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->up.frag_handling,
804 ofproto->up.forward_bpdu,
805 connmgr_has_in_band(ofproto->up.connmgr),
806 ofproto->netflow != NULL);
808 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
809 xlate_bundle_set(ofproto, bundle, bundle->name,
810 bundle->vlan_mode, bundle->vlan,
811 bundle->trunks, bundle->use_priority_tags,
812 bundle->bond, bundle->lacp,
816 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
817 int stp_port = ofport->stp_port
818 ? stp_port_no(ofport->stp_port)
820 xlate_ofport_set(ofproto, ofport->bundle, ofport,
821 ofport->up.ofp_port, ofport->odp_port,
822 ofport->up.netdev, ofport->cfm,
823 ofport->bfd, ofport->peer, stp_port,
824 ofport->qdscp, ofport->n_qdscp,
825 ofport->up.pp.config, ofport->up.pp.state,
826 ofport->is_tunnel, ofport->may_enable);
828 ovs_rwlock_unlock(&xlate_rwlock);
830 /* Only ofproto-dpif cares about the facet classifier so we just
831 * lock cls_cursor_init() to appease the thread safety analysis. */
832 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
833 cls_cursor_init(&cursor, &ofproto->facets, NULL);
834 ovs_rwlock_unlock(&ofproto->facets.rwlock);
835 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
836 facet_revalidate(facet);
841 udpif_revalidate(backer->udpif);
844 if (!backer->recv_set_enable) {
845 /* Wake up before a max of 1000ms. */
846 timer_set_duration(&backer->next_expiration, 1000);
847 } else if (timer_expired(&backer->next_expiration)) {
848 int delay = expire(backer);
849 timer_set_duration(&backer->next_expiration, delay);
852 process_dpif_port_changes(backer);
854 if (backer->governor) {
857 governor_run(backer->governor);
859 /* If the governor has shrunk to its minimum size and the number of
860 * subfacets has dwindled, then drop the governor entirely.
862 * For hysteresis, the number of subfacets to drop the governor is
863 * smaller than the number needed to trigger its creation. */
864 n_subfacets = hmap_count(&backer->subfacets);
865 if (n_subfacets * 4 < flow_eviction_threshold
866 && governor_is_idle(backer->governor)) {
867 governor_destroy(backer->governor);
868 backer->governor = NULL;
875 /* Check for and handle port changes in 'backer''s dpif. */
877 process_dpif_port_changes(struct dpif_backer *backer)
883 error = dpif_port_poll(backer->dpif, &devname);
889 process_dpif_all_ports_changed(backer);
893 process_dpif_port_change(backer, devname);
898 process_dpif_port_error(backer, error);
905 process_dpif_all_ports_changed(struct dpif_backer *backer)
907 struct ofproto_dpif *ofproto;
908 struct dpif_port dpif_port;
909 struct dpif_port_dump dump;
910 struct sset devnames;
913 sset_init(&devnames);
914 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
915 if (ofproto->backer == backer) {
916 struct ofport *ofport;
918 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
919 sset_add(&devnames, netdev_get_name(ofport->netdev));
923 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
924 sset_add(&devnames, dpif_port.name);
927 SSET_FOR_EACH (devname, &devnames) {
928 process_dpif_port_change(backer, devname);
930 sset_destroy(&devnames);
934 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
936 struct ofproto_dpif *ofproto;
937 struct dpif_port port;
939 /* Don't report on the datapath's device. */
940 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
944 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
945 &all_ofproto_dpifs) {
946 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
951 ofproto = lookup_ofproto_dpif_by_port_name(devname);
952 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
953 /* The port was removed. If we know the datapath,
954 * report it through poll_set(). If we don't, it may be
955 * notifying us of a removal we initiated, so ignore it.
956 * If there's a pending ENOBUFS, let it stand, since
957 * everything will be reevaluated. */
958 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
959 sset_add(&ofproto->port_poll_set, devname);
960 ofproto->port_poll_errno = 0;
962 } else if (!ofproto) {
963 /* The port was added, but we don't know with which
964 * ofproto we should associate it. Delete it. */
965 dpif_port_del(backer->dpif, port.port_no);
967 struct ofport_dpif *ofport;
969 ofport = ofport_dpif_cast(shash_find_data(
970 &ofproto->up.port_by_name, devname));
972 && ofport->odp_port != port.port_no
973 && !odp_port_to_ofport(backer, port.port_no))
975 /* 'ofport''s datapath port number has changed from
976 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
977 * structures to match. */
978 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
979 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
980 ofport->odp_port = port.port_no;
981 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
982 hash_odp_port(port.port_no));
983 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
984 backer->need_revalidate = REV_RECONFIGURE;
987 dpif_port_destroy(&port);
990 /* Propagate 'error' to all ofprotos based on 'backer'. */
992 process_dpif_port_error(struct dpif_backer *backer, int error)
994 struct ofproto_dpif *ofproto;
996 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
997 if (ofproto->backer == backer) {
998 sset_clear(&ofproto->port_poll_set);
999 ofproto->port_poll_errno = error;
1005 dpif_backer_run_fast(struct dpif_backer *backer)
1007 handle_upcalls(backer);
1013 type_run_fast(const char *type)
1015 struct dpif_backer *backer;
1017 backer = shash_find_data(&all_dpif_backers, type);
1019 /* This is not necessarily a problem, since backers are only
1020 * created on demand. */
1024 return dpif_backer_run_fast(backer);
1030 static long long int port_rl = LLONG_MIN;
1032 if (time_msec() >= port_rl) {
1033 struct ofproto_dpif *ofproto;
1035 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1036 run_fast(&ofproto->up);
1038 port_rl = time_msec() + 200;
1043 type_wait(const char *type)
1045 struct dpif_backer *backer;
1047 backer = shash_find_data(&all_dpif_backers, type);
1049 /* This is not necessarily a problem, since backers are only
1050 * created on demand. */
1054 if (backer->governor) {
1055 governor_wait(backer->governor);
1058 timer_wait(&backer->next_expiration);
1059 dpif_wait(backer->dpif);
1060 udpif_wait(backer->udpif);
1063 /* Basic life-cycle. */
1065 static int add_internal_flows(struct ofproto_dpif *);
1067 static struct ofproto *
1070 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1071 return &ofproto->up;
1075 dealloc(struct ofproto *ofproto_)
1077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1082 close_dpif_backer(struct dpif_backer *backer)
1084 ovs_assert(backer->refcount > 0);
1086 if (--backer->refcount) {
1090 drop_key_clear(backer);
1091 hmap_destroy(&backer->drop_keys);
1093 udpif_destroy(backer->udpif);
1095 simap_destroy(&backer->tnl_backers);
1096 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1097 hmap_destroy(&backer->odp_to_ofport_map);
1098 shash_find_and_delete(&all_dpif_backers, backer->type);
1100 dpif_close(backer->dpif);
1102 ovs_assert(hmap_is_empty(&backer->subfacets));
1103 hmap_destroy(&backer->subfacets);
1104 governor_destroy(backer->governor);
1109 /* Datapath port slated for removal from datapath. */
1110 struct odp_garbage {
1111 struct list list_node;
1112 odp_port_t odp_port;
1116 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1118 struct dpif_backer *backer;
1119 struct dpif_port_dump port_dump;
1120 struct dpif_port port;
1121 struct shash_node *node;
1122 struct list garbage_list;
1123 struct odp_garbage *garbage, *next;
1129 backer = shash_find_data(&all_dpif_backers, type);
1136 backer_name = xasprintf("ovs-%s", type);
1138 /* Remove any existing datapaths, since we assume we're the only
1139 * userspace controlling the datapath. */
1141 dp_enumerate_names(type, &names);
1142 SSET_FOR_EACH(name, &names) {
1143 struct dpif *old_dpif;
1145 /* Don't remove our backer if it exists. */
1146 if (!strcmp(name, backer_name)) {
1150 if (dpif_open(name, type, &old_dpif)) {
1151 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1153 dpif_delete(old_dpif);
1154 dpif_close(old_dpif);
1157 sset_destroy(&names);
1159 backer = xmalloc(sizeof *backer);
1161 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1164 VLOG_ERR("failed to open datapath of type %s: %s", type,
1165 ovs_strerror(error));
1169 backer->udpif = udpif_create(backer, backer->dpif);
1171 backer->type = xstrdup(type);
1172 backer->governor = NULL;
1173 backer->refcount = 1;
1174 hmap_init(&backer->odp_to_ofport_map);
1175 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1176 hmap_init(&backer->drop_keys);
1177 hmap_init(&backer->subfacets);
1178 timer_set_duration(&backer->next_expiration, 1000);
1179 backer->need_revalidate = 0;
1180 simap_init(&backer->tnl_backers);
1181 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1184 if (backer->recv_set_enable) {
1185 dpif_flow_flush(backer->dpif);
1188 /* Loop through the ports already on the datapath and remove any
1189 * that we don't need anymore. */
1190 list_init(&garbage_list);
1191 dpif_port_dump_start(&port_dump, backer->dpif);
1192 while (dpif_port_dump_next(&port_dump, &port)) {
1193 node = shash_find(&init_ofp_ports, port.name);
1194 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1195 garbage = xmalloc(sizeof *garbage);
1196 garbage->odp_port = port.port_no;
1197 list_push_front(&garbage_list, &garbage->list_node);
1200 dpif_port_dump_done(&port_dump);
1202 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1203 dpif_port_del(backer->dpif, garbage->odp_port);
1204 list_remove(&garbage->list_node);
1208 shash_add(&all_dpif_backers, type, backer);
1210 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1212 VLOG_ERR("failed to listen on datapath of type %s: %s",
1213 type, ovs_strerror(error));
1214 close_dpif_backer(backer);
1217 udpif_recv_set(backer->udpif, n_handler_threads,
1218 backer->recv_set_enable);
1219 backer->n_handler_threads = n_handler_threads;
1221 backer->max_n_subfacet = 0;
1222 backer->avg_n_subfacet = 0;
1223 backer->avg_subfacet_life = 0;
1229 construct(struct ofproto *ofproto_)
1231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1232 struct shash_node *node, *next;
1235 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1240 ofproto->netflow = NULL;
1241 ofproto->sflow = NULL;
1242 ofproto->ipfix = NULL;
1243 ofproto->stp = NULL;
1244 hmap_init(&ofproto->bundles);
1245 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1246 ofproto->mbridge = mbridge_create();
1247 ofproto->has_bonded_bundles = false;
1248 ofproto->lacp_enabled = false;
1249 ovs_mutex_init(&ofproto->stats_mutex);
1250 ovs_mutex_init(&ofproto->vsp_mutex);
1252 classifier_init(&ofproto->facets, NULL);
1253 ofproto->consistency_rl = LLONG_MIN;
1255 guarded_list_init(&ofproto->pins);
1257 ofproto_dpif_unixctl_init();
1259 hmap_init(&ofproto->vlandev_map);
1260 hmap_init(&ofproto->realdev_vid_map);
1262 sset_init(&ofproto->ports);
1263 sset_init(&ofproto->ghost_ports);
1264 sset_init(&ofproto->port_poll_set);
1265 ofproto->port_poll_errno = 0;
1266 ofproto->change_seq = 0;
1268 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1269 struct iface_hint *iface_hint = node->data;
1271 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1272 /* Check if the datapath already has this port. */
1273 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1274 sset_add(&ofproto->ports, node->name);
1277 free(iface_hint->br_name);
1278 free(iface_hint->br_type);
1280 shash_delete(&init_ofp_ports, node);
1284 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1285 hash_string(ofproto->up.name, 0));
1286 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1288 ofproto_init_tables(ofproto_, N_TABLES);
1289 error = add_internal_flows(ofproto);
1290 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1293 ofproto->n_missed = 0;
1299 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1300 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1302 struct ofputil_flow_mod fm;
1305 match_init_catchall(&fm.match);
1307 match_set_reg(&fm.match, 0, id);
1308 fm.new_cookie = htonll(0);
1309 fm.cookie = htonll(0);
1310 fm.cookie_mask = htonll(0);
1311 fm.modify_cookie = false;
1312 fm.table_id = TBL_INTERNAL;
1313 fm.command = OFPFC_ADD;
1314 fm.idle_timeout = 0;
1315 fm.hard_timeout = 0;
1319 fm.ofpacts = ofpacts->data;
1320 fm.ofpacts_len = ofpacts->size;
1322 error = ofproto_flow_mod(&ofproto->up, &fm);
1324 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1325 id, ofperr_to_string(error));
1329 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1331 rule_dpif_unref(*rulep);
1340 add_internal_flows(struct ofproto_dpif *ofproto)
1342 struct ofpact_controller *controller;
1343 uint64_t ofpacts_stub[128 / 8];
1344 struct ofpbuf ofpacts;
1348 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1351 controller = ofpact_put_CONTROLLER(&ofpacts);
1352 controller->max_len = UINT16_MAX;
1353 controller->controller_id = 0;
1354 controller->reason = OFPR_NO_MATCH;
1355 ofpact_pad(&ofpacts);
1357 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1362 ofpbuf_clear(&ofpacts);
1363 error = add_internal_flow(ofproto, id++, &ofpacts,
1364 &ofproto->no_packet_in_rule);
1369 error = add_internal_flow(ofproto, id++, &ofpacts,
1370 &ofproto->drop_frags_rule);
1375 destruct(struct ofproto *ofproto_)
1377 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1378 struct rule_dpif *rule, *next_rule;
1379 struct ofproto_packet_in *pin, *next_pin;
1380 struct facet *facet, *next_facet;
1381 struct cls_cursor cursor;
1382 struct oftable *table;
1385 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1386 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1387 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1388 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1389 facet_remove(facet);
1392 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1393 ovs_rwlock_wrlock(&xlate_rwlock);
1394 xlate_remove_ofproto(ofproto);
1395 ovs_rwlock_unlock(&xlate_rwlock);
1397 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1398 * use-after-free error. */
1399 udpif_revalidate(ofproto->backer->udpif);
1401 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1403 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1404 struct cls_cursor cursor;
1406 ovs_rwlock_rdlock(&table->cls.rwlock);
1407 cls_cursor_init(&cursor, &table->cls, NULL);
1408 ovs_rwlock_unlock(&table->cls.rwlock);
1409 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1410 ofproto_rule_delete(&ofproto->up, &rule->up);
1414 guarded_list_pop_all(&ofproto->pins, &pins);
1415 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1416 list_remove(&pin->list_node);
1417 free(CONST_CAST(void *, pin->up.packet));
1420 guarded_list_destroy(&ofproto->pins);
1422 mbridge_unref(ofproto->mbridge);
1424 netflow_destroy(ofproto->netflow);
1425 dpif_sflow_unref(ofproto->sflow);
1426 hmap_destroy(&ofproto->bundles);
1427 mac_learning_unref(ofproto->ml);
1429 classifier_destroy(&ofproto->facets);
1431 hmap_destroy(&ofproto->vlandev_map);
1432 hmap_destroy(&ofproto->realdev_vid_map);
1434 sset_destroy(&ofproto->ports);
1435 sset_destroy(&ofproto->ghost_ports);
1436 sset_destroy(&ofproto->port_poll_set);
1438 ovs_mutex_destroy(&ofproto->stats_mutex);
1439 ovs_mutex_destroy(&ofproto->vsp_mutex);
1441 close_dpif_backer(ofproto->backer);
1445 run_fast(struct ofproto *ofproto_)
1447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1448 struct ofproto_packet_in *pin, *next_pin;
1451 /* Do not perform any periodic activity required by 'ofproto' while
1452 * waiting for flow restore to complete. */
1453 if (ofproto_get_flow_restore_wait()) {
1457 guarded_list_pop_all(&ofproto->pins, &pins);
1458 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1459 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1460 list_remove(&pin->list_node);
1461 free(CONST_CAST(void *, pin->up.packet));
1469 run(struct ofproto *ofproto_)
1471 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1475 if (mbridge_need_revalidate(ofproto->mbridge)) {
1476 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1477 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1478 mac_learning_flush(ofproto->ml);
1479 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1482 /* Do not perform any periodic activity below required by 'ofproto' while
1483 * waiting for flow restore to complete. */
1484 if (ofproto_get_flow_restore_wait()) {
1488 error = run_fast(ofproto_);
1493 if (ofproto->netflow) {
1494 netflow_run(ofproto->netflow);
1496 if (ofproto->sflow) {
1497 dpif_sflow_run(ofproto->sflow);
1499 if (ofproto->ipfix) {
1500 dpif_ipfix_run(ofproto->ipfix);
1503 new_seq = seq_read(connectivity_seq_get());
1504 if (ofproto->change_seq != new_seq) {
1505 struct ofport_dpif *ofport;
1507 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1511 ofproto->change_seq = new_seq;
1513 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1514 struct ofbundle *bundle;
1516 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1522 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1523 if (mac_learning_run(ofproto->ml)) {
1524 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1526 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1528 /* Check the consistency of a random facet, to aid debugging. */
1529 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1530 if (time_msec() >= ofproto->consistency_rl
1531 && !classifier_is_empty(&ofproto->facets)
1532 && !ofproto->backer->need_revalidate) {
1533 struct cls_subtable *table;
1534 struct cls_rule *cr;
1535 struct facet *facet;
1537 ofproto->consistency_rl = time_msec() + 250;
1539 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1540 struct cls_subtable, hmap_node);
1541 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1543 facet = CONTAINER_OF(cr, struct facet, cr);
1545 if (!facet_check_consistency(facet)) {
1546 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1549 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1555 wait(struct ofproto *ofproto_)
1557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1559 if (ofproto_get_flow_restore_wait()) {
1563 if (ofproto->sflow) {
1564 dpif_sflow_wait(ofproto->sflow);
1566 if (ofproto->ipfix) {
1567 dpif_ipfix_wait(ofproto->ipfix);
1569 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1570 struct ofbundle *bundle;
1572 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1573 bundle_wait(bundle);
1576 if (ofproto->netflow) {
1577 netflow_wait(ofproto->netflow);
1579 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1580 mac_learning_wait(ofproto->ml);
1581 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1583 if (ofproto->backer->need_revalidate) {
1584 /* Shouldn't happen, but if it does just go around again. */
1585 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1586 poll_immediate_wake();
1591 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1593 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1594 struct cls_cursor cursor;
1595 size_t n_subfacets = 0;
1596 struct facet *facet;
1598 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1599 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1600 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1602 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1603 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1604 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1605 n_subfacets += list_size(&facet->subfacets);
1607 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1608 simap_increase(usage, "subfacets", n_subfacets);
1612 flush(struct ofproto *ofproto_)
1614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1615 struct subfacet *subfacet, *next_subfacet;
1616 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1620 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1621 &ofproto->backer->subfacets) {
1622 if (subfacet->facet->ofproto != ofproto) {
1626 if (subfacet->path != SF_NOT_INSTALLED) {
1627 batch[n_batch++] = subfacet;
1628 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1629 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1633 subfacet_destroy(subfacet);
1638 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1643 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1644 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1646 *arp_match_ip = true;
1647 *actions = (OFPUTIL_A_OUTPUT |
1648 OFPUTIL_A_SET_VLAN_VID |
1649 OFPUTIL_A_SET_VLAN_PCP |
1650 OFPUTIL_A_STRIP_VLAN |
1651 OFPUTIL_A_SET_DL_SRC |
1652 OFPUTIL_A_SET_DL_DST |
1653 OFPUTIL_A_SET_NW_SRC |
1654 OFPUTIL_A_SET_NW_DST |
1655 OFPUTIL_A_SET_NW_TOS |
1656 OFPUTIL_A_SET_TP_SRC |
1657 OFPUTIL_A_SET_TP_DST |
1662 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1665 struct dpif_dp_stats s;
1666 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1669 strcpy(ots->name, "classifier");
1671 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1672 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1673 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1674 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1676 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1677 ots->lookup_count = htonll(n_lookup);
1678 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1681 static struct ofport *
1684 struct ofport_dpif *port = xmalloc(sizeof *port);
1689 port_dealloc(struct ofport *port_)
1691 struct ofport_dpif *port = ofport_dpif_cast(port_);
1696 port_construct(struct ofport *port_)
1698 struct ofport_dpif *port = ofport_dpif_cast(port_);
1699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1700 const struct netdev *netdev = port->up.netdev;
1701 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1702 struct dpif_port dpif_port;
1705 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1706 port->bundle = NULL;
1709 port->may_enable = true;
1710 port->stp_port = NULL;
1711 port->stp_state = STP_DISABLED;
1712 port->is_tunnel = false;
1716 port->realdev_ofp_port = 0;
1717 port->vlandev_vid = 0;
1718 port->carrier_seq = netdev_get_carrier_resets(netdev);
1719 port->is_layer3 = netdev_vport_is_layer3(netdev);
1721 if (netdev_vport_is_patch(netdev)) {
1722 /* By bailing out here, we don't submit the port to the sFlow module
1723 * to be considered for counter polling export. This is correct
1724 * because the patch port represents an interface that sFlow considers
1725 * to be "internal" to the switch as a whole, and therefore not an
1726 * candidate for counter polling. */
1727 port->odp_port = ODPP_NONE;
1728 ofport_update_peer(port);
1732 error = dpif_port_query_by_name(ofproto->backer->dpif,
1733 netdev_vport_get_dpif_port(netdev, namebuf,
1740 port->odp_port = dpif_port.port_no;
1742 if (netdev_get_tunnel_config(netdev)) {
1743 tnl_port_add(port, port->up.netdev, port->odp_port);
1744 port->is_tunnel = true;
1746 /* Sanity-check that a mapping doesn't already exist. This
1747 * shouldn't happen for non-tunnel ports. */
1748 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1749 VLOG_ERR("port %s already has an OpenFlow port number",
1751 dpif_port_destroy(&dpif_port);
1755 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1756 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1757 hash_odp_port(port->odp_port));
1758 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1760 dpif_port_destroy(&dpif_port);
1762 if (ofproto->sflow) {
1763 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1770 port_destruct(struct ofport *port_)
1772 struct ofport_dpif *port = ofport_dpif_cast(port_);
1773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1774 const char *devname = netdev_get_name(port->up.netdev);
1775 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1776 const char *dp_port_name;
1778 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1779 ovs_rwlock_wrlock(&xlate_rwlock);
1780 xlate_ofport_remove(port);
1781 ovs_rwlock_unlock(&xlate_rwlock);
1783 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1785 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1786 /* The underlying device is still there, so delete it. This
1787 * happens when the ofproto is being destroyed, since the caller
1788 * assumes that removal of attached ports will happen as part of
1790 if (!port->is_tunnel) {
1791 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1796 port->peer->peer = NULL;
1800 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1801 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1802 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1803 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1807 sset_find_and_delete(&ofproto->ports, devname);
1808 sset_find_and_delete(&ofproto->ghost_ports, devname);
1809 bundle_remove(port_);
1810 set_cfm(port_, NULL);
1811 set_bfd(port_, NULL);
1812 if (ofproto->sflow) {
1813 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1820 port_modified(struct ofport *port_)
1822 struct ofport_dpif *port = ofport_dpif_cast(port_);
1824 if (port->bundle && port->bundle->bond) {
1825 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1829 cfm_set_netdev(port->cfm, port->up.netdev);
1833 bfd_set_netdev(port->bfd, port->up.netdev);
1836 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1837 port->up.pp.hw_addr);
1839 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1841 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1845 ofport_update_peer(port);
1849 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1851 struct ofport_dpif *port = ofport_dpif_cast(port_);
1852 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1853 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1855 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1856 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1857 OFPUTIL_PC_NO_PACKET_IN)) {
1858 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1861 bundle_update(port->bundle);
1867 set_sflow(struct ofproto *ofproto_,
1868 const struct ofproto_sflow_options *sflow_options)
1870 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1871 struct dpif_sflow *ds = ofproto->sflow;
1873 if (sflow_options) {
1875 struct ofport_dpif *ofport;
1877 ds = ofproto->sflow = dpif_sflow_create();
1878 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1879 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1881 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883 dpif_sflow_set_options(ds, sflow_options);
1886 dpif_sflow_unref(ds);
1887 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1888 ofproto->sflow = NULL;
1896 struct ofproto *ofproto_,
1897 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1898 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1899 size_t n_flow_exporters_options)
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1902 struct dpif_ipfix *di = ofproto->ipfix;
1903 bool has_options = bridge_exporter_options || flow_exporters_options;
1905 if (has_options && !di) {
1906 di = ofproto->ipfix = dpif_ipfix_create();
1910 /* Call set_options in any case to cleanly flush the flow
1911 * caches in the last exporters that are to be destroyed. */
1912 dpif_ipfix_set_options(
1913 di, bridge_exporter_options, flow_exporters_options,
1914 n_flow_exporters_options);
1917 dpif_ipfix_unref(di);
1918 ofproto->ipfix = NULL;
1926 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1928 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1933 struct ofproto_dpif *ofproto;
1935 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1936 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1937 ofport->cfm = cfm_create(ofport->up.netdev);
1940 if (cfm_configure(ofport->cfm, s)) {
1947 cfm_unref(ofport->cfm);
1950 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1951 ofport->up.pp.hw_addr);
1956 get_cfm_status(const struct ofport *ofport_,
1957 struct ofproto_cfm_status *status)
1959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1962 status->faults = cfm_get_fault(ofport->cfm);
1963 status->flap_count = cfm_get_flap_count(ofport->cfm);
1964 status->remote_opstate = cfm_get_opup(ofport->cfm);
1965 status->health = cfm_get_health(ofport->cfm);
1966 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1974 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1976 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1977 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1981 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1982 cfg, ofport->up.netdev);
1983 if (ofport->bfd != old) {
1984 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1986 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1987 ofport->up.pp.hw_addr);
1992 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1994 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997 bfd_get_status(ofport->bfd, smap);
2004 /* Spanning Tree. */
2007 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2009 struct ofproto_dpif *ofproto = ofproto_;
2010 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2011 struct ofport_dpif *ofport;
2013 ofport = stp_port_get_aux(sp);
2015 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2016 ofproto->up.name, port_num);
2018 struct eth_header *eth = pkt->l2;
2020 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2021 if (eth_addr_is_zero(eth->eth_src)) {
2022 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2023 "with unknown MAC", ofproto->up.name, port_num);
2025 ofproto_dpif_send_packet(ofport, pkt);
2031 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2033 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2037 /* Only revalidate flows if the configuration changed. */
2038 if (!s != !ofproto->stp) {
2039 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043 if (!ofproto->stp) {
2044 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2045 send_bpdu_cb, ofproto);
2046 ofproto->stp_last_tick = time_msec();
2049 stp_set_bridge_id(ofproto->stp, s->system_id);
2050 stp_set_bridge_priority(ofproto->stp, s->priority);
2051 stp_set_hello_time(ofproto->stp, s->hello_time);
2052 stp_set_max_age(ofproto->stp, s->max_age);
2053 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2055 struct ofport *ofport;
2057 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2058 set_stp_port(ofport, NULL);
2061 stp_unref(ofproto->stp);
2062 ofproto->stp = NULL;
2069 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2075 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2076 s->designated_root = stp_get_designated_root(ofproto->stp);
2077 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2086 update_stp_port_state(struct ofport_dpif *ofport)
2088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2089 enum stp_state state;
2091 /* Figure out new state. */
2092 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2096 if (ofport->stp_state != state) {
2097 enum ofputil_port_state of_state;
2100 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2101 netdev_get_name(ofport->up.netdev),
2102 stp_state_name(ofport->stp_state),
2103 stp_state_name(state));
2104 if (stp_learn_in_state(ofport->stp_state)
2105 != stp_learn_in_state(state)) {
2106 /* xxx Learning action flows should also be flushed. */
2107 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2108 mac_learning_flush(ofproto->ml);
2109 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2111 fwd_change = stp_forward_in_state(ofport->stp_state)
2112 != stp_forward_in_state(state);
2114 ofproto->backer->need_revalidate = REV_STP;
2115 ofport->stp_state = state;
2116 ofport->stp_state_entered = time_msec();
2118 if (fwd_change && ofport->bundle) {
2119 bundle_update(ofport->bundle);
2122 /* Update the STP state bits in the OpenFlow port description. */
2123 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2124 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2125 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2126 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2127 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2129 ofproto_port_set_state(&ofport->up, of_state);
2133 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2134 * caller is responsible for assigning STP port numbers and ensuring
2135 * there are no duplicates. */
2137 set_stp_port(struct ofport *ofport_,
2138 const struct ofproto_port_stp_settings *s)
2140 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2142 struct stp_port *sp = ofport->stp_port;
2144 if (!s || !s->enable) {
2146 ofport->stp_port = NULL;
2147 stp_port_disable(sp);
2148 update_stp_port_state(ofport);
2151 } else if (sp && stp_port_no(sp) != s->port_num
2152 && ofport == stp_port_get_aux(sp)) {
2153 /* The port-id changed, so disable the old one if it's not
2154 * already in use by another port. */
2155 stp_port_disable(sp);
2158 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2159 stp_port_enable(sp);
2161 stp_port_set_aux(sp, ofport);
2162 stp_port_set_priority(sp, s->priority);
2163 stp_port_set_path_cost(sp, s->path_cost);
2165 update_stp_port_state(ofport);
2171 get_stp_port_status(struct ofport *ofport_,
2172 struct ofproto_port_stp_status *s)
2174 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2176 struct stp_port *sp = ofport->stp_port;
2178 if (!ofproto->stp || !sp) {
2184 s->port_id = stp_port_get_id(sp);
2185 s->state = stp_port_get_state(sp);
2186 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2187 s->role = stp_port_get_role(sp);
2193 get_stp_port_stats(struct ofport *ofport_,
2194 struct ofproto_port_stp_stats *s)
2196 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2198 struct stp_port *sp = ofport->stp_port;
2200 if (!ofproto->stp || !sp) {
2206 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2212 stp_run(struct ofproto_dpif *ofproto)
2215 long long int now = time_msec();
2216 long long int elapsed = now - ofproto->stp_last_tick;
2217 struct stp_port *sp;
2220 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2221 ofproto->stp_last_tick = now;
2223 while (stp_get_changed_port(ofproto->stp, &sp)) {
2224 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2227 update_stp_port_state(ofport);
2231 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2232 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2233 mac_learning_flush(ofproto->ml);
2234 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2240 stp_wait(struct ofproto_dpif *ofproto)
2243 poll_timer_wait(1000);
2248 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2251 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2254 if (ofport->n_qdscp != n_qdscp
2255 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2256 n_qdscp * sizeof *qdscp))) {
2257 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2258 free(ofport->qdscp);
2259 ofport->qdscp = n_qdscp
2260 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2262 ofport->n_qdscp = n_qdscp;
2270 /* Expires all MAC learning entries associated with 'bundle' and forces its
2271 * ofproto to revalidate every flow.
2273 * Normally MAC learning entries are removed only from the ofproto associated
2274 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2275 * are removed from every ofproto. When patch ports and SLB bonds are in use
2276 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2277 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2278 * with the host from which it migrated. */
2280 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2282 struct ofproto_dpif *ofproto = bundle->ofproto;
2283 struct mac_learning *ml = ofproto->ml;
2284 struct mac_entry *mac, *next_mac;
2286 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2287 ovs_rwlock_wrlock(&ml->rwlock);
2288 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2289 if (mac->port.p == bundle) {
2291 struct ofproto_dpif *o;
2293 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2295 struct mac_entry *e;
2297 ovs_rwlock_wrlock(&o->ml->rwlock);
2298 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2300 mac_learning_expire(o->ml, e);
2302 ovs_rwlock_unlock(&o->ml->rwlock);
2307 mac_learning_expire(ml, mac);
2310 ovs_rwlock_unlock(&ml->rwlock);
2313 static struct ofbundle *
2314 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2316 struct ofbundle *bundle;
2318 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2319 &ofproto->bundles) {
2320 if (bundle->aux == aux) {
2328 bundle_update(struct ofbundle *bundle)
2330 struct ofport_dpif *port;
2332 bundle->floodable = true;
2333 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2334 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2336 || !stp_forward_in_state(port->stp_state)) {
2337 bundle->floodable = false;
2344 bundle_del_port(struct ofport_dpif *port)
2346 struct ofbundle *bundle = port->bundle;
2348 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2350 list_remove(&port->bundle_node);
2351 port->bundle = NULL;
2354 lacp_slave_unregister(bundle->lacp, port);
2357 bond_slave_unregister(bundle->bond, port);
2360 bundle_update(bundle);
2364 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2365 struct lacp_slave_settings *lacp)
2367 struct ofport_dpif *port;
2369 port = get_ofp_port(bundle->ofproto, ofp_port);
2374 if (port->bundle != bundle) {
2375 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2377 bundle_remove(&port->up);
2380 port->bundle = bundle;
2381 list_push_back(&bundle->ports, &port->bundle_node);
2382 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2384 || !stp_forward_in_state(port->stp_state)) {
2385 bundle->floodable = false;
2389 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2390 lacp_slave_register(bundle->lacp, port, lacp);
2397 bundle_destroy(struct ofbundle *bundle)
2399 struct ofproto_dpif *ofproto;
2400 struct ofport_dpif *port, *next_port;
2406 ofproto = bundle->ofproto;
2407 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2409 ovs_rwlock_wrlock(&xlate_rwlock);
2410 xlate_bundle_remove(bundle);
2411 ovs_rwlock_unlock(&xlate_rwlock);
2413 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2414 bundle_del_port(port);
2417 bundle_flush_macs(bundle, true);
2418 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2420 free(bundle->trunks);
2421 lacp_unref(bundle->lacp);
2422 bond_unref(bundle->bond);
2427 bundle_set(struct ofproto *ofproto_, void *aux,
2428 const struct ofproto_bundle_settings *s)
2430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2431 bool need_flush = false;
2432 struct ofport_dpif *port;
2433 struct ofbundle *bundle;
2434 unsigned long *trunks;
2440 bundle_destroy(bundle_lookup(ofproto, aux));
2444 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2445 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2447 bundle = bundle_lookup(ofproto, aux);
2449 bundle = xmalloc(sizeof *bundle);
2451 bundle->ofproto = ofproto;
2452 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2453 hash_pointer(aux, 0));
2455 bundle->name = NULL;
2457 list_init(&bundle->ports);
2458 bundle->vlan_mode = PORT_VLAN_TRUNK;
2460 bundle->trunks = NULL;
2461 bundle->use_priority_tags = s->use_priority_tags;
2462 bundle->lacp = NULL;
2463 bundle->bond = NULL;
2465 bundle->floodable = true;
2466 mbridge_register_bundle(ofproto->mbridge, bundle);
2469 if (!bundle->name || strcmp(s->name, bundle->name)) {
2471 bundle->name = xstrdup(s->name);
2476 ofproto->lacp_enabled = true;
2477 if (!bundle->lacp) {
2478 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2479 bundle->lacp = lacp_create();
2481 lacp_configure(bundle->lacp, s->lacp);
2483 lacp_unref(bundle->lacp);
2484 bundle->lacp = NULL;
2487 /* Update set of ports. */
2489 for (i = 0; i < s->n_slaves; i++) {
2490 if (!bundle_add_port(bundle, s->slaves[i],
2491 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2495 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2496 struct ofport_dpif *next_port;
2498 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2499 for (i = 0; i < s->n_slaves; i++) {
2500 if (s->slaves[i] == port->up.ofp_port) {
2505 bundle_del_port(port);
2509 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2511 if (list_is_empty(&bundle->ports)) {
2512 bundle_destroy(bundle);
2516 /* Set VLAN tagging mode */
2517 if (s->vlan_mode != bundle->vlan_mode
2518 || s->use_priority_tags != bundle->use_priority_tags) {
2519 bundle->vlan_mode = s->vlan_mode;
2520 bundle->use_priority_tags = s->use_priority_tags;
2525 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2526 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2528 if (vlan != bundle->vlan) {
2529 bundle->vlan = vlan;
2533 /* Get trunked VLANs. */
2534 switch (s->vlan_mode) {
2535 case PORT_VLAN_ACCESS:
2539 case PORT_VLAN_TRUNK:
2540 trunks = CONST_CAST(unsigned long *, s->trunks);
2543 case PORT_VLAN_NATIVE_UNTAGGED:
2544 case PORT_VLAN_NATIVE_TAGGED:
2545 if (vlan != 0 && (!s->trunks
2546 || !bitmap_is_set(s->trunks, vlan)
2547 || bitmap_is_set(s->trunks, 0))) {
2548 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2550 trunks = bitmap_clone(s->trunks, 4096);
2552 trunks = bitmap_allocate1(4096);
2554 bitmap_set1(trunks, vlan);
2555 bitmap_set0(trunks, 0);
2557 trunks = CONST_CAST(unsigned long *, s->trunks);
2564 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2565 free(bundle->trunks);
2566 if (trunks == s->trunks) {
2567 bundle->trunks = vlan_bitmap_clone(trunks);
2569 bundle->trunks = trunks;
2574 if (trunks != s->trunks) {
2579 if (!list_is_short(&bundle->ports)) {
2580 bundle->ofproto->has_bonded_bundles = true;
2582 if (bond_reconfigure(bundle->bond, s->bond)) {
2583 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2586 bundle->bond = bond_create(s->bond);
2587 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2590 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2591 bond_slave_register(bundle->bond, port, port->up.netdev);
2594 bond_unref(bundle->bond);
2595 bundle->bond = NULL;
2598 /* If we changed something that would affect MAC learning, un-learn
2599 * everything on this port and force flow revalidation. */
2601 bundle_flush_macs(bundle, false);
2608 bundle_remove(struct ofport *port_)
2610 struct ofport_dpif *port = ofport_dpif_cast(port_);
2611 struct ofbundle *bundle = port->bundle;
2614 bundle_del_port(port);
2615 if (list_is_empty(&bundle->ports)) {
2616 bundle_destroy(bundle);
2617 } else if (list_is_short(&bundle->ports)) {
2618 bond_unref(bundle->bond);
2619 bundle->bond = NULL;
2625 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2627 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2628 struct ofport_dpif *port = port_;
2629 uint8_t ea[ETH_ADDR_LEN];
2632 error = netdev_get_etheraddr(port->up.netdev, ea);
2634 struct ofpbuf packet;
2637 ofpbuf_init(&packet, 0);
2638 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2640 memcpy(packet_pdu, pdu, pdu_size);
2642 ofproto_dpif_send_packet(port, &packet);
2643 ofpbuf_uninit(&packet);
2645 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2646 "%s (%s)", port->bundle->name,
2647 netdev_get_name(port->up.netdev), ovs_strerror(error));
2652 bundle_send_learning_packets(struct ofbundle *bundle)
2654 struct ofproto_dpif *ofproto = bundle->ofproto;
2655 struct ofpbuf *learning_packet;
2656 int error, n_packets, n_errors;
2657 struct mac_entry *e;
2658 struct list packets;
2660 list_init(&packets);
2661 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2662 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2663 if (e->port.p != bundle) {
2666 learning_packet = bond_compose_learning_packet(bundle->bond,
2669 learning_packet->private_p = port_void;
2670 list_push_back(&packets, &learning_packet->list_node);
2673 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2675 error = n_packets = n_errors = 0;
2676 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2679 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2686 ofpbuf_list_delete(&packets);
2689 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2690 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2691 "packets, last error was: %s",
2692 bundle->name, n_errors, n_packets, ovs_strerror(error));
2694 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2695 bundle->name, n_packets);
2700 bundle_run(struct ofbundle *bundle)
2703 lacp_run(bundle->lacp, send_pdu_cb);
2706 struct ofport_dpif *port;
2708 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2709 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2712 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2713 bundle->ofproto->backer->need_revalidate = REV_BOND;
2716 if (bond_should_send_learning_packets(bundle->bond)) {
2717 bundle_send_learning_packets(bundle);
2723 bundle_wait(struct ofbundle *bundle)
2726 lacp_wait(bundle->lacp);
2729 bond_wait(bundle->bond);
2736 mirror_set__(struct ofproto *ofproto_, void *aux,
2737 const struct ofproto_mirror_settings *s)
2739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2740 struct ofbundle **srcs, **dsts;
2745 mirror_destroy(ofproto->mbridge, aux);
2749 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2750 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2752 for (i = 0; i < s->n_srcs; i++) {
2753 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2756 for (i = 0; i < s->n_dsts; i++) {
2757 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2760 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2761 s->n_dsts, s->src_vlans,
2762 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2769 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2770 uint64_t *packets, uint64_t *bytes)
2773 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2778 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2780 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2781 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2782 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2783 mac_learning_flush(ofproto->ml);
2785 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2790 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2793 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2794 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2798 forward_bpdu_changed(struct ofproto *ofproto_)
2800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2805 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2809 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2810 mac_learning_set_idle_time(ofproto->ml, idle_time);
2811 mac_learning_set_max_entries(ofproto->ml, max_entries);
2812 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2817 static struct ofport_dpif *
2818 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2820 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2821 return ofport ? ofport_dpif_cast(ofport) : NULL;
2824 static struct ofport_dpif *
2825 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2827 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2828 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2832 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2833 struct ofproto_port *ofproto_port,
2834 struct dpif_port *dpif_port)
2836 ofproto_port->name = dpif_port->name;
2837 ofproto_port->type = dpif_port->type;
2838 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2842 ofport_update_peer(struct ofport_dpif *ofport)
2844 const struct ofproto_dpif *ofproto;
2845 struct dpif_backer *backer;
2848 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2852 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2853 backer->need_revalidate = REV_RECONFIGURE;
2856 ofport->peer->peer = NULL;
2857 ofport->peer = NULL;
2860 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2865 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2866 struct ofport *peer_ofport;
2867 struct ofport_dpif *peer;
2870 if (ofproto->backer != backer) {
2874 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2879 peer = ofport_dpif_cast(peer_ofport);
2880 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2881 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2883 ofport->peer = peer;
2884 ofport->peer->peer = ofport;
2894 port_run(struct ofport_dpif *ofport)
2896 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2897 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2898 bool enable = netdev_get_carrier(ofport->up.netdev);
2899 bool cfm_enable = false;
2900 bool bfd_enable = false;
2902 ofport->carrier_seq = carrier_seq;
2905 int cfm_opup = cfm_get_opup(ofport->cfm);
2907 cfm_enable = !cfm_get_fault(ofport->cfm);
2909 if (cfm_opup >= 0) {
2910 cfm_enable = cfm_enable && cfm_opup;
2915 bfd_enable = bfd_forwarding(ofport->bfd);
2918 if (ofport->bfd || ofport->cfm) {
2919 enable = enable && (cfm_enable || bfd_enable);
2922 if (ofport->bundle) {
2923 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2924 if (carrier_changed) {
2925 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2929 if (ofport->may_enable != enable) {
2930 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2931 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2934 ofport->may_enable = enable;
2938 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2939 struct ofproto_port *ofproto_port)
2941 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2942 struct dpif_port dpif_port;
2945 if (sset_contains(&ofproto->ghost_ports, devname)) {
2946 const char *type = netdev_get_type_from_name(devname);
2948 /* We may be called before ofproto->up.port_by_name is populated with
2949 * the appropriate ofport. For this reason, we must get the name and
2950 * type from the netdev layer directly. */
2952 const struct ofport *ofport;
2954 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2955 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2956 ofproto_port->name = xstrdup(devname);
2957 ofproto_port->type = xstrdup(type);
2963 if (!sset_contains(&ofproto->ports, devname)) {
2966 error = dpif_port_query_by_name(ofproto->backer->dpif,
2967 devname, &dpif_port);
2969 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2975 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2978 const char *devname = netdev_get_name(netdev);
2979 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2980 const char *dp_port_name;
2982 if (netdev_vport_is_patch(netdev)) {
2983 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2987 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2988 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2989 odp_port_t port_no = ODPP_NONE;
2992 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2996 if (netdev_get_tunnel_config(netdev)) {
2997 simap_put(&ofproto->backer->tnl_backers,
2998 dp_port_name, odp_to_u32(port_no));
3002 if (netdev_get_tunnel_config(netdev)) {
3003 sset_add(&ofproto->ghost_ports, devname);
3005 sset_add(&ofproto->ports, devname);
3011 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3013 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3014 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3021 sset_find_and_delete(&ofproto->ghost_ports,
3022 netdev_get_name(ofport->up.netdev));
3023 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3024 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3025 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3027 /* The caller is going to close ofport->up.netdev. If this is a
3028 * bonded port, then the bond is using that netdev, so remove it
3029 * from the bond. The client will need to reconfigure everything
3030 * after deleting ports, so then the slave will get re-added. */
3031 bundle_remove(&ofport->up);
3038 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3040 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3045 error = netdev_get_stats(ofport->up.netdev, stats);
3047 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3050 ovs_mutex_lock(&ofproto->stats_mutex);
3051 /* ofproto->stats.tx_packets represents packets that we created
3052 * internally and sent to some port (e.g. packets sent with
3053 * ofproto_dpif_send_packet()). Account for them as if they had
3054 * come from OFPP_LOCAL and got forwarded. */
3056 if (stats->rx_packets != UINT64_MAX) {
3057 stats->rx_packets += ofproto->stats.tx_packets;
3060 if (stats->rx_bytes != UINT64_MAX) {
3061 stats->rx_bytes += ofproto->stats.tx_bytes;
3064 /* ofproto->stats.rx_packets represents packets that were received on
3065 * some port and we processed internally and dropped (e.g. STP).
3066 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3068 if (stats->tx_packets != UINT64_MAX) {
3069 stats->tx_packets += ofproto->stats.rx_packets;
3072 if (stats->tx_bytes != UINT64_MAX) {
3073 stats->tx_bytes += ofproto->stats.rx_bytes;
3075 ovs_mutex_unlock(&ofproto->stats_mutex);
3081 struct port_dump_state {
3086 struct ofproto_port port;
3091 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3093 *statep = xzalloc(sizeof(struct port_dump_state));
3098 port_dump_next(const struct ofproto *ofproto_, void *state_,
3099 struct ofproto_port *port)
3101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102 struct port_dump_state *state = state_;
3103 const struct sset *sset;
3104 struct sset_node *node;
3106 if (state->has_port) {
3107 ofproto_port_destroy(&state->port);
3108 state->has_port = false;
3110 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3111 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3114 error = port_query_by_name(ofproto_, node->name, &state->port);
3116 *port = state->port;
3117 state->has_port = true;
3119 } else if (error != ENODEV) {
3124 if (!state->ghost) {
3125 state->ghost = true;
3128 return port_dump_next(ofproto_, state_, port);
3135 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3137 struct port_dump_state *state = state_;
3139 if (state->has_port) {
3140 ofproto_port_destroy(&state->port);
3147 port_poll(const struct ofproto *ofproto_, char **devnamep)
3149 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3151 if (ofproto->port_poll_errno) {
3152 int error = ofproto->port_poll_errno;
3153 ofproto->port_poll_errno = 0;
3157 if (sset_is_empty(&ofproto->port_poll_set)) {
3161 *devnamep = sset_pop(&ofproto->port_poll_set);
3166 port_poll_wait(const struct ofproto *ofproto_)
3168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3169 dpif_port_poll_wait(ofproto->backer->dpif);
3173 port_is_lacp_current(const struct ofport *ofport_)
3175 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3176 return (ofport->bundle && ofport->bundle->lacp
3177 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3181 /* Upcall handling. */
3183 struct flow_miss_op {
3184 struct dpif_op dpif_op;
3186 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3187 struct xlate_out xout;
3188 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3190 struct ofpbuf mask; /* Flow mask for "put" ops. */
3191 struct odputil_keybuf maskbuf;
3193 /* If this is a "put" op, then a pointer to the subfacet that should
3194 * be marked as uninstalled if the operation fails. */
3195 struct subfacet *subfacet;
3198 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3199 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3200 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3201 * return value of true). However, for short flows the cost of bookkeeping is
3202 * much higher than the benefits, so when the datapath holds a large number of
3203 * flows we impose some heuristics to decide which flows are likely to be worth
3206 flow_miss_should_make_facet(struct flow_miss *miss)
3208 struct dpif_backer *backer = miss->ofproto->backer;
3211 switch (flow_miss_model) {
3212 case OFPROTO_HANDLE_MISS_AUTO:
3214 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3216 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3220 if (!backer->governor) {
3223 n_subfacets = hmap_count(&backer->subfacets);
3224 if (n_subfacets * 2 <= flow_eviction_threshold) {
3228 backer->governor = governor_create();
3231 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3232 return governor_should_install_flow(backer->governor, hash,
3233 miss->stats.n_packets);
3236 /* Handles 'miss', which matches 'facet'. May add any required datapath
3237 * operations to 'ops', incrementing '*n_ops' for each new op.
3239 * All of the packets in 'miss' are considered to have arrived at time
3240 * 'miss->stats.used'. This is really important only for new facets: if we
3241 * just called time_msec() here, then the new subfacet or its packets could
3242 * look (occasionally) as though it was used some time after the facet was
3243 * used. That can make a one-packet flow look like it has a nonzero duration,
3244 * which looks odd in e.g. NetFlow statistics. */
3246 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3247 struct flow_miss_op *ops, size_t *n_ops)
3249 enum subfacet_path want_path;
3250 struct subfacet *subfacet;
3253 /* Update facet stats. */
3254 facet->packet_count += miss->stats.n_packets;
3255 facet->prev_packet_count += miss->stats.n_packets;
3256 facet->byte_count += miss->stats.n_bytes;
3257 facet->prev_byte_count += miss->stats.n_bytes;
3259 /* Look for an existing subfacet. If we find one, update its used time. */
3260 key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3261 if (!list_is_empty(&facet->subfacets)) {
3262 subfacet = subfacet_find(miss->ofproto->backer,
3263 miss->key, miss->key_len, key_hash);
3265 if (subfacet->facet == facet) {
3266 subfacet->used = MAX(subfacet->used, miss->stats.used);
3268 /* This shouldn't happen. */
3269 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3270 subfacet_destroy(subfacet);
3278 /* Don't install the flow if it's the result of the "userspace"
3279 * action for an already installed facet. This can occur when a
3280 * datapath flow with wildcards has a "userspace" action and flows
3281 * sent to userspace result in a different subfacet, which will then
3282 * be rejected as overlapping by the datapath. */
3283 if (miss->upcall_type == DPIF_UC_ACTION
3284 && !list_is_empty(&facet->subfacets)) {
3288 /* Create a subfacet, if we don't already have one. */
3290 subfacet = subfacet_create(facet, miss, key_hash);
3293 /* Install the subfacet, if it's not already installed. */
3294 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3295 if (subfacet->path != want_path) {
3296 struct flow_miss_op *op = &ops[(*n_ops)++];
3297 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3299 subfacet->path = want_path;
3301 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3302 if (enable_megaflows) {
3303 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3304 &miss->flow, UINT32_MAX);
3307 op->xout_garbage = false;
3308 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3309 op->subfacet = subfacet;
3310 put->flags = DPIF_FP_CREATE;
3311 put->key = miss->key;
3312 put->key_len = miss->key_len;
3313 put->mask = op->mask.data;
3314 put->mask_len = op->mask.size;
3316 if (want_path == SF_FAST_PATH) {
3317 put->actions = facet->xout.odp_actions.data;
3318 put->actions_len = facet->xout.odp_actions.size;
3320 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3321 op->slow_stub, sizeof op->slow_stub,
3322 &put->actions, &put->actions_len);
3328 /* Handles flow miss 'miss'. May add any required datapath operations
3329 * to 'ops', incrementing '*n_ops' for each new op. */
3331 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3334 struct facet *facet;
3336 miss->ofproto->n_missed += miss->stats.n_packets;
3338 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3340 /* There does not exist a bijection between 'struct flow' and datapath
3341 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3342 * assumption used throughout the facet and subfacet handling code.
3343 * Since we have to handle these misses in userspace anyway, we simply
3344 * skip facet creation, avoiding the problem altogether. */
3345 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3346 || !flow_miss_should_make_facet(miss)) {
3350 facet = facet_create(miss);
3352 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3355 static struct drop_key *
3356 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3359 struct drop_key *drop_key;
3361 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3362 &backer->drop_keys) {
3363 if (drop_key->key_len == key_len
3364 && !memcmp(drop_key->key, key, key_len)) {
3372 drop_key_clear(struct dpif_backer *backer)
3374 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3375 struct drop_key *drop_key, *next;
3377 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3380 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3382 if (error && !VLOG_DROP_WARN(&rl)) {
3383 struct ds ds = DS_EMPTY_INITIALIZER;
3384 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3385 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3386 ovs_strerror(error), ds_cstr(&ds));
3390 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3391 drop_key_destroy(drop_key);
3394 udpif_drop_key_clear(backer->udpif);
3398 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3400 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3401 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3402 struct flow_miss *miss;
3405 /* Process each element in the to-do list, constructing the set of
3406 * operations to batch. */
3408 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3409 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3411 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3413 /* Execute batch. */
3414 for (i = 0; i < n_ops; i++) {
3415 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3417 dpif_operate(backer->dpif, dpif_ops, n_ops);
3419 for (i = 0; i < n_ops; i++) {
3420 if (dpif_ops[i]->error != 0
3421 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3422 && flow_miss_ops[i].subfacet) {
3423 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3425 COVERAGE_INC(subfacet_install_fail);
3427 /* Zero-out subfacet counters when installation failed, but
3428 * datapath reported hits. This should not happen and
3429 * indicates a bug, since if the datapath flow exists, we
3430 * should not be attempting to create a new subfacet. A
3431 * buggy datapath could trigger this, so just zero out the
3432 * counters and log an error. */
3433 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3434 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3435 "datapath reported hits");
3436 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3439 subfacet->path = SF_NOT_INSTALLED;
3445 handle_upcalls(struct dpif_backer *backer)
3447 struct flow_miss_batch *fmb;
3450 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3451 struct drop_key *drop_key = drop_key_next(backer->udpif);
3456 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3457 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3458 hash_bytes(drop_key->key, drop_key->key_len, 0));
3459 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3460 drop_key->key, drop_key->key_len,
3461 NULL, 0, NULL, 0, NULL);
3463 drop_key_destroy(drop_key);
3467 fmb = flow_miss_batch_next(backer->udpif);
3469 handle_flow_misses(backer, fmb);
3470 flow_miss_batch_destroy(fmb);
3474 /* Flow expiration. */
3476 static int subfacet_max_idle(const struct dpif_backer *);
3477 static void update_stats(struct dpif_backer *);
3478 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3479 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3481 /* This function is called periodically by run(). Its job is to collect
3482 * updates for the flows that have been installed into the datapath, most
3483 * importantly when they last were used, and then use that information to
3484 * expire flows that have not been used recently.
3486 * Returns the number of milliseconds after which it should be called again. */
3488 expire(struct dpif_backer *backer)
3490 struct ofproto_dpif *ofproto;
3494 /* Periodically clear out the drop keys in an effort to keep them
3495 * relatively few. */
3496 drop_key_clear(backer);
3498 /* Update stats for each flow in the backer. */
3499 update_stats(backer);
3501 n_subfacets = hmap_count(&backer->subfacets);
3503 struct subfacet *subfacet;
3504 long long int total, now;
3508 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3509 total += now - subfacet->created;
3511 backer->avg_subfacet_life += total / n_subfacets;
3513 backer->avg_subfacet_life /= 2;
3515 backer->avg_n_subfacet += n_subfacets;
3516 backer->avg_n_subfacet /= 2;
3518 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3520 max_idle = subfacet_max_idle(backer);
3521 expire_subfacets(backer, max_idle);
3523 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3524 struct rule *rule, *next_rule;
3526 if (ofproto->backer != backer) {
3530 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3532 ovs_mutex_lock(&ofproto_mutex);
3533 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3534 &ofproto->up.expirable) {
3535 rule_expire(rule_dpif_cast(rule));
3537 ovs_mutex_unlock(&ofproto_mutex);
3539 /* All outstanding data in existing flows has been accounted, so it's a
3540 * good time to do bond rebalancing. */
3541 if (ofproto->has_bonded_bundles) {
3542 struct ofbundle *bundle;
3544 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3546 bond_rebalance(bundle->bond);
3552 return MIN(max_idle, 1000);
3555 /* Updates flow table statistics given that the datapath just reported 'stats'
3556 * as 'subfacet''s statistics. */
3558 update_subfacet_stats(struct subfacet *subfacet,
3559 const struct dpif_flow_stats *stats)
3561 struct facet *facet = subfacet->facet;
3562 struct dpif_flow_stats diff;
3564 diff.tcp_flags = stats->tcp_flags;
3565 diff.used = stats->used;
3567 if (stats->n_packets >= subfacet->dp_packet_count) {
3568 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3570 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3574 if (stats->n_bytes >= subfacet->dp_byte_count) {
3575 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3577 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3581 facet->ofproto->n_hit += diff.n_packets;
3582 subfacet->dp_packet_count = stats->n_packets;
3583 subfacet->dp_byte_count = stats->n_bytes;
3584 subfacet_update_stats(subfacet, &diff);
3586 if (facet->accounted_bytes < facet->byte_count) {
3588 facet_account(facet);
3589 facet->accounted_bytes = facet->byte_count;
3593 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3594 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3596 delete_unexpected_flow(struct dpif_backer *backer,
3597 const struct nlattr *key, size_t key_len)
3599 if (!VLOG_DROP_WARN(&rl)) {
3603 odp_flow_key_format(key, key_len, &s);
3604 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3608 COVERAGE_INC(facet_unexpected);
3609 dpif_flow_del(backer->dpif, key, key_len, NULL);
3612 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3614 * This function also pushes statistics updates to rules which each facet
3615 * resubmits into. Generally these statistics will be accurate. However, if a
3616 * facet changes the rule it resubmits into at some time in between
3617 * update_stats() runs, it is possible that statistics accrued to the
3618 * old rule will be incorrectly attributed to the new rule. This could be
3619 * avoided by calling update_stats() whenever rules are created or
3620 * deleted. However, the performance impact of making so many calls to the
3621 * datapath do not justify the benefit of having perfectly accurate statistics.
3623 * In addition, this function maintains per ofproto flow hit counts. The patch
3624 * port is not treated specially. e.g. A packet ingress from br0 patched into
3625 * br1 will increase the hit count of br0 by 1, however, does not affect
3626 * the hit or miss counts of br1.
3629 update_stats(struct dpif_backer *backer)
3631 const struct dpif_flow_stats *stats;
3632 struct dpif_flow_dump dump;
3633 const struct nlattr *key, *mask;
3634 size_t key_len, mask_len;
3636 dpif_flow_dump_start(&dump, backer->dpif);
3637 while (dpif_flow_dump_next(&dump, &key, &key_len,
3638 &mask, &mask_len, NULL, NULL, &stats)) {
3639 struct subfacet *subfacet;
3642 key_hash = odp_flow_key_hash(key, key_len);
3643 subfacet = subfacet_find(backer, key, key_len, key_hash);
3644 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3646 update_subfacet_stats(subfacet, stats);
3650 /* Stats are updated per-packet. */
3653 case SF_NOT_INSTALLED:
3655 delete_unexpected_flow(backer, key, key_len);
3660 dpif_flow_dump_done(&dump);
3663 /* Calculates and returns the number of milliseconds of idle time after which
3664 * subfacets should expire from the datapath. When a subfacet expires, we fold
3665 * its statistics into its facet, and when a facet's last subfacet expires, we
3666 * fold its statistic into its rule. */
3668 subfacet_max_idle(const struct dpif_backer *backer)
3671 * Idle time histogram.
3673 * Most of the time a switch has a relatively small number of subfacets.
3674 * When this is the case we might as well keep statistics for all of them
3675 * in userspace and to cache them in the kernel datapath for performance as
3678 * As the number of subfacets increases, the memory required to maintain
3679 * statistics about them in userspace and in the kernel becomes
3680 * significant. However, with a large number of subfacets it is likely
3681 * that only a few of them are "heavy hitters" that consume a large amount
3682 * of bandwidth. At this point, only heavy hitters are worth caching in
3683 * the kernel and maintaining in userspaces; other subfacets we can
3686 * The technique used to compute the idle time is to build a histogram with
3687 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3688 * that is installed in the kernel gets dropped in the appropriate bucket.
3689 * After the histogram has been built, we compute the cutoff so that only
3690 * the most-recently-used 1% of subfacets (but at least
3691 * flow_eviction_threshold flows) are kept cached. At least
3692 * the most-recently-used bucket of subfacets is kept, so actually an
3693 * arbitrary number of subfacets can be kept in any given expiration run
3694 * (though the next run will delete most of those unless they receive
3697 * This requires a second pass through the subfacets, in addition to the
3698 * pass made by update_stats(), because the former function never looks at
3699 * uninstallable subfacets.
3701 enum { BUCKET_WIDTH = 100 };
3702 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3703 int buckets[N_BUCKETS] = { 0 };
3704 int total, subtotal, bucket;
3705 struct subfacet *subfacet;
3709 total = hmap_count(&backer->subfacets);
3710 if (total <= flow_eviction_threshold) {
3711 return N_BUCKETS * BUCKET_WIDTH;
3714 /* Build histogram. */
3716 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3717 long long int idle = now - subfacet->used;
3718 int bucket = (idle <= 0 ? 0
3719 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3720 : (unsigned int) idle / BUCKET_WIDTH);
3724 /* Find the first bucket whose flows should be expired. */
3725 subtotal = bucket = 0;
3727 subtotal += buckets[bucket++];
3728 } while (bucket < N_BUCKETS &&
3729 subtotal < MAX(flow_eviction_threshold, total / 100));
3731 if (VLOG_IS_DBG_ENABLED()) {
3735 ds_put_cstr(&s, "keep");
3736 for (i = 0; i < N_BUCKETS; i++) {
3738 ds_put_cstr(&s, ", drop");
3741 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3744 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3748 return bucket * BUCKET_WIDTH;
3752 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3754 /* Cutoff time for most flows. */
3755 long long int normal_cutoff = time_msec() - dp_max_idle;
3757 /* We really want to keep flows for special protocols around, so use a more
3758 * conservative cutoff. */
3759 long long int special_cutoff = time_msec() - 10000;
3761 struct subfacet *subfacet, *next_subfacet;
3762 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3766 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3767 &backer->subfacets) {
3768 long long int cutoff;
3770 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3774 if (subfacet->used < cutoff) {
3775 if (subfacet->path != SF_NOT_INSTALLED) {
3776 batch[n_batch++] = subfacet;
3777 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3778 subfacet_destroy_batch(backer, batch, n_batch);
3782 subfacet_destroy(subfacet);
3788 subfacet_destroy_batch(backer, batch, n_batch);
3792 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3793 * then delete it entirely. */
3795 rule_expire(struct rule_dpif *rule)
3796 OVS_REQUIRES(ofproto_mutex)
3798 uint16_t idle_timeout, hard_timeout;
3799 long long int now = time_msec();
3802 ovs_assert(!rule->up.pending);
3804 /* Has 'rule' expired? */
3805 ovs_mutex_lock(&rule->up.mutex);
3806 hard_timeout = rule->up.hard_timeout;
3807 idle_timeout = rule->up.idle_timeout;
3808 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3809 reason = OFPRR_HARD_TIMEOUT;
3810 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3811 reason = OFPRR_IDLE_TIMEOUT;
3815 ovs_mutex_unlock(&rule->up.mutex);
3818 COVERAGE_INC(ofproto_dpif_expired);
3819 ofproto_rule_expire(&rule->up, reason);
3825 /* Creates and returns a new facet based on 'miss'.
3827 * The caller must already have determined that no facet with an identical
3828 * 'miss->flow' exists in 'miss->ofproto'.
3830 * 'rule' and 'xout' must have been created based on 'miss'.
3832 * 'facet'' statistics are initialized based on 'stats'.
3834 * The facet will initially have no subfacets. The caller should create (at
3835 * least) one subfacet with subfacet_create(). */
3836 static struct facet *
3837 facet_create(const struct flow_miss *miss)
3839 struct ofproto_dpif *ofproto = miss->ofproto;
3840 struct facet *facet;
3843 COVERAGE_INC(facet_create);
3844 facet = xzalloc(sizeof *facet);
3845 facet->ofproto = miss->ofproto;
3846 facet->used = miss->stats.used;
3847 facet->flow = miss->flow;
3848 facet->learn_rl = time_msec() + 500;
3850 list_init(&facet->subfacets);
3852 xlate_out_copy(&facet->xout, &miss->xout);
3854 match_init(&match, &facet->flow, &facet->xout.wc);
3855 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3856 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3857 classifier_insert(&ofproto->facets, &facet->cr);
3858 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3860 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3861 netflow_flow_update(ofproto->netflow, &facet->flow,
3862 facet->xout.nf_output_iface, &miss->stats);
3869 facet_free(struct facet *facet)
3872 xlate_out_uninit(&facet->xout);
3877 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3878 * 'flow' must reflect the data in 'packet'. */
3880 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3881 const struct flow *flow,
3882 struct rule_dpif *rule,
3883 const struct ofpact *ofpacts, size_t ofpacts_len,
3884 struct ofpbuf *packet)
3886 struct odputil_keybuf keybuf;
3887 struct dpif_flow_stats stats;
3888 struct xlate_out xout;
3889 struct xlate_in xin;
3894 ovs_assert((rule != NULL) != (ofpacts != NULL));
3896 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3898 rule_dpif_credit_stats(rule, &stats);
3901 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3902 xin.ofpacts = ofpacts;
3903 xin.ofpacts_len = ofpacts_len;
3904 xin.resubmit_stats = &stats;
3905 xlate_actions(&xin, &xout);
3907 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3908 in_port = flow->in_port.ofp_port;
3909 if (in_port == OFPP_NONE) {
3910 in_port = OFPP_LOCAL;
3912 odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3914 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3915 xout.odp_actions.data, xout.odp_actions.size, packet,
3916 (xout.slow & SLOW_ACTION) != 0);
3917 xlate_out_uninit(&xout);
3922 /* Remove 'facet' from its ofproto and free up the associated memory:
3924 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3925 * rule's statistics, via subfacet_uninstall().
3927 * - Removes 'facet' from its rule and from ofproto->facets.
3930 facet_remove(struct facet *facet)
3932 struct subfacet *subfacet, *next_subfacet;
3934 COVERAGE_INC(facet_remove);
3935 ovs_assert(!list_is_empty(&facet->subfacets));
3937 /* First uninstall all of the subfacets to get final statistics. */
3938 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3939 subfacet_uninstall(subfacet);
3942 /* Flush the final stats to the rule.
3944 * This might require us to have at least one subfacet around so that we
3945 * can use its actions for accounting in facet_account(), which is why we
3946 * have uninstalled but not yet destroyed the subfacets. */
3947 facet_flush_stats(facet);
3949 /* Now we're really all done so destroy everything. */
3950 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3951 &facet->subfacets) {
3952 subfacet_destroy__(subfacet);
3954 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3955 classifier_remove(&facet->ofproto->facets, &facet->cr);
3956 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3957 cls_rule_destroy(&facet->cr);
3961 /* Feed information from 'facet' back into the learning table to keep it in
3962 * sync with what is actually flowing through the datapath. */
3964 facet_learn(struct facet *facet)
3966 long long int now = time_msec();
3968 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3972 facet->learn_rl = now + 500;
3974 if (!facet->xout.has_learn
3975 && !facet->xout.has_normal
3976 && (!facet->xout.has_fin_timeout
3977 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3981 facet_push_stats(facet, true);
3985 facet_account(struct facet *facet)
3987 const struct nlattr *a;
3992 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3995 n_bytes = facet->byte_count - facet->accounted_bytes;
3997 /* This loop feeds byte counters to bond_account() for rebalancing to use
3998 * as a basis. We also need to track the actual VLAN on which the packet
3999 * is going to be sent to ensure that it matches the one passed to
4000 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4003 * We use the actions from an arbitrary subfacet because they should all
4004 * be equally valid for our purpose. */
4005 vlan_tci = facet->flow.vlan_tci;
4006 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4007 facet->xout.odp_actions.size) {
4008 const struct ovs_action_push_vlan *vlan;
4009 struct ofport_dpif *port;
4011 switch (nl_attr_type(a)) {
4012 case OVS_ACTION_ATTR_OUTPUT:
4013 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4014 if (port && port->bundle && port->bundle->bond) {
4015 bond_account(port->bundle->bond, &facet->flow,
4016 vlan_tci_to_vid(vlan_tci), n_bytes);
4020 case OVS_ACTION_ATTR_POP_VLAN:
4021 vlan_tci = htons(0);
4024 case OVS_ACTION_ATTR_PUSH_VLAN:
4025 vlan = nl_attr_get(a);
4026 vlan_tci = vlan->vlan_tci;
4032 /* Returns true if the only action for 'facet' is to send to the controller.
4033 * (We don't report NetFlow expiration messages for such facets because they
4034 * are just part of the control logic for the network, not real traffic). */
4036 facet_is_controller_flow(struct facet *facet)
4039 struct ofproto_dpif *ofproto = facet->ofproto;
4040 const struct ofpact *ofpacts;
4041 struct rule_actions *actions;
4042 struct rule_dpif *rule;
4046 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4047 actions = rule_dpif_get_actions(rule);
4048 rule_dpif_unref(rule);
4050 ofpacts_len = actions->ofpacts_len;
4051 ofpacts = actions->ofpacts;
4052 is_controller = ofpacts_len > 0
4053 && ofpacts->type == OFPACT_CONTROLLER
4054 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4055 rule_actions_unref(actions);
4057 return is_controller;
4062 /* Folds all of 'facet''s statistics into its rule. Also updates the
4063 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4064 * 'facet''s statistics in the datapath should have been zeroed and folded into
4065 * its packet and byte counts before this function is called. */
4067 facet_flush_stats(struct facet *facet)
4069 struct ofproto_dpif *ofproto = facet->ofproto;
4070 struct subfacet *subfacet;
4072 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4073 ovs_assert(!subfacet->dp_byte_count);
4074 ovs_assert(!subfacet->dp_packet_count);
4077 facet_push_stats(facet, false);
4078 if (facet->accounted_bytes < facet->byte_count) {
4079 facet_account(facet);
4080 facet->accounted_bytes = facet->byte_count;
4083 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4084 netflow_expire(ofproto->netflow, &facet->flow);
4085 netflow_flow_clear(ofproto->netflow, &facet->flow);
4088 /* Reset counters to prevent double counting if 'facet' ever gets
4090 facet_reset_counters(facet);
4091 facet->tcp_flags = 0;
4094 /* Searches 'ofproto''s table of facets for one which would be responsible for
4095 * 'flow'. Returns it if found, otherwise a null pointer.
4097 * The returned facet might need revalidation; use facet_lookup_valid()
4098 * instead if that is important. */
4099 static struct facet *
4100 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4102 struct cls_rule *cr;
4104 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4105 cr = classifier_lookup(&ofproto->facets, flow, NULL);
4106 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4107 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4110 /* Searches 'ofproto''s table of facets for one capable that covers
4111 * 'flow'. Returns it if found, otherwise a null pointer.
4113 * The returned facet is guaranteed to be valid. */
4114 static struct facet *
4115 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4117 struct facet *facet;
4119 facet = facet_find(ofproto, flow);
4121 && ofproto->backer->need_revalidate
4122 && !facet_revalidate(facet)) {
4130 facet_check_consistency(struct facet *facet)
4132 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4134 struct xlate_out xout;
4135 struct xlate_in xin;
4138 /* Check the datapath actions for consistency. */
4139 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4140 xlate_actions(&xin, &xout);
4142 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4143 && facet->xout.slow == xout.slow;
4144 if (!ok && !VLOG_DROP_WARN(&rl)) {
4145 struct ds s = DS_EMPTY_INITIALIZER;
4147 flow_format(&s, &facet->flow);
4148 ds_put_cstr(&s, ": inconsistency in facet");
4150 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4151 ds_put_cstr(&s, " (actions were: ");
4152 format_odp_actions(&s, facet->xout.odp_actions.data,
4153 facet->xout.odp_actions.size);
4154 ds_put_cstr(&s, ") (correct actions: ");
4155 format_odp_actions(&s, xout.odp_actions.data,
4156 xout.odp_actions.size);
4157 ds_put_char(&s, ')');
4160 if (facet->xout.slow != xout.slow) {
4161 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4166 xlate_out_uninit(&xout);
4171 /* Re-searches the classifier for 'facet':
4173 * - If the rule found is different from 'facet''s current rule, moves
4174 * 'facet' to the new rule and recompiles its actions.
4176 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4177 * where it is and recompiles its actions anyway.
4179 * - If any of 'facet''s subfacets correspond to a new flow according to
4180 * xlate_receive(), 'facet' is removed.
4182 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4184 facet_revalidate(struct facet *facet)
4186 struct ofproto_dpif *ofproto = facet->ofproto;
4187 struct rule_dpif *new_rule;
4188 struct subfacet *subfacet;
4189 struct flow_wildcards wc;
4190 struct xlate_out xout;
4191 struct xlate_in xin;
4193 COVERAGE_INC(facet_revalidate);
4195 /* Check that child subfacets still correspond to this facet. Tunnel
4196 * configuration changes could cause a subfacet's OpenFlow in_port to
4198 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4199 struct ofproto_dpif *recv_ofproto;
4200 struct flow recv_flow;
4203 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4204 subfacet->key_len, &recv_flow, NULL,
4205 &recv_ofproto, NULL);
4207 || recv_ofproto != ofproto
4208 || facet != facet_find(ofproto, &recv_flow)) {
4209 facet_remove(facet);
4214 flow_wildcards_init_catchall(&wc);
4215 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4217 /* Calculate new datapath actions.
4219 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4220 * emit a NetFlow expiration and, if so, we need to have the old state
4221 * around to properly compose it. */
4222 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4223 xlate_actions(&xin, &xout);
4224 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4225 /* Make sure non -packet fields are not masked. If not cleared,
4226 * the memcmp() below may fail, causing an otherwise valid facet
4228 flow_wildcards_clear_non_packet_fields(&xout.wc);
4230 /* A facet's slow path reason should only change under dramatic
4231 * circumstances. Rather than try to update everything, it's simpler to
4232 * remove the facet and start over.
4234 * More importantly, if a facet's wildcards change, it will be relatively
4235 * difficult to figure out if its subfacets still belong to it, and if not
4236 * which facet they may belong to. Again, to avoid the complexity, we
4237 * simply give up instead. */
4238 if (facet->xout.slow != xout.slow
4239 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4240 facet_remove(facet);
4241 xlate_out_uninit(&xout);
4242 rule_dpif_unref(new_rule);
4246 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4247 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4248 if (subfacet->path == SF_FAST_PATH) {
4249 struct dpif_flow_stats stats;
4251 subfacet_install(subfacet, &xout.odp_actions, &stats);
4252 subfacet_update_stats(subfacet, &stats);
4256 facet_flush_stats(facet);
4258 ofpbuf_clear(&facet->xout.odp_actions);
4259 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4260 xout.odp_actions.size);
4263 /* Update 'facet' now that we've taken care of all the old state. */
4264 facet->xout.slow = xout.slow;
4265 facet->xout.has_learn = xout.has_learn;
4266 facet->xout.has_normal = xout.has_normal;
4267 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4268 facet->xout.nf_output_iface = xout.nf_output_iface;
4269 facet->xout.mirrors = xout.mirrors;
4271 ovs_mutex_lock(&new_rule->up.mutex);
4272 facet->used = MAX(facet->used, new_rule->up.created);
4273 ovs_mutex_unlock(&new_rule->up.mutex);
4275 xlate_out_uninit(&xout);
4276 rule_dpif_unref(new_rule);
4281 facet_reset_counters(struct facet *facet)
4283 facet->packet_count = 0;
4284 facet->byte_count = 0;
4285 facet->prev_packet_count = 0;
4286 facet->prev_byte_count = 0;
4287 facet->accounted_bytes = 0;
4291 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4292 struct dpif_flow_stats *stats, bool may_learn)
4294 struct ofport_dpif *in_port;
4295 struct xlate_in xin;
4297 in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4298 if (in_port && in_port->is_tunnel) {
4299 netdev_vport_inc_rx(in_port->up.netdev, stats);
4301 bfd_account_rx(in_port->bfd, stats);
4305 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4306 xin.resubmit_stats = stats;
4307 xin.may_learn = may_learn;
4308 xlate_actions_for_side_effects(&xin);
4312 facet_push_stats(struct facet *facet, bool may_learn)
4314 struct dpif_flow_stats stats;
4316 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4317 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4318 ovs_assert(facet->used >= facet->prev_used);
4320 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4321 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4322 stats.used = facet->used;
4323 stats.tcp_flags = facet->tcp_flags;
4325 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4326 facet->prev_packet_count = facet->packet_count;
4327 facet->prev_byte_count = facet->byte_count;
4328 facet->prev_used = facet->used;
4330 if (facet->ofproto->netflow && !facet_is_controller_flow(facet)) {
4331 netflow_flow_update(facet->ofproto->netflow, &facet->flow,
4332 facet->xout.nf_output_iface, &stats);
4334 mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4335 stats.n_packets, stats.n_bytes);
4336 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4341 push_all_stats__(bool run_fast)
4343 static long long int rl = LLONG_MIN;
4344 struct ofproto_dpif *ofproto;
4346 if (time_msec() < rl) {
4350 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4351 struct cls_cursor cursor;
4352 struct facet *facet;
4354 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4355 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4356 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4357 facet_push_stats(facet, false);
4362 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4365 rl = time_msec() + 100;
4369 push_all_stats(void)
4371 push_all_stats__(true);
4375 rule_dpif_credit_stats(struct rule_dpif *rule,
4376 const struct dpif_flow_stats *stats)
4378 ovs_mutex_lock(&rule->stats_mutex);
4379 rule->packet_count += stats->n_packets;
4380 rule->byte_count += stats->n_bytes;
4381 rule->up.used = MAX(rule->up.used, stats->used);
4382 ovs_mutex_unlock(&rule->stats_mutex);
4386 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4388 return is_fail_open_rule(&rule->up);
4392 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4394 return rule_is_table_miss(&rule->up);
4398 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4399 OVS_REQUIRES(rule->up.mutex)
4401 return rule->up.flow_cookie;
4405 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4406 uint16_t hard_timeout)
4408 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4411 /* Returns 'rule''s actions. The caller owns a reference on the returned
4412 * actions and must eventually release it (with rule_actions_unref()) to avoid
4414 struct rule_actions *
4415 rule_dpif_get_actions(const struct rule_dpif *rule)
4417 return rule_get_actions(&rule->up);
4422 static struct subfacet *
4423 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4424 size_t key_len, uint32_t key_hash)
4426 struct subfacet *subfacet;
4428 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4429 &backer->subfacets) {
4430 if (subfacet->key_len == key_len
4431 && !memcmp(key, subfacet->key, key_len)) {
4439 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4440 * 'key_hash' must be a hash over miss->key. The caller must have already
4441 * ensured that no subfacet subfacet already exists. */
4442 static struct subfacet *
4443 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4445 struct dpif_backer *backer = miss->ofproto->backer;
4446 const struct nlattr *key = miss->key;
4447 size_t key_len = miss->key_len;
4448 struct subfacet *subfacet;
4450 subfacet = (list_is_empty(&facet->subfacets)
4451 ? &facet->one_subfacet
4452 : xmalloc(sizeof *subfacet));
4454 COVERAGE_INC(subfacet_create);
4455 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4456 list_push_back(&facet->subfacets, &subfacet->list_node);
4457 subfacet->facet = facet;
4458 subfacet->key = xmemdup(key, key_len);
4459 subfacet->key_len = key_len;
4460 subfacet->used = miss->stats.used;
4461 subfacet->created = subfacet->used;
4462 subfacet->dp_packet_count = 0;
4463 subfacet->dp_byte_count = 0;
4464 subfacet->path = SF_NOT_INSTALLED;
4465 subfacet->backer = backer;
4470 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4471 * its facet within 'ofproto', and frees it. */
4473 subfacet_destroy__(struct subfacet *subfacet)
4475 struct facet *facet = subfacet->facet;
4477 COVERAGE_INC(subfacet_destroy);
4478 subfacet_uninstall(subfacet);
4479 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4480 list_remove(&subfacet->list_node);
4481 free(subfacet->key);
4482 if (subfacet != &facet->one_subfacet) {
4487 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4488 * last remaining subfacet in its facet destroys the facet too. */
4490 subfacet_destroy(struct subfacet *subfacet)
4492 struct facet *facet = subfacet->facet;
4494 if (list_is_singleton(&facet->subfacets)) {
4495 /* facet_remove() needs at least one subfacet (it will remove it). */
4496 facet_remove(facet);
4498 subfacet_destroy__(subfacet);
4503 subfacet_destroy_batch(struct dpif_backer *backer,
4504 struct subfacet **subfacets, int n)
4506 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4507 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4508 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4511 for (i = 0; i < n; i++) {
4512 ops[i].type = DPIF_OP_FLOW_DEL;
4513 ops[i].u.flow_del.key = subfacets[i]->key;
4514 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4515 ops[i].u.flow_del.stats = &stats[i];
4519 dpif_operate(backer->dpif, opsp, n);
4520 for (i = 0; i < n; i++) {
4521 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4522 subfacets[i]->path = SF_NOT_INSTALLED;
4523 subfacet_destroy(subfacets[i]);
4528 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4529 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4530 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4531 * since 'subfacet' was last updated.
4533 * Returns 0 if successful, otherwise a positive errno value. */
4535 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4536 struct dpif_flow_stats *stats)
4538 struct facet *facet = subfacet->facet;
4539 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4540 const struct nlattr *actions = odp_actions->data;
4541 size_t actions_len = odp_actions->size;
4542 struct odputil_keybuf maskbuf;
4545 uint64_t slow_path_stub[128 / 8];
4546 enum dpif_flow_put_flags flags;
4549 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4552 flags |= DPIF_FP_ZERO_STATS;
4555 if (path == SF_SLOW_PATH) {
4556 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4557 slow_path_stub, sizeof slow_path_stub,
4558 &actions, &actions_len);
4561 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4562 if (enable_megaflows) {
4563 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4564 &facet->flow, UINT32_MAX);
4567 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4568 subfacet->key_len, mask.data, mask.size,
4569 actions, actions_len, stats);
4572 subfacet_reset_dp_stats(subfacet, stats);
4576 COVERAGE_INC(subfacet_install_fail);
4578 subfacet->path = path;
4583 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4585 subfacet_uninstall(struct subfacet *subfacet)
4587 if (subfacet->path != SF_NOT_INSTALLED) {
4588 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4589 struct dpif_flow_stats stats;
4592 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4593 subfacet->key_len, &stats);
4594 subfacet_reset_dp_stats(subfacet, &stats);
4596 subfacet_update_stats(subfacet, &stats);
4598 subfacet->path = SF_NOT_INSTALLED;
4600 ovs_assert(subfacet->dp_packet_count == 0);
4601 ovs_assert(subfacet->dp_byte_count == 0);
4605 /* Resets 'subfacet''s datapath statistics counters. This should be called
4606 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4607 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4608 * was reset in the datapath. 'stats' will be modified to include only
4609 * statistics new since 'subfacet' was last updated. */
4611 subfacet_reset_dp_stats(struct subfacet *subfacet,
4612 struct dpif_flow_stats *stats)
4615 && subfacet->dp_packet_count <= stats->n_packets
4616 && subfacet->dp_byte_count <= stats->n_bytes) {
4617 stats->n_packets -= subfacet->dp_packet_count;
4618 stats->n_bytes -= subfacet->dp_byte_count;
4621 subfacet->dp_packet_count = 0;
4622 subfacet->dp_byte_count = 0;
4625 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4627 * Because of the meaning of a subfacet's counters, it only makes sense to do
4628 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4629 * represents a packet that was sent by hand or if it represents statistics
4630 * that have been cleared out of the datapath. */
4632 subfacet_update_stats(struct subfacet *subfacet,
4633 const struct dpif_flow_stats *stats)
4635 if (stats->n_packets || stats->used > subfacet->used) {
4636 struct facet *facet = subfacet->facet;
4638 subfacet->used = MAX(subfacet->used, stats->used);
4639 facet->used = MAX(facet->used, stats->used);
4640 facet->packet_count += stats->n_packets;
4641 facet->byte_count += stats->n_bytes;
4642 facet->tcp_flags |= stats->tcp_flags;
4648 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4649 * the fields that were relevant as part of the lookup. */
4651 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4652 struct flow_wildcards *wc, struct rule_dpif **rule)
4654 struct ofport_dpif *port;
4656 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4659 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4661 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4662 flow->in_port.ofp_port);
4665 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4666 ofproto->no_packet_in_rule, rule);
4670 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4671 const struct flow *flow, struct flow_wildcards *wc,
4672 uint8_t table_id, struct rule_dpif **rule)
4674 const struct cls_rule *cls_rule;
4675 struct classifier *cls;
4679 if (table_id >= N_TABLES) {
4684 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4685 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4688 cls = &ofproto->up.tables[table_id].cls;
4689 ovs_rwlock_rdlock(&cls->rwlock);
4690 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4691 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4692 /* We must pretend that transport ports are unavailable. */
4693 struct flow ofpc_normal_flow = *flow;
4694 ofpc_normal_flow.tp_src = htons(0);
4695 ofpc_normal_flow.tp_dst = htons(0);
4696 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4697 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4698 cls_rule = &ofproto->drop_frags_rule->up.cr;
4699 /* Frag mask in wc already set above. */
4701 cls_rule = classifier_lookup(cls, flow, wc);
4704 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4705 rule_dpif_ref(*rule);
4706 ovs_rwlock_unlock(&cls->rwlock);
4708 return *rule != NULL;
4711 /* Given a port configuration (specified as zero if there's no port), chooses
4712 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4713 * flow table miss. */
4715 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4716 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4718 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4719 rule_dpif_ref(*rule);
4723 rule_dpif_ref(struct rule_dpif *rule)
4726 ofproto_rule_ref(&rule->up);
4731 rule_dpif_unref(struct rule_dpif *rule)
4734 ofproto_rule_unref(&rule->up);
4739 complete_operation(struct rule_dpif *rule)
4740 OVS_REQUIRES(ofproto_mutex)
4742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4744 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4745 ofoperation_complete(rule->up.pending, 0);
4748 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4750 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4753 static struct rule *
4756 struct rule_dpif *rule = xmalloc(sizeof *rule);
4761 rule_dealloc(struct rule *rule_)
4763 struct rule_dpif *rule = rule_dpif_cast(rule_);
4768 rule_construct(struct rule *rule_)
4770 struct rule_dpif *rule = rule_dpif_cast(rule_);
4771 ovs_mutex_init(&rule->stats_mutex);
4772 ovs_mutex_lock(&rule->stats_mutex);
4773 rule->packet_count = 0;
4774 rule->byte_count = 0;
4775 ovs_mutex_unlock(&rule->stats_mutex);
4780 rule_insert(struct rule *rule_)
4781 OVS_REQUIRES(ofproto_mutex)
4783 struct rule_dpif *rule = rule_dpif_cast(rule_);
4784 complete_operation(rule);
4788 rule_delete(struct rule *rule_)
4789 OVS_REQUIRES(ofproto_mutex)
4791 struct rule_dpif *rule = rule_dpif_cast(rule_);
4792 complete_operation(rule);
4796 rule_destruct(struct rule *rule_)
4798 struct rule_dpif *rule = rule_dpif_cast(rule_);
4799 ovs_mutex_destroy(&rule->stats_mutex);
4803 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4805 struct rule_dpif *rule = rule_dpif_cast(rule_);
4807 /* push_all_stats() can handle flow misses which, when using the learn
4808 * action, can cause rules to be added and deleted. This can corrupt our
4809 * caller's datastructures which assume that rule_get_stats() doesn't have
4810 * an impact on the flow table. To be safe, we disable miss handling. */
4811 push_all_stats__(false);
4813 /* Start from historical data for 'rule' itself that are no longer tracked
4814 * in facets. This counts, for example, facets that have expired. */
4815 ovs_mutex_lock(&rule->stats_mutex);
4816 *packets = rule->packet_count;
4817 *bytes = rule->byte_count;
4818 ovs_mutex_unlock(&rule->stats_mutex);
4822 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4823 struct ofpbuf *packet)
4825 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4827 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4831 rule_execute(struct rule *rule, const struct flow *flow,
4832 struct ofpbuf *packet)
4834 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4835 ofpbuf_delete(packet);
4840 rule_modify_actions(struct rule *rule_, bool reset_counters)
4841 OVS_REQUIRES(ofproto_mutex)
4843 struct rule_dpif *rule = rule_dpif_cast(rule_);
4845 if (reset_counters) {
4846 ovs_mutex_lock(&rule->stats_mutex);
4847 rule->packet_count = 0;
4848 rule->byte_count = 0;
4849 ovs_mutex_unlock(&rule->stats_mutex);
4852 complete_operation(rule);
4855 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4857 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4860 static struct ofgroup *
4863 struct group_dpif *group = xzalloc(sizeof *group);
4868 group_dealloc(struct ofgroup *group_)
4870 struct group_dpif *group = group_dpif_cast(group_);
4875 group_construct_stats(struct group_dpif *group)
4876 OVS_REQUIRES(group->stats_mutex)
4878 group->packet_count = 0;
4879 group->byte_count = 0;
4880 if (!group->bucket_stats) {
4881 group->bucket_stats = xcalloc(group->up.n_buckets,
4882 sizeof *group->bucket_stats);
4884 memset(group->bucket_stats, 0, group->up.n_buckets *
4885 sizeof *group->bucket_stats);
4890 group_construct(struct ofgroup *group_)
4892 struct group_dpif *group = group_dpif_cast(group_);
4893 ovs_mutex_init(&group->stats_mutex);
4894 ovs_mutex_lock(&group->stats_mutex);
4895 group_construct_stats(group);
4896 ovs_mutex_unlock(&group->stats_mutex);
4901 group_destruct__(struct group_dpif *group)
4902 OVS_REQUIRES(group->stats_mutex)
4904 free(group->bucket_stats);
4905 group->bucket_stats = NULL;
4909 group_destruct(struct ofgroup *group_)
4911 struct group_dpif *group = group_dpif_cast(group_);
4912 ovs_mutex_lock(&group->stats_mutex);
4913 group_destruct__(group);
4914 ovs_mutex_unlock(&group->stats_mutex);
4915 ovs_mutex_destroy(&group->stats_mutex);
4919 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4921 struct group_dpif *group = group_dpif_cast(group_);
4922 struct group_dpif *victim = group_dpif_cast(victim_);
4924 ovs_mutex_lock(&group->stats_mutex);
4925 if (victim->up.n_buckets < group->up.n_buckets) {
4926 group_destruct__(group);
4928 group_construct_stats(group);
4929 ovs_mutex_unlock(&group->stats_mutex);
4935 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4937 struct group_dpif *group = group_dpif_cast(group_);
4939 /* Start from historical data for 'group' itself that are no longer tracked
4940 * in facets. This counts, for example, facets that have expired. */
4941 ovs_mutex_lock(&group->stats_mutex);
4942 ogs->packet_count = group->packet_count;
4943 ogs->byte_count = group->byte_count;
4944 memcpy(ogs->bucket_stats, group->bucket_stats,
4945 group->up.n_buckets * sizeof *group->bucket_stats);
4946 ovs_mutex_unlock(&group->stats_mutex);
4952 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4953 struct group_dpif **group)
4954 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4956 struct ofgroup *ofgroup;
4960 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4961 *group = found ? group_dpif_cast(ofgroup) : NULL;
4967 group_dpif_release(struct group_dpif *group)
4968 OVS_RELEASES(group->up.rwlock)
4970 ofproto_group_release(&group->up);
4974 group_dpif_get_buckets(const struct group_dpif *group,
4975 const struct list **buckets)
4977 *buckets = &group->up.buckets;
4980 enum ofp11_group_type
4981 group_dpif_get_type(const struct group_dpif *group)
4983 return group->up.type;
4986 /* Sends 'packet' out 'ofport'.
4987 * May modify 'packet'.
4988 * Returns 0 if successful, otherwise a positive errno value. */
4990 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4995 error = xlate_send_packet(ofport, packet);
4997 ovs_mutex_lock(&ofproto->stats_mutex);
4998 ofproto->stats.tx_packets++;
4999 ofproto->stats.tx_bytes += packet->size;
5000 ovs_mutex_unlock(&ofproto->stats_mutex);
5004 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5005 * The action will state 'slow' as the reason that the action is in the slow
5006 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5007 * dump-flows" output to see why a flow is in the slow path.)
5009 * The 'stub_size' bytes in 'stub' will be used to store the action.
5010 * 'stub_size' must be large enough for the action.
5012 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5015 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5016 enum slow_path_reason slow,
5017 uint64_t *stub, size_t stub_size,
5018 const struct nlattr **actionsp, size_t *actions_lenp)
5020 union user_action_cookie cookie;
5023 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5024 cookie.slow_path.unused = 0;
5025 cookie.slow_path.reason = slow;
5027 ofpbuf_use_stack(&buf, stub, stub_size);
5028 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5029 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5031 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5033 odp_port_t odp_port;
5036 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5037 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5038 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5040 *actionsp = buf.data;
5041 *actions_lenp = buf.size;
5045 set_frag_handling(struct ofproto *ofproto_,
5046 enum ofp_config_flags frag_handling)
5048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5049 if (frag_handling != OFPC_FRAG_REASM) {
5050 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5058 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5059 const struct flow *flow,
5060 const struct ofpact *ofpacts, size_t ofpacts_len)
5062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5064 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5065 ofpacts_len, packet);
5072 set_netflow(struct ofproto *ofproto_,
5073 const struct netflow_options *netflow_options)
5075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5077 if (netflow_options) {
5078 if (!ofproto->netflow) {
5079 ofproto->netflow = netflow_create();
5080 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5082 return netflow_set_options(ofproto->netflow, netflow_options);
5083 } else if (ofproto->netflow) {
5084 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5085 netflow_destroy(ofproto->netflow);
5086 ofproto->netflow = NULL;
5093 get_netflow_ids(const struct ofproto *ofproto_,
5094 uint8_t *engine_type, uint8_t *engine_id)
5096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5098 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5101 static struct ofproto_dpif *
5102 ofproto_dpif_lookup(const char *name)
5104 struct ofproto_dpif *ofproto;
5106 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5107 hash_string(name, 0), &all_ofproto_dpifs) {
5108 if (!strcmp(ofproto->up.name, name)) {
5116 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5117 const char *argv[], void *aux OVS_UNUSED)
5119 struct ofproto_dpif *ofproto;
5122 ofproto = ofproto_dpif_lookup(argv[1]);
5124 unixctl_command_reply_error(conn, "no such bridge");
5127 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5128 mac_learning_flush(ofproto->ml);
5129 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5131 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5132 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5133 mac_learning_flush(ofproto->ml);
5134 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5138 unixctl_command_reply(conn, "table successfully flushed");
5141 static struct ofport_dpif *
5142 ofbundle_get_a_port(const struct ofbundle *bundle)
5144 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5149 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5150 const char *argv[], void *aux OVS_UNUSED)
5152 struct ds ds = DS_EMPTY_INITIALIZER;
5153 const struct ofproto_dpif *ofproto;
5154 const struct mac_entry *e;
5156 ofproto = ofproto_dpif_lookup(argv[1]);
5158 unixctl_command_reply_error(conn, "no such bridge");
5162 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5163 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5164 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5165 struct ofbundle *bundle = e->port.p;
5166 char name[OFP_MAX_PORT_NAME_LEN];
5168 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5170 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5171 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5172 mac_entry_age(ofproto->ml, e));
5174 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5175 unixctl_command_reply(conn, ds_cstr(&ds));
5180 struct xlate_out xout;
5181 struct xlate_in xin;
5187 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5189 struct rule_actions *actions;
5192 ds_put_char_multiple(result, '\t', level);
5194 ds_put_cstr(result, "No match\n");
5198 ovs_mutex_lock(&rule->up.mutex);
5199 cookie = rule->up.flow_cookie;
5200 ovs_mutex_unlock(&rule->up.mutex);
5202 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5203 rule ? rule->up.table_id : 0, ntohll(cookie));
5204 cls_rule_format(&rule->up.cr, result);
5205 ds_put_char(result, '\n');
5207 actions = rule_dpif_get_actions(rule);
5209 ds_put_char_multiple(result, '\t', level);
5210 ds_put_cstr(result, "OpenFlow actions=");
5211 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5212 ds_put_char(result, '\n');
5214 rule_actions_unref(actions);
5218 trace_format_flow(struct ds *result, int level, const char *title,
5219 struct trace_ctx *trace)
5221 ds_put_char_multiple(result, '\t', level);
5222 ds_put_format(result, "%s: ", title);
5223 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5224 ds_put_cstr(result, "unchanged");
5226 flow_format(result, &trace->xin.flow);
5227 trace->flow = trace->xin.flow;
5229 ds_put_char(result, '\n');
5233 trace_format_regs(struct ds *result, int level, const char *title,
5234 struct trace_ctx *trace)
5238 ds_put_char_multiple(result, '\t', level);
5239 ds_put_format(result, "%s:", title);
5240 for (i = 0; i < FLOW_N_REGS; i++) {
5241 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5243 ds_put_char(result, '\n');
5247 trace_format_odp(struct ds *result, int level, const char *title,
5248 struct trace_ctx *trace)
5250 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5252 ds_put_char_multiple(result, '\t', level);
5253 ds_put_format(result, "%s: ", title);
5254 format_odp_actions(result, odp_actions->data, odp_actions->size);
5255 ds_put_char(result, '\n');
5259 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5261 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5262 struct ds *result = trace->result;
5264 ds_put_char(result, '\n');
5265 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5266 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5267 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5268 trace_format_rule(result, recurse + 1, rule);
5272 trace_report(struct xlate_in *xin, const char *s, int recurse)
5274 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5275 struct ds *result = trace->result;
5277 ds_put_char_multiple(result, '\t', recurse);
5278 ds_put_cstr(result, s);
5279 ds_put_char(result, '\n');
5282 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
5283 * forms are supported:
5285 * - [dpname] odp_flow [-generate | packet]
5286 * - bridge br_flow [-generate | packet]
5288 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
5289 * returns a nonnull error message. */
5291 parse_flow_and_packet(int argc, const char *argv[],
5292 struct ofproto_dpif **ofprotop, struct flow *flow,
5293 struct ofpbuf **packetp)
5295 const struct dpif_backer *backer = NULL;
5296 const char *error = NULL;
5297 struct simap port_names = SIMAP_INITIALIZER(&port_names);
5298 struct ofpbuf *packet;
5299 struct ofpbuf odp_key;
5300 struct ofpbuf odp_mask;
5302 ofpbuf_init(&odp_key, 0);
5303 ofpbuf_init(&odp_mask, 0);
5305 /* Handle "-generate" or a hex string as the last argument. */
5306 if (!strcmp(argv[argc - 1], "-generate")) {
5307 packet = ofpbuf_new(0);
5310 error = eth_from_hex(argv[argc - 1], &packet);
5313 } else if (argc == 4) {
5314 /* The 3-argument form must end in "-generate' or a hex string. */
5319 /* odp_flow can have its in_port specified as a name instead of port no.
5320 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5321 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5322 * we need to create a simap of name to port no. */
5324 const char *dp_type;
5325 if (!strncmp(argv[1], "ovs-", 4)) {
5326 dp_type = argv[1] + 4;
5330 backer = shash_find_data(&all_dpif_backers, dp_type);
5331 } else if (argc == 2) {
5332 struct shash_node *node;
5333 if (shash_count(&all_dpif_backers) == 1) {
5334 node = shash_first(&all_dpif_backers);
5335 backer = node->data;
5338 error = "Syntax error";
5341 if (backer && backer->dpif) {
5342 struct dpif_port dpif_port;
5343 struct dpif_port_dump port_dump;
5344 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5345 simap_put(&port_names, dpif_port.name,
5346 odp_to_u32(dpif_port.port_no));
5350 /* Parse the flow and determine whether a datapath or
5351 * bridge is specified. If function odp_flow_key_from_string()
5352 * returns 0, the flow is a odp_flow. If function
5353 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5354 if (!odp_flow_from_string(argv[argc - 1], &port_names,
5355 &odp_key, &odp_mask)) {
5357 error = "Cannot find the datapath";
5361 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5362 NULL, ofprotop, NULL)) {
5363 error = "Invalid datapath flow";
5366 } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5368 error = "Must specify bridge name";
5372 *ofprotop = ofproto_dpif_lookup(argv[1]);
5374 error = "Unknown bridge name";
5378 error = "Bad flow syntax";
5382 /* Generate a packet, if requested. */
5384 if (!packet->size) {
5385 flow_compose(packet, flow);
5387 union flow_in_port in_port = flow->in_port;
5389 /* Use the metadata from the flow and the packet argument
5390 * to reconstruct the flow. */
5391 flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5400 ofpbuf_delete(packet);
5404 ofpbuf_uninit(&odp_key);
5405 ofpbuf_uninit(&odp_mask);
5406 simap_destroy(&port_names);
5411 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5412 void *aux OVS_UNUSED)
5414 struct ofproto_dpif *ofproto;
5415 struct ofpbuf *packet;
5419 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5424 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5425 unixctl_command_reply(conn, ds_cstr(&result));
5426 ds_destroy(&result);
5427 ofpbuf_delete(packet);
5429 unixctl_command_reply_error(conn, error);
5434 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5435 const char *argv[], void *aux OVS_UNUSED)
5437 enum ofputil_protocol usable_protocols;
5438 struct ofproto_dpif *ofproto;
5439 bool enforce_consistency;
5440 struct ofpbuf ofpacts;
5441 struct ofpbuf *packet;
5446 /* Three kinds of error return values! */
5453 ofpbuf_init(&ofpacts, 0);
5455 /* Parse actions. */
5456 rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5458 unixctl_command_reply_error(conn, rw_error);
5463 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5464 * consistency between the flow and the actions. With -consistent, we
5465 * enforce consistency even for a flow supported in OpenFlow 1.0. */
5466 if (!strcmp(argv[1], "-consistent")) {
5467 enforce_consistency = true;
5471 enforce_consistency = false;
5474 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5476 unixctl_command_reply_error(conn, error);
5480 /* Do the same checks as handle_packet_out() in ofproto.c.
5482 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5483 * strictly correct because these actions aren't in any table, but it's OK
5484 * because it 'table_id' is used only to check goto_table instructions, but
5485 * packet-outs take a list of actions and therefore it can't include
5488 * We skip the "meter" check here because meter is an instruction, not an
5489 * action, and thus cannot appear in ofpacts. */
5490 in_port = ofp_to_u16(flow.in_port.ofp_port);
5491 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5492 unixctl_command_reply_error(conn, "invalid in_port");
5495 if (enforce_consistency) {
5496 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5497 u16_to_ofp(ofproto->up.max_ports),
5498 0, 0, usable_protocols);
5500 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5501 u16_to_ofp(ofproto->up.max_ports), 0, 0,
5507 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5508 unixctl_command_reply_error(conn, ds_cstr(&result));
5512 ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5513 unixctl_command_reply(conn, ds_cstr(&result));
5516 ds_destroy(&result);
5517 ofpbuf_delete(packet);
5518 ofpbuf_uninit(&ofpacts);
5521 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5522 * description of the results to 'ds'.
5524 * The trace follows a packet with the specified 'flow' through the flow
5525 * table. 'packet' may be nonnull to trace an actual packet, with consequent
5526 * side effects (if it is nonnull then its flow must be 'flow').
5528 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5529 * trace, otherwise the actions are determined by a flow table lookup. */
5531 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5532 const struct ofpbuf *packet,
5533 const struct ofpact ofpacts[], size_t ofpacts_len,
5536 struct rule_dpif *rule;
5537 struct flow_wildcards wc;
5539 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5540 ds_put_cstr(ds, "Flow: ");
5541 flow_format(ds, flow);
5542 ds_put_char(ds, '\n');
5544 flow_wildcards_init_catchall(&wc);
5548 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5550 trace_format_rule(ds, 0, rule);
5551 if (rule == ofproto->miss_rule) {
5552 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5553 } else if (rule == ofproto->no_packet_in_rule) {
5554 ds_put_cstr(ds, "\nNo match, packets dropped because "
5555 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5556 } else if (rule == ofproto->drop_frags_rule) {
5557 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5558 "and the fragment handling mode is \"drop\".\n");
5562 if (rule || ofpacts) {
5563 uint64_t odp_actions_stub[1024 / 8];
5564 struct ofpbuf odp_actions;
5565 struct trace_ctx trace;
5569 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5572 ofpbuf_use_stub(&odp_actions,
5573 odp_actions_stub, sizeof odp_actions_stub);
5574 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5576 trace.xin.ofpacts = ofpacts;
5577 trace.xin.ofpacts_len = ofpacts_len;
5579 trace.xin.resubmit_hook = trace_resubmit;
5580 trace.xin.report_hook = trace_report;
5582 xlate_actions(&trace.xin, &trace.xout);
5583 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5585 ds_put_char(ds, '\n');
5586 trace_format_flow(ds, 0, "Final flow", &trace);
5588 match_init(&match, flow, &trace.xout.wc);
5589 ds_put_cstr(ds, "Relevant fields: ");
5590 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5591 ds_put_char(ds, '\n');
5593 ds_put_cstr(ds, "Datapath actions: ");
5594 format_odp_actions(ds, trace.xout.odp_actions.data,
5595 trace.xout.odp_actions.size);
5597 if (trace.xout.slow) {
5598 enum slow_path_reason slow;
5600 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5601 "slow path because it:");
5603 slow = trace.xout.slow;
5605 enum slow_path_reason bit = rightmost_1bit(slow);
5607 ds_put_format(ds, "\n\t- %s.",
5608 slow_path_reason_to_explanation(bit));
5614 xlate_out_uninit(&trace.xout);
5617 rule_dpif_unref(rule);
5620 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5621 * 'reply' describing the results. */
5623 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5625 struct cls_cursor cursor;
5626 struct facet *facet;
5630 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5631 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5632 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5633 if (!facet_check_consistency(facet)) {
5637 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5639 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5643 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5644 ofproto->up.name, errors);
5646 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5651 ofproto_dpif_self_check(struct unixctl_conn *conn,
5652 int argc, const char *argv[], void *aux OVS_UNUSED)
5654 struct ds reply = DS_EMPTY_INITIALIZER;
5655 struct ofproto_dpif *ofproto;
5658 ofproto = ofproto_dpif_lookup(argv[1]);
5660 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5661 "ofproto/list for help)");
5664 ofproto_dpif_self_check__(ofproto, &reply);
5666 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5667 ofproto_dpif_self_check__(ofproto, &reply);
5671 unixctl_command_reply(conn, ds_cstr(&reply));
5675 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5676 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5677 * to destroy 'ofproto_shash' and free the returned value. */
5678 static const struct shash_node **
5679 get_ofprotos(struct shash *ofproto_shash)
5681 const struct ofproto_dpif *ofproto;
5683 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5684 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5685 shash_add_nocopy(ofproto_shash, name, ofproto);
5688 return shash_sort(ofproto_shash);
5692 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5693 const char *argv[] OVS_UNUSED,
5694 void *aux OVS_UNUSED)
5696 struct ds ds = DS_EMPTY_INITIALIZER;
5697 struct shash ofproto_shash;
5698 const struct shash_node **sorted_ofprotos;
5701 shash_init(&ofproto_shash);
5702 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5703 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5704 const struct shash_node *node = sorted_ofprotos[i];
5705 ds_put_format(&ds, "%s\n", node->name);
5708 shash_destroy(&ofproto_shash);
5709 free(sorted_ofprotos);
5711 unixctl_command_reply(conn, ds_cstr(&ds));
5716 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5718 const struct shash_node **ofprotos;
5719 struct ofproto_dpif *ofproto;
5720 struct shash ofproto_shash;
5721 uint64_t n_hit, n_missed;
5724 n_hit = n_missed = 0;
5725 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5726 if (ofproto->backer == backer) {
5727 n_missed += ofproto->n_missed;
5728 n_hit += ofproto->n_hit;
5732 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5733 dpif_name(backer->dpif), n_hit, n_missed);
5734 ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5735 " life span: %lldms\n", hmap_count(&backer->subfacets),
5736 backer->avg_n_subfacet, backer->max_n_subfacet,
5737 backer->avg_subfacet_life);
5739 shash_init(&ofproto_shash);
5740 ofprotos = get_ofprotos(&ofproto_shash);
5741 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5742 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5743 const struct shash_node **ports;
5746 if (ofproto->backer != backer) {
5750 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5751 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5753 ports = shash_sort(&ofproto->up.port_by_name);
5754 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5755 const struct shash_node *node = ports[j];
5756 struct ofport *ofport = node->data;
5758 odp_port_t odp_port;
5760 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5763 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5764 if (odp_port != ODPP_NONE) {
5765 ds_put_format(ds, "%"PRIu32":", odp_port);
5767 ds_put_cstr(ds, "none:");
5770 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5773 if (!netdev_get_config(ofport->netdev, &config)) {
5774 const struct smap_node **nodes;
5777 nodes = smap_sort(&config);
5778 for (i = 0; i < smap_count(&config); i++) {
5779 const struct smap_node *node = nodes[i];
5780 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5781 node->key, node->value);
5785 smap_destroy(&config);
5787 ds_put_char(ds, ')');
5788 ds_put_char(ds, '\n');
5792 shash_destroy(&ofproto_shash);
5797 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5798 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5800 struct ds ds = DS_EMPTY_INITIALIZER;
5801 const struct shash_node **backers;
5804 backers = shash_sort(&all_dpif_backers);
5805 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5806 dpif_show_backer(backers[i]->data, &ds);
5810 unixctl_command_reply(conn, ds_cstr(&ds));
5814 /* Dump the megaflow (facet) cache. This is useful to check the
5815 * correctness of flow wildcarding, since the same mechanism is used for
5816 * both xlate caching and kernel wildcarding.
5818 * It's important to note that in the output the flow description uses
5819 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5821 * This command is only needed for advanced debugging, so it's not
5822 * documented in the man page. */
5824 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5825 int argc OVS_UNUSED, const char *argv[],
5826 void *aux OVS_UNUSED)
5828 struct ds ds = DS_EMPTY_INITIALIZER;
5829 const struct ofproto_dpif *ofproto;
5830 long long int now = time_msec();
5831 struct cls_cursor cursor;
5832 struct facet *facet;
5834 ofproto = ofproto_dpif_lookup(argv[1]);
5836 unixctl_command_reply_error(conn, "no such bridge");
5840 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5841 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5842 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5843 cls_rule_format(&facet->cr, &ds);
5844 ds_put_cstr(&ds, ", ");
5845 ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5846 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5847 ds_put_cstr(&ds, "Datapath actions: ");
5848 if (facet->xout.slow) {
5849 uint64_t slow_path_stub[128 / 8];
5850 const struct nlattr *actions;
5853 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5854 slow_path_stub, sizeof slow_path_stub,
5855 &actions, &actions_len);
5856 format_odp_actions(&ds, actions, actions_len);
5858 format_odp_actions(&ds, facet->xout.odp_actions.data,
5859 facet->xout.odp_actions.size);
5861 ds_put_cstr(&ds, "\n");
5863 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5865 ds_chomp(&ds, '\n');
5866 unixctl_command_reply(conn, ds_cstr(&ds));
5870 /* Disable using the megaflows.
5872 * This command is only needed for advanced debugging, so it's not
5873 * documented in the man page. */
5875 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5876 int argc OVS_UNUSED,
5877 const char *argv[] OVS_UNUSED,
5878 void *aux OVS_UNUSED)
5880 struct ofproto_dpif *ofproto;
5882 enable_megaflows = false;
5884 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5885 flush(&ofproto->up);
5888 unixctl_command_reply(conn, "megaflows disabled");
5891 /* Re-enable using megaflows.
5893 * This command is only needed for advanced debugging, so it's not
5894 * documented in the man page. */
5896 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5897 int argc OVS_UNUSED,
5898 const char *argv[] OVS_UNUSED,
5899 void *aux OVS_UNUSED)
5901 struct ofproto_dpif *ofproto;
5903 enable_megaflows = true;
5905 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5906 flush(&ofproto->up);
5909 unixctl_command_reply(conn, "megaflows enabled");
5913 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5914 int argc OVS_UNUSED, const char *argv[],
5915 void *aux OVS_UNUSED)
5917 struct ds ds = DS_EMPTY_INITIALIZER;
5918 const struct ofproto_dpif *ofproto;
5919 struct subfacet *subfacet;
5921 ofproto = ofproto_dpif_lookup(argv[1]);
5923 unixctl_command_reply_error(conn, "no such bridge");
5927 update_stats(ofproto->backer);
5929 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5930 struct facet *facet = subfacet->facet;
5931 struct odputil_keybuf maskbuf;
5934 if (facet->ofproto != ofproto) {
5938 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5939 if (enable_megaflows) {
5940 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5941 &facet->flow, UINT32_MAX);
5944 odp_flow_format(subfacet->key, subfacet->key_len,
5945 mask.data, mask.size, NULL, &ds, false);
5947 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5948 subfacet->dp_packet_count, subfacet->dp_byte_count);
5949 if (subfacet->used) {
5950 ds_put_format(&ds, "%.3fs",
5951 (time_msec() - subfacet->used) / 1000.0);
5953 ds_put_format(&ds, "never");
5955 if (subfacet->facet->tcp_flags) {
5956 ds_put_cstr(&ds, ", flags:");
5957 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5960 ds_put_cstr(&ds, ", actions:");
5961 if (facet->xout.slow) {
5962 uint64_t slow_path_stub[128 / 8];
5963 const struct nlattr *actions;
5966 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5967 slow_path_stub, sizeof slow_path_stub,
5968 &actions, &actions_len);
5969 format_odp_actions(&ds, actions, actions_len);
5971 format_odp_actions(&ds, facet->xout.odp_actions.data,
5972 facet->xout.odp_actions.size);
5974 ds_put_char(&ds, '\n');
5977 unixctl_command_reply(conn, ds_cstr(&ds));
5982 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5983 int argc OVS_UNUSED, const char *argv[],
5984 void *aux OVS_UNUSED)
5986 struct ds ds = DS_EMPTY_INITIALIZER;
5987 struct ofproto_dpif *ofproto;
5989 ofproto = ofproto_dpif_lookup(argv[1]);
5991 unixctl_command_reply_error(conn, "no such bridge");
5995 flush(&ofproto->up);
5997 unixctl_command_reply(conn, ds_cstr(&ds));
6002 ofproto_dpif_unixctl_init(void)
6004 static bool registered;
6010 unixctl_command_register(
6012 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6013 1, 3, ofproto_unixctl_trace, NULL);
6014 unixctl_command_register(
6015 "ofproto/trace-packet-out",
6016 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6017 2, 6, ofproto_unixctl_trace_actions, NULL);
6018 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6019 ofproto_unixctl_fdb_flush, NULL);
6020 unixctl_command_register("fdb/show", "bridge", 1, 1,
6021 ofproto_unixctl_fdb_show, NULL);
6022 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6023 ofproto_dpif_self_check, NULL);
6024 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6025 ofproto_unixctl_dpif_dump_dps, NULL);
6026 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6028 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6029 ofproto_unixctl_dpif_dump_flows, NULL);
6030 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6031 ofproto_unixctl_dpif_del_flows, NULL);
6032 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6033 ofproto_unixctl_dpif_dump_megaflows, NULL);
6034 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6035 ofproto_unixctl_dpif_disable_megaflows, NULL);
6036 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6037 ofproto_unixctl_dpif_enable_megaflows, NULL);
6040 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6042 * This is deprecated. It is only for compatibility with broken device drivers
6043 * in old versions of Linux that do not properly support VLANs when VLAN
6044 * devices are not used. When broken device drivers are no longer in
6045 * widespread use, we will delete these interfaces. */
6048 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6051 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6053 if (realdev_ofp_port == ofport->realdev_ofp_port
6054 && vid == ofport->vlandev_vid) {
6058 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6060 if (ofport->realdev_ofp_port) {
6063 if (realdev_ofp_port && ofport->bundle) {
6064 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6065 * themselves be part of a bundle. */
6066 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6069 ofport->realdev_ofp_port = realdev_ofp_port;
6070 ofport->vlandev_vid = vid;
6072 if (realdev_ofp_port) {
6073 vsp_add(ofport, realdev_ofp_port, vid);
6080 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6082 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6086 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6087 OVS_EXCLUDED(ofproto->vsp_mutex)
6091 ovs_mutex_lock(&ofproto->vsp_mutex);
6092 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6093 ovs_mutex_unlock(&ofproto->vsp_mutex);
6098 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6099 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6100 OVS_REQUIRES(ofproto->vsp_mutex)
6102 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6103 int vid = vlan_tci_to_vid(vlan_tci);
6104 const struct vlan_splinter *vsp;
6106 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6107 hash_realdev_vid(realdev_ofp_port, vid),
6108 &ofproto->realdev_vid_map) {
6109 if (vsp->realdev_ofp_port == realdev_ofp_port
6110 && vsp->vid == vid) {
6111 return vsp->vlandev_ofp_port;
6115 return realdev_ofp_port;
6118 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6119 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6120 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6121 * 'vlan_tci' 9, it would return the port number of eth0.9.
6123 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6124 * function just returns its 'realdev_ofp_port' argument. */
6126 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6127 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6128 OVS_EXCLUDED(ofproto->vsp_mutex)
6132 ovs_mutex_lock(&ofproto->vsp_mutex);
6133 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6134 ovs_mutex_unlock(&ofproto->vsp_mutex);
6138 static struct vlan_splinter *
6139 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6141 struct vlan_splinter *vsp;
6143 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6144 hash_ofp_port(vlandev_ofp_port),
6145 &ofproto->vlandev_map) {
6146 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6154 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6155 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6156 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6157 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6158 * eth0 and store 9 in '*vid'.
6160 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6161 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6164 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6165 ofp_port_t vlandev_ofp_port, int *vid)
6166 OVS_REQUIRES(ofproto->vsp_mutex)
6168 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6169 const struct vlan_splinter *vsp;
6171 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6176 return vsp->realdev_ofp_port;
6182 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6183 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6184 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6185 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6186 * always the case unless VLAN splinters are enabled), returns false without
6187 * making any changes. */
6189 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6190 OVS_EXCLUDED(ofproto->vsp_mutex)
6195 ovs_mutex_lock(&ofproto->vsp_mutex);
6196 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6197 ovs_mutex_unlock(&ofproto->vsp_mutex);
6202 /* Cause the flow to be processed as if it came in on the real device with
6203 * the VLAN device's VLAN ID. */
6204 flow->in_port.ofp_port = realdev;
6205 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6210 vsp_remove(struct ofport_dpif *port)
6212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6213 struct vlan_splinter *vsp;
6215 ovs_mutex_lock(&ofproto->vsp_mutex);
6216 vsp = vlandev_find(ofproto, port->up.ofp_port);
6218 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6219 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6222 port->realdev_ofp_port = 0;
6224 VLOG_ERR("missing vlan device record");
6226 ovs_mutex_unlock(&ofproto->vsp_mutex);
6230 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6234 ovs_mutex_lock(&ofproto->vsp_mutex);
6235 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6236 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6237 == realdev_ofp_port)) {
6238 struct vlan_splinter *vsp;
6240 vsp = xmalloc(sizeof *vsp);
6241 vsp->realdev_ofp_port = realdev_ofp_port;
6242 vsp->vlandev_ofp_port = port->up.ofp_port;
6245 port->realdev_ofp_port = realdev_ofp_port;
6247 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6248 hash_ofp_port(port->up.ofp_port));
6249 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6250 hash_realdev_vid(realdev_ofp_port, vid));
6252 VLOG_ERR("duplicate vlan device record");
6254 ovs_mutex_unlock(&ofproto->vsp_mutex);
6258 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6260 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6261 return ofport ? ofport->odp_port : ODPP_NONE;
6264 struct ofport_dpif *
6265 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6267 struct ofport_dpif *port;
6269 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6270 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6271 &backer->odp_to_ofport_map) {
6272 if (port->odp_port == odp_port) {
6273 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6278 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6283 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6285 struct ofport_dpif *port;
6287 port = odp_port_to_ofport(ofproto->backer, odp_port);
6288 if (port && &ofproto->up == port->up.ofproto) {
6289 return port->up.ofp_port;
6295 const struct ofproto_class ofproto_dpif_class = {
6330 port_is_lacp_current,
6331 NULL, /* rule_choose_table */
6340 rule_modify_actions,
6354 get_stp_port_status,
6362 is_mirror_output_bundle,
6363 forward_bpdu_changed,
6364 set_mac_table_config,
6366 NULL, /* meter_get_features */
6367 NULL, /* meter_set */
6368 NULL, /* meter_get */
6369 NULL, /* meter_del */
6370 group_alloc, /* group_alloc */
6371 group_construct, /* group_construct */
6372 group_destruct, /* group_destruct */
6373 group_dealloc, /* group_dealloc */
6374 group_modify, /* group_modify */
6375 group_get_stats, /* group_get_stats */