(&wc);
4735 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4737 /* Calculate new datapath actions.
4739 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4740 * emit a NetFlow expiration and, if so, we need to have the old state
4741 * around to properly compose it. */
4742 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4743 xlate_actions(&xin, &xout);
4744 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4746 /* A facet's slow path reason should only change under dramatic
4747 * circumstances. Rather than try to update everything, it's simpler to
4748 * remove the facet and start over.
4750 * More importantly, if a facet's wildcards change, it will be relatively
4751 * difficult to figure out if its subfacets still belong to it, and if not
4752 * which facet they may belong to. Again, to avoid the complexity, we
4753 * simply give up instead. */
4754 if (facet->xout.slow != xout.slow
4755 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4756 facet_remove(facet);
4757 xlate_out_uninit(&xout);
4758 rule_release(new_rule);
4762 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4763 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4764 if (subfacet->path == SF_FAST_PATH) {
4765 struct dpif_flow_stats stats;
4767 subfacet_install(subfacet, &xout.odp_actions, &stats);
4768 subfacet_update_stats(subfacet, &stats);
4772 facet_flush_stats(facet);
4774 ofpbuf_clear(&facet->xout.odp_actions);
4775 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4776 xout.odp_actions.size);
4779 /* Update 'facet' now that we've taken care of all the old state. */
4780 facet->xout.slow = xout.slow;
4781 facet->xout.has_learn = xout.has_learn;
4782 facet->xout.has_normal = xout.has_normal;
4783 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4784 facet->xout.nf_output_iface = xout.nf_output_iface;
4785 facet->xout.mirrors = xout.mirrors;
4786 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4787 facet->used = MAX(facet->used, new_rule->up.created);
4788 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4790 xlate_out_uninit(&xout);
4791 rule_release(new_rule);
4796 facet_reset_counters(struct facet *facet)
4798 facet->packet_count = 0;
4799 facet->byte_count = 0;
4800 facet->prev_packet_count = 0;
4801 facet->prev_byte_count = 0;
4802 facet->accounted_bytes = 0;
4806 facet_push_stats(struct facet *facet, bool may_learn)
4808 struct dpif_flow_stats stats;
4810 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4811 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4812 ovs_assert(facet->used >= facet->prev_used);
4814 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4815 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4816 stats.used = facet->used;
4817 stats.tcp_flags = facet->tcp_flags;
4819 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4820 struct ofproto_dpif *ofproto = facet->ofproto;
4821 struct ofport_dpif *in_port;
4822 struct rule_dpif *rule;
4823 struct xlate_in xin;
4825 facet->prev_packet_count = facet->packet_count;
4826 facet->prev_byte_count = facet->byte_count;
4827 facet->prev_used = facet->used;
4829 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4830 if (in_port && in_port->is_tunnel) {
4831 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4834 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4835 rule_credit_stats(rule, &stats);
4836 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4838 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4839 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4840 stats.n_packets, stats.n_bytes);
4842 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4844 xin.resubmit_stats = &stats;
4845 xin.may_learn = may_learn;
4846 xlate_actions_for_side_effects(&xin);
4852 push_all_stats__(bool run_fast)
4854 static long long int rl = LLONG_MIN;
4855 struct ofproto_dpif *ofproto;
4857 if (time_msec() < rl) {
4861 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4862 struct cls_cursor cursor;
4863 struct facet *facet;
4865 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4866 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4867 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4868 facet_push_stats(facet, false);
4873 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4876 rl = time_msec() + 100;
4880 push_all_stats(void)
4882 push_all_stats__(true);
4886 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4888 ovs_mutex_lock(&rule->stats_mutex);
4889 rule->packet_count += stats->n_packets;
4890 rule->byte_count += stats->n_bytes;
4891 ofproto_rule_update_used(&rule->up, stats->used);
4892 ovs_mutex_unlock(&rule->stats_mutex);
4897 static struct subfacet *
4898 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4899 size_t key_len, uint32_t key_hash)
4901 struct subfacet *subfacet;
4903 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4904 &backer->subfacets) {
4905 if (subfacet->key_len == key_len
4906 && !memcmp(key, subfacet->key, key_len)) {
4914 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4915 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4916 * existing subfacet if there is one, otherwise creates and returns a
4918 static struct subfacet *
4919 subfacet_create(struct facet *facet, struct flow_miss *miss,
4922 struct dpif_backer *backer = miss->ofproto->backer;
4923 enum odp_key_fitness key_fitness = miss->key_fitness;
4924 const struct nlattr *key = miss->key;
4925 size_t key_len = miss->key_len;
4927 struct subfacet *subfacet;
4929 key_hash = odp_flow_key_hash(key, key_len);
4931 if (list_is_empty(&facet->subfacets)) {
4932 subfacet = &facet->one_subfacet;
4934 subfacet = subfacet_find(backer, key, key_len, key_hash);
4936 if (subfacet->facet == facet) {
4940 /* This shouldn't happen. */
4941 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4942 subfacet_destroy(subfacet);
4945 subfacet = xmalloc(sizeof *subfacet);
4948 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4949 list_push_back(&facet->subfacets, &subfacet->list_node);
4950 subfacet->facet = facet;
4951 subfacet->key_fitness = key_fitness;
4952 subfacet->key = xmemdup(key, key_len);
4953 subfacet->key_len = key_len;
4954 subfacet->used = now;
4955 subfacet->created = now;
4956 subfacet->dp_packet_count = 0;
4957 subfacet->dp_byte_count = 0;
4958 subfacet->path = SF_NOT_INSTALLED;
4959 subfacet->backer = backer;
4961 backer->subfacet_add_count++;
4965 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4966 * its facet within 'ofproto', and frees it. */
4968 subfacet_destroy__(struct subfacet *subfacet)
4970 struct facet *facet = subfacet->facet;
4971 struct ofproto_dpif *ofproto = facet->ofproto;
4973 /* Update ofproto stats before uninstall the subfacet. */
4974 ofproto->backer->subfacet_del_count++;
4976 subfacet_uninstall(subfacet);
4977 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4978 list_remove(&subfacet->list_node);
4979 free(subfacet->key);
4980 if (subfacet != &facet->one_subfacet) {
4985 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4986 * last remaining subfacet in its facet destroys the facet too. */
4988 subfacet_destroy(struct subfacet *subfacet)
4990 struct facet *facet = subfacet->facet;
4992 if (list_is_singleton(&facet->subfacets)) {
4993 /* facet_remove() needs at least one subfacet (it will remove it). */
4994 facet_remove(facet);
4996 subfacet_destroy__(subfacet);
5001 subfacet_destroy_batch(struct dpif_backer *backer,
5002 struct subfacet **subfacets, int n)
5004 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5005 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5006 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5009 for (i = 0; i < n; i++) {
5010 ops[i].type = DPIF_OP_FLOW_DEL;
5011 ops[i].u.flow_del.key = subfacets[i]->key;
5012 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5013 ops[i].u.flow_del.stats = &stats[i];
5017 dpif_operate(backer->dpif, opsp, n);
5018 for (i = 0; i < n; i++) {
5019 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5020 subfacets[i]->path = SF_NOT_INSTALLED;
5021 subfacet_destroy(subfacets[i]);
5026 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5027 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5028 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5029 * since 'subfacet' was last updated.
5031 * Returns 0 if successful, otherwise a positive errno value. */
5033 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5034 struct dpif_flow_stats *stats)
5036 struct facet *facet = subfacet->facet;
5037 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5038 const struct nlattr *actions = odp_actions->data;
5039 size_t actions_len = odp_actions->size;
5040 struct odputil_keybuf maskbuf;
5043 uint64_t slow_path_stub[128 / 8];
5044 enum dpif_flow_put_flags flags;
5047 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5050 flags |= DPIF_FP_ZERO_STATS;
5053 if (path == SF_SLOW_PATH) {
5054 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5055 slow_path_stub, sizeof slow_path_stub,
5056 &actions, &actions_len);
5059 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5060 if (enable_megaflows) {
5061 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5062 &facet->flow, UINT32_MAX);
5065 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5066 subfacet->key_len, mask.data, mask.size,
5067 actions, actions_len, stats);
5070 subfacet_reset_dp_stats(subfacet, stats);
5074 COVERAGE_INC(subfacet_install_fail);
5076 subfacet->path = path;
5081 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5083 subfacet_uninstall(struct subfacet *subfacet)
5085 if (subfacet->path != SF_NOT_INSTALLED) {
5086 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5087 struct dpif_flow_stats stats;
5090 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5091 subfacet->key_len, &stats);
5092 subfacet_reset_dp_stats(subfacet, &stats);
5094 subfacet_update_stats(subfacet, &stats);
5096 subfacet->path = SF_NOT_INSTALLED;
5098 ovs_assert(subfacet->dp_packet_count == 0);
5099 ovs_assert(subfacet->dp_byte_count == 0);
5103 /* Resets 'subfacet''s datapath statistics counters. This should be called
5104 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5105 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5106 * was reset in the datapath. 'stats' will be modified to include only
5107 * statistics new since 'subfacet' was last updated. */
5109 subfacet_reset_dp_stats(struct subfacet *subfacet,
5110 struct dpif_flow_stats *stats)
5113 && subfacet->dp_packet_count <= stats->n_packets
5114 && subfacet->dp_byte_count <= stats->n_bytes) {
5115 stats->n_packets -= subfacet->dp_packet_count;
5116 stats->n_bytes -= subfacet->dp_byte_count;
5119 subfacet->dp_packet_count = 0;
5120 subfacet->dp_byte_count = 0;
5123 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5125 * Because of the meaning of a subfacet's counters, it only makes sense to do
5126 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5127 * represents a packet that was sent by hand or if it represents statistics
5128 * that have been cleared out of the datapath. */
5130 subfacet_update_stats(struct subfacet *subfacet,
5131 const struct dpif_flow_stats *stats)
5133 if (stats->n_packets || stats->used > subfacet->used) {
5134 struct facet *facet = subfacet->facet;
5136 subfacet->used = MAX(subfacet->used, stats->used);
5137 facet->used = MAX(facet->used, stats->used);
5138 facet->packet_count += stats->n_packets;
5139 facet->byte_count += stats->n_bytes;
5140 facet->tcp_flags |= stats->tcp_flags;
5146 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5147 * the fields that were relevant as part of the lookup. */
5149 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5150 struct flow_wildcards *wc, struct rule_dpif **rule)
5152 struct ofport_dpif *port;
5154 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
5157 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5159 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5160 flow->in_port.ofp_port);
5163 *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
5164 ofproto->no_packet_in_rule);
5165 ovs_rwlock_rdlock(&(*rule)->up.evict);
5169 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5170 const struct flow *flow, struct flow_wildcards *wc,
5171 uint8_t table_id, struct rule_dpif **rule)
5173 struct cls_rule *cls_rule;
5174 struct classifier *cls;
5178 if (table_id >= N_TABLES) {
5183 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5184 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5187 cls = &ofproto->up.tables[table_id].cls;
5188 ovs_rwlock_rdlock(&cls->rwlock);
5189 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5190 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5191 /* We must pretend that transport ports are unavailable. */
5192 struct flow ofpc_normal_flow = *flow;
5193 ofpc_normal_flow.tp_src = htons(0);
5194 ofpc_normal_flow.tp_dst = htons(0);
5195 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5196 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5197 cls_rule = &ofproto->drop_frags_rule->up.cr;
5199 flow_wildcards_init_exact(wc);
5202 cls_rule = classifier_lookup(cls, flow, wc);
5205 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
5206 if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
5207 /* The rule is in the process of being removed. Best we can do is
5208 * pretend it isn't there. */
5211 ovs_rwlock_unlock(&cls->rwlock);
5213 return *rule != NULL;
5216 /* Given a port configuration (specified as zero if there's no port), chooses
5217 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
5218 * flow table miss. */
5220 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
5221 struct rule_dpif *no_packet_in_rule)
5223 return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
5227 rule_release(struct rule_dpif *rule)
5230 ovs_rwlock_unlock(&rule->up.evict);
5235 complete_operation(struct rule_dpif *rule)
5237 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5239 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5241 struct dpif_completion *c = xmalloc(sizeof *c);
5242 c->op = rule->up.pending;
5243 list_push_back(&ofproto->completions, &c->list_node);
5245 ofoperation_complete(rule->up.pending, 0);
5249 static struct rule *
5252 struct rule_dpif *rule = xmalloc(sizeof *rule);
5257 rule_dealloc(struct rule *rule_)
5259 struct rule_dpif *rule = rule_dpif_cast(rule_);
5264 rule_construct(struct rule *rule_)
5266 struct rule_dpif *rule = rule_dpif_cast(rule_);
5267 ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
5268 ovs_mutex_lock(&rule->stats_mutex);
5269 rule->packet_count = 0;
5270 rule->byte_count = 0;
5271 ovs_mutex_unlock(&rule->stats_mutex);
5272 complete_operation(rule);
5277 rule_destruct(struct rule *rule_)
5279 struct rule_dpif *rule = rule_dpif_cast(rule_);
5280 complete_operation(rule);
5281 ovs_mutex_destroy(&rule->stats_mutex);
5285 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5287 struct rule_dpif *rule = rule_dpif_cast(rule_);
5289 /* push_all_stats() can handle flow misses which, when using the learn
5290 * action, can cause rules to be added and deleted. This can corrupt our
5291 * caller's datastructures which assume that rule_get_stats() doesn't have
5292 * an impact on the flow table. To be safe, we disable miss handling. */
5293 push_all_stats__(false);
5295 /* Start from historical data for 'rule' itself that are no longer tracked
5296 * in facets. This counts, for example, facets that have expired. */
5297 ovs_mutex_lock(&rule->stats_mutex);
5298 *packets = rule->packet_count;
5299 *bytes = rule->byte_count;
5300 ovs_mutex_unlock(&rule->stats_mutex);
5304 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5305 struct ofpbuf *packet)
5307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5308 struct dpif_flow_stats stats;
5309 struct xlate_out xout;
5310 struct xlate_in xin;
5312 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5313 rule_credit_stats(rule, &stats);
5315 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5316 xin.resubmit_stats = &stats;
5317 xlate_actions(&xin, &xout);
5319 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5320 xout.odp_actions.size, packet);
5322 xlate_out_uninit(&xout);
5326 rule_execute(struct rule *rule, const struct flow *flow,
5327 struct ofpbuf *packet)
5329 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5330 ofpbuf_delete(packet);
5335 rule_modify_actions(struct rule *rule_)
5337 struct rule_dpif *rule = rule_dpif_cast(rule_);
5339 complete_operation(rule);
5342 /* Sends 'packet' out 'ofport'.
5343 * May modify 'packet'.
5344 * Returns 0 if successful, otherwise a positive errno value. */
5346 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5349 uint64_t odp_actions_stub[1024 / 8];
5350 struct ofpbuf key, odp_actions;
5351 struct dpif_flow_stats stats;
5352 struct odputil_keybuf keybuf;
5353 struct ofpact_output output;
5354 struct xlate_out xout;
5355 struct xlate_in xin;
5357 union flow_in_port in_port_;
5360 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5361 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5363 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5364 in_port_.ofp_port = OFPP_NONE;
5365 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5366 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5368 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5370 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5371 output.port = ofport->up.ofp_port;
5374 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5375 xin.ofpacts_len = sizeof output;
5376 xin.ofpacts = &output.ofpact;
5377 xin.resubmit_stats = &stats;
5378 xlate_actions(&xin, &xout);
5380 error = dpif_execute(ofproto->backer->dpif,
5382 xout.odp_actions.data, xout.odp_actions.size,
5384 xlate_out_uninit(&xout);
5387 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5388 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5389 ovs_strerror(error));
5392 ofproto->stats.tx_packets++;
5393 ofproto->stats.tx_bytes += packet->size;
5397 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5398 * The action will state 'slow' as the reason that the action is in the slow
5399 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5400 * dump-flows" output to see why a flow is in the slow path.)
5402 * The 'stub_size' bytes in 'stub' will be used to store the action.
5403 * 'stub_size' must be large enough for the action.
5405 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5408 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5409 enum slow_path_reason slow,
5410 uint64_t *stub, size_t stub_size,
5411 const struct nlattr **actionsp, size_t *actions_lenp)
5413 union user_action_cookie cookie;
5416 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5417 cookie.slow_path.unused = 0;
5418 cookie.slow_path.reason = slow;
5420 ofpbuf_use_stack(&buf, stub, stub_size);
5421 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5422 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5424 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5426 odp_port_t odp_port;
5429 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5430 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5431 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5433 *actionsp = buf.data;
5434 *actions_lenp = buf.size;
5438 set_frag_handling(struct ofproto *ofproto_,
5439 enum ofp_config_flags frag_handling)
5441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5442 if (frag_handling != OFPC_FRAG_REASM) {
5443 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5451 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5452 const struct flow *flow,
5453 const struct ofpact *ofpacts, size_t ofpacts_len)
5455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5456 struct odputil_keybuf keybuf;
5457 struct dpif_flow_stats stats;
5458 struct xlate_out xout;
5459 struct xlate_in xin;
5463 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5464 odp_flow_key_from_flow(&key, flow,
5465 ofp_port_to_odp_port(ofproto,
5466 flow->in_port.ofp_port));
5468 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5470 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5471 xin.resubmit_stats = &stats;
5472 xin.ofpacts_len = ofpacts_len;
5473 xin.ofpacts = ofpacts;
5475 xlate_actions(&xin, &xout);
5476 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5477 xout.odp_actions.data, xout.odp_actions.size, packet);
5478 xlate_out_uninit(&xout);
5486 set_netflow(struct ofproto *ofproto_,
5487 const struct netflow_options *netflow_options)
5489 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5491 if (netflow_options) {
5492 if (!ofproto->netflow) {
5493 ofproto->netflow = netflow_create();
5494 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5496 return netflow_set_options(ofproto->netflow, netflow_options);
5497 } else if (ofproto->netflow) {
5498 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5499 netflow_destroy(ofproto->netflow);
5500 ofproto->netflow = NULL;
5507 get_netflow_ids(const struct ofproto *ofproto_,
5508 uint8_t *engine_type, uint8_t *engine_id)
5510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5512 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5516 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5518 if (!facet_is_controller_flow(facet) &&
5519 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5520 struct subfacet *subfacet;
5521 struct ofexpired expired;
5523 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5524 if (subfacet->path == SF_FAST_PATH) {
5525 struct dpif_flow_stats stats;
5527 subfacet_install(subfacet, &facet->xout.odp_actions,
5529 subfacet_update_stats(subfacet, &stats);
5533 expired.flow = facet->flow;
5534 expired.packet_count = facet->packet_count;
5535 expired.byte_count = facet->byte_count;
5536 expired.used = facet->used;
5537 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5542 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5544 struct cls_cursor cursor;
5545 struct facet *facet;
5547 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5548 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5549 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5550 send_active_timeout(ofproto, facet);
5552 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5555 static struct ofproto_dpif *
5556 ofproto_dpif_lookup(const char *name)
5558 struct ofproto_dpif *ofproto;
5560 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5561 hash_string(name, 0), &all_ofproto_dpifs) {
5562 if (!strcmp(ofproto->up.name, name)) {
5570 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5571 const char *argv[], void *aux OVS_UNUSED)
5573 struct ofproto_dpif *ofproto;
5576 ofproto = ofproto_dpif_lookup(argv[1]);
5578 unixctl_command_reply_error(conn, "no such bridge");
5581 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5582 mac_learning_flush(ofproto->ml);
5583 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5585 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5586 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5587 mac_learning_flush(ofproto->ml);
5588 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5592 unixctl_command_reply(conn, "table successfully flushed");
5595 static struct ofport_dpif *
5596 ofbundle_get_a_port(const struct ofbundle *bundle)
5598 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5603 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5604 const char *argv[], void *aux OVS_UNUSED)
5606 struct ds ds = DS_EMPTY_INITIALIZER;
5607 const struct ofproto_dpif *ofproto;
5608 const struct mac_entry *e;
5610 ofproto = ofproto_dpif_lookup(argv[1]);
5612 unixctl_command_reply_error(conn, "no such bridge");
5616 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5617 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5618 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5619 struct ofbundle *bundle = e->port.p;
5620 char name[OFP_MAX_PORT_NAME_LEN];
5622 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5624 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5625 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5626 mac_entry_age(ofproto->ml, e));
5628 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5629 unixctl_command_reply(conn, ds_cstr(&ds));
5634 struct xlate_out xout;
5635 struct xlate_in xin;
5641 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5643 ds_put_char_multiple(result, '\t', level);
5645 ds_put_cstr(result, "No match\n");
5649 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5650 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5651 cls_rule_format(&rule->up.cr, result);
5652 ds_put_char(result, '\n');
5654 ds_put_char_multiple(result, '\t', level);
5655 ds_put_cstr(result, "OpenFlow ");
5656 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5657 ds_put_char(result, '\n');
5661 trace_format_flow(struct ds *result, int level, const char *title,
5662 struct trace_ctx *trace)
5664 ds_put_char_multiple(result, '\t', level);
5665 ds_put_format(result, "%s: ", title);
5666 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5667 ds_put_cstr(result, "unchanged");
5669 flow_format(result, &trace->xin.flow);
5670 trace->flow = trace->xin.flow;
5672 ds_put_char(result, '\n');
5676 trace_format_regs(struct ds *result, int level, const char *title,
5677 struct trace_ctx *trace)
5681 ds_put_char_multiple(result, '\t', level);
5682 ds_put_format(result, "%s:", title);
5683 for (i = 0; i < FLOW_N_REGS; i++) {
5684 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5686 ds_put_char(result, '\n');
5690 trace_format_odp(struct ds *result, int level, const char *title,
5691 struct trace_ctx *trace)
5693 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5695 ds_put_char_multiple(result, '\t', level);
5696 ds_put_format(result, "%s: ", title);
5697 format_odp_actions(result, odp_actions->data, odp_actions->size);
5698 ds_put_char(result, '\n');
5702 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5704 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5705 struct ds *result = trace->result;
5707 ds_put_char(result, '\n');
5708 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5709 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5710 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5711 trace_format_rule(result, recurse + 1, rule);
5715 trace_report(struct xlate_in *xin, const char *s, int recurse)
5717 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5718 struct ds *result = trace->result;
5720 ds_put_char_multiple(result, '\t', recurse);
5721 ds_put_cstr(result, s);
5722 ds_put_char(result, '\n');
5726 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5727 void *aux OVS_UNUSED)
5729 const struct dpif_backer *backer;
5730 struct ofproto_dpif *ofproto;
5731 struct ofpbuf odp_key, odp_mask;
5732 struct ofpbuf *packet;
5740 ofpbuf_init(&odp_key, 0);
5741 ofpbuf_init(&odp_mask, 0);
5743 /* Handle "-generate" or a hex string as the last argument. */
5744 if (!strcmp(argv[argc - 1], "-generate")) {
5745 packet = ofpbuf_new(0);
5748 const char *error = eth_from_hex(argv[argc - 1], &packet);
5751 } else if (argc == 4) {
5752 /* The 3-argument form must end in "-generate' or a hex string. */
5753 unixctl_command_reply_error(conn, error);
5758 /* Parse the flow and determine whether a datapath or
5759 * bridge is specified. If function odp_flow_key_from_string()
5760 * returns 0, the flow is a odp_flow. If function
5761 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5762 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5763 /* If the odp_flow is the second argument,
5764 * the datapath name is the first argument. */
5766 const char *dp_type;
5767 if (!strncmp(argv[1], "ovs-", 4)) {
5768 dp_type = argv[1] + 4;
5772 backer = shash_find_data(&all_dpif_backers, dp_type);
5774 unixctl_command_reply_error(conn, "Cannot find datapath "
5779 /* No datapath name specified, so there should be only one
5781 struct shash_node *node;
5782 if (shash_count(&all_dpif_backers) != 1) {
5783 unixctl_command_reply_error(conn, "Must specify datapath "
5784 "name, there is more than one type of datapath");
5787 node = shash_first(&all_dpif_backers);
5788 backer = node->data;
5791 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5792 NULL, &ofproto, NULL)) {
5793 unixctl_command_reply_error(conn, "Invalid datapath flow");
5796 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5797 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5799 unixctl_command_reply_error(conn, "Must specify bridge name");
5803 ofproto = ofproto_dpif_lookup(argv[1]);
5805 unixctl_command_reply_error(conn, "Unknown bridge name");
5809 unixctl_command_reply_error(conn, "Bad flow syntax");
5813 /* Generate a packet, if requested. */
5815 if (!packet->size) {
5816 flow_compose(packet, &flow);
5818 union flow_in_port in_port_;
5820 in_port_ = flow.in_port;
5821 ds_put_cstr(&result, "Packet: ");
5822 s = ofp_packet_to_string(packet->data, packet->size);
5823 ds_put_cstr(&result, s);
5826 /* Use the metadata from the flow and the packet argument
5827 * to reconstruct the flow. */
5828 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5833 ofproto_trace(ofproto, &flow, packet, &result);
5834 unixctl_command_reply(conn, ds_cstr(&result));
5837 ds_destroy(&result);
5838 ofpbuf_delete(packet);
5839 ofpbuf_uninit(&odp_key);
5840 ofpbuf_uninit(&odp_mask);
5844 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5845 const struct ofpbuf *packet, struct ds *ds)
5847 struct rule_dpif *rule;
5849 ds_put_cstr(ds, "Flow: ");
5850 flow_format(ds, flow);
5851 ds_put_char(ds, '\n');
5853 rule_dpif_lookup(ofproto, flow, NULL, &rule);
5855 trace_format_rule(ds, 0, rule);
5856 if (rule == ofproto->miss_rule) {
5857 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5858 } else if (rule == ofproto->no_packet_in_rule) {
5859 ds_put_cstr(ds, "\nNo match, packets dropped because "
5860 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5861 } else if (rule == ofproto->drop_frags_rule) {
5862 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5863 "and the fragment handling mode is \"drop\".\n");
5867 uint64_t odp_actions_stub[1024 / 8];
5868 struct ofpbuf odp_actions;
5869 struct trace_ctx trace;
5873 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5876 ofpbuf_use_stub(&odp_actions,
5877 odp_actions_stub, sizeof odp_actions_stub);
5878 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5879 trace.xin.resubmit_hook = trace_resubmit;
5880 trace.xin.report_hook = trace_report;
5882 xlate_actions(&trace.xin, &trace.xout);
5884 ds_put_char(ds, '\n');
5885 trace_format_flow(ds, 0, "Final flow", &trace);
5887 match_init(&match, flow, &trace.xout.wc);
5888 ds_put_cstr(ds, "Relevant fields: ");
5889 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5890 ds_put_char(ds, '\n');
5892 ds_put_cstr(ds, "Datapath actions: ");
5893 format_odp_actions(ds, trace.xout.odp_actions.data,
5894 trace.xout.odp_actions.size);
5896 if (trace.xout.slow) {
5897 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5898 "slow path because it:");
5899 switch (trace.xout.slow) {
5901 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5904 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5907 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5910 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5912 case SLOW_CONTROLLER:
5913 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5914 "to the OpenFlow controller.");
5921 xlate_out_uninit(&trace.xout);
5928 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5929 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5932 unixctl_command_reply(conn, NULL);
5936 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5937 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5940 unixctl_command_reply(conn, NULL);
5943 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5944 * 'reply' describing the results. */
5946 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5948 struct cls_cursor cursor;
5949 struct facet *facet;
5953 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5954 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5955 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5956 if (!facet_check_consistency(facet)) {
5960 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5962 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5966 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5967 ofproto->up.name, errors);
5969 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5974 ofproto_dpif_self_check(struct unixctl_conn *conn,
5975 int argc, const char *argv[], void *aux OVS_UNUSED)
5977 struct ds reply = DS_EMPTY_INITIALIZER;
5978 struct ofproto_dpif *ofproto;
5981 ofproto = ofproto_dpif_lookup(argv[1]);
5983 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5984 "ofproto/list for help)");
5987 ofproto_dpif_self_check__(ofproto, &reply);
5989 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5990 ofproto_dpif_self_check__(ofproto, &reply);
5994 unixctl_command_reply(conn, ds_cstr(&reply));
5998 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5999 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6000 * to destroy 'ofproto_shash' and free the returned value. */
6001 static const struct shash_node **
6002 get_ofprotos(struct shash *ofproto_shash)
6004 const struct ofproto_dpif *ofproto;
6006 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6007 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6008 shash_add_nocopy(ofproto_shash, name, ofproto);
6011 return shash_sort(ofproto_shash);
6015 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6016 const char *argv[] OVS_UNUSED,
6017 void *aux OVS_UNUSED)
6019 struct ds ds = DS_EMPTY_INITIALIZER;
6020 struct shash ofproto_shash;
6021 const struct shash_node **sorted_ofprotos;
6024 shash_init(&ofproto_shash);
6025 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6026 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6027 const struct shash_node *node = sorted_ofprotos[i];
6028 ds_put_format(&ds, "%s\n", node->name);
6031 shash_destroy(&ofproto_shash);
6032 free(sorted_ofprotos);
6034 unixctl_command_reply(conn, ds_cstr(&ds));
6039 show_dp_rates(struct ds *ds, const char *heading,
6040 const struct avg_subfacet_rates *rates)
6042 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6043 heading, rates->add_rate, rates->del_rate);
6047 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6049 const struct shash_node **ofprotos;
6050 struct ofproto_dpif *ofproto;
6051 struct shash ofproto_shash;
6052 uint64_t n_hit, n_missed;
6053 long long int minutes;
6056 n_hit = n_missed = 0;
6057 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6058 if (ofproto->backer == backer) {
6059 n_missed += ofproto->n_missed;
6060 n_hit += ofproto->n_hit;
6064 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6065 dpif_name(backer->dpif), n_hit, n_missed);
6066 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6067 " life span: %lldms\n", hmap_count(&backer->subfacets),
6068 backer->avg_n_subfacet, backer->max_n_subfacet,
6069 backer->avg_subfacet_life);
6071 minutes = (time_msec() - backer->created) / (1000 * 60);
6072 if (minutes >= 60) {
6073 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6075 if (minutes >= 60 * 24) {
6076 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6078 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6080 shash_init(&ofproto_shash);
6081 ofprotos = get_ofprotos(&ofproto_shash);
6082 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6083 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6084 const struct shash_node **ports;
6087 if (ofproto->backer != backer) {
6091 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6092 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6094 ports = shash_sort(&ofproto->up.port_by_name);
6095 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6096 const struct shash_node *node = ports[j];
6097 struct ofport *ofport = node->data;
6099 odp_port_t odp_port;
6101 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6104 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6105 if (odp_port != ODPP_NONE) {
6106 ds_put_format(ds, "%"PRIu32":", odp_port);
6108 ds_put_cstr(ds, "none:");
6111 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6114 if (!netdev_get_config(ofport->netdev, &config)) {
6115 const struct smap_node **nodes;
6118 nodes = smap_sort(&config);
6119 for (i = 0; i < smap_count(&config); i++) {
6120 const struct smap_node *node = nodes[i];
6121 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6122 node->key, node->value);
6126 smap_destroy(&config);
6128 ds_put_char(ds, ')');
6129 ds_put_char(ds, '\n');
6133 shash_destroy(&ofproto_shash);
6138 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6139 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6141 struct ds ds = DS_EMPTY_INITIALIZER;
6142 const struct shash_node **backers;
6145 backers = shash_sort(&all_dpif_backers);
6146 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6147 dpif_show_backer(backers[i]->data, &ds);
6151 unixctl_command_reply(conn, ds_cstr(&ds));
6155 /* Dump the megaflow (facet) cache. This is useful to check the
6156 * correctness of flow wildcarding, since the same mechanism is used for
6157 * both xlate caching and kernel wildcarding.
6159 * It's important to note that in the output the flow description uses
6160 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6162 * This command is only needed for advanced debugging, so it's not
6163 * documented in the man page. */
6165 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6166 int argc OVS_UNUSED, const char *argv[],
6167 void *aux OVS_UNUSED)
6169 struct ds ds = DS_EMPTY_INITIALIZER;
6170 const struct ofproto_dpif *ofproto;
6171 long long int now = time_msec();
6172 struct cls_cursor cursor;
6173 struct facet *facet;
6175 ofproto = ofproto_dpif_lookup(argv[1]);
6177 unixctl_command_reply_error(conn, "no such bridge");
6181 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
6182 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6183 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6184 cls_rule_format(&facet->cr, &ds);
6185 ds_put_cstr(&ds, ", ");
6186 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6187 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6188 ds_put_cstr(&ds, "Datapath actions: ");
6189 if (facet->xout.slow) {
6190 uint64_t slow_path_stub[128 / 8];
6191 const struct nlattr *actions;
6194 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6195 slow_path_stub, sizeof slow_path_stub,
6196 &actions, &actions_len);
6197 format_odp_actions(&ds, actions, actions_len);
6199 format_odp_actions(&ds, facet->xout.odp_actions.data,
6200 facet->xout.odp_actions.size);
6202 ds_put_cstr(&ds, "\n");
6204 ovs_rwlock_unlock(&ofproto->facets.rwlock);
6206 ds_chomp(&ds, '\n');
6207 unixctl_command_reply(conn, ds_cstr(&ds));
6211 /* Disable using the megaflows.
6213 * This command is only needed for advanced debugging, so it's not
6214 * documented in the man page. */
6216 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6217 int argc OVS_UNUSED,
6218 const char *argv[] OVS_UNUSED,
6219 void *aux OVS_UNUSED)
6221 struct ofproto_dpif *ofproto;
6223 enable_megaflows = false;
6225 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6226 flush(&ofproto->up);
6229 unixctl_command_reply(conn, "megaflows disabled");
6232 /* Re-enable using megaflows.
6234 * This command is only needed for advanced debugging, so it's not
6235 * documented in the man page. */
6237 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6238 int argc OVS_UNUSED,
6239 const char *argv[] OVS_UNUSED,
6240 void *aux OVS_UNUSED)
6242 struct ofproto_dpif *ofproto;
6244 enable_megaflows = true;
6246 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6247 flush(&ofproto->up);
6250 unixctl_command_reply(conn, "megaflows enabled");
6254 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6255 int argc OVS_UNUSED, const char *argv[],
6256 void *aux OVS_UNUSED)
6258 struct ds ds = DS_EMPTY_INITIALIZER;
6259 const struct ofproto_dpif *ofproto;
6260 struct subfacet *subfacet;
6262 ofproto = ofproto_dpif_lookup(argv[1]);
6264 unixctl_command_reply_error(conn, "no such bridge");
6268 update_stats(ofproto->backer);
6270 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6271 struct facet *facet = subfacet->facet;
6272 struct odputil_keybuf maskbuf;
6275 if (facet->ofproto != ofproto) {
6279 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6280 if (enable_megaflows) {
6281 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6282 &facet->flow, UINT32_MAX);
6285 odp_flow_format(subfacet->key, subfacet->key_len,
6286 mask.data, mask.size, &ds, false);
6288 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6289 subfacet->dp_packet_count, subfacet->dp_byte_count);
6290 if (subfacet->used) {
6291 ds_put_format(&ds, "%.3fs",
6292 (time_msec() - subfacet->used) / 1000.0);
6294 ds_put_format(&ds, "never");
6296 if (subfacet->facet->tcp_flags) {
6297 ds_put_cstr(&ds, ", flags:");
6298 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6301 ds_put_cstr(&ds, ", actions:");
6302 if (facet->xout.slow) {
6303 uint64_t slow_path_stub[128 / 8];
6304 const struct nlattr *actions;
6307 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6308 slow_path_stub, sizeof slow_path_stub,
6309 &actions, &actions_len);
6310 format_odp_actions(&ds, actions, actions_len);
6312 format_odp_actions(&ds, facet->xout.odp_actions.data,
6313 facet->xout.odp_actions.size);
6315 ds_put_char(&ds, '\n');
6318 unixctl_command_reply(conn, ds_cstr(&ds));
6323 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6324 int argc OVS_UNUSED, const char *argv[],
6325 void *aux OVS_UNUSED)
6327 struct ds ds = DS_EMPTY_INITIALIZER;
6328 struct ofproto_dpif *ofproto;
6330 ofproto = ofproto_dpif_lookup(argv[1]);
6332 unixctl_command_reply_error(conn, "no such bridge");
6336 flush(&ofproto->up);
6338 unixctl_command_reply(conn, ds_cstr(&ds));
6343 ofproto_dpif_unixctl_init(void)
6345 static bool registered;
6351 unixctl_command_register(
6353 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6354 1, 3, ofproto_unixctl_trace, NULL);
6355 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6356 ofproto_unixctl_fdb_flush, NULL);
6357 unixctl_command_register("fdb/show", "bridge", 1, 1,
6358 ofproto_unixctl_fdb_show, NULL);
6359 unixctl_command_register("ofproto/clog", "", 0, 0,
6360 ofproto_dpif_clog, NULL);
6361 unixctl_command_register("ofproto/unclog", "", 0, 0,
6362 ofproto_dpif_unclog, NULL);
6363 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6364 ofproto_dpif_self_check, NULL);
6365 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6366 ofproto_unixctl_dpif_dump_dps, NULL);
6367 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6369 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6370 ofproto_unixctl_dpif_dump_flows, NULL);
6371 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6372 ofproto_unixctl_dpif_del_flows, NULL);
6373 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6374 ofproto_unixctl_dpif_dump_megaflows, NULL);
6375 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6376 ofproto_unixctl_dpif_disable_megaflows, NULL);
6377 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6378 ofproto_unixctl_dpif_enable_megaflows, NULL);
6381 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6383 * This is deprecated. It is only for compatibility with broken device drivers
6384 * in old versions of Linux that do not properly support VLANs when VLAN
6385 * devices are not used. When broken device drivers are no longer in
6386 * widespread use, we will delete these interfaces. */
6389 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6392 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6394 if (realdev_ofp_port == ofport->realdev_ofp_port
6395 && vid == ofport->vlandev_vid) {
6399 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6401 if (ofport->realdev_ofp_port) {
6404 if (realdev_ofp_port && ofport->bundle) {
6405 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6406 * themselves be part of a bundle. */
6407 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6410 ofport->realdev_ofp_port = realdev_ofp_port;
6411 ofport->vlandev_vid = vid;
6413 if (realdev_ofp_port) {
6414 vsp_add(ofport, realdev_ofp_port, vid);
6421 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6423 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6427 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6428 OVS_EXCLUDED(ofproto->vsp_mutex)
6432 ovs_mutex_lock(&ofproto->vsp_mutex);
6433 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6434 ovs_mutex_unlock(&ofproto->vsp_mutex);
6439 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6440 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6441 OVS_REQUIRES(ofproto->vsp_mutex)
6443 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6444 int vid = vlan_tci_to_vid(vlan_tci);
6445 const struct vlan_splinter *vsp;
6447 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6448 hash_realdev_vid(realdev_ofp_port, vid),
6449 &ofproto->realdev_vid_map) {
6450 if (vsp->realdev_ofp_port == realdev_ofp_port
6451 && vsp->vid == vid) {
6452 return vsp->vlandev_ofp_port;
6456 return realdev_ofp_port;
6459 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6460 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6461 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6462 * 'vlan_tci' 9, it would return the port number of eth0.9.
6464 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6465 * function just returns its 'realdev_ofp_port' argument. */
6467 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6468 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6469 OVS_EXCLUDED(ofproto->vsp_mutex)
6473 ovs_mutex_lock(&ofproto->vsp_mutex);
6474 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6475 ovs_mutex_unlock(&ofproto->vsp_mutex);
6479 static struct vlan_splinter *
6480 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6482 struct vlan_splinter *vsp;
6484 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6485 hash_ofp_port(vlandev_ofp_port),
6486 &ofproto->vlandev_map) {
6487 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6495 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6496 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6497 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6498 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6499 * eth0 and store 9 in '*vid'.
6501 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6502 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6505 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6506 ofp_port_t vlandev_ofp_port, int *vid)
6507 OVS_REQUIRES(ofproto->vsp_mutex)
6509 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6510 const struct vlan_splinter *vsp;
6512 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6517 return vsp->realdev_ofp_port;
6523 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6524 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6525 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6526 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6527 * always the case unless VLAN splinters are enabled), returns false without
6528 * making any changes. */
6530 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6531 OVS_EXCLUDED(ofproto->vsp_mutex)
6536 ovs_mutex_lock(&ofproto->vsp_mutex);
6537 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6538 ovs_mutex_unlock(&ofproto->vsp_mutex);
6543 /* Cause the flow to be processed as if it came in on the real device with
6544 * the VLAN device's VLAN ID. */
6545 flow->in_port.ofp_port = realdev;
6546 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6551 vsp_remove(struct ofport_dpif *port)
6553 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6554 struct vlan_splinter *vsp;
6556 ovs_mutex_lock(&ofproto->vsp_mutex);
6557 vsp = vlandev_find(ofproto, port->up.ofp_port);
6559 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6560 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6563 port->realdev_ofp_port = 0;
6565 VLOG_ERR("missing vlan device record");
6567 ovs_mutex_unlock(&ofproto->vsp_mutex);
6571 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6575 ovs_mutex_lock(&ofproto->vsp_mutex);
6576 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6577 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6578 == realdev_ofp_port)) {
6579 struct vlan_splinter *vsp;
6581 vsp = xmalloc(sizeof *vsp);
6582 vsp->realdev_ofp_port = realdev_ofp_port;
6583 vsp->vlandev_ofp_port = port->up.ofp_port;
6586 port->realdev_ofp_port = realdev_ofp_port;
6588 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6589 hash_ofp_port(port->up.ofp_port));
6590 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6591 hash_realdev_vid(realdev_ofp_port, vid));
6593 VLOG_ERR("duplicate vlan device record");
6595 ovs_mutex_unlock(&ofproto->vsp_mutex);
6599 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6601 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6602 return ofport ? ofport->odp_port : ODPP_NONE;
6605 struct ofport_dpif *
6606 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6608 struct ofport_dpif *port;
6610 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6611 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6612 &backer->odp_to_ofport_map) {
6613 if (port->odp_port == odp_port) {
6614 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6619 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6624 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6626 struct ofport_dpif *port;
6628 port = odp_port_to_ofport(ofproto->backer, odp_port);
6629 if (port && &ofproto->up == port->up.ofproto) {
6630 return port->up.ofp_port;
6636 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6637 * most heavily weighted element. 'base' designates the rate of decay: after
6638 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6641 exp_mavg(double *avg, int base, double new)
6643 *avg = (*avg * (base - 1) + new) / base;
6647 update_moving_averages(struct dpif_backer *backer)
6649 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6650 long long int minutes = (time_msec() - backer->created) / min_ms;
6653 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6655 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6658 backer->lifetime.add_rate = 0.0;
6659 backer->lifetime.del_rate = 0.0;
6662 /* Update hourly averages on the minute boundaries. */
6663 if (time_msec() - backer->last_minute >= min_ms) {
6664 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6665 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6667 /* Update daily averages on the hour boundaries. */
6668 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6669 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6670 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6673 backer->total_subfacet_add_count += backer->subfacet_add_count;
6674 backer->total_subfacet_del_count += backer->subfacet_del_count;
6675 backer->subfacet_add_count = 0;
6676 backer->subfacet_del_count = 0;
6677 backer->last_minute += min_ms;
6681 const struct ofproto_class ofproto_dpif_class = {
6716 port_is_lacp_current,
6717 NULL, /* rule_choose_table */
6724 rule_modify_actions,
6738 get_stp_port_status,
6745 is_mirror_output_bundle,
6746 forward_bpdu_changed,
6747 set_mac_table_config,
6749 NULL, /* meter_get_features */
6750 NULL, /* meter_set */
6751 NULL, /* meter_get */
6752 NULL, /* meter_del */