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"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-governor.h"
54 #include "ofproto-dpif-ipfix.h"
55 #include "ofproto-dpif-mirror.h"
56 #include "ofproto-dpif-monitor.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
60 #include "poll-loop.h"
66 #include "unaligned.h"
68 #include "vlan-bitmap.h"
71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
73 COVERAGE_DEFINE(ofproto_dpif_expired);
74 COVERAGE_DEFINE(facet_revalidate);
75 COVERAGE_DEFINE(facet_unexpected);
76 COVERAGE_DEFINE(facet_create);
77 COVERAGE_DEFINE(facet_remove);
78 COVERAGE_DEFINE(subfacet_create);
79 COVERAGE_DEFINE(subfacet_destroy);
80 COVERAGE_DEFINE(subfacet_install_fail);
81 COVERAGE_DEFINE(packet_in_overflow);
83 /* Number of implemented OpenFlow tables. */
84 enum { N_TABLES = 255 };
85 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
86 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
96 * - Do include packets and bytes from facets that have been deleted or
97 * whose own statistics have been folded into the rule.
99 * - Do include packets and bytes sent "by hand" that were accounted to
100 * the rule without any facet being involved (this is a rare corner
101 * case in rule_execute()).
103 * - Do not include packet or bytes that can be obtained from any facet's
104 * packet_count or byte_count member or that can be obtained from the
105 * datapath by, e.g., dpif_flow_get() for any subfacet.
107 struct ovs_mutex stats_mutex;
108 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
109 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
112 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
120 * - Do include packets and bytes from facets that have been deleted or
121 * whose own statistics have been folded into the rule.
123 * - Do include packets and bytes sent "by hand" that were accounted to
124 * the rule without any facet being involved (this is a rare corner
125 * case in rule_execute()).
127 * - Do not include packet or bytes that can be obtained from any facet's
128 * packet_count or byte_count member or that can be obtained from the
129 * datapath by, e.g., dpif_flow_get() for any subfacet.
131 struct ovs_mutex stats_mutex;
132 uint64_t packet_count OVS_GUARDED; /* Number of packets received. */
133 uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */
134 struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */
138 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
139 struct ofproto_dpif *ofproto; /* Owning ofproto. */
140 void *aux; /* Key supplied by ofproto's client. */
141 char *name; /* Identifier for log messages. */
144 struct list ports; /* Contains "struct ofport"s. */
145 enum port_vlan_mode vlan_mode; /* VLAN mode */
146 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
147 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
148 * NULL if all VLANs are trunked. */
149 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
150 struct bond *bond; /* Nonnull iff more than one port. */
151 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
154 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
157 static void bundle_remove(struct ofport *);
158 static void bundle_update(struct ofbundle *);
159 static void bundle_destroy(struct ofbundle *);
160 static void bundle_del_port(struct ofport_dpif *);
161 static void bundle_run(struct ofbundle *);
162 static void bundle_wait(struct ofbundle *);
164 static void stp_run(struct ofproto_dpif *ofproto);
165 static void stp_wait(struct ofproto_dpif *ofproto);
166 static int set_stp_port(struct ofport *,
167 const struct ofproto_port_stp_settings *);
169 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
170 enum slow_path_reason,
171 uint64_t *stub, size_t stub_size,
172 const struct nlattr **actionsp,
173 size_t *actions_lenp);
175 /* A subfacet (see "struct subfacet" below) has three possible installation
178 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
179 * case just after the subfacet is created, just before the subfacet is
180 * destroyed, or if the datapath returns an error when we try to install a
183 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
185 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
186 * ofproto_dpif is installed in the datapath.
189 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
190 SF_FAST_PATH, /* Full actions are installed. */
191 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
194 /* A dpif flow and actions associated with a facet.
196 * See also the large comment on struct facet. */
199 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
200 struct list list_node; /* In struct facet's 'facets' list. */
201 struct facet *facet; /* Owning facet. */
202 struct dpif_backer *backer; /* Owning backer. */
207 long long int used; /* Time last used; time created if not used. */
208 long long int created; /* Time created. */
210 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
211 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
213 enum subfacet_path path; /* Installed in datapath? */
216 #define SUBFACET_DESTROY_MAX_BATCH 50
218 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *,
220 static struct subfacet *subfacet_find(struct dpif_backer *,
221 const struct nlattr *key, size_t key_len,
223 static void subfacet_destroy(struct subfacet *);
224 static void subfacet_destroy__(struct subfacet *);
225 static void subfacet_destroy_batch(struct dpif_backer *,
226 struct subfacet **, int n);
227 static void subfacet_reset_dp_stats(struct subfacet *,
228 struct dpif_flow_stats *);
229 static void subfacet_update_stats(struct subfacet *,
230 const struct dpif_flow_stats *);
231 static int subfacet_install(struct subfacet *,
232 const struct ofpbuf *odp_actions,
233 struct dpif_flow_stats *);
234 static void subfacet_uninstall(struct subfacet *);
236 /* A unique, non-overlapping instantiation of an OpenFlow flow.
238 * A facet associates a "struct flow", which represents the Open vSwitch
239 * userspace idea of an exact-match flow, with one or more subfacets.
240 * While the facet is created based on an exact-match flow, it is stored
241 * within the ofproto based on the wildcards that could be expressed
242 * based on the flow table and other configuration. (See the 'wc'
243 * description in "struct xlate_out" for more details.)
245 * Each subfacet tracks the datapath's idea of the flow equivalent to
246 * the facet. When the kernel module (or other dpif implementation) and
247 * Open vSwitch userspace agree on the definition of a flow key, there
248 * is exactly one subfacet per facet. If the dpif implementation
249 * supports more-specific flow matching than userspace, however, a facet
250 * can have more than one subfacet. Examples include the dpif
251 * implementation not supporting the same wildcards as userspace or some
252 * distinction in flow that userspace simply doesn't understand.
254 * Flow expiration works in terms of subfacets, so a facet must have at
255 * least one subfacet or it will never expire, leaking memory. */
258 struct ofproto_dpif *ofproto;
261 struct list subfacets;
262 long long int used; /* Time last used; time created if not used. */
265 struct flow flow; /* Flow of the creating subfacet. */
266 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
270 * - Do include packets and bytes sent "by hand", e.g. with
273 * - Do include packets and bytes that were obtained from the datapath
274 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
275 * DPIF_FP_ZERO_STATS).
277 * - Do not include packets or bytes that can be obtained from the
278 * datapath for any existing subfacet.
280 uint64_t packet_count; /* Number of packets received. */
281 uint64_t byte_count; /* Number of bytes received. */
283 /* Resubmit statistics. */
284 uint64_t prev_packet_count; /* Number of packets from last stats push. */
285 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
286 long long int prev_used; /* Used time from last stats push. */
289 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
290 uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */
292 struct xlate_out xout;
294 /* Storage for a single subfacet, to reduce malloc() time and space
295 * overhead. (A facet always has at least one subfacet and in the common
296 * case has exactly one subfacet. However, 'one_subfacet' may not
297 * always be valid, since it could have been removed after newer
298 * subfacets were pushed onto the 'subfacets' list.) */
299 struct subfacet one_subfacet;
301 long long int learn_rl; /* Rate limiter for facet_learn(). */
304 static struct facet *facet_create(const struct flow_miss *);
305 static void facet_remove(struct facet *);
306 static void facet_free(struct facet *);
308 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
309 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
310 const struct flow *);
311 static bool facet_revalidate(struct facet *);
312 static bool facet_check_consistency(struct facet *);
314 static void facet_flush_stats(struct facet *);
316 static void facet_reset_counters(struct facet *);
317 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
318 struct dpif_flow_stats *, bool may_learn);
319 static void facet_push_stats(struct facet *, bool may_learn);
320 static void facet_learn(struct facet *);
321 static void facet_account(struct facet *);
322 static void push_all_stats(void);
324 static bool facet_is_controller_flow(struct facet *);
327 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
331 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
332 struct list bundle_node; /* In struct ofbundle's "ports" list. */
333 struct cfm *cfm; /* Connectivity Fault Management, if any. */
334 struct bfd *bfd; /* BFD, if any. */
335 bool may_enable; /* May be enabled in bonds. */
336 bool is_tunnel; /* This port is a tunnel. */
337 bool is_layer3; /* This is a layer 3 port. */
338 long long int carrier_seq; /* Carrier status changes. */
339 struct ofport_dpif *peer; /* Peer if patch port. */
342 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
343 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
344 long long int stp_state_entered;
346 /* Queue to DSCP mapping. */
347 struct ofproto_port_queue *qdscp;
350 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
352 * This is deprecated. It is only for compatibility with broken device
353 * drivers in old versions of Linux that do not properly support VLANs when
354 * VLAN devices are not used. When broken device drivers are no longer in
355 * widespread use, we will delete these interfaces. */
356 ofp_port_t realdev_ofp_port;
360 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
362 * This is deprecated. It is only for compatibility with broken device drivers
363 * in old versions of Linux that do not properly support VLANs when VLAN
364 * devices are not used. When broken device drivers are no longer in
365 * widespread use, we will delete these interfaces. */
366 struct vlan_splinter {
367 struct hmap_node realdev_vid_node;
368 struct hmap_node vlandev_node;
369 ofp_port_t realdev_ofp_port;
370 ofp_port_t vlandev_ofp_port;
374 static void vsp_remove(struct ofport_dpif *);
375 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
377 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
380 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
383 static struct ofport_dpif *
384 ofport_dpif_cast(const struct ofport *ofport)
386 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
389 static void port_run(struct ofport_dpif *);
390 static int set_bfd(struct ofport *, const struct smap *);
391 static int set_cfm(struct ofport *, const struct cfm_settings *);
392 static void ofport_update_peer(struct ofport_dpif *);
393 static void run_fast_rl(void);
394 static int run_fast(struct ofproto *);
396 struct dpif_completion {
397 struct list list_node;
398 struct ofoperation *op;
401 /* Reasons that we might need to revalidate every facet, and corresponding
404 * A value of 0 means that there is no need to revalidate.
406 * It would be nice to have some cleaner way to integrate with coverage
407 * counters, but with only a few reasons I guess this is good enough for
409 enum revalidate_reason {
410 REV_RECONFIGURE = 1, /* Switch configuration changed. */
411 REV_STP, /* Spanning tree protocol port status change. */
412 REV_BOND, /* Bonding changed. */
413 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
414 REV_FLOW_TABLE, /* Flow table changed. */
415 REV_MAC_LEARNING, /* Mac learning changed. */
416 REV_INCONSISTENCY /* Facet self-check failed. */
418 COVERAGE_DEFINE(rev_reconfigure);
419 COVERAGE_DEFINE(rev_stp);
420 COVERAGE_DEFINE(rev_bond);
421 COVERAGE_DEFINE(rev_port_toggled);
422 COVERAGE_DEFINE(rev_flow_table);
423 COVERAGE_DEFINE(rev_mac_learning);
424 COVERAGE_DEFINE(rev_inconsistency);
426 /* All datapaths of a given type share a single dpif backer instance. */
432 struct timer next_expiration;
434 struct ovs_rwlock odp_to_ofport_lock;
435 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
437 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
439 /* Facet revalidation flags applying to facets which use this backer. */
440 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
442 struct hmap drop_keys; /* Set of dropped odp keys. */
443 bool recv_set_enable; /* Enables or disables receiving packets. */
445 struct hmap subfacets;
446 struct governor *governor;
448 /* Subfacet statistics.
450 * These keep track of the total number of subfacets added and deleted and
451 * flow life span. They are useful for computing the flow rates stats
452 * exposed via "ovs-appctl dpif/show". The goal is to learn about
453 * traffic patterns in ways that we can use later to improve Open vSwitch
454 * performance in new situations. */
455 unsigned max_n_subfacet; /* Maximum number of flows */
456 unsigned avg_n_subfacet; /* Average number of flows. */
457 long long int avg_subfacet_life; /* Average life span of subfacets. */
459 /* Number of upcall handling threads. */
460 unsigned int n_handler_threads;
463 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
464 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
466 static void drop_key_clear(struct dpif_backer *);
468 struct ofproto_dpif {
469 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
471 struct dpif_backer *backer;
473 /* Special OpenFlow rules. */
474 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
475 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
476 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
479 struct netflow *netflow;
480 struct dpif_sflow *sflow;
481 struct dpif_ipfix *ipfix;
482 struct hmap bundles; /* Contains "struct ofbundle"s. */
483 struct mac_learning *ml;
484 bool has_bonded_bundles;
486 struct mbridge *mbridge;
489 struct classifier facets; /* Contains 'struct facet's. */
490 long long int consistency_rl;
492 struct ovs_mutex stats_mutex;
493 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
494 * consumed in userspace. */
498 long long int stp_last_tick;
500 /* VLAN splinters. */
501 struct ovs_mutex vsp_mutex;
502 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
503 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
506 struct sset ports; /* Set of standard port names. */
507 struct sset ghost_ports; /* Ports with no datapath port. */
508 struct sset port_poll_set; /* Queued names for port_poll() reply. */
509 int port_poll_errno; /* Last errno for port_poll() reply. */
510 uint64_t change_seq; /* Connectivity status changes. */
512 /* Per ofproto's dpif stats. */
517 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
520 /* By default, flows in the datapath are wildcarded (megaflows). They
521 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
522 static bool enable_megaflows = true;
524 /* All existing ofproto_dpif instances, indexed by ->up.name. */
525 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
527 static void ofproto_dpif_unixctl_init(void);
529 static inline struct ofproto_dpif *
530 ofproto_dpif_cast(const struct ofproto *ofproto)
532 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
533 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
536 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
537 ofp_port_t ofp_port);
538 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
539 const struct ofpbuf *packet,
540 const struct ofpact[], size_t ofpacts_len,
544 static void handle_upcalls(struct dpif_backer *);
546 /* Flow expiration. */
547 static int expire(struct dpif_backer *);
549 /* Global variables. */
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
552 /* Initial mappings of port to bridge mappings. */
553 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
555 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
558 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
559 struct ofputil_flow_mod *fm)
561 ofproto_flow_mod(&ofproto->up, fm);
564 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
565 * Takes ownership of 'pin' and pin->packet. */
567 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
568 struct ofproto_packet_in *pin)
570 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
571 COVERAGE_INC(packet_in_overflow);
572 free(CONST_CAST(void *, pin->up.packet));
577 /* Factory functions. */
580 init(const struct shash *iface_hints)
582 struct shash_node *node;
584 /* Make a local copy, since we don't own 'iface_hints' elements. */
585 SHASH_FOR_EACH(node, iface_hints) {
586 const struct iface_hint *orig_hint = node->data;
587 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
589 new_hint->br_name = xstrdup(orig_hint->br_name);
590 new_hint->br_type = xstrdup(orig_hint->br_type);
591 new_hint->ofp_port = orig_hint->ofp_port;
593 shash_add(&init_ofp_ports, node->name, new_hint);
598 enumerate_types(struct sset *types)
600 dp_enumerate_types(types);
604 enumerate_names(const char *type, struct sset *names)
606 struct ofproto_dpif *ofproto;
609 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
610 if (strcmp(type, ofproto->up.type)) {
613 sset_add(names, ofproto->up.name);
620 del(const char *type, const char *name)
625 error = dpif_open(name, type, &dpif);
627 error = dpif_delete(dpif);
634 port_open_type(const char *datapath_type, const char *port_type)
636 return dpif_port_open_type(datapath_type, port_type);
639 /* Type functions. */
641 static void process_dpif_port_changes(struct dpif_backer *);
642 static void process_dpif_all_ports_changed(struct dpif_backer *);
643 static void process_dpif_port_change(struct dpif_backer *,
644 const char *devname);
645 static void process_dpif_port_error(struct dpif_backer *, int error);
647 static struct ofproto_dpif *
648 lookup_ofproto_dpif_by_port_name(const char *name)
650 struct ofproto_dpif *ofproto;
652 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
653 if (sset_contains(&ofproto->ports, name)) {
662 type_run(const char *type)
664 static long long int push_timer = LLONG_MIN;
665 struct dpif_backer *backer;
667 backer = shash_find_data(&all_dpif_backers, type);
669 /* This is not necessarily a problem, since backers are only
670 * created on demand. */
674 dpif_run(backer->dpif);
676 /* The most natural place to push facet statistics is when they're pulled
677 * from the datapath. However, when there are many flows in the datapath,
678 * this expensive operation can occur so frequently, that it reduces our
679 * ability to quickly set up flows. To reduce the cost, we push statistics
681 if (time_msec() > push_timer) {
682 push_timer = time_msec() + 2000;
686 /* If vswitchd started with other_config:flow_restore_wait set as "true",
687 * and the configuration has now changed to "false", enable receiving
688 * packets from the datapath. */
689 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
692 backer->recv_set_enable = true;
694 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
696 udpif_recv_set(backer->udpif, 0, false);
697 VLOG_ERR("Failed to enable receiving packets in dpif.");
700 udpif_recv_set(backer->udpif, n_handler_threads,
701 backer->recv_set_enable);
702 dpif_flow_flush(backer->dpif);
703 backer->need_revalidate = REV_RECONFIGURE;
706 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
707 * to reset the handler threads. */
708 if (backer->n_handler_threads != n_handler_threads) {
709 udpif_recv_set(backer->udpif, n_handler_threads,
710 backer->recv_set_enable);
711 backer->n_handler_threads = n_handler_threads;
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->netflow, ofproto->up.frag_handling,
804 ofproto->up.forward_bpdu,
805 connmgr_has_in_band(ofproto->up.connmgr));
807 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
808 xlate_bundle_set(ofproto, bundle, bundle->name,
809 bundle->vlan_mode, bundle->vlan,
810 bundle->trunks, bundle->use_priority_tags,
811 bundle->bond, bundle->lacp,
815 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
816 int stp_port = ofport->stp_port
817 ? stp_port_no(ofport->stp_port)
819 xlate_ofport_set(ofproto, ofport->bundle, ofport,
820 ofport->up.ofp_port, ofport->odp_port,
821 ofport->up.netdev, ofport->cfm,
822 ofport->bfd, ofport->peer, stp_port,
823 ofport->qdscp, ofport->n_qdscp,
824 ofport->up.pp.config, ofport->up.pp.state,
825 ofport->is_tunnel, ofport->may_enable);
827 ovs_rwlock_unlock(&xlate_rwlock);
829 /* Only ofproto-dpif cares about the facet classifier so we just
830 * lock cls_cursor_init() to appease the thread safety analysis. */
831 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
832 cls_cursor_init(&cursor, &ofproto->facets, NULL);
833 ovs_rwlock_unlock(&ofproto->facets.rwlock);
834 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
835 facet_revalidate(facet);
840 udpif_revalidate(backer->udpif);
843 if (!backer->recv_set_enable) {
844 /* Wake up before a max of 1000ms. */
845 timer_set_duration(&backer->next_expiration, 1000);
846 } else if (timer_expired(&backer->next_expiration)) {
847 int delay = expire(backer);
848 timer_set_duration(&backer->next_expiration, delay);
851 process_dpif_port_changes(backer);
853 if (backer->governor) {
856 governor_run(backer->governor);
858 /* If the governor has shrunk to its minimum size and the number of
859 * subfacets has dwindled, then drop the governor entirely.
861 * For hysteresis, the number of subfacets to drop the governor is
862 * smaller than the number needed to trigger its creation. */
863 n_subfacets = hmap_count(&backer->subfacets);
864 if (n_subfacets * 4 < flow_eviction_threshold
865 && governor_is_idle(backer->governor)) {
866 governor_destroy(backer->governor);
867 backer->governor = NULL;
874 /* Check for and handle port changes in 'backer''s dpif. */
876 process_dpif_port_changes(struct dpif_backer *backer)
882 error = dpif_port_poll(backer->dpif, &devname);
888 process_dpif_all_ports_changed(backer);
892 process_dpif_port_change(backer, devname);
897 process_dpif_port_error(backer, error);
904 process_dpif_all_ports_changed(struct dpif_backer *backer)
906 struct ofproto_dpif *ofproto;
907 struct dpif_port dpif_port;
908 struct dpif_port_dump dump;
909 struct sset devnames;
912 sset_init(&devnames);
913 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
914 if (ofproto->backer == backer) {
915 struct ofport *ofport;
917 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
918 sset_add(&devnames, netdev_get_name(ofport->netdev));
922 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
923 sset_add(&devnames, dpif_port.name);
926 SSET_FOR_EACH (devname, &devnames) {
927 process_dpif_port_change(backer, devname);
929 sset_destroy(&devnames);
933 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
935 struct ofproto_dpif *ofproto;
936 struct dpif_port port;
938 /* Don't report on the datapath's device. */
939 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
943 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
944 &all_ofproto_dpifs) {
945 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
950 ofproto = lookup_ofproto_dpif_by_port_name(devname);
951 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
952 /* The port was removed. If we know the datapath,
953 * report it through poll_set(). If we don't, it may be
954 * notifying us of a removal we initiated, so ignore it.
955 * If there's a pending ENOBUFS, let it stand, since
956 * everything will be reevaluated. */
957 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
958 sset_add(&ofproto->port_poll_set, devname);
959 ofproto->port_poll_errno = 0;
961 } else if (!ofproto) {
962 /* The port was added, but we don't know with which
963 * ofproto we should associate it. Delete it. */
964 dpif_port_del(backer->dpif, port.port_no);
966 struct ofport_dpif *ofport;
968 ofport = ofport_dpif_cast(shash_find_data(
969 &ofproto->up.port_by_name, devname));
971 && ofport->odp_port != port.port_no
972 && !odp_port_to_ofport(backer, port.port_no))
974 /* 'ofport''s datapath port number has changed from
975 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
976 * structures to match. */
977 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
978 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
979 ofport->odp_port = port.port_no;
980 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
981 hash_odp_port(port.port_no));
982 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
983 backer->need_revalidate = REV_RECONFIGURE;
986 dpif_port_destroy(&port);
989 /* Propagate 'error' to all ofprotos based on 'backer'. */
991 process_dpif_port_error(struct dpif_backer *backer, int error)
993 struct ofproto_dpif *ofproto;
995 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
996 if (ofproto->backer == backer) {
997 sset_clear(&ofproto->port_poll_set);
998 ofproto->port_poll_errno = error;
1004 dpif_backer_run_fast(struct dpif_backer *backer)
1006 handle_upcalls(backer);
1012 type_run_fast(const char *type)
1014 struct dpif_backer *backer;
1016 backer = shash_find_data(&all_dpif_backers, type);
1018 /* This is not necessarily a problem, since backers are only
1019 * created on demand. */
1023 return dpif_backer_run_fast(backer);
1029 static long long int port_rl = LLONG_MIN;
1031 if (time_msec() >= port_rl) {
1032 struct ofproto_dpif *ofproto;
1034 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1035 run_fast(&ofproto->up);
1037 port_rl = time_msec() + 200;
1042 type_wait(const char *type)
1044 struct dpif_backer *backer;
1046 backer = shash_find_data(&all_dpif_backers, type);
1048 /* This is not necessarily a problem, since backers are only
1049 * created on demand. */
1053 if (backer->governor) {
1054 governor_wait(backer->governor);
1057 timer_wait(&backer->next_expiration);
1058 dpif_wait(backer->dpif);
1059 udpif_wait(backer->udpif);
1062 /* Basic life-cycle. */
1064 static int add_internal_flows(struct ofproto_dpif *);
1066 static struct ofproto *
1069 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1070 return &ofproto->up;
1074 dealloc(struct ofproto *ofproto_)
1076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1081 close_dpif_backer(struct dpif_backer *backer)
1083 ovs_assert(backer->refcount > 0);
1085 if (--backer->refcount) {
1089 drop_key_clear(backer);
1090 hmap_destroy(&backer->drop_keys);
1092 udpif_destroy(backer->udpif);
1094 simap_destroy(&backer->tnl_backers);
1095 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1096 hmap_destroy(&backer->odp_to_ofport_map);
1097 shash_find_and_delete(&all_dpif_backers, backer->type);
1099 dpif_close(backer->dpif);
1101 ovs_assert(hmap_is_empty(&backer->subfacets));
1102 hmap_destroy(&backer->subfacets);
1103 governor_destroy(backer->governor);
1108 /* Datapath port slated for removal from datapath. */
1109 struct odp_garbage {
1110 struct list list_node;
1111 odp_port_t odp_port;
1115 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1117 struct dpif_backer *backer;
1118 struct dpif_port_dump port_dump;
1119 struct dpif_port port;
1120 struct shash_node *node;
1121 struct list garbage_list;
1122 struct odp_garbage *garbage, *next;
1128 backer = shash_find_data(&all_dpif_backers, type);
1135 backer_name = xasprintf("ovs-%s", type);
1137 /* Remove any existing datapaths, since we assume we're the only
1138 * userspace controlling the datapath. */
1140 dp_enumerate_names(type, &names);
1141 SSET_FOR_EACH(name, &names) {
1142 struct dpif *old_dpif;
1144 /* Don't remove our backer if it exists. */
1145 if (!strcmp(name, backer_name)) {
1149 if (dpif_open(name, type, &old_dpif)) {
1150 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1152 dpif_delete(old_dpif);
1153 dpif_close(old_dpif);
1156 sset_destroy(&names);
1158 backer = xmalloc(sizeof *backer);
1160 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1163 VLOG_ERR("failed to open datapath of type %s: %s", type,
1164 ovs_strerror(error));
1168 backer->udpif = udpif_create(backer, backer->dpif);
1170 backer->type = xstrdup(type);
1171 backer->governor = NULL;
1172 backer->refcount = 1;
1173 hmap_init(&backer->odp_to_ofport_map);
1174 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1175 hmap_init(&backer->drop_keys);
1176 hmap_init(&backer->subfacets);
1177 timer_set_duration(&backer->next_expiration, 1000);
1178 backer->need_revalidate = 0;
1179 simap_init(&backer->tnl_backers);
1180 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1183 if (backer->recv_set_enable) {
1184 dpif_flow_flush(backer->dpif);
1187 /* Loop through the ports already on the datapath and remove any
1188 * that we don't need anymore. */
1189 list_init(&garbage_list);
1190 dpif_port_dump_start(&port_dump, backer->dpif);
1191 while (dpif_port_dump_next(&port_dump, &port)) {
1192 node = shash_find(&init_ofp_ports, port.name);
1193 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1194 garbage = xmalloc(sizeof *garbage);
1195 garbage->odp_port = port.port_no;
1196 list_push_front(&garbage_list, &garbage->list_node);
1199 dpif_port_dump_done(&port_dump);
1201 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1202 dpif_port_del(backer->dpif, garbage->odp_port);
1203 list_remove(&garbage->list_node);
1207 shash_add(&all_dpif_backers, type, backer);
1209 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1211 VLOG_ERR("failed to listen on datapath of type %s: %s",
1212 type, ovs_strerror(error));
1213 close_dpif_backer(backer);
1216 udpif_recv_set(backer->udpif, n_handler_threads,
1217 backer->recv_set_enable);
1218 backer->n_handler_threads = n_handler_threads;
1220 backer->max_n_subfacet = 0;
1221 backer->avg_n_subfacet = 0;
1222 backer->avg_subfacet_life = 0;
1228 construct(struct ofproto *ofproto_)
1230 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1231 struct shash_node *node, *next;
1234 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1239 ofproto->netflow = NULL;
1240 ofproto->sflow = NULL;
1241 ofproto->ipfix = NULL;
1242 ofproto->stp = NULL;
1243 hmap_init(&ofproto->bundles);
1244 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1245 ofproto->mbridge = mbridge_create();
1246 ofproto->has_bonded_bundles = false;
1247 ofproto->lacp_enabled = false;
1248 ovs_mutex_init(&ofproto->stats_mutex);
1249 ovs_mutex_init(&ofproto->vsp_mutex);
1251 classifier_init(&ofproto->facets, NULL);
1252 ofproto->consistency_rl = LLONG_MIN;
1254 guarded_list_init(&ofproto->pins);
1256 ofproto_dpif_unixctl_init();
1258 hmap_init(&ofproto->vlandev_map);
1259 hmap_init(&ofproto->realdev_vid_map);
1261 sset_init(&ofproto->ports);
1262 sset_init(&ofproto->ghost_ports);
1263 sset_init(&ofproto->port_poll_set);
1264 ofproto->port_poll_errno = 0;
1265 ofproto->change_seq = 0;
1267 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1268 struct iface_hint *iface_hint = node->data;
1270 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1271 /* Check if the datapath already has this port. */
1272 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1273 sset_add(&ofproto->ports, node->name);
1276 free(iface_hint->br_name);
1277 free(iface_hint->br_type);
1279 shash_delete(&init_ofp_ports, node);
1283 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1284 hash_string(ofproto->up.name, 0));
1285 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1287 ofproto_init_tables(ofproto_, N_TABLES);
1288 error = add_internal_flows(ofproto);
1289 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1292 ofproto->n_missed = 0;
1298 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1299 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1301 struct ofputil_flow_mod fm;
1304 match_init_catchall(&fm.match);
1306 match_set_reg(&fm.match, 0, id);
1307 fm.new_cookie = htonll(0);
1308 fm.cookie = htonll(0);
1309 fm.cookie_mask = htonll(0);
1310 fm.modify_cookie = false;
1311 fm.table_id = TBL_INTERNAL;
1312 fm.command = OFPFC_ADD;
1313 fm.idle_timeout = 0;
1314 fm.hard_timeout = 0;
1318 fm.ofpacts = ofpacts->data;
1319 fm.ofpacts_len = ofpacts->size;
1321 error = ofproto_flow_mod(&ofproto->up, &fm);
1323 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1324 id, ofperr_to_string(error));
1328 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1330 rule_dpif_unref(*rulep);
1339 add_internal_flows(struct ofproto_dpif *ofproto)
1341 struct ofpact_controller *controller;
1342 uint64_t ofpacts_stub[128 / 8];
1343 struct ofpbuf ofpacts;
1347 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1350 controller = ofpact_put_CONTROLLER(&ofpacts);
1351 controller->max_len = UINT16_MAX;
1352 controller->controller_id = 0;
1353 controller->reason = OFPR_NO_MATCH;
1354 ofpact_pad(&ofpacts);
1356 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1361 ofpbuf_clear(&ofpacts);
1362 error = add_internal_flow(ofproto, id++, &ofpacts,
1363 &ofproto->no_packet_in_rule);
1368 error = add_internal_flow(ofproto, id++, &ofpacts,
1369 &ofproto->drop_frags_rule);
1374 destruct(struct ofproto *ofproto_)
1376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1377 struct rule_dpif *rule, *next_rule;
1378 struct ofproto_packet_in *pin, *next_pin;
1379 struct facet *facet, *next_facet;
1380 struct cls_cursor cursor;
1381 struct oftable *table;
1384 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1385 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1386 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1387 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1388 facet_remove(facet);
1391 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1392 ovs_rwlock_wrlock(&xlate_rwlock);
1393 xlate_remove_ofproto(ofproto);
1394 ovs_rwlock_unlock(&xlate_rwlock);
1396 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1397 * use-after-free error. */
1398 udpif_revalidate(ofproto->backer->udpif);
1400 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1402 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1403 struct cls_cursor cursor;
1405 ovs_rwlock_rdlock(&table->cls.rwlock);
1406 cls_cursor_init(&cursor, &table->cls, NULL);
1407 ovs_rwlock_unlock(&table->cls.rwlock);
1408 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1409 ofproto_rule_delete(&ofproto->up, &rule->up);
1413 guarded_list_pop_all(&ofproto->pins, &pins);
1414 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1415 list_remove(&pin->list_node);
1416 free(CONST_CAST(void *, pin->up.packet));
1419 guarded_list_destroy(&ofproto->pins);
1421 mbridge_unref(ofproto->mbridge);
1423 netflow_unref(ofproto->netflow);
1424 dpif_sflow_unref(ofproto->sflow);
1425 hmap_destroy(&ofproto->bundles);
1426 mac_learning_unref(ofproto->ml);
1428 classifier_destroy(&ofproto->facets);
1430 hmap_destroy(&ofproto->vlandev_map);
1431 hmap_destroy(&ofproto->realdev_vid_map);
1433 sset_destroy(&ofproto->ports);
1434 sset_destroy(&ofproto->ghost_ports);
1435 sset_destroy(&ofproto->port_poll_set);
1437 ovs_mutex_destroy(&ofproto->stats_mutex);
1438 ovs_mutex_destroy(&ofproto->vsp_mutex);
1440 close_dpif_backer(ofproto->backer);
1444 run_fast(struct ofproto *ofproto_)
1446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1447 struct ofproto_packet_in *pin, *next_pin;
1450 /* Do not perform any periodic activity required by 'ofproto' while
1451 * waiting for flow restore to complete. */
1452 if (ofproto_get_flow_restore_wait()) {
1456 guarded_list_pop_all(&ofproto->pins, &pins);
1457 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1458 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1459 list_remove(&pin->list_node);
1460 free(CONST_CAST(void *, pin->up.packet));
1468 run(struct ofproto *ofproto_)
1470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1474 if (mbridge_need_revalidate(ofproto->mbridge)) {
1475 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1476 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1477 mac_learning_flush(ofproto->ml);
1478 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1481 /* Do not perform any periodic activity below required by 'ofproto' while
1482 * waiting for flow restore to complete. */
1483 if (ofproto_get_flow_restore_wait()) {
1487 error = run_fast(ofproto_);
1492 if (ofproto->netflow) {
1493 netflow_run(ofproto->netflow);
1495 if (ofproto->sflow) {
1496 dpif_sflow_run(ofproto->sflow);
1498 if (ofproto->ipfix) {
1499 dpif_ipfix_run(ofproto->ipfix);
1502 new_seq = seq_read(connectivity_seq_get());
1503 if (ofproto->change_seq != new_seq) {
1504 struct ofport_dpif *ofport;
1506 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1510 ofproto->change_seq = new_seq;
1512 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1513 struct ofbundle *bundle;
1515 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1521 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1522 if (mac_learning_run(ofproto->ml)) {
1523 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1525 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1527 /* Check the consistency of a random facet, to aid debugging. */
1528 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1529 if (time_msec() >= ofproto->consistency_rl
1530 && !classifier_is_empty(&ofproto->facets)
1531 && !ofproto->backer->need_revalidate) {
1532 struct cls_subtable *table;
1533 struct cls_rule *cr;
1534 struct facet *facet;
1536 ofproto->consistency_rl = time_msec() + 250;
1538 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1539 struct cls_subtable, hmap_node);
1540 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1542 facet = CONTAINER_OF(cr, struct facet, cr);
1544 if (!facet_check_consistency(facet)) {
1545 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1548 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1554 wait(struct ofproto *ofproto_)
1556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1558 if (ofproto_get_flow_restore_wait()) {
1562 if (ofproto->sflow) {
1563 dpif_sflow_wait(ofproto->sflow);
1565 if (ofproto->ipfix) {
1566 dpif_ipfix_wait(ofproto->ipfix);
1568 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1569 struct ofbundle *bundle;
1571 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1572 bundle_wait(bundle);
1575 if (ofproto->netflow) {
1576 netflow_wait(ofproto->netflow);
1578 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1579 mac_learning_wait(ofproto->ml);
1580 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1582 if (ofproto->backer->need_revalidate) {
1583 /* Shouldn't happen, but if it does just go around again. */
1584 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1585 poll_immediate_wake();
1590 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1592 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 struct cls_cursor cursor;
1594 size_t n_subfacets = 0;
1595 struct facet *facet;
1597 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1598 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1599 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1601 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1602 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1603 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1604 n_subfacets += list_size(&facet->subfacets);
1606 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1607 simap_increase(usage, "subfacets", n_subfacets);
1611 flush(struct ofproto *ofproto_)
1613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1614 struct subfacet *subfacet, *next_subfacet;
1615 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1619 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1620 &ofproto->backer->subfacets) {
1621 if (subfacet->facet->ofproto != ofproto) {
1625 if (subfacet->path != SF_NOT_INSTALLED) {
1626 batch[n_batch++] = subfacet;
1627 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1628 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1632 subfacet_destroy(subfacet);
1637 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1642 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1643 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1645 *arp_match_ip = true;
1646 *actions = (OFPUTIL_A_OUTPUT |
1647 OFPUTIL_A_SET_VLAN_VID |
1648 OFPUTIL_A_SET_VLAN_PCP |
1649 OFPUTIL_A_STRIP_VLAN |
1650 OFPUTIL_A_SET_DL_SRC |
1651 OFPUTIL_A_SET_DL_DST |
1652 OFPUTIL_A_SET_NW_SRC |
1653 OFPUTIL_A_SET_NW_DST |
1654 OFPUTIL_A_SET_NW_TOS |
1655 OFPUTIL_A_SET_TP_SRC |
1656 OFPUTIL_A_SET_TP_DST |
1661 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1664 struct dpif_dp_stats s;
1665 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1668 strcpy(ots->name, "classifier");
1670 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1671 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1672 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1673 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1675 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1676 ots->lookup_count = htonll(n_lookup);
1677 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1680 static struct ofport *
1683 struct ofport_dpif *port = xmalloc(sizeof *port);
1688 port_dealloc(struct ofport *port_)
1690 struct ofport_dpif *port = ofport_dpif_cast(port_);
1695 port_construct(struct ofport *port_)
1697 struct ofport_dpif *port = ofport_dpif_cast(port_);
1698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1699 const struct netdev *netdev = port->up.netdev;
1700 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1701 struct dpif_port dpif_port;
1704 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1705 port->bundle = NULL;
1708 port->may_enable = true;
1709 port->stp_port = NULL;
1710 port->stp_state = STP_DISABLED;
1711 port->is_tunnel = false;
1715 port->realdev_ofp_port = 0;
1716 port->vlandev_vid = 0;
1717 port->carrier_seq = netdev_get_carrier_resets(netdev);
1718 port->is_layer3 = netdev_vport_is_layer3(netdev);
1720 if (netdev_vport_is_patch(netdev)) {
1721 /* By bailing out here, we don't submit the port to the sFlow module
1722 * to be considered for counter polling export. This is correct
1723 * because the patch port represents an interface that sFlow considers
1724 * to be "internal" to the switch as a whole, and therefore not an
1725 * candidate for counter polling. */
1726 port->odp_port = ODPP_NONE;
1727 ofport_update_peer(port);
1731 error = dpif_port_query_by_name(ofproto->backer->dpif,
1732 netdev_vport_get_dpif_port(netdev, namebuf,
1739 port->odp_port = dpif_port.port_no;
1741 if (netdev_get_tunnel_config(netdev)) {
1742 tnl_port_add(port, port->up.netdev, port->odp_port);
1743 port->is_tunnel = true;
1745 /* Sanity-check that a mapping doesn't already exist. This
1746 * shouldn't happen for non-tunnel ports. */
1747 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1748 VLOG_ERR("port %s already has an OpenFlow port number",
1750 dpif_port_destroy(&dpif_port);
1754 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1755 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1756 hash_odp_port(port->odp_port));
1757 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1759 dpif_port_destroy(&dpif_port);
1761 if (ofproto->sflow) {
1762 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1769 port_destruct(struct ofport *port_)
1771 struct ofport_dpif *port = ofport_dpif_cast(port_);
1772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1773 const char *devname = netdev_get_name(port->up.netdev);
1774 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1775 const char *dp_port_name;
1777 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1778 ovs_rwlock_wrlock(&xlate_rwlock);
1779 xlate_ofport_remove(port);
1780 ovs_rwlock_unlock(&xlate_rwlock);
1782 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1784 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1785 /* The underlying device is still there, so delete it. This
1786 * happens when the ofproto is being destroyed, since the caller
1787 * assumes that removal of attached ports will happen as part of
1789 if (!port->is_tunnel) {
1790 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1795 port->peer->peer = NULL;
1799 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1800 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1801 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1802 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1806 sset_find_and_delete(&ofproto->ports, devname);
1807 sset_find_and_delete(&ofproto->ghost_ports, devname);
1808 bundle_remove(port_);
1809 set_cfm(port_, NULL);
1810 set_bfd(port_, NULL);
1811 if (ofproto->sflow) {
1812 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1819 port_modified(struct ofport *port_)
1821 struct ofport_dpif *port = ofport_dpif_cast(port_);
1823 if (port->bundle && port->bundle->bond) {
1824 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1828 cfm_set_netdev(port->cfm, port->up.netdev);
1832 bfd_set_netdev(port->bfd, port->up.netdev);
1835 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1836 port->up.pp.hw_addr);
1838 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1840 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1844 ofport_update_peer(port);
1848 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1850 struct ofport_dpif *port = ofport_dpif_cast(port_);
1851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1852 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1854 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1855 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1856 OFPUTIL_PC_NO_PACKET_IN)) {
1857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1859 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1860 bundle_update(port->bundle);
1866 set_sflow(struct ofproto *ofproto_,
1867 const struct ofproto_sflow_options *sflow_options)
1869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1870 struct dpif_sflow *ds = ofproto->sflow;
1872 if (sflow_options) {
1874 struct ofport_dpif *ofport;
1876 ds = ofproto->sflow = dpif_sflow_create();
1877 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1878 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1880 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 dpif_sflow_set_options(ds, sflow_options);
1885 dpif_sflow_unref(ds);
1886 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1887 ofproto->sflow = NULL;
1895 struct ofproto *ofproto_,
1896 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1897 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1898 size_t n_flow_exporters_options)
1900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1901 struct dpif_ipfix *di = ofproto->ipfix;
1902 bool has_options = bridge_exporter_options || flow_exporters_options;
1904 if (has_options && !di) {
1905 di = ofproto->ipfix = dpif_ipfix_create();
1909 /* Call set_options in any case to cleanly flush the flow
1910 * caches in the last exporters that are to be destroyed. */
1911 dpif_ipfix_set_options(
1912 di, bridge_exporter_options, flow_exporters_options,
1913 n_flow_exporters_options);
1916 dpif_ipfix_unref(di);
1917 ofproto->ipfix = NULL;
1925 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1927 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1932 struct ofproto_dpif *ofproto;
1934 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1935 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1936 ofport->cfm = cfm_create(ofport->up.netdev);
1939 if (cfm_configure(ofport->cfm, s)) {
1946 cfm_unref(ofport->cfm);
1949 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1950 ofport->up.pp.hw_addr);
1955 get_cfm_status(const struct ofport *ofport_,
1956 struct ofproto_cfm_status *status)
1958 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961 status->faults = cfm_get_fault(ofport->cfm);
1962 status->flap_count = cfm_get_flap_count(ofport->cfm);
1963 status->remote_opstate = cfm_get_opup(ofport->cfm);
1964 status->health = cfm_get_health(ofport->cfm);
1965 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1973 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1976 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1980 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1981 cfg, ofport->up.netdev);
1982 if (ofport->bfd != old) {
1983 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1985 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1986 ofport->up.pp.hw_addr);
1991 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1993 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1996 bfd_get_status(ofport->bfd, smap);
2003 /* Spanning Tree. */
2006 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2008 struct ofproto_dpif *ofproto = ofproto_;
2009 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2010 struct ofport_dpif *ofport;
2012 ofport = stp_port_get_aux(sp);
2014 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2015 ofproto->up.name, port_num);
2017 struct eth_header *eth = pkt->l2;
2019 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2020 if (eth_addr_is_zero(eth->eth_src)) {
2021 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2022 "with unknown MAC", ofproto->up.name, port_num);
2024 ofproto_dpif_send_packet(ofport, pkt);
2030 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2032 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2036 /* Only revalidate flows if the configuration changed. */
2037 if (!s != !ofproto->stp) {
2038 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2042 if (!ofproto->stp) {
2043 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2044 send_bpdu_cb, ofproto);
2045 ofproto->stp_last_tick = time_msec();
2048 stp_set_bridge_id(ofproto->stp, s->system_id);
2049 stp_set_bridge_priority(ofproto->stp, s->priority);
2050 stp_set_hello_time(ofproto->stp, s->hello_time);
2051 stp_set_max_age(ofproto->stp, s->max_age);
2052 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2054 struct ofport *ofport;
2056 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2057 set_stp_port(ofport, NULL);
2060 stp_unref(ofproto->stp);
2061 ofproto->stp = NULL;
2068 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2074 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2075 s->designated_root = stp_get_designated_root(ofproto->stp);
2076 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2085 update_stp_port_state(struct ofport_dpif *ofport)
2087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2088 enum stp_state state;
2090 /* Figure out new state. */
2091 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2095 if (ofport->stp_state != state) {
2096 enum ofputil_port_state of_state;
2099 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2100 netdev_get_name(ofport->up.netdev),
2101 stp_state_name(ofport->stp_state),
2102 stp_state_name(state));
2103 if (stp_learn_in_state(ofport->stp_state)
2104 != stp_learn_in_state(state)) {
2105 /* xxx Learning action flows should also be flushed. */
2106 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2107 mac_learning_flush(ofproto->ml);
2108 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2110 fwd_change = stp_forward_in_state(ofport->stp_state)
2111 != stp_forward_in_state(state);
2113 ofproto->backer->need_revalidate = REV_STP;
2114 ofport->stp_state = state;
2115 ofport->stp_state_entered = time_msec();
2117 if (fwd_change && ofport->bundle) {
2118 bundle_update(ofport->bundle);
2121 /* Update the STP state bits in the OpenFlow port description. */
2122 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2123 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2124 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2125 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2126 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2128 ofproto_port_set_state(&ofport->up, of_state);
2132 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2133 * caller is responsible for assigning STP port numbers and ensuring
2134 * there are no duplicates. */
2136 set_stp_port(struct ofport *ofport_,
2137 const struct ofproto_port_stp_settings *s)
2139 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2141 struct stp_port *sp = ofport->stp_port;
2143 if (!s || !s->enable) {
2145 ofport->stp_port = NULL;
2146 stp_port_disable(sp);
2147 update_stp_port_state(ofport);
2150 } else if (sp && stp_port_no(sp) != s->port_num
2151 && ofport == stp_port_get_aux(sp)) {
2152 /* The port-id changed, so disable the old one if it's not
2153 * already in use by another port. */
2154 stp_port_disable(sp);
2157 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2158 stp_port_enable(sp);
2160 stp_port_set_aux(sp, ofport);
2161 stp_port_set_priority(sp, s->priority);
2162 stp_port_set_path_cost(sp, s->path_cost);
2164 update_stp_port_state(ofport);
2170 get_stp_port_status(struct ofport *ofport_,
2171 struct ofproto_port_stp_status *s)
2173 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2175 struct stp_port *sp = ofport->stp_port;
2177 if (!ofproto->stp || !sp) {
2183 s->port_id = stp_port_get_id(sp);
2184 s->state = stp_port_get_state(sp);
2185 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2186 s->role = stp_port_get_role(sp);
2192 get_stp_port_stats(struct ofport *ofport_,
2193 struct ofproto_port_stp_stats *s)
2195 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2197 struct stp_port *sp = ofport->stp_port;
2199 if (!ofproto->stp || !sp) {
2205 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2211 stp_run(struct ofproto_dpif *ofproto)
2214 long long int now = time_msec();
2215 long long int elapsed = now - ofproto->stp_last_tick;
2216 struct stp_port *sp;
2219 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2220 ofproto->stp_last_tick = now;
2222 while (stp_get_changed_port(ofproto->stp, &sp)) {
2223 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2226 update_stp_port_state(ofport);
2230 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2231 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2232 mac_learning_flush(ofproto->ml);
2233 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2239 stp_wait(struct ofproto_dpif *ofproto)
2242 poll_timer_wait(1000);
2247 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2250 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2251 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2253 if (ofport->n_qdscp != n_qdscp
2254 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2255 n_qdscp * sizeof *qdscp))) {
2256 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2257 free(ofport->qdscp);
2258 ofport->qdscp = n_qdscp
2259 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2261 ofport->n_qdscp = n_qdscp;
2269 /* Expires all MAC learning entries associated with 'bundle' and forces its
2270 * ofproto to revalidate every flow.
2272 * Normally MAC learning entries are removed only from the ofproto associated
2273 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2274 * are removed from every ofproto. When patch ports and SLB bonds are in use
2275 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2276 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2277 * with the host from which it migrated. */
2279 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2281 struct ofproto_dpif *ofproto = bundle->ofproto;
2282 struct mac_learning *ml = ofproto->ml;
2283 struct mac_entry *mac, *next_mac;
2285 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2286 ovs_rwlock_wrlock(&ml->rwlock);
2287 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2288 if (mac->port.p == bundle) {
2290 struct ofproto_dpif *o;
2292 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2294 struct mac_entry *e;
2296 ovs_rwlock_wrlock(&o->ml->rwlock);
2297 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2299 mac_learning_expire(o->ml, e);
2301 ovs_rwlock_unlock(&o->ml->rwlock);
2306 mac_learning_expire(ml, mac);
2309 ovs_rwlock_unlock(&ml->rwlock);
2312 static struct ofbundle *
2313 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2315 struct ofbundle *bundle;
2317 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2318 &ofproto->bundles) {
2319 if (bundle->aux == aux) {
2327 bundle_update(struct ofbundle *bundle)
2329 struct ofport_dpif *port;
2331 bundle->floodable = true;
2332 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2333 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2335 || !stp_forward_in_state(port->stp_state)) {
2336 bundle->floodable = false;
2343 bundle_del_port(struct ofport_dpif *port)
2345 struct ofbundle *bundle = port->bundle;
2347 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2349 list_remove(&port->bundle_node);
2350 port->bundle = NULL;
2353 lacp_slave_unregister(bundle->lacp, port);
2356 bond_slave_unregister(bundle->bond, port);
2359 bundle_update(bundle);
2363 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2364 struct lacp_slave_settings *lacp)
2366 struct ofport_dpif *port;
2368 port = get_ofp_port(bundle->ofproto, ofp_port);
2373 if (port->bundle != bundle) {
2374 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2376 bundle_remove(&port->up);
2379 port->bundle = bundle;
2380 list_push_back(&bundle->ports, &port->bundle_node);
2381 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2383 || !stp_forward_in_state(port->stp_state)) {
2384 bundle->floodable = false;
2388 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2389 lacp_slave_register(bundle->lacp, port, lacp);
2396 bundle_destroy(struct ofbundle *bundle)
2398 struct ofproto_dpif *ofproto;
2399 struct ofport_dpif *port, *next_port;
2405 ofproto = bundle->ofproto;
2406 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2408 ovs_rwlock_wrlock(&xlate_rwlock);
2409 xlate_bundle_remove(bundle);
2410 ovs_rwlock_unlock(&xlate_rwlock);
2412 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2413 bundle_del_port(port);
2416 bundle_flush_macs(bundle, true);
2417 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2419 free(bundle->trunks);
2420 lacp_unref(bundle->lacp);
2421 bond_unref(bundle->bond);
2426 bundle_set(struct ofproto *ofproto_, void *aux,
2427 const struct ofproto_bundle_settings *s)
2429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2430 bool need_flush = false;
2431 struct ofport_dpif *port;
2432 struct ofbundle *bundle;
2433 unsigned long *trunks;
2439 bundle_destroy(bundle_lookup(ofproto, aux));
2443 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2444 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2446 bundle = bundle_lookup(ofproto, aux);
2448 bundle = xmalloc(sizeof *bundle);
2450 bundle->ofproto = ofproto;
2451 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2452 hash_pointer(aux, 0));
2454 bundle->name = NULL;
2456 list_init(&bundle->ports);
2457 bundle->vlan_mode = PORT_VLAN_TRUNK;
2459 bundle->trunks = NULL;
2460 bundle->use_priority_tags = s->use_priority_tags;
2461 bundle->lacp = NULL;
2462 bundle->bond = NULL;
2464 bundle->floodable = true;
2465 mbridge_register_bundle(ofproto->mbridge, bundle);
2468 if (!bundle->name || strcmp(s->name, bundle->name)) {
2470 bundle->name = xstrdup(s->name);
2475 ofproto->lacp_enabled = true;
2476 if (!bundle->lacp) {
2477 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2478 bundle->lacp = lacp_create();
2480 lacp_configure(bundle->lacp, s->lacp);
2482 lacp_unref(bundle->lacp);
2483 bundle->lacp = NULL;
2486 /* Update set of ports. */
2488 for (i = 0; i < s->n_slaves; i++) {
2489 if (!bundle_add_port(bundle, s->slaves[i],
2490 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2494 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2495 struct ofport_dpif *next_port;
2497 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2498 for (i = 0; i < s->n_slaves; i++) {
2499 if (s->slaves[i] == port->up.ofp_port) {
2504 bundle_del_port(port);
2508 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2510 if (list_is_empty(&bundle->ports)) {
2511 bundle_destroy(bundle);
2515 /* Set VLAN tagging mode */
2516 if (s->vlan_mode != bundle->vlan_mode
2517 || s->use_priority_tags != bundle->use_priority_tags) {
2518 bundle->vlan_mode = s->vlan_mode;
2519 bundle->use_priority_tags = s->use_priority_tags;
2524 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2525 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2527 if (vlan != bundle->vlan) {
2528 bundle->vlan = vlan;
2532 /* Get trunked VLANs. */
2533 switch (s->vlan_mode) {
2534 case PORT_VLAN_ACCESS:
2538 case PORT_VLAN_TRUNK:
2539 trunks = CONST_CAST(unsigned long *, s->trunks);
2542 case PORT_VLAN_NATIVE_UNTAGGED:
2543 case PORT_VLAN_NATIVE_TAGGED:
2544 if (vlan != 0 && (!s->trunks
2545 || !bitmap_is_set(s->trunks, vlan)
2546 || bitmap_is_set(s->trunks, 0))) {
2547 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2549 trunks = bitmap_clone(s->trunks, 4096);
2551 trunks = bitmap_allocate1(4096);
2553 bitmap_set1(trunks, vlan);
2554 bitmap_set0(trunks, 0);
2556 trunks = CONST_CAST(unsigned long *, s->trunks);
2563 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2564 free(bundle->trunks);
2565 if (trunks == s->trunks) {
2566 bundle->trunks = vlan_bitmap_clone(trunks);
2568 bundle->trunks = trunks;
2573 if (trunks != s->trunks) {
2578 if (!list_is_short(&bundle->ports)) {
2579 bundle->ofproto->has_bonded_bundles = true;
2581 if (bond_reconfigure(bundle->bond, s->bond)) {
2582 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2585 bundle->bond = bond_create(s->bond);
2586 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2589 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2590 bond_slave_register(bundle->bond, port, port->up.netdev);
2593 bond_unref(bundle->bond);
2594 bundle->bond = NULL;
2597 /* If we changed something that would affect MAC learning, un-learn
2598 * everything on this port and force flow revalidation. */
2600 bundle_flush_macs(bundle, false);
2607 bundle_remove(struct ofport *port_)
2609 struct ofport_dpif *port = ofport_dpif_cast(port_);
2610 struct ofbundle *bundle = port->bundle;
2613 bundle_del_port(port);
2614 if (list_is_empty(&bundle->ports)) {
2615 bundle_destroy(bundle);
2616 } else if (list_is_short(&bundle->ports)) {
2617 bond_unref(bundle->bond);
2618 bundle->bond = NULL;
2624 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2626 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2627 struct ofport_dpif *port = port_;
2628 uint8_t ea[ETH_ADDR_LEN];
2631 error = netdev_get_etheraddr(port->up.netdev, ea);
2633 struct ofpbuf packet;
2636 ofpbuf_init(&packet, 0);
2637 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2639 memcpy(packet_pdu, pdu, pdu_size);
2641 ofproto_dpif_send_packet(port, &packet);
2642 ofpbuf_uninit(&packet);
2644 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2645 "%s (%s)", port->bundle->name,
2646 netdev_get_name(port->up.netdev), ovs_strerror(error));
2651 bundle_send_learning_packets(struct ofbundle *bundle)
2653 struct ofproto_dpif *ofproto = bundle->ofproto;
2654 struct ofpbuf *learning_packet;
2655 int error, n_packets, n_errors;
2656 struct mac_entry *e;
2657 struct list packets;
2659 list_init(&packets);
2660 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2661 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2662 if (e->port.p != bundle) {
2665 learning_packet = bond_compose_learning_packet(bundle->bond,
2668 learning_packet->private_p = port_void;
2669 list_push_back(&packets, &learning_packet->list_node);
2672 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2674 error = n_packets = n_errors = 0;
2675 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2678 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2685 ofpbuf_list_delete(&packets);
2688 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2689 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2690 "packets, last error was: %s",
2691 bundle->name, n_errors, n_packets, ovs_strerror(error));
2693 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2694 bundle->name, n_packets);
2699 bundle_run(struct ofbundle *bundle)
2702 lacp_run(bundle->lacp, send_pdu_cb);
2705 struct ofport_dpif *port;
2707 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2708 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2711 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2712 bundle->ofproto->backer->need_revalidate = REV_BOND;
2715 if (bond_should_send_learning_packets(bundle->bond)) {
2716 bundle_send_learning_packets(bundle);
2722 bundle_wait(struct ofbundle *bundle)
2725 lacp_wait(bundle->lacp);
2728 bond_wait(bundle->bond);
2735 mirror_set__(struct ofproto *ofproto_, void *aux,
2736 const struct ofproto_mirror_settings *s)
2738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2739 struct ofbundle **srcs, **dsts;
2744 mirror_destroy(ofproto->mbridge, aux);
2748 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2749 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2751 for (i = 0; i < s->n_srcs; i++) {
2752 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2755 for (i = 0; i < s->n_dsts; i++) {
2756 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2759 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2760 s->n_dsts, s->src_vlans,
2761 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2768 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2769 uint64_t *packets, uint64_t *bytes)
2772 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2777 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2781 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2782 mac_learning_flush(ofproto->ml);
2784 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2789 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2792 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2793 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2797 forward_bpdu_changed(struct ofproto *ofproto_)
2799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2800 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2804 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2808 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2809 mac_learning_set_idle_time(ofproto->ml, idle_time);
2810 mac_learning_set_max_entries(ofproto->ml, max_entries);
2811 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2816 static struct ofport_dpif *
2817 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2819 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2820 return ofport ? ofport_dpif_cast(ofport) : NULL;
2823 static struct ofport_dpif *
2824 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2826 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2827 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2831 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2832 struct ofproto_port *ofproto_port,
2833 struct dpif_port *dpif_port)
2835 ofproto_port->name = dpif_port->name;
2836 ofproto_port->type = dpif_port->type;
2837 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2841 ofport_update_peer(struct ofport_dpif *ofport)
2843 const struct ofproto_dpif *ofproto;
2844 struct dpif_backer *backer;
2847 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2851 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2852 backer->need_revalidate = REV_RECONFIGURE;
2855 ofport->peer->peer = NULL;
2856 ofport->peer = NULL;
2859 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2864 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2865 struct ofport *peer_ofport;
2866 struct ofport_dpif *peer;
2869 if (ofproto->backer != backer) {
2873 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2878 peer = ofport_dpif_cast(peer_ofport);
2879 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2880 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2882 ofport->peer = peer;
2883 ofport->peer->peer = ofport;
2893 port_run(struct ofport_dpif *ofport)
2895 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2896 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2897 bool enable = netdev_get_carrier(ofport->up.netdev);
2898 bool cfm_enable = false;
2899 bool bfd_enable = false;
2901 ofport->carrier_seq = carrier_seq;
2904 int cfm_opup = cfm_get_opup(ofport->cfm);
2906 cfm_enable = !cfm_get_fault(ofport->cfm);
2908 if (cfm_opup >= 0) {
2909 cfm_enable = cfm_enable && cfm_opup;
2914 bfd_enable = bfd_forwarding(ofport->bfd);
2917 if (ofport->bfd || ofport->cfm) {
2918 enable = enable && (cfm_enable || bfd_enable);
2921 if (ofport->bundle) {
2922 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2923 if (carrier_changed) {
2924 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2928 if (ofport->may_enable != enable) {
2929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2930 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2933 ofport->may_enable = enable;
2937 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2938 struct ofproto_port *ofproto_port)
2940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2941 struct dpif_port dpif_port;
2944 if (sset_contains(&ofproto->ghost_ports, devname)) {
2945 const char *type = netdev_get_type_from_name(devname);
2947 /* We may be called before ofproto->up.port_by_name is populated with
2948 * the appropriate ofport. For this reason, we must get the name and
2949 * type from the netdev layer directly. */
2951 const struct ofport *ofport;
2953 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2954 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2955 ofproto_port->name = xstrdup(devname);
2956 ofproto_port->type = xstrdup(type);
2962 if (!sset_contains(&ofproto->ports, devname)) {
2965 error = dpif_port_query_by_name(ofproto->backer->dpif,
2966 devname, &dpif_port);
2968 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2974 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2977 const char *devname = netdev_get_name(netdev);
2978 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2979 const char *dp_port_name;
2981 if (netdev_vport_is_patch(netdev)) {
2982 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2986 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2987 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2988 odp_port_t port_no = ODPP_NONE;
2991 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2995 if (netdev_get_tunnel_config(netdev)) {
2996 simap_put(&ofproto->backer->tnl_backers,
2997 dp_port_name, odp_to_u32(port_no));
3001 if (netdev_get_tunnel_config(netdev)) {
3002 sset_add(&ofproto->ghost_ports, devname);
3004 sset_add(&ofproto->ports, devname);
3010 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3013 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3020 sset_find_and_delete(&ofproto->ghost_ports,
3021 netdev_get_name(ofport->up.netdev));
3022 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3023 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3024 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3026 /* The caller is going to close ofport->up.netdev. If this is a
3027 * bonded port, then the bond is using that netdev, so remove it
3028 * from the bond. The client will need to reconfigure everything
3029 * after deleting ports, so then the slave will get re-added. */
3030 bundle_remove(&ofport->up);
3037 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3039 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3044 error = netdev_get_stats(ofport->up.netdev, stats);
3046 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3049 ovs_mutex_lock(&ofproto->stats_mutex);
3050 /* ofproto->stats.tx_packets represents packets that we created
3051 * internally and sent to some port (e.g. packets sent with
3052 * ofproto_dpif_send_packet()). Account for them as if they had
3053 * come from OFPP_LOCAL and got forwarded. */
3055 if (stats->rx_packets != UINT64_MAX) {
3056 stats->rx_packets += ofproto->stats.tx_packets;
3059 if (stats->rx_bytes != UINT64_MAX) {
3060 stats->rx_bytes += ofproto->stats.tx_bytes;
3063 /* ofproto->stats.rx_packets represents packets that were received on
3064 * some port and we processed internally and dropped (e.g. STP).
3065 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3067 if (stats->tx_packets != UINT64_MAX) {
3068 stats->tx_packets += ofproto->stats.rx_packets;
3071 if (stats->tx_bytes != UINT64_MAX) {
3072 stats->tx_bytes += ofproto->stats.rx_bytes;
3074 ovs_mutex_unlock(&ofproto->stats_mutex);
3080 struct port_dump_state {
3085 struct ofproto_port port;
3090 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3092 *statep = xzalloc(sizeof(struct port_dump_state));
3097 port_dump_next(const struct ofproto *ofproto_, void *state_,
3098 struct ofproto_port *port)
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 struct port_dump_state *state = state_;
3102 const struct sset *sset;
3103 struct sset_node *node;
3105 if (state->has_port) {
3106 ofproto_port_destroy(&state->port);
3107 state->has_port = false;
3109 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3110 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3113 error = port_query_by_name(ofproto_, node->name, &state->port);
3115 *port = state->port;
3116 state->has_port = true;
3118 } else if (error != ENODEV) {
3123 if (!state->ghost) {
3124 state->ghost = true;
3127 return port_dump_next(ofproto_, state_, port);
3134 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3136 struct port_dump_state *state = state_;
3138 if (state->has_port) {
3139 ofproto_port_destroy(&state->port);
3146 port_poll(const struct ofproto *ofproto_, char **devnamep)
3148 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3150 if (ofproto->port_poll_errno) {
3151 int error = ofproto->port_poll_errno;
3152 ofproto->port_poll_errno = 0;
3156 if (sset_is_empty(&ofproto->port_poll_set)) {
3160 *devnamep = sset_pop(&ofproto->port_poll_set);
3165 port_poll_wait(const struct ofproto *ofproto_)
3167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3168 dpif_port_poll_wait(ofproto->backer->dpif);
3172 port_is_lacp_current(const struct ofport *ofport_)
3174 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3175 return (ofport->bundle && ofport->bundle->lacp
3176 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3180 /* Upcall handling. */
3182 struct flow_miss_op {
3183 struct dpif_op dpif_op;
3185 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3186 struct xlate_out xout;
3187 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3189 struct ofpbuf mask; /* Flow mask for "put" ops. */
3190 struct odputil_keybuf maskbuf;
3192 /* If this is a "put" op, then a pointer to the subfacet that should
3193 * be marked as uninstalled if the operation fails. */
3194 struct subfacet *subfacet;
3197 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3198 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3199 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3200 * return value of true). However, for short flows the cost of bookkeeping is
3201 * much higher than the benefits, so when the datapath holds a large number of
3202 * flows we impose some heuristics to decide which flows are likely to be worth
3205 flow_miss_should_make_facet(struct flow_miss *miss)
3207 struct dpif_backer *backer = miss->ofproto->backer;
3210 switch (flow_miss_model) {
3211 case OFPROTO_HANDLE_MISS_AUTO:
3213 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3215 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3219 if (!backer->governor) {
3222 n_subfacets = hmap_count(&backer->subfacets);
3223 if (n_subfacets * 2 <= flow_eviction_threshold) {
3227 backer->governor = governor_create();
3230 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3231 return governor_should_install_flow(backer->governor, hash,
3232 miss->stats.n_packets);
3235 /* Handles 'miss', which matches 'facet'. May add any required datapath
3236 * operations to 'ops', incrementing '*n_ops' for each new op.
3238 * All of the packets in 'miss' are considered to have arrived at time
3239 * 'miss->stats.used'. This is really important only for new facets: if we
3240 * just called time_msec() here, then the new subfacet or its packets could
3241 * look (occasionally) as though it was used some time after the facet was
3242 * used. That can make a one-packet flow look like it has a nonzero duration,
3243 * which looks odd in e.g. NetFlow statistics. */
3245 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3246 struct flow_miss_op *ops, size_t *n_ops)
3248 enum subfacet_path want_path;
3249 struct subfacet *subfacet;
3252 /* Update facet stats. */
3253 facet->packet_count += miss->stats.n_packets;
3254 facet->prev_packet_count += miss->stats.n_packets;
3255 facet->byte_count += miss->stats.n_bytes;
3256 facet->prev_byte_count += miss->stats.n_bytes;
3258 /* Look for an existing subfacet. If we find one, update its used time. */
3259 key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3260 if (!list_is_empty(&facet->subfacets)) {
3261 subfacet = subfacet_find(miss->ofproto->backer,
3262 miss->key, miss->key_len, key_hash);
3264 if (subfacet->facet == facet) {
3265 subfacet->used = MAX(subfacet->used, miss->stats.used);
3267 /* This shouldn't happen. */
3268 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3269 subfacet_destroy(subfacet);
3277 /* Don't install the flow if it's the result of the "userspace"
3278 * action for an already installed facet. This can occur when a
3279 * datapath flow with wildcards has a "userspace" action and flows
3280 * sent to userspace result in a different subfacet, which will then
3281 * be rejected as overlapping by the datapath. */
3282 if (miss->upcall_type == DPIF_UC_ACTION
3283 && !list_is_empty(&facet->subfacets)) {
3287 /* Create a subfacet, if we don't already have one. */
3289 subfacet = subfacet_create(facet, miss, key_hash);
3292 /* Install the subfacet, if it's not already installed. */
3293 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3294 if (subfacet->path != want_path) {
3295 struct flow_miss_op *op = &ops[(*n_ops)++];
3296 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3298 subfacet->path = want_path;
3300 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3301 if (enable_megaflows) {
3302 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3303 &miss->flow, UINT32_MAX);
3306 op->xout_garbage = false;
3307 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3308 op->subfacet = subfacet;
3309 put->flags = DPIF_FP_CREATE;
3310 put->key = miss->key;
3311 put->key_len = miss->key_len;
3312 put->mask = op->mask.data;
3313 put->mask_len = op->mask.size;
3315 if (want_path == SF_FAST_PATH) {
3316 put->actions = facet->xout.odp_actions.data;
3317 put->actions_len = facet->xout.odp_actions.size;
3319 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3320 op->slow_stub, sizeof op->slow_stub,
3321 &put->actions, &put->actions_len);
3327 /* Handles flow miss 'miss'. May add any required datapath operations
3328 * to 'ops', incrementing '*n_ops' for each new op. */
3330 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3333 struct facet *facet;
3335 miss->ofproto->n_missed += miss->stats.n_packets;
3337 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3339 /* There does not exist a bijection between 'struct flow' and datapath
3340 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3341 * assumption used throughout the facet and subfacet handling code.
3342 * Since we have to handle these misses in userspace anyway, we simply
3343 * skip facet creation, avoiding the problem altogether. */
3344 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3345 || !flow_miss_should_make_facet(miss)) {
3349 facet = facet_create(miss);
3351 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3354 static struct drop_key *
3355 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3358 struct drop_key *drop_key;
3360 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3361 &backer->drop_keys) {
3362 if (drop_key->key_len == key_len
3363 && !memcmp(drop_key->key, key, key_len)) {
3371 drop_key_clear(struct dpif_backer *backer)
3373 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3374 struct drop_key *drop_key, *next;
3376 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3379 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3381 if (error && !VLOG_DROP_WARN(&rl)) {
3382 struct ds ds = DS_EMPTY_INITIALIZER;
3383 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3384 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3385 ovs_strerror(error), ds_cstr(&ds));
3389 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3390 drop_key_destroy(drop_key);
3393 udpif_drop_key_clear(backer->udpif);
3397 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3399 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3400 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3401 struct flow_miss *miss;
3404 /* Process each element in the to-do list, constructing the set of
3405 * operations to batch. */
3407 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3408 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3410 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3412 /* Execute batch. */
3413 for (i = 0; i < n_ops; i++) {
3414 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3416 dpif_operate(backer->dpif, dpif_ops, n_ops);
3418 for (i = 0; i < n_ops; i++) {
3419 if (dpif_ops[i]->error != 0
3420 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3421 && flow_miss_ops[i].subfacet) {
3422 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3424 COVERAGE_INC(subfacet_install_fail);
3426 /* Zero-out subfacet counters when installation failed, but
3427 * datapath reported hits. This should not happen and
3428 * indicates a bug, since if the datapath flow exists, we
3429 * should not be attempting to create a new subfacet. A
3430 * buggy datapath could trigger this, so just zero out the
3431 * counters and log an error. */
3432 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3433 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3434 "datapath reported hits");
3435 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3438 subfacet->path = SF_NOT_INSTALLED;
3444 handle_upcalls(struct dpif_backer *backer)
3446 struct flow_miss_batch *fmb;
3449 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3450 struct drop_key *drop_key = drop_key_next(backer->udpif);
3455 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3456 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3457 hash_bytes(drop_key->key, drop_key->key_len, 0));
3458 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3459 drop_key->key, drop_key->key_len,
3460 NULL, 0, NULL, 0, NULL);
3462 drop_key_destroy(drop_key);
3466 fmb = flow_miss_batch_next(backer->udpif);
3468 handle_flow_misses(backer, fmb);
3469 flow_miss_batch_destroy(fmb);
3473 /* Flow expiration. */
3475 static int subfacet_max_idle(const struct dpif_backer *);
3476 static void update_stats(struct dpif_backer *);
3477 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3478 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3480 /* This function is called periodically by run(). Its job is to collect
3481 * updates for the flows that have been installed into the datapath, most
3482 * importantly when they last were used, and then use that information to
3483 * expire flows that have not been used recently.
3485 * Returns the number of milliseconds after which it should be called again. */
3487 expire(struct dpif_backer *backer)
3489 struct ofproto_dpif *ofproto;
3493 /* Periodically clear out the drop keys in an effort to keep them
3494 * relatively few. */
3495 drop_key_clear(backer);
3497 /* Update stats for each flow in the backer. */
3498 update_stats(backer);
3500 n_subfacets = hmap_count(&backer->subfacets);
3502 struct subfacet *subfacet;
3503 long long int total, now;
3507 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3508 total += now - subfacet->created;
3510 backer->avg_subfacet_life += total / n_subfacets;
3512 backer->avg_subfacet_life /= 2;
3514 backer->avg_n_subfacet += n_subfacets;
3515 backer->avg_n_subfacet /= 2;
3517 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3519 max_idle = subfacet_max_idle(backer);
3520 expire_subfacets(backer, max_idle);
3522 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3523 struct rule *rule, *next_rule;
3525 if (ofproto->backer != backer) {
3529 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3531 ovs_mutex_lock(&ofproto_mutex);
3532 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3533 &ofproto->up.expirable) {
3534 rule_expire(rule_dpif_cast(rule));
3536 ovs_mutex_unlock(&ofproto_mutex);
3538 /* All outstanding data in existing flows has been accounted, so it's a
3539 * good time to do bond rebalancing. */
3540 if (ofproto->has_bonded_bundles) {
3541 struct ofbundle *bundle;
3543 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3545 bond_rebalance(bundle->bond);
3551 return MIN(max_idle, 1000);
3554 /* Updates flow table statistics given that the datapath just reported 'stats'
3555 * as 'subfacet''s statistics. */
3557 update_subfacet_stats(struct subfacet *subfacet,
3558 const struct dpif_flow_stats *stats)
3560 struct facet *facet = subfacet->facet;
3561 struct dpif_flow_stats diff;
3563 diff.tcp_flags = stats->tcp_flags;
3564 diff.used = stats->used;
3566 if (stats->n_packets >= subfacet->dp_packet_count) {
3567 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3569 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3573 if (stats->n_bytes >= subfacet->dp_byte_count) {
3574 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3576 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3580 facet->ofproto->n_hit += diff.n_packets;
3581 subfacet->dp_packet_count = stats->n_packets;
3582 subfacet->dp_byte_count = stats->n_bytes;
3583 subfacet_update_stats(subfacet, &diff);
3585 if (facet->accounted_bytes < facet->byte_count) {
3587 facet_account(facet);
3588 facet->accounted_bytes = facet->byte_count;
3592 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3593 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3595 delete_unexpected_flow(struct dpif_backer *backer,
3596 const struct nlattr *key, size_t key_len)
3598 if (!VLOG_DROP_WARN(&rl)) {
3602 odp_flow_key_format(key, key_len, &s);
3603 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3607 COVERAGE_INC(facet_unexpected);
3608 dpif_flow_del(backer->dpif, key, key_len, NULL);
3611 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3613 * This function also pushes statistics updates to rules which each facet
3614 * resubmits into. Generally these statistics will be accurate. However, if a
3615 * facet changes the rule it resubmits into at some time in between
3616 * update_stats() runs, it is possible that statistics accrued to the
3617 * old rule will be incorrectly attributed to the new rule. This could be
3618 * avoided by calling update_stats() whenever rules are created or
3619 * deleted. However, the performance impact of making so many calls to the
3620 * datapath do not justify the benefit of having perfectly accurate statistics.
3622 * In addition, this function maintains per ofproto flow hit counts. The patch
3623 * port is not treated specially. e.g. A packet ingress from br0 patched into
3624 * br1 will increase the hit count of br0 by 1, however, does not affect
3625 * the hit or miss counts of br1.
3628 update_stats(struct dpif_backer *backer)
3630 const struct dpif_flow_stats *stats;
3631 struct dpif_flow_dump dump;
3632 const struct nlattr *key, *mask;
3633 size_t key_len, mask_len;
3635 dpif_flow_dump_start(&dump, backer->dpif);
3636 while (dpif_flow_dump_next(&dump, &key, &key_len,
3637 &mask, &mask_len, NULL, NULL, &stats)) {
3638 struct subfacet *subfacet;
3641 key_hash = odp_flow_key_hash(key, key_len);
3642 subfacet = subfacet_find(backer, key, key_len, key_hash);
3643 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3645 update_subfacet_stats(subfacet, stats);
3649 /* Stats are updated per-packet. */
3652 case SF_NOT_INSTALLED:
3654 delete_unexpected_flow(backer, key, key_len);
3659 dpif_flow_dump_done(&dump);
3662 /* Calculates and returns the number of milliseconds of idle time after which
3663 * subfacets should expire from the datapath. When a subfacet expires, we fold
3664 * its statistics into its facet, and when a facet's last subfacet expires, we
3665 * fold its statistic into its rule. */
3667 subfacet_max_idle(const struct dpif_backer *backer)
3670 * Idle time histogram.
3672 * Most of the time a switch has a relatively small number of subfacets.
3673 * When this is the case we might as well keep statistics for all of them
3674 * in userspace and to cache them in the kernel datapath for performance as
3677 * As the number of subfacets increases, the memory required to maintain
3678 * statistics about them in userspace and in the kernel becomes
3679 * significant. However, with a large number of subfacets it is likely
3680 * that only a few of them are "heavy hitters" that consume a large amount
3681 * of bandwidth. At this point, only heavy hitters are worth caching in
3682 * the kernel and maintaining in userspaces; other subfacets we can
3685 * The technique used to compute the idle time is to build a histogram with
3686 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3687 * that is installed in the kernel gets dropped in the appropriate bucket.
3688 * After the histogram has been built, we compute the cutoff so that only
3689 * the most-recently-used 1% of subfacets (but at least
3690 * flow_eviction_threshold flows) are kept cached. At least
3691 * the most-recently-used bucket of subfacets is kept, so actually an
3692 * arbitrary number of subfacets can be kept in any given expiration run
3693 * (though the next run will delete most of those unless they receive
3696 * This requires a second pass through the subfacets, in addition to the
3697 * pass made by update_stats(), because the former function never looks at
3698 * uninstallable subfacets.
3700 enum { BUCKET_WIDTH = 100 };
3701 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3702 int buckets[N_BUCKETS] = { 0 };
3703 int total, subtotal, bucket;
3704 struct subfacet *subfacet;
3708 total = hmap_count(&backer->subfacets);
3709 if (total <= flow_eviction_threshold) {
3710 return N_BUCKETS * BUCKET_WIDTH;
3713 /* Build histogram. */
3715 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3716 long long int idle = now - subfacet->used;
3717 int bucket = (idle <= 0 ? 0
3718 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3719 : (unsigned int) idle / BUCKET_WIDTH);
3723 /* Find the first bucket whose flows should be expired. */
3724 subtotal = bucket = 0;
3726 subtotal += buckets[bucket++];
3727 } while (bucket < N_BUCKETS &&
3728 subtotal < MAX(flow_eviction_threshold, total / 100));
3730 if (VLOG_IS_DBG_ENABLED()) {
3734 ds_put_cstr(&s, "keep");
3735 for (i = 0; i < N_BUCKETS; i++) {
3737 ds_put_cstr(&s, ", drop");
3740 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3743 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3747 return bucket * BUCKET_WIDTH;
3751 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3753 /* Cutoff time for most flows. */
3754 long long int normal_cutoff = time_msec() - dp_max_idle;
3756 /* We really want to keep flows for special protocols around, so use a more
3757 * conservative cutoff. */
3758 long long int special_cutoff = time_msec() - 10000;
3760 struct subfacet *subfacet, *next_subfacet;
3761 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3765 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3766 &backer->subfacets) {
3767 long long int cutoff;
3769 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3773 if (subfacet->used < cutoff) {
3774 if (subfacet->path != SF_NOT_INSTALLED) {
3775 batch[n_batch++] = subfacet;
3776 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3777 subfacet_destroy_batch(backer, batch, n_batch);
3781 subfacet_destroy(subfacet);
3787 subfacet_destroy_batch(backer, batch, n_batch);
3791 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3792 * then delete it entirely. */
3794 rule_expire(struct rule_dpif *rule)
3795 OVS_REQUIRES(ofproto_mutex)
3797 uint16_t idle_timeout, hard_timeout;
3798 long long int now = time_msec();
3801 ovs_assert(!rule->up.pending);
3803 /* Has 'rule' expired? */
3804 ovs_mutex_lock(&rule->up.mutex);
3805 hard_timeout = rule->up.hard_timeout;
3806 idle_timeout = rule->up.idle_timeout;
3807 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3808 reason = OFPRR_HARD_TIMEOUT;
3809 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3810 reason = OFPRR_IDLE_TIMEOUT;
3814 ovs_mutex_unlock(&rule->up.mutex);
3817 COVERAGE_INC(ofproto_dpif_expired);
3818 ofproto_rule_expire(&rule->up, reason);
3824 /* Creates and returns a new facet based on 'miss'.
3826 * The caller must already have determined that no facet with an identical
3827 * 'miss->flow' exists in 'miss->ofproto'.
3829 * 'rule' and 'xout' must have been created based on 'miss'.
3831 * 'facet'' statistics are initialized based on 'stats'.
3833 * The facet will initially have no subfacets. The caller should create (at
3834 * least) one subfacet with subfacet_create(). */
3835 static struct facet *
3836 facet_create(const struct flow_miss *miss)
3838 struct ofproto_dpif *ofproto = miss->ofproto;
3839 struct facet *facet;
3842 COVERAGE_INC(facet_create);
3843 facet = xzalloc(sizeof *facet);
3844 facet->ofproto = miss->ofproto;
3845 facet->used = miss->stats.used;
3846 facet->flow = miss->flow;
3847 facet->learn_rl = time_msec() + 500;
3849 list_init(&facet->subfacets);
3851 xlate_out_copy(&facet->xout, &miss->xout);
3853 match_init(&match, &facet->flow, &facet->xout.wc);
3854 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3855 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3856 classifier_insert(&ofproto->facets, &facet->cr);
3857 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3859 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3860 netflow_flow_update(ofproto->netflow, &facet->flow,
3861 facet->xout.nf_output_iface, &miss->stats);
3868 facet_free(struct facet *facet)
3871 xlate_out_uninit(&facet->xout);
3876 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3877 * 'flow' must reflect the data in 'packet'. */
3879 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3880 const struct flow *flow,
3881 struct rule_dpif *rule,
3882 const struct ofpact *ofpacts, size_t ofpacts_len,
3883 struct ofpbuf *packet)
3885 struct odputil_keybuf keybuf;
3886 struct dpif_flow_stats stats;
3887 struct xlate_out xout;
3888 struct xlate_in xin;
3893 ovs_assert((rule != NULL) != (ofpacts != NULL));
3895 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3897 rule_dpif_credit_stats(rule, &stats);
3900 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3901 xin.ofpacts = ofpacts;
3902 xin.ofpacts_len = ofpacts_len;
3903 xin.resubmit_stats = &stats;
3904 xlate_actions(&xin, &xout);
3906 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3907 in_port = flow->in_port.ofp_port;
3908 if (in_port == OFPP_NONE) {
3909 in_port = OFPP_LOCAL;
3911 odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3913 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3914 xout.odp_actions.data, xout.odp_actions.size, packet,
3915 (xout.slow & SLOW_ACTION) != 0);
3916 xlate_out_uninit(&xout);
3921 /* Remove 'facet' from its ofproto and free up the associated memory:
3923 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3924 * rule's statistics, via subfacet_uninstall().
3926 * - Removes 'facet' from its rule and from ofproto->facets.
3929 facet_remove(struct facet *facet)
3931 struct subfacet *subfacet, *next_subfacet;
3933 COVERAGE_INC(facet_remove);
3934 ovs_assert(!list_is_empty(&facet->subfacets));
3936 /* First uninstall all of the subfacets to get final statistics. */
3937 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3938 subfacet_uninstall(subfacet);
3941 /* Flush the final stats to the rule.
3943 * This might require us to have at least one subfacet around so that we
3944 * can use its actions for accounting in facet_account(), which is why we
3945 * have uninstalled but not yet destroyed the subfacets. */
3946 facet_flush_stats(facet);
3948 /* Now we're really all done so destroy everything. */
3949 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3950 &facet->subfacets) {
3951 subfacet_destroy__(subfacet);
3953 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3954 classifier_remove(&facet->ofproto->facets, &facet->cr);
3955 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3956 cls_rule_destroy(&facet->cr);
3960 /* Feed information from 'facet' back into the learning table to keep it in
3961 * sync with what is actually flowing through the datapath. */
3963 facet_learn(struct facet *facet)
3965 long long int now = time_msec();
3967 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3971 facet->learn_rl = now + 500;
3973 if (!facet->xout.has_learn
3974 && !facet->xout.has_normal
3975 && (!facet->xout.has_fin_timeout
3976 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3980 facet_push_stats(facet, true);
3984 facet_account(struct facet *facet)
3986 const struct nlattr *a;
3991 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3994 n_bytes = facet->byte_count - facet->accounted_bytes;
3996 /* This loop feeds byte counters to bond_account() for rebalancing to use
3997 * as a basis. We also need to track the actual VLAN on which the packet
3998 * is going to be sent to ensure that it matches the one passed to
3999 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4002 * We use the actions from an arbitrary subfacet because they should all
4003 * be equally valid for our purpose. */
4004 vlan_tci = facet->flow.vlan_tci;
4005 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4006 facet->xout.odp_actions.size) {
4007 const struct ovs_action_push_vlan *vlan;
4008 struct ofport_dpif *port;
4010 switch (nl_attr_type(a)) {
4011 case OVS_ACTION_ATTR_OUTPUT:
4012 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4013 if (port && port->bundle && port->bundle->bond) {
4014 bond_account(port->bundle->bond, &facet->flow,
4015 vlan_tci_to_vid(vlan_tci), n_bytes);
4019 case OVS_ACTION_ATTR_POP_VLAN:
4020 vlan_tci = htons(0);
4023 case OVS_ACTION_ATTR_PUSH_VLAN:
4024 vlan = nl_attr_get(a);
4025 vlan_tci = vlan->vlan_tci;
4031 /* Returns true if the only action for 'facet' is to send to the controller.
4032 * (We don't report NetFlow expiration messages for such facets because they
4033 * are just part of the control logic for the network, not real traffic). */
4035 facet_is_controller_flow(struct facet *facet)
4038 struct ofproto_dpif *ofproto = facet->ofproto;
4039 const struct ofpact *ofpacts;
4040 struct rule_actions *actions;
4041 struct rule_dpif *rule;
4045 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4046 actions = rule_dpif_get_actions(rule);
4047 rule_dpif_unref(rule);
4049 ofpacts_len = actions->ofpacts_len;
4050 ofpacts = actions->ofpacts;
4051 is_controller = ofpacts_len > 0
4052 && ofpacts->type == OFPACT_CONTROLLER
4053 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4054 rule_actions_unref(actions);
4056 return is_controller;
4061 /* Folds all of 'facet''s statistics into its rule. Also updates the
4062 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4063 * 'facet''s statistics in the datapath should have been zeroed and folded into
4064 * its packet and byte counts before this function is called. */
4066 facet_flush_stats(struct facet *facet)
4068 struct ofproto_dpif *ofproto = facet->ofproto;
4069 struct subfacet *subfacet;
4071 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4072 ovs_assert(!subfacet->dp_byte_count);
4073 ovs_assert(!subfacet->dp_packet_count);
4076 facet_push_stats(facet, false);
4077 if (facet->accounted_bytes < facet->byte_count) {
4078 facet_account(facet);
4079 facet->accounted_bytes = facet->byte_count;
4082 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4083 netflow_expire(ofproto->netflow, &facet->flow);
4084 netflow_flow_clear(ofproto->netflow, &facet->flow);
4087 /* Reset counters to prevent double counting if 'facet' ever gets
4089 facet_reset_counters(facet);
4090 facet->tcp_flags = 0;
4093 /* Searches 'ofproto''s table of facets for one which would be responsible for
4094 * 'flow'. Returns it if found, otherwise a null pointer.
4096 * The returned facet might need revalidation; use facet_lookup_valid()
4097 * instead if that is important. */
4098 static struct facet *
4099 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4101 struct cls_rule *cr;
4103 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4104 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4105 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4106 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4109 /* Searches 'ofproto''s table of facets for one capable that covers
4110 * 'flow'. Returns it if found, otherwise a null pointer.
4112 * The returned facet is guaranteed to be valid. */
4113 static struct facet *
4114 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4116 struct facet *facet;
4118 facet = facet_find(ofproto, flow);
4120 && ofproto->backer->need_revalidate
4121 && !facet_revalidate(facet)) {
4129 facet_check_consistency(struct facet *facet)
4131 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4133 struct xlate_out xout;
4134 struct xlate_in xin;
4137 /* Check the datapath actions for consistency. */
4138 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4139 xlate_actions(&xin, &xout);
4141 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4142 && facet->xout.slow == xout.slow;
4143 if (!ok && !VLOG_DROP_WARN(&rl)) {
4144 struct ds s = DS_EMPTY_INITIALIZER;
4146 flow_format(&s, &facet->flow);
4147 ds_put_cstr(&s, ": inconsistency in facet");
4149 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4150 ds_put_cstr(&s, " (actions were: ");
4151 format_odp_actions(&s, facet->xout.odp_actions.data,
4152 facet->xout.odp_actions.size);
4153 ds_put_cstr(&s, ") (correct actions: ");
4154 format_odp_actions(&s, xout.odp_actions.data,
4155 xout.odp_actions.size);
4156 ds_put_char(&s, ')');
4159 if (facet->xout.slow != xout.slow) {
4160 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4165 xlate_out_uninit(&xout);
4170 /* Re-searches the classifier for 'facet':
4172 * - If the rule found is different from 'facet''s current rule, moves
4173 * 'facet' to the new rule and recompiles its actions.
4175 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4176 * where it is and recompiles its actions anyway.
4178 * - If any of 'facet''s subfacets correspond to a new flow according to
4179 * xlate_receive(), 'facet' is removed.
4181 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4183 facet_revalidate(struct facet *facet)
4185 struct ofproto_dpif *ofproto = facet->ofproto;
4186 struct rule_dpif *new_rule;
4187 struct subfacet *subfacet;
4188 struct flow_wildcards wc;
4189 struct xlate_out xout;
4190 struct xlate_in xin;
4192 COVERAGE_INC(facet_revalidate);
4194 /* Check that child subfacets still correspond to this facet. Tunnel
4195 * configuration changes could cause a subfacet's OpenFlow in_port to
4197 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4198 struct ofproto_dpif *recv_ofproto;
4199 struct flow recv_flow;
4202 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4203 subfacet->key_len, &recv_flow, NULL,
4204 &recv_ofproto, NULL, NULL, NULL, NULL);
4206 || recv_ofproto != ofproto
4207 || facet != facet_find(ofproto, &recv_flow)) {
4208 facet_remove(facet);
4213 flow_wildcards_init_catchall(&wc);
4214 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4216 /* Calculate new datapath actions.
4218 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4219 * emit a NetFlow expiration and, if so, we need to have the old state
4220 * around to properly compose it. */
4221 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4222 xlate_actions(&xin, &xout);
4223 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4224 /* Make sure non -packet fields are not masked. If not cleared,
4225 * the memcmp() below may fail, causing an otherwise valid facet
4227 flow_wildcards_clear_non_packet_fields(&xout.wc);
4229 /* A facet's slow path reason should only change under dramatic
4230 * circumstances. Rather than try to update everything, it's simpler to
4231 * remove the facet and start over.
4233 * More importantly, if a facet's wildcards change, it will be relatively
4234 * difficult to figure out if its subfacets still belong to it, and if not
4235 * which facet they may belong to. Again, to avoid the complexity, we
4236 * simply give up instead. */
4237 if (facet->xout.slow != xout.slow
4238 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4239 facet_remove(facet);
4240 xlate_out_uninit(&xout);
4241 rule_dpif_unref(new_rule);
4245 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4246 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4247 if (subfacet->path == SF_FAST_PATH) {
4248 struct dpif_flow_stats stats;
4250 subfacet_install(subfacet, &xout.odp_actions, &stats);
4251 subfacet_update_stats(subfacet, &stats);
4255 facet_flush_stats(facet);
4257 ofpbuf_clear(&facet->xout.odp_actions);
4258 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4259 xout.odp_actions.size);
4262 /* Update 'facet' now that we've taken care of all the old state. */
4263 facet->xout.slow = xout.slow;
4264 facet->xout.has_learn = xout.has_learn;
4265 facet->xout.has_normal = xout.has_normal;
4266 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4267 facet->xout.nf_output_iface = xout.nf_output_iface;
4268 facet->xout.mirrors = xout.mirrors;
4270 ovs_mutex_lock(&new_rule->up.mutex);
4271 facet->used = MAX(facet->used, new_rule->up.created);
4272 ovs_mutex_unlock(&new_rule->up.mutex);
4274 xlate_out_uninit(&xout);
4275 rule_dpif_unref(new_rule);
4280 facet_reset_counters(struct facet *facet)
4282 facet->packet_count = 0;
4283 facet->byte_count = 0;
4284 facet->prev_packet_count = 0;
4285 facet->prev_byte_count = 0;
4286 facet->accounted_bytes = 0;
4290 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4291 struct dpif_flow_stats *stats, bool may_learn)
4293 struct ofport_dpif *in_port;
4294 struct xlate_in xin;
4296 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4297 if (in_port && in_port->is_tunnel) {
4298 netdev_vport_inc_rx(in_port->up.netdev, stats);
4300 bfd_account_rx(in_port->bfd, stats);
4304 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4305 xin.resubmit_stats = stats;
4306 xin.may_learn = may_learn;
4307 xlate_actions_for_side_effects(&xin);
4311 facet_push_stats(struct facet *facet, bool may_learn)
4313 struct dpif_flow_stats stats;
4315 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4316 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4317 ovs_assert(facet->used >= facet->prev_used);
4319 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4320 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4321 stats.used = facet->used;
4322 stats.tcp_flags = facet->tcp_flags;
4324 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4325 facet->prev_packet_count = facet->packet_count;
4326 facet->prev_byte_count = facet->byte_count;
4327 facet->prev_used = facet->used;
4329 if (facet->ofproto->netflow && !facet_is_controller_flow(facet)) {
4330 netflow_flow_update(facet->ofproto->netflow, &facet->flow,
4331 facet->xout.nf_output_iface, &stats);
4333 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4334 stats.n_packets, stats.n_bytes);
4335 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4340 push_all_stats__(bool run_fast)
4342 static long long int rl = LLONG_MIN;
4343 struct ofproto_dpif *ofproto;
4345 if (time_msec() < rl) {
4349 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4350 struct cls_cursor cursor;
4351 struct facet *facet;
4353 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4354 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4355 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4356 facet_push_stats(facet, false);
4361 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4364 rl = time_msec() + 100;
4368 push_all_stats(void)
4370 push_all_stats__(true);
4374 rule_dpif_credit_stats(struct rule_dpif *rule,
4375 const struct dpif_flow_stats *stats)
4377 ovs_mutex_lock(&rule->stats_mutex);
4378 rule->packet_count += stats->n_packets;
4379 rule->byte_count += stats->n_bytes;
4380 rule->up.used = MAX(rule->up.used, stats->used);
4381 ovs_mutex_unlock(&rule->stats_mutex);
4385 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4387 return is_fail_open_rule(&rule->up);
4391 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4393 return rule_is_table_miss(&rule->up);
4397 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4398 OVS_REQUIRES(rule->up.mutex)
4400 return rule->up.flow_cookie;
4404 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4405 uint16_t hard_timeout)
4407 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4410 /* Returns 'rule''s actions. The caller owns a reference on the returned
4411 * actions and must eventually release it (with rule_actions_unref()) to avoid
4413 struct rule_actions *
4414 rule_dpif_get_actions(const struct rule_dpif *rule)
4416 return rule_get_actions(&rule->up);
4421 static struct subfacet *
4422 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4423 size_t key_len, uint32_t key_hash)
4425 struct subfacet *subfacet;
4427 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4428 &backer->subfacets) {
4429 if (subfacet->key_len == key_len
4430 && !memcmp(key, subfacet->key, key_len)) {
4438 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4439 * 'key_hash' must be a hash over miss->key. The caller must have already
4440 * ensured that no subfacet subfacet already exists. */
4441 static struct subfacet *
4442 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4444 struct dpif_backer *backer = miss->ofproto->backer;
4445 const struct nlattr *key = miss->key;
4446 size_t key_len = miss->key_len;
4447 struct subfacet *subfacet;
4449 subfacet = (list_is_empty(&facet->subfacets)
4450 ? &facet->one_subfacet
4451 : xmalloc(sizeof *subfacet));
4453 COVERAGE_INC(subfacet_create);
4454 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4455 list_push_back(&facet->subfacets, &subfacet->list_node);
4456 subfacet->facet = facet;
4457 subfacet->key = xmemdup(key, key_len);
4458 subfacet->key_len = key_len;
4459 subfacet->used = miss->stats.used;
4460 subfacet->created = subfacet->used;
4461 subfacet->dp_packet_count = 0;
4462 subfacet->dp_byte_count = 0;
4463 subfacet->path = SF_NOT_INSTALLED;
4464 subfacet->backer = backer;
4469 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4470 * its facet within 'ofproto', and frees it. */
4472 subfacet_destroy__(struct subfacet *subfacet)
4474 struct facet *facet = subfacet->facet;
4476 COVERAGE_INC(subfacet_destroy);
4477 subfacet_uninstall(subfacet);
4478 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4479 list_remove(&subfacet->list_node);
4480 free(subfacet->key);
4481 if (subfacet != &facet->one_subfacet) {
4486 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4487 * last remaining subfacet in its facet destroys the facet too. */
4489 subfacet_destroy(struct subfacet *subfacet)
4491 struct facet *facet = subfacet->facet;
4493 if (list_is_singleton(&facet->subfacets)) {
4494 /* facet_remove() needs at least one subfacet (it will remove it). */
4495 facet_remove(facet);
4497 subfacet_destroy__(subfacet);
4502 subfacet_destroy_batch(struct dpif_backer *backer,
4503 struct subfacet **subfacets, int n)
4505 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4506 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4507 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4510 for (i = 0; i < n; i++) {
4511 ops[i].type = DPIF_OP_FLOW_DEL;
4512 ops[i].u.flow_del.key = subfacets[i]->key;
4513 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4514 ops[i].u.flow_del.stats = &stats[i];
4518 dpif_operate(backer->dpif, opsp, n);
4519 for (i = 0; i < n; i++) {
4520 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4521 subfacets[i]->path = SF_NOT_INSTALLED;
4522 subfacet_destroy(subfacets[i]);
4527 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4528 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4529 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4530 * since 'subfacet' was last updated.
4532 * Returns 0 if successful, otherwise a positive errno value. */
4534 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4535 struct dpif_flow_stats *stats)
4537 struct facet *facet = subfacet->facet;
4538 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4539 const struct nlattr *actions = odp_actions->data;
4540 size_t actions_len = odp_actions->size;
4541 struct odputil_keybuf maskbuf;
4544 uint64_t slow_path_stub[128 / 8];
4545 enum dpif_flow_put_flags flags;
4548 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4551 flags |= DPIF_FP_ZERO_STATS;
4554 if (path == SF_SLOW_PATH) {
4555 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4556 slow_path_stub, sizeof slow_path_stub,
4557 &actions, &actions_len);
4560 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4561 if (enable_megaflows) {
4562 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4563 &facet->flow, UINT32_MAX);
4566 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4567 subfacet->key_len, mask.data, mask.size,
4568 actions, actions_len, stats);
4571 subfacet_reset_dp_stats(subfacet, stats);
4575 COVERAGE_INC(subfacet_install_fail);
4577 subfacet->path = path;
4582 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4584 subfacet_uninstall(struct subfacet *subfacet)
4586 if (subfacet->path != SF_NOT_INSTALLED) {
4587 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4588 struct dpif_flow_stats stats;
4591 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4592 subfacet->key_len, &stats);
4593 subfacet_reset_dp_stats(subfacet, &stats);
4595 subfacet_update_stats(subfacet, &stats);
4597 subfacet->path = SF_NOT_INSTALLED;
4599 ovs_assert(subfacet->dp_packet_count == 0);
4600 ovs_assert(subfacet->dp_byte_count == 0);
4604 /* Resets 'subfacet''s datapath statistics counters. This should be called
4605 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4606 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4607 * was reset in the datapath. 'stats' will be modified to include only
4608 * statistics new since 'subfacet' was last updated. */
4610 subfacet_reset_dp_stats(struct subfacet *subfacet,
4611 struct dpif_flow_stats *stats)
4614 && subfacet->dp_packet_count <= stats->n_packets
4615 && subfacet->dp_byte_count <= stats->n_bytes) {
4616 stats->n_packets -= subfacet->dp_packet_count;
4617 stats->n_bytes -= subfacet->dp_byte_count;
4620 subfacet->dp_packet_count = 0;
4621 subfacet->dp_byte_count = 0;
4624 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4626 * Because of the meaning of a subfacet's counters, it only makes sense to do
4627 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4628 * represents a packet that was sent by hand or if it represents statistics
4629 * that have been cleared out of the datapath. */
4631 subfacet_update_stats(struct subfacet *subfacet,
4632 const struct dpif_flow_stats *stats)
4634 if (stats->n_packets || stats->used > subfacet->used) {
4635 struct facet *facet = subfacet->facet;
4637 subfacet->used = MAX(subfacet->used, stats->used);
4638 facet->used = MAX(facet->used, stats->used);
4639 facet->packet_count += stats->n_packets;
4640 facet->byte_count += stats->n_bytes;
4641 facet->tcp_flags |= stats->tcp_flags;
4647 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4648 * the fields that were relevant as part of the lookup. */
4650 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4651 struct flow_wildcards *wc, struct rule_dpif **rule)
4653 struct ofport_dpif *port;
4655 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4658 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4660 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4661 flow->in_port.ofp_port);
4664 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4665 ofproto->no_packet_in_rule, rule);
4669 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4670 const struct flow *flow, struct flow_wildcards *wc,
4671 uint8_t table_id, struct rule_dpif **rule)
4673 const struct cls_rule *cls_rule;
4674 struct classifier *cls;
4678 if (table_id >= N_TABLES) {
4683 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4684 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4687 cls = &ofproto->up.tables[table_id].cls;
4688 ovs_rwlock_rdlock(&cls->rwlock);
4689 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4690 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4691 /* We must pretend that transport ports are unavailable. */
4692 struct flow ofpc_normal_flow = *flow;
4693 ofpc_normal_flow.tp_src = htons(0);
4694 ofpc_normal_flow.tp_dst = htons(0);
4695 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4696 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4697 cls_rule = &ofproto->drop_frags_rule->up.cr;
4698 /* Frag mask in wc already set above. */
4700 cls_rule = classifier_lookup(cls, flow, wc);
4703 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4704 rule_dpif_ref(*rule);
4705 ovs_rwlock_unlock(&cls->rwlock);
4707 return *rule != NULL;
4710 /* Given a port configuration (specified as zero if there's no port), chooses
4711 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4712 * flow table miss. */
4714 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4715 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4717 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4718 rule_dpif_ref(*rule);
4722 rule_dpif_ref(struct rule_dpif *rule)
4725 ofproto_rule_ref(&rule->up);
4730 rule_dpif_unref(struct rule_dpif *rule)
4733 ofproto_rule_unref(&rule->up);
4738 complete_operation(struct rule_dpif *rule)
4739 OVS_REQUIRES(ofproto_mutex)
4741 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4743 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4744 ofoperation_complete(rule->up.pending, 0);
4747 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4749 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4752 static struct rule *
4755 struct rule_dpif *rule = xmalloc(sizeof *rule);
4760 rule_dealloc(struct rule *rule_)
4762 struct rule_dpif *rule = rule_dpif_cast(rule_);
4767 rule_construct(struct rule *rule_)
4769 struct rule_dpif *rule = rule_dpif_cast(rule_);
4770 ovs_mutex_init(&rule->stats_mutex);
4771 ovs_mutex_lock(&rule->stats_mutex);
4772 rule->packet_count = 0;
4773 rule->byte_count = 0;
4774 ovs_mutex_unlock(&rule->stats_mutex);
4779 rule_insert(struct rule *rule_)
4780 OVS_REQUIRES(ofproto_mutex)
4782 struct rule_dpif *rule = rule_dpif_cast(rule_);
4783 complete_operation(rule);
4787 rule_delete(struct rule *rule_)
4788 OVS_REQUIRES(ofproto_mutex)
4790 struct rule_dpif *rule = rule_dpif_cast(rule_);
4791 complete_operation(rule);
4795 rule_destruct(struct rule *rule_)
4797 struct rule_dpif *rule = rule_dpif_cast(rule_);
4798 ovs_mutex_destroy(&rule->stats_mutex);
4802 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4804 struct rule_dpif *rule = rule_dpif_cast(rule_);
4806 /* push_all_stats() can handle flow misses which, when using the learn
4807 * action, can cause rules to be added and deleted. This can corrupt our
4808 * caller's datastructures which assume that rule_get_stats() doesn't have
4809 * an impact on the flow table. To be safe, we disable miss handling. */
4810 push_all_stats__(false);
4812 /* Start from historical data for 'rule' itself that are no longer tracked
4813 * in facets. This counts, for example, facets that have expired. */
4814 ovs_mutex_lock(&rule->stats_mutex);
4815 *packets = rule->packet_count;
4816 *bytes = rule->byte_count;
4817 ovs_mutex_unlock(&rule->stats_mutex);
4821 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4822 struct ofpbuf *packet)
4824 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4826 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4830 rule_execute(struct rule *rule, const struct flow *flow,
4831 struct ofpbuf *packet)
4833 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4834 ofpbuf_delete(packet);
4839 rule_modify_actions(struct rule *rule_, bool reset_counters)
4840 OVS_REQUIRES(ofproto_mutex)
4842 struct rule_dpif *rule = rule_dpif_cast(rule_);
4844 if (reset_counters) {
4845 ovs_mutex_lock(&rule->stats_mutex);
4846 rule->packet_count = 0;
4847 rule->byte_count = 0;
4848 ovs_mutex_unlock(&rule->stats_mutex);
4851 complete_operation(rule);
4854 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4856 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4859 static struct ofgroup *
4862 struct group_dpif *group = xzalloc(sizeof *group);
4867 group_dealloc(struct ofgroup *group_)
4869 struct group_dpif *group = group_dpif_cast(group_);
4874 group_construct_stats(struct group_dpif *group)
4875 OVS_REQUIRES(group->stats_mutex)
4877 group->packet_count = 0;
4878 group->byte_count = 0;
4879 if (!group->bucket_stats) {
4880 group->bucket_stats = xcalloc(group->up.n_buckets,
4881 sizeof *group->bucket_stats);
4883 memset(group->bucket_stats, 0, group->up.n_buckets *
4884 sizeof *group->bucket_stats);
4889 group_construct(struct ofgroup *group_)
4891 struct group_dpif *group = group_dpif_cast(group_);
4892 ovs_mutex_init(&group->stats_mutex);
4893 ovs_mutex_lock(&group->stats_mutex);
4894 group_construct_stats(group);
4895 ovs_mutex_unlock(&group->stats_mutex);
4900 group_destruct__(struct group_dpif *group)
4901 OVS_REQUIRES(group->stats_mutex)
4903 free(group->bucket_stats);
4904 group->bucket_stats = NULL;
4908 group_destruct(struct ofgroup *group_)
4910 struct group_dpif *group = group_dpif_cast(group_);
4911 ovs_mutex_lock(&group->stats_mutex);
4912 group_destruct__(group);
4913 ovs_mutex_unlock(&group->stats_mutex);
4914 ovs_mutex_destroy(&group->stats_mutex);
4918 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4920 struct group_dpif *group = group_dpif_cast(group_);
4921 struct group_dpif *victim = group_dpif_cast(victim_);
4923 ovs_mutex_lock(&group->stats_mutex);
4924 if (victim->up.n_buckets < group->up.n_buckets) {
4925 group_destruct__(group);
4927 group_construct_stats(group);
4928 ovs_mutex_unlock(&group->stats_mutex);
4934 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4936 struct group_dpif *group = group_dpif_cast(group_);
4938 /* Start from historical data for 'group' itself that are no longer tracked
4939 * in facets. This counts, for example, facets that have expired. */
4940 ovs_mutex_lock(&group->stats_mutex);
4941 ogs->packet_count = group->packet_count;
4942 ogs->byte_count = group->byte_count;
4943 memcpy(ogs->bucket_stats, group->bucket_stats,
4944 group->up.n_buckets * sizeof *group->bucket_stats);
4945 ovs_mutex_unlock(&group->stats_mutex);
4951 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4952 struct group_dpif **group)
4953 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4955 struct ofgroup *ofgroup;
4959 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4960 *group = found ? group_dpif_cast(ofgroup) : NULL;
4966 group_dpif_release(struct group_dpif *group)
4967 OVS_RELEASES(group->up.rwlock)
4969 ofproto_group_release(&group->up);
4973 group_dpif_get_buckets(const struct group_dpif *group,
4974 const struct list **buckets)
4976 *buckets = &group->up.buckets;
4979 enum ofp11_group_type
4980 group_dpif_get_type(const struct group_dpif *group)
4982 return group->up.type;
4985 /* Sends 'packet' out 'ofport'.
4986 * May modify 'packet'.
4987 * Returns 0 if successful, otherwise a positive errno value. */
4989 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4994 error = xlate_send_packet(ofport, packet);
4996 ovs_mutex_lock(&ofproto->stats_mutex);
4997 ofproto->stats.tx_packets++;
4998 ofproto->stats.tx_bytes += packet->size;
4999 ovs_mutex_unlock(&ofproto->stats_mutex);
5003 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5004 * The action will state 'slow' as the reason that the action is in the slow
5005 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5006 * dump-flows" output to see why a flow is in the slow path.)
5008 * The 'stub_size' bytes in 'stub' will be used to store the action.
5009 * 'stub_size' must be large enough for the action.
5011 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5014 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5015 enum slow_path_reason slow,
5016 uint64_t *stub, size_t stub_size,
5017 const struct nlattr **actionsp, size_t *actions_lenp)
5019 union user_action_cookie cookie;
5022 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5023 cookie.slow_path.unused = 0;
5024 cookie.slow_path.reason = slow;
5026 ofpbuf_use_stack(&buf, stub, stub_size);
5027 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5028 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5030 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5032 odp_port_t odp_port;
5035 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5036 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5037 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5039 *actionsp = buf.data;
5040 *actions_lenp = buf.size;
5044 set_frag_handling(struct ofproto *ofproto_,
5045 enum ofp_config_flags frag_handling)
5047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5048 if (frag_handling != OFPC_FRAG_REASM) {
5049 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5057 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5058 const struct flow *flow,
5059 const struct ofpact *ofpacts, size_t ofpacts_len)
5061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5063 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5064 ofpacts_len, packet);
5071 set_netflow(struct ofproto *ofproto_,
5072 const struct netflow_options *netflow_options)
5074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5076 if (netflow_options) {
5077 if (!ofproto->netflow) {
5078 ofproto->netflow = netflow_create();
5079 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5081 return netflow_set_options(ofproto->netflow, netflow_options);
5082 } else if (ofproto->netflow) {
5083 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5084 netflow_unref(ofproto->netflow);
5085 ofproto->netflow = NULL;
5092 get_netflow_ids(const struct ofproto *ofproto_,
5093 uint8_t *engine_type, uint8_t *engine_id)
5095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5097 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5100 static struct ofproto_dpif *
5101 ofproto_dpif_lookup(const char *name)
5103 struct ofproto_dpif *ofproto;
5105 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5106 hash_string(name, 0), &all_ofproto_dpifs) {
5107 if (!strcmp(ofproto->up.name, name)) {
5115 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5116 const char *argv[], void *aux OVS_UNUSED)
5118 struct ofproto_dpif *ofproto;
5121 ofproto = ofproto_dpif_lookup(argv[1]);
5123 unixctl_command_reply_error(conn, "no such bridge");
5126 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5127 mac_learning_flush(ofproto->ml);
5128 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5130 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5131 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5132 mac_learning_flush(ofproto->ml);
5133 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5137 unixctl_command_reply(conn, "table successfully flushed");
5140 static struct ofport_dpif *
5141 ofbundle_get_a_port(const struct ofbundle *bundle)
5143 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5148 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5149 const char *argv[], void *aux OVS_UNUSED)
5151 struct ds ds = DS_EMPTY_INITIALIZER;
5152 const struct ofproto_dpif *ofproto;
5153 const struct mac_entry *e;
5155 ofproto = ofproto_dpif_lookup(argv[1]);
5157 unixctl_command_reply_error(conn, "no such bridge");
5161 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5162 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5163 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5164 struct ofbundle *bundle = e->port.p;
5165 char name[OFP_MAX_PORT_NAME_LEN];
5167 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5169 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5170 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5171 mac_entry_age(ofproto->ml, e));
5173 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5174 unixctl_command_reply(conn, ds_cstr(&ds));
5179 struct xlate_out xout;
5180 struct xlate_in xin;
5186 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5188 struct rule_actions *actions;
5191 ds_put_char_multiple(result, '\t', level);
5193 ds_put_cstr(result, "No match\n");
5197 ovs_mutex_lock(&rule->up.mutex);
5198 cookie = rule->up.flow_cookie;
5199 ovs_mutex_unlock(&rule->up.mutex);
5201 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5202 rule ? rule->up.table_id : 0, ntohll(cookie));
5203 cls_rule_format(&rule->up.cr, result);
5204 ds_put_char(result, '\n');
5206 actions = rule_dpif_get_actions(rule);
5208 ds_put_char_multiple(result, '\t', level);
5209 ds_put_cstr(result, "OpenFlow actions=");
5210 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5211 ds_put_char(result, '\n');
5213 rule_actions_unref(actions);
5217 trace_format_flow(struct ds *result, int level, const char *title,
5218 struct trace_ctx *trace)
5220 ds_put_char_multiple(result, '\t', level);
5221 ds_put_format(result, "%s: ", title);
5222 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5223 ds_put_cstr(result, "unchanged");
5225 flow_format(result, &trace->xin.flow);
5226 trace->flow = trace->xin.flow;
5228 ds_put_char(result, '\n');
5232 trace_format_regs(struct ds *result, int level, const char *title,
5233 struct trace_ctx *trace)
5237 ds_put_char_multiple(result, '\t', level);
5238 ds_put_format(result, "%s:", title);
5239 for (i = 0; i < FLOW_N_REGS; i++) {
5240 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5242 ds_put_char(result, '\n');
5246 trace_format_odp(struct ds *result, int level, const char *title,
5247 struct trace_ctx *trace)
5249 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5251 ds_put_char_multiple(result, '\t', level);
5252 ds_put_format(result, "%s: ", title);
5253 format_odp_actions(result, odp_actions->data, odp_actions->size);
5254 ds_put_char(result, '\n');
5258 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5260 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5261 struct ds *result = trace->result;
5263 ds_put_char(result, '\n');
5264 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5265 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5266 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5267 trace_format_rule(result, recurse + 1, rule);
5271 trace_report(struct xlate_in *xin, const char *s, int recurse)
5273 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5274 struct ds *result = trace->result;
5276 ds_put_char_multiple(result, '\t', recurse);
5277 ds_put_cstr(result, s);
5278 ds_put_char(result, '\n');
5281 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
5282 * forms are supported:
5284 * - [dpname] odp_flow [-generate | packet]
5285 * - bridge br_flow [-generate | packet]
5287 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
5288 * returns a nonnull error message. */
5290 parse_flow_and_packet(int argc, const char *argv[],
5291 struct ofproto_dpif **ofprotop, struct flow *flow,
5292 struct ofpbuf **packetp)
5294 const struct dpif_backer *backer = NULL;
5295 const char *error = NULL;
5296 struct simap port_names = SIMAP_INITIALIZER(&port_names);
5297 struct ofpbuf *packet;
5298 struct ofpbuf odp_key;
5299 struct ofpbuf odp_mask;
5301 ofpbuf_init(&odp_key, 0);
5302 ofpbuf_init(&odp_mask, 0);
5304 /* Handle "-generate" or a hex string as the last argument. */
5305 if (!strcmp(argv[argc - 1], "-generate")) {
5306 packet = ofpbuf_new(0);
5309 error = eth_from_hex(argv[argc - 1], &packet);
5312 } else if (argc == 4) {
5313 /* The 3-argument form must end in "-generate' or a hex string. */
5318 /* odp_flow can have its in_port specified as a name instead of port no.
5319 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5320 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5321 * we need to create a simap of name to port no. */
5323 const char *dp_type;
5324 if (!strncmp(argv[1], "ovs-", 4)) {
5325 dp_type = argv[1] + 4;
5329 backer = shash_find_data(&all_dpif_backers, dp_type);
5330 } else if (argc == 2) {
5331 struct shash_node *node;
5332 if (shash_count(&all_dpif_backers) == 1) {
5333 node = shash_first(&all_dpif_backers);
5334 backer = node->data;
5337 error = "Syntax error";
5340 if (backer && backer->dpif) {
5341 struct dpif_port dpif_port;
5342 struct dpif_port_dump port_dump;
5343 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5344 simap_put(&port_names, dpif_port.name,
5345 odp_to_u32(dpif_port.port_no));
5349 /* Parse the flow and determine whether a datapath or
5350 * bridge is specified. If function odp_flow_key_from_string()
5351 * returns 0, the flow is a odp_flow. If function
5352 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5353 if (!odp_flow_from_string(argv[argc - 1], &port_names,
5354 &odp_key, &odp_mask)) {
5356 error = "Cannot find the datapath";
5360 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5361 NULL, ofprotop, NULL, NULL, NULL, NULL)) {
5362 error = "Invalid datapath flow";
5365 } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5367 error = "Must specify bridge name";
5371 *ofprotop = ofproto_dpif_lookup(argv[1]);
5373 error = "Unknown bridge name";
5377 error = "Bad flow syntax";
5381 /* Generate a packet, if requested. */
5383 if (!packet->size) {
5384 flow_compose(packet, flow);
5386 union flow_in_port in_port = flow->in_port;
5388 /* Use the metadata from the flow and the packet argument
5389 * to reconstruct the flow. */
5390 flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5399 ofpbuf_delete(packet);
5403 ofpbuf_uninit(&odp_key);
5404 ofpbuf_uninit(&odp_mask);
5405 simap_destroy(&port_names);
5410 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5411 void *aux OVS_UNUSED)
5413 struct ofproto_dpif *ofproto;
5414 struct ofpbuf *packet;
5418 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5423 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5424 unixctl_command_reply(conn, ds_cstr(&result));
5425 ds_destroy(&result);
5426 ofpbuf_delete(packet);
5428 unixctl_command_reply_error(conn, error);
5433 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5434 const char *argv[], void *aux OVS_UNUSED)
5436 enum ofputil_protocol usable_protocols;
5437 struct ofproto_dpif *ofproto;
5438 bool enforce_consistency;
5439 struct ofpbuf ofpacts;
5440 struct ofpbuf *packet;
5445 /* Three kinds of error return values! */
5452 ofpbuf_init(&ofpacts, 0);
5454 /* Parse actions. */
5455 rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5457 unixctl_command_reply_error(conn, rw_error);
5462 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5463 * consistency between the flow and the actions. With -consistent, we
5464 * enforce consistency even for a flow supported in OpenFlow 1.0. */
5465 if (!strcmp(argv[1], "-consistent")) {
5466 enforce_consistency = true;
5470 enforce_consistency = false;
5473 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5475 unixctl_command_reply_error(conn, error);
5479 /* Do the same checks as handle_packet_out() in ofproto.c.
5481 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5482 * strictly correct because these actions aren't in any table, but it's OK
5483 * because it 'table_id' is used only to check goto_table instructions, but
5484 * packet-outs take a list of actions and therefore it can't include
5487 * We skip the "meter" check here because meter is an instruction, not an
5488 * action, and thus cannot appear in ofpacts. */
5489 in_port = ofp_to_u16(flow.in_port.ofp_port);
5490 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5491 unixctl_command_reply_error(conn, "invalid in_port");
5494 if (enforce_consistency) {
5495 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5496 u16_to_ofp(ofproto->up.max_ports),
5497 0, 0, usable_protocols);
5499 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5500 u16_to_ofp(ofproto->up.max_ports), 0, 0,
5506 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5507 unixctl_command_reply_error(conn, ds_cstr(&result));
5511 ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5512 unixctl_command_reply(conn, ds_cstr(&result));
5515 ds_destroy(&result);
5516 ofpbuf_delete(packet);
5517 ofpbuf_uninit(&ofpacts);
5520 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5521 * description of the results to 'ds'.
5523 * The trace follows a packet with the specified 'flow' through the flow
5524 * table. 'packet' may be nonnull to trace an actual packet, with consequent
5525 * side effects (if it is nonnull then its flow must be 'flow').
5527 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5528 * trace, otherwise the actions are determined by a flow table lookup. */
5530 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5531 const struct ofpbuf *packet,
5532 const struct ofpact ofpacts[], size_t ofpacts_len,
5535 struct rule_dpif *rule;
5536 struct flow_wildcards wc;
5538 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5539 ds_put_cstr(ds, "Flow: ");
5540 flow_format(ds, flow);
5541 ds_put_char(ds, '\n');
5543 flow_wildcards_init_catchall(&wc);
5547 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5549 trace_format_rule(ds, 0, rule);
5550 if (rule == ofproto->miss_rule) {
5551 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5552 } else if (rule == ofproto->no_packet_in_rule) {
5553 ds_put_cstr(ds, "\nNo match, packets dropped because "
5554 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5555 } else if (rule == ofproto->drop_frags_rule) {
5556 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5557 "and the fragment handling mode is \"drop\".\n");
5561 if (rule || ofpacts) {
5562 uint64_t odp_actions_stub[1024 / 8];
5563 struct ofpbuf odp_actions;
5564 struct trace_ctx trace;
5568 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5571 ofpbuf_use_stub(&odp_actions,
5572 odp_actions_stub, sizeof odp_actions_stub);
5573 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5575 trace.xin.ofpacts = ofpacts;
5576 trace.xin.ofpacts_len = ofpacts_len;
5578 trace.xin.resubmit_hook = trace_resubmit;
5579 trace.xin.report_hook = trace_report;
5581 xlate_actions(&trace.xin, &trace.xout);
5582 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5584 ds_put_char(ds, '\n');
5585 trace_format_flow(ds, 0, "Final flow", &trace);
5587 match_init(&match, flow, &trace.xout.wc);
5588 ds_put_cstr(ds, "Relevant fields: ");
5589 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5590 ds_put_char(ds, '\n');
5592 ds_put_cstr(ds, "Datapath actions: ");
5593 format_odp_actions(ds, trace.xout.odp_actions.data,
5594 trace.xout.odp_actions.size);
5596 if (trace.xout.slow) {
5597 enum slow_path_reason slow;
5599 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5600 "slow path because it:");
5602 slow = trace.xout.slow;
5604 enum slow_path_reason bit = rightmost_1bit(slow);
5606 ds_put_format(ds, "\n\t- %s.",
5607 slow_path_reason_to_explanation(bit));
5613 xlate_out_uninit(&trace.xout);
5616 rule_dpif_unref(rule);
5619 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5620 * 'reply' describing the results. */
5622 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5624 struct cls_cursor cursor;
5625 struct facet *facet;
5629 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5630 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5631 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5632 if (!facet_check_consistency(facet)) {
5636 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5638 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5642 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5643 ofproto->up.name, errors);
5645 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5650 ofproto_dpif_self_check(struct unixctl_conn *conn,
5651 int argc, const char *argv[], void *aux OVS_UNUSED)
5653 struct ds reply = DS_EMPTY_INITIALIZER;
5654 struct ofproto_dpif *ofproto;
5657 ofproto = ofproto_dpif_lookup(argv[1]);
5659 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5660 "ofproto/list for help)");
5663 ofproto_dpif_self_check__(ofproto, &reply);
5665 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5666 ofproto_dpif_self_check__(ofproto, &reply);
5670 unixctl_command_reply(conn, ds_cstr(&reply));
5674 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5675 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5676 * to destroy 'ofproto_shash' and free the returned value. */
5677 static const struct shash_node **
5678 get_ofprotos(struct shash *ofproto_shash)
5680 const struct ofproto_dpif *ofproto;
5682 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5683 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5684 shash_add_nocopy(ofproto_shash, name, ofproto);
5687 return shash_sort(ofproto_shash);
5691 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5692 const char *argv[] OVS_UNUSED,
5693 void *aux OVS_UNUSED)
5695 struct ds ds = DS_EMPTY_INITIALIZER;
5696 struct shash ofproto_shash;
5697 const struct shash_node **sorted_ofprotos;
5700 shash_init(&ofproto_shash);
5701 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5702 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5703 const struct shash_node *node = sorted_ofprotos[i];
5704 ds_put_format(&ds, "%s\n", node->name);
5707 shash_destroy(&ofproto_shash);
5708 free(sorted_ofprotos);
5710 unixctl_command_reply(conn, ds_cstr(&ds));
5715 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5717 const struct shash_node **ofprotos;
5718 struct ofproto_dpif *ofproto;
5719 struct shash ofproto_shash;
5720 uint64_t n_hit, n_missed;
5723 n_hit = n_missed = 0;
5724 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5725 if (ofproto->backer == backer) {
5726 n_missed += ofproto->n_missed;
5727 n_hit += ofproto->n_hit;
5731 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5732 dpif_name(backer->dpif), n_hit, n_missed);
5733 ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5734 " life span: %lldms\n", hmap_count(&backer->subfacets),
5735 backer->avg_n_subfacet, backer->max_n_subfacet,
5736 backer->avg_subfacet_life);
5738 shash_init(&ofproto_shash);
5739 ofprotos = get_ofprotos(&ofproto_shash);
5740 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5741 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5742 const struct shash_node **ports;
5745 if (ofproto->backer != backer) {
5749 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5750 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5752 ports = shash_sort(&ofproto->up.port_by_name);
5753 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5754 const struct shash_node *node = ports[j];
5755 struct ofport *ofport = node->data;
5757 odp_port_t odp_port;
5759 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5762 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5763 if (odp_port != ODPP_NONE) {
5764 ds_put_format(ds, "%"PRIu32":", odp_port);
5766 ds_put_cstr(ds, "none:");
5769 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5772 if (!netdev_get_config(ofport->netdev, &config)) {
5773 const struct smap_node **nodes;
5776 nodes = smap_sort(&config);
5777 for (i = 0; i < smap_count(&config); i++) {
5778 const struct smap_node *node = nodes[i];
5779 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5780 node->key, node->value);
5784 smap_destroy(&config);
5786 ds_put_char(ds, ')');
5787 ds_put_char(ds, '\n');
5791 shash_destroy(&ofproto_shash);
5796 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5797 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5799 struct ds ds = DS_EMPTY_INITIALIZER;
5800 const struct shash_node **backers;
5803 backers = shash_sort(&all_dpif_backers);
5804 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5805 dpif_show_backer(backers[i]->data, &ds);
5809 unixctl_command_reply(conn, ds_cstr(&ds));
5813 /* Dump the megaflow (facet) cache. This is useful to check the
5814 * correctness of flow wildcarding, since the same mechanism is used for
5815 * both xlate caching and kernel wildcarding.
5817 * It's important to note that in the output the flow description uses
5818 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5820 * This command is only needed for advanced debugging, so it's not
5821 * documented in the man page. */
5823 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5824 int argc OVS_UNUSED, const char *argv[],
5825 void *aux OVS_UNUSED)
5827 struct ds ds = DS_EMPTY_INITIALIZER;
5828 const struct ofproto_dpif *ofproto;
5829 long long int now = time_msec();
5830 struct cls_cursor cursor;
5831 struct facet *facet;
5833 ofproto = ofproto_dpif_lookup(argv[1]);
5835 unixctl_command_reply_error(conn, "no such bridge");
5839 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5840 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5841 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5842 cls_rule_format(&facet->cr, &ds);
5843 ds_put_cstr(&ds, ", ");
5844 ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5845 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5846 ds_put_cstr(&ds, "Datapath actions: ");
5847 if (facet->xout.slow) {
5848 uint64_t slow_path_stub[128 / 8];
5849 const struct nlattr *actions;
5852 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5853 slow_path_stub, sizeof slow_path_stub,
5854 &actions, &actions_len);
5855 format_odp_actions(&ds, actions, actions_len);
5857 format_odp_actions(&ds, facet->xout.odp_actions.data,
5858 facet->xout.odp_actions.size);
5860 ds_put_cstr(&ds, "\n");
5862 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5864 ds_chomp(&ds, '\n');
5865 unixctl_command_reply(conn, ds_cstr(&ds));
5869 /* Disable using the megaflows.
5871 * This command is only needed for advanced debugging, so it's not
5872 * documented in the man page. */
5874 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5875 int argc OVS_UNUSED,
5876 const char *argv[] OVS_UNUSED,
5877 void *aux OVS_UNUSED)
5879 struct ofproto_dpif *ofproto;
5881 enable_megaflows = false;
5883 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5884 flush(&ofproto->up);
5887 unixctl_command_reply(conn, "megaflows disabled");
5890 /* Re-enable using megaflows.
5892 * This command is only needed for advanced debugging, so it's not
5893 * documented in the man page. */
5895 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5896 int argc OVS_UNUSED,
5897 const char *argv[] OVS_UNUSED,
5898 void *aux OVS_UNUSED)
5900 struct ofproto_dpif *ofproto;
5902 enable_megaflows = true;
5904 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5905 flush(&ofproto->up);
5908 unixctl_command_reply(conn, "megaflows enabled");
5912 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5913 int argc OVS_UNUSED, const char *argv[],
5914 void *aux OVS_UNUSED)
5916 struct ds ds = DS_EMPTY_INITIALIZER;
5917 const struct ofproto_dpif *ofproto;
5918 struct subfacet *subfacet;
5920 ofproto = ofproto_dpif_lookup(argv[1]);
5922 unixctl_command_reply_error(conn, "no such bridge");
5926 update_stats(ofproto->backer);
5928 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5929 struct facet *facet = subfacet->facet;
5930 struct odputil_keybuf maskbuf;
5933 if (facet->ofproto != ofproto) {
5937 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5938 if (enable_megaflows) {
5939 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5940 &facet->flow, UINT32_MAX);
5943 odp_flow_format(subfacet->key, subfacet->key_len,
5944 mask.data, mask.size, NULL, &ds, false);
5946 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5947 subfacet->dp_packet_count, subfacet->dp_byte_count);
5948 if (subfacet->used) {
5949 ds_put_format(&ds, "%.3fs",
5950 (time_msec() - subfacet->used) / 1000.0);
5952 ds_put_format(&ds, "never");
5954 if (subfacet->facet->tcp_flags) {
5955 ds_put_cstr(&ds, ", flags:");
5956 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5959 ds_put_cstr(&ds, ", actions:");
5960 if (facet->xout.slow) {
5961 uint64_t slow_path_stub[128 / 8];
5962 const struct nlattr *actions;
5965 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5966 slow_path_stub, sizeof slow_path_stub,
5967 &actions, &actions_len);
5968 format_odp_actions(&ds, actions, actions_len);
5970 format_odp_actions(&ds, facet->xout.odp_actions.data,
5971 facet->xout.odp_actions.size);
5973 ds_put_char(&ds, '\n');
5976 unixctl_command_reply(conn, ds_cstr(&ds));
5981 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5982 int argc OVS_UNUSED, const char *argv[],
5983 void *aux OVS_UNUSED)
5985 struct ds ds = DS_EMPTY_INITIALIZER;
5986 struct ofproto_dpif *ofproto;
5988 ofproto = ofproto_dpif_lookup(argv[1]);
5990 unixctl_command_reply_error(conn, "no such bridge");
5994 flush(&ofproto->up);
5996 unixctl_command_reply(conn, ds_cstr(&ds));
6001 ofproto_dpif_unixctl_init(void)
6003 static bool registered;
6009 unixctl_command_register(
6011 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6012 1, 3, ofproto_unixctl_trace, NULL);
6013 unixctl_command_register(
6014 "ofproto/trace-packet-out",
6015 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6016 2, 6, ofproto_unixctl_trace_actions, NULL);
6017 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6018 ofproto_unixctl_fdb_flush, NULL);
6019 unixctl_command_register("fdb/show", "bridge", 1, 1,
6020 ofproto_unixctl_fdb_show, NULL);
6021 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6022 ofproto_dpif_self_check, NULL);
6023 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6024 ofproto_unixctl_dpif_dump_dps, NULL);
6025 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6027 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6028 ofproto_unixctl_dpif_dump_flows, NULL);
6029 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6030 ofproto_unixctl_dpif_del_flows, NULL);
6031 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6032 ofproto_unixctl_dpif_dump_megaflows, NULL);
6033 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6034 ofproto_unixctl_dpif_disable_megaflows, NULL);
6035 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6036 ofproto_unixctl_dpif_enable_megaflows, NULL);
6039 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6041 * This is deprecated. It is only for compatibility with broken device drivers
6042 * in old versions of Linux that do not properly support VLANs when VLAN
6043 * devices are not used. When broken device drivers are no longer in
6044 * widespread use, we will delete these interfaces. */
6047 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6050 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6052 if (realdev_ofp_port == ofport->realdev_ofp_port
6053 && vid == ofport->vlandev_vid) {
6057 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6059 if (ofport->realdev_ofp_port) {
6062 if (realdev_ofp_port && ofport->bundle) {
6063 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6064 * themselves be part of a bundle. */
6065 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6068 ofport->realdev_ofp_port = realdev_ofp_port;
6069 ofport->vlandev_vid = vid;
6071 if (realdev_ofp_port) {
6072 vsp_add(ofport, realdev_ofp_port, vid);
6079 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6081 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6085 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6086 OVS_EXCLUDED(ofproto->vsp_mutex)
6090 ovs_mutex_lock(&ofproto->vsp_mutex);
6091 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6092 ovs_mutex_unlock(&ofproto->vsp_mutex);
6097 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6098 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6099 OVS_REQUIRES(ofproto->vsp_mutex)
6101 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6102 int vid = vlan_tci_to_vid(vlan_tci);
6103 const struct vlan_splinter *vsp;
6105 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6106 hash_realdev_vid(realdev_ofp_port, vid),
6107 &ofproto->realdev_vid_map) {
6108 if (vsp->realdev_ofp_port == realdev_ofp_port
6109 && vsp->vid == vid) {
6110 return vsp->vlandev_ofp_port;
6114 return realdev_ofp_port;
6117 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6118 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6119 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6120 * 'vlan_tci' 9, it would return the port number of eth0.9.
6122 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6123 * function just returns its 'realdev_ofp_port' argument. */
6125 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6126 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6127 OVS_EXCLUDED(ofproto->vsp_mutex)
6131 ovs_mutex_lock(&ofproto->vsp_mutex);
6132 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6133 ovs_mutex_unlock(&ofproto->vsp_mutex);
6137 static struct vlan_splinter *
6138 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6140 struct vlan_splinter *vsp;
6142 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6143 hash_ofp_port(vlandev_ofp_port),
6144 &ofproto->vlandev_map) {
6145 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6153 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6154 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6155 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6156 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6157 * eth0 and store 9 in '*vid'.
6159 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6160 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6163 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6164 ofp_port_t vlandev_ofp_port, int *vid)
6165 OVS_REQUIRES(ofproto->vsp_mutex)
6167 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6168 const struct vlan_splinter *vsp;
6170 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6175 return vsp->realdev_ofp_port;
6181 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6182 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6183 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6184 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6185 * always the case unless VLAN splinters are enabled), returns false without
6186 * making any changes. */
6188 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6189 OVS_EXCLUDED(ofproto->vsp_mutex)
6194 ovs_mutex_lock(&ofproto->vsp_mutex);
6195 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6196 ovs_mutex_unlock(&ofproto->vsp_mutex);
6201 /* Cause the flow to be processed as if it came in on the real device with
6202 * the VLAN device's VLAN ID. */
6203 flow->in_port.ofp_port = realdev;
6204 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6209 vsp_remove(struct ofport_dpif *port)
6211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6212 struct vlan_splinter *vsp;
6214 ovs_mutex_lock(&ofproto->vsp_mutex);
6215 vsp = vlandev_find(ofproto, port->up.ofp_port);
6217 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6218 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6221 port->realdev_ofp_port = 0;
6223 VLOG_ERR("missing vlan device record");
6225 ovs_mutex_unlock(&ofproto->vsp_mutex);
6229 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6233 ovs_mutex_lock(&ofproto->vsp_mutex);
6234 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6235 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6236 == realdev_ofp_port)) {
6237 struct vlan_splinter *vsp;
6239 vsp = xmalloc(sizeof *vsp);
6240 vsp->realdev_ofp_port = realdev_ofp_port;
6241 vsp->vlandev_ofp_port = port->up.ofp_port;
6244 port->realdev_ofp_port = realdev_ofp_port;
6246 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6247 hash_ofp_port(port->up.ofp_port));
6248 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6249 hash_realdev_vid(realdev_ofp_port, vid));
6251 VLOG_ERR("duplicate vlan device record");
6253 ovs_mutex_unlock(&ofproto->vsp_mutex);
6257 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6259 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6260 return ofport ? ofport->odp_port : ODPP_NONE;
6263 struct ofport_dpif *
6264 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6266 struct ofport_dpif *port;
6268 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6269 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6270 &backer->odp_to_ofport_map) {
6271 if (port->odp_port == odp_port) {
6272 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6277 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6282 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6284 struct ofport_dpif *port;
6286 port = odp_port_to_ofport(ofproto->backer, odp_port);
6287 if (port && &ofproto->up == port->up.ofproto) {
6288 return port->up.ofp_port;
6294 const struct ofproto_class ofproto_dpif_class = {
6329 port_is_lacp_current,
6330 NULL, /* rule_choose_table */
6339 rule_modify_actions,
6353 get_stp_port_status,
6361 is_mirror_output_bundle,
6362 forward_bpdu_changed,
6363 set_mac_table_config,
6365 NULL, /* meter_get_features */
6366 NULL, /* meter_set */
6367 NULL, /* meter_get */
6368 NULL, /* meter_del */
6369 group_alloc, /* group_alloc */
6370 group_construct, /* group_construct */
6371 group_destruct, /* group_destruct */
6372 group_dealloc, /* group_dealloc */
6373 group_modify, /* group_modify */
6374 group_get_stats, /* group_get_stats */