2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "meta-flow.h"
39 #include "multipath.h"
40 #include "netdev-vport.h"
45 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofp-parse.h"
50 #include "ofp-print.h"
51 #include "ofproto-dpif-governor.h"
52 #include "ofproto-dpif-ipfix.h"
53 #include "ofproto-dpif-mirror.h"
54 #include "ofproto-dpif-sflow.h"
55 #include "ofproto-dpif-xlate.h"
56 #include "poll-loop.h"
61 #include "unaligned.h"
63 #include "vlan-bitmap.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
68 COVERAGE_DEFINE(ofproto_dpif_expired);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(facet_unexpected);
72 COVERAGE_DEFINE(facet_suppress);
73 COVERAGE_DEFINE(subfacet_install_fail);
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
83 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
85 struct flow_wildcards *wc);
87 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
90 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
91 struct ofproto_dpif *ofproto; /* Owning ofproto. */
92 void *aux; /* Key supplied by ofproto's client. */
93 char *name; /* Identifier for log messages. */
96 struct list ports; /* Contains "struct ofport"s. */
97 enum port_vlan_mode vlan_mode; /* VLAN mode */
98 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
99 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
100 * NULL if all VLANs are trunked. */
101 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
102 struct bond *bond; /* Nonnull iff more than one port. */
103 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
106 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
109 static void bundle_remove(struct ofport *);
110 static void bundle_update(struct ofbundle *);
111 static void bundle_destroy(struct ofbundle *);
112 static void bundle_del_port(struct ofport_dpif *);
113 static void bundle_run(struct ofbundle *);
114 static void bundle_wait(struct ofbundle *);
116 static void stp_run(struct ofproto_dpif *ofproto);
117 static void stp_wait(struct ofproto_dpif *ofproto);
118 static int set_stp_port(struct ofport *,
119 const struct ofproto_port_stp_settings *);
121 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
122 enum slow_path_reason,
123 uint64_t *stub, size_t stub_size,
124 const struct nlattr **actionsp,
125 size_t *actions_lenp);
127 /* A subfacet (see "struct subfacet" below) has three possible installation
130 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
131 * case just after the subfacet is created, just before the subfacet is
132 * destroyed, or if the datapath returns an error when we try to install a
135 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
137 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
138 * ofproto_dpif is installed in the datapath.
141 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
142 SF_FAST_PATH, /* Full actions are installed. */
143 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
146 /* A dpif flow and actions associated with a facet.
148 * See also the large comment on struct facet. */
151 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
152 struct list list_node; /* In struct facet's 'facets' list. */
153 struct facet *facet; /* Owning facet. */
154 struct dpif_backer *backer; /* Owning backer. */
156 enum odp_key_fitness key_fitness;
160 long long int used; /* Time last used; time created if not used. */
161 long long int created; /* Time created. */
163 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
164 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
166 enum subfacet_path path; /* Installed in datapath? */
169 #define SUBFACET_DESTROY_MAX_BATCH 50
171 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
173 static struct subfacet *subfacet_find(struct dpif_backer *,
174 const struct nlattr *key, size_t key_len,
176 static void subfacet_destroy(struct subfacet *);
177 static void subfacet_destroy__(struct subfacet *);
178 static void subfacet_destroy_batch(struct dpif_backer *,
179 struct subfacet **, int n);
180 static void subfacet_reset_dp_stats(struct subfacet *,
181 struct dpif_flow_stats *);
182 static void subfacet_update_stats(struct subfacet *,
183 const struct dpif_flow_stats *);
184 static int subfacet_install(struct subfacet *,
185 const struct ofpbuf *odp_actions,
186 struct dpif_flow_stats *);
187 static void subfacet_uninstall(struct subfacet *);
189 /* A unique, non-overlapping instantiation of an OpenFlow flow.
191 * A facet associates a "struct flow", which represents the Open vSwitch
192 * userspace idea of an exact-match flow, with one or more subfacets.
193 * While the facet is created based on an exact-match flow, it is stored
194 * within the ofproto based on the wildcards that could be expressed
195 * based on the flow table and other configuration. (See the 'wc'
196 * description in "struct xlate_out" for more details.)
198 * Each subfacet tracks the datapath's idea of the flow equivalent to
199 * the facet. When the kernel module (or other dpif implementation) and
200 * Open vSwitch userspace agree on the definition of a flow key, there
201 * is exactly one subfacet per facet. If the dpif implementation
202 * supports more-specific flow matching than userspace, however, a facet
203 * can have more than one subfacet. Examples include the dpif
204 * implementation not supporting the same wildcards as userspace or some
205 * distinction in flow that userspace simply doesn't understand.
207 * Flow expiration works in terms of subfacets, so a facet must have at
208 * least one subfacet or it will never expire, leaking memory. */
211 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
212 struct ofproto_dpif *ofproto;
215 struct list subfacets;
216 long long int used; /* Time last used; time created if not used. */
219 struct flow flow; /* Flow of the creating subfacet. */
220 struct cls_rule cr; /* In 'ofproto_dpif's facets classifier. */
224 * - Do include packets and bytes sent "by hand", e.g. with
227 * - Do include packets and bytes that were obtained from the datapath
228 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
229 * DPIF_FP_ZERO_STATS).
231 * - Do not include packets or bytes that can be obtained from the
232 * datapath for any existing subfacet.
234 uint64_t packet_count; /* Number of packets received. */
235 uint64_t byte_count; /* Number of bytes received. */
237 /* Resubmit statistics. */
238 uint64_t prev_packet_count; /* Number of packets from last stats push. */
239 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
240 long long int prev_used; /* Used time from last stats push. */
243 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
244 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
245 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
247 struct xlate_out xout;
248 bool fail_open; /* Facet matched the fail open rule. */
250 /* Storage for a single subfacet, to reduce malloc() time and space
251 * overhead. (A facet always has at least one subfacet and in the common
252 * case has exactly one subfacet. However, 'one_subfacet' may not
253 * always be valid, since it could have been removed after newer
254 * subfacets were pushed onto the 'subfacets' list.) */
255 struct subfacet one_subfacet;
257 long long int learn_rl; /* Rate limiter for facet_learn(). */
260 static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *,
262 struct dpif_flow_stats *);
263 static void facet_remove(struct facet *);
264 static void facet_free(struct facet *);
266 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
267 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
268 const struct flow *);
269 static bool facet_revalidate(struct facet *);
270 static bool facet_check_consistency(struct facet *);
272 static void facet_flush_stats(struct facet *);
274 static void facet_reset_counters(struct facet *);
275 static void facet_push_stats(struct facet *, bool may_learn);
276 static void facet_learn(struct facet *);
277 static void facet_account(struct facet *);
278 static void push_all_stats(void);
280 static bool facet_is_controller_flow(struct facet *);
283 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
287 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
288 struct list bundle_node; /* In struct ofbundle's "ports" list. */
289 struct cfm *cfm; /* Connectivity Fault Management, if any. */
290 struct bfd *bfd; /* BFD, if any. */
291 bool may_enable; /* May be enabled in bonds. */
292 bool is_tunnel; /* This port is a tunnel. */
293 long long int carrier_seq; /* Carrier status changes. */
294 struct ofport_dpif *peer; /* Peer if patch port. */
297 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
298 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
299 long long int stp_state_entered;
301 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
303 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
305 * This is deprecated. It is only for compatibility with broken device
306 * drivers in old versions of Linux that do not properly support VLANs when
307 * VLAN devices are not used. When broken device drivers are no longer in
308 * widespread use, we will delete these interfaces. */
309 ofp_port_t realdev_ofp_port;
313 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
314 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
315 * traffic egressing the 'ofport' with that priority should be marked with. */
316 struct priority_to_dscp {
317 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
318 uint32_t priority; /* Priority of this queue (see struct flow). */
320 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
323 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
325 * This is deprecated. It is only for compatibility with broken device drivers
326 * in old versions of Linux that do not properly support VLANs when VLAN
327 * devices are not used. When broken device drivers are no longer in
328 * widespread use, we will delete these interfaces. */
329 struct vlan_splinter {
330 struct hmap_node realdev_vid_node;
331 struct hmap_node vlandev_node;
332 ofp_port_t realdev_ofp_port;
333 ofp_port_t vlandev_ofp_port;
337 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
338 static void vsp_remove(struct ofport_dpif *);
339 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
341 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
344 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
347 static struct ofport_dpif *
348 ofport_dpif_cast(const struct ofport *ofport)
350 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
353 static void port_run(struct ofport_dpif *);
354 static void port_run_fast(struct ofport_dpif *);
355 static void port_wait(struct ofport_dpif *);
356 static int set_bfd(struct ofport *, const struct smap *);
357 static int set_cfm(struct ofport *, const struct cfm_settings *);
358 static void ofport_clear_priorities(struct ofport_dpif *);
359 static void ofport_update_peer(struct ofport_dpif *);
360 static void run_fast_rl(void);
362 struct dpif_completion {
363 struct list list_node;
364 struct ofoperation *op;
367 /* Reasons that we might need to revalidate every facet, and corresponding
370 * A value of 0 means that there is no need to revalidate.
372 * It would be nice to have some cleaner way to integrate with coverage
373 * counters, but with only a few reasons I guess this is good enough for
375 enum revalidate_reason {
376 REV_RECONFIGURE = 1, /* Switch configuration changed. */
377 REV_STP, /* Spanning tree protocol port status change. */
378 REV_BOND, /* Bonding changed. */
379 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
380 REV_FLOW_TABLE, /* Flow table changed. */
381 REV_MAC_LEARNING, /* Mac learning changed. */
382 REV_INCONSISTENCY /* Facet self-check failed. */
384 COVERAGE_DEFINE(rev_reconfigure);
385 COVERAGE_DEFINE(rev_stp);
386 COVERAGE_DEFINE(rev_bond);
387 COVERAGE_DEFINE(rev_port_toggled);
388 COVERAGE_DEFINE(rev_flow_table);
389 COVERAGE_DEFINE(rev_mac_learning);
390 COVERAGE_DEFINE(rev_inconsistency);
392 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
393 * These are datapath flows which have no associated ofproto, if they did we
394 * would use facets. */
396 struct hmap_node hmap_node;
401 struct avg_subfacet_rates {
402 double add_rate; /* Moving average of new flows created per minute. */
403 double del_rate; /* Moving average of flows deleted per minute. */
406 /* All datapaths of a given type share a single dpif backer instance. */
411 struct timer next_expiration;
412 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
414 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
416 /* Facet revalidation flags applying to facets which use this backer. */
417 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
419 struct hmap drop_keys; /* Set of dropped odp keys. */
420 bool recv_set_enable; /* Enables or disables receiving packets. */
422 struct hmap subfacets;
423 struct governor *governor;
425 /* Subfacet statistics.
427 * These keep track of the total number of subfacets added and deleted and
428 * flow life span. They are useful for computing the flow rates stats
429 * exposed via "ovs-appctl dpif/show". The goal is to learn about
430 * traffic patterns in ways that we can use later to improve Open vSwitch
431 * performance in new situations. */
432 long long int created; /* Time when it is created. */
433 unsigned max_n_subfacet; /* Maximum number of flows */
434 unsigned avg_n_subfacet; /* Average number of flows. */
435 long long int avg_subfacet_life; /* Average life span of subfacets. */
437 /* The average number of subfacets... */
438 struct avg_subfacet_rates hourly; /* ...over the last hour. */
439 struct avg_subfacet_rates daily; /* ...over the last day. */
440 struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
441 long long int last_minute; /* Last time 'hourly' was updated. */
443 /* Number of subfacets added or deleted since 'last_minute'. */
444 unsigned subfacet_add_count;
445 unsigned subfacet_del_count;
447 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
448 unsigned long long int total_subfacet_add_count;
449 unsigned long long int total_subfacet_del_count;
452 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
453 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
455 static void drop_key_clear(struct dpif_backer *);
456 static struct ofport_dpif *
457 odp_port_to_ofport(const struct dpif_backer *, odp_port_t odp_port);
458 static void update_moving_averages(struct dpif_backer *backer);
460 struct ofproto_dpif {
461 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
463 struct dpif_backer *backer;
465 /* Special OpenFlow rules. */
466 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
467 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
468 struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
471 struct netflow *netflow;
472 struct dpif_sflow *sflow;
473 struct dpif_ipfix *ipfix;
474 struct hmap bundles; /* Contains "struct ofbundle"s. */
475 struct mac_learning *ml;
476 bool has_bonded_bundles;
477 struct mbridge *mbridge;
480 struct classifier facets; /* Contains 'struct facet's. */
481 long long int consistency_rl;
483 /* Support for debugging async flow mods. */
484 struct list completions;
486 struct netdev_stats stats; /* To account packets generated and consumed in
491 long long int stp_last_tick;
493 /* VLAN splinters. */
494 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
495 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
498 struct sset ports; /* Set of standard port names. */
499 struct sset ghost_ports; /* Ports with no datapath port. */
500 struct sset port_poll_set; /* Queued names for port_poll() reply. */
501 int port_poll_errno; /* Last errno for port_poll() reply. */
503 /* Per ofproto's dpif stats. */
508 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
509 * for debugging the asynchronous flow_mod implementation.) */
512 /* By default, flows in the datapath are wildcarded (megaflows). They
513 * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
514 static bool enable_megaflows = true;
516 /* All existing ofproto_dpif instances, indexed by ->up.name. */
517 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
519 static void ofproto_dpif_unixctl_init(void);
521 static inline struct ofproto_dpif *
522 ofproto_dpif_cast(const struct ofproto *ofproto)
524 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
525 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
528 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
529 ofp_port_t ofp_port);
532 #define FLOW_MISS_MAX_BATCH 50
533 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
535 /* Flow expiration. */
536 static int expire(struct dpif_backer *);
539 static void send_netflow_active_timeouts(struct ofproto_dpif *);
542 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
544 /* Global variables. */
545 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
547 /* Initial mappings of port to bridge mappings. */
548 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
551 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
552 struct ofputil_flow_mod *fm)
554 return ofproto_flow_mod(&ofproto->up, fm);
558 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
559 struct ofputil_packet_in *pin)
561 connmgr_send_packet_in(ofproto->up.connmgr, pin);
564 /* Factory functions. */
567 init(const struct shash *iface_hints)
569 struct shash_node *node;
571 /* Make a local copy, since we don't own 'iface_hints' elements. */
572 SHASH_FOR_EACH(node, iface_hints) {
573 const struct iface_hint *orig_hint = node->data;
574 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
576 new_hint->br_name = xstrdup(orig_hint->br_name);
577 new_hint->br_type = xstrdup(orig_hint->br_type);
578 new_hint->ofp_port = orig_hint->ofp_port;
580 shash_add(&init_ofp_ports, node->name, new_hint);
585 enumerate_types(struct sset *types)
587 dp_enumerate_types(types);
591 enumerate_names(const char *type, struct sset *names)
593 struct ofproto_dpif *ofproto;
596 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
597 if (strcmp(type, ofproto->up.type)) {
600 sset_add(names, ofproto->up.name);
607 del(const char *type, const char *name)
612 error = dpif_open(name, type, &dpif);
614 error = dpif_delete(dpif);
621 port_open_type(const char *datapath_type, const char *port_type)
623 return dpif_port_open_type(datapath_type, port_type);
626 /* Type functions. */
628 static void process_dpif_port_changes(struct dpif_backer *);
629 static void process_dpif_all_ports_changed(struct dpif_backer *);
630 static void process_dpif_port_change(struct dpif_backer *,
631 const char *devname);
632 static void process_dpif_port_error(struct dpif_backer *, int error);
634 static struct ofproto_dpif *
635 lookup_ofproto_dpif_by_port_name(const char *name)
637 struct ofproto_dpif *ofproto;
639 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
640 if (sset_contains(&ofproto->ports, name)) {
649 type_run(const char *type)
651 static long long int push_timer = LLONG_MIN;
652 struct dpif_backer *backer;
654 backer = shash_find_data(&all_dpif_backers, type);
656 /* This is not necessarily a problem, since backers are only
657 * created on demand. */
661 dpif_run(backer->dpif);
663 /* The most natural place to push facet statistics is when they're pulled
664 * from the datapath. However, when there are many flows in the datapath,
665 * this expensive operation can occur so frequently, that it reduces our
666 * ability to quickly set up flows. To reduce the cost, we push statistics
668 if (time_msec() > push_timer) {
669 push_timer = time_msec() + 2000;
673 /* If vswitchd started with other_config:flow_restore_wait set as "true",
674 * and the configuration has now changed to "false", enable receiving
675 * packets from the datapath. */
676 if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
679 backer->recv_set_enable = true;
681 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
683 VLOG_ERR("Failed to enable receiving packets in dpif.");
686 dpif_flow_flush(backer->dpif);
687 backer->need_revalidate = REV_RECONFIGURE;
690 if (backer->need_revalidate) {
691 struct ofproto_dpif *ofproto;
692 struct simap_node *node;
693 struct simap tmp_backers;
695 /* Handle tunnel garbage collection. */
696 simap_init(&tmp_backers);
697 simap_swap(&backer->tnl_backers, &tmp_backers);
699 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
700 struct ofport_dpif *iter;
702 if (backer != ofproto->backer) {
706 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
707 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
710 if (!iter->is_tunnel) {
714 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
715 namebuf, sizeof namebuf);
716 node = simap_find(&tmp_backers, dp_port);
718 simap_put(&backer->tnl_backers, dp_port, node->data);
719 simap_delete(&tmp_backers, node);
720 node = simap_find(&backer->tnl_backers, dp_port);
722 node = simap_find(&backer->tnl_backers, dp_port);
724 odp_port_t odp_port = ODPP_NONE;
726 if (!dpif_port_add(backer->dpif, iter->up.netdev,
728 simap_put(&backer->tnl_backers, dp_port,
729 odp_to_u32(odp_port));
730 node = simap_find(&backer->tnl_backers, dp_port);
735 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
736 if (tnl_port_reconfigure(iter, iter->up.netdev,
738 backer->need_revalidate = REV_RECONFIGURE;
743 SIMAP_FOR_EACH (node, &tmp_backers) {
744 dpif_port_del(backer->dpif, u32_to_odp(node->data));
746 simap_destroy(&tmp_backers);
748 switch (backer->need_revalidate) {
749 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
750 case REV_STP: COVERAGE_INC(rev_stp); break;
751 case REV_BOND: COVERAGE_INC(rev_bond); break;
752 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
753 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
754 case REV_MAC_LEARNING: COVERAGE_INC(rev_mac_learning); break;
755 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
757 backer->need_revalidate = 0;
759 /* Clear the drop_keys in case we should now be accepting some
760 * formerly dropped flows. */
761 drop_key_clear(backer);
763 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
764 struct facet *facet, *next;
765 struct ofport_dpif *ofport;
766 struct cls_cursor cursor;
767 struct ofbundle *bundle;
769 if (ofproto->backer != backer) {
773 xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
774 ofproto->mbridge, ofproto->sflow,
775 ofproto->ipfix, ofproto->up.frag_handling,
776 ofproto->up.forward_bpdu,
777 connmgr_has_in_band(ofproto->up.connmgr),
778 ofproto->netflow != NULL,
779 ofproto->stp != NULL);
781 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
782 xlate_bundle_set(ofproto, bundle, bundle->name,
783 bundle->vlan_mode, bundle->vlan,
784 bundle->trunks, bundle->use_priority_tags,
785 bundle->bond, bundle->lacp,
789 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
790 xlate_ofport_set(ofproto, ofport->bundle, ofport,
791 ofport->up.ofp_port, ofport->odp_port,
792 ofport->up.netdev, ofport->cfm,
793 ofport->bfd, ofport->peer,
794 ofport->up.pp.config, ofport->stp_state,
795 ofport->is_tunnel, ofport->may_enable);
798 cls_cursor_init(&cursor, &ofproto->facets, NULL);
799 CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
800 facet_revalidate(facet);
806 if (!backer->recv_set_enable) {
807 /* Wake up before a max of 1000ms. */
808 timer_set_duration(&backer->next_expiration, 1000);
809 } else if (timer_expired(&backer->next_expiration)) {
810 int delay = expire(backer);
811 timer_set_duration(&backer->next_expiration, delay);
814 process_dpif_port_changes(backer);
816 if (backer->governor) {
819 governor_run(backer->governor);
821 /* If the governor has shrunk to its minimum size and the number of
822 * subfacets has dwindled, then drop the governor entirely.
824 * For hysteresis, the number of subfacets to drop the governor is
825 * smaller than the number needed to trigger its creation. */
826 n_subfacets = hmap_count(&backer->subfacets);
827 if (n_subfacets * 4 < flow_eviction_threshold
828 && governor_is_idle(backer->governor)) {
829 governor_destroy(backer->governor);
830 backer->governor = NULL;
837 /* Check for and handle port changes in 'backer''s dpif. */
839 process_dpif_port_changes(struct dpif_backer *backer)
845 error = dpif_port_poll(backer->dpif, &devname);
851 process_dpif_all_ports_changed(backer);
855 process_dpif_port_change(backer, devname);
860 process_dpif_port_error(backer, error);
867 process_dpif_all_ports_changed(struct dpif_backer *backer)
869 struct ofproto_dpif *ofproto;
870 struct dpif_port dpif_port;
871 struct dpif_port_dump dump;
872 struct sset devnames;
875 sset_init(&devnames);
876 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
877 if (ofproto->backer == backer) {
878 struct ofport *ofport;
880 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
881 sset_add(&devnames, netdev_get_name(ofport->netdev));
885 DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
886 sset_add(&devnames, dpif_port.name);
889 SSET_FOR_EACH (devname, &devnames) {
890 process_dpif_port_change(backer, devname);
892 sset_destroy(&devnames);
896 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
898 struct ofproto_dpif *ofproto;
899 struct dpif_port port;
901 /* Don't report on the datapath's device. */
902 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
906 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
907 &all_ofproto_dpifs) {
908 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
913 ofproto = lookup_ofproto_dpif_by_port_name(devname);
914 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
915 /* The port was removed. If we know the datapath,
916 * report it through poll_set(). If we don't, it may be
917 * notifying us of a removal we initiated, so ignore it.
918 * If there's a pending ENOBUFS, let it stand, since
919 * everything will be reevaluated. */
920 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
921 sset_add(&ofproto->port_poll_set, devname);
922 ofproto->port_poll_errno = 0;
924 } else if (!ofproto) {
925 /* The port was added, but we don't know with which
926 * ofproto we should associate it. Delete it. */
927 dpif_port_del(backer->dpif, port.port_no);
929 struct ofport_dpif *ofport;
931 ofport = ofport_dpif_cast(shash_find_data(
932 &ofproto->up.port_by_name, devname));
934 && ofport->odp_port != port.port_no
935 && !odp_port_to_ofport(backer, port.port_no))
937 /* 'ofport''s datapath port number has changed from
938 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
939 * structures to match. */
940 hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
941 ofport->odp_port = port.port_no;
942 hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
943 hash_odp_port(port.port_no));
944 backer->need_revalidate = REV_RECONFIGURE;
947 dpif_port_destroy(&port);
950 /* Propagate 'error' to all ofprotos based on 'backer'. */
952 process_dpif_port_error(struct dpif_backer *backer, int error)
954 struct ofproto_dpif *ofproto;
956 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
957 if (ofproto->backer == backer) {
958 sset_clear(&ofproto->port_poll_set);
959 ofproto->port_poll_errno = error;
965 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
969 /* If recv_set_enable is false, we should not handle upcalls. */
970 if (!backer->recv_set_enable) {
974 /* Handle one or more batches of upcalls, until there's nothing left to do
975 * or until we do a fixed total amount of work.
977 * We do work in batches because it can be much cheaper to set up a number
978 * of flows and fire off their patches all at once. We do multiple batches
979 * because in some cases handling a packet can cause another packet to be
980 * queued almost immediately as part of the return flow. Both
981 * optimizations can make major improvements on some benchmarks and
982 * presumably for real traffic as well. */
984 while (work < max_batch) {
985 int retval = handle_upcalls(backer, max_batch - work);
996 type_run_fast(const char *type)
998 struct dpif_backer *backer;
1000 backer = shash_find_data(&all_dpif_backers, type);
1002 /* This is not necessarily a problem, since backers are only
1003 * created on demand. */
1007 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1013 static long long int port_rl = LLONG_MIN;
1014 static unsigned int backer_rl = 0;
1016 if (time_msec() >= port_rl) {
1017 struct ofproto_dpif *ofproto;
1018 struct ofport_dpif *ofport;
1020 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1022 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1023 port_run_fast(ofport);
1026 port_rl = time_msec() + 200;
1029 /* XXX: We have to be careful not to do too much work in this function. If
1030 * we call dpif_backer_run_fast() too often, or with too large a batch,
1031 * performance improves signifcantly, but at a cost. It's possible for the
1032 * number of flows in the datapath to increase without bound, and for poll
1033 * loops to take 10s of seconds. The correct solution to this problem,
1034 * long term, is to separate flow miss handling into it's own thread so it
1035 * isn't affected by revalidations, and expirations. Until then, this is
1036 * the best we can do. */
1037 if (++backer_rl >= 10) {
1038 struct shash_node *node;
1041 SHASH_FOR_EACH (node, &all_dpif_backers) {
1042 dpif_backer_run_fast(node->data, 1);
1048 type_wait(const char *type)
1050 struct dpif_backer *backer;
1052 backer = shash_find_data(&all_dpif_backers, type);
1054 /* This is not necessarily a problem, since backers are only
1055 * created on demand. */
1059 if (backer->governor) {
1060 governor_wait(backer->governor);
1063 timer_wait(&backer->next_expiration);
1066 /* Basic life-cycle. */
1068 static int add_internal_flows(struct ofproto_dpif *);
1070 static struct ofproto *
1073 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1074 return &ofproto->up;
1078 dealloc(struct ofproto *ofproto_)
1080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085 close_dpif_backer(struct dpif_backer *backer)
1087 struct shash_node *node;
1089 ovs_assert(backer->refcount > 0);
1091 if (--backer->refcount) {
1095 drop_key_clear(backer);
1096 hmap_destroy(&backer->drop_keys);
1098 simap_destroy(&backer->tnl_backers);
1099 hmap_destroy(&backer->odp_to_ofport_map);
1100 node = shash_find(&all_dpif_backers, backer->type);
1102 shash_delete(&all_dpif_backers, node);
1103 dpif_close(backer->dpif);
1105 ovs_assert(hmap_is_empty(&backer->subfacets));
1106 hmap_destroy(&backer->subfacets);
1107 governor_destroy(backer->governor);
1112 /* Datapath port slated for removal from datapath. */
1113 struct odp_garbage {
1114 struct list list_node;
1115 odp_port_t odp_port;
1119 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1121 struct dpif_backer *backer;
1122 struct dpif_port_dump port_dump;
1123 struct dpif_port port;
1124 struct shash_node *node;
1125 struct list garbage_list;
1126 struct odp_garbage *garbage, *next;
1132 backer = shash_find_data(&all_dpif_backers, type);
1139 backer_name = xasprintf("ovs-%s", type);
1141 /* Remove any existing datapaths, since we assume we're the only
1142 * userspace controlling the datapath. */
1144 dp_enumerate_names(type, &names);
1145 SSET_FOR_EACH(name, &names) {
1146 struct dpif *old_dpif;
1148 /* Don't remove our backer if it exists. */
1149 if (!strcmp(name, backer_name)) {
1153 if (dpif_open(name, type, &old_dpif)) {
1154 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1156 dpif_delete(old_dpif);
1157 dpif_close(old_dpif);
1160 sset_destroy(&names);
1162 backer = xmalloc(sizeof *backer);
1164 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1167 VLOG_ERR("failed to open datapath of type %s: %s", type,
1168 ovs_strerror(error));
1173 backer->type = xstrdup(type);
1174 backer->governor = NULL;
1175 backer->refcount = 1;
1176 hmap_init(&backer->odp_to_ofport_map);
1177 hmap_init(&backer->drop_keys);
1178 hmap_init(&backer->subfacets);
1179 timer_set_duration(&backer->next_expiration, 1000);
1180 backer->need_revalidate = 0;
1181 simap_init(&backer->tnl_backers);
1182 backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1185 if (backer->recv_set_enable) {
1186 dpif_flow_flush(backer->dpif);
1189 /* Loop through the ports already on the datapath and remove any
1190 * that we don't need anymore. */
1191 list_init(&garbage_list);
1192 dpif_port_dump_start(&port_dump, backer->dpif);
1193 while (dpif_port_dump_next(&port_dump, &port)) {
1194 node = shash_find(&init_ofp_ports, port.name);
1195 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1196 garbage = xmalloc(sizeof *garbage);
1197 garbage->odp_port = port.port_no;
1198 list_push_front(&garbage_list, &garbage->list_node);
1201 dpif_port_dump_done(&port_dump);
1203 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1204 dpif_port_del(backer->dpif, garbage->odp_port);
1205 list_remove(&garbage->list_node);
1209 shash_add(&all_dpif_backers, type, backer);
1211 error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1213 VLOG_ERR("failed to listen on datapath of type %s: %s",
1214 type, ovs_strerror(error));
1215 close_dpif_backer(backer);
1219 backer->max_n_subfacet = 0;
1220 backer->created = time_msec();
1221 backer->last_minute = backer->created;
1222 memset(&backer->hourly, 0, sizeof backer->hourly);
1223 memset(&backer->daily, 0, sizeof backer->daily);
1224 memset(&backer->lifetime, 0, sizeof backer->lifetime);
1225 backer->subfacet_add_count = 0;
1226 backer->subfacet_del_count = 0;
1227 backer->total_subfacet_add_count = 0;
1228 backer->total_subfacet_del_count = 0;
1229 backer->avg_n_subfacet = 0;
1230 backer->avg_subfacet_life = 0;
1236 construct(struct ofproto *ofproto_)
1238 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1239 struct shash_node *node, *next;
1240 odp_port_t max_ports;
1243 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1248 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1249 ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1250 ofp_to_u16(OFPP_MAX))));
1252 ofproto->netflow = NULL;
1253 ofproto->sflow = NULL;
1254 ofproto->ipfix = NULL;
1255 ofproto->stp = NULL;
1256 hmap_init(&ofproto->bundles);
1257 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1258 ofproto->mbridge = mbridge_create();
1259 ofproto->has_bonded_bundles = false;
1261 classifier_init(&ofproto->facets);
1262 ofproto->consistency_rl = LLONG_MIN;
1264 list_init(&ofproto->completions);
1266 ofproto_dpif_unixctl_init();
1268 hmap_init(&ofproto->vlandev_map);
1269 hmap_init(&ofproto->realdev_vid_map);
1271 sset_init(&ofproto->ports);
1272 sset_init(&ofproto->ghost_ports);
1273 sset_init(&ofproto->port_poll_set);
1274 ofproto->port_poll_errno = 0;
1276 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1277 struct iface_hint *iface_hint = node->data;
1279 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1280 /* Check if the datapath already has this port. */
1281 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1282 sset_add(&ofproto->ports, node->name);
1285 free(iface_hint->br_name);
1286 free(iface_hint->br_type);
1288 shash_delete(&init_ofp_ports, node);
1292 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1293 hash_string(ofproto->up.name, 0));
1294 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1296 ofproto_init_tables(ofproto_, N_TABLES);
1297 error = add_internal_flows(ofproto);
1298 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1301 ofproto->n_missed = 0;
1307 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1308 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1310 struct ofputil_flow_mod fm;
1313 match_init_catchall(&fm.match);
1315 match_set_reg(&fm.match, 0, id);
1316 fm.new_cookie = htonll(0);
1317 fm.cookie = htonll(0);
1318 fm.cookie_mask = htonll(0);
1319 fm.modify_cookie = false;
1320 fm.table_id = TBL_INTERNAL;
1321 fm.command = OFPFC_ADD;
1322 fm.idle_timeout = 0;
1323 fm.hard_timeout = 0;
1327 fm.ofpacts = ofpacts->data;
1328 fm.ofpacts_len = ofpacts->size;
1330 error = ofproto_flow_mod(&ofproto->up, &fm);
1332 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1333 id, ofperr_to_string(error));
1337 *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL,
1339 ovs_assert(*rulep != NULL);
1345 add_internal_flows(struct ofproto_dpif *ofproto)
1347 struct ofpact_controller *controller;
1348 uint64_t ofpacts_stub[128 / 8];
1349 struct ofpbuf ofpacts;
1353 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1356 controller = ofpact_put_CONTROLLER(&ofpacts);
1357 controller->max_len = UINT16_MAX;
1358 controller->controller_id = 0;
1359 controller->reason = OFPR_NO_MATCH;
1360 ofpact_pad(&ofpacts);
1362 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1367 ofpbuf_clear(&ofpacts);
1368 error = add_internal_flow(ofproto, id++, &ofpacts,
1369 &ofproto->no_packet_in_rule);
1374 error = add_internal_flow(ofproto, id++, &ofpacts,
1375 &ofproto->drop_frags_rule);
1380 complete_operations(struct ofproto_dpif *ofproto)
1382 struct dpif_completion *c, *next;
1384 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1385 ofoperation_complete(c->op, 0);
1386 list_remove(&c->list_node);
1392 destruct(struct ofproto *ofproto_)
1394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1395 struct rule_dpif *rule, *next_rule;
1396 struct oftable *table;
1398 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1399 xlate_remove_ofproto(ofproto);
1401 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1402 complete_operations(ofproto);
1404 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1405 struct cls_cursor cursor;
1407 cls_cursor_init(&cursor, &table->cls, NULL);
1408 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1409 ofproto_rule_destroy(&rule->up);
1413 mbridge_unref(ofproto->mbridge);
1415 netflow_destroy(ofproto->netflow);
1416 dpif_sflow_unref(ofproto->sflow);
1417 hmap_destroy(&ofproto->bundles);
1418 mac_learning_unref(ofproto->ml);
1420 classifier_destroy(&ofproto->facets);
1422 hmap_destroy(&ofproto->vlandev_map);
1423 hmap_destroy(&ofproto->realdev_vid_map);
1425 sset_destroy(&ofproto->ports);
1426 sset_destroy(&ofproto->ghost_ports);
1427 sset_destroy(&ofproto->port_poll_set);
1429 close_dpif_backer(ofproto->backer);
1433 run_fast(struct ofproto *ofproto_)
1435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1436 struct ofport_dpif *ofport;
1438 /* Do not perform any periodic activity required by 'ofproto' while
1439 * waiting for flow restore to complete. */
1440 if (ofproto_get_flow_restore_wait()) {
1444 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1445 port_run_fast(ofport);
1452 run(struct ofproto *ofproto_)
1454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1455 struct ofport_dpif *ofport;
1456 struct ofbundle *bundle;
1460 complete_operations(ofproto);
1463 if (mbridge_need_revalidate(ofproto->mbridge)) {
1464 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1465 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1466 mac_learning_flush(ofproto->ml);
1467 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1470 /* Do not perform any periodic activity below required by 'ofproto' while
1471 * waiting for flow restore to complete. */
1472 if (ofproto_get_flow_restore_wait()) {
1476 error = run_fast(ofproto_);
1481 if (ofproto->netflow) {
1482 if (netflow_run(ofproto->netflow)) {
1483 send_netflow_active_timeouts(ofproto);
1486 if (ofproto->sflow) {
1487 dpif_sflow_run(ofproto->sflow);
1490 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1493 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1498 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1499 if (mac_learning_run(ofproto->ml)) {
1500 ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1502 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1504 /* Check the consistency of a random facet, to aid debugging. */
1505 if (time_msec() >= ofproto->consistency_rl
1506 && !classifier_is_empty(&ofproto->facets)
1507 && !ofproto->backer->need_revalidate) {
1508 struct cls_table *table;
1509 struct cls_rule *cr;
1510 struct facet *facet;
1512 ofproto->consistency_rl = time_msec() + 250;
1514 table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1515 struct cls_table, hmap_node);
1516 cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1518 facet = CONTAINER_OF(cr, struct facet, cr);
1520 if (!facet_check_consistency(facet)) {
1521 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1529 wait(struct ofproto *ofproto_)
1531 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1532 struct ofport_dpif *ofport;
1533 struct ofbundle *bundle;
1535 if (!clogged && !list_is_empty(&ofproto->completions)) {
1536 poll_immediate_wake();
1539 if (ofproto_get_flow_restore_wait()) {
1543 dpif_wait(ofproto->backer->dpif);
1544 dpif_recv_wait(ofproto->backer->dpif);
1545 if (ofproto->sflow) {
1546 dpif_sflow_wait(ofproto->sflow);
1548 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1551 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1552 bundle_wait(bundle);
1554 if (ofproto->netflow) {
1555 netflow_wait(ofproto->netflow);
1557 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1558 mac_learning_wait(ofproto->ml);
1559 ovs_rwlock_unlock(&ofproto->ml->rwlock);
1561 if (ofproto->backer->need_revalidate) {
1562 /* Shouldn't happen, but if it does just go around again. */
1563 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1564 poll_immediate_wake();
1569 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1571 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1572 struct cls_cursor cursor;
1573 size_t n_subfacets = 0;
1574 struct facet *facet;
1576 simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1578 cls_cursor_init(&cursor, &ofproto->facets, NULL);
1579 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1580 n_subfacets += list_size(&facet->subfacets);
1582 simap_increase(usage, "subfacets", n_subfacets);
1586 flush(struct ofproto *ofproto_)
1588 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1589 struct subfacet *subfacet, *next_subfacet;
1590 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1594 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1595 &ofproto->backer->subfacets) {
1596 if (subfacet->facet->ofproto != ofproto) {
1600 if (subfacet->path != SF_NOT_INSTALLED) {
1601 batch[n_batch++] = subfacet;
1602 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1603 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1607 subfacet_destroy(subfacet);
1612 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1617 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1618 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1620 *arp_match_ip = true;
1621 *actions = (OFPUTIL_A_OUTPUT |
1622 OFPUTIL_A_SET_VLAN_VID |
1623 OFPUTIL_A_SET_VLAN_PCP |
1624 OFPUTIL_A_STRIP_VLAN |
1625 OFPUTIL_A_SET_DL_SRC |
1626 OFPUTIL_A_SET_DL_DST |
1627 OFPUTIL_A_SET_NW_SRC |
1628 OFPUTIL_A_SET_NW_DST |
1629 OFPUTIL_A_SET_NW_TOS |
1630 OFPUTIL_A_SET_TP_SRC |
1631 OFPUTIL_A_SET_TP_DST |
1636 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1639 struct dpif_dp_stats s;
1640 uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1643 strcpy(ots->name, "classifier");
1645 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1646 rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1647 rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1648 rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1650 n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1651 ots->lookup_count = htonll(n_lookup);
1652 ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1655 static struct ofport *
1658 struct ofport_dpif *port = xmalloc(sizeof *port);
1663 port_dealloc(struct ofport *port_)
1665 struct ofport_dpif *port = ofport_dpif_cast(port_);
1670 port_construct(struct ofport *port_)
1672 struct ofport_dpif *port = ofport_dpif_cast(port_);
1673 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1674 const struct netdev *netdev = port->up.netdev;
1675 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1676 struct dpif_port dpif_port;
1679 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1680 port->bundle = NULL;
1683 port->may_enable = true;
1684 port->stp_port = NULL;
1685 port->stp_state = STP_DISABLED;
1686 port->is_tunnel = false;
1688 hmap_init(&port->priorities);
1689 port->realdev_ofp_port = 0;
1690 port->vlandev_vid = 0;
1691 port->carrier_seq = netdev_get_carrier_resets(netdev);
1693 if (netdev_vport_is_patch(netdev)) {
1694 /* By bailing out here, we don't submit the port to the sFlow module
1695 * to be considered for counter polling export. This is correct
1696 * because the patch port represents an interface that sFlow considers
1697 * to be "internal" to the switch as a whole, and therefore not an
1698 * candidate for counter polling. */
1699 port->odp_port = ODPP_NONE;
1700 ofport_update_peer(port);
1704 error = dpif_port_query_by_name(ofproto->backer->dpif,
1705 netdev_vport_get_dpif_port(netdev, namebuf,
1712 port->odp_port = dpif_port.port_no;
1714 if (netdev_get_tunnel_config(netdev)) {
1715 tnl_port_add(port, port->up.netdev, port->odp_port);
1716 port->is_tunnel = true;
1718 /* Sanity-check that a mapping doesn't already exist. This
1719 * shouldn't happen for non-tunnel ports. */
1720 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1721 VLOG_ERR("port %s already has an OpenFlow port number",
1723 dpif_port_destroy(&dpif_port);
1727 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1728 hash_odp_port(port->odp_port));
1730 dpif_port_destroy(&dpif_port);
1732 if (ofproto->sflow) {
1733 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1740 port_destruct(struct ofport *port_)
1742 struct ofport_dpif *port = ofport_dpif_cast(port_);
1743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1744 const char *devname = netdev_get_name(port->up.netdev);
1745 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1746 const char *dp_port_name;
1748 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1749 xlate_ofport_remove(port);
1751 dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1753 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1754 /* The underlying device is still there, so delete it. This
1755 * happens when the ofproto is being destroyed, since the caller
1756 * assumes that removal of attached ports will happen as part of
1758 if (!port->is_tunnel) {
1759 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1764 port->peer->peer = NULL;
1768 if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1769 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1773 sset_find_and_delete(&ofproto->ports, devname);
1774 sset_find_and_delete(&ofproto->ghost_ports, devname);
1775 bundle_remove(port_);
1776 set_cfm(port_, NULL);
1777 set_bfd(port_, NULL);
1778 if (ofproto->sflow) {
1779 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1782 ofport_clear_priorities(port);
1783 hmap_destroy(&port->priorities);
1787 port_modified(struct ofport *port_)
1789 struct ofport_dpif *port = ofport_dpif_cast(port_);
1791 if (port->bundle && port->bundle->bond) {
1792 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1796 cfm_set_netdev(port->cfm, port->up.netdev);
1799 if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1801 ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1805 ofport_update_peer(port);
1809 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1811 struct ofport_dpif *port = ofport_dpif_cast(port_);
1812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1813 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1815 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1816 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1817 OFPUTIL_PC_NO_PACKET_IN)) {
1818 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1820 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1821 bundle_update(port->bundle);
1827 set_sflow(struct ofproto *ofproto_,
1828 const struct ofproto_sflow_options *sflow_options)
1830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1831 struct dpif_sflow *ds = ofproto->sflow;
1833 if (sflow_options) {
1835 struct ofport_dpif *ofport;
1837 ds = ofproto->sflow = dpif_sflow_create();
1838 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1839 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1841 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1843 dpif_sflow_set_options(ds, sflow_options);
1846 dpif_sflow_unref(ds);
1847 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1848 ofproto->sflow = NULL;
1856 struct ofproto *ofproto_,
1857 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1858 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1859 size_t n_flow_exporters_options)
1861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1862 struct dpif_ipfix *di = ofproto->ipfix;
1864 if (bridge_exporter_options || flow_exporters_options) {
1866 di = ofproto->ipfix = dpif_ipfix_create();
1868 dpif_ipfix_set_options(
1869 di, bridge_exporter_options, flow_exporters_options,
1870 n_flow_exporters_options);
1873 dpif_ipfix_unref(di);
1874 ofproto->ipfix = NULL;
1881 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1883 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1890 struct ofproto_dpif *ofproto;
1892 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1893 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894 ofport->cfm = cfm_create(ofport->up.netdev);
1897 if (cfm_configure(ofport->cfm, s)) {
1903 cfm_unref(ofport->cfm);
1909 get_cfm_status(const struct ofport *ofport_,
1910 struct ofproto_cfm_status *status)
1912 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1915 status->faults = cfm_get_fault(ofport->cfm);
1916 status->remote_opstate = cfm_get_opup(ofport->cfm);
1917 status->health = cfm_get_health(ofport->cfm);
1918 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1926 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1929 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1933 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
1934 if (ofport->bfd != old) {
1935 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1942 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1944 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1947 bfd_get_status(ofport->bfd, smap);
1954 /* Spanning Tree. */
1957 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1959 struct ofproto_dpif *ofproto = ofproto_;
1960 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1961 struct ofport_dpif *ofport;
1963 ofport = stp_port_get_aux(sp);
1965 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1966 ofproto->up.name, port_num);
1968 struct eth_header *eth = pkt->l2;
1970 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1971 if (eth_addr_is_zero(eth->eth_src)) {
1972 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1973 "with unknown MAC", ofproto->up.name, port_num);
1975 send_packet(ofport, pkt);
1981 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1983 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1987 /* Only revalidate flows if the configuration changed. */
1988 if (!s != !ofproto->stp) {
1989 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1993 if (!ofproto->stp) {
1994 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1995 send_bpdu_cb, ofproto);
1996 ofproto->stp_last_tick = time_msec();
1999 stp_set_bridge_id(ofproto->stp, s->system_id);
2000 stp_set_bridge_priority(ofproto->stp, s->priority);
2001 stp_set_hello_time(ofproto->stp, s->hello_time);
2002 stp_set_max_age(ofproto->stp, s->max_age);
2003 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2005 struct ofport *ofport;
2007 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2008 set_stp_port(ofport, NULL);
2011 stp_unref(ofproto->stp);
2012 ofproto->stp = NULL;
2019 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2021 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2025 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2026 s->designated_root = stp_get_designated_root(ofproto->stp);
2027 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2036 update_stp_port_state(struct ofport_dpif *ofport)
2038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2039 enum stp_state state;
2041 /* Figure out new state. */
2042 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2046 if (ofport->stp_state != state) {
2047 enum ofputil_port_state of_state;
2050 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2051 netdev_get_name(ofport->up.netdev),
2052 stp_state_name(ofport->stp_state),
2053 stp_state_name(state));
2054 if (stp_learn_in_state(ofport->stp_state)
2055 != stp_learn_in_state(state)) {
2056 /* xxx Learning action flows should also be flushed. */
2057 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2058 mac_learning_flush(ofproto->ml);
2059 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2061 fwd_change = stp_forward_in_state(ofport->stp_state)
2062 != stp_forward_in_state(state);
2064 ofproto->backer->need_revalidate = REV_STP;
2065 ofport->stp_state = state;
2066 ofport->stp_state_entered = time_msec();
2068 if (fwd_change && ofport->bundle) {
2069 bundle_update(ofport->bundle);
2072 /* Update the STP state bits in the OpenFlow port description. */
2073 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2074 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2075 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2076 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2077 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2079 ofproto_port_set_state(&ofport->up, of_state);
2083 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2084 * caller is responsible for assigning STP port numbers and ensuring
2085 * there are no duplicates. */
2087 set_stp_port(struct ofport *ofport_,
2088 const struct ofproto_port_stp_settings *s)
2090 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2091 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2092 struct stp_port *sp = ofport->stp_port;
2094 if (!s || !s->enable) {
2096 ofport->stp_port = NULL;
2097 stp_port_disable(sp);
2098 update_stp_port_state(ofport);
2101 } else if (sp && stp_port_no(sp) != s->port_num
2102 && ofport == stp_port_get_aux(sp)) {
2103 /* The port-id changed, so disable the old one if it's not
2104 * already in use by another port. */
2105 stp_port_disable(sp);
2108 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2109 stp_port_enable(sp);
2111 stp_port_set_aux(sp, ofport);
2112 stp_port_set_priority(sp, s->priority);
2113 stp_port_set_path_cost(sp, s->path_cost);
2115 update_stp_port_state(ofport);
2121 get_stp_port_status(struct ofport *ofport_,
2122 struct ofproto_port_stp_status *s)
2124 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2125 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2126 struct stp_port *sp = ofport->stp_port;
2128 if (!ofproto->stp || !sp) {
2134 s->port_id = stp_port_get_id(sp);
2135 s->state = stp_port_get_state(sp);
2136 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2137 s->role = stp_port_get_role(sp);
2138 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2144 stp_run(struct ofproto_dpif *ofproto)
2147 long long int now = time_msec();
2148 long long int elapsed = now - ofproto->stp_last_tick;
2149 struct stp_port *sp;
2152 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2153 ofproto->stp_last_tick = now;
2155 while (stp_get_changed_port(ofproto->stp, &sp)) {
2156 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2159 update_stp_port_state(ofport);
2163 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2164 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2165 mac_learning_flush(ofproto->ml);
2166 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2172 stp_wait(struct ofproto_dpif *ofproto)
2175 poll_timer_wait(1000);
2179 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
2180 * were used to make the determination.*/
2182 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
2184 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2185 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2189 stp_process_packet(const struct ofport_dpif *ofport,
2190 const struct ofpbuf *packet)
2192 struct ofpbuf payload = *packet;
2193 struct eth_header *eth = payload.data;
2194 struct stp_port *sp = ofport->stp_port;
2196 /* Sink packets on ports that have STP disabled when the bridge has
2198 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2202 /* Trim off padding on payload. */
2203 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2204 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2207 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2208 stp_received_bpdu(sp, payload.data, payload.size);
2213 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
2214 uint32_t queue_id, uint32_t *priority)
2216 return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
2219 static struct priority_to_dscp *
2220 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2222 struct priority_to_dscp *pdscp;
2225 hash = hash_int(priority, 0);
2226 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2227 if (pdscp->priority == priority) {
2235 ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
2236 uint32_t priority, uint8_t *dscp)
2238 struct priority_to_dscp *pdscp = get_priority(ofport, priority);
2239 *dscp = pdscp ? pdscp->dscp : 0;
2240 return pdscp != NULL;
2244 ofport_clear_priorities(struct ofport_dpif *ofport)
2246 struct priority_to_dscp *pdscp, *next;
2248 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2249 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2255 set_queues(struct ofport *ofport_,
2256 const struct ofproto_port_queue *qdscp_list,
2259 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2260 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2261 struct hmap new = HMAP_INITIALIZER(&new);
2264 for (i = 0; i < n_qdscp; i++) {
2265 struct priority_to_dscp *pdscp;
2269 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2270 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2275 pdscp = get_priority(ofport, priority);
2277 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2279 pdscp = xmalloc(sizeof *pdscp);
2280 pdscp->priority = priority;
2282 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2285 if (pdscp->dscp != dscp) {
2287 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2290 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2293 if (!hmap_is_empty(&ofport->priorities)) {
2294 ofport_clear_priorities(ofport);
2295 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2298 hmap_swap(&new, &ofport->priorities);
2306 /* Expires all MAC learning entries associated with 'bundle' and forces its
2307 * ofproto to revalidate every flow.
2309 * Normally MAC learning entries are removed only from the ofproto associated
2310 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2311 * are removed from every ofproto. When patch ports and SLB bonds are in use
2312 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2313 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2314 * with the host from which it migrated. */
2316 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2318 struct ofproto_dpif *ofproto = bundle->ofproto;
2319 struct mac_learning *ml = ofproto->ml;
2320 struct mac_entry *mac, *next_mac;
2322 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2323 ovs_rwlock_wrlock(&ml->rwlock);
2324 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2325 if (mac->port.p == bundle) {
2327 struct ofproto_dpif *o;
2329 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2331 struct mac_entry *e;
2333 ovs_rwlock_wrlock(&o->ml->rwlock);
2334 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2336 mac_learning_expire(o->ml, e);
2338 ovs_rwlock_unlock(&o->ml->rwlock);
2343 mac_learning_expire(ml, mac);
2346 ovs_rwlock_unlock(&ml->rwlock);
2349 static struct ofbundle *
2350 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2352 struct ofbundle *bundle;
2354 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2355 &ofproto->bundles) {
2356 if (bundle->aux == aux) {
2364 bundle_update(struct ofbundle *bundle)
2366 struct ofport_dpif *port;
2368 bundle->floodable = true;
2369 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2370 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2371 || !stp_forward_in_state(port->stp_state)) {
2372 bundle->floodable = false;
2379 bundle_del_port(struct ofport_dpif *port)
2381 struct ofbundle *bundle = port->bundle;
2383 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2385 list_remove(&port->bundle_node);
2386 port->bundle = NULL;
2389 lacp_slave_unregister(bundle->lacp, port);
2392 bond_slave_unregister(bundle->bond, port);
2395 bundle_update(bundle);
2399 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2400 struct lacp_slave_settings *lacp)
2402 struct ofport_dpif *port;
2404 port = get_ofp_port(bundle->ofproto, ofp_port);
2409 if (port->bundle != bundle) {
2410 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2412 bundle_del_port(port);
2415 port->bundle = bundle;
2416 list_push_back(&bundle->ports, &port->bundle_node);
2417 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2418 || !stp_forward_in_state(port->stp_state)) {
2419 bundle->floodable = false;
2423 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2424 lacp_slave_register(bundle->lacp, port, lacp);
2431 bundle_destroy(struct ofbundle *bundle)
2433 struct ofproto_dpif *ofproto;
2434 struct ofport_dpif *port, *next_port;
2440 ofproto = bundle->ofproto;
2441 mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2443 xlate_bundle_remove(bundle);
2445 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2446 bundle_del_port(port);
2449 bundle_flush_macs(bundle, true);
2450 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2452 free(bundle->trunks);
2453 lacp_unref(bundle->lacp);
2454 bond_unref(bundle->bond);
2459 bundle_set(struct ofproto *ofproto_, void *aux,
2460 const struct ofproto_bundle_settings *s)
2462 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2463 bool need_flush = false;
2464 struct ofport_dpif *port;
2465 struct ofbundle *bundle;
2466 unsigned long *trunks;
2472 bundle_destroy(bundle_lookup(ofproto, aux));
2476 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2477 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2479 bundle = bundle_lookup(ofproto, aux);
2481 bundle = xmalloc(sizeof *bundle);
2483 bundle->ofproto = ofproto;
2484 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2485 hash_pointer(aux, 0));
2487 bundle->name = NULL;
2489 list_init(&bundle->ports);
2490 bundle->vlan_mode = PORT_VLAN_TRUNK;
2492 bundle->trunks = NULL;
2493 bundle->use_priority_tags = s->use_priority_tags;
2494 bundle->lacp = NULL;
2495 bundle->bond = NULL;
2497 bundle->floodable = true;
2498 mbridge_register_bundle(ofproto->mbridge, bundle);
2501 if (!bundle->name || strcmp(s->name, bundle->name)) {
2503 bundle->name = xstrdup(s->name);
2508 if (!bundle->lacp) {
2509 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2510 bundle->lacp = lacp_create();
2512 lacp_configure(bundle->lacp, s->lacp);
2514 lacp_unref(bundle->lacp);
2515 bundle->lacp = NULL;
2518 /* Update set of ports. */
2520 for (i = 0; i < s->n_slaves; i++) {
2521 if (!bundle_add_port(bundle, s->slaves[i],
2522 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2526 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2527 struct ofport_dpif *next_port;
2529 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2530 for (i = 0; i < s->n_slaves; i++) {
2531 if (s->slaves[i] == port->up.ofp_port) {
2536 bundle_del_port(port);
2540 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2542 if (list_is_empty(&bundle->ports)) {
2543 bundle_destroy(bundle);
2547 /* Set VLAN tagging mode */
2548 if (s->vlan_mode != bundle->vlan_mode
2549 || s->use_priority_tags != bundle->use_priority_tags) {
2550 bundle->vlan_mode = s->vlan_mode;
2551 bundle->use_priority_tags = s->use_priority_tags;
2556 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2557 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2559 if (vlan != bundle->vlan) {
2560 bundle->vlan = vlan;
2564 /* Get trunked VLANs. */
2565 switch (s->vlan_mode) {
2566 case PORT_VLAN_ACCESS:
2570 case PORT_VLAN_TRUNK:
2571 trunks = CONST_CAST(unsigned long *, s->trunks);
2574 case PORT_VLAN_NATIVE_UNTAGGED:
2575 case PORT_VLAN_NATIVE_TAGGED:
2576 if (vlan != 0 && (!s->trunks
2577 || !bitmap_is_set(s->trunks, vlan)
2578 || bitmap_is_set(s->trunks, 0))) {
2579 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2581 trunks = bitmap_clone(s->trunks, 4096);
2583 trunks = bitmap_allocate1(4096);
2585 bitmap_set1(trunks, vlan);
2586 bitmap_set0(trunks, 0);
2588 trunks = CONST_CAST(unsigned long *, s->trunks);
2595 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2596 free(bundle->trunks);
2597 if (trunks == s->trunks) {
2598 bundle->trunks = vlan_bitmap_clone(trunks);
2600 bundle->trunks = trunks;
2605 if (trunks != s->trunks) {
2610 if (!list_is_short(&bundle->ports)) {
2611 bundle->ofproto->has_bonded_bundles = true;
2613 if (bond_reconfigure(bundle->bond, s->bond)) {
2614 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2617 bundle->bond = bond_create(s->bond);
2618 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2621 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2622 bond_slave_register(bundle->bond, port, port->up.netdev);
2625 bond_unref(bundle->bond);
2626 bundle->bond = NULL;
2629 /* If we changed something that would affect MAC learning, un-learn
2630 * everything on this port and force flow revalidation. */
2632 bundle_flush_macs(bundle, false);
2639 bundle_remove(struct ofport *port_)
2641 struct ofport_dpif *port = ofport_dpif_cast(port_);
2642 struct ofbundle *bundle = port->bundle;
2645 bundle_del_port(port);
2646 if (list_is_empty(&bundle->ports)) {
2647 bundle_destroy(bundle);
2648 } else if (list_is_short(&bundle->ports)) {
2649 bond_unref(bundle->bond);
2650 bundle->bond = NULL;
2656 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2658 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2659 struct ofport_dpif *port = port_;
2660 uint8_t ea[ETH_ADDR_LEN];
2663 error = netdev_get_etheraddr(port->up.netdev, ea);
2665 struct ofpbuf packet;
2668 ofpbuf_init(&packet, 0);
2669 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2671 memcpy(packet_pdu, pdu, pdu_size);
2673 send_packet(port, &packet);
2674 ofpbuf_uninit(&packet);
2676 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2677 "%s (%s)", port->bundle->name,
2678 netdev_get_name(port->up.netdev), ovs_strerror(error));
2683 bundle_send_learning_packets(struct ofbundle *bundle)
2685 struct ofproto_dpif *ofproto = bundle->ofproto;
2686 int error, n_packets, n_errors;
2687 struct mac_entry *e;
2689 error = n_packets = n_errors = 0;
2690 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2691 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2692 if (e->port.p != bundle) {
2693 struct ofpbuf *learning_packet;
2694 struct ofport_dpif *port;
2698 /* The assignment to "port" is unnecessary but makes "grep"ing for
2699 * struct ofport_dpif more effective. */
2700 learning_packet = bond_compose_learning_packet(bundle->bond,
2704 ret = send_packet(port, learning_packet);
2705 ofpbuf_delete(learning_packet);
2713 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2716 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2717 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2718 "packets, last error was: %s",
2719 bundle->name, n_errors, n_packets, ovs_strerror(error));
2721 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2722 bundle->name, n_packets);
2727 bundle_run(struct ofbundle *bundle)
2730 lacp_run(bundle->lacp, send_pdu_cb);
2733 struct ofport_dpif *port;
2735 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2736 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2739 if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2740 bundle->ofproto->backer->need_revalidate = REV_BOND;
2743 if (bond_should_send_learning_packets(bundle->bond)) {
2744 bundle_send_learning_packets(bundle);
2750 bundle_wait(struct ofbundle *bundle)
2753 lacp_wait(bundle->lacp);
2756 bond_wait(bundle->bond);
2763 mirror_set__(struct ofproto *ofproto_, void *aux,
2764 const struct ofproto_mirror_settings *s)
2766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2767 struct ofbundle **srcs, **dsts;
2772 mirror_destroy(ofproto->mbridge, aux);
2776 srcs = xmalloc(s->n_srcs * sizeof *srcs);
2777 dsts = xmalloc(s->n_dsts * sizeof *dsts);
2779 for (i = 0; i < s->n_srcs; i++) {
2780 srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2783 for (i = 0; i < s->n_dsts; i++) {
2784 dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2787 error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2788 s->n_dsts, s->src_vlans,
2789 bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2796 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2797 uint64_t *packets, uint64_t *bytes)
2800 return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2805 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2808 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2809 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2810 mac_learning_flush(ofproto->ml);
2812 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2817 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2820 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2821 return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2825 forward_bpdu_changed(struct ofproto *ofproto_)
2827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2828 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2832 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2836 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2837 mac_learning_set_idle_time(ofproto->ml, idle_time);
2838 mac_learning_set_max_entries(ofproto->ml, max_entries);
2839 ovs_rwlock_unlock(&ofproto->ml->rwlock);
2844 static struct ofport_dpif *
2845 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2847 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2848 return ofport ? ofport_dpif_cast(ofport) : NULL;
2851 static struct ofport_dpif *
2852 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2854 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2855 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2859 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2860 struct ofproto_port *ofproto_port,
2861 struct dpif_port *dpif_port)
2863 ofproto_port->name = dpif_port->name;
2864 ofproto_port->type = dpif_port->type;
2865 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2869 ofport_update_peer(struct ofport_dpif *ofport)
2871 const struct ofproto_dpif *ofproto;
2872 struct dpif_backer *backer;
2873 const char *peer_name;
2875 if (!netdev_vport_is_patch(ofport->up.netdev)) {
2879 backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2880 backer->need_revalidate = REV_RECONFIGURE;
2883 ofport->peer->peer = NULL;
2884 ofport->peer = NULL;
2887 peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2892 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2893 struct ofport *peer_ofport;
2894 struct ofport_dpif *peer;
2895 const char *peer_peer;
2897 if (ofproto->backer != backer) {
2901 peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2906 peer = ofport_dpif_cast(peer_ofport);
2907 peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2908 if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2910 ofport->peer = peer;
2911 ofport->peer->peer = ofport;
2919 port_run_fast(struct ofport_dpif *ofport)
2921 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2922 struct ofpbuf packet;
2924 ofpbuf_init(&packet, 0);
2925 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2926 send_packet(ofport, &packet);
2927 ofpbuf_uninit(&packet);
2930 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2931 struct ofpbuf packet;
2933 ofpbuf_init(&packet, 0);
2934 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2935 send_packet(ofport, &packet);
2936 ofpbuf_uninit(&packet);
2941 port_run(struct ofport_dpif *ofport)
2943 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2944 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2945 bool enable = netdev_get_carrier(ofport->up.netdev);
2947 ofport->carrier_seq = carrier_seq;
2949 port_run_fast(ofport);
2952 int cfm_opup = cfm_get_opup(ofport->cfm);
2954 cfm_run(ofport->cfm);
2955 enable = enable && !cfm_get_fault(ofport->cfm);
2957 if (cfm_opup >= 0) {
2958 enable = enable && cfm_opup;
2963 bfd_run(ofport->bfd);
2964 enable = enable && bfd_forwarding(ofport->bfd);
2967 if (ofport->bundle) {
2968 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2969 if (carrier_changed) {
2970 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2974 if (ofport->may_enable != enable) {
2975 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2976 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2979 ofport->may_enable = enable;
2983 port_wait(struct ofport_dpif *ofport)
2986 cfm_wait(ofport->cfm);
2990 bfd_wait(ofport->bfd);
2995 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2996 struct ofproto_port *ofproto_port)
2998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2999 struct dpif_port dpif_port;
3002 if (sset_contains(&ofproto->ghost_ports, devname)) {
3003 const char *type = netdev_get_type_from_name(devname);
3005 /* We may be called before ofproto->up.port_by_name is populated with
3006 * the appropriate ofport. For this reason, we must get the name and
3007 * type from the netdev layer directly. */
3009 const struct ofport *ofport;
3011 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3012 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3013 ofproto_port->name = xstrdup(devname);
3014 ofproto_port->type = xstrdup(type);
3020 if (!sset_contains(&ofproto->ports, devname)) {
3023 error = dpif_port_query_by_name(ofproto->backer->dpif,
3024 devname, &dpif_port);
3026 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3032 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3035 const char *devname = netdev_get_name(netdev);
3036 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3037 const char *dp_port_name;
3039 if (netdev_vport_is_patch(netdev)) {
3040 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3044 dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3045 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3046 odp_port_t port_no = ODPP_NONE;
3049 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3053 if (netdev_get_tunnel_config(netdev)) {
3054 simap_put(&ofproto->backer->tnl_backers,
3055 dp_port_name, odp_to_u32(port_no));
3059 if (netdev_get_tunnel_config(netdev)) {
3060 sset_add(&ofproto->ghost_ports, devname);
3062 sset_add(&ofproto->ports, devname);
3068 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3071 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3078 sset_find_and_delete(&ofproto->ghost_ports,
3079 netdev_get_name(ofport->up.netdev));
3080 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3081 if (!ofport->is_tunnel) {
3082 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3084 /* The caller is going to close ofport->up.netdev. If this is a
3085 * bonded port, then the bond is using that netdev, so remove it
3086 * from the bond. The client will need to reconfigure everything
3087 * after deleting ports, so then the slave will get re-added. */
3088 bundle_remove(&ofport->up);
3095 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3097 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3102 error = netdev_get_stats(ofport->up.netdev, stats);
3104 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3107 /* ofproto->stats.tx_packets represents packets that we created
3108 * internally and sent to some port (e.g. packets sent with
3109 * send_packet()). Account for them as if they had come from
3110 * OFPP_LOCAL and got forwarded. */
3112 if (stats->rx_packets != UINT64_MAX) {
3113 stats->rx_packets += ofproto->stats.tx_packets;
3116 if (stats->rx_bytes != UINT64_MAX) {
3117 stats->rx_bytes += ofproto->stats.tx_bytes;
3120 /* ofproto->stats.rx_packets represents packets that were received on
3121 * some port and we processed internally and dropped (e.g. STP).
3122 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3124 if (stats->tx_packets != UINT64_MAX) {
3125 stats->tx_packets += ofproto->stats.rx_packets;
3128 if (stats->tx_bytes != UINT64_MAX) {
3129 stats->tx_bytes += ofproto->stats.rx_bytes;
3136 struct port_dump_state {
3141 struct ofproto_port port;
3146 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3148 *statep = xzalloc(sizeof(struct port_dump_state));
3153 port_dump_next(const struct ofproto *ofproto_, void *state_,
3154 struct ofproto_port *port)
3156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3157 struct port_dump_state *state = state_;
3158 const struct sset *sset;
3159 struct sset_node *node;
3161 if (state->has_port) {
3162 ofproto_port_destroy(&state->port);
3163 state->has_port = false;
3165 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3166 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3169 error = port_query_by_name(ofproto_, node->name, &state->port);
3171 *port = state->port;
3172 state->has_port = true;
3174 } else if (error != ENODEV) {
3179 if (!state->ghost) {
3180 state->ghost = true;
3183 return port_dump_next(ofproto_, state_, port);
3190 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3192 struct port_dump_state *state = state_;
3194 if (state->has_port) {
3195 ofproto_port_destroy(&state->port);
3202 port_poll(const struct ofproto *ofproto_, char **devnamep)
3204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3206 if (ofproto->port_poll_errno) {
3207 int error = ofproto->port_poll_errno;
3208 ofproto->port_poll_errno = 0;
3212 if (sset_is_empty(&ofproto->port_poll_set)) {
3216 *devnamep = sset_pop(&ofproto->port_poll_set);
3221 port_poll_wait(const struct ofproto *ofproto_)
3223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3224 dpif_port_poll_wait(ofproto->backer->dpif);
3228 port_is_lacp_current(const struct ofport *ofport_)
3230 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3231 return (ofport->bundle && ofport->bundle->lacp
3232 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3236 /* Upcall handling. */
3238 /* Flow miss batching.
3240 * Some dpifs implement operations faster when you hand them off in a batch.
3241 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3242 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3243 * more packets, plus possibly installing the flow in the dpif.
3245 * So far we only batch the operations that affect flow setup time the most.
3246 * It's possible to batch more than that, but the benefit might be minimal. */
3248 struct hmap_node hmap_node;
3249 struct ofproto_dpif *ofproto;
3251 enum odp_key_fitness key_fitness;
3252 const struct nlattr *key;
3254 struct list packets;
3255 enum dpif_upcall_type upcall_type;
3258 struct flow_miss_op {
3259 struct dpif_op dpif_op;
3261 uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3262 struct xlate_out xout;
3263 bool xout_garbage; /* 'xout' needs to be uninitialized? */
3265 struct ofpbuf mask; /* Flow mask for "put" ops. */
3266 struct odputil_keybuf maskbuf;
3268 /* If this is a "put" op, then a pointer to the subfacet that should
3269 * be marked as uninstalled if the operation fails. */
3270 struct subfacet *subfacet;
3273 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3274 * OpenFlow controller as necessary according to their individual
3275 * configurations. */
3277 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3278 const struct flow *flow)
3280 struct ofputil_packet_in pin;
3282 pin.packet = packet->data;
3283 pin.packet_len = packet->size;
3284 pin.reason = OFPR_NO_MATCH;
3285 pin.controller_id = 0;
3290 pin.send_len = 0; /* not used for flow table misses */
3292 flow_get_metadata(flow, &pin.fmd);
3294 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3297 static struct flow_miss *
3298 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3299 const struct flow *flow, uint32_t hash)
3301 struct flow_miss *miss;
3303 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3304 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3312 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3313 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3314 * 'miss' is associated with a subfacet the caller must also initialize the
3315 * returned op->subfacet, and if anything needs to be freed after processing
3316 * the op, the caller must initialize op->garbage also. */
3318 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3319 struct flow_miss_op *op)
3321 if (miss->flow.in_port.ofp_port
3322 != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port,
3323 miss->flow.vlan_tci)) {
3324 /* This packet was received on a VLAN splinter port. We
3325 * added a VLAN to the packet to make the packet resemble
3326 * the flow, but the actions were composed assuming that
3327 * the packet contained no VLAN. So, we must remove the
3328 * VLAN header from the packet before trying to execute the
3330 eth_pop_vlan(packet);
3333 op->subfacet = NULL;
3334 op->xout_garbage = false;
3335 op->dpif_op.type = DPIF_OP_EXECUTE;
3336 op->dpif_op.u.execute.key = miss->key;
3337 op->dpif_op.u.execute.key_len = miss->key_len;
3338 op->dpif_op.u.execute.packet = packet;
3339 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3342 /* Helper for handle_flow_miss_without_facet() and
3343 * handle_flow_miss_with_facet(). */
3345 handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
3346 const struct flow *flow, bool fail_open)
3350 * Extra-special case for fail-open mode.
3352 * We are in fail-open mode and the packet matched the fail-open
3353 * rule, but we are connected to a controller too. We should send
3354 * the packet up to the controller in the hope that it will try to
3355 * set up a flow and thereby allow us to exit fail-open.
3357 * See the top-level comment in fail-open.c for more information.
3359 send_packet_in_miss(ofproto, packet, flow);
3363 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3364 * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3365 * and (usually) installing a datapath flow. The answer is usually "yes" (a
3366 * return value of true). However, for short flows the cost of bookkeeping is
3367 * much higher than the benefits, so when the datapath holds a large number of
3368 * flows we impose some heuristics to decide which flows are likely to be worth
3371 flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc)
3373 struct dpif_backer *backer = miss->ofproto->backer;
3376 switch (flow_miss_model) {
3377 case OFPROTO_HANDLE_MISS_AUTO:
3379 case OFPROTO_HANDLE_MISS_WITH_FACETS:
3381 case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3385 if (!backer->governor) {
3388 n_subfacets = hmap_count(&backer->subfacets);
3389 if (n_subfacets * 2 <= flow_eviction_threshold) {
3393 backer->governor = governor_create();
3396 hash = flow_hash_in_wildcards(&miss->flow, wc, 0);
3397 return governor_should_install_flow(backer->governor, hash,
3398 list_size(&miss->packets));
3401 /* Handles 'miss' without creating a facet or subfacet or creating any datapath
3402 * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'.
3403 * May add an "execute" operation to 'ops' and increment '*n_ops'. */
3405 handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout,
3406 struct flow_miss *miss,
3407 struct flow_miss_op *ops, size_t *n_ops)
3409 struct ofpbuf *packet;
3411 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3413 COVERAGE_INC(facet_suppress);
3415 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3416 rule->up.cr.priority == FAIL_OPEN_PRIORITY);
3419 struct xlate_in xin;
3421 xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet);
3422 xlate_actions_for_side_effects(&xin);
3425 if (xout->odp_actions.size) {
3426 struct flow_miss_op *op = &ops[*n_ops];
3427 struct dpif_execute *execute = &op->dpif_op.u.execute;
3429 init_flow_miss_execute_op(miss, packet, op);
3430 xlate_out_copy(&op->xout, xout);
3431 execute->actions = op->xout.odp_actions.data;
3432 execute->actions_len = op->xout.odp_actions.size;
3433 op->xout_garbage = true;
3440 /* Handles 'miss', which matches 'facet'. May add any required datapath
3441 * operations to 'ops', incrementing '*n_ops' for each new op.
3443 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3444 * This is really important only for new facets: if we just called time_msec()
3445 * here, then the new subfacet or its packets could look (occasionally) as
3446 * though it was used some time after the facet was used. That can make a
3447 * one-packet flow look like it has a nonzero duration, which looks odd in
3448 * e.g. NetFlow statistics.
3450 * If non-null, 'stats' will be folded into 'facet'. */
3452 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3453 long long int now, struct dpif_flow_stats *stats,
3454 struct flow_miss_op *ops, size_t *n_ops)
3456 enum subfacet_path want_path;
3457 struct subfacet *subfacet;
3458 struct ofpbuf *packet;
3460 want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3462 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3463 struct flow_miss_op *op = &ops[*n_ops];
3465 handle_flow_miss_common(miss->ofproto, packet, &miss->flow,
3468 if (want_path != SF_FAST_PATH) {
3469 struct rule_dpif *rule;
3470 struct xlate_in xin;
3472 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
3473 xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet);
3474 xlate_actions_for_side_effects(&xin);
3477 if (facet->xout.odp_actions.size) {
3478 struct dpif_execute *execute = &op->dpif_op.u.execute;
3480 init_flow_miss_execute_op(miss, packet, op);
3481 execute->actions = facet->xout.odp_actions.data,
3482 execute->actions_len = facet->xout.odp_actions.size;
3487 /* Don't install the flow if it's the result of the "userspace"
3488 * action for an already installed facet. This can occur when a
3489 * datapath flow with wildcards has a "userspace" action and flows
3490 * sent to userspace result in a different subfacet, which will then
3491 * be rejected as overlapping by the datapath. */
3492 if (miss->upcall_type == DPIF_UC_ACTION
3493 && !list_is_empty(&facet->subfacets)) {
3495 facet->used = MAX(facet->used, stats->used);
3496 facet->packet_count += stats->n_packets;
3497 facet->byte_count += stats->n_bytes;
3498 facet->tcp_flags |= stats->tcp_flags;
3503 subfacet = subfacet_create(facet, miss, now);
3505 subfacet_update_stats(subfacet, stats);
3508 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3509 struct flow_miss_op *op = &ops[(*n_ops)++];
3510 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3512 subfacet->path = want_path;
3514 ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3515 if (enable_megaflows) {
3516 odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3517 &miss->flow, UINT32_MAX);
3520 op->xout_garbage = false;
3521 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3522 op->subfacet = subfacet;
3523 put->flags = DPIF_FP_CREATE;
3524 put->key = miss->key;
3525 put->key_len = miss->key_len;
3526 put->mask = op->mask.data;
3527 put->mask_len = op->mask.size;
3529 if (want_path == SF_FAST_PATH) {
3530 put->actions = facet->xout.odp_actions.data;
3531 put->actions_len = facet->xout.odp_actions.size;
3533 compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3534 op->slow_stub, sizeof op->slow_stub,
3535 &put->actions, &put->actions_len);
3541 /* Handles flow miss 'miss'. May add any required datapath operations
3542 * to 'ops', incrementing '*n_ops' for each new op. */
3544 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3547 struct ofproto_dpif *ofproto = miss->ofproto;
3548 struct dpif_flow_stats stats__;
3549 struct dpif_flow_stats *stats = &stats__;
3550 struct ofpbuf *packet;
3551 struct facet *facet;
3555 memset(stats, 0, sizeof *stats);
3557 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3558 stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow);
3559 stats->n_bytes += packet->size;
3563 facet = facet_lookup_valid(ofproto, &miss->flow);
3565 struct flow_wildcards wc;
3566 struct rule_dpif *rule;
3567 struct xlate_out xout;
3568 struct xlate_in xin;
3570 flow_wildcards_init_catchall(&wc);
3571 rule = rule_dpif_lookup(ofproto, &miss->flow, &wc);
3572 rule_credit_stats(rule, stats);
3574 xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags,
3576 xin.resubmit_stats = stats;
3577 xin.may_learn = true;
3578 xlate_actions(&xin, &xout);
3579 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
3581 /* There does not exist a bijection between 'struct flow' and datapath
3582 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3583 * assumption used throughout the facet and subfacet handling code.
3584 * Since we have to handle these misses in userspace anyway, we simply
3585 * skip facet creation, avoiding the problem altogether. */
3586 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3587 || !flow_miss_should_make_facet(miss, &xout.wc)) {
3588 handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops);
3592 facet = facet_create(miss, rule, &xout, stats);
3595 handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops);
3598 static struct drop_key *
3599 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3602 struct drop_key *drop_key;
3604 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3605 &backer->drop_keys) {
3606 if (drop_key->key_len == key_len
3607 && !memcmp(drop_key->key, key, key_len)) {
3615 drop_key_clear(struct dpif_backer *backer)
3617 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3618 struct drop_key *drop_key, *next;
3620 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3623 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3625 if (error && !VLOG_DROP_WARN(&rl)) {
3626 struct ds ds = DS_EMPTY_INITIALIZER;
3627 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3628 VLOG_WARN("Failed to delete drop key (%s) (%s)",
3629 ovs_strerror(error), ds_cstr(&ds));
3633 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3634 free(drop_key->key);
3639 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3640 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3641 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3642 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3643 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3644 * 'packet' ingressed.
3646 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3647 * 'flow''s in_port to OFPP_NONE.
3649 * This function does post-processing on data returned from
3650 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3651 * of the upcall processing logic. In particular, if the extracted in_port is
3652 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3653 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3654 * a VLAN header onto 'packet' (if it is nonnull).
3656 * Similarly, this function also includes some logic to help with tunnels. It
3657 * may modify 'flow' as necessary to make the tunneling implementation
3658 * transparent to the upcall processing logic.
3660 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3661 * or some other positive errno if there are other problems. */
3663 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3664 const struct nlattr *key, size_t key_len,
3665 struct flow *flow, enum odp_key_fitness *fitnessp,
3666 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
3668 const struct ofport_dpif *port;
3669 enum odp_key_fitness fitness;
3672 fitness = odp_flow_key_to_flow(key, key_len, flow);
3673 if (fitness == ODP_FIT_ERROR) {
3679 *odp_in_port = flow->in_port.odp_port;
3682 port = (tnl_port_should_receive(flow)
3683 ? tnl_port_receive(flow)
3684 : odp_port_to_ofport(backer, flow->in_port.odp_port));
3685 flow->in_port.ofp_port = port ? port->up.ofp_port : OFPP_NONE;
3690 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3691 * it's theoretically possible that we'll receive an ofport belonging to an
3692 * entirely different datapath. In practice, this can't happen because no
3693 * platforms has two separate datapaths which each support tunneling. */
3694 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3696 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3698 /* Make the packet resemble the flow, so that it gets sent to
3699 * an OpenFlow controller properly, so that it looks correct
3700 * for sFlow, and so that flow_extract() will get the correct
3701 * vlan_tci if it is called on 'packet'.
3703 * The allocated space inside 'packet' probably also contains
3704 * 'key', that is, both 'packet' and 'key' are probably part of
3705 * a struct dpif_upcall (see the large comment on that
3706 * structure definition), so pushing data on 'packet' is in
3707 * general not a good idea since it could overwrite 'key' or
3708 * free it as a side effect. However, it's OK in this special
3709 * case because we know that 'packet' is inside a Netlink
3710 * attribute: pushing 4 bytes will just overwrite the 4-byte
3711 * "struct nlattr", which is fine since we don't need that
3712 * header anymore. */
3713 eth_push_vlan(packet, flow->vlan_tci);
3715 /* We can't reproduce 'key' from 'flow'. */
3716 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3721 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3726 *fitnessp = fitness;
3732 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3735 struct dpif_upcall *upcall;
3736 struct flow_miss *miss;
3737 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3738 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3739 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3749 /* Construct the to-do list.
3751 * This just amounts to extracting the flow from each packet and sticking
3752 * the packets that have the same flow in the same "flow_miss" structure so
3753 * that we can process them together. */
3756 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3757 struct flow_miss *miss = &misses[n_misses];
3758 struct flow_miss *existing_miss;
3759 struct ofproto_dpif *ofproto;
3760 odp_port_t odp_in_port;
3765 error = ofproto_receive(backer, upcall->packet, upcall->key,
3766 upcall->key_len, &flow, &miss->key_fitness,
3767 &ofproto, &odp_in_port);
3768 if (error == ENODEV) {
3769 struct drop_key *drop_key;
3771 /* Received packet on datapath port for which we couldn't
3772 * associate an ofproto. This can happen if a port is removed
3773 * while traffic is being received. Print a rate-limited message
3774 * in case it happens frequently. Install a drop flow so
3775 * that future packets of the flow are inexpensively dropped
3777 VLOG_INFO_RL(&rl, "received packet on unassociated datapath port "
3778 "%"PRIu32, odp_in_port);
3780 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3783 ret = dpif_flow_put(backer->dpif,
3784 DPIF_FP_CREATE | DPIF_FP_MODIFY,
3785 upcall->key, upcall->key_len,
3786 NULL, 0, NULL, 0, NULL);
3789 drop_key = xmalloc(sizeof *drop_key);
3790 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3791 drop_key->key_len = upcall->key_len;
3793 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3794 hash_bytes(drop_key->key, drop_key->key_len, 0));
3803 ofproto->n_missed++;
3804 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3805 &flow.tunnel, &flow.in_port, &miss->flow);
3807 /* Add other packets to a to-do list. */
3808 hash = flow_hash(&miss->flow, 0);
3809 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3810 if (!existing_miss) {
3811 hmap_insert(&todo, &miss->hmap_node, hash);
3812 miss->ofproto = ofproto;
3813 miss->key = upcall->key;
3814 miss->key_len = upcall->key_len;
3815 miss->upcall_type = upcall->type;
3816 list_init(&miss->packets);
3820 miss = existing_miss;
3822 list_push_back(&miss->packets, &upcall->packet->list_node);
3825 /* Process each element in the to-do list, constructing the set of
3826 * operations to batch. */
3828 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3829 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3831 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3833 /* Execute batch. */
3834 for (i = 0; i < n_ops; i++) {
3835 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3837 dpif_operate(backer->dpif, dpif_ops, n_ops);
3839 for (i = 0; i < n_ops; i++) {
3840 if (dpif_ops[i]->error != 0
3841 && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3842 && flow_miss_ops[i].subfacet) {
3843 struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3845 COVERAGE_INC(subfacet_install_fail);
3847 /* Zero-out subfacet counters when installation failed, but
3848 * datapath reported hits. This should not happen and
3849 * indicates a bug, since if the datapath flow exists, we
3850 * should not be attempting to create a new subfacet. A
3851 * buggy datapath could trigger this, so just zero out the
3852 * counters and log an error. */
3853 if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3854 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3855 "datapath reported hits");
3856 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3859 subfacet->path = SF_NOT_INSTALLED;
3863 if (flow_miss_ops[i].xout_garbage) {
3864 xlate_out_uninit(&flow_miss_ops[i].xout);
3867 hmap_destroy(&todo);
3870 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
3872 classify_upcall(const struct dpif_upcall *upcall)
3874 size_t userdata_len;
3875 union user_action_cookie cookie;
3877 /* First look at the upcall type. */
3878 switch (upcall->type) {
3879 case DPIF_UC_ACTION:
3885 case DPIF_N_UC_TYPES:
3887 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3891 /* "action" upcalls need a closer look. */
3892 if (!upcall->userdata) {
3893 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3896 userdata_len = nl_attr_get_size(upcall->userdata);
3897 if (userdata_len < sizeof cookie.type
3898 || userdata_len > sizeof cookie) {
3899 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3903 memset(&cookie, 0, sizeof cookie);
3904 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
3905 if (userdata_len == sizeof cookie.sflow
3906 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
3907 return SFLOW_UPCALL;
3908 } else if (userdata_len == sizeof cookie.slow_path
3909 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
3911 } else if (userdata_len == sizeof cookie.flow_sample
3912 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
3913 return FLOW_SAMPLE_UPCALL;
3914 } else if (userdata_len == sizeof cookie.ipfix
3915 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
3916 return IPFIX_UPCALL;
3918 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
3919 " and size %zu", cookie.type, userdata_len);
3925 handle_sflow_upcall(struct dpif_backer *backer,
3926 const struct dpif_upcall *upcall)
3928 struct ofproto_dpif *ofproto;
3929 union user_action_cookie cookie;
3931 odp_port_t odp_in_port;
3933 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3934 &flow, NULL, &ofproto, &odp_in_port)
3935 || !ofproto->sflow) {
3939 memset(&cookie, 0, sizeof cookie);
3940 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3941 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3942 odp_in_port, &cookie);
3946 handle_flow_sample_upcall(struct dpif_backer *backer,
3947 const struct dpif_upcall *upcall)
3949 struct ofproto_dpif *ofproto;
3950 union user_action_cookie cookie;
3953 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3954 &flow, NULL, &ofproto, NULL)
3955 || !ofproto->ipfix) {
3959 memset(&cookie, 0, sizeof cookie);
3960 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3962 /* The flow reflects exactly the contents of the packet. Sample
3963 * the packet using it. */
3964 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3965 cookie.flow_sample.collector_set_id,
3966 cookie.flow_sample.probability,
3967 cookie.flow_sample.obs_domain_id,
3968 cookie.flow_sample.obs_point_id);
3972 handle_ipfix_upcall(struct dpif_backer *backer,
3973 const struct dpif_upcall *upcall)
3975 struct ofproto_dpif *ofproto;
3978 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3979 &flow, NULL, &ofproto, NULL)
3980 || !ofproto->ipfix) {
3984 /* The flow reflects exactly the contents of the packet. Sample
3985 * the packet using it. */
3986 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3990 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3992 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3993 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3994 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3999 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4002 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4003 struct dpif_upcall *upcall = &misses[n_misses];
4004 struct ofpbuf *buf = &miss_bufs[n_misses];
4007 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4008 sizeof miss_buf_stubs[n_misses]);
4009 error = dpif_recv(backer->dpif, upcall, buf);
4015 switch (classify_upcall(upcall)) {
4017 /* Handle it later. */
4022 handle_sflow_upcall(backer, upcall);
4026 case FLOW_SAMPLE_UPCALL:
4027 handle_flow_sample_upcall(backer, upcall);
4032 handle_ipfix_upcall(backer, upcall);
4042 /* Handle deferred MISS_UPCALL processing. */
4043 handle_miss_upcalls(backer, misses, n_misses);
4044 for (i = 0; i < n_misses; i++) {
4045 ofpbuf_uninit(&miss_bufs[i]);
4051 /* Flow expiration. */
4053 static int subfacet_max_idle(const struct dpif_backer *);
4054 static void update_stats(struct dpif_backer *);
4055 static void rule_expire(struct rule_dpif *);
4056 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
4058 /* This function is called periodically by run(). Its job is to collect
4059 * updates for the flows that have been installed into the datapath, most
4060 * importantly when they last were used, and then use that information to
4061 * expire flows that have not been used recently.
4063 * Returns the number of milliseconds after which it should be called again. */
4065 expire(struct dpif_backer *backer)
4067 struct ofproto_dpif *ofproto;
4071 /* Periodically clear out the drop keys in an effort to keep them
4072 * relatively few. */
4073 drop_key_clear(backer);
4075 /* Update stats for each flow in the backer. */
4076 update_stats(backer);
4078 n_subfacets = hmap_count(&backer->subfacets);
4080 struct subfacet *subfacet;
4081 long long int total, now;
4085 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4086 total += now - subfacet->created;
4088 backer->avg_subfacet_life += total / n_subfacets;
4090 backer->avg_subfacet_life /= 2;
4092 backer->avg_n_subfacet += n_subfacets;
4093 backer->avg_n_subfacet /= 2;
4095 backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
4097 max_idle = subfacet_max_idle(backer);
4098 expire_subfacets(backer, max_idle);
4100 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4101 struct rule *rule, *next_rule;
4103 if (ofproto->backer != backer) {
4107 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4109 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4110 &ofproto->up.expirable) {
4111 rule_expire(rule_dpif_cast(rule));
4114 /* All outstanding data in existing flows has been accounted, so it's a
4115 * good time to do bond rebalancing. */
4116 if (ofproto->has_bonded_bundles) {
4117 struct ofbundle *bundle;
4119 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4121 bond_rebalance(bundle->bond);
4127 return MIN(max_idle, 1000);
4130 /* Updates flow table statistics given that the datapath just reported 'stats'
4131 * as 'subfacet''s statistics. */
4133 update_subfacet_stats(struct subfacet *subfacet,
4134 const struct dpif_flow_stats *stats)
4136 struct facet *facet = subfacet->facet;
4137 struct dpif_flow_stats diff;
4139 diff.tcp_flags = stats->tcp_flags;
4140 diff.used = stats->used;
4142 if (stats->n_packets >= subfacet->dp_packet_count) {
4143 diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
4145 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4149 if (stats->n_bytes >= subfacet->dp_byte_count) {
4150 diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
4152 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4156 facet->ofproto->n_hit += diff.n_packets;
4157 subfacet->dp_packet_count = stats->n_packets;
4158 subfacet->dp_byte_count = stats->n_bytes;
4159 subfacet_update_stats(subfacet, &diff);
4161 if (facet->accounted_bytes < facet->byte_count) {
4163 facet_account(facet);
4164 facet->accounted_bytes = facet->byte_count;
4168 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4169 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4171 delete_unexpected_flow(struct dpif_backer *backer,
4172 const struct nlattr *key, size_t key_len)
4174 if (!VLOG_DROP_WARN(&rl)) {
4178 odp_flow_key_format(key, key_len, &s);
4179 VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
4183 COVERAGE_INC(facet_unexpected);
4184 dpif_flow_del(backer->dpif, key, key_len, NULL);
4187 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4189 * This function also pushes statistics updates to rules which each facet
4190 * resubmits into. Generally these statistics will be accurate. However, if a
4191 * facet changes the rule it resubmits into at some time in between
4192 * update_stats() runs, it is possible that statistics accrued to the
4193 * old rule will be incorrectly attributed to the new rule. This could be
4194 * avoided by calling update_stats() whenever rules are created or
4195 * deleted. However, the performance impact of making so many calls to the
4196 * datapath do not justify the benefit of having perfectly accurate statistics.
4198 * In addition, this function maintains per ofproto flow hit counts. The patch
4199 * port is not treated specially. e.g. A packet ingress from br0 patched into
4200 * br1 will increase the hit count of br0 by 1, however, does not affect
4201 * the hit or miss counts of br1.
4204 update_stats(struct dpif_backer *backer)
4206 const struct dpif_flow_stats *stats;
4207 struct dpif_flow_dump dump;
4208 const struct nlattr *key, *mask;
4209 size_t key_len, mask_len;
4211 dpif_flow_dump_start(&dump, backer->dpif);
4212 while (dpif_flow_dump_next(&dump, &key, &key_len,
4213 &mask, &mask_len, NULL, NULL, &stats)) {
4214 struct subfacet *subfacet;
4217 key_hash = odp_flow_key_hash(key, key_len);
4218 subfacet = subfacet_find(backer, key, key_len, key_hash);
4219 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4221 update_subfacet_stats(subfacet, stats);
4225 /* Stats are updated per-packet. */
4228 case SF_NOT_INSTALLED:
4230 delete_unexpected_flow(backer, key, key_len);
4235 dpif_flow_dump_done(&dump);
4237 update_moving_averages(backer);
4240 /* Calculates and returns the number of milliseconds of idle time after which
4241 * subfacets should expire from the datapath. When a subfacet expires, we fold
4242 * its statistics into its facet, and when a facet's last subfacet expires, we
4243 * fold its statistic into its rule. */
4245 subfacet_max_idle(const struct dpif_backer *backer)
4248 * Idle time histogram.
4250 * Most of the time a switch has a relatively small number of subfacets.
4251 * When this is the case we might as well keep statistics for all of them
4252 * in userspace and to cache them in the kernel datapath for performance as
4255 * As the number of subfacets increases, the memory required to maintain
4256 * statistics about them in userspace and in the kernel becomes
4257 * significant. However, with a large number of subfacets it is likely
4258 * that only a few of them are "heavy hitters" that consume a large amount
4259 * of bandwidth. At this point, only heavy hitters are worth caching in
4260 * the kernel and maintaining in userspaces; other subfacets we can
4263 * The technique used to compute the idle time is to build a histogram with
4264 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4265 * that is installed in the kernel gets dropped in the appropriate bucket.
4266 * After the histogram has been built, we compute the cutoff so that only
4267 * the most-recently-used 1% of subfacets (but at least
4268 * flow_eviction_threshold flows) are kept cached. At least
4269 * the most-recently-used bucket of subfacets is kept, so actually an
4270 * arbitrary number of subfacets can be kept in any given expiration run
4271 * (though the next run will delete most of those unless they receive
4274 * This requires a second pass through the subfacets, in addition to the
4275 * pass made by update_stats(), because the former function never looks at
4276 * uninstallable subfacets.
4278 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4279 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4280 int buckets[N_BUCKETS] = { 0 };
4281 int total, subtotal, bucket;
4282 struct subfacet *subfacet;
4286 total = hmap_count(&backer->subfacets);
4287 if (total <= flow_eviction_threshold) {
4288 return N_BUCKETS * BUCKET_WIDTH;
4291 /* Build histogram. */
4293 HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
4294 long long int idle = now - subfacet->used;
4295 int bucket = (idle <= 0 ? 0
4296 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4297 : (unsigned int) idle / BUCKET_WIDTH);
4301 /* Find the first bucket whose flows should be expired. */
4302 subtotal = bucket = 0;
4304 subtotal += buckets[bucket++];
4305 } while (bucket < N_BUCKETS &&
4306 subtotal < MAX(flow_eviction_threshold, total / 100));
4308 if (VLOG_IS_DBG_ENABLED()) {
4312 ds_put_cstr(&s, "keep");
4313 for (i = 0; i < N_BUCKETS; i++) {
4315 ds_put_cstr(&s, ", drop");
4318 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4321 VLOG_INFO("%s (msec:count)", ds_cstr(&s));
4325 return bucket * BUCKET_WIDTH;
4329 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
4331 /* Cutoff time for most flows. */
4332 long long int normal_cutoff = time_msec() - dp_max_idle;
4334 /* We really want to keep flows for special protocols around, so use a more
4335 * conservative cutoff. */
4336 long long int special_cutoff = time_msec() - 10000;
4338 struct subfacet *subfacet, *next_subfacet;
4339 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4343 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4344 &backer->subfacets) {
4345 long long int cutoff;
4347 cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
4351 if (subfacet->used < cutoff) {
4352 if (subfacet->path != SF_NOT_INSTALLED) {
4353 batch[n_batch++] = subfacet;
4354 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4355 subfacet_destroy_batch(backer, batch, n_batch);
4359 subfacet_destroy(subfacet);
4365 subfacet_destroy_batch(backer, batch, n_batch);
4369 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4370 * then delete it entirely. */
4372 rule_expire(struct rule_dpif *rule)
4377 if (rule->up.pending) {
4378 /* We'll have to expire it later. */
4382 /* Has 'rule' expired? */
4384 if (rule->up.hard_timeout
4385 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4386 reason = OFPRR_HARD_TIMEOUT;
4387 } else if (rule->up.idle_timeout
4388 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4389 reason = OFPRR_IDLE_TIMEOUT;
4394 COVERAGE_INC(ofproto_dpif_expired);
4396 /* Get rid of the rule. */
4397 ofproto_rule_expire(&rule->up, reason);
4402 /* Creates and returns a new facet based on 'miss'.
4404 * The caller must already have determined that no facet with an identical
4405 * 'miss->flow' exists in 'miss->ofproto'.
4407 * 'rule' and 'xout' must have been created based on 'miss'.
4409 * 'facet'' statistics are initialized based on 'stats'.
4411 * The facet will initially have no subfacets. The caller should create (at
4412 * least) one subfacet with subfacet_create(). */
4413 static struct facet *
4414 facet_create(const struct flow_miss *miss, struct rule_dpif *rule,
4415 struct xlate_out *xout, struct dpif_flow_stats *stats)
4417 struct ofproto_dpif *ofproto = miss->ofproto;
4418 struct facet *facet;
4421 facet = xzalloc(sizeof *facet);
4422 facet->ofproto = miss->ofproto;
4423 facet->packet_count = facet->prev_packet_count = stats->n_packets;
4424 facet->byte_count = facet->prev_byte_count = stats->n_bytes;
4425 facet->tcp_flags = stats->tcp_flags;
4426 facet->used = stats->used;
4427 facet->flow = miss->flow;
4428 facet->learn_rl = time_msec() + 500;
4430 list_init(&facet->subfacets);
4431 netflow_flow_init(&facet->nf_flow);
4432 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4434 xlate_out_copy(&facet->xout, xout);
4436 match_init(&match, &facet->flow, &facet->xout.wc);
4437 cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4438 classifier_insert(&ofproto->facets, &facet->cr);
4440 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4441 facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4447 facet_free(struct facet *facet)
4450 xlate_out_uninit(&facet->xout);
4455 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4456 * 'packet', which arrived on 'in_port'. */
4458 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4459 const struct nlattr *odp_actions, size_t actions_len,
4460 struct ofpbuf *packet)
4462 struct odputil_keybuf keybuf;
4466 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4467 odp_flow_key_from_flow(&key, flow,
4468 ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4470 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4471 odp_actions, actions_len, packet);
4475 /* Remove 'facet' from its ofproto and free up the associated memory:
4477 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4478 * rule's statistics, via subfacet_uninstall().
4480 * - Removes 'facet' from its rule and from ofproto->facets.
4483 facet_remove(struct facet *facet)
4485 struct subfacet *subfacet, *next_subfacet;
4487 ovs_assert(!list_is_empty(&facet->subfacets));
4489 /* First uninstall all of the subfacets to get final statistics. */
4490 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4491 subfacet_uninstall(subfacet);
4494 /* Flush the final stats to the rule.
4496 * This might require us to have at least one subfacet around so that we
4497 * can use its actions for accounting in facet_account(), which is why we
4498 * have uninstalled but not yet destroyed the subfacets. */
4499 facet_flush_stats(facet);
4501 /* Now we're really all done so destroy everything. */
4502 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4503 &facet->subfacets) {
4504 subfacet_destroy__(subfacet);
4506 classifier_remove(&facet->ofproto->facets, &facet->cr);
4507 cls_rule_destroy(&facet->cr);
4511 /* Feed information from 'facet' back into the learning table to keep it in
4512 * sync with what is actually flowing through the datapath. */
4514 facet_learn(struct facet *facet)
4516 long long int now = time_msec();
4518 if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4522 facet->learn_rl = now + 500;
4524 if (!facet->xout.has_learn
4525 && !facet->xout.has_normal
4526 && (!facet->xout.has_fin_timeout
4527 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4531 facet_push_stats(facet, true);
4535 facet_account(struct facet *facet)
4537 const struct nlattr *a;
4542 if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4545 n_bytes = facet->byte_count - facet->accounted_bytes;
4547 /* This loop feeds byte counters to bond_account() for rebalancing to use
4548 * as a basis. We also need to track the actual VLAN on which the packet
4549 * is going to be sent to ensure that it matches the one passed to
4550 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4553 * We use the actions from an arbitrary subfacet because they should all
4554 * be equally valid for our purpose. */
4555 vlan_tci = facet->flow.vlan_tci;
4556 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4557 facet->xout.odp_actions.size) {
4558 const struct ovs_action_push_vlan *vlan;
4559 struct ofport_dpif *port;
4561 switch (nl_attr_type(a)) {
4562 case OVS_ACTION_ATTR_OUTPUT:
4563 port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4564 if (port && port->bundle && port->bundle->bond) {
4565 bond_account(port->bundle->bond, &facet->flow,
4566 vlan_tci_to_vid(vlan_tci), n_bytes);
4570 case OVS_ACTION_ATTR_POP_VLAN:
4571 vlan_tci = htons(0);
4574 case OVS_ACTION_ATTR_PUSH_VLAN:
4575 vlan = nl_attr_get(a);
4576 vlan_tci = vlan->vlan_tci;
4582 /* Returns true if the only action for 'facet' is to send to the controller.
4583 * (We don't report NetFlow expiration messages for such facets because they
4584 * are just part of the control logic for the network, not real traffic). */
4586 facet_is_controller_flow(struct facet *facet)
4589 struct ofproto_dpif *ofproto = facet->ofproto;
4590 const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow,
4592 const struct ofpact *ofpacts = rule->up.ofpacts;
4593 size_t ofpacts_len = rule->up.ofpacts_len;
4595 if (ofpacts_len > 0 &&
4596 ofpacts->type == OFPACT_CONTROLLER &&
4597 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4604 /* Folds all of 'facet''s statistics into its rule. Also updates the
4605 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4606 * 'facet''s statistics in the datapath should have been zeroed and folded into
4607 * its packet and byte counts before this function is called. */
4609 facet_flush_stats(struct facet *facet)
4611 struct ofproto_dpif *ofproto = facet->ofproto;
4612 struct subfacet *subfacet;
4614 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4615 ovs_assert(!subfacet->dp_byte_count);
4616 ovs_assert(!subfacet->dp_packet_count);
4619 facet_push_stats(facet, false);
4620 if (facet->accounted_bytes < facet->byte_count) {
4621 facet_account(facet);
4622 facet->accounted_bytes = facet->byte_count;
4625 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4626 struct ofexpired expired;
4627 expired.flow = facet->flow;
4628 expired.packet_count = facet->packet_count;
4629 expired.byte_count = facet->byte_count;
4630 expired.used = facet->used;
4631 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4634 /* Reset counters to prevent double counting if 'facet' ever gets
4636 facet_reset_counters(facet);
4638 netflow_flow_clear(&facet->nf_flow);
4639 facet->tcp_flags = 0;
4642 /* Searches 'ofproto''s table of facets for one which would be responsible for
4643 * 'flow'. Returns it if found, otherwise a null pointer.
4645 * The returned facet might need revalidation; use facet_lookup_valid()
4646 * instead if that is important. */
4647 static struct facet *
4648 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4650 struct cls_rule *cr = classifier_lookup(&ofproto->facets, flow, NULL);
4651 return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4654 /* Searches 'ofproto''s table of facets for one capable that covers
4655 * 'flow'. Returns it if found, otherwise a null pointer.
4657 * The returned facet is guaranteed to be valid. */
4658 static struct facet *
4659 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4661 struct facet *facet;
4663 facet = facet_find(ofproto, flow);
4665 && ofproto->backer->need_revalidate
4666 && !facet_revalidate(facet)) {
4674 facet_check_consistency(struct facet *facet)
4676 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4678 struct xlate_out xout;
4679 struct xlate_in xin;
4681 struct rule_dpif *rule;
4684 /* Check the datapath actions for consistency. */
4685 rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL);
4686 xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4687 xlate_actions(&xin, &xout);
4689 fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4690 ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4691 && facet->xout.slow == xout.slow
4692 && facet->fail_open == fail_open;
4693 if (!ok && !VLOG_DROP_WARN(&rl)) {
4694 struct ds s = DS_EMPTY_INITIALIZER;
4696 flow_format(&s, &facet->flow);
4697 ds_put_cstr(&s, ": inconsistency in facet");
4699 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4700 ds_put_cstr(&s, " (actions were: ");
4701 format_odp_actions(&s, facet->xout.odp_actions.data,
4702 facet->xout.odp_actions.size);
4703 ds_put_cstr(&s, ") (correct actions: ");
4704 format_odp_actions(&s, xout.odp_actions.data,
4705 xout.odp_actions.size);
4706 ds_put_char(&s, ')');
4709 if (facet->xout.slow != xout.slow) {
4710 ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4713 if (facet->fail_open != fail_open) {
4714 ds_put_format(&s, " fail open incorrect. should be %s",
4715 fail_open ? "true" : "false");
4719 xlate_out_uninit(&xout);
4724 /* Re-searches the classifier for 'facet':
4726 * - If the rule found is different from 'facet''s current rule, moves
4727 * 'facet' to the new rule and recompiles its actions.
4729 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4730 * where it is and recompiles its actions anyway.
4732 * - If any of 'facet''s subfacets correspond to a new flow according to
4733 * ofproto_receive(), 'facet' is removed.
4735 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
4737 facet_revalidate(struct facet *facet)
4739 struct ofproto_dpif *ofproto = facet->ofproto;
4740 struct rule_dpif *new_rule;
4741 struct subfacet *subfacet;
4742 struct flow_wildcards wc;
4743 struct xlate_out xout;
4744 struct xlate_in xin;
4746 COVERAGE_INC(facet_revalidate);
4748 /* Check that child subfacets still correspond to this facet. Tunnel
4749 * configuration changes could cause a subfacet's OpenFlow in_port to
4751 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4752 struct ofproto_dpif *recv_ofproto;
4753 struct flow recv_flow;
4756 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4757 subfacet->key_len, &recv_flow, NULL,
4758 &recv_ofproto, NULL);
4760 || recv_ofproto != ofproto
4761 || facet != facet_find(ofproto, &recv_flow)) {
4762 facet_remove(facet);
4767 flow_wildcards_init_catchall(&wc);
4768 new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc);
4770 /* Calculate new datapath actions.
4772 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4773 * emit a NetFlow expiration and, if so, we need to have the old state
4774 * around to properly compose it. */
4775 xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4776 xlate_actions(&xin, &xout);
4777 flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4779 /* A facet's slow path reason should only change under dramatic
4780 * circumstances. Rather than try to update everything, it's simpler to
4781 * remove the facet and start over.
4783 * More importantly, if a facet's wildcards change, it will be relatively
4784 * difficult to figure out if its subfacets still belong to it, and if not
4785 * which facet they may belong to. Again, to avoid the complexity, we
4786 * simply give up instead. */
4787 if (facet->xout.slow != xout.slow
4788 || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4789 facet_remove(facet);
4790 xlate_out_uninit(&xout);
4794 if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4795 LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4796 if (subfacet->path == SF_FAST_PATH) {
4797 struct dpif_flow_stats stats;
4799 subfacet_install(subfacet, &xout.odp_actions, &stats);
4800 subfacet_update_stats(subfacet, &stats);
4804 facet_flush_stats(facet);
4806 ofpbuf_clear(&facet->xout.odp_actions);
4807 ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4808 xout.odp_actions.size);
4811 /* Update 'facet' now that we've taken care of all the old state. */
4812 facet->xout.slow = xout.slow;
4813 facet->xout.has_learn = xout.has_learn;
4814 facet->xout.has_normal = xout.has_normal;
4815 facet->xout.has_fin_timeout = xout.has_fin_timeout;
4816 facet->xout.nf_output_iface = xout.nf_output_iface;
4817 facet->xout.mirrors = xout.mirrors;
4818 facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4819 facet->used = MAX(facet->used, new_rule->up.created);
4820 facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4822 xlate_out_uninit(&xout);
4827 facet_reset_counters(struct facet *facet)
4829 facet->packet_count = 0;
4830 facet->byte_count = 0;
4831 facet->prev_packet_count = 0;
4832 facet->prev_byte_count = 0;
4833 facet->accounted_bytes = 0;
4837 facet_push_stats(struct facet *facet, bool may_learn)
4839 struct dpif_flow_stats stats;
4841 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4842 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4843 ovs_assert(facet->used >= facet->prev_used);
4845 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4846 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4847 stats.used = facet->used;
4848 stats.tcp_flags = facet->tcp_flags;
4850 if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4851 struct ofproto_dpif *ofproto = facet->ofproto;
4852 struct ofport_dpif *in_port;
4853 struct rule_dpif *rule;
4854 struct xlate_in xin;
4856 facet->prev_packet_count = facet->packet_count;
4857 facet->prev_byte_count = facet->byte_count;
4858 facet->prev_used = facet->used;
4860 in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port);
4861 if (in_port && in_port->is_tunnel) {
4862 netdev_vport_inc_rx(in_port->up.netdev, &stats);
4865 rule = rule_dpif_lookup(ofproto, &facet->flow, NULL);
4866 rule_credit_stats(rule, &stats);
4867 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow,
4869 netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4870 mirror_update_stats(ofproto->mbridge, facet->xout.mirrors,
4871 stats.n_packets, stats.n_bytes);
4873 xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags,
4875 xin.resubmit_stats = &stats;
4876 xin.may_learn = may_learn;
4877 xlate_actions_for_side_effects(&xin);
4882 push_all_stats__(bool run_fast)
4884 static long long int rl = LLONG_MIN;
4885 struct ofproto_dpif *ofproto;
4887 if (time_msec() < rl) {
4891 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4892 struct cls_cursor cursor;
4893 struct facet *facet;
4895 cls_cursor_init(&cursor, &ofproto->facets, NULL);
4896 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4897 facet_push_stats(facet, false);
4904 rl = time_msec() + 100;
4908 push_all_stats(void)
4910 push_all_stats__(true);
4914 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4916 rule->packet_count += stats->n_packets;
4917 rule->byte_count += stats->n_bytes;
4918 ofproto_rule_update_used(&rule->up, stats->used);
4923 static struct subfacet *
4924 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4925 size_t key_len, uint32_t key_hash)
4927 struct subfacet *subfacet;
4929 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4930 &backer->subfacets) {
4931 if (subfacet->key_len == key_len
4932 && !memcmp(key, subfacet->key, key_len)) {
4940 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4941 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4942 * existing subfacet if there is one, otherwise creates and returns a
4944 static struct subfacet *
4945 subfacet_create(struct facet *facet, struct flow_miss *miss,
4948 struct dpif_backer *backer = miss->ofproto->backer;
4949 enum odp_key_fitness key_fitness = miss->key_fitness;
4950 const struct nlattr *key = miss->key;
4951 size_t key_len = miss->key_len;
4953 struct subfacet *subfacet;
4955 key_hash = odp_flow_key_hash(key, key_len);
4957 if (list_is_empty(&facet->subfacets)) {
4958 subfacet = &facet->one_subfacet;
4960 subfacet = subfacet_find(backer, key, key_len, key_hash);
4962 if (subfacet->facet == facet) {
4966 /* This shouldn't happen. */
4967 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4968 subfacet_destroy(subfacet);
4971 subfacet = xmalloc(sizeof *subfacet);
4974 hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4975 list_push_back(&facet->subfacets, &subfacet->list_node);
4976 subfacet->facet = facet;
4977 subfacet->key_fitness = key_fitness;
4978 subfacet->key = xmemdup(key, key_len);
4979 subfacet->key_len = key_len;
4980 subfacet->used = now;
4981 subfacet->created = now;
4982 subfacet->dp_packet_count = 0;
4983 subfacet->dp_byte_count = 0;
4984 subfacet->path = SF_NOT_INSTALLED;
4985 subfacet->backer = backer;
4987 backer->subfacet_add_count++;
4991 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4992 * its facet within 'ofproto', and frees it. */
4994 subfacet_destroy__(struct subfacet *subfacet)
4996 struct facet *facet = subfacet->facet;
4997 struct ofproto_dpif *ofproto = facet->ofproto;
4999 /* Update ofproto stats before uninstall the subfacet. */
5000 ofproto->backer->subfacet_del_count++;
5002 subfacet_uninstall(subfacet);
5003 hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
5004 list_remove(&subfacet->list_node);
5005 free(subfacet->key);
5006 if (subfacet != &facet->one_subfacet) {
5011 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5012 * last remaining subfacet in its facet destroys the facet too. */
5014 subfacet_destroy(struct subfacet *subfacet)
5016 struct facet *facet = subfacet->facet;
5018 if (list_is_singleton(&facet->subfacets)) {
5019 /* facet_remove() needs at least one subfacet (it will remove it). */
5020 facet_remove(facet);
5022 subfacet_destroy__(subfacet);
5027 subfacet_destroy_batch(struct dpif_backer *backer,
5028 struct subfacet **subfacets, int n)
5030 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5031 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5032 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5035 for (i = 0; i < n; i++) {
5036 ops[i].type = DPIF_OP_FLOW_DEL;
5037 ops[i].u.flow_del.key = subfacets[i]->key;
5038 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5039 ops[i].u.flow_del.stats = &stats[i];
5043 dpif_operate(backer->dpif, opsp, n);
5044 for (i = 0; i < n; i++) {
5045 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5046 subfacets[i]->path = SF_NOT_INSTALLED;
5047 subfacet_destroy(subfacets[i]);
5052 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5053 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5054 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5055 * since 'subfacet' was last updated.
5057 * Returns 0 if successful, otherwise a positive errno value. */
5059 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
5060 struct dpif_flow_stats *stats)
5062 struct facet *facet = subfacet->facet;
5063 enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
5064 const struct nlattr *actions = odp_actions->data;
5065 size_t actions_len = odp_actions->size;
5066 struct odputil_keybuf maskbuf;
5069 uint64_t slow_path_stub[128 / 8];
5070 enum dpif_flow_put_flags flags;
5073 flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
5076 flags |= DPIF_FP_ZERO_STATS;
5079 if (path == SF_SLOW_PATH) {
5080 compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
5081 slow_path_stub, sizeof slow_path_stub,
5082 &actions, &actions_len);
5085 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5086 if (enable_megaflows) {
5087 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5088 &facet->flow, UINT32_MAX);
5091 ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
5092 subfacet->key_len, mask.data, mask.size,
5093 actions, actions_len, stats);
5096 subfacet_reset_dp_stats(subfacet, stats);
5100 COVERAGE_INC(subfacet_install_fail);
5102 subfacet->path = path;
5107 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5109 subfacet_uninstall(struct subfacet *subfacet)
5111 if (subfacet->path != SF_NOT_INSTALLED) {
5112 struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
5113 struct dpif_flow_stats stats;
5116 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5117 subfacet->key_len, &stats);
5118 subfacet_reset_dp_stats(subfacet, &stats);
5120 subfacet_update_stats(subfacet, &stats);
5122 subfacet->path = SF_NOT_INSTALLED;
5124 ovs_assert(subfacet->dp_packet_count == 0);
5125 ovs_assert(subfacet->dp_byte_count == 0);
5129 /* Resets 'subfacet''s datapath statistics counters. This should be called
5130 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5131 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5132 * was reset in the datapath. 'stats' will be modified to include only
5133 * statistics new since 'subfacet' was last updated. */
5135 subfacet_reset_dp_stats(struct subfacet *subfacet,
5136 struct dpif_flow_stats *stats)
5139 && subfacet->dp_packet_count <= stats->n_packets
5140 && subfacet->dp_byte_count <= stats->n_bytes) {
5141 stats->n_packets -= subfacet->dp_packet_count;
5142 stats->n_bytes -= subfacet->dp_byte_count;
5145 subfacet->dp_packet_count = 0;
5146 subfacet->dp_byte_count = 0;
5149 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5151 * Because of the meaning of a subfacet's counters, it only makes sense to do
5152 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5153 * represents a packet that was sent by hand or if it represents statistics
5154 * that have been cleared out of the datapath. */
5156 subfacet_update_stats(struct subfacet *subfacet,
5157 const struct dpif_flow_stats *stats)
5159 if (stats->n_packets || stats->used > subfacet->used) {
5160 struct facet *facet = subfacet->facet;
5162 subfacet->used = MAX(subfacet->used, stats->used);
5163 facet->used = MAX(facet->used, stats->used);
5164 facet->packet_count += stats->n_packets;
5165 facet->byte_count += stats->n_bytes;
5166 facet->tcp_flags |= stats->tcp_flags;
5172 /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets
5173 * the fields that were relevant as part of the lookup. */
5174 static struct rule_dpif *
5175 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
5176 struct flow_wildcards *wc)
5178 struct rule_dpif *rule;
5180 rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0);
5185 return rule_dpif_miss_rule(ofproto, flow);
5189 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
5190 const struct flow *flow, struct flow_wildcards *wc,
5193 struct cls_rule *cls_rule;
5194 struct classifier *cls;
5197 if (table_id >= N_TABLES) {
5202 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
5203 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
5206 cls = &ofproto->up.tables[table_id].cls;
5207 frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
5208 if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5209 /* We must pretend that transport ports are unavailable. */
5210 struct flow ofpc_normal_flow = *flow;
5211 ofpc_normal_flow.tp_src = htons(0);
5212 ofpc_normal_flow.tp_dst = htons(0);
5213 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
5214 } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
5215 cls_rule = &ofproto->drop_frags_rule->up.cr;
5217 flow_wildcards_init_exact(wc);
5220 cls_rule = classifier_lookup(cls, flow, wc);
5222 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5226 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5228 struct ofport_dpif *port;
5230 port = get_ofp_port(ofproto, flow->in_port.ofp_port);
5232 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
5233 flow->in_port.ofp_port);
5234 return ofproto->miss_rule;
5237 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5238 return ofproto->no_packet_in_rule;
5240 return ofproto->miss_rule;
5244 complete_operation(struct rule_dpif *rule)
5246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5248 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
5250 struct dpif_completion *c = xmalloc(sizeof *c);
5251 c->op = rule->up.pending;
5252 list_push_back(&ofproto->completions, &c->list_node);
5254 ofoperation_complete(rule->up.pending, 0);
5258 static struct rule *
5261 struct rule_dpif *rule = xmalloc(sizeof *rule);
5266 rule_dealloc(struct rule *rule_)
5268 struct rule_dpif *rule = rule_dpif_cast(rule_);
5273 rule_construct(struct rule *rule_)
5275 struct rule_dpif *rule = rule_dpif_cast(rule_);
5276 rule->packet_count = 0;
5277 rule->byte_count = 0;
5278 complete_operation(rule);
5283 rule_destruct(struct rule *rule)
5285 complete_operation(rule_dpif_cast(rule));
5289 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5291 struct rule_dpif *rule = rule_dpif_cast(rule_);
5293 /* push_all_stats() can handle flow misses which, when using the learn
5294 * action, can cause rules to be added and deleted. This can corrupt our
5295 * caller's datastructures which assume that rule_get_stats() doesn't have
5296 * an impact on the flow table. To be safe, we disable miss handling. */
5297 push_all_stats__(false);
5299 /* Start from historical data for 'rule' itself that are no longer tracked
5300 * in facets. This counts, for example, facets that have expired. */
5301 *packets = rule->packet_count;
5302 *bytes = rule->byte_count;
5306 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5307 struct ofpbuf *packet)
5309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5310 struct dpif_flow_stats stats;
5311 struct xlate_out xout;
5312 struct xlate_in xin;
5314 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5315 rule_credit_stats(rule, &stats);
5317 xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5318 xin.resubmit_stats = &stats;
5319 xlate_actions(&xin, &xout);
5321 execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5322 xout.odp_actions.size, packet);
5324 xlate_out_uninit(&xout);
5328 rule_execute(struct rule *rule, const struct flow *flow,
5329 struct ofpbuf *packet)
5331 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5332 ofpbuf_delete(packet);
5337 rule_modify_actions(struct rule *rule_)
5339 struct rule_dpif *rule = rule_dpif_cast(rule_);
5341 complete_operation(rule);
5344 /* Sends 'packet' out 'ofport'.
5345 * May modify 'packet'.
5346 * Returns 0 if successful, otherwise a positive errno value. */
5348 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5351 uint64_t odp_actions_stub[1024 / 8];
5352 struct ofpbuf key, odp_actions;
5353 struct dpif_flow_stats stats;
5354 struct odputil_keybuf keybuf;
5355 struct ofpact_output output;
5356 struct xlate_out xout;
5357 struct xlate_in xin;
5359 union flow_in_port in_port_;
5362 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5363 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5365 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5366 in_port_.ofp_port = OFPP_NONE;
5367 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5368 odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5370 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5372 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5373 output.port = ofport->up.ofp_port;
5376 xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5377 xin.ofpacts_len = sizeof output;
5378 xin.ofpacts = &output.ofpact;
5379 xin.resubmit_stats = &stats;
5380 xlate_actions(&xin, &xout);
5382 error = dpif_execute(ofproto->backer->dpif,
5384 xout.odp_actions.data, xout.odp_actions.size,
5386 xlate_out_uninit(&xout);
5389 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5390 ofproto->up.name, netdev_get_name(ofport->up.netdev),
5391 ovs_strerror(error));
5394 ofproto->stats.tx_packets++;
5395 ofproto->stats.tx_bytes += packet->size;
5399 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5400 * The action will state 'slow' as the reason that the action is in the slow
5401 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5402 * dump-flows" output to see why a flow is in the slow path.)
5404 * The 'stub_size' bytes in 'stub' will be used to store the action.
5405 * 'stub_size' must be large enough for the action.
5407 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5410 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5411 enum slow_path_reason slow,
5412 uint64_t *stub, size_t stub_size,
5413 const struct nlattr **actionsp, size_t *actions_lenp)
5415 union user_action_cookie cookie;
5418 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5419 cookie.slow_path.unused = 0;
5420 cookie.slow_path.reason = slow;
5422 ofpbuf_use_stack(&buf, stub, stub_size);
5423 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5424 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5426 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5428 put_userspace_action(ofproto, &buf, flow, &cookie,
5429 sizeof cookie.slow_path);
5431 *actionsp = buf.data;
5432 *actions_lenp = buf.size;
5436 put_userspace_action(const struct ofproto_dpif *ofproto,
5437 struct ofpbuf *odp_actions,
5438 const struct flow *flow,
5439 const union user_action_cookie *cookie,
5440 const size_t cookie_size)
5444 pid = dpif_port_get_pid(ofproto->backer->dpif,
5445 ofp_port_to_odp_port(ofproto,
5446 flow->in_port.ofp_port));
5448 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5452 set_frag_handling(struct ofproto *ofproto_,
5453 enum ofp_config_flags frag_handling)
5455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5456 if (frag_handling != OFPC_FRAG_REASM) {
5457 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5465 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5466 const struct flow *flow,
5467 const struct ofpact *ofpacts, size_t ofpacts_len)
5469 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5470 struct odputil_keybuf keybuf;
5471 struct dpif_flow_stats stats;
5472 struct xlate_out xout;
5473 struct xlate_in xin;
5477 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5478 odp_flow_key_from_flow(&key, flow,
5479 ofp_port_to_odp_port(ofproto,
5480 flow->in_port.ofp_port));
5482 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5484 xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5485 xin.resubmit_stats = &stats;
5486 xin.ofpacts_len = ofpacts_len;
5487 xin.ofpacts = ofpacts;
5489 xlate_actions(&xin, &xout);
5490 dpif_execute(ofproto->backer->dpif, key.data, key.size,
5491 xout.odp_actions.data, xout.odp_actions.size, packet);
5492 xlate_out_uninit(&xout);
5500 set_netflow(struct ofproto *ofproto_,
5501 const struct netflow_options *netflow_options)
5503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5505 if (netflow_options) {
5506 if (!ofproto->netflow) {
5507 ofproto->netflow = netflow_create();
5508 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5510 return netflow_set_options(ofproto->netflow, netflow_options);
5511 } else if (ofproto->netflow) {
5512 ofproto->backer->need_revalidate = REV_RECONFIGURE;
5513 netflow_destroy(ofproto->netflow);
5514 ofproto->netflow = NULL;
5521 get_netflow_ids(const struct ofproto *ofproto_,
5522 uint8_t *engine_type, uint8_t *engine_id)
5524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5526 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5530 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5532 if (!facet_is_controller_flow(facet) &&
5533 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5534 struct subfacet *subfacet;
5535 struct ofexpired expired;
5537 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5538 if (subfacet->path == SF_FAST_PATH) {
5539 struct dpif_flow_stats stats;
5541 subfacet_install(subfacet, &facet->xout.odp_actions,
5543 subfacet_update_stats(subfacet, &stats);
5547 expired.flow = facet->flow;
5548 expired.packet_count = facet->packet_count;
5549 expired.byte_count = facet->byte_count;
5550 expired.used = facet->used;
5551 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5556 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5558 struct cls_cursor cursor;
5559 struct facet *facet;
5561 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5562 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5563 send_active_timeout(ofproto, facet);
5567 static struct ofproto_dpif *
5568 ofproto_dpif_lookup(const char *name)
5570 struct ofproto_dpif *ofproto;
5572 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5573 hash_string(name, 0), &all_ofproto_dpifs) {
5574 if (!strcmp(ofproto->up.name, name)) {
5582 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5583 const char *argv[], void *aux OVS_UNUSED)
5585 struct ofproto_dpif *ofproto;
5588 ofproto = ofproto_dpif_lookup(argv[1]);
5590 unixctl_command_reply_error(conn, "no such bridge");
5593 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5594 mac_learning_flush(ofproto->ml);
5595 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5597 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5598 ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5599 mac_learning_flush(ofproto->ml);
5600 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5604 unixctl_command_reply(conn, "table successfully flushed");
5607 static struct ofport_dpif *
5608 ofbundle_get_a_port(const struct ofbundle *bundle)
5610 return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5615 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5616 const char *argv[], void *aux OVS_UNUSED)
5618 struct ds ds = DS_EMPTY_INITIALIZER;
5619 const struct ofproto_dpif *ofproto;
5620 const struct mac_entry *e;
5622 ofproto = ofproto_dpif_lookup(argv[1]);
5624 unixctl_command_reply_error(conn, "no such bridge");
5628 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5629 ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5630 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5631 struct ofbundle *bundle = e->port.p;
5632 char name[OFP_MAX_PORT_NAME_LEN];
5634 ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5636 ds_put_format(&ds, "%5s %4d "ETH_ADDR_FMT" %3d\n",
5637 name, e->vlan, ETH_ADDR_ARGS(e->mac),
5638 mac_entry_age(ofproto->ml, e));
5640 ovs_rwlock_unlock(&ofproto->ml->rwlock);
5641 unixctl_command_reply(conn, ds_cstr(&ds));
5646 struct xlate_out xout;
5647 struct xlate_in xin;
5653 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5655 ds_put_char_multiple(result, '\t', level);
5657 ds_put_cstr(result, "No match\n");
5661 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5662 rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5663 cls_rule_format(&rule->up.cr, result);
5664 ds_put_char(result, '\n');
5666 ds_put_char_multiple(result, '\t', level);
5667 ds_put_cstr(result, "OpenFlow ");
5668 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5669 ds_put_char(result, '\n');
5673 trace_format_flow(struct ds *result, int level, const char *title,
5674 struct trace_ctx *trace)
5676 ds_put_char_multiple(result, '\t', level);
5677 ds_put_format(result, "%s: ", title);
5678 if (flow_equal(&trace->xin.flow, &trace->flow)) {
5679 ds_put_cstr(result, "unchanged");
5681 flow_format(result, &trace->xin.flow);
5682 trace->flow = trace->xin.flow;
5684 ds_put_char(result, '\n');
5688 trace_format_regs(struct ds *result, int level, const char *title,
5689 struct trace_ctx *trace)
5693 ds_put_char_multiple(result, '\t', level);
5694 ds_put_format(result, "%s:", title);
5695 for (i = 0; i < FLOW_N_REGS; i++) {
5696 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5698 ds_put_char(result, '\n');
5702 trace_format_odp(struct ds *result, int level, const char *title,
5703 struct trace_ctx *trace)
5705 struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5707 ds_put_char_multiple(result, '\t', level);
5708 ds_put_format(result, "%s: ", title);
5709 format_odp_actions(result, odp_actions->data, odp_actions->size);
5710 ds_put_char(result, '\n');
5714 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5716 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5717 struct ds *result = trace->result;
5719 ds_put_char(result, '\n');
5720 trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5721 trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5722 trace_format_odp(result, recurse + 1, "Resubmitted odp", trace);
5723 trace_format_rule(result, recurse + 1, rule);
5727 trace_report(struct xlate_in *xin, const char *s, int recurse)
5729 struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5730 struct ds *result = trace->result;
5732 ds_put_char_multiple(result, '\t', recurse);
5733 ds_put_cstr(result, s);
5734 ds_put_char(result, '\n');
5738 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5739 void *aux OVS_UNUSED)
5741 const struct dpif_backer *backer;
5742 struct ofproto_dpif *ofproto;
5743 struct ofpbuf odp_key, odp_mask;
5744 struct ofpbuf *packet;
5752 ofpbuf_init(&odp_key, 0);
5753 ofpbuf_init(&odp_mask, 0);
5755 /* Handle "-generate" or a hex string as the last argument. */
5756 if (!strcmp(argv[argc - 1], "-generate")) {
5757 packet = ofpbuf_new(0);
5760 const char *error = eth_from_hex(argv[argc - 1], &packet);
5763 } else if (argc == 4) {
5764 /* The 3-argument form must end in "-generate' or a hex string. */
5765 unixctl_command_reply_error(conn, error);
5770 /* Parse the flow and determine whether a datapath or
5771 * bridge is specified. If function odp_flow_key_from_string()
5772 * returns 0, the flow is a odp_flow. If function
5773 * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5774 if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5775 /* If the odp_flow is the second argument,
5776 * the datapath name is the first argument. */
5778 const char *dp_type;
5779 if (!strncmp(argv[1], "ovs-", 4)) {
5780 dp_type = argv[1] + 4;
5784 backer = shash_find_data(&all_dpif_backers, dp_type);
5786 unixctl_command_reply_error(conn, "Cannot find datapath "
5791 /* No datapath name specified, so there should be only one
5793 struct shash_node *node;
5794 if (shash_count(&all_dpif_backers) != 1) {
5795 unixctl_command_reply_error(conn, "Must specify datapath "
5796 "name, there is more than one type of datapath");
5799 node = shash_first(&all_dpif_backers);
5800 backer = node->data;
5803 /* Extract the ofproto_dpif object from the ofproto_receive()
5805 if (ofproto_receive(backer, NULL, odp_key.data,
5806 odp_key.size, &flow, NULL, &ofproto, NULL)) {
5807 unixctl_command_reply_error(conn, "Invalid datapath flow");
5810 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5811 } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5813 unixctl_command_reply_error(conn, "Must specify bridge name");
5817 ofproto = ofproto_dpif_lookup(argv[1]);
5819 unixctl_command_reply_error(conn, "Unknown bridge name");
5823 unixctl_command_reply_error(conn, "Bad flow syntax");
5827 /* Generate a packet, if requested. */
5829 if (!packet->size) {
5830 flow_compose(packet, &flow);
5832 union flow_in_port in_port_;
5834 in_port_ = flow.in_port;
5835 ds_put_cstr(&result, "Packet: ");
5836 s = ofp_packet_to_string(packet->data, packet->size);
5837 ds_put_cstr(&result, s);
5840 /* Use the metadata from the flow and the packet argument
5841 * to reconstruct the flow. */
5842 flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL,
5847 ofproto_trace(ofproto, &flow, packet, &result);
5848 unixctl_command_reply(conn, ds_cstr(&result));
5851 ds_destroy(&result);
5852 ofpbuf_delete(packet);
5853 ofpbuf_uninit(&odp_key);
5854 ofpbuf_uninit(&odp_mask);
5858 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5859 const struct ofpbuf *packet, struct ds *ds)
5861 struct rule_dpif *rule;
5863 ds_put_cstr(ds, "Flow: ");
5864 flow_format(ds, flow);
5865 ds_put_char(ds, '\n');
5867 rule = rule_dpif_lookup(ofproto, flow, NULL);
5869 trace_format_rule(ds, 0, rule);
5870 if (rule == ofproto->miss_rule) {
5871 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5872 } else if (rule == ofproto->no_packet_in_rule) {
5873 ds_put_cstr(ds, "\nNo match, packets dropped because "
5874 "OFPPC_NO_PACKET_IN is set on in_port.\n");
5875 } else if (rule == ofproto->drop_frags_rule) {
5876 ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5877 "and the fragment handling mode is \"drop\".\n");
5881 uint64_t odp_actions_stub[1024 / 8];
5882 struct ofpbuf odp_actions;
5883 struct trace_ctx trace;
5887 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5890 ofpbuf_use_stub(&odp_actions,
5891 odp_actions_stub, sizeof odp_actions_stub);
5892 xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5893 trace.xin.resubmit_hook = trace_resubmit;
5894 trace.xin.report_hook = trace_report;
5896 xlate_actions(&trace.xin, &trace.xout);
5898 ds_put_char(ds, '\n');
5899 trace_format_flow(ds, 0, "Final flow", &trace);
5901 match_init(&match, flow, &trace.xout.wc);
5902 ds_put_cstr(ds, "Relevant fields: ");
5903 match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5904 ds_put_char(ds, '\n');
5906 ds_put_cstr(ds, "Datapath actions: ");
5907 format_odp_actions(ds, trace.xout.odp_actions.data,
5908 trace.xout.odp_actions.size);
5910 if (trace.xout.slow) {
5911 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5912 "slow path because it:");
5913 switch (trace.xout.slow) {
5915 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5918 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5921 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5924 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5926 case SLOW_CONTROLLER:
5927 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5928 "to the OpenFlow controller.");
5935 xlate_out_uninit(&trace.xout);
5940 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5941 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5944 unixctl_command_reply(conn, NULL);
5948 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5949 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5952 unixctl_command_reply(conn, NULL);
5955 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
5956 * 'reply' describing the results. */
5958 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5960 struct cls_cursor cursor;
5961 struct facet *facet;
5965 cls_cursor_init(&cursor, &ofproto->facets, NULL);
5966 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5967 if (!facet_check_consistency(facet)) {
5972 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5976 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5977 ofproto->up.name, errors);
5979 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5984 ofproto_dpif_self_check(struct unixctl_conn *conn,
5985 int argc, const char *argv[], void *aux OVS_UNUSED)
5987 struct ds reply = DS_EMPTY_INITIALIZER;
5988 struct ofproto_dpif *ofproto;
5991 ofproto = ofproto_dpif_lookup(argv[1]);
5993 unixctl_command_reply_error(conn, "Unknown ofproto (use "
5994 "ofproto/list for help)");
5997 ofproto_dpif_self_check__(ofproto, &reply);
5999 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6000 ofproto_dpif_self_check__(ofproto, &reply);
6004 unixctl_command_reply(conn, ds_cstr(&reply));
6008 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
6009 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
6010 * to destroy 'ofproto_shash' and free the returned value. */
6011 static const struct shash_node **
6012 get_ofprotos(struct shash *ofproto_shash)
6014 const struct ofproto_dpif *ofproto;
6016 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6017 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
6018 shash_add_nocopy(ofproto_shash, name, ofproto);
6021 return shash_sort(ofproto_shash);
6025 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
6026 const char *argv[] OVS_UNUSED,
6027 void *aux OVS_UNUSED)
6029 struct ds ds = DS_EMPTY_INITIALIZER;
6030 struct shash ofproto_shash;
6031 const struct shash_node **sorted_ofprotos;
6034 shash_init(&ofproto_shash);
6035 sorted_ofprotos = get_ofprotos(&ofproto_shash);
6036 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6037 const struct shash_node *node = sorted_ofprotos[i];
6038 ds_put_format(&ds, "%s\n", node->name);
6041 shash_destroy(&ofproto_shash);
6042 free(sorted_ofprotos);
6044 unixctl_command_reply(conn, ds_cstr(&ds));
6049 show_dp_rates(struct ds *ds, const char *heading,
6050 const struct avg_subfacet_rates *rates)
6052 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
6053 heading, rates->add_rate, rates->del_rate);
6057 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
6059 const struct shash_node **ofprotos;
6060 struct ofproto_dpif *ofproto;
6061 struct shash ofproto_shash;
6062 uint64_t n_hit, n_missed;
6063 long long int minutes;
6066 n_hit = n_missed = 0;
6067 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6068 if (ofproto->backer == backer) {
6069 n_missed += ofproto->n_missed;
6070 n_hit += ofproto->n_hit;
6074 ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6075 dpif_name(backer->dpif), n_hit, n_missed);
6076 ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
6077 " life span: %lldms\n", hmap_count(&backer->subfacets),
6078 backer->avg_n_subfacet, backer->max_n_subfacet,
6079 backer->avg_subfacet_life);
6081 minutes = (time_msec() - backer->created) / (1000 * 60);
6082 if (minutes >= 60) {
6083 show_dp_rates(ds, "\thourly avg:", &backer->hourly);
6085 if (minutes >= 60 * 24) {
6086 show_dp_rates(ds, "\tdaily avg:", &backer->daily);
6088 show_dp_rates(ds, "\toverall avg:", &backer->lifetime);
6090 shash_init(&ofproto_shash);
6091 ofprotos = get_ofprotos(&ofproto_shash);
6092 for (i = 0; i < shash_count(&ofproto_shash); i++) {
6093 struct ofproto_dpif *ofproto = ofprotos[i]->data;
6094 const struct shash_node **ports;
6097 if (ofproto->backer != backer) {
6101 ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
6102 ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
6104 ports = shash_sort(&ofproto->up.port_by_name);
6105 for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
6106 const struct shash_node *node = ports[j];
6107 struct ofport *ofport = node->data;
6109 odp_port_t odp_port;
6111 ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
6114 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
6115 if (odp_port != ODPP_NONE) {
6116 ds_put_format(ds, "%"PRIu32":", odp_port);
6118 ds_put_cstr(ds, "none:");
6121 ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
6124 if (!netdev_get_config(ofport->netdev, &config)) {
6125 const struct smap_node **nodes;
6128 nodes = smap_sort(&config);
6129 for (i = 0; i < smap_count(&config); i++) {
6130 const struct smap_node *node = nodes[i];
6131 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
6132 node->key, node->value);
6136 smap_destroy(&config);
6138 ds_put_char(ds, ')');
6139 ds_put_char(ds, '\n');
6143 shash_destroy(&ofproto_shash);
6148 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6149 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6151 struct ds ds = DS_EMPTY_INITIALIZER;
6152 const struct shash_node **backers;
6155 backers = shash_sort(&all_dpif_backers);
6156 for (i = 0; i < shash_count(&all_dpif_backers); i++) {
6157 dpif_show_backer(backers[i]->data, &ds);
6161 unixctl_command_reply(conn, ds_cstr(&ds));
6165 /* Dump the megaflow (facet) cache. This is useful to check the
6166 * correctness of flow wildcarding, since the same mechanism is used for
6167 * both xlate caching and kernel wildcarding.
6169 * It's important to note that in the output the flow description uses
6170 * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
6172 * This command is only needed for advanced debugging, so it's not
6173 * documented in the man page. */
6175 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
6176 int argc OVS_UNUSED, const char *argv[],
6177 void *aux OVS_UNUSED)
6179 struct ds ds = DS_EMPTY_INITIALIZER;
6180 const struct ofproto_dpif *ofproto;
6181 long long int now = time_msec();
6182 struct cls_cursor cursor;
6183 struct facet *facet;
6185 ofproto = ofproto_dpif_lookup(argv[1]);
6187 unixctl_command_reply_error(conn, "no such bridge");
6191 cls_cursor_init(&cursor, &ofproto->facets, NULL);
6192 CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
6193 cls_rule_format(&facet->cr, &ds);
6194 ds_put_cstr(&ds, ", ");
6195 ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
6196 ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
6197 ds_put_cstr(&ds, "Datapath actions: ");
6198 if (facet->xout.slow) {
6199 uint64_t slow_path_stub[128 / 8];
6200 const struct nlattr *actions;
6203 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6204 slow_path_stub, sizeof slow_path_stub,
6205 &actions, &actions_len);
6206 format_odp_actions(&ds, actions, actions_len);
6208 format_odp_actions(&ds, facet->xout.odp_actions.data,
6209 facet->xout.odp_actions.size);
6211 ds_put_cstr(&ds, "\n");
6214 ds_chomp(&ds, '\n');
6215 unixctl_command_reply(conn, ds_cstr(&ds));
6219 /* Disable using the megaflows.
6221 * This command is only needed for advanced debugging, so it's not
6222 * documented in the man page. */
6224 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
6225 int argc OVS_UNUSED,
6226 const char *argv[] OVS_UNUSED,
6227 void *aux OVS_UNUSED)
6229 struct ofproto_dpif *ofproto;
6231 enable_megaflows = false;
6233 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6234 flush(&ofproto->up);
6237 unixctl_command_reply(conn, "megaflows disabled");
6240 /* Re-enable using megaflows.
6242 * This command is only needed for advanced debugging, so it's not
6243 * documented in the man page. */
6245 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
6246 int argc OVS_UNUSED,
6247 const char *argv[] OVS_UNUSED,
6248 void *aux OVS_UNUSED)
6250 struct ofproto_dpif *ofproto;
6252 enable_megaflows = true;
6254 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6255 flush(&ofproto->up);
6258 unixctl_command_reply(conn, "megaflows enabled");
6262 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
6263 int argc OVS_UNUSED, const char *argv[],
6264 void *aux OVS_UNUSED)
6266 struct ds ds = DS_EMPTY_INITIALIZER;
6267 const struct ofproto_dpif *ofproto;
6268 struct subfacet *subfacet;
6270 ofproto = ofproto_dpif_lookup(argv[1]);
6272 unixctl_command_reply_error(conn, "no such bridge");
6276 update_stats(ofproto->backer);
6278 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
6279 struct facet *facet = subfacet->facet;
6280 struct odputil_keybuf maskbuf;
6283 if (facet->ofproto != ofproto) {
6287 ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
6288 if (enable_megaflows) {
6289 odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
6290 &facet->flow, UINT32_MAX);
6293 odp_flow_format(subfacet->key, subfacet->key_len,
6294 mask.data, mask.size, &ds);
6296 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
6297 subfacet->dp_packet_count, subfacet->dp_byte_count);
6298 if (subfacet->used) {
6299 ds_put_format(&ds, "%.3fs",
6300 (time_msec() - subfacet->used) / 1000.0);
6302 ds_put_format(&ds, "never");
6304 if (subfacet->facet->tcp_flags) {
6305 ds_put_cstr(&ds, ", flags:");
6306 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6309 ds_put_cstr(&ds, ", actions:");
6310 if (facet->xout.slow) {
6311 uint64_t slow_path_stub[128 / 8];
6312 const struct nlattr *actions;
6315 compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6316 slow_path_stub, sizeof slow_path_stub,
6317 &actions, &actions_len);
6318 format_odp_actions(&ds, actions, actions_len);
6320 format_odp_actions(&ds, facet->xout.odp_actions.data,
6321 facet->xout.odp_actions.size);
6323 ds_put_char(&ds, '\n');
6326 unixctl_command_reply(conn, ds_cstr(&ds));
6331 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6332 int argc OVS_UNUSED, const char *argv[],
6333 void *aux OVS_UNUSED)
6335 struct ds ds = DS_EMPTY_INITIALIZER;
6336 struct ofproto_dpif *ofproto;
6338 ofproto = ofproto_dpif_lookup(argv[1]);
6340 unixctl_command_reply_error(conn, "no such bridge");
6344 flush(&ofproto->up);
6346 unixctl_command_reply(conn, ds_cstr(&ds));
6351 ofproto_dpif_unixctl_init(void)
6353 static bool registered;
6359 unixctl_command_register(
6361 "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6362 1, 3, ofproto_unixctl_trace, NULL);
6363 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6364 ofproto_unixctl_fdb_flush, NULL);
6365 unixctl_command_register("fdb/show", "bridge", 1, 1,
6366 ofproto_unixctl_fdb_show, NULL);
6367 unixctl_command_register("ofproto/clog", "", 0, 0,
6368 ofproto_dpif_clog, NULL);
6369 unixctl_command_register("ofproto/unclog", "", 0, 0,
6370 ofproto_dpif_unclog, NULL);
6371 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6372 ofproto_dpif_self_check, NULL);
6373 unixctl_command_register("dpif/dump-dps", "", 0, 0,
6374 ofproto_unixctl_dpif_dump_dps, NULL);
6375 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6377 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6378 ofproto_unixctl_dpif_dump_flows, NULL);
6379 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6380 ofproto_unixctl_dpif_del_flows, NULL);
6381 unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6382 ofproto_unixctl_dpif_dump_megaflows, NULL);
6383 unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6384 ofproto_unixctl_dpif_disable_megaflows, NULL);
6385 unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6386 ofproto_unixctl_dpif_enable_megaflows, NULL);
6389 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6391 * This is deprecated. It is only for compatibility with broken device drivers
6392 * in old versions of Linux that do not properly support VLANs when VLAN
6393 * devices are not used. When broken device drivers are no longer in
6394 * widespread use, we will delete these interfaces. */
6397 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6400 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6402 if (realdev_ofp_port == ofport->realdev_ofp_port
6403 && vid == ofport->vlandev_vid) {
6407 ofproto->backer->need_revalidate = REV_RECONFIGURE;
6409 if (ofport->realdev_ofp_port) {
6412 if (realdev_ofp_port && ofport->bundle) {
6413 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6414 * themselves be part of a bundle. */
6415 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6418 ofport->realdev_ofp_port = realdev_ofp_port;
6419 ofport->vlandev_vid = vid;
6421 if (realdev_ofp_port) {
6422 vsp_add(ofport, realdev_ofp_port, vid);
6429 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6431 return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6435 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6437 return !hmap_is_empty(&ofproto->realdev_vid_map);
6440 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6441 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6442 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
6443 * 'vlan_tci' 9, it would return the port number of eth0.9.
6445 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6446 * function just returns its 'realdev_ofp_port' argument. */
6448 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6449 ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6451 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6452 int vid = vlan_tci_to_vid(vlan_tci);
6453 const struct vlan_splinter *vsp;
6455 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6456 hash_realdev_vid(realdev_ofp_port, vid),
6457 &ofproto->realdev_vid_map) {
6458 if (vsp->realdev_ofp_port == realdev_ofp_port
6459 && vsp->vid == vid) {
6460 return vsp->vlandev_ofp_port;
6464 return realdev_ofp_port;
6467 static struct vlan_splinter *
6468 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6470 struct vlan_splinter *vsp;
6472 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6473 hash_ofp_port(vlandev_ofp_port),
6474 &ofproto->vlandev_map) {
6475 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6483 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6484 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6485 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6486 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6487 * eth0 and store 9 in '*vid'.
6489 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6490 * VLAN device. Unless VLAN splinters are enabled, this is what this function
6493 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6494 ofp_port_t vlandev_ofp_port, int *vid)
6496 if (!hmap_is_empty(&ofproto->vlandev_map)) {
6497 const struct vlan_splinter *vsp;
6499 vsp = vlandev_find(ofproto, vlandev_ofp_port);
6504 return vsp->realdev_ofp_port;
6510 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6511 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
6512 * 'flow->in_port' to the "real" device backing the VLAN device, sets
6513 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
6514 * always the case unless VLAN splinters are enabled), returns false without
6515 * making any changes. */
6517 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6522 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6527 /* Cause the flow to be processed as if it came in on the real device with
6528 * the VLAN device's VLAN ID. */
6529 flow->in_port.ofp_port = realdev;
6530 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6535 vsp_remove(struct ofport_dpif *port)
6537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6538 struct vlan_splinter *vsp;
6540 vsp = vlandev_find(ofproto, port->up.ofp_port);
6542 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6543 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6546 port->realdev_ofp_port = 0;
6548 VLOG_ERR("missing vlan device record");
6553 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6557 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6558 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
6559 == realdev_ofp_port)) {
6560 struct vlan_splinter *vsp;
6562 vsp = xmalloc(sizeof *vsp);
6563 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6564 hash_ofp_port(port->up.ofp_port));
6565 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6566 hash_realdev_vid(realdev_ofp_port, vid));
6567 vsp->realdev_ofp_port = realdev_ofp_port;
6568 vsp->vlandev_ofp_port = port->up.ofp_port;
6571 port->realdev_ofp_port = realdev_ofp_port;
6573 VLOG_ERR("duplicate vlan device record");
6578 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6580 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6581 return ofport ? ofport->odp_port : ODPP_NONE;
6584 static struct ofport_dpif *
6585 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6587 struct ofport_dpif *port;
6589 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6590 &backer->odp_to_ofport_map) {
6591 if (port->odp_port == odp_port) {
6600 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6602 struct ofport_dpif *port;
6604 port = odp_port_to_ofport(ofproto->backer, odp_port);
6605 if (port && &ofproto->up == port->up.ofproto) {
6606 return port->up.ofp_port;
6612 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6613 * most heavily weighted element. 'base' designates the rate of decay: after
6614 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6617 exp_mavg(double *avg, int base, double new)
6619 *avg = (*avg * (base - 1) + new) / base;
6623 update_moving_averages(struct dpif_backer *backer)
6625 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6626 long long int minutes = (time_msec() - backer->created) / min_ms;
6629 backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6631 backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6634 backer->lifetime.add_rate = 0.0;
6635 backer->lifetime.del_rate = 0.0;
6638 /* Update hourly averages on the minute boundaries. */
6639 if (time_msec() - backer->last_minute >= min_ms) {
6640 exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6641 exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6643 /* Update daily averages on the hour boundaries. */
6644 if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6645 exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6646 exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6649 backer->total_subfacet_add_count += backer->subfacet_add_count;
6650 backer->total_subfacet_del_count += backer->subfacet_del_count;
6651 backer->subfacet_add_count = 0;
6652 backer->subfacet_del_count = 0;
6653 backer->last_minute += min_ms;
6657 const struct ofproto_class ofproto_dpif_class = {
6692 port_is_lacp_current,
6693 NULL, /* rule_choose_table */
6700 rule_modify_actions,
6714 get_stp_port_status,
6721 is_mirror_output_bundle,
6722 forward_bpdu_changed,
6723 set_mac_table_config,
6725 NULL, /* meter_get_features */
6726 NULL, /* meter_set */
6727 NULL, /* meter_get */
6728 NULL, /* meter_del */