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 uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */
291 struct xlate_out xout;
293 /* Storage for a single subfacet, to reduce malloc() time and space
294 * overhead. (A facet always has at least one subfacet and in the common
295 * case has exactly one subfacet. However, 'one_subfacet' may not
296 * always be valid, since it could have been removed after newer
297 * subfacets were pushed onto the 'subfacets' list.) */
298 struct subfacet one_subfacet;
300 long long int learn_rl; /* Rate limiter for facet_learn(). */
303 static struct facet *facet_create(const struct flow_miss *);
304 static void facet_remove(struct facet *);
305 static void facet_free(struct facet *);
307 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
308 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
309 const struct flow *);
310 static bool facet_revalidate(struct facet *);
311 static bool facet_check_consistency(struct facet *);
313 static void facet_flush_stats(struct facet *);
315 static void facet_reset_counters(struct facet *);
316 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
317 struct dpif_flow_stats *, bool may_learn);
318 static void facet_push_stats(struct facet *, bool may_learn);
319 static void facet_learn(struct facet *);
320 static void push_all_stats(void);
322 static bool facet_is_controller_flow(struct facet *);
325 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
329 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
330 struct list bundle_node; /* In struct ofbundle's "ports" list. */
331 struct cfm *cfm; /* Connectivity Fault Management, if any. */
332 struct bfd *bfd; /* BFD, if any. */
333 bool may_enable; /* May be enabled in bonds. */
334 bool is_tunnel; /* This port is a tunnel. */
335 bool is_layer3; /* This is a layer 3 port. */
336 long long int carrier_seq; /* Carrier status changes. */
337 struct ofport_dpif *peer; /* Peer if patch port. */
340 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
341 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
342 long long int stp_state_entered;
344 /* Queue to DSCP mapping. */
345 struct ofproto_port_queue *qdscp;
348 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
350 * This is deprecated. It is only for compatibility with broken device
351 * drivers in old versions of Linux that do not properly support VLANs when
352 * VLAN devices are not used. When broken device drivers are no longer in
353 * widespread use, we will delete these interfaces. */
354 ofp_port_t realdev_ofp_port;
358 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
360 * This is deprecated. It is only for compatibility with broken device drivers
361 * in old versions of Linux that do not properly support VLANs when VLAN
362 * devices are not used. When broken device drivers are no longer in
363 * widespread use, we will delete these interfaces. */
364 struct vlan_splinter {
365 struct hmap_node realdev_vid_node;
366 struct hmap_node vlandev_node;
367 ofp_port_t realdev_ofp_port;
368 ofp_port_t vlandev_ofp_port;
372 static void vsp_remove(struct ofport_dpif *);
373 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
375 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
378 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
381 static struct ofport_dpif *
382 ofport_dpif_cast(const struct ofport *ofport)
384 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
387 static void port_run(struct ofport_dpif *);
388 static int set_bfd(struct ofport *, const struct smap *);
389 static int set_cfm(struct ofport *, const struct cfm_settings *);
390 static void ofport_update_peer(struct ofport_dpif *);
392 struct dpif_completion {
393 struct list list_node;
394 struct ofoperation *op;
397 /* Reasons that we might need to revalidate every facet, and corresponding
400 * A value of 0 means that there is no need to revalidate.
402 * It would be nice to have some cleaner way to integrate with coverage
403 * counters, but with only a few reasons I guess this is good enough for
405 enum revalidate_reason {
406 REV_RECONFIGURE = 1, /* Switch configuration changed. */
407 REV_STP, /* Spanning tree protocol port status change. */
408 REV_BOND, /* Bonding changed. */
409 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
410 REV_FLOW_TABLE, /* Flow table changed. */
411 REV_MAC_LEARNING, /* Mac learning changed. */
412 REV_INCONSISTENCY /* Facet self-check failed. */
414 COVERAGE_DEFINE(rev_reconfigure);
415 COVERAGE_DEFINE(rev_stp);
416 COVERAGE_DEFINE(rev_bond);
417 COVERAGE_DEFINE(rev_port_toggled);
418 COVERAGE_DEFINE(rev_flow_table);
419 COVERAGE_DEFINE(rev_mac_learning);
420 COVERAGE_DEFINE(rev_inconsistency);
422 /* All datapaths of a given type share a single dpif backer instance. */
428 struct timer next_expiration;
430 struct ovs_rwlock odp_to_ofport_lock;
431 struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
433 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
435 /* Facet revalidation flags applying to facets which use this backer. */
436 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
438 struct hmap drop_keys; /* Set of dropped odp keys. */
439 bool recv_set_enable; /* Enables or disables receiving packets. */
441 struct hmap subfacets;
442 struct governor *governor;
444 /* Subfacet statistics.
446 * These keep track of the total number of subfacets added and deleted and
447 * flow life span. They are useful for computing the flow rates stats
448 * exposed via "ovs-appctl dpif/show". The goal is to learn about
449 * traffic patterns in ways that we can use later to improve Open vSwitch
450 * performance in new situations. */
451 unsigned max_n_subfacet; /* Maximum number of flows */
452 unsigned avg_n_subfacet; /* Average number of flows. */
453 long long int avg_subfacet_life; /* Average life span of subfacets. */
455 /* Number of upcall handling threads. */
456 unsigned int n_handler_threads;
459 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
460 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
462 static void drop_key_clear(struct dpif_backer *);
464 struct ofproto_dpif {
465 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
467 struct dpif_backer *backer;
469 /* Special OpenFlow rules. */
470 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
471 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
472 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
475 struct netflow *netflow;
476 struct dpif_sflow *sflow;
477 struct dpif_ipfix *ipfix;
478 struct hmap bundles; /* Contains "struct ofbundle"s. */
479 struct mac_learning *ml;
480 bool has_bonded_bundles;
482 struct mbridge *mbridge;
485 struct classifier facets; /* Contains 'struct facet's. */
486 long long int consistency_rl;
488 struct ovs_mutex stats_mutex;
489 struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
490 * consumed in userspace. */
494 long long int stp_last_tick;
496 /* VLAN splinters. */
497 struct ovs_mutex vsp_mutex;
498 struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
499 struct hmap vlandev_map OVS_GUARDED; /* vlandev -> (realdev,vid). */
502 struct sset ports; /* Set of standard port names. */
503 struct sset ghost_ports; /* Ports with no datapath port. */
504 struct sset port_poll_set; /* Queued names for port_poll() reply. */
505 int port_poll_errno; /* Last errno for port_poll() reply. */
506 uint64_t change_seq; /* Connectivity status changes. */
508 /* Per ofproto's dpif stats. */
513 struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */
516 /* By default, flows in the datapath are wildcarded (megaflows). They
517 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
518 static bool enable_megaflows = true;
520 /* All existing ofproto_dpif instances, indexed by ->up.name. */
521 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
523 static void ofproto_dpif_unixctl_init(void);
525 static inline struct ofproto_dpif *
526 ofproto_dpif_cast(const struct ofproto *ofproto)
528 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
529 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
532 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
533 ofp_port_t ofp_port);
534 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
535 const struct ofpbuf *packet,
536 const struct ofpact[], size_t ofpacts_len,
540 static void handle_upcalls(struct dpif_backer *);
542 /* Flow expiration. */
543 static int expire(struct dpif_backer *);
545 /* Global variables. */
546 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
548 /* Initial mappings of port to bridge mappings. */
549 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
551 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
554 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
555 struct ofputil_flow_mod *fm)
557 ofproto_flow_mod(&ofproto->up, fm);
560 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
561 * Takes ownership of 'pin' and pin->packet. */
563 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
564 struct ofproto_packet_in *pin)
566 if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
567 COVERAGE_INC(packet_in_overflow);
568 free(CONST_CAST(void *, pin->up.packet));
573 /* Factory functions. */
576 init(const struct shash *iface_hints)
578 struct shash_node *node;
580 /* Make a local copy, since we don't own 'iface_hints' elements. */
581 SHASH_FOR_EACH(node, iface_hints) {
582 const struct iface_hint *orig_hint = node->data;
583 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
585 new_hint->br_name = xstrdup(orig_hint->br_name);
586 new_hint->br_type = xstrdup(orig_hint->br_type);
587 new_hint->ofp_port = orig_hint->ofp_port;
589 shash_add(&init_ofp_ports, node->name, new_hint);
594 enumerate_types(struct sset *types)
596 dp_enumerate_types(types);
600 enumerate_names(const char *type, struct sset *names)
602 struct ofproto_dpif *ofproto;
605 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
606 if (strcmp(type, ofproto->up.type)) {
609 sset_add(names, ofproto->up.name);
616 del(const char *type, const char *name)
621 error = dpif_open(name, type, &dpif);
623 error = dpif_delete(dpif);
630 port_open_type(const char *datapath_type, const char *port_type)
632 return dpif_port_open_type(datapath_type, port_type);
635 /* Type functions. */
637 static void process_dpif_port_changes(struct dpif_backer *);
638 static void process_dpif_all_ports_changed(struct dpif_backer *);
639 static void process_dpif_port_change(struct dpif_backer *,
640 const char *devname);
641 static void process_dpif_port_error(struct dpif_backer *, int error);
643 static struct ofproto_dpif *
644 lookup_ofproto_dpif_by_port_name(const char *name)
646 struct ofproto_dpif *ofproto;
648 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
649 if (sset_contains(&ofproto->ports, name)) {
658 type_run(const char *type)
660 static long long int push_timer = LLONG_MIN;
661 struct dpif_backer *backer;
663 backer = shash_find_data(&all_dpif_backers, type);
665 /* This is not necessarily a problem, since backers are only
666 * created on demand. */
670 dpif_run(backer->dpif);
672 handle_upcalls(backer);
674 /* The most natural place to push facet statistics is when they're pulled
675 * from the datapath. However, when there are many flows in the datapath,
676 * this expensive operation can occur so frequently, that it reduces our
677 * ability to quickly set up flows. To reduce the cost, we push statistics
679 if (time_msec() > push_timer) {
680 push_timer = time_msec() + 2000;
684 /* If vswitchd started with other_config:flow_restore_wait set as "true",
685 * and the configuration has now changed to "false", enable receiving
686 * packets from the datapath. */
687 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
690 backer->recv_set_enable = true;
692 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
694 udpif_recv_set(backer->udpif, 0, false);
695 VLOG_ERR("Failed to enable receiving packets in dpif.");
698 udpif_recv_set(backer->udpif, n_handler_threads,
699 backer->recv_set_enable);
700 dpif_flow_flush(backer->dpif);
701 backer->need_revalidate = REV_RECONFIGURE;
704 /* If the n_handler_threads is reconfigured, call udpif_recv_set()
705 * to reset the handler threads. */
706 if (backer->n_handler_threads != n_handler_threads) {
707 udpif_recv_set(backer->udpif, n_handler_threads,
708 backer->recv_set_enable);
709 backer->n_handler_threads = n_handler_threads;
712 if (backer->need_revalidate) {
713 struct ofproto_dpif *ofproto;
714 struct simap_node *node;
715 struct simap tmp_backers;
717 /* Handle tunnel garbage collection. */
718 simap_init(&tmp_backers);
719 simap_swap(&backer->tnl_backers, &tmp_backers);
721 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
722 struct ofport_dpif *iter;
724 if (backer != ofproto->backer) {
728 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
729 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
732 if (!iter->is_tunnel) {
736 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
737 namebuf, sizeof namebuf);
738 node = simap_find(&tmp_backers, dp_port);
740 simap_put(&backer->tnl_backers, dp_port, node->data);
741 simap_delete(&tmp_backers, node);
742 node = simap_find(&backer->tnl_backers, dp_port);
744 node = simap_find(&backer->tnl_backers, dp_port);
746 odp_port_t odp_port = ODPP_NONE;
748 if (!dpif_port_add(backer->dpif, iter->up.netdev,
750 simap_put(&backer->tnl_backers, dp_port,
751 odp_to_u32(odp_port));
752 node = simap_find(&backer->tnl_backers, dp_port);
757 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
758 if (tnl_port_reconfigure(iter, iter->up.netdev,
760 backer->need_revalidate = REV_RECONFIGURE;
765 SIMAP_FOR_EACH (node, &tmp_backers) {
766 dpif_port_del(backer->dpif, u32_to_odp(node->data));
768 simap_destroy(&tmp_backers);
770 switch (backer->need_revalidate) {
771 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
772 case REV_STP: COVERAGE_INC(rev_stp); break;
773 case REV_BOND: COVERAGE_INC(rev_bond); break;
774 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
775 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
776 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
777 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
779 backer->need_revalidate = 0;
781 /* Clear the drop_keys in case we should now be accepting some
782 * formerly dropped flows. */
783 drop_key_clear(backer);
785 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
786 struct facet *facet, *next;
787 struct ofport_dpif *ofport;
788 struct cls_cursor cursor;
789 struct ofbundle *bundle;
791 if (ofproto->backer != backer) {
795 ovs_rwlock_wrlock(&xlate_rwlock);
796 xlate_ofproto_set(ofproto, ofproto->up.name,
797 ofproto->backer->dpif, ofproto->miss_rule,
798 ofproto->no_packet_in_rule, ofproto->ml,
799 ofproto->stp, ofproto->mbridge,
800 ofproto->sflow, ofproto->ipfix,
801 ofproto->netflow, ofproto->up.frag_handling,
802 ofproto->up.forward_bpdu,
803 connmgr_has_in_band(ofproto->up.connmgr));
805 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
806 xlate_bundle_set(ofproto, bundle, bundle->name,
807 bundle->vlan_mode, bundle->vlan,
808 bundle->trunks, bundle->use_priority_tags,
809 bundle->bond, bundle->lacp,
813 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
814 int stp_port = ofport->stp_port
815 ? stp_port_no(ofport->stp_port)
817 xlate_ofport_set(ofproto, ofport->bundle, ofport,
818 ofport->up.ofp_port, ofport->odp_port,
819 ofport->up.netdev, ofport->cfm,
820 ofport->bfd, ofport->peer, stp_port,
821 ofport->qdscp, ofport->n_qdscp,
822 ofport->up.pp.config, ofport->up.pp.state,
823 ofport->is_tunnel, ofport->may_enable);
825 ovs_rwlock_unlock(&xlate_rwlock);
827 /* Only ofproto-dpif cares about the facet classifier so we just
828 * lock cls_cursor_init() to appease the thread safety analysis. */
829 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
830 cls_cursor_init(&cursor, &ofproto->facets, NULL);
831 ovs_rwlock_unlock(&ofproto->facets.rwlock);
832 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
833 facet_revalidate(facet);
837 udpif_revalidate(backer->udpif);
840 if (!backer->recv_set_enable) {
841 /* Wake up before a max of 1000ms. */
842 timer_set_duration(&backer->next_expiration, 1000);
843 } else if (timer_expired(&backer->next_expiration)) {
844 int delay = expire(backer);
845 timer_set_duration(&backer->next_expiration, delay);
848 process_dpif_port_changes(backer);
850 if (backer->governor) {
853 governor_run(backer->governor);
855 /* If the governor has shrunk to its minimum size and the number of
856 * subfacets has dwindled, then drop the governor entirely.
858 * For hysteresis, the number of subfacets to drop the governor is
859 * smaller than the number needed to trigger its creation. */
860 n_subfacets = hmap_count(&backer->subfacets);
861 if (n_subfacets * 4 < flow_eviction_threshold
862 && governor_is_idle(backer->governor)) {
863 governor_destroy(backer->governor);
864 backer->governor = NULL;
871 /* Check for and handle port changes in 'backer''s dpif. */
873 process_dpif_port_changes(struct dpif_backer *backer)
879 error = dpif_port_poll(backer->dpif, &devname);
885 process_dpif_all_ports_changed(backer);
889 process_dpif_port_change(backer, devname);
894 process_dpif_port_error(backer, error);
901 process_dpif_all_ports_changed(struct dpif_backer *backer)
903 struct ofproto_dpif *ofproto;
904 struct dpif_port dpif_port;
905 struct dpif_port_dump dump;
906 struct sset devnames;
909 sset_init(&devnames);
910 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
911 if (ofproto->backer == backer) {
912 struct ofport *ofport;
914 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
915 sset_add(&devnames, netdev_get_name(ofport->netdev));
919 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
920 sset_add(&devnames, dpif_port.name);
923 SSET_FOR_EACH (devname, &devnames) {
924 process_dpif_port_change(backer, devname);
926 sset_destroy(&devnames);
930 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
932 struct ofproto_dpif *ofproto;
933 struct dpif_port port;
935 /* Don't report on the datapath's device. */
936 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
940 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
941 &all_ofproto_dpifs) {
942 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
947 ofproto = lookup_ofproto_dpif_by_port_name(devname);
948 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
949 /* The port was removed. If we know the datapath,
950 * report it through poll_set(). If we don't, it may be
951 * notifying us of a removal we initiated, so ignore it.
952 * If there's a pending ENOBUFS, let it stand, since
953 * everything will be reevaluated. */
954 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
955 sset_add(&ofproto->port_poll_set, devname);
956 ofproto->port_poll_errno = 0;
958 } else if (!ofproto) {
959 /* The port was added, but we don't know with which
960 * ofproto we should associate it. Delete it. */
961 dpif_port_del(backer->dpif, port.port_no);
963 struct ofport_dpif *ofport;
965 ofport = ofport_dpif_cast(shash_find_data(
966 &ofproto->up.port_by_name, devname));
968 && ofport->odp_port != port.port_no
969 && !odp_port_to_ofport(backer, port.port_no))
971 /* 'ofport''s datapath port number has changed from
972 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
973 * structures to match. */
974 ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
975 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
976 ofport->odp_port = port.port_no;
977 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
978 hash_odp_port(port.port_no));
979 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
980 backer->need_revalidate = REV_RECONFIGURE;
983 dpif_port_destroy(&port);
986 /* Propagate 'error' to all ofprotos based on 'backer'. */
988 process_dpif_port_error(struct dpif_backer *backer, int error)
990 struct ofproto_dpif *ofproto;
992 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
993 if (ofproto->backer == backer) {
994 sset_clear(&ofproto->port_poll_set);
995 ofproto->port_poll_errno = error;
1001 type_wait(const char *type)
1003 struct dpif_backer *backer;
1005 backer = shash_find_data(&all_dpif_backers, type);
1007 /* This is not necessarily a problem, since backers are only
1008 * created on demand. */
1012 if (backer->governor) {
1013 governor_wait(backer->governor);
1016 timer_wait(&backer->next_expiration);
1017 dpif_wait(backer->dpif);
1018 udpif_wait(backer->udpif);
1021 /* Basic life-cycle. */
1023 static int add_internal_flows(struct ofproto_dpif *);
1025 static struct ofproto *
1028 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1029 return &ofproto->up;
1033 dealloc(struct ofproto *ofproto_)
1035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1040 close_dpif_backer(struct dpif_backer *backer)
1042 ovs_assert(backer->refcount > 0);
1044 if (--backer->refcount) {
1048 drop_key_clear(backer);
1049 hmap_destroy(&backer->drop_keys);
1051 udpif_destroy(backer->udpif);
1053 simap_destroy(&backer->tnl_backers);
1054 ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1055 hmap_destroy(&backer->odp_to_ofport_map);
1056 shash_find_and_delete(&all_dpif_backers, backer->type);
1058 dpif_close(backer->dpif);
1060 ovs_assert(hmap_is_empty(&backer->subfacets));
1061 hmap_destroy(&backer->subfacets);
1062 governor_destroy(backer->governor);
1067 /* Datapath port slated for removal from datapath. */
1068 struct odp_garbage {
1069 struct list list_node;
1070 odp_port_t odp_port;
1074 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1076 struct dpif_backer *backer;
1077 struct dpif_port_dump port_dump;
1078 struct dpif_port port;
1079 struct shash_node *node;
1080 struct list garbage_list;
1081 struct odp_garbage *garbage, *next;
1087 backer = shash_find_data(&all_dpif_backers, type);
1094 backer_name = xasprintf("ovs-%s", type);
1096 /* Remove any existing datapaths, since we assume we're the only
1097 * userspace controlling the datapath. */
1099 dp_enumerate_names(type, &names);
1100 SSET_FOR_EACH(name, &names) {
1101 struct dpif *old_dpif;
1103 /* Don't remove our backer if it exists. */
1104 if (!strcmp(name, backer_name)) {
1108 if (dpif_open(name, type, &old_dpif)) {
1109 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1111 dpif_delete(old_dpif);
1112 dpif_close(old_dpif);
1115 sset_destroy(&names);
1117 backer = xmalloc(sizeof *backer);
1119 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1122 VLOG_ERR("failed to open datapath of type %s: %s", type,
1123 ovs_strerror(error));
1127 backer->udpif = udpif_create(backer, backer->dpif);
1129 backer->type = xstrdup(type);
1130 backer->governor = NULL;
1131 backer->refcount = 1;
1132 hmap_init(&backer->odp_to_ofport_map);
1133 ovs_rwlock_init(&backer->odp_to_ofport_lock);
1134 hmap_init(&backer->drop_keys);
1135 hmap_init(&backer->subfacets);
1136 timer_set_duration(&backer->next_expiration, 1000);
1137 backer->need_revalidate = 0;
1138 simap_init(&backer->tnl_backers);
1139 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1142 if (backer->recv_set_enable) {
1143 dpif_flow_flush(backer->dpif);
1146 /* Loop through the ports already on the datapath and remove any
1147 * that we don't need anymore. */
1148 list_init(&garbage_list);
1149 dpif_port_dump_start(&port_dump, backer->dpif);
1150 while (dpif_port_dump_next(&port_dump, &port)) {
1151 node = shash_find(&init_ofp_ports, port.name);
1152 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1153 garbage = xmalloc(sizeof *garbage);
1154 garbage->odp_port = port.port_no;
1155 list_push_front(&garbage_list, &garbage->list_node);
1158 dpif_port_dump_done(&port_dump);
1160 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1161 dpif_port_del(backer->dpif, garbage->odp_port);
1162 list_remove(&garbage->list_node);
1166 shash_add(&all_dpif_backers, type, backer);
1168 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1170 VLOG_ERR("failed to listen on datapath of type %s: %s",
1171 type, ovs_strerror(error));
1172 close_dpif_backer(backer);
1175 udpif_recv_set(backer->udpif, n_handler_threads,
1176 backer->recv_set_enable);
1177 backer->n_handler_threads = n_handler_threads;
1179 backer->max_n_subfacet = 0;
1180 backer->avg_n_subfacet = 0;
1181 backer->avg_subfacet_life = 0;
1187 construct(struct ofproto *ofproto_)
1189 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1190 struct shash_node *node, *next;
1193 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1198 ofproto->netflow = NULL;
1199 ofproto->sflow = NULL;
1200 ofproto->ipfix = NULL;
1201 ofproto->stp = NULL;
1202 hmap_init(&ofproto->bundles);
1203 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1204 ofproto->mbridge = mbridge_create();
1205 ofproto->has_bonded_bundles = false;
1206 ofproto->lacp_enabled = false;
1207 ovs_mutex_init(&ofproto->stats_mutex);
1208 ovs_mutex_init(&ofproto->vsp_mutex);
1210 classifier_init(&ofproto->facets, NULL);
1211 ofproto->consistency_rl = LLONG_MIN;
1213 guarded_list_init(&ofproto->pins);
1215 ofproto_dpif_unixctl_init();
1217 hmap_init(&ofproto->vlandev_map);
1218 hmap_init(&ofproto->realdev_vid_map);
1220 sset_init(&ofproto->ports);
1221 sset_init(&ofproto->ghost_ports);
1222 sset_init(&ofproto->port_poll_set);
1223 ofproto->port_poll_errno = 0;
1224 ofproto->change_seq = 0;
1226 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1227 struct iface_hint *iface_hint = node->data;
1229 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1230 /* Check if the datapath already has this port. */
1231 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1232 sset_add(&ofproto->ports, node->name);
1235 free(iface_hint->br_name);
1236 free(iface_hint->br_type);
1238 shash_delete(&init_ofp_ports, node);
1242 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1243 hash_string(ofproto->up.name, 0));
1244 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1246 ofproto_init_tables(ofproto_, N_TABLES);
1247 error = add_internal_flows(ofproto);
1248 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1251 ofproto->n_missed = 0;
1257 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1258 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1260 struct ofputil_flow_mod fm;
1263 match_init_catchall(&fm.match);
1265 match_set_reg(&fm.match, 0, id);
1266 fm.new_cookie = htonll(0);
1267 fm.cookie = htonll(0);
1268 fm.cookie_mask = htonll(0);
1269 fm.modify_cookie = false;
1270 fm.table_id = TBL_INTERNAL;
1271 fm.command = OFPFC_ADD;
1272 fm.idle_timeout = 0;
1273 fm.hard_timeout = 0;
1277 fm.ofpacts = ofpacts->data;
1278 fm.ofpacts_len = ofpacts->size;
1280 error = ofproto_flow_mod(&ofproto->up, &fm);
1282 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1283 id, ofperr_to_string(error));
1287 if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1289 rule_dpif_unref(*rulep);
1298 add_internal_flows(struct ofproto_dpif *ofproto)
1300 struct ofpact_controller *controller;
1301 uint64_t ofpacts_stub[128 / 8];
1302 struct ofpbuf ofpacts;
1306 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1309 controller = ofpact_put_CONTROLLER(&ofpacts);
1310 controller->max_len = UINT16_MAX;
1311 controller->controller_id = 0;
1312 controller->reason = OFPR_NO_MATCH;
1313 ofpact_pad(&ofpacts);
1315 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1320 ofpbuf_clear(&ofpacts);
1321 error = add_internal_flow(ofproto, id++, &ofpacts,
1322 &ofproto->no_packet_in_rule);
1327 error = add_internal_flow(ofproto, id++, &ofpacts,
1328 &ofproto->drop_frags_rule);
1333 destruct(struct ofproto *ofproto_)
1335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336 struct rule_dpif *rule, *next_rule;
1337 struct ofproto_packet_in *pin, *next_pin;
1338 struct facet *facet, *next_facet;
1339 struct cls_cursor cursor;
1340 struct oftable *table;
1343 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1344 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1345 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1346 CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1347 facet_remove(facet);
1350 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1351 ovs_rwlock_wrlock(&xlate_rwlock);
1352 xlate_remove_ofproto(ofproto);
1353 ovs_rwlock_unlock(&xlate_rwlock);
1355 /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1356 * use-after-free error. */
1357 udpif_revalidate(ofproto->backer->udpif);
1359 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1361 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1362 struct cls_cursor cursor;
1364 ovs_rwlock_rdlock(&table->cls.rwlock);
1365 cls_cursor_init(&cursor, &table->cls, NULL);
1366 ovs_rwlock_unlock(&table->cls.rwlock);
1367 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1368 ofproto_rule_delete(&ofproto->up, &rule->up);
1372 guarded_list_pop_all(&ofproto->pins, &pins);
1373 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1374 list_remove(&pin->list_node);
1375 free(CONST_CAST(void *, pin->up.packet));
1378 guarded_list_destroy(&ofproto->pins);
1380 mbridge_unref(ofproto->mbridge);
1382 netflow_unref(ofproto->netflow);
1383 dpif_sflow_unref(ofproto->sflow);
1384 hmap_destroy(&ofproto->bundles);
1385 mac_learning_unref(ofproto->ml);
1387 classifier_destroy(&ofproto->facets);
1389 hmap_destroy(&ofproto->vlandev_map);
1390 hmap_destroy(&ofproto->realdev_vid_map);
1392 sset_destroy(&ofproto->ports);
1393 sset_destroy(&ofproto->ghost_ports);
1394 sset_destroy(&ofproto->port_poll_set);
1396 ovs_mutex_destroy(&ofproto->stats_mutex);
1397 ovs_mutex_destroy(&ofproto->vsp_mutex);
1399 close_dpif_backer(ofproto->backer);
1403 run(struct ofproto *ofproto_)
1405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1408 if (mbridge_need_revalidate(ofproto->mbridge)) {
1409 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1410 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1411 mac_learning_flush(ofproto->ml);
1412 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1415 /* Do not perform any periodic activity required by 'ofproto' while
1416 * waiting for flow restore to complete. */
1417 if (!ofproto_get_flow_restore_wait()) {
1418 struct ofproto_packet_in *pin, *next_pin;
1421 guarded_list_pop_all(&ofproto->pins, &pins);
1422 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1423 connmgr_send_packet_in(ofproto->up.connmgr, pin);
1424 list_remove(&pin->list_node);
1425 free(CONST_CAST(void *, pin->up.packet));
1430 if (ofproto->netflow) {
1431 netflow_run(ofproto->netflow);
1433 if (ofproto->sflow) {
1434 dpif_sflow_run(ofproto->sflow);
1436 if (ofproto->ipfix) {
1437 dpif_ipfix_run(ofproto->ipfix);
1440 new_seq = seq_read(connectivity_seq_get());
1441 if (ofproto->change_seq != new_seq) {
1442 struct ofport_dpif *ofport;
1444 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1448 ofproto->change_seq = new_seq;
1450 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1451 struct ofbundle *bundle;
1453 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1459 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1460 if (mac_learning_run(ofproto->ml)) {
1461 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1463 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1465 /* Check the consistency of a random facet, to aid debugging. */
1466 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1467 if (time_msec() >= ofproto->consistency_rl
1468 && !classifier_is_empty(&ofproto->facets)
1469 && !ofproto->backer->need_revalidate) {
1470 struct cls_subtable *table;
1471 struct cls_rule *cr;
1472 struct facet *facet;
1474 ofproto->consistency_rl = time_msec() + 250;
1476 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1477 struct cls_subtable, hmap_node);
1478 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1480 facet = CONTAINER_OF(cr, struct facet, cr);
1482 if (!facet_check_consistency(facet)) {
1483 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1486 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1492 wait(struct ofproto *ofproto_)
1494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1496 if (ofproto_get_flow_restore_wait()) {
1500 if (ofproto->sflow) {
1501 dpif_sflow_wait(ofproto->sflow);
1503 if (ofproto->ipfix) {
1504 dpif_ipfix_wait(ofproto->ipfix);
1506 if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1507 struct ofbundle *bundle;
1509 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1510 bundle_wait(bundle);
1513 if (ofproto->netflow) {
1514 netflow_wait(ofproto->netflow);
1516 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1517 mac_learning_wait(ofproto->ml);
1518 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1520 if (ofproto->backer->need_revalidate) {
1521 /* Shouldn't happen, but if it does just go around again. */
1522 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1523 poll_immediate_wake();
1528 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1530 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1531 struct cls_cursor cursor;
1532 size_t n_subfacets = 0;
1533 struct facet *facet;
1535 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1536 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1537 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1539 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1540 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1541 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1542 n_subfacets += list_size(&facet->subfacets);
1544 ovs_rwlock_unlock(&ofproto->facets.rwlock);
1545 simap_increase(usage, "subfacets", n_subfacets);
1549 type_get_memory_usage(const char *type, struct simap *usage)
1551 struct dpif_backer *backer;
1553 backer = shash_find_data(&all_dpif_backers, type);
1555 udpif_get_memory_usage(backer->udpif, usage);
1560 flush(struct ofproto *ofproto_)
1562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1563 struct subfacet *subfacet, *next_subfacet;
1564 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1568 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1569 &ofproto->backer->subfacets) {
1570 if (subfacet->facet->ofproto != ofproto) {
1574 if (subfacet->path != SF_NOT_INSTALLED) {
1575 batch[n_batch++] = subfacet;
1576 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1577 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1581 subfacet_destroy(subfacet);
1586 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1591 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1592 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1594 *arp_match_ip = true;
1595 *actions = (OFPUTIL_A_OUTPUT |
1596 OFPUTIL_A_SET_VLAN_VID |
1597 OFPUTIL_A_SET_VLAN_PCP |
1598 OFPUTIL_A_STRIP_VLAN |
1599 OFPUTIL_A_SET_DL_SRC |
1600 OFPUTIL_A_SET_DL_DST |
1601 OFPUTIL_A_SET_NW_SRC |
1602 OFPUTIL_A_SET_NW_DST |
1603 OFPUTIL_A_SET_NW_TOS |
1604 OFPUTIL_A_SET_TP_SRC |
1605 OFPUTIL_A_SET_TP_DST |
1610 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1613 struct dpif_dp_stats s;
1614 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1617 strcpy(ots->name, "classifier");
1619 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1620 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1621 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1622 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1624 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1625 ots->lookup_count = htonll(n_lookup);
1626 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1629 static struct ofport *
1632 struct ofport_dpif *port = xmalloc(sizeof *port);
1637 port_dealloc(struct ofport *port_)
1639 struct ofport_dpif *port = ofport_dpif_cast(port_);
1644 port_construct(struct ofport *port_)
1646 struct ofport_dpif *port = ofport_dpif_cast(port_);
1647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1648 const struct netdev *netdev = port->up.netdev;
1649 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1650 struct dpif_port dpif_port;
1653 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1654 port->bundle = NULL;
1657 port->may_enable = true;
1658 port->stp_port = NULL;
1659 port->stp_state = STP_DISABLED;
1660 port->is_tunnel = false;
1664 port->realdev_ofp_port = 0;
1665 port->vlandev_vid = 0;
1666 port->carrier_seq = netdev_get_carrier_resets(netdev);
1667 port->is_layer3 = netdev_vport_is_layer3(netdev);
1669 if (netdev_vport_is_patch(netdev)) {
1670 /* By bailing out here, we don't submit the port to the sFlow module
1671 * to be considered for counter polling export. This is correct
1672 * because the patch port represents an interface that sFlow considers
1673 * to be "internal" to the switch as a whole, and therefore not an
1674 * candidate for counter polling. */
1675 port->odp_port = ODPP_NONE;
1676 ofport_update_peer(port);
1680 error = dpif_port_query_by_name(ofproto->backer->dpif,
1681 netdev_vport_get_dpif_port(netdev, namebuf,
1688 port->odp_port = dpif_port.port_no;
1690 if (netdev_get_tunnel_config(netdev)) {
1691 tnl_port_add(port, port->up.netdev, port->odp_port);
1692 port->is_tunnel = true;
1694 /* Sanity-check that a mapping doesn't already exist. This
1695 * shouldn't happen for non-tunnel ports. */
1696 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1697 VLOG_ERR("port %s already has an OpenFlow port number",
1699 dpif_port_destroy(&dpif_port);
1703 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1704 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1705 hash_odp_port(port->odp_port));
1706 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1708 dpif_port_destroy(&dpif_port);
1710 if (ofproto->sflow) {
1711 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1718 port_destruct(struct ofport *port_)
1720 struct ofport_dpif *port = ofport_dpif_cast(port_);
1721 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1722 const char *devname = netdev_get_name(port->up.netdev);
1723 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1724 const char *dp_port_name;
1726 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1727 ovs_rwlock_wrlock(&xlate_rwlock);
1728 xlate_ofport_remove(port);
1729 ovs_rwlock_unlock(&xlate_rwlock);
1731 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1733 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1734 /* The underlying device is still there, so delete it. This
1735 * happens when the ofproto is being destroyed, since the caller
1736 * assumes that removal of attached ports will happen as part of
1738 if (!port->is_tunnel) {
1739 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1744 port->peer->peer = NULL;
1748 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1749 ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1750 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1751 ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1755 sset_find_and_delete(&ofproto->ports, devname);
1756 sset_find_and_delete(&ofproto->ghost_ports, devname);
1757 bundle_remove(port_);
1758 set_cfm(port_, NULL);
1759 set_bfd(port_, NULL);
1760 if (ofproto->sflow) {
1761 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1768 port_modified(struct ofport *port_)
1770 struct ofport_dpif *port = ofport_dpif_cast(port_);
1772 if (port->bundle && port->bundle->bond) {
1773 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1777 cfm_set_netdev(port->cfm, port->up.netdev);
1781 bfd_set_netdev(port->bfd, port->up.netdev);
1784 ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1785 port->up.pp.hw_addr);
1787 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1789 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1793 ofport_update_peer(port);
1797 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1799 struct ofport_dpif *port = ofport_dpif_cast(port_);
1800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1801 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1803 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1804 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1805 OFPUTIL_PC_NO_PACKET_IN)) {
1806 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1808 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1809 bundle_update(port->bundle);
1815 set_sflow(struct ofproto *ofproto_,
1816 const struct ofproto_sflow_options *sflow_options)
1818 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1819 struct dpif_sflow *ds = ofproto->sflow;
1821 if (sflow_options) {
1823 struct ofport_dpif *ofport;
1825 ds = ofproto->sflow = dpif_sflow_create();
1826 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1827 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1829 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1831 dpif_sflow_set_options(ds, sflow_options);
1834 dpif_sflow_unref(ds);
1835 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 ofproto->sflow = NULL;
1844 struct ofproto *ofproto_,
1845 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1846 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1847 size_t n_flow_exporters_options)
1849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1850 struct dpif_ipfix *di = ofproto->ipfix;
1851 bool has_options = bridge_exporter_options || flow_exporters_options;
1853 if (has_options && !di) {
1854 di = ofproto->ipfix = dpif_ipfix_create();
1858 /* Call set_options in any case to cleanly flush the flow
1859 * caches in the last exporters that are to be destroyed. */
1860 dpif_ipfix_set_options(
1861 di, bridge_exporter_options, flow_exporters_options,
1862 n_flow_exporters_options);
1865 dpif_ipfix_unref(di);
1866 ofproto->ipfix = NULL;
1874 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1876 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1881 struct ofproto_dpif *ofproto;
1883 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1884 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885 ofport->cfm = cfm_create(ofport->up.netdev);
1888 if (cfm_configure(ofport->cfm, s)) {
1895 cfm_unref(ofport->cfm);
1898 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1899 ofport->up.pp.hw_addr);
1904 get_cfm_status(const struct ofport *ofport_,
1905 struct ofproto_cfm_status *status)
1907 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1910 status->faults = cfm_get_fault(ofport->cfm);
1911 status->flap_count = cfm_get_flap_count(ofport->cfm);
1912 status->remote_opstate = cfm_get_opup(ofport->cfm);
1913 status->health = cfm_get_health(ofport->cfm);
1914 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1922 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1924 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1925 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1929 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1930 cfg, ofport->up.netdev);
1931 if (ofport->bfd != old) {
1932 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1934 ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1935 ofport->up.pp.hw_addr);
1940 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1942 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1945 bfd_get_status(ofport->bfd, smap);
1952 /* Spanning Tree. */
1955 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1957 struct ofproto_dpif *ofproto = ofproto_;
1958 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1959 struct ofport_dpif *ofport;
1961 ofport = stp_port_get_aux(sp);
1963 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1964 ofproto->up.name, port_num);
1966 struct eth_header *eth = pkt->l2;
1968 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1969 if (eth_addr_is_zero(eth->eth_src)) {
1970 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1971 "with unknown MAC", ofproto->up.name, port_num);
1973 ofproto_dpif_send_packet(ofport, pkt);
1979 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1981 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1985 /* Only revalidate flows if the configuration changed. */
1986 if (!s != !ofproto->stp) {
1987 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1991 if (!ofproto->stp) {
1992 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1993 send_bpdu_cb, ofproto);
1994 ofproto->stp_last_tick = time_msec();
1997 stp_set_bridge_id(ofproto->stp, s->system_id);
1998 stp_set_bridge_priority(ofproto->stp, s->priority);
1999 stp_set_hello_time(ofproto->stp, s->hello_time);
2000 stp_set_max_age(ofproto->stp, s->max_age);
2001 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2003 struct ofport *ofport;
2005 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2006 set_stp_port(ofport, NULL);
2009 stp_unref(ofproto->stp);
2010 ofproto->stp = NULL;
2017 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2023 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2024 s->designated_root = stp_get_designated_root(ofproto->stp);
2025 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2034 update_stp_port_state(struct ofport_dpif *ofport)
2036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2037 enum stp_state state;
2039 /* Figure out new state. */
2040 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2044 if (ofport->stp_state != state) {
2045 enum ofputil_port_state of_state;
2048 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2049 netdev_get_name(ofport->up.netdev),
2050 stp_state_name(ofport->stp_state),
2051 stp_state_name(state));
2052 if (stp_learn_in_state(ofport->stp_state)
2053 != stp_learn_in_state(state)) {
2054 /* xxx Learning action flows should also be flushed. */
2055 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2056 mac_learning_flush(ofproto->ml);
2057 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2059 fwd_change = stp_forward_in_state(ofport->stp_state)
2060 != stp_forward_in_state(state);
2062 ofproto->backer->need_revalidate = REV_STP;
2063 ofport->stp_state = state;
2064 ofport->stp_state_entered = time_msec();
2066 if (fwd_change && ofport->bundle) {
2067 bundle_update(ofport->bundle);
2070 /* Update the STP state bits in the OpenFlow port description. */
2071 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2072 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2073 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2074 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2075 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2077 ofproto_port_set_state(&ofport->up, of_state);
2081 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2082 * caller is responsible for assigning STP port numbers and ensuring
2083 * there are no duplicates. */
2085 set_stp_port(struct ofport *ofport_,
2086 const struct ofproto_port_stp_settings *s)
2088 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2089 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2090 struct stp_port *sp = ofport->stp_port;
2092 if (!s || !s->enable) {
2094 ofport->stp_port = NULL;
2095 stp_port_disable(sp);
2096 update_stp_port_state(ofport);
2099 } else if (sp && stp_port_no(sp) != s->port_num
2100 && ofport == stp_port_get_aux(sp)) {
2101 /* The port-id changed, so disable the old one if it's not
2102 * already in use by another port. */
2103 stp_port_disable(sp);
2106 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2107 stp_port_enable(sp);
2109 stp_port_set_aux(sp, ofport);
2110 stp_port_set_priority(sp, s->priority);
2111 stp_port_set_path_cost(sp, s->path_cost);
2113 update_stp_port_state(ofport);
2119 get_stp_port_status(struct ofport *ofport_,
2120 struct ofproto_port_stp_status *s)
2122 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2123 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2124 struct stp_port *sp = ofport->stp_port;
2126 if (!ofproto->stp || !sp) {
2132 s->port_id = stp_port_get_id(sp);
2133 s->state = stp_port_get_state(sp);
2134 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2135 s->role = stp_port_get_role(sp);
2141 get_stp_port_stats(struct ofport *ofport_,
2142 struct ofproto_port_stp_stats *s)
2144 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2146 struct stp_port *sp = ofport->stp_port;
2148 if (!ofproto->stp || !sp) {
2154 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2160 stp_run(struct ofproto_dpif *ofproto)
2163 long long int now = time_msec();
2164 long long int elapsed = now - ofproto->stp_last_tick;
2165 struct stp_port *sp;
2168 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2169 ofproto->stp_last_tick = now;
2171 while (stp_get_changed_port(ofproto->stp, &sp)) {
2172 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2175 update_stp_port_state(ofport);
2179 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2180 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2181 mac_learning_flush(ofproto->ml);
2182 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2188 stp_wait(struct ofproto_dpif *ofproto)
2191 poll_timer_wait(1000);
2196 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2199 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2202 if (ofport->n_qdscp != n_qdscp
2203 || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2204 n_qdscp * sizeof *qdscp))) {
2205 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2206 free(ofport->qdscp);
2207 ofport->qdscp = n_qdscp
2208 ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2210 ofport->n_qdscp = n_qdscp;
2218 /* Expires all MAC learning entries associated with 'bundle' and forces its
2219 * ofproto to revalidate every flow.
2221 * Normally MAC learning entries are removed only from the ofproto associated
2222 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2223 * are removed from every ofproto. When patch ports and SLB bonds are in use
2224 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2225 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2226 * with the host from which it migrated. */
2228 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2230 struct ofproto_dpif *ofproto = bundle->ofproto;
2231 struct mac_learning *ml = ofproto->ml;
2232 struct mac_entry *mac, *next_mac;
2234 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2235 ovs_rwlock_wrlock(&ml->rwlock);
2236 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2237 if (mac->port.p == bundle) {
2239 struct ofproto_dpif *o;
2241 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2243 struct mac_entry *e;
2245 ovs_rwlock_wrlock(&o->ml->rwlock);
2246 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2248 mac_learning_expire(o->ml, e);
2250 ovs_rwlock_unlock(&o->ml->rwlock);
2255 mac_learning_expire(ml, mac);
2258 ovs_rwlock_unlock(&ml->rwlock);
2261 static struct ofbundle *
2262 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2264 struct ofbundle *bundle;
2266 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2267 &ofproto->bundles) {
2268 if (bundle->aux == aux) {
2276 bundle_update(struct ofbundle *bundle)
2278 struct ofport_dpif *port;
2280 bundle->floodable = true;
2281 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2282 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2284 || !stp_forward_in_state(port->stp_state)) {
2285 bundle->floodable = false;
2292 bundle_del_port(struct ofport_dpif *port)
2294 struct ofbundle *bundle = port->bundle;
2296 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2298 list_remove(&port->bundle_node);
2299 port->bundle = NULL;
2302 lacp_slave_unregister(bundle->lacp, port);
2305 bond_slave_unregister(bundle->bond, port);
2308 bundle_update(bundle);
2312 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2313 struct lacp_slave_settings *lacp)
2315 struct ofport_dpif *port;
2317 port = get_ofp_port(bundle->ofproto, ofp_port);
2322 if (port->bundle != bundle) {
2323 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2325 bundle_remove(&port->up);
2328 port->bundle = bundle;
2329 list_push_back(&bundle->ports, &port->bundle_node);
2330 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2332 || !stp_forward_in_state(port->stp_state)) {
2333 bundle->floodable = false;
2337 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2338 lacp_slave_register(bundle->lacp, port, lacp);
2345 bundle_destroy(struct ofbundle *bundle)
2347 struct ofproto_dpif *ofproto;
2348 struct ofport_dpif *port, *next_port;
2354 ofproto = bundle->ofproto;
2355 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2357 ovs_rwlock_wrlock(&xlate_rwlock);
2358 xlate_bundle_remove(bundle);
2359 ovs_rwlock_unlock(&xlate_rwlock);
2361 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2362 bundle_del_port(port);
2365 bundle_flush_macs(bundle, true);
2366 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2368 free(bundle->trunks);
2369 lacp_unref(bundle->lacp);
2370 bond_unref(bundle->bond);
2375 bundle_set(struct ofproto *ofproto_, void *aux,
2376 const struct ofproto_bundle_settings *s)
2378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2379 bool need_flush = false;
2380 struct ofport_dpif *port;
2381 struct ofbundle *bundle;
2382 unsigned long *trunks;
2388 bundle_destroy(bundle_lookup(ofproto, aux));
2392 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2393 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2395 bundle = bundle_lookup(ofproto, aux);
2397 bundle = xmalloc(sizeof *bundle);
2399 bundle->ofproto = ofproto;
2400 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2401 hash_pointer(aux, 0));
2403 bundle->name = NULL;
2405 list_init(&bundle->ports);
2406 bundle->vlan_mode = PORT_VLAN_TRUNK;
2408 bundle->trunks = NULL;
2409 bundle->use_priority_tags = s->use_priority_tags;
2410 bundle->lacp = NULL;
2411 bundle->bond = NULL;
2413 bundle->floodable = true;
2414 mbridge_register_bundle(ofproto->mbridge, bundle);
2417 if (!bundle->name || strcmp(s->name, bundle->name)) {
2419 bundle->name = xstrdup(s->name);
2424 ofproto->lacp_enabled = true;
2425 if (!bundle->lacp) {
2426 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2427 bundle->lacp = lacp_create();
2429 lacp_configure(bundle->lacp, s->lacp);
2431 lacp_unref(bundle->lacp);
2432 bundle->lacp = NULL;
2435 /* Update set of ports. */
2437 for (i = 0; i < s->n_slaves; i++) {
2438 if (!bundle_add_port(bundle, s->slaves[i],
2439 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2443 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2444 struct ofport_dpif *next_port;
2446 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2447 for (i = 0; i < s->n_slaves; i++) {
2448 if (s->slaves[i] == port->up.ofp_port) {
2453 bundle_del_port(port);
2457 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2459 if (list_is_empty(&bundle->ports)) {
2460 bundle_destroy(bundle);
2464 /* Set VLAN tagging mode */
2465 if (s->vlan_mode != bundle->vlan_mode
2466 || s->use_priority_tags != bundle->use_priority_tags) {
2467 bundle->vlan_mode = s->vlan_mode;
2468 bundle->use_priority_tags = s->use_priority_tags;
2473 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2474 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2476 if (vlan != bundle->vlan) {
2477 bundle->vlan = vlan;
2481 /* Get trunked VLANs. */
2482 switch (s->vlan_mode) {
2483 case PORT_VLAN_ACCESS:
2487 case PORT_VLAN_TRUNK:
2488 trunks = CONST_CAST(unsigned long *, s->trunks);
2491 case PORT_VLAN_NATIVE_UNTAGGED:
2492 case PORT_VLAN_NATIVE_TAGGED:
2493 if (vlan != 0 && (!s->trunks
2494 || !bitmap_is_set(s->trunks, vlan)
2495 || bitmap_is_set(s->trunks, 0))) {
2496 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2498 trunks = bitmap_clone(s->trunks, 4096);
2500 trunks = bitmap_allocate1(4096);
2502 bitmap_set1(trunks, vlan);
2503 bitmap_set0(trunks, 0);
2505 trunks = CONST_CAST(unsigned long *, s->trunks);
2512 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2513 free(bundle->trunks);
2514 if (trunks == s->trunks) {
2515 bundle->trunks = vlan_bitmap_clone(trunks);
2517 bundle->trunks = trunks;
2522 if (trunks != s->trunks) {
2527 if (!list_is_short(&bundle->ports)) {
2528 bundle->ofproto->has_bonded_bundles = true;
2530 if (bond_reconfigure(bundle->bond, s->bond)) {
2531 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2534 bundle->bond = bond_create(s->bond);
2535 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2538 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2539 bond_slave_register(bundle->bond, port, port->up.netdev);
2542 bond_unref(bundle->bond);
2543 bundle->bond = NULL;
2546 /* If we changed something that would affect MAC learning, un-learn
2547 * everything on this port and force flow revalidation. */
2549 bundle_flush_macs(bundle, false);
2556 bundle_remove(struct ofport *port_)
2558 struct ofport_dpif *port = ofport_dpif_cast(port_);
2559 struct ofbundle *bundle = port->bundle;
2562 bundle_del_port(port);
2563 if (list_is_empty(&bundle->ports)) {
2564 bundle_destroy(bundle);
2565 } else if (list_is_short(&bundle->ports)) {
2566 bond_unref(bundle->bond);
2567 bundle->bond = NULL;
2573 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2576 struct ofport_dpif *port = port_;
2577 uint8_t ea[ETH_ADDR_LEN];
2580 error = netdev_get_etheraddr(port->up.netdev, ea);
2582 struct ofpbuf packet;
2585 ofpbuf_init(&packet, 0);
2586 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2588 memcpy(packet_pdu, pdu, pdu_size);
2590 ofproto_dpif_send_packet(port, &packet);
2591 ofpbuf_uninit(&packet);
2593 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2594 "%s (%s)", port->bundle->name,
2595 netdev_get_name(port->up.netdev), ovs_strerror(error));
2600 bundle_send_learning_packets(struct ofbundle *bundle)
2602 struct ofproto_dpif *ofproto = bundle->ofproto;
2603 struct ofpbuf *learning_packet;
2604 int error, n_packets, n_errors;
2605 struct mac_entry *e;
2606 struct list packets;
2608 list_init(&packets);
2609 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2610 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2611 if (e->port.p != bundle) {
2614 learning_packet = bond_compose_learning_packet(bundle->bond,
2617 learning_packet->private_p = port_void;
2618 list_push_back(&packets, &learning_packet->list_node);
2621 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2623 error = n_packets = n_errors = 0;
2624 LIST_FOR_EACH (learning_packet, list_node, &packets) {
2627 ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2634 ofpbuf_list_delete(&packets);
2637 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2638 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2639 "packets, last error was: %s",
2640 bundle->name, n_errors, n_packets, ovs_strerror(error));
2642 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2643 bundle->name, n_packets);
2648 bundle_run(struct ofbundle *bundle)
2651 lacp_run(bundle->lacp, send_pdu_cb);
2654 struct ofport_dpif *port;
2656 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2657 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2660 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2661 bundle->ofproto->backer->need_revalidate = REV_BOND;
2664 if (bond_should_send_learning_packets(bundle->bond)) {
2665 bundle_send_learning_packets(bundle);
2671 bundle_wait(struct ofbundle *bundle)
2674 lacp_wait(bundle->lacp);
2677 bond_wait(bundle->bond);
2684 mirror_set__(struct ofproto *ofproto_, void *aux,
2685 const struct ofproto_mirror_settings *s)
2687 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2688 struct ofbundle **srcs, **dsts;
2693 mirror_destroy(ofproto->mbridge, aux);
2697 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2698 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2700 for (i = 0; i < s->n_srcs; i++) {
2701 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2704 for (i = 0; i < s->n_dsts; i++) {
2705 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2708 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2709 s->n_dsts, s->src_vlans,
2710 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2717 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2718 uint64_t *packets, uint64_t *bytes)
2721 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2726 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2728 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2729 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2730 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2731 mac_learning_flush(ofproto->ml);
2733 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2738 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2741 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2742 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2746 forward_bpdu_changed(struct ofproto *ofproto_)
2748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2749 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2753 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2757 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2758 mac_learning_set_idle_time(ofproto->ml, idle_time);
2759 mac_learning_set_max_entries(ofproto->ml, max_entries);
2760 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2765 static struct ofport_dpif *
2766 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2768 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2769 return ofport ? ofport_dpif_cast(ofport) : NULL;
2773 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2774 struct ofproto_port *ofproto_port,
2775 struct dpif_port *dpif_port)
2777 ofproto_port->name = dpif_port->name;
2778 ofproto_port->type = dpif_port->type;
2779 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2783 ofport_update_peer(struct ofport_dpif *ofport)
2785 const struct ofproto_dpif *ofproto;
2786 struct dpif_backer *backer;
2789 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2793 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2794 backer->need_revalidate = REV_RECONFIGURE;
2797 ofport->peer->peer = NULL;
2798 ofport->peer = NULL;
2801 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2806 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2807 struct ofport *peer_ofport;
2808 struct ofport_dpif *peer;
2811 if (ofproto->backer != backer) {
2815 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2820 peer = ofport_dpif_cast(peer_ofport);
2821 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2822 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2824 ofport->peer = peer;
2825 ofport->peer->peer = ofport;
2835 port_run(struct ofport_dpif *ofport)
2837 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2838 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2839 bool enable = netdev_get_carrier(ofport->up.netdev);
2840 bool cfm_enable = false;
2841 bool bfd_enable = false;
2843 ofport->carrier_seq = carrier_seq;
2846 int cfm_opup = cfm_get_opup(ofport->cfm);
2848 cfm_enable = !cfm_get_fault(ofport->cfm);
2850 if (cfm_opup >= 0) {
2851 cfm_enable = cfm_enable && cfm_opup;
2856 bfd_enable = bfd_forwarding(ofport->bfd);
2859 if (ofport->bfd || ofport->cfm) {
2860 enable = enable && (cfm_enable || bfd_enable);
2863 if (ofport->bundle) {
2864 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2865 if (carrier_changed) {
2866 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2870 if (ofport->may_enable != enable) {
2871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2872 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2875 ofport->may_enable = enable;
2879 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2880 struct ofproto_port *ofproto_port)
2882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2883 struct dpif_port dpif_port;
2886 if (sset_contains(&ofproto->ghost_ports, devname)) {
2887 const char *type = netdev_get_type_from_name(devname);
2889 /* We may be called before ofproto->up.port_by_name is populated with
2890 * the appropriate ofport. For this reason, we must get the name and
2891 * type from the netdev layer directly. */
2893 const struct ofport *ofport;
2895 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2896 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2897 ofproto_port->name = xstrdup(devname);
2898 ofproto_port->type = xstrdup(type);
2904 if (!sset_contains(&ofproto->ports, devname)) {
2907 error = dpif_port_query_by_name(ofproto->backer->dpif,
2908 devname, &dpif_port);
2910 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2916 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2919 const char *devname = netdev_get_name(netdev);
2920 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2921 const char *dp_port_name;
2923 if (netdev_vport_is_patch(netdev)) {
2924 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2928 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2929 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2930 odp_port_t port_no = ODPP_NONE;
2933 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2937 if (netdev_get_tunnel_config(netdev)) {
2938 simap_put(&ofproto->backer->tnl_backers,
2939 dp_port_name, odp_to_u32(port_no));
2943 if (netdev_get_tunnel_config(netdev)) {
2944 sset_add(&ofproto->ghost_ports, devname);
2946 sset_add(&ofproto->ports, devname);
2952 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2955 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2962 sset_find_and_delete(&ofproto->ghost_ports,
2963 netdev_get_name(ofport->up.netdev));
2964 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2965 if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2966 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2968 /* The caller is going to close ofport->up.netdev. If this is a
2969 * bonded port, then the bond is using that netdev, so remove it
2970 * from the bond. The client will need to reconfigure everything
2971 * after deleting ports, so then the slave will get re-added. */
2972 bundle_remove(&ofport->up);
2979 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2981 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2986 error = netdev_get_stats(ofport->up.netdev, stats);
2988 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2989 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2991 ovs_mutex_lock(&ofproto->stats_mutex);
2992 /* ofproto->stats.tx_packets represents packets that we created
2993 * internally and sent to some port (e.g. packets sent with
2994 * ofproto_dpif_send_packet()). Account for them as if they had
2995 * come from OFPP_LOCAL and got forwarded. */
2997 if (stats->rx_packets != UINT64_MAX) {
2998 stats->rx_packets += ofproto->stats.tx_packets;
3001 if (stats->rx_bytes != UINT64_MAX) {
3002 stats->rx_bytes += ofproto->stats.tx_bytes;
3005 /* ofproto->stats.rx_packets represents packets that were received on
3006 * some port and we processed internally and dropped (e.g. STP).
3007 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3009 if (stats->tx_packets != UINT64_MAX) {
3010 stats->tx_packets += ofproto->stats.rx_packets;
3013 if (stats->tx_bytes != UINT64_MAX) {
3014 stats->tx_bytes += ofproto->stats.rx_bytes;
3016 ovs_mutex_unlock(&ofproto->stats_mutex);
3022 struct port_dump_state {
3027 struct ofproto_port port;
3032 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3034 *statep = xzalloc(sizeof(struct port_dump_state));
3039 port_dump_next(const struct ofproto *ofproto_, void *state_,
3040 struct ofproto_port *port)
3042 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3043 struct port_dump_state *state = state_;
3044 const struct sset *sset;
3045 struct sset_node *node;
3047 if (state->has_port) {
3048 ofproto_port_destroy(&state->port);
3049 state->has_port = false;
3051 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3052 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3055 error = port_query_by_name(ofproto_, node->name, &state->port);
3057 *port = state->port;
3058 state->has_port = true;
3060 } else if (error != ENODEV) {
3065 if (!state->ghost) {
3066 state->ghost = true;
3069 return port_dump_next(ofproto_, state_, port);
3076 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3078 struct port_dump_state *state = state_;
3080 if (state->has_port) {
3081 ofproto_port_destroy(&state->port);
3088 port_poll(const struct ofproto *ofproto_, char **devnamep)
3090 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3092 if (ofproto->port_poll_errno) {
3093 int error = ofproto->port_poll_errno;
3094 ofproto->port_poll_errno = 0;
3098 if (sset_is_empty(&ofproto->port_poll_set)) {
3102 *devnamep = sset_pop(&ofproto->port_poll_set);
3107 port_poll_wait(const struct ofproto *ofproto_)
3109 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3110 dpif_port_poll_wait(ofproto->backer->dpif);
3114 port_is_lacp_current(const struct ofport *ofport_)
3116 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3117 return (ofport->bundle && ofport->bundle->lacp
3118 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3122 /* Upcall handling. */
3124 struct flow_miss_op {
3125 struct dpif_op dpif_op;
3127 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3128 struct xlate_out xout;
3129 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3131 struct ofpbuf mask; /* Flow mask for "put" ops. */
3132 struct odputil_keybuf maskbuf;
3134 /* If this is a "put" op, then a pointer to the subfacet that should
3135 * be marked as uninstalled if the operation fails. */
3136 struct subfacet *subfacet;
3139 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3140 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3141 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3142 * return value of true). However, for short flows the cost of bookkeeping is
3143 * much higher than the benefits, so when the datapath holds a large number of
3144 * flows we impose some heuristics to decide which flows are likely to be worth
3147 flow_miss_should_make_facet(struct flow_miss *miss)
3149 struct dpif_backer *backer = miss->ofproto->backer;
3152 switch (flow_miss_model) {
3153 case OFPROTO_HANDLE_MISS_AUTO:
3155 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3157 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3161 if (!backer->governor) {
3164 n_subfacets = hmap_count(&backer->subfacets);
3165 if (n_subfacets * 2 <= flow_eviction_threshold) {
3169 backer->governor = governor_create();
3172 hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3173 return governor_should_install_flow(backer->governor, hash,
3174 miss->stats.n_packets);
3177 /* Handles 'miss', which matches 'facet'. May add any required datapath
3178 * operations to 'ops', incrementing '*n_ops' for each new op.
3180 * All of the packets in 'miss' are considered to have arrived at time
3181 * 'miss->stats.used'. This is really important only for new facets: if we
3182 * just called time_msec() here, then the new subfacet or its packets could
3183 * look (occasionally) as though it was used some time after the facet was
3184 * used. That can make a one-packet flow look like it has a nonzero duration,
3185 * which looks odd in e.g. NetFlow statistics. */
3187 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3188 struct flow_miss_op *ops, size_t *n_ops)
3190 enum subfacet_path want_path;
3191 struct subfacet *subfacet;
3194 /* Update facet stats. */
3195 facet->packet_count += miss->stats.n_packets;
3196 facet->prev_packet_count += miss->stats.n_packets;
3197 facet->byte_count += miss->stats.n_bytes;
3198 facet->prev_byte_count += miss->stats.n_bytes;
3200 /* Look for an existing subfacet. If we find one, update its used time. */
3201 key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3202 if (!list_is_empty(&facet->subfacets)) {
3203 subfacet = subfacet_find(miss->ofproto->backer,
3204 miss->key, miss->key_len, key_hash);
3206 if (subfacet->facet == facet) {
3207 subfacet->used = MAX(subfacet->used, miss->stats.used);
3209 /* This shouldn't happen. */
3210 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3211 subfacet_destroy(subfacet);
3219 /* Don't install the flow if it's the result of the "userspace"
3220 * action for an already installed facet. This can occur when a
3221 * datapath flow with wildcards has a "userspace" action and flows
3222 * sent to userspace result in a different subfacet, which will then
3223 * be rejected as overlapping by the datapath. */
3224 if (miss->upcall_type == DPIF_UC_ACTION
3225 && !list_is_empty(&facet->subfacets)) {
3229 /* Create a subfacet, if we don't already have one. */
3231 subfacet = subfacet_create(facet, miss, key_hash);
3234 /* Install the subfacet, if it's not already installed. */
3235 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3236 if (subfacet->path != want_path) {
3237 struct flow_miss_op *op = &ops[(*n_ops)++];
3238 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3240 subfacet->path = want_path;
3242 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3243 if (enable_megaflows) {
3244 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3245 &miss->flow, UINT32_MAX);
3248 op->xout_garbage = false;
3249 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3250 op->subfacet = subfacet;
3251 put->flags = DPIF_FP_CREATE;
3252 put->key = miss->key;
3253 put->key_len = miss->key_len;
3254 put->mask = op->mask.data;
3255 put->mask_len = op->mask.size;
3257 if (want_path == SF_FAST_PATH) {
3258 put->actions = facet->xout.odp_actions.data;
3259 put->actions_len = facet->xout.odp_actions.size;
3261 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3262 op->slow_stub, sizeof op->slow_stub,
3263 &put->actions, &put->actions_len);
3269 /* Handles flow miss 'miss'. May add any required datapath operations
3270 * to 'ops', incrementing '*n_ops' for each new op. */
3272 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3275 struct facet *facet;
3277 miss->ofproto->n_missed += miss->stats.n_packets;
3279 facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3281 /* There does not exist a bijection between 'struct flow' and datapath
3282 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3283 * assumption used throughout the facet and subfacet handling code.
3284 * Since we have to handle these misses in userspace anyway, we simply
3285 * skip facet creation, avoiding the problem altogether. */
3286 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3287 || !flow_miss_should_make_facet(miss)) {
3291 facet = facet_create(miss);
3293 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3296 static struct drop_key *
3297 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3300 struct drop_key *drop_key;
3302 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3303 &backer->drop_keys) {
3304 if (drop_key->key_len == key_len
3305 && !memcmp(drop_key->key, key, key_len)) {
3313 drop_key_clear(struct dpif_backer *backer)
3315 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3316 struct drop_key *drop_key, *next;
3318 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3321 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3323 if (error && !VLOG_DROP_WARN(&rl)) {
3324 struct ds ds = DS_EMPTY_INITIALIZER;
3325 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3326 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3327 ovs_strerror(error), ds_cstr(&ds));
3331 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3332 drop_key_destroy(drop_key);
3335 udpif_drop_key_clear(backer->udpif);
3339 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3341 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3342 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3343 struct flow_miss *miss;
3346 /* Process each element in the to-do list, constructing the set of
3347 * operations to batch. */
3349 HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3350 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3352 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3354 /* Execute batch. */
3355 for (i = 0; i < n_ops; i++) {
3356 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3358 dpif_operate(backer->dpif, dpif_ops, n_ops);
3360 for (i = 0; i < n_ops; i++) {
3361 if (dpif_ops[i]->error != 0
3362 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3363 && flow_miss_ops[i].subfacet) {
3364 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3366 COVERAGE_INC(subfacet_install_fail);
3368 /* Zero-out subfacet counters when installation failed, but
3369 * datapath reported hits. This should not happen and
3370 * indicates a bug, since if the datapath flow exists, we
3371 * should not be attempting to create a new subfacet. A
3372 * buggy datapath could trigger this, so just zero out the
3373 * counters and log an error. */
3374 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3375 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3376 "datapath reported hits");
3377 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3380 subfacet->path = SF_NOT_INSTALLED;
3386 handle_upcalls(struct dpif_backer *backer)
3388 struct flow_miss_batch *fmb;
3391 for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3392 struct drop_key *drop_key = drop_key_next(backer->udpif);
3397 if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3398 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3399 hash_bytes(drop_key->key, drop_key->key_len, 0));
3400 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3401 drop_key->key, drop_key->key_len,
3402 NULL, 0, NULL, 0, NULL);
3404 drop_key_destroy(drop_key);
3408 fmb = flow_miss_batch_next(backer->udpif);
3410 handle_flow_misses(backer, fmb);
3411 flow_miss_batch_destroy(fmb);
3415 /* Flow expiration. */
3417 static int subfacet_max_idle(const struct dpif_backer *);
3418 static void update_stats(struct dpif_backer *);
3419 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3420 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3422 /* This function is called periodically by run(). Its job is to collect
3423 * updates for the flows that have been installed into the datapath, most
3424 * importantly when they last were used, and then use that information to
3425 * expire flows that have not been used recently.
3427 * Returns the number of milliseconds after which it should be called again. */
3429 expire(struct dpif_backer *backer)
3431 struct ofproto_dpif *ofproto;
3435 /* Periodically clear out the drop keys in an effort to keep them
3436 * relatively few. */
3437 drop_key_clear(backer);
3439 /* Update stats for each flow in the backer. */
3440 update_stats(backer);
3442 n_subfacets = hmap_count(&backer->subfacets);
3444 struct subfacet *subfacet;
3445 long long int total, now;
3449 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3450 total += now - subfacet->created;
3452 backer->avg_subfacet_life += total / n_subfacets;
3454 backer->avg_subfacet_life /= 2;
3456 backer->avg_n_subfacet += n_subfacets;
3457 backer->avg_n_subfacet /= 2;
3459 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3461 max_idle = subfacet_max_idle(backer);
3462 expire_subfacets(backer, max_idle);
3464 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3465 struct rule *rule, *next_rule;
3467 if (ofproto->backer != backer) {
3471 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3473 ovs_mutex_lock(&ofproto_mutex);
3474 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3475 &ofproto->up.expirable) {
3476 rule_expire(rule_dpif_cast(rule));
3478 ovs_mutex_unlock(&ofproto_mutex);
3480 /* All outstanding data in existing flows has been accounted, so it's a
3481 * good time to do bond rebalancing. */
3482 if (ofproto->has_bonded_bundles) {
3483 struct ofbundle *bundle;
3485 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3487 bond_rebalance(bundle->bond);
3493 return MIN(max_idle, 1000);
3496 /* Updates flow table statistics given that the datapath just reported 'stats'
3497 * as 'subfacet''s statistics. */
3499 update_subfacet_stats(struct subfacet *subfacet,
3500 const struct dpif_flow_stats *stats)
3502 struct facet *facet = subfacet->facet;
3503 struct dpif_flow_stats diff;
3505 diff.tcp_flags = stats->tcp_flags;
3506 diff.used = stats->used;
3508 if (stats->n_packets >= subfacet->dp_packet_count) {
3509 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3511 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3515 if (stats->n_bytes >= subfacet->dp_byte_count) {
3516 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3518 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3522 facet->ofproto->n_hit += diff.n_packets;
3523 subfacet->dp_packet_count = stats->n_packets;
3524 subfacet->dp_byte_count = stats->n_bytes;
3525 subfacet_update_stats(subfacet, &diff);
3527 if (diff.n_packets) {
3532 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3533 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3535 delete_unexpected_flow(struct dpif_backer *backer,
3536 const struct nlattr *key, size_t key_len)
3538 if (!VLOG_DROP_WARN(&rl)) {
3542 odp_flow_key_format(key, key_len, &s);
3543 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3547 COVERAGE_INC(facet_unexpected);
3548 dpif_flow_del(backer->dpif, key, key_len, NULL);
3551 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3553 * This function also pushes statistics updates to rules which each facet
3554 * resubmits into. Generally these statistics will be accurate. However, if a
3555 * facet changes the rule it resubmits into at some time in between
3556 * update_stats() runs, it is possible that statistics accrued to the
3557 * old rule will be incorrectly attributed to the new rule. This could be
3558 * avoided by calling update_stats() whenever rules are created or
3559 * deleted. However, the performance impact of making so many calls to the
3560 * datapath do not justify the benefit of having perfectly accurate statistics.
3562 * In addition, this function maintains per ofproto flow hit counts. The patch
3563 * port is not treated specially. e.g. A packet ingress from br0 patched into
3564 * br1 will increase the hit count of br0 by 1, however, does not affect
3565 * the hit or miss counts of br1.
3568 update_stats(struct dpif_backer *backer)
3570 const struct dpif_flow_stats *stats;
3571 struct dpif_flow_dump dump;
3572 const struct nlattr *key, *mask;
3573 size_t key_len, mask_len;
3575 dpif_flow_dump_start(&dump, backer->dpif);
3576 while (dpif_flow_dump_next(&dump, &key, &key_len,
3577 &mask, &mask_len, NULL, NULL, &stats)) {
3578 struct subfacet *subfacet;
3581 key_hash = odp_flow_key_hash(key, key_len);
3582 subfacet = subfacet_find(backer, key, key_len, key_hash);
3583 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3585 update_subfacet_stats(subfacet, stats);
3589 /* Stats are updated per-packet. */
3592 case SF_NOT_INSTALLED:
3594 delete_unexpected_flow(backer, key, key_len);
3598 dpif_flow_dump_done(&dump);
3601 /* Calculates and returns the number of milliseconds of idle time after which
3602 * subfacets should expire from the datapath. When a subfacet expires, we fold
3603 * its statistics into its facet, and when a facet's last subfacet expires, we
3604 * fold its statistic into its rule. */
3606 subfacet_max_idle(const struct dpif_backer *backer)
3609 * Idle time histogram.
3611 * Most of the time a switch has a relatively small number of subfacets.
3612 * When this is the case we might as well keep statistics for all of them
3613 * in userspace and to cache them in the kernel datapath for performance as
3616 * As the number of subfacets increases, the memory required to maintain
3617 * statistics about them in userspace and in the kernel becomes
3618 * significant. However, with a large number of subfacets it is likely
3619 * that only a few of them are "heavy hitters" that consume a large amount
3620 * of bandwidth. At this point, only heavy hitters are worth caching in
3621 * the kernel and maintaining in userspaces; other subfacets we can
3624 * The technique used to compute the idle time is to build a histogram with
3625 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3626 * that is installed in the kernel gets dropped in the appropriate bucket.
3627 * After the histogram has been built, we compute the cutoff so that only
3628 * the most-recently-used 1% of subfacets (but at least
3629 * flow_eviction_threshold flows) are kept cached. At least
3630 * the most-recently-used bucket of subfacets is kept, so actually an
3631 * arbitrary number of subfacets can be kept in any given expiration run
3632 * (though the next run will delete most of those unless they receive
3635 * This requires a second pass through the subfacets, in addition to the
3636 * pass made by update_stats(), because the former function never looks at
3637 * uninstallable subfacets.
3639 enum { BUCKET_WIDTH = 100 };
3640 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3641 int buckets[N_BUCKETS] = { 0 };
3642 int total, subtotal, bucket;
3643 struct subfacet *subfacet;
3647 total = hmap_count(&backer->subfacets);
3648 if (total <= flow_eviction_threshold) {
3649 return N_BUCKETS * BUCKET_WIDTH;
3652 /* Build histogram. */
3654 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3655 long long int idle = now - subfacet->used;
3656 int bucket = (idle <= 0 ? 0
3657 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3658 : (unsigned int) idle / BUCKET_WIDTH);
3662 /* Find the first bucket whose flows should be expired. */
3663 subtotal = bucket = 0;
3665 subtotal += buckets[bucket++];
3666 } while (bucket < N_BUCKETS &&
3667 subtotal < MAX(flow_eviction_threshold, total / 100));
3669 if (VLOG_IS_DBG_ENABLED()) {
3673 ds_put_cstr(&s, "keep");
3674 for (i = 0; i < N_BUCKETS; i++) {
3676 ds_put_cstr(&s, ", drop");
3679 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3682 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3686 return bucket * BUCKET_WIDTH;
3690 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3692 /* Cutoff time for most flows. */
3693 long long int normal_cutoff = time_msec() - dp_max_idle;
3695 /* We really want to keep flows for special protocols around, so use a more
3696 * conservative cutoff. */
3697 long long int special_cutoff = time_msec() - 10000;
3699 struct subfacet *subfacet, *next_subfacet;
3700 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3704 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3705 &backer->subfacets) {
3706 long long int cutoff;
3708 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3712 if (subfacet->used < cutoff) {
3713 if (subfacet->path != SF_NOT_INSTALLED) {
3714 batch[n_batch++] = subfacet;
3715 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3716 subfacet_destroy_batch(backer, batch, n_batch);
3720 subfacet_destroy(subfacet);
3726 subfacet_destroy_batch(backer, batch, n_batch);
3730 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3731 * then delete it entirely. */
3733 rule_expire(struct rule_dpif *rule)
3734 OVS_REQUIRES(ofproto_mutex)
3736 uint16_t idle_timeout, hard_timeout;
3737 long long int now = time_msec();
3740 ovs_assert(!rule->up.pending);
3742 /* Has 'rule' expired? */
3743 ovs_mutex_lock(&rule->up.mutex);
3744 hard_timeout = rule->up.hard_timeout;
3745 idle_timeout = rule->up.idle_timeout;
3746 if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3747 reason = OFPRR_HARD_TIMEOUT;
3748 } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3749 reason = OFPRR_IDLE_TIMEOUT;
3753 ovs_mutex_unlock(&rule->up.mutex);
3756 COVERAGE_INC(ofproto_dpif_expired);
3757 ofproto_rule_expire(&rule->up, reason);
3763 /* Creates and returns a new facet based on 'miss'.
3765 * The caller must already have determined that no facet with an identical
3766 * 'miss->flow' exists in 'miss->ofproto'.
3768 * 'rule' and 'xout' must have been created based on 'miss'.
3770 * 'facet'' statistics are initialized based on 'stats'.
3772 * The facet will initially have no subfacets. The caller should create (at
3773 * least) one subfacet with subfacet_create(). */
3774 static struct facet *
3775 facet_create(const struct flow_miss *miss)
3777 struct ofproto_dpif *ofproto = miss->ofproto;
3778 struct facet *facet;
3781 COVERAGE_INC(facet_create);
3782 facet = xzalloc(sizeof *facet);
3783 facet->ofproto = miss->ofproto;
3784 facet->used = miss->stats.used;
3785 facet->flow = miss->flow;
3786 facet->learn_rl = time_msec() + 500;
3788 list_init(&facet->subfacets);
3790 xlate_out_copy(&facet->xout, &miss->xout);
3792 match_init(&match, &facet->flow, &facet->xout.wc);
3793 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3794 ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3795 classifier_insert(&ofproto->facets, &facet->cr);
3796 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3802 facet_free(struct facet *facet)
3805 xlate_out_uninit(&facet->xout);
3810 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3811 * 'flow' must reflect the data in 'packet'. */
3813 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3814 const struct flow *flow,
3815 struct rule_dpif *rule,
3816 const struct ofpact *ofpacts, size_t ofpacts_len,
3817 struct ofpbuf *packet)
3819 struct odputil_keybuf keybuf;
3820 struct dpif_flow_stats stats;
3821 struct xlate_out xout;
3822 struct xlate_in xin;
3827 ovs_assert((rule != NULL) != (ofpacts != NULL));
3829 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3831 rule_dpif_credit_stats(rule, &stats);
3834 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3835 xin.ofpacts = ofpacts;
3836 xin.ofpacts_len = ofpacts_len;
3837 xin.resubmit_stats = &stats;
3838 xlate_actions(&xin, &xout);
3840 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3841 in_port = flow->in_port.ofp_port;
3842 if (in_port == OFPP_NONE) {
3843 in_port = OFPP_LOCAL;
3845 odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3847 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3848 xout.odp_actions.data, xout.odp_actions.size, packet,
3849 (xout.slow & SLOW_ACTION) != 0);
3850 xlate_out_uninit(&xout);
3855 /* Remove 'facet' from its ofproto and free up the associated memory:
3857 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3858 * rule's statistics, via subfacet_uninstall().
3860 * - Removes 'facet' from its rule and from ofproto->facets.
3863 facet_remove(struct facet *facet)
3865 struct subfacet *subfacet, *next_subfacet;
3867 COVERAGE_INC(facet_remove);
3868 ovs_assert(!list_is_empty(&facet->subfacets));
3870 /* First uninstall all of the subfacets to get final statistics. */
3871 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3872 subfacet_uninstall(subfacet);
3875 /* Flush the final stats to the rule.
3877 * This might require us to have at least one subfacet around so that we
3878 * can use its actions for accounting in facet_account(), which is why we
3879 * have uninstalled but not yet destroyed the subfacets. */
3880 facet_flush_stats(facet);
3882 /* Now we're really all done so destroy everything. */
3883 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3884 &facet->subfacets) {
3885 subfacet_destroy__(subfacet);
3887 ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3888 classifier_remove(&facet->ofproto->facets, &facet->cr);
3889 ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3890 cls_rule_destroy(&facet->cr);
3894 /* Feed information from 'facet' back into the learning table to keep it in
3895 * sync with what is actually flowing through the datapath. */
3897 facet_learn(struct facet *facet)
3899 long long int now = time_msec();
3901 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3905 facet->learn_rl = now + 500;
3907 if (!facet->xout.has_learn
3908 && !facet->xout.has_normal
3909 && (!facet->xout.has_fin_timeout
3910 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3914 facet_push_stats(facet, true);
3917 /* Returns true if the only action for 'facet' is to send to the controller.
3918 * (We don't report NetFlow expiration messages for such facets because they
3919 * are just part of the control logic for the network, not real traffic). */
3921 facet_is_controller_flow(struct facet *facet)
3924 struct ofproto_dpif *ofproto = facet->ofproto;
3925 const struct ofpact *ofpacts;
3926 struct rule_actions *actions;
3927 struct rule_dpif *rule;
3931 rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3932 actions = rule_dpif_get_actions(rule);
3933 rule_dpif_unref(rule);
3935 ofpacts_len = actions->ofpacts_len;
3936 ofpacts = actions->ofpacts;
3937 is_controller = ofpacts_len > 0
3938 && ofpacts->type == OFPACT_CONTROLLER
3939 && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3940 rule_actions_unref(actions);
3942 return is_controller;
3947 /* Folds all of 'facet''s statistics into its rule. Also updates the
3948 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3949 * 'facet''s statistics in the datapath should have been zeroed and folded into
3950 * its packet and byte counts before this function is called. */
3952 facet_flush_stats(struct facet *facet)
3954 struct ofproto_dpif *ofproto = facet->ofproto;
3955 struct subfacet *subfacet;
3957 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3958 ovs_assert(!subfacet->dp_byte_count);
3959 ovs_assert(!subfacet->dp_packet_count);
3962 facet_push_stats(facet, false);
3964 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3965 netflow_expire(ofproto->netflow, &facet->flow);
3966 netflow_flow_clear(ofproto->netflow, &facet->flow);
3969 /* Reset counters to prevent double counting if 'facet' ever gets
3971 facet_reset_counters(facet);
3972 facet->tcp_flags = 0;
3975 /* Searches 'ofproto''s table of facets for one which would be responsible for
3976 * 'flow'. Returns it if found, otherwise a null pointer.
3978 * The returned facet might need revalidation; use facet_lookup_valid()
3979 * instead if that is important. */
3980 static struct facet *
3981 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3983 struct cls_rule *cr;
3985 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
3986 cr = classifier_lookup(&ofproto->facets, flow, NULL);
3987 ovs_rwlock_unlock(&ofproto->facets.rwlock);
3988 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
3991 /* Searches 'ofproto''s table of facets for one capable that covers
3992 * 'flow'. Returns it if found, otherwise a null pointer.
3994 * The returned facet is guaranteed to be valid. */
3995 static struct facet *
3996 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3998 struct facet *facet;
4000 facet = facet_find(ofproto, flow);
4002 && ofproto->backer->need_revalidate
4003 && !facet_revalidate(facet)) {
4011 facet_check_consistency(struct facet *facet)
4013 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4015 struct xlate_out xout;
4016 struct xlate_in xin;
4019 /* Check the datapath actions for consistency. */
4020 xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4021 xlate_actions(&xin, &xout);
4023 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4024 && facet->xout.slow == xout.slow;
4025 if (!ok && !VLOG_DROP_WARN(&rl)) {
4026 struct ds s = DS_EMPTY_INITIALIZER;
4028 flow_format(&s, &facet->flow);
4029 ds_put_cstr(&s, ": inconsistency in facet");
4031 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4032 ds_put_cstr(&s, " (actions were: ");
4033 format_odp_actions(&s, facet->xout.odp_actions.data,
4034 facet->xout.odp_actions.size);
4035 ds_put_cstr(&s, ") (correct actions: ");
4036 format_odp_actions(&s, xout.odp_actions.data,
4037 xout.odp_actions.size);
4038 ds_put_char(&s, ')');
4041 if (facet->xout.slow != xout.slow) {
4042 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4047 xlate_out_uninit(&xout);
4052 /* Re-searches the classifier for 'facet':
4054 * - If the rule found is different from 'facet''s current rule, moves
4055 * 'facet' to the new rule and recompiles its actions.
4057 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4058 * where it is and recompiles its actions anyway.
4060 * - If any of 'facet''s subfacets correspond to a new flow according to
4061 * xlate_receive(), 'facet' is removed.
4063 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4065 facet_revalidate(struct facet *facet)
4067 struct ofproto_dpif *ofproto = facet->ofproto;
4068 struct rule_dpif *new_rule;
4069 struct subfacet *subfacet;
4070 struct flow_wildcards wc;
4071 struct xlate_out xout;
4072 struct xlate_in xin;
4074 COVERAGE_INC(facet_revalidate);
4076 /* Check that child subfacets still correspond to this facet. Tunnel
4077 * configuration changes could cause a subfacet's OpenFlow in_port to
4079 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4080 struct ofproto_dpif *recv_ofproto;
4081 struct flow recv_flow;
4084 error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4085 subfacet->key_len, &recv_flow, NULL,
4086 &recv_ofproto, NULL, NULL, NULL, NULL);
4088 || recv_ofproto != ofproto
4089 || facet != facet_find(ofproto, &recv_flow)) {
4090 facet_remove(facet);
4095 flow_wildcards_init_catchall(&wc);
4096 rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4098 /* Calculate new datapath actions.
4100 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4101 * emit a NetFlow expiration and, if so, we need to have the old state
4102 * around to properly compose it. */
4103 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4104 xlate_actions(&xin, &xout);
4105 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4106 /* Make sure non -packet fields are not masked. If not cleared,
4107 * the memcmp() below may fail, causing an otherwise valid facet
4109 flow_wildcards_clear_non_packet_fields(&xout.wc);
4111 /* A facet's slow path reason should only change under dramatic
4112 * circumstances. Rather than try to update everything, it's simpler to
4113 * remove the facet and start over.
4115 * More importantly, if a facet's wildcards change, it will be relatively
4116 * difficult to figure out if its subfacets still belong to it, and if not
4117 * which facet they may belong to. Again, to avoid the complexity, we
4118 * simply give up instead. */
4119 if (facet->xout.slow != xout.slow
4120 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4121 facet_remove(facet);
4122 xlate_out_uninit(&xout);
4123 rule_dpif_unref(new_rule);
4127 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4128 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4129 if (subfacet->path == SF_FAST_PATH) {
4130 struct dpif_flow_stats stats;
4132 subfacet_install(subfacet, &xout.odp_actions, &stats);
4133 subfacet_update_stats(subfacet, &stats);
4137 facet_flush_stats(facet);
4139 ofpbuf_clear(&facet->xout.odp_actions);
4140 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4141 xout.odp_actions.size);
4144 /* Update 'facet' now that we've taken care of all the old state. */
4145 facet->xout.slow = xout.slow;
4146 facet->xout.has_learn = xout.has_learn;
4147 facet->xout.has_normal = xout.has_normal;
4148 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4149 facet->xout.nf_output_iface = xout.nf_output_iface;
4150 facet->xout.mirrors = xout.mirrors;
4152 ovs_mutex_lock(&new_rule->up.mutex);
4153 facet->used = MAX(facet->used, new_rule->up.created);
4154 ovs_mutex_unlock(&new_rule->up.mutex);
4156 xlate_out_uninit(&xout);
4157 rule_dpif_unref(new_rule);
4162 facet_reset_counters(struct facet *facet)
4164 facet->packet_count = 0;
4165 facet->byte_count = 0;
4166 facet->prev_packet_count = 0;
4167 facet->prev_byte_count = 0;
4171 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4172 struct dpif_flow_stats *stats, bool may_learn)
4174 struct xlate_in xin;
4176 xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4177 xin.resubmit_stats = stats;
4178 xin.may_learn = may_learn;
4179 xlate_actions_for_side_effects(&xin);
4183 facet_push_stats(struct facet *facet, bool may_learn)
4185 struct dpif_flow_stats stats;
4187 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4188 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4189 ovs_assert(facet->used >= facet->prev_used);
4191 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4192 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4193 stats.used = facet->used;
4194 stats.tcp_flags = facet->tcp_flags;
4196 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4197 facet->prev_packet_count = facet->packet_count;
4198 facet->prev_byte_count = facet->byte_count;
4199 facet->prev_used = facet->used;
4200 flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4205 push_all_stats(void)
4207 static long long int rl = LLONG_MIN;
4208 struct ofproto_dpif *ofproto;
4210 if (time_msec() < rl) {
4214 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4215 struct cls_cursor cursor;
4216 struct facet *facet;
4218 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4219 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4220 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4221 facet_push_stats(facet, false);
4223 ovs_rwlock_unlock(&ofproto->facets.rwlock);
4226 rl = time_msec() + 100;
4230 rule_dpif_credit_stats(struct rule_dpif *rule,
4231 const struct dpif_flow_stats *stats)
4233 ovs_mutex_lock(&rule->stats_mutex);
4234 rule->packet_count += stats->n_packets;
4235 rule->byte_count += stats->n_bytes;
4236 rule->up.used = MAX(rule->up.used, stats->used);
4237 ovs_mutex_unlock(&rule->stats_mutex);
4241 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4243 return is_fail_open_rule(&rule->up);
4247 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4249 return rule_is_table_miss(&rule->up);
4253 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4254 OVS_REQUIRES(rule->up.mutex)
4256 return rule->up.flow_cookie;
4260 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4261 uint16_t hard_timeout)
4263 ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4266 /* Returns 'rule''s actions. The caller owns a reference on the returned
4267 * actions and must eventually release it (with rule_actions_unref()) to avoid
4269 struct rule_actions *
4270 rule_dpif_get_actions(const struct rule_dpif *rule)
4272 return rule_get_actions(&rule->up);
4277 static struct subfacet *
4278 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4279 size_t key_len, uint32_t key_hash)
4281 struct subfacet *subfacet;
4283 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4284 &backer->subfacets) {
4285 if (subfacet->key_len == key_len
4286 && !memcmp(key, subfacet->key, key_len)) {
4294 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4295 * 'key_hash' must be a hash over miss->key. The caller must have already
4296 * ensured that no subfacet subfacet already exists. */
4297 static struct subfacet *
4298 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4300 struct dpif_backer *backer = miss->ofproto->backer;
4301 const struct nlattr *key = miss->key;
4302 size_t key_len = miss->key_len;
4303 struct subfacet *subfacet;
4305 subfacet = (list_is_empty(&facet->subfacets)
4306 ? &facet->one_subfacet
4307 : xmalloc(sizeof *subfacet));
4309 COVERAGE_INC(subfacet_create);
4310 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4311 list_push_back(&facet->subfacets, &subfacet->list_node);
4312 subfacet->facet = facet;
4313 subfacet->key = xmemdup(key, key_len);
4314 subfacet->key_len = key_len;
4315 subfacet->used = miss->stats.used;
4316 subfacet->created = subfacet->used;
4317 subfacet->dp_packet_count = 0;
4318 subfacet->dp_byte_count = 0;
4319 subfacet->path = SF_NOT_INSTALLED;
4320 subfacet->backer = backer;
4325 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4326 * its facet within 'ofproto', and frees it. */
4328 subfacet_destroy__(struct subfacet *subfacet)
4330 struct facet *facet = subfacet->facet;
4332 COVERAGE_INC(subfacet_destroy);
4333 subfacet_uninstall(subfacet);
4334 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4335 list_remove(&subfacet->list_node);
4336 free(subfacet->key);
4337 if (subfacet != &facet->one_subfacet) {
4342 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4343 * last remaining subfacet in its facet destroys the facet too. */
4345 subfacet_destroy(struct subfacet *subfacet)
4347 struct facet *facet = subfacet->facet;
4349 if (list_is_singleton(&facet->subfacets)) {
4350 /* facet_remove() needs at least one subfacet (it will remove it). */
4351 facet_remove(facet);
4353 subfacet_destroy__(subfacet);
4358 subfacet_destroy_batch(struct dpif_backer *backer,
4359 struct subfacet **subfacets, int n)
4361 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4362 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4363 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4366 for (i = 0; i < n; i++) {
4367 ops[i].type = DPIF_OP_FLOW_DEL;
4368 ops[i].u.flow_del.key = subfacets[i]->key;
4369 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4370 ops[i].u.flow_del.stats = &stats[i];
4374 dpif_operate(backer->dpif, opsp, n);
4375 for (i = 0; i < n; i++) {
4376 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4377 subfacets[i]->path = SF_NOT_INSTALLED;
4378 subfacet_destroy(subfacets[i]);
4382 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4383 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4384 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4385 * since 'subfacet' was last updated.
4387 * Returns 0 if successful, otherwise a positive errno value. */
4389 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4390 struct dpif_flow_stats *stats)
4392 struct facet *facet = subfacet->facet;
4393 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4394 const struct nlattr *actions = odp_actions->data;
4395 size_t actions_len = odp_actions->size;
4396 struct odputil_keybuf maskbuf;
4399 uint64_t slow_path_stub[128 / 8];
4400 enum dpif_flow_put_flags flags;
4403 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4406 flags |= DPIF_FP_ZERO_STATS;
4409 if (path == SF_SLOW_PATH) {
4410 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4411 slow_path_stub, sizeof slow_path_stub,
4412 &actions, &actions_len);
4415 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4416 if (enable_megaflows) {
4417 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4418 &facet->flow, UINT32_MAX);
4421 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4422 subfacet->key_len, mask.data, mask.size,
4423 actions, actions_len, stats);
4426 subfacet_reset_dp_stats(subfacet, stats);
4430 COVERAGE_INC(subfacet_install_fail);
4432 subfacet->path = path;
4437 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4439 subfacet_uninstall(struct subfacet *subfacet)
4441 if (subfacet->path != SF_NOT_INSTALLED) {
4442 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4443 struct dpif_flow_stats stats;
4446 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4447 subfacet->key_len, &stats);
4448 subfacet_reset_dp_stats(subfacet, &stats);
4450 subfacet_update_stats(subfacet, &stats);
4452 subfacet->path = SF_NOT_INSTALLED;
4454 ovs_assert(subfacet->dp_packet_count == 0);
4455 ovs_assert(subfacet->dp_byte_count == 0);
4459 /* Resets 'subfacet''s datapath statistics counters. This should be called
4460 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4461 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4462 * was reset in the datapath. 'stats' will be modified to include only
4463 * statistics new since 'subfacet' was last updated. */
4465 subfacet_reset_dp_stats(struct subfacet *subfacet,
4466 struct dpif_flow_stats *stats)
4469 && subfacet->dp_packet_count <= stats->n_packets
4470 && subfacet->dp_byte_count <= stats->n_bytes) {
4471 stats->n_packets -= subfacet->dp_packet_count;
4472 stats->n_bytes -= subfacet->dp_byte_count;
4475 subfacet->dp_packet_count = 0;
4476 subfacet->dp_byte_count = 0;
4479 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4481 * Because of the meaning of a subfacet's counters, it only makes sense to do
4482 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4483 * represents a packet that was sent by hand or if it represents statistics
4484 * that have been cleared out of the datapath. */
4486 subfacet_update_stats(struct subfacet *subfacet,
4487 const struct dpif_flow_stats *stats)
4489 if (stats->n_packets || stats->used > subfacet->used) {
4490 struct facet *facet = subfacet->facet;
4492 subfacet->used = MAX(subfacet->used, stats->used);
4493 facet->used = MAX(facet->used, stats->used);
4494 facet->packet_count += stats->n_packets;
4495 facet->byte_count += stats->n_bytes;
4496 facet->tcp_flags |= stats->tcp_flags;
4502 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
4503 * the fields that were relevant as part of the lookup. */
4505 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4506 struct flow_wildcards *wc, struct rule_dpif **rule)
4508 struct ofport_dpif *port;
4510 if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4513 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4515 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4516 flow->in_port.ofp_port);
4519 choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4520 ofproto->no_packet_in_rule, rule);
4524 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4525 const struct flow *flow, struct flow_wildcards *wc,
4526 uint8_t table_id, struct rule_dpif **rule)
4528 const struct cls_rule *cls_rule;
4529 struct classifier *cls;
4533 if (table_id >= N_TABLES) {
4538 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4539 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4542 cls = &ofproto->up.tables[table_id].cls;
4543 ovs_rwlock_rdlock(&cls->rwlock);
4544 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4545 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4546 /* We must pretend that transport ports are unavailable. */
4547 struct flow ofpc_normal_flow = *flow;
4548 ofpc_normal_flow.tp_src = htons(0);
4549 ofpc_normal_flow.tp_dst = htons(0);
4550 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4551 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4552 cls_rule = &ofproto->drop_frags_rule->up.cr;
4553 /* Frag mask in wc already set above. */
4555 cls_rule = classifier_lookup(cls, flow, wc);
4558 *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4559 rule_dpif_ref(*rule);
4560 ovs_rwlock_unlock(&cls->rwlock);
4562 return *rule != NULL;
4565 /* Given a port configuration (specified as zero if there's no port), chooses
4566 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4567 * flow table miss. */
4569 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4570 struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4572 *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4573 rule_dpif_ref(*rule);
4577 rule_dpif_ref(struct rule_dpif *rule)
4580 ofproto_rule_ref(&rule->up);
4585 rule_dpif_unref(struct rule_dpif *rule)
4588 ofproto_rule_unref(&rule->up);
4593 complete_operation(struct rule_dpif *rule)
4594 OVS_REQUIRES(ofproto_mutex)
4596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4598 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4599 ofoperation_complete(rule->up.pending, 0);
4602 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4604 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4607 static struct rule *
4610 struct rule_dpif *rule = xmalloc(sizeof *rule);
4615 rule_dealloc(struct rule *rule_)
4617 struct rule_dpif *rule = rule_dpif_cast(rule_);
4622 rule_construct(struct rule *rule_)
4624 struct rule_dpif *rule = rule_dpif_cast(rule_);
4625 ovs_mutex_init(&rule->stats_mutex);
4626 ovs_mutex_lock(&rule->stats_mutex);
4627 rule->packet_count = 0;
4628 rule->byte_count = 0;
4629 ovs_mutex_unlock(&rule->stats_mutex);
4634 rule_insert(struct rule *rule_)
4635 OVS_REQUIRES(ofproto_mutex)
4637 struct rule_dpif *rule = rule_dpif_cast(rule_);
4638 complete_operation(rule);
4642 rule_delete(struct rule *rule_)
4643 OVS_REQUIRES(ofproto_mutex)
4645 struct rule_dpif *rule = rule_dpif_cast(rule_);
4646 complete_operation(rule);
4650 rule_destruct(struct rule *rule_)
4652 struct rule_dpif *rule = rule_dpif_cast(rule_);
4653 ovs_mutex_destroy(&rule->stats_mutex);
4657 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4659 struct rule_dpif *rule = rule_dpif_cast(rule_);
4663 /* Start from historical data for 'rule' itself that are no longer tracked
4664 * in facets. This counts, for example, facets that have expired. */
4665 ovs_mutex_lock(&rule->stats_mutex);
4666 *packets = rule->packet_count;
4667 *bytes = rule->byte_count;
4668 ovs_mutex_unlock(&rule->stats_mutex);
4672 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4673 struct ofpbuf *packet)
4675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4677 ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4681 rule_execute(struct rule *rule, const struct flow *flow,
4682 struct ofpbuf *packet)
4684 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4685 ofpbuf_delete(packet);
4690 rule_modify_actions(struct rule *rule_, bool reset_counters)
4691 OVS_REQUIRES(ofproto_mutex)
4693 struct rule_dpif *rule = rule_dpif_cast(rule_);
4695 if (reset_counters) {
4696 ovs_mutex_lock(&rule->stats_mutex);
4697 rule->packet_count = 0;
4698 rule->byte_count = 0;
4699 ovs_mutex_unlock(&rule->stats_mutex);
4702 complete_operation(rule);
4705 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4707 return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4710 static struct ofgroup *
4713 struct group_dpif *group = xzalloc(sizeof *group);
4718 group_dealloc(struct ofgroup *group_)
4720 struct group_dpif *group = group_dpif_cast(group_);
4725 group_construct_stats(struct group_dpif *group)
4726 OVS_REQUIRES(group->stats_mutex)
4728 group->packet_count = 0;
4729 group->byte_count = 0;
4730 if (!group->bucket_stats) {
4731 group->bucket_stats = xcalloc(group->up.n_buckets,
4732 sizeof *group->bucket_stats);
4734 memset(group->bucket_stats, 0, group->up.n_buckets *
4735 sizeof *group->bucket_stats);
4740 group_construct(struct ofgroup *group_)
4742 struct group_dpif *group = group_dpif_cast(group_);
4743 ovs_mutex_init(&group->stats_mutex);
4744 ovs_mutex_lock(&group->stats_mutex);
4745 group_construct_stats(group);
4746 ovs_mutex_unlock(&group->stats_mutex);
4751 group_destruct__(struct group_dpif *group)
4752 OVS_REQUIRES(group->stats_mutex)
4754 free(group->bucket_stats);
4755 group->bucket_stats = NULL;
4759 group_destruct(struct ofgroup *group_)
4761 struct group_dpif *group = group_dpif_cast(group_);
4762 ovs_mutex_lock(&group->stats_mutex);
4763 group_destruct__(group);
4764 ovs_mutex_unlock(&group->stats_mutex);
4765 ovs_mutex_destroy(&group->stats_mutex);
4769 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4771 struct group_dpif *group = group_dpif_cast(group_);
4772 struct group_dpif *victim = group_dpif_cast(victim_);
4774 ovs_mutex_lock(&group->stats_mutex);
4775 if (victim->up.n_buckets < group->up.n_buckets) {
4776 group_destruct__(group);
4778 group_construct_stats(group);
4779 ovs_mutex_unlock(&group->stats_mutex);
4785 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4787 struct group_dpif *group = group_dpif_cast(group_);
4789 /* Start from historical data for 'group' itself that are no longer tracked
4790 * in facets. This counts, for example, facets that have expired. */
4791 ovs_mutex_lock(&group->stats_mutex);
4792 ogs->packet_count = group->packet_count;
4793 ogs->byte_count = group->byte_count;
4794 memcpy(ogs->bucket_stats, group->bucket_stats,
4795 group->up.n_buckets * sizeof *group->bucket_stats);
4796 ovs_mutex_unlock(&group->stats_mutex);
4802 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4803 struct group_dpif **group)
4804 OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4806 struct ofgroup *ofgroup;
4810 found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4811 *group = found ? group_dpif_cast(ofgroup) : NULL;
4817 group_dpif_release(struct group_dpif *group)
4818 OVS_RELEASES(group->up.rwlock)
4820 ofproto_group_release(&group->up);
4824 group_dpif_get_buckets(const struct group_dpif *group,
4825 const struct list **buckets)
4827 *buckets = &group->up.buckets;
4830 enum ofp11_group_type
4831 group_dpif_get_type(const struct group_dpif *group)
4833 return group->up.type;
4836 /* Sends 'packet' out 'ofport'.
4837 * May modify 'packet'.
4838 * Returns 0 if successful, otherwise a positive errno value. */
4840 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4845 error = xlate_send_packet(ofport, packet);
4847 ovs_mutex_lock(&ofproto->stats_mutex);
4848 ofproto->stats.tx_packets++;
4849 ofproto->stats.tx_bytes += packet->size;
4850 ovs_mutex_unlock(&ofproto->stats_mutex);
4854 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4855 * The action will state 'slow' as the reason that the action is in the slow
4856 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4857 * dump-flows" output to see why a flow is in the slow path.)
4859 * The 'stub_size' bytes in 'stub' will be used to store the action.
4860 * 'stub_size' must be large enough for the action.
4862 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4865 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4866 enum slow_path_reason slow,
4867 uint64_t *stub, size_t stub_size,
4868 const struct nlattr **actionsp, size_t *actions_lenp)
4870 union user_action_cookie cookie;
4873 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4874 cookie.slow_path.unused = 0;
4875 cookie.slow_path.reason = slow;
4877 ofpbuf_use_stack(&buf, stub, stub_size);
4878 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4879 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4881 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4883 odp_port_t odp_port;
4886 odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4887 pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4888 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4890 *actionsp = buf.data;
4891 *actions_lenp = buf.size;
4895 set_frag_handling(struct ofproto *ofproto_,
4896 enum ofp_config_flags frag_handling)
4898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4899 if (frag_handling != OFPC_FRAG_REASM) {
4900 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4908 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4909 const struct flow *flow,
4910 const struct ofpact *ofpacts, size_t ofpacts_len)
4912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4914 ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4915 ofpacts_len, packet);
4922 set_netflow(struct ofproto *ofproto_,
4923 const struct netflow_options *netflow_options)
4925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4927 if (netflow_options) {
4928 if (!ofproto->netflow) {
4929 ofproto->netflow = netflow_create();
4930 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4932 return netflow_set_options(ofproto->netflow, netflow_options);
4933 } else if (ofproto->netflow) {
4934 ofproto->backer->need_revalidate = REV_RECONFIGURE;
4935 netflow_unref(ofproto->netflow);
4936 ofproto->netflow = NULL;
4943 get_netflow_ids(const struct ofproto *ofproto_,
4944 uint8_t *engine_type, uint8_t *engine_id)
4946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4948 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4951 static struct ofproto_dpif *
4952 ofproto_dpif_lookup(const char *name)
4954 struct ofproto_dpif *ofproto;
4956 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4957 hash_string(name, 0), &all_ofproto_dpifs) {
4958 if (!strcmp(ofproto->up.name, name)) {
4966 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4967 const char *argv[], void *aux OVS_UNUSED)
4969 struct ofproto_dpif *ofproto;
4972 ofproto = ofproto_dpif_lookup(argv[1]);
4974 unixctl_command_reply_error(conn, "no such bridge");
4977 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4978 mac_learning_flush(ofproto->ml);
4979 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4981 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4982 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4983 mac_learning_flush(ofproto->ml);
4984 ovs_rwlock_unlock(&ofproto->ml->rwlock);
4988 unixctl_command_reply(conn, "table successfully flushed");
4991 static struct ofport_dpif *
4992 ofbundle_get_a_port(const struct ofbundle *bundle)
4994 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4999 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5000 const char *argv[], void *aux OVS_UNUSED)
5002 struct ds ds = DS_EMPTY_INITIALIZER;
5003 const struct ofproto_dpif *ofproto;
5004 const struct mac_entry *e;
5006 ofproto = ofproto_dpif_lookup(argv[1]);
5008 unixctl_command_reply_error(conn, "no such bridge");
5012 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5013 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5014 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5015 struct ofbundle *bundle = e->port.p;
5016 char name[OFP_MAX_PORT_NAME_LEN];
5018 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5020 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5021 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5022 mac_entry_age(ofproto->ml, e));
5024 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5025 unixctl_command_reply(conn, ds_cstr(&ds));
5030 struct xlate_out xout;
5031 struct xlate_in xin;
5037 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5039 struct rule_actions *actions;
5042 ds_put_char_multiple(result, '\t', level);
5044 ds_put_cstr(result, "No match\n");
5048 ovs_mutex_lock(&rule->up.mutex);
5049 cookie = rule->up.flow_cookie;
5050 ovs_mutex_unlock(&rule->up.mutex);
5052 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5053 rule ? rule->up.table_id : 0, ntohll(cookie));
5054 cls_rule_format(&rule->up.cr, result);
5055 ds_put_char(result, '\n');
5057 actions = rule_dpif_get_actions(rule);
5059 ds_put_char_multiple(result, '\t', level);
5060 ds_put_cstr(result, "OpenFlow actions=");
5061 ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5062 ds_put_char(result, '\n');
5064 rule_actions_unref(actions);
5068 trace_format_flow(struct ds *result, int level, const char *title,
5069 struct trace_ctx *trace)
5071 ds_put_char_multiple(result, '\t', level);
5072 ds_put_format(result, "%s: ", title);
5073 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5074 ds_put_cstr(result, "unchanged");
5076 flow_format(result, &trace->xin.flow);
5077 trace->flow = trace->xin.flow;
5079 ds_put_char(result, '\n');
5083 trace_format_regs(struct ds *result, int level, const char *title,
5084 struct trace_ctx *trace)
5088 ds_put_char_multiple(result, '\t', level);
5089 ds_put_format(result, "%s:", title);
5090 for (i = 0; i < FLOW_N_REGS; i++) {
5091 ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5093 ds_put_char(result, '\n');
5097 trace_format_odp(struct ds *result, int level, const char *title,
5098 struct trace_ctx *trace)
5100 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5102 ds_put_char_multiple(result, '\t', level);
5103 ds_put_format(result, "%s: ", title);
5104 format_odp_actions(result, odp_actions->data, odp_actions->size);
5105 ds_put_char(result, '\n');
5109 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5111 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5112 struct ds *result = trace->result;
5114 ds_put_char(result, '\n');
5115 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5116 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5117 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5118 trace_format_rule(result, recurse + 1, rule);
5122 trace_report(struct xlate_in *xin, const char *s, int recurse)
5124 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5125 struct ds *result = trace->result;
5127 ds_put_char_multiple(result, '\t', recurse);
5128 ds_put_cstr(result, s);
5129 ds_put_char(result, '\n');
5132 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
5133 * forms are supported:
5135 * - [dpname] odp_flow [-generate | packet]
5136 * - bridge br_flow [-generate | packet]
5138 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
5139 * returns a nonnull error message. */
5141 parse_flow_and_packet(int argc, const char *argv[],
5142 struct ofproto_dpif **ofprotop, struct flow *flow,
5143 struct ofpbuf **packetp)
5145 const struct dpif_backer *backer = NULL;
5146 const char *error = NULL;
5147 struct simap port_names = SIMAP_INITIALIZER(&port_names);
5148 struct ofpbuf *packet;
5149 struct ofpbuf odp_key;
5150 struct ofpbuf odp_mask;
5152 ofpbuf_init(&odp_key, 0);
5153 ofpbuf_init(&odp_mask, 0);
5155 /* Handle "-generate" or a hex string as the last argument. */
5156 if (!strcmp(argv[argc - 1], "-generate")) {
5157 packet = ofpbuf_new(0);
5160 error = eth_from_hex(argv[argc - 1], &packet);
5163 } else if (argc == 4) {
5164 /* The 3-argument form must end in "-generate' or a hex string. */
5169 /* odp_flow can have its in_port specified as a name instead of port no.
5170 * We do not yet know whether a given flow is a odp_flow or a br_flow.
5171 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5172 * we need to create a simap of name to port no. */
5174 const char *dp_type;
5175 if (!strncmp(argv[1], "ovs-", 4)) {
5176 dp_type = argv[1] + 4;
5180 backer = shash_find_data(&all_dpif_backers, dp_type);
5181 } else if (argc == 2) {
5182 struct shash_node *node;
5183 if (shash_count(&all_dpif_backers) == 1) {
5184 node = shash_first(&all_dpif_backers);
5185 backer = node->data;
5188 error = "Syntax error";
5191 if (backer && backer->dpif) {
5192 struct dpif_port dpif_port;
5193 struct dpif_port_dump port_dump;
5194 DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5195 simap_put(&port_names, dpif_port.name,
5196 odp_to_u32(dpif_port.port_no));
5200 /* Parse the flow and determine whether a datapath or
5201 * bridge is specified. If function odp_flow_key_from_string()
5202 * returns 0, the flow is a odp_flow. If function
5203 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5204 if (!odp_flow_from_string(argv[argc - 1], &port_names,
5205 &odp_key, &odp_mask)) {
5207 error = "Cannot find the datapath";
5211 if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5212 NULL, ofprotop, NULL, NULL, NULL, NULL)) {
5213 error = "Invalid datapath flow";
5216 } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5218 error = "Must specify bridge name";
5222 *ofprotop = ofproto_dpif_lookup(argv[1]);
5224 error = "Unknown bridge name";
5228 error = "Bad flow syntax";
5232 /* Generate a packet, if requested. */
5234 if (!packet->size) {
5235 flow_compose(packet, flow);
5237 union flow_in_port in_port = flow->in_port;
5239 /* Use the metadata from the flow and the packet argument
5240 * to reconstruct the flow. */
5241 flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5250 ofpbuf_delete(packet);
5254 ofpbuf_uninit(&odp_key);
5255 ofpbuf_uninit(&odp_mask);
5256 simap_destroy(&port_names);
5261 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5262 void *aux OVS_UNUSED)
5264 struct ofproto_dpif *ofproto;
5265 struct ofpbuf *packet;
5269 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5274 ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5275 unixctl_command_reply(conn, ds_cstr(&result));
5276 ds_destroy(&result);
5277 ofpbuf_delete(packet);
5279 unixctl_command_reply_error(conn, error);
5284 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5285 const char *argv[], void *aux OVS_UNUSED)
5287 enum ofputil_protocol usable_protocols;
5288 struct ofproto_dpif *ofproto;
5289 bool enforce_consistency;
5290 struct ofpbuf ofpacts;
5291 struct ofpbuf *packet;
5296 /* Three kinds of error return values! */
5303 ofpbuf_init(&ofpacts, 0);
5305 /* Parse actions. */
5306 rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5308 unixctl_command_reply_error(conn, rw_error);
5313 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5314 * consistency between the flow and the actions. With -consistent, we
5315 * enforce consistency even for a flow supported in OpenFlow 1.0. */
5316 if (!strcmp(argv[1], "-consistent")) {
5317 enforce_consistency = true;
5321 enforce_consistency = false;
5324 error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5326 unixctl_command_reply_error(conn, error);
5330 /* Do the same checks as handle_packet_out() in ofproto.c.
5332 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5333 * strictly correct because these actions aren't in any table, but it's OK
5334 * because it 'table_id' is used only to check goto_table instructions, but
5335 * packet-outs take a list of actions and therefore it can't include
5338 * We skip the "meter" check here because meter is an instruction, not an
5339 * action, and thus cannot appear in ofpacts. */
5340 in_port = ofp_to_u16(flow.in_port.ofp_port);
5341 if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5342 unixctl_command_reply_error(conn, "invalid in_port");
5345 if (enforce_consistency) {
5346 retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5347 u16_to_ofp(ofproto->up.max_ports),
5348 0, 0, usable_protocols);
5350 retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5351 u16_to_ofp(ofproto->up.max_ports), 0, 0,
5357 ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5358 unixctl_command_reply_error(conn, ds_cstr(&result));
5362 ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5363 unixctl_command_reply(conn, ds_cstr(&result));
5366 ds_destroy(&result);
5367 ofpbuf_delete(packet);
5368 ofpbuf_uninit(&ofpacts);
5371 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5372 * description of the results to 'ds'.
5374 * The trace follows a packet with the specified 'flow' through the flow
5375 * table. 'packet' may be nonnull to trace an actual packet, with consequent
5376 * side effects (if it is nonnull then its flow must be 'flow').
5378 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5379 * trace, otherwise the actions are determined by a flow table lookup. */
5381 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5382 const struct ofpbuf *packet,
5383 const struct ofpact ofpacts[], size_t ofpacts_len,
5386 struct rule_dpif *rule;
5387 struct flow_wildcards wc;
5389 ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5390 ds_put_cstr(ds, "Flow: ");
5391 flow_format(ds, flow);
5392 ds_put_char(ds, '\n');
5394 flow_wildcards_init_catchall(&wc);
5398 rule_dpif_lookup(ofproto, flow, &wc, &rule);
5400 trace_format_rule(ds, 0, rule);
5401 if (rule == ofproto->miss_rule) {
5402 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5403 } else if (rule == ofproto->no_packet_in_rule) {
5404 ds_put_cstr(ds, "\nNo match, packets dropped because "
5405 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5406 } else if (rule == ofproto->drop_frags_rule) {
5407 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5408 "and the fragment handling mode is \"drop\".\n");
5412 if (rule || ofpacts) {
5413 uint64_t odp_actions_stub[1024 / 8];
5414 struct ofpbuf odp_actions;
5415 struct trace_ctx trace;
5419 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5422 ofpbuf_use_stub(&odp_actions,
5423 odp_actions_stub, sizeof odp_actions_stub);
5424 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5426 trace.xin.ofpacts = ofpacts;
5427 trace.xin.ofpacts_len = ofpacts_len;
5429 trace.xin.resubmit_hook = trace_resubmit;
5430 trace.xin.report_hook = trace_report;
5432 xlate_actions(&trace.xin, &trace.xout);
5433 flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5435 ds_put_char(ds, '\n');
5436 trace_format_flow(ds, 0, "Final flow", &trace);
5438 match_init(&match, flow, &trace.xout.wc);
5439 ds_put_cstr(ds, "Relevant fields: ");
5440 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5441 ds_put_char(ds, '\n');
5443 ds_put_cstr(ds, "Datapath actions: ");
5444 format_odp_actions(ds, trace.xout.odp_actions.data,
5445 trace.xout.odp_actions.size);
5447 if (trace.xout.slow) {
5448 enum slow_path_reason slow;
5450 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5451 "slow path because it:");
5453 slow = trace.xout.slow;
5455 enum slow_path_reason bit = rightmost_1bit(slow);
5457 ds_put_format(ds, "\n\t- %s.",
5458 slow_path_reason_to_explanation(bit));
5464 xlate_out_uninit(&trace.xout);
5467 rule_dpif_unref(rule);
5470 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5471 * 'reply' describing the results. */
5473 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5475 struct cls_cursor cursor;
5476 struct facet *facet;
5480 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5481 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5482 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5483 if (!facet_check_consistency(facet)) {
5487 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5489 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5493 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5494 ofproto->up.name, errors);
5496 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5501 ofproto_dpif_self_check(struct unixctl_conn *conn,
5502 int argc, const char *argv[], void *aux OVS_UNUSED)
5504 struct ds reply = DS_EMPTY_INITIALIZER;
5505 struct ofproto_dpif *ofproto;
5508 ofproto = ofproto_dpif_lookup(argv[1]);
5510 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5511 "ofproto/list for help)");
5514 ofproto_dpif_self_check__(ofproto, &reply);
5516 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5517 ofproto_dpif_self_check__(ofproto, &reply);
5521 unixctl_command_reply(conn, ds_cstr(&reply));
5525 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
5526 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
5527 * to destroy 'ofproto_shash' and free the returned value. */
5528 static const struct shash_node **
5529 get_ofprotos(struct shash *ofproto_shash)
5531 const struct ofproto_dpif *ofproto;
5533 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5534 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5535 shash_add_nocopy(ofproto_shash, name, ofproto);
5538 return shash_sort(ofproto_shash);
5542 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5543 const char *argv[] OVS_UNUSED,
5544 void *aux OVS_UNUSED)
5546 struct ds ds = DS_EMPTY_INITIALIZER;
5547 struct shash ofproto_shash;
5548 const struct shash_node **sorted_ofprotos;
5551 shash_init(&ofproto_shash);
5552 sorted_ofprotos = get_ofprotos(&ofproto_shash);
5553 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5554 const struct shash_node *node = sorted_ofprotos[i];
5555 ds_put_format(&ds, "%s\n", node->name);
5558 shash_destroy(&ofproto_shash);
5559 free(sorted_ofprotos);
5561 unixctl_command_reply(conn, ds_cstr(&ds));
5566 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5568 const struct shash_node **ofprotos;
5569 struct ofproto_dpif *ofproto;
5570 struct shash ofproto_shash;
5571 uint64_t n_hit, n_missed;
5574 n_hit = n_missed = 0;
5575 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5576 if (ofproto->backer == backer) {
5577 n_missed += ofproto->n_missed;
5578 n_hit += ofproto->n_hit;
5582 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5583 dpif_name(backer->dpif), n_hit, n_missed);
5584 ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5585 " life span: %lldms\n", hmap_count(&backer->subfacets),
5586 backer->avg_n_subfacet, backer->max_n_subfacet,
5587 backer->avg_subfacet_life);
5589 shash_init(&ofproto_shash);
5590 ofprotos = get_ofprotos(&ofproto_shash);
5591 for (i = 0; i < shash_count(&ofproto_shash); i++) {
5592 struct ofproto_dpif *ofproto = ofprotos[i]->data;
5593 const struct shash_node **ports;
5596 if (ofproto->backer != backer) {
5600 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5601 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5603 ports = shash_sort(&ofproto->up.port_by_name);
5604 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5605 const struct shash_node *node = ports[j];
5606 struct ofport *ofport = node->data;
5608 odp_port_t odp_port;
5610 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5613 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5614 if (odp_port != ODPP_NONE) {
5615 ds_put_format(ds, "%"PRIu32":", odp_port);
5617 ds_put_cstr(ds, "none:");
5620 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5623 if (!netdev_get_config(ofport->netdev, &config)) {
5624 const struct smap_node **nodes;
5627 nodes = smap_sort(&config);
5628 for (i = 0; i < smap_count(&config); i++) {
5629 const struct smap_node *node = nodes[i];
5630 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5631 node->key, node->value);
5635 smap_destroy(&config);
5637 ds_put_char(ds, ')');
5638 ds_put_char(ds, '\n');
5642 shash_destroy(&ofproto_shash);
5647 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5648 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5650 struct ds ds = DS_EMPTY_INITIALIZER;
5651 const struct shash_node **backers;
5654 backers = shash_sort(&all_dpif_backers);
5655 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5656 dpif_show_backer(backers[i]->data, &ds);
5660 unixctl_command_reply(conn, ds_cstr(&ds));
5664 /* Dump the megaflow (facet) cache. This is useful to check the
5665 * correctness of flow wildcarding, since the same mechanism is used for
5666 * both xlate caching and kernel wildcarding.
5668 * It's important to note that in the output the flow description uses
5669 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5671 * This command is only needed for advanced debugging, so it's not
5672 * documented in the man page. */
5674 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5675 int argc OVS_UNUSED, const char *argv[],
5676 void *aux OVS_UNUSED)
5678 struct ds ds = DS_EMPTY_INITIALIZER;
5679 const struct ofproto_dpif *ofproto;
5680 long long int now = time_msec();
5681 struct cls_cursor cursor;
5682 struct facet *facet;
5684 ofproto = ofproto_dpif_lookup(argv[1]);
5686 unixctl_command_reply_error(conn, "no such bridge");
5690 ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5691 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5692 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5693 cls_rule_format(&facet->cr, &ds);
5694 ds_put_cstr(&ds, ", ");
5695 ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5696 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5697 ds_put_cstr(&ds, "Datapath actions: ");
5698 if (facet->xout.slow) {
5699 uint64_t slow_path_stub[128 / 8];
5700 const struct nlattr *actions;
5703 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5704 slow_path_stub, sizeof slow_path_stub,
5705 &actions, &actions_len);
5706 format_odp_actions(&ds, actions, actions_len);
5708 format_odp_actions(&ds, facet->xout.odp_actions.data,
5709 facet->xout.odp_actions.size);
5711 ds_put_cstr(&ds, "\n");
5713 ovs_rwlock_unlock(&ofproto->facets.rwlock);
5715 ds_chomp(&ds, '\n');
5716 unixctl_command_reply(conn, ds_cstr(&ds));
5720 /* Disable using the megaflows.
5722 * This command is only needed for advanced debugging, so it's not
5723 * documented in the man page. */
5725 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5726 int argc OVS_UNUSED,
5727 const char *argv[] OVS_UNUSED,
5728 void *aux OVS_UNUSED)
5730 struct ofproto_dpif *ofproto;
5732 enable_megaflows = false;
5734 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5735 flush(&ofproto->up);
5738 unixctl_command_reply(conn, "megaflows disabled");
5741 /* Re-enable using megaflows.
5743 * This command is only needed for advanced debugging, so it's not
5744 * documented in the man page. */
5746 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5747 int argc OVS_UNUSED,
5748 const char *argv[] OVS_UNUSED,
5749 void *aux OVS_UNUSED)
5751 struct ofproto_dpif *ofproto;
5753 enable_megaflows = true;
5755 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5756 flush(&ofproto->up);
5759 unixctl_command_reply(conn, "megaflows enabled");
5763 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5764 int argc OVS_UNUSED, const char *argv[],
5765 void *aux OVS_UNUSED)
5767 struct ds ds = DS_EMPTY_INITIALIZER;
5768 const struct ofproto_dpif *ofproto;
5769 struct subfacet *subfacet;
5771 ofproto = ofproto_dpif_lookup(argv[1]);
5773 unixctl_command_reply_error(conn, "no such bridge");
5777 update_stats(ofproto->backer);
5779 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5780 struct facet *facet = subfacet->facet;
5781 struct odputil_keybuf maskbuf;
5784 if (facet->ofproto != ofproto) {
5788 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5789 if (enable_megaflows) {
5790 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5791 &facet->flow, UINT32_MAX);
5794 odp_flow_format(subfacet->key, subfacet->key_len,
5795 mask.data, mask.size, NULL, &ds, false);
5797 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5798 subfacet->dp_packet_count, subfacet->dp_byte_count);
5799 if (subfacet->used) {
5800 ds_put_format(&ds, "%.3fs",
5801 (time_msec() - subfacet->used) / 1000.0);
5803 ds_put_format(&ds, "never");
5805 if (subfacet->facet->tcp_flags) {
5806 ds_put_cstr(&ds, ", flags:");
5807 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5810 ds_put_cstr(&ds, ", actions:");
5811 if (facet->xout.slow) {
5812 uint64_t slow_path_stub[128 / 8];
5813 const struct nlattr *actions;
5816 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5817 slow_path_stub, sizeof slow_path_stub,
5818 &actions, &actions_len);
5819 format_odp_actions(&ds, actions, actions_len);
5821 format_odp_actions(&ds, facet->xout.odp_actions.data,
5822 facet->xout.odp_actions.size);
5824 ds_put_char(&ds, '\n');
5827 unixctl_command_reply(conn, ds_cstr(&ds));
5832 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5833 int argc OVS_UNUSED, const char *argv[],
5834 void *aux OVS_UNUSED)
5836 struct ds ds = DS_EMPTY_INITIALIZER;
5837 struct ofproto_dpif *ofproto;
5839 ofproto = ofproto_dpif_lookup(argv[1]);
5841 unixctl_command_reply_error(conn, "no such bridge");
5845 flush(&ofproto->up);
5847 unixctl_command_reply(conn, ds_cstr(&ds));
5852 ofproto_dpif_unixctl_init(void)
5854 static bool registered;
5860 unixctl_command_register(
5862 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5863 1, 3, ofproto_unixctl_trace, NULL);
5864 unixctl_command_register(
5865 "ofproto/trace-packet-out",
5866 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5867 2, 6, ofproto_unixctl_trace_actions, NULL);
5868 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5869 ofproto_unixctl_fdb_flush, NULL);
5870 unixctl_command_register("fdb/show", "bridge", 1, 1,
5871 ofproto_unixctl_fdb_show, NULL);
5872 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5873 ofproto_dpif_self_check, NULL);
5874 unixctl_command_register("dpif/dump-dps", "", 0, 0,
5875 ofproto_unixctl_dpif_dump_dps, NULL);
5876 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5878 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5879 ofproto_unixctl_dpif_dump_flows, NULL);
5880 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5881 ofproto_unixctl_dpif_del_flows, NULL);
5882 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5883 ofproto_unixctl_dpif_dump_megaflows, NULL);
5884 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5885 ofproto_unixctl_dpif_disable_megaflows, NULL);
5886 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5887 ofproto_unixctl_dpif_enable_megaflows, NULL);
5890 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5892 * This is deprecated. It is only for compatibility with broken device drivers
5893 * in old versions of Linux that do not properly support VLANs when VLAN
5894 * devices are not used. When broken device drivers are no longer in
5895 * widespread use, we will delete these interfaces. */
5898 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5901 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5903 if (realdev_ofp_port == ofport->realdev_ofp_port
5904 && vid == ofport->vlandev_vid) {
5908 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5910 if (ofport->realdev_ofp_port) {
5913 if (realdev_ofp_port && ofport->bundle) {
5914 /* vlandevs are enslaved to their realdevs, so they are not allowed to
5915 * themselves be part of a bundle. */
5916 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5919 ofport->realdev_ofp_port = realdev_ofp_port;
5920 ofport->vlandev_vid = vid;
5922 if (realdev_ofp_port) {
5923 vsp_add(ofport, realdev_ofp_port, vid);
5930 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5932 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5936 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5937 OVS_EXCLUDED(ofproto->vsp_mutex)
5941 ovs_mutex_lock(&ofproto->vsp_mutex);
5942 ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5943 ovs_mutex_unlock(&ofproto->vsp_mutex);
5948 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5949 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5950 OVS_REQUIRES(ofproto->vsp_mutex)
5952 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5953 int vid = vlan_tci_to_vid(vlan_tci);
5954 const struct vlan_splinter *vsp;
5956 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5957 hash_realdev_vid(realdev_ofp_port, vid),
5958 &ofproto->realdev_vid_map) {
5959 if (vsp->realdev_ofp_port == realdev_ofp_port
5960 && vsp->vid == vid) {
5961 return vsp->vlandev_ofp_port;
5965 return realdev_ofp_port;
5968 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5969 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5970 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
5971 * 'vlan_tci' 9, it would return the port number of eth0.9.
5973 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5974 * function just returns its 'realdev_ofp_port' argument. */
5976 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5977 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5978 OVS_EXCLUDED(ofproto->vsp_mutex)
5982 ovs_mutex_lock(&ofproto->vsp_mutex);
5983 ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5984 ovs_mutex_unlock(&ofproto->vsp_mutex);
5988 static struct vlan_splinter *
5989 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5991 struct vlan_splinter *vsp;
5993 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5994 hash_ofp_port(vlandev_ofp_port),
5995 &ofproto->vlandev_map) {
5996 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6004 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6005 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6006 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6007 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6008 * eth0 and store 9 in '*vid'.
6010 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6011 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6014 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6015 ofp_port_t vlandev_ofp_port, int *vid)
6016 OVS_REQUIRES(ofproto->vsp_mutex)
6018 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6019 const struct vlan_splinter *vsp;
6021 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6026 return vsp->realdev_ofp_port;
6032 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6033 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6034 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6035 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6036 * always the case unless VLAN splinters are enabled), returns false without
6037 * making any changes. */
6039 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6040 OVS_EXCLUDED(ofproto->vsp_mutex)
6045 ovs_mutex_lock(&ofproto->vsp_mutex);
6046 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6047 ovs_mutex_unlock(&ofproto->vsp_mutex);
6052 /* Cause the flow to be processed as if it came in on the real device with
6053 * the VLAN device's VLAN ID. */
6054 flow->in_port.ofp_port = realdev;
6055 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6060 vsp_remove(struct ofport_dpif *port)
6062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6063 struct vlan_splinter *vsp;
6065 ovs_mutex_lock(&ofproto->vsp_mutex);
6066 vsp = vlandev_find(ofproto, port->up.ofp_port);
6068 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6069 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6072 port->realdev_ofp_port = 0;
6074 VLOG_ERR("missing vlan device record");
6076 ovs_mutex_unlock(&ofproto->vsp_mutex);
6080 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6084 ovs_mutex_lock(&ofproto->vsp_mutex);
6085 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6086 && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6087 == realdev_ofp_port)) {
6088 struct vlan_splinter *vsp;
6090 vsp = xmalloc(sizeof *vsp);
6091 vsp->realdev_ofp_port = realdev_ofp_port;
6092 vsp->vlandev_ofp_port = port->up.ofp_port;
6095 port->realdev_ofp_port = realdev_ofp_port;
6097 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6098 hash_ofp_port(port->up.ofp_port));
6099 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6100 hash_realdev_vid(realdev_ofp_port, vid));
6102 VLOG_ERR("duplicate vlan device record");
6104 ovs_mutex_unlock(&ofproto->vsp_mutex);
6108 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6110 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6111 return ofport ? ofport->odp_port : ODPP_NONE;
6114 struct ofport_dpif *
6115 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6117 struct ofport_dpif *port;
6119 ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6120 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6121 &backer->odp_to_ofport_map) {
6122 if (port->odp_port == odp_port) {
6123 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6128 ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6133 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6135 struct ofport_dpif *port;
6137 port = odp_port_to_ofport(ofproto->backer, odp_port);
6138 if (port && &ofproto->up == port->up.ofproto) {
6139 return port->up.ofp_port;
6145 const struct ofproto_class ofproto_dpif_class = {
6160 type_get_memory_usage,
6179 port_is_lacp_current,
6180 NULL, /* rule_choose_table */
6189 rule_modify_actions,
6203 get_stp_port_status,
6211 is_mirror_output_bundle,
6212 forward_bpdu_changed,
6213 set_mac_table_config,
6215 NULL, /* meter_get_features */
6216 NULL, /* meter_set */
6217 NULL, /* meter_get */
6218 NULL, /* meter_del */
6219 group_alloc, /* group_alloc */
6220 group_construct, /* group_construct */
6221 group_destruct, /* group_destruct */
6222 group_dealloc, /* group_dealloc */
6223 group_modify, /* group_modify */
6224 group_get_stats, /* group_get_stats */