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-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
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);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
211 struct action_xlate_ctx {
212 /* action_xlate_ctx_init() initializes these members. */
215 struct ofproto_dpif *ofproto;
217 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
218 * this flow when actions change header fields. */
221 /* stack for the push and pop actions.
222 * Each stack element is of the type "union mf_subvalue". */
224 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
226 /* The packet corresponding to 'flow', or a null pointer if we are
227 * revalidating without a packet to refer to. */
228 const struct ofpbuf *packet;
230 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
231 * actions update the flow table?
233 * We want to update these tables if we are actually processing a packet,
234 * or if we are accounting for packets that the datapath has processed, but
235 * not if we are just revalidating. */
238 /* The rule that we are currently translating, or NULL. */
239 struct rule_dpif *rule;
241 /* Union of the set of TCP flags seen so far in this flow. (Used only by
242 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
246 /* If nonnull, flow translation calls this function just before executing a
247 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
248 * when the recursion depth is exceeded.
250 * 'rule' is the rule being submitted into. It will be null if the
251 * resubmit or OFPP_TABLE action didn't find a matching rule.
253 * This is normally null so the client has to set it manually after
254 * calling action_xlate_ctx_init(). */
255 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
257 /* If nonnull, flow translation calls this function to report some
258 * significant decision, e.g. to explain why OFPP_NORMAL translation
259 * dropped a packet. */
260 void (*report_hook)(struct action_xlate_ctx *, const char *s);
262 /* If nonnull, flow translation credits the specified statistics to each
263 * rule reached through a resubmit or OFPP_TABLE action.
265 * This is normally null so the client has to set it manually after
266 * calling action_xlate_ctx_init(). */
267 const struct dpif_flow_stats *resubmit_stats;
269 /* xlate_actions() initializes and uses these members. The client might want
270 * to look at them after it returns. */
272 struct ofpbuf *odp_actions; /* Datapath actions. */
273 tag_type tags; /* Tags associated with actions. */
274 enum slow_path_reason slow; /* 0 if fast path may be used. */
275 bool has_learn; /* Actions include NXAST_LEARN? */
276 bool has_normal; /* Actions output to OFPP_NORMAL? */
277 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
278 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
279 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
281 /* xlate_actions() initializes and uses these members, but the client has no
282 * reason to look at them. */
284 int recurse; /* Recursion level, via xlate_table_action. */
285 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
286 struct flow base_flow; /* Flow at the last commit. */
287 uint32_t orig_skb_priority; /* Priority when packet arrived. */
288 uint8_t table_id; /* OpenFlow table ID where flow was found. */
289 uint32_t sflow_n_outputs; /* Number of output ports. */
290 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
291 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
292 bool exit; /* No further actions should be processed. */
295 /* Initial values of fields of the packet that may be changed during
296 * flow processing and needed later. */
297 struct initial_vals {
298 /* This is the value of vlan_tci in the packet as actually received from
299 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
300 * was received via a VLAN splinter. In that case, this value is 0
301 * (because the packet as actually received from the dpif had no 802.1Q
302 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
305 * This member should be removed when the VLAN splinters feature is no
310 static void action_xlate_ctx_init(struct action_xlate_ctx *,
311 struct ofproto_dpif *, const struct flow *,
312 const struct initial_vals *initial_vals,
314 uint8_t tcp_flags, const struct ofpbuf *);
315 static void xlate_actions(struct action_xlate_ctx *,
316 const struct ofpact *ofpacts, size_t ofpacts_len,
317 struct ofpbuf *odp_actions);
318 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
319 const struct ofpact *ofpacts,
321 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
322 uint8_t table_id, bool may_packet_in);
324 static size_t put_userspace_action(const struct ofproto_dpif *,
325 struct ofpbuf *odp_actions,
327 const union user_action_cookie *,
330 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
331 enum slow_path_reason,
332 uint64_t *stub, size_t stub_size,
333 const struct nlattr **actionsp,
334 size_t *actions_lenp);
336 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
338 /* A subfacet (see "struct subfacet" below) has three possible installation
341 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
342 * case just after the subfacet is created, just before the subfacet is
343 * destroyed, or if the datapath returns an error when we try to install a
346 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
348 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
349 * ofproto_dpif is installed in the datapath.
352 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
353 SF_FAST_PATH, /* Full actions are installed. */
354 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
357 static const char *subfacet_path_to_string(enum subfacet_path);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
368 enum odp_key_fitness key_fitness;
372 long long int used; /* Time last used; time created if not used. */
373 long long int created; /* Time created. */
375 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
376 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
380 * These should be essentially identical for every subfacet in a facet, but
381 * may differ in trivial ways due to VLAN splinters. */
382 size_t actions_len; /* Number of bytes in actions[]. */
383 struct nlattr *actions; /* Datapath actions. */
385 enum slow_path_reason slow; /* 0 if fast path may be used. */
386 enum subfacet_path path; /* Installed in datapath? */
388 /* Initial values of the packet that may be needed later. */
389 struct initial_vals initial_vals;
391 /* Datapath port the packet arrived on. This is needed to remove
392 * flows for ports that are no longer part of the bridge. Since the
393 * flow definition only has the OpenFlow port number and the port is
394 * no longer part of the bridge, we can't determine the datapath port
395 * number needed to delete the flow from the datapath. */
396 uint32_t odp_in_port;
399 #define SUBFACET_DESTROY_MAX_BATCH 50
401 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
403 static struct subfacet *subfacet_find(struct ofproto_dpif *,
404 const struct nlattr *key, size_t key_len,
406 static void subfacet_destroy(struct subfacet *);
407 static void subfacet_destroy__(struct subfacet *);
408 static void subfacet_destroy_batch(struct ofproto_dpif *,
409 struct subfacet **, int n);
410 static void subfacet_reset_dp_stats(struct subfacet *,
411 struct dpif_flow_stats *);
412 static void subfacet_update_time(struct subfacet *, long long int used);
413 static void subfacet_update_stats(struct subfacet *,
414 const struct dpif_flow_stats *);
415 static void subfacet_make_actions(struct subfacet *,
416 const struct ofpbuf *packet,
417 struct ofpbuf *odp_actions);
418 static int subfacet_install(struct subfacet *,
419 const struct nlattr *actions, size_t actions_len,
420 struct dpif_flow_stats *, enum slow_path_reason);
421 static void subfacet_uninstall(struct subfacet *);
423 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
425 /* An exact-match instantiation of an OpenFlow flow.
427 * A facet associates a "struct flow", which represents the Open vSwitch
428 * userspace idea of an exact-match flow, with one or more subfacets. Each
429 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
430 * the facet. When the kernel module (or other dpif implementation) and Open
431 * vSwitch userspace agree on the definition of a flow key, there is exactly
432 * one subfacet per facet. If the dpif implementation supports more-specific
433 * flow matching than userspace, however, a facet can have more than one
434 * subfacet, each of which corresponds to some distinction in flow that
435 * userspace simply doesn't understand.
437 * Flow expiration works in terms of subfacets, so a facet must have at least
438 * one subfacet or it will never expire, leaking memory. */
441 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
442 struct list list_node; /* In owning rule's 'facets' list. */
443 struct rule_dpif *rule; /* Owning rule. */
446 struct list subfacets;
447 long long int used; /* Time last used; time created if not used. */
454 * - Do include packets and bytes sent "by hand", e.g. with
457 * - Do include packets and bytes that were obtained from the datapath
458 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
459 * DPIF_FP_ZERO_STATS).
461 * - Do not include packets or bytes that can be obtained from the
462 * datapath for any existing subfacet.
464 uint64_t packet_count; /* Number of packets received. */
465 uint64_t byte_count; /* Number of bytes received. */
467 /* Resubmit statistics. */
468 uint64_t prev_packet_count; /* Number of packets from last stats push. */
469 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
470 long long int prev_used; /* Used time from last stats push. */
473 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
474 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
475 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
477 /* Properties of datapath actions.
479 * Every subfacet has its own actions because actions can differ slightly
480 * between splintered and non-splintered subfacets due to the VLAN tag
481 * being initially different (present vs. absent). All of them have these
482 * properties in common so we just store one copy of them here. */
483 bool has_learn; /* Actions include NXAST_LEARN? */
484 bool has_normal; /* Actions output to OFPP_NORMAL? */
485 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
486 tag_type tags; /* Tags that would require revalidation. */
487 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
489 /* Storage for a single subfacet, to reduce malloc() time and space
490 * overhead. (A facet always has at least one subfacet and in the common
491 * case has exactly one subfacet. However, 'one_subfacet' may not
492 * always be valid, since it could have been removed after newer
493 * subfacets were pushed onto the 'subfacets' list.) */
494 struct subfacet one_subfacet;
496 long long int learn_rl; /* Rate limiter for facet_learn(). */
499 static struct facet *facet_create(struct rule_dpif *,
500 const struct flow *, uint32_t hash);
501 static void facet_remove(struct facet *);
502 static void facet_free(struct facet *);
504 static struct facet *facet_find(struct ofproto_dpif *,
505 const struct flow *, uint32_t hash);
506 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
507 const struct flow *, uint32_t hash);
508 static void facet_revalidate(struct facet *);
509 static bool facet_check_consistency(struct facet *);
511 static void facet_flush_stats(struct facet *);
513 static void facet_update_time(struct facet *, long long int used);
514 static void facet_reset_counters(struct facet *);
515 static void facet_push_stats(struct facet *);
516 static void facet_learn(struct facet *);
517 static void facet_account(struct facet *);
518 static void push_all_stats(void);
520 static struct subfacet *facet_get_subfacet(struct facet *);
522 static bool facet_is_controller_flow(struct facet *);
525 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
529 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
530 struct list bundle_node; /* In struct ofbundle's "ports" list. */
531 struct cfm *cfm; /* Connectivity Fault Management, if any. */
532 struct bfd *bfd; /* BFD, if any. */
533 tag_type tag; /* Tag associated with this port. */
534 bool may_enable; /* May be enabled in bonds. */
535 long long int carrier_seq; /* Carrier status changes. */
536 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
539 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
540 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
541 long long int stp_state_entered;
543 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
545 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
547 * This is deprecated. It is only for compatibility with broken device
548 * drivers in old versions of Linux that do not properly support VLANs when
549 * VLAN devices are not used. When broken device drivers are no longer in
550 * widespread use, we will delete these interfaces. */
551 uint16_t realdev_ofp_port;
555 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
556 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
557 * traffic egressing the 'ofport' with that priority should be marked with. */
558 struct priority_to_dscp {
559 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
560 uint32_t priority; /* Priority of this queue (see struct flow). */
562 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
565 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
567 * This is deprecated. It is only for compatibility with broken device drivers
568 * in old versions of Linux that do not properly support VLANs when VLAN
569 * devices are not used. When broken device drivers are no longer in
570 * widespread use, we will delete these interfaces. */
571 struct vlan_splinter {
572 struct hmap_node realdev_vid_node;
573 struct hmap_node vlandev_node;
574 uint16_t realdev_ofp_port;
575 uint16_t vlandev_ofp_port;
579 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
580 uint32_t realdev, ovs_be16 vlan_tci);
581 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
582 static void vsp_remove(struct ofport_dpif *);
583 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
585 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
587 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
590 static struct ofport_dpif *
591 ofport_dpif_cast(const struct ofport *ofport)
593 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
594 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
597 static void port_run(struct ofport_dpif *);
598 static void port_run_fast(struct ofport_dpif *);
599 static void port_wait(struct ofport_dpif *);
600 static int set_cfm(struct ofport *, const struct cfm_settings *);
601 static void ofport_clear_priorities(struct ofport_dpif *);
602 static void run_fast_rl(void);
604 struct dpif_completion {
605 struct list list_node;
606 struct ofoperation *op;
609 /* Extra information about a classifier table.
610 * Currently used just for optimized flow revalidation. */
612 /* If either of these is nonnull, then this table has a form that allows
613 * flows to be tagged to avoid revalidating most flows for the most common
614 * kinds of flow table changes. */
615 struct cls_table *catchall_table; /* Table that wildcards all fields. */
616 struct cls_table *other_table; /* Table with any other wildcard set. */
617 uint32_t basis; /* Keeps each table's tags separate. */
620 /* Reasons that we might need to revalidate every facet, and corresponding
623 * A value of 0 means that there is no need to revalidate.
625 * It would be nice to have some cleaner way to integrate with coverage
626 * counters, but with only a few reasons I guess this is good enough for
628 enum revalidate_reason {
629 REV_RECONFIGURE = 1, /* Switch configuration changed. */
630 REV_STP, /* Spanning tree protocol port status change. */
631 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
632 REV_FLOW_TABLE, /* Flow table changed. */
633 REV_INCONSISTENCY /* Facet self-check failed. */
635 COVERAGE_DEFINE(rev_reconfigure);
636 COVERAGE_DEFINE(rev_stp);
637 COVERAGE_DEFINE(rev_port_toggled);
638 COVERAGE_DEFINE(rev_flow_table);
639 COVERAGE_DEFINE(rev_inconsistency);
641 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
642 * These are datapath flows which have no associated ofproto, if they did we
643 * would use facets. */
645 struct hmap_node hmap_node;
650 /* All datapaths of a given type share a single dpif backer instance. */
655 struct timer next_expiration;
656 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
658 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
660 /* Facet revalidation flags applying to facets which use this backer. */
661 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
662 struct tag_set revalidate_set; /* Revalidate only matching facets. */
664 struct hmap drop_keys; /* Set of dropped odp keys. */
667 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
668 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
670 static void drop_key_clear(struct dpif_backer *);
671 static struct ofport_dpif *
672 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
674 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
676 struct avg_subfacet_rates {
677 double add_rate; /* Moving average of new flows created per minute. */
678 double del_rate; /* Moving average of flows deleted per minute. */
680 static void show_dp_rates(struct ds *ds, const char *heading,
681 const struct avg_subfacet_rates *rates);
682 static void exp_mavg(double *avg, int base, double new);
684 struct ofproto_dpif {
685 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
687 struct dpif_backer *backer;
689 /* Special OpenFlow rules. */
690 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
691 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
697 struct netflow *netflow;
698 struct dpif_sflow *sflow;
699 struct dpif_ipfix *ipfix;
700 struct hmap bundles; /* Contains "struct ofbundle"s. */
701 struct mac_learning *ml;
702 struct ofmirror *mirrors[MAX_MIRRORS];
704 bool has_bonded_bundles;
708 struct hmap subfacets;
709 struct governor *governor;
710 long long int consistency_rl;
713 struct table_dpif tables[N_TABLES];
715 /* Support for debugging async flow mods. */
716 struct list completions;
718 bool has_bundle_action; /* True when the first bundle action appears. */
719 struct netdev_stats stats; /* To account packets generated and consumed in
724 long long int stp_last_tick;
726 /* VLAN splinters. */
727 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
728 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
731 struct sset ports; /* Set of standard port names. */
732 struct sset ghost_ports; /* Ports with no datapath port. */
733 struct sset port_poll_set; /* Queued names for port_poll() reply. */
734 int port_poll_errno; /* Last errno for port_poll() reply. */
736 /* Per ofproto's dpif stats. */
740 /* Subfacet statistics.
742 * These keep track of the total number of subfacets added and deleted and
743 * flow life span. They are useful for computing the flow rates stats
744 * exposed via "ovs-appctl dpif/show". The goal is to learn about
745 * traffic patterns in ways that we can use later to improve Open vSwitch
746 * performance in new situations. */
747 long long int created; /* Time when it is created. */
748 unsigned int max_n_subfacet; /* Maximum number of flows */
750 /* The average number of subfacets... */
751 struct avg_subfacet_rates hourly; /* ...over the last hour. */
752 struct avg_subfacet_rates daily; /* ...over the last day. */
753 long long int last_minute; /* Last time 'hourly' was updated. */
755 /* Number of subfacets added or deleted since 'last_minute'. */
756 unsigned int subfacet_add_count;
757 unsigned int subfacet_del_count;
759 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
760 unsigned long long int total_subfacet_add_count;
761 unsigned long long int total_subfacet_del_count;
763 /* Sum of the number of milliseconds that each subfacet existed,
764 * over the subfacets that have been added and then later deleted. */
765 unsigned long long int total_subfacet_life_span;
767 /* Incremented by the number of currently existing subfacets, each
768 * time we pull statistics from the kernel. */
769 unsigned long long int total_subfacet_count;
771 /* Number of times we pull statistics from the kernel. */
772 unsigned long long int n_update_stats;
774 static unsigned long long int avg_subfacet_life_span(
775 const struct ofproto_dpif *);
776 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
777 static void update_moving_averages(struct ofproto_dpif *ofproto);
778 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
780 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
782 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
783 * for debugging the asynchronous flow_mod implementation.) */
786 /* All existing ofproto_dpif instances, indexed by ->up.name. */
787 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
789 static void ofproto_dpif_unixctl_init(void);
791 static struct ofproto_dpif *
792 ofproto_dpif_cast(const struct ofproto *ofproto)
794 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
795 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
798 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
800 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
802 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
803 const struct ofpbuf *,
804 const struct initial_vals *, struct ds *);
806 /* Packet processing. */
807 static void update_learning_table(struct ofproto_dpif *,
808 const struct flow *, int vlan,
811 #define FLOW_MISS_MAX_BATCH 50
812 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
814 /* Flow expiration. */
815 static int expire(struct dpif_backer *);
818 static void send_netflow_active_timeouts(struct ofproto_dpif *);
821 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
822 static size_t compose_sflow_action(const struct ofproto_dpif *,
823 struct ofpbuf *odp_actions,
824 const struct flow *, uint32_t odp_port);
825 static void compose_ipfix_action(const struct ofproto_dpif *,
826 struct ofpbuf *odp_actions,
827 const struct flow *);
828 static void add_mirror_actions(struct action_xlate_ctx *ctx,
829 const struct flow *flow);
830 /* Global variables. */
831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
833 /* Initial mappings of port to bridge mappings. */
834 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
836 /* Factory functions. */
839 init(const struct shash *iface_hints)
841 struct shash_node *node;
843 /* Make a local copy, since we don't own 'iface_hints' elements. */
844 SHASH_FOR_EACH(node, iface_hints) {
845 const struct iface_hint *orig_hint = node->data;
846 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
848 new_hint->br_name = xstrdup(orig_hint->br_name);
849 new_hint->br_type = xstrdup(orig_hint->br_type);
850 new_hint->ofp_port = orig_hint->ofp_port;
852 shash_add(&init_ofp_ports, node->name, new_hint);
857 enumerate_types(struct sset *types)
859 dp_enumerate_types(types);
863 enumerate_names(const char *type, struct sset *names)
865 struct ofproto_dpif *ofproto;
868 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
869 if (strcmp(type, ofproto->up.type)) {
872 sset_add(names, ofproto->up.name);
879 del(const char *type, const char *name)
884 error = dpif_open(name, type, &dpif);
886 error = dpif_delete(dpif);
893 port_open_type(const char *datapath_type, const char *port_type)
895 return dpif_port_open_type(datapath_type, port_type);
898 /* Type functions. */
900 static struct ofproto_dpif *
901 lookup_ofproto_dpif_by_port_name(const char *name)
903 struct ofproto_dpif *ofproto;
905 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
906 if (sset_contains(&ofproto->ports, name)) {
915 type_run(const char *type)
917 static long long int push_timer = LLONG_MIN;
918 struct dpif_backer *backer;
922 backer = shash_find_data(&all_dpif_backers, type);
924 /* This is not necessarily a problem, since backers are only
925 * created on demand. */
929 dpif_run(backer->dpif);
931 /* The most natural place to push facet statistics is when they're pulled
932 * from the datapath. However, when there are many flows in the datapath,
933 * this expensive operation can occur so frequently, that it reduces our
934 * ability to quickly set up flows. To reduce the cost, we push statistics
936 if (time_msec() > push_timer) {
937 push_timer = time_msec() + 2000;
941 if (backer->need_revalidate
942 || !tag_set_is_empty(&backer->revalidate_set)) {
943 struct tag_set revalidate_set = backer->revalidate_set;
944 bool need_revalidate = backer->need_revalidate;
945 struct ofproto_dpif *ofproto;
946 struct simap_node *node;
947 struct simap tmp_backers;
949 /* Handle tunnel garbage collection. */
950 simap_init(&tmp_backers);
951 simap_swap(&backer->tnl_backers, &tmp_backers);
953 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
954 struct ofport_dpif *iter;
956 if (backer != ofproto->backer) {
960 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
963 if (!iter->tnl_port) {
967 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
968 node = simap_find(&tmp_backers, dp_port);
970 simap_put(&backer->tnl_backers, dp_port, node->data);
971 simap_delete(&tmp_backers, node);
972 node = simap_find(&backer->tnl_backers, dp_port);
974 node = simap_find(&backer->tnl_backers, dp_port);
976 uint32_t odp_port = UINT32_MAX;
978 if (!dpif_port_add(backer->dpif, iter->up.netdev,
980 simap_put(&backer->tnl_backers, dp_port, odp_port);
981 node = simap_find(&backer->tnl_backers, dp_port);
986 iter->odp_port = node ? node->data : OVSP_NONE;
987 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
989 backer->need_revalidate = REV_RECONFIGURE;
994 SIMAP_FOR_EACH (node, &tmp_backers) {
995 dpif_port_del(backer->dpif, node->data);
997 simap_destroy(&tmp_backers);
999 switch (backer->need_revalidate) {
1000 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1001 case REV_STP: COVERAGE_INC(rev_stp); break;
1002 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1003 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1004 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1007 if (backer->need_revalidate) {
1008 /* Clear the drop_keys in case we should now be accepting some
1009 * formerly dropped flows. */
1010 drop_key_clear(backer);
1013 /* Clear the revalidation flags. */
1014 tag_set_init(&backer->revalidate_set);
1015 backer->need_revalidate = 0;
1017 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1018 struct facet *facet, *next;
1020 if (ofproto->backer != backer) {
1024 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1026 || tag_set_intersects(&revalidate_set, facet->tags)) {
1027 facet_revalidate(facet);
1034 if (timer_expired(&backer->next_expiration)) {
1035 int delay = expire(backer);
1036 timer_set_duration(&backer->next_expiration, delay);
1039 /* Check for port changes in the dpif. */
1040 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1041 struct ofproto_dpif *ofproto;
1042 struct dpif_port port;
1044 /* Don't report on the datapath's device. */
1045 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1049 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1050 &all_ofproto_dpifs) {
1051 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1056 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1057 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1058 /* The port was removed. If we know the datapath,
1059 * report it through poll_set(). If we don't, it may be
1060 * notifying us of a removal we initiated, so ignore it.
1061 * If there's a pending ENOBUFS, let it stand, since
1062 * everything will be reevaluated. */
1063 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1064 sset_add(&ofproto->port_poll_set, devname);
1065 ofproto->port_poll_errno = 0;
1067 } else if (!ofproto) {
1068 /* The port was added, but we don't know with which
1069 * ofproto we should associate it. Delete it. */
1070 dpif_port_del(backer->dpif, port.port_no);
1072 dpif_port_destroy(&port);
1078 if (error != EAGAIN) {
1079 struct ofproto_dpif *ofproto;
1081 /* There was some sort of error, so propagate it to all
1082 * ofprotos that use this backer. */
1083 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1084 &all_ofproto_dpifs) {
1085 if (ofproto->backer == backer) {
1086 sset_clear(&ofproto->port_poll_set);
1087 ofproto->port_poll_errno = error;
1096 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1100 /* Handle one or more batches of upcalls, until there's nothing left to do
1101 * or until we do a fixed total amount of work.
1103 * We do work in batches because it can be much cheaper to set up a number
1104 * of flows and fire off their patches all at once. We do multiple batches
1105 * because in some cases handling a packet can cause another packet to be
1106 * queued almost immediately as part of the return flow. Both
1107 * optimizations can make major improvements on some benchmarks and
1108 * presumably for real traffic as well. */
1110 while (work < max_batch) {
1111 int retval = handle_upcalls(backer, max_batch - work);
1122 type_run_fast(const char *type)
1124 struct dpif_backer *backer;
1126 backer = shash_find_data(&all_dpif_backers, type);
1128 /* This is not necessarily a problem, since backers are only
1129 * created on demand. */
1133 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1139 static long long int port_rl = LLONG_MIN;
1140 static unsigned int backer_rl = 0;
1142 if (time_msec() >= port_rl) {
1143 struct ofproto_dpif *ofproto;
1144 struct ofport_dpif *ofport;
1146 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1148 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1149 port_run_fast(ofport);
1152 port_rl = time_msec() + 200;
1155 /* XXX: We have to be careful not to do too much work in this function. If
1156 * we call dpif_backer_run_fast() too often, or with too large a batch,
1157 * performance improves signifcantly, but at a cost. It's possible for the
1158 * number of flows in the datapath to increase without bound, and for poll
1159 * loops to take 10s of seconds. The correct solution to this problem,
1160 * long term, is to separate flow miss handling into it's own thread so it
1161 * isn't affected by revalidations, and expirations. Until then, this is
1162 * the best we can do. */
1163 if (++backer_rl >= 10) {
1164 struct shash_node *node;
1167 SHASH_FOR_EACH (node, &all_dpif_backers) {
1168 dpif_backer_run_fast(node->data, 1);
1174 type_wait(const char *type)
1176 struct dpif_backer *backer;
1178 backer = shash_find_data(&all_dpif_backers, type);
1180 /* This is not necessarily a problem, since backers are only
1181 * created on demand. */
1185 timer_wait(&backer->next_expiration);
1188 /* Basic life-cycle. */
1190 static int add_internal_flows(struct ofproto_dpif *);
1192 static struct ofproto *
1195 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1196 return &ofproto->up;
1200 dealloc(struct ofproto *ofproto_)
1202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1207 close_dpif_backer(struct dpif_backer *backer)
1209 struct shash_node *node;
1211 ovs_assert(backer->refcount > 0);
1213 if (--backer->refcount) {
1217 drop_key_clear(backer);
1218 hmap_destroy(&backer->drop_keys);
1220 simap_destroy(&backer->tnl_backers);
1221 hmap_destroy(&backer->odp_to_ofport_map);
1222 node = shash_find(&all_dpif_backers, backer->type);
1224 shash_delete(&all_dpif_backers, node);
1225 dpif_close(backer->dpif);
1230 /* Datapath port slated for removal from datapath. */
1231 struct odp_garbage {
1232 struct list list_node;
1237 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1239 struct dpif_backer *backer;
1240 struct dpif_port_dump port_dump;
1241 struct dpif_port port;
1242 struct shash_node *node;
1243 struct list garbage_list;
1244 struct odp_garbage *garbage, *next;
1250 backer = shash_find_data(&all_dpif_backers, type);
1257 backer_name = xasprintf("ovs-%s", type);
1259 /* Remove any existing datapaths, since we assume we're the only
1260 * userspace controlling the datapath. */
1262 dp_enumerate_names(type, &names);
1263 SSET_FOR_EACH(name, &names) {
1264 struct dpif *old_dpif;
1266 /* Don't remove our backer if it exists. */
1267 if (!strcmp(name, backer_name)) {
1271 if (dpif_open(name, type, &old_dpif)) {
1272 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1274 dpif_delete(old_dpif);
1275 dpif_close(old_dpif);
1278 sset_destroy(&names);
1280 backer = xmalloc(sizeof *backer);
1282 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1285 VLOG_ERR("failed to open datapath of type %s: %s", type,
1291 backer->type = xstrdup(type);
1292 backer->refcount = 1;
1293 hmap_init(&backer->odp_to_ofport_map);
1294 hmap_init(&backer->drop_keys);
1295 timer_set_duration(&backer->next_expiration, 1000);
1296 backer->need_revalidate = 0;
1297 simap_init(&backer->tnl_backers);
1298 tag_set_init(&backer->revalidate_set);
1301 dpif_flow_flush(backer->dpif);
1303 /* Loop through the ports already on the datapath and remove any
1304 * that we don't need anymore. */
1305 list_init(&garbage_list);
1306 dpif_port_dump_start(&port_dump, backer->dpif);
1307 while (dpif_port_dump_next(&port_dump, &port)) {
1308 node = shash_find(&init_ofp_ports, port.name);
1309 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1310 garbage = xmalloc(sizeof *garbage);
1311 garbage->odp_port = port.port_no;
1312 list_push_front(&garbage_list, &garbage->list_node);
1315 dpif_port_dump_done(&port_dump);
1317 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1318 dpif_port_del(backer->dpif, garbage->odp_port);
1319 list_remove(&garbage->list_node);
1323 shash_add(&all_dpif_backers, type, backer);
1325 error = dpif_recv_set(backer->dpif, true);
1327 VLOG_ERR("failed to listen on datapath of type %s: %s",
1328 type, strerror(error));
1329 close_dpif_backer(backer);
1337 construct(struct ofproto *ofproto_)
1339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1340 struct shash_node *node, *next;
1345 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1350 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1351 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1353 ofproto->n_matches = 0;
1355 ofproto->netflow = NULL;
1356 ofproto->sflow = NULL;
1357 ofproto->ipfix = NULL;
1358 ofproto->stp = NULL;
1359 hmap_init(&ofproto->bundles);
1360 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1361 for (i = 0; i < MAX_MIRRORS; i++) {
1362 ofproto->mirrors[i] = NULL;
1364 ofproto->has_bonded_bundles = false;
1366 hmap_init(&ofproto->facets);
1367 hmap_init(&ofproto->subfacets);
1368 ofproto->governor = NULL;
1369 ofproto->consistency_rl = LLONG_MIN;
1371 for (i = 0; i < N_TABLES; i++) {
1372 struct table_dpif *table = &ofproto->tables[i];
1374 table->catchall_table = NULL;
1375 table->other_table = NULL;
1376 table->basis = random_uint32();
1379 list_init(&ofproto->completions);
1381 ofproto_dpif_unixctl_init();
1383 ofproto->has_mirrors = false;
1384 ofproto->has_bundle_action = false;
1386 hmap_init(&ofproto->vlandev_map);
1387 hmap_init(&ofproto->realdev_vid_map);
1389 sset_init(&ofproto->ports);
1390 sset_init(&ofproto->ghost_ports);
1391 sset_init(&ofproto->port_poll_set);
1392 ofproto->port_poll_errno = 0;
1394 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1395 struct iface_hint *iface_hint = node->data;
1397 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1398 /* Check if the datapath already has this port. */
1399 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1400 sset_add(&ofproto->ports, node->name);
1403 free(iface_hint->br_name);
1404 free(iface_hint->br_type);
1406 shash_delete(&init_ofp_ports, node);
1410 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1411 hash_string(ofproto->up.name, 0));
1412 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1414 ofproto_init_tables(ofproto_, N_TABLES);
1415 error = add_internal_flows(ofproto);
1416 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1419 ofproto->n_missed = 0;
1421 ofproto->max_n_subfacet = 0;
1422 ofproto->created = time_msec();
1423 ofproto->last_minute = ofproto->created;
1424 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1425 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1426 ofproto->subfacet_add_count = 0;
1427 ofproto->subfacet_del_count = 0;
1428 ofproto->total_subfacet_add_count = 0;
1429 ofproto->total_subfacet_del_count = 0;
1430 ofproto->total_subfacet_life_span = 0;
1431 ofproto->total_subfacet_count = 0;
1432 ofproto->n_update_stats = 0;
1438 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1439 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1441 struct ofputil_flow_mod fm;
1444 match_init_catchall(&fm.match);
1446 match_set_reg(&fm.match, 0, id);
1447 fm.new_cookie = htonll(0);
1448 fm.cookie = htonll(0);
1449 fm.cookie_mask = htonll(0);
1450 fm.table_id = TBL_INTERNAL;
1451 fm.command = OFPFC_ADD;
1452 fm.idle_timeout = 0;
1453 fm.hard_timeout = 0;
1457 fm.ofpacts = ofpacts->data;
1458 fm.ofpacts_len = ofpacts->size;
1460 error = ofproto_flow_mod(&ofproto->up, &fm);
1462 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1463 id, ofperr_to_string(error));
1467 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1468 ovs_assert(*rulep != NULL);
1474 add_internal_flows(struct ofproto_dpif *ofproto)
1476 struct ofpact_controller *controller;
1477 uint64_t ofpacts_stub[128 / 8];
1478 struct ofpbuf ofpacts;
1482 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1485 controller = ofpact_put_CONTROLLER(&ofpacts);
1486 controller->max_len = UINT16_MAX;
1487 controller->controller_id = 0;
1488 controller->reason = OFPR_NO_MATCH;
1489 ofpact_pad(&ofpacts);
1491 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1496 ofpbuf_clear(&ofpacts);
1497 error = add_internal_flow(ofproto, id++, &ofpacts,
1498 &ofproto->no_packet_in_rule);
1503 complete_operations(struct ofproto_dpif *ofproto)
1505 struct dpif_completion *c, *next;
1507 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1508 ofoperation_complete(c->op, 0);
1509 list_remove(&c->list_node);
1515 destruct(struct ofproto *ofproto_)
1517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1518 struct rule_dpif *rule, *next_rule;
1519 struct oftable *table;
1522 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1523 complete_operations(ofproto);
1525 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1526 struct cls_cursor cursor;
1528 cls_cursor_init(&cursor, &table->cls, NULL);
1529 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1530 ofproto_rule_destroy(&rule->up);
1534 for (i = 0; i < MAX_MIRRORS; i++) {
1535 mirror_destroy(ofproto->mirrors[i]);
1538 netflow_destroy(ofproto->netflow);
1539 dpif_sflow_destroy(ofproto->sflow);
1540 hmap_destroy(&ofproto->bundles);
1541 mac_learning_destroy(ofproto->ml);
1543 hmap_destroy(&ofproto->facets);
1544 hmap_destroy(&ofproto->subfacets);
1545 governor_destroy(ofproto->governor);
1547 hmap_destroy(&ofproto->vlandev_map);
1548 hmap_destroy(&ofproto->realdev_vid_map);
1550 sset_destroy(&ofproto->ports);
1551 sset_destroy(&ofproto->ghost_ports);
1552 sset_destroy(&ofproto->port_poll_set);
1554 close_dpif_backer(ofproto->backer);
1558 run_fast(struct ofproto *ofproto_)
1560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1561 struct ofport_dpif *ofport;
1563 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1564 port_run_fast(ofport);
1571 run(struct ofproto *ofproto_)
1573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1574 struct ofport_dpif *ofport;
1575 struct ofbundle *bundle;
1579 complete_operations(ofproto);
1582 error = run_fast(ofproto_);
1587 if (ofproto->netflow) {
1588 if (netflow_run(ofproto->netflow)) {
1589 send_netflow_active_timeouts(ofproto);
1592 if (ofproto->sflow) {
1593 dpif_sflow_run(ofproto->sflow);
1596 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1599 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1604 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1606 /* Check the consistency of a random facet, to aid debugging. */
1607 if (time_msec() >= ofproto->consistency_rl
1608 && !hmap_is_empty(&ofproto->facets)
1609 && !ofproto->backer->need_revalidate) {
1610 struct facet *facet;
1612 ofproto->consistency_rl = time_msec() + 250;
1614 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1615 struct facet, hmap_node);
1616 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1618 if (!facet_check_consistency(facet)) {
1619 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1624 if (ofproto->governor) {
1627 governor_run(ofproto->governor);
1629 /* If the governor has shrunk to its minimum size and the number of
1630 * subfacets has dwindled, then drop the governor entirely.
1632 * For hysteresis, the number of subfacets to drop the governor is
1633 * smaller than the number needed to trigger its creation. */
1634 n_subfacets = hmap_count(&ofproto->subfacets);
1635 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1636 && governor_is_idle(ofproto->governor)) {
1637 governor_destroy(ofproto->governor);
1638 ofproto->governor = NULL;
1646 wait(struct ofproto *ofproto_)
1648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1649 struct ofport_dpif *ofport;
1650 struct ofbundle *bundle;
1652 if (!clogged && !list_is_empty(&ofproto->completions)) {
1653 poll_immediate_wake();
1656 dpif_wait(ofproto->backer->dpif);
1657 dpif_recv_wait(ofproto->backer->dpif);
1658 if (ofproto->sflow) {
1659 dpif_sflow_wait(ofproto->sflow);
1661 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1662 poll_immediate_wake();
1664 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1667 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1668 bundle_wait(bundle);
1670 if (ofproto->netflow) {
1671 netflow_wait(ofproto->netflow);
1673 mac_learning_wait(ofproto->ml);
1675 if (ofproto->backer->need_revalidate) {
1676 /* Shouldn't happen, but if it does just go around again. */
1677 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1678 poll_immediate_wake();
1680 if (ofproto->governor) {
1681 governor_wait(ofproto->governor);
1686 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1688 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1691 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1695 flush(struct ofproto *ofproto_)
1697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1698 struct subfacet *subfacet, *next_subfacet;
1699 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1703 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1704 &ofproto->subfacets) {
1705 if (subfacet->path != SF_NOT_INSTALLED) {
1706 batch[n_batch++] = subfacet;
1707 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1708 subfacet_destroy_batch(ofproto, batch, n_batch);
1712 subfacet_destroy(subfacet);
1717 subfacet_destroy_batch(ofproto, batch, n_batch);
1722 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1723 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1725 *arp_match_ip = true;
1726 *actions = (OFPUTIL_A_OUTPUT |
1727 OFPUTIL_A_SET_VLAN_VID |
1728 OFPUTIL_A_SET_VLAN_PCP |
1729 OFPUTIL_A_STRIP_VLAN |
1730 OFPUTIL_A_SET_DL_SRC |
1731 OFPUTIL_A_SET_DL_DST |
1732 OFPUTIL_A_SET_NW_SRC |
1733 OFPUTIL_A_SET_NW_DST |
1734 OFPUTIL_A_SET_NW_TOS |
1735 OFPUTIL_A_SET_TP_SRC |
1736 OFPUTIL_A_SET_TP_DST |
1741 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1744 struct dpif_dp_stats s;
1746 strcpy(ots->name, "classifier");
1748 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1750 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1751 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1754 static struct ofport *
1757 struct ofport_dpif *port = xmalloc(sizeof *port);
1762 port_dealloc(struct ofport *port_)
1764 struct ofport_dpif *port = ofport_dpif_cast(port_);
1769 port_construct(struct ofport *port_)
1771 struct ofport_dpif *port = ofport_dpif_cast(port_);
1772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1773 const struct netdev *netdev = port->up.netdev;
1774 struct dpif_port dpif_port;
1777 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1778 port->bundle = NULL;
1781 port->tag = tag_create_random();
1782 port->may_enable = true;
1783 port->stp_port = NULL;
1784 port->stp_state = STP_DISABLED;
1785 port->tnl_port = NULL;
1786 hmap_init(&port->priorities);
1787 port->realdev_ofp_port = 0;
1788 port->vlandev_vid = 0;
1789 port->carrier_seq = netdev_get_carrier_resets(netdev);
1791 if (netdev_vport_is_patch(netdev)) {
1792 /* By bailing out here, we don't submit the port to the sFlow module
1793 * to be considered for counter polling export. This is correct
1794 * because the patch port represents an interface that sFlow considers
1795 * to be "internal" to the switch as a whole, and therefore not an
1796 * candidate for counter polling. */
1797 port->odp_port = OVSP_NONE;
1801 error = dpif_port_query_by_name(ofproto->backer->dpif,
1802 netdev_vport_get_dpif_port(netdev),
1808 port->odp_port = dpif_port.port_no;
1810 if (netdev_get_tunnel_config(netdev)) {
1811 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1813 /* Sanity-check that a mapping doesn't already exist. This
1814 * shouldn't happen for non-tunnel ports. */
1815 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1816 VLOG_ERR("port %s already has an OpenFlow port number",
1818 dpif_port_destroy(&dpif_port);
1822 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1823 hash_int(port->odp_port, 0));
1825 dpif_port_destroy(&dpif_port);
1827 if (ofproto->sflow) {
1828 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1835 port_destruct(struct ofport *port_)
1837 struct ofport_dpif *port = ofport_dpif_cast(port_);
1838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1839 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1840 const char *devname = netdev_get_name(port->up.netdev);
1842 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1843 /* The underlying device is still there, so delete it. This
1844 * happens when the ofproto is being destroyed, since the caller
1845 * assumes that removal of attached ports will happen as part of
1847 if (!port->tnl_port) {
1848 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1850 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1854 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1857 tnl_port_del(port->tnl_port);
1858 sset_find_and_delete(&ofproto->ports, devname);
1859 sset_find_and_delete(&ofproto->ghost_ports, devname);
1860 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861 bundle_remove(port_);
1862 set_cfm(port_, NULL);
1863 if (ofproto->sflow) {
1864 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1867 ofport_clear_priorities(port);
1868 hmap_destroy(&port->priorities);
1872 port_modified(struct ofport *port_)
1874 struct ofport_dpif *port = ofport_dpif_cast(port_);
1876 if (port->bundle && port->bundle->bond) {
1877 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1882 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1884 struct ofport_dpif *port = ofport_dpif_cast(port_);
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1886 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1888 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1889 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1890 OFPUTIL_PC_NO_PACKET_IN)) {
1891 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1893 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1894 bundle_update(port->bundle);
1900 set_sflow(struct ofproto *ofproto_,
1901 const struct ofproto_sflow_options *sflow_options)
1903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1904 struct dpif_sflow *ds = ofproto->sflow;
1906 if (sflow_options) {
1908 struct ofport_dpif *ofport;
1910 ds = ofproto->sflow = dpif_sflow_create();
1911 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1912 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1914 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1916 dpif_sflow_set_options(ds, sflow_options);
1919 dpif_sflow_destroy(ds);
1920 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1921 ofproto->sflow = NULL;
1929 struct ofproto *ofproto_,
1930 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1931 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1932 size_t n_flow_exporters_options)
1934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1935 struct dpif_ipfix *di = ofproto->ipfix;
1937 if (bridge_exporter_options || flow_exporters_options) {
1939 di = ofproto->ipfix = dpif_ipfix_create();
1941 dpif_ipfix_set_options(
1942 di, bridge_exporter_options, flow_exporters_options,
1943 n_flow_exporters_options);
1946 dpif_ipfix_destroy(di);
1947 ofproto->ipfix = NULL;
1954 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1956 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1963 struct ofproto_dpif *ofproto;
1965 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1966 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1967 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1970 if (cfm_configure(ofport->cfm, s)) {
1976 cfm_destroy(ofport->cfm);
1982 get_cfm_status(const struct ofport *ofport_,
1983 struct ofproto_cfm_status *status)
1985 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1988 status->faults = cfm_get_fault(ofport->cfm);
1989 status->remote_opstate = cfm_get_opup(ofport->cfm);
1990 status->health = cfm_get_health(ofport->cfm);
1991 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1999 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2002 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2006 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2007 if (ofport->bfd != old) {
2008 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2015 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2017 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2020 bfd_get_status(ofport->bfd, smap);
2027 /* Spanning Tree. */
2030 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2032 struct ofproto_dpif *ofproto = ofproto_;
2033 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2034 struct ofport_dpif *ofport;
2036 ofport = stp_port_get_aux(sp);
2038 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2039 ofproto->up.name, port_num);
2041 struct eth_header *eth = pkt->l2;
2043 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2044 if (eth_addr_is_zero(eth->eth_src)) {
2045 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2046 "with unknown MAC", ofproto->up.name, port_num);
2048 send_packet(ofport, pkt);
2054 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2056 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2060 /* Only revalidate flows if the configuration changed. */
2061 if (!s != !ofproto->stp) {
2062 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2066 if (!ofproto->stp) {
2067 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2068 send_bpdu_cb, ofproto);
2069 ofproto->stp_last_tick = time_msec();
2072 stp_set_bridge_id(ofproto->stp, s->system_id);
2073 stp_set_bridge_priority(ofproto->stp, s->priority);
2074 stp_set_hello_time(ofproto->stp, s->hello_time);
2075 stp_set_max_age(ofproto->stp, s->max_age);
2076 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2078 struct ofport *ofport;
2080 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2081 set_stp_port(ofport, NULL);
2084 stp_destroy(ofproto->stp);
2085 ofproto->stp = NULL;
2092 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2098 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2099 s->designated_root = stp_get_designated_root(ofproto->stp);
2100 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2109 update_stp_port_state(struct ofport_dpif *ofport)
2111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2112 enum stp_state state;
2114 /* Figure out new state. */
2115 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2119 if (ofport->stp_state != state) {
2120 enum ofputil_port_state of_state;
2123 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2124 netdev_get_name(ofport->up.netdev),
2125 stp_state_name(ofport->stp_state),
2126 stp_state_name(state));
2127 if (stp_learn_in_state(ofport->stp_state)
2128 != stp_learn_in_state(state)) {
2129 /* xxx Learning action flows should also be flushed. */
2130 mac_learning_flush(ofproto->ml,
2131 &ofproto->backer->revalidate_set);
2133 fwd_change = stp_forward_in_state(ofport->stp_state)
2134 != stp_forward_in_state(state);
2136 ofproto->backer->need_revalidate = REV_STP;
2137 ofport->stp_state = state;
2138 ofport->stp_state_entered = time_msec();
2140 if (fwd_change && ofport->bundle) {
2141 bundle_update(ofport->bundle);
2144 /* Update the STP state bits in the OpenFlow port description. */
2145 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2146 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2147 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2148 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2149 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2151 ofproto_port_set_state(&ofport->up, of_state);
2155 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2156 * caller is responsible for assigning STP port numbers and ensuring
2157 * there are no duplicates. */
2159 set_stp_port(struct ofport *ofport_,
2160 const struct ofproto_port_stp_settings *s)
2162 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2163 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2164 struct stp_port *sp = ofport->stp_port;
2166 if (!s || !s->enable) {
2168 ofport->stp_port = NULL;
2169 stp_port_disable(sp);
2170 update_stp_port_state(ofport);
2173 } else if (sp && stp_port_no(sp) != s->port_num
2174 && ofport == stp_port_get_aux(sp)) {
2175 /* The port-id changed, so disable the old one if it's not
2176 * already in use by another port. */
2177 stp_port_disable(sp);
2180 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2181 stp_port_enable(sp);
2183 stp_port_set_aux(sp, ofport);
2184 stp_port_set_priority(sp, s->priority);
2185 stp_port_set_path_cost(sp, s->path_cost);
2187 update_stp_port_state(ofport);
2193 get_stp_port_status(struct ofport *ofport_,
2194 struct ofproto_port_stp_status *s)
2196 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2198 struct stp_port *sp = ofport->stp_port;
2200 if (!ofproto->stp || !sp) {
2206 s->port_id = stp_port_get_id(sp);
2207 s->state = stp_port_get_state(sp);
2208 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2209 s->role = stp_port_get_role(sp);
2210 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2216 stp_run(struct ofproto_dpif *ofproto)
2219 long long int now = time_msec();
2220 long long int elapsed = now - ofproto->stp_last_tick;
2221 struct stp_port *sp;
2224 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2225 ofproto->stp_last_tick = now;
2227 while (stp_get_changed_port(ofproto->stp, &sp)) {
2228 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2231 update_stp_port_state(ofport);
2235 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2236 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2242 stp_wait(struct ofproto_dpif *ofproto)
2245 poll_timer_wait(1000);
2249 /* Returns true if STP should process 'flow'. */
2251 stp_should_process_flow(const struct flow *flow)
2253 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2257 stp_process_packet(const struct ofport_dpif *ofport,
2258 const struct ofpbuf *packet)
2260 struct ofpbuf payload = *packet;
2261 struct eth_header *eth = payload.data;
2262 struct stp_port *sp = ofport->stp_port;
2264 /* Sink packets on ports that have STP disabled when the bridge has
2266 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2270 /* Trim off padding on payload. */
2271 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2272 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2275 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2276 stp_received_bpdu(sp, payload.data, payload.size);
2280 static struct priority_to_dscp *
2281 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2283 struct priority_to_dscp *pdscp;
2286 hash = hash_int(priority, 0);
2287 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2288 if (pdscp->priority == priority) {
2296 ofport_clear_priorities(struct ofport_dpif *ofport)
2298 struct priority_to_dscp *pdscp, *next;
2300 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2301 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2307 set_queues(struct ofport *ofport_,
2308 const struct ofproto_port_queue *qdscp_list,
2311 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2313 struct hmap new = HMAP_INITIALIZER(&new);
2316 for (i = 0; i < n_qdscp; i++) {
2317 struct priority_to_dscp *pdscp;
2321 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2322 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2327 pdscp = get_priority(ofport, priority);
2329 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2331 pdscp = xmalloc(sizeof *pdscp);
2332 pdscp->priority = priority;
2334 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2337 if (pdscp->dscp != dscp) {
2339 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2342 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2345 if (!hmap_is_empty(&ofport->priorities)) {
2346 ofport_clear_priorities(ofport);
2347 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2350 hmap_swap(&new, &ofport->priorities);
2358 /* Expires all MAC learning entries associated with 'bundle' and forces its
2359 * ofproto to revalidate every flow.
2361 * Normally MAC learning entries are removed only from the ofproto associated
2362 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2363 * are removed from every ofproto. When patch ports and SLB bonds are in use
2364 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2365 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2366 * with the host from which it migrated. */
2368 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2370 struct ofproto_dpif *ofproto = bundle->ofproto;
2371 struct mac_learning *ml = ofproto->ml;
2372 struct mac_entry *mac, *next_mac;
2374 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2375 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2376 if (mac->port.p == bundle) {
2378 struct ofproto_dpif *o;
2380 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2382 struct mac_entry *e;
2384 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2387 mac_learning_expire(o->ml, e);
2393 mac_learning_expire(ml, mac);
2398 static struct ofbundle *
2399 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2401 struct ofbundle *bundle;
2403 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2404 &ofproto->bundles) {
2405 if (bundle->aux == aux) {
2412 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2413 * ones that are found to 'bundles'. */
2415 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2416 void **auxes, size_t n_auxes,
2417 struct hmapx *bundles)
2421 hmapx_init(bundles);
2422 for (i = 0; i < n_auxes; i++) {
2423 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2425 hmapx_add(bundles, bundle);
2431 bundle_update(struct ofbundle *bundle)
2433 struct ofport_dpif *port;
2435 bundle->floodable = true;
2436 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2437 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2438 || !stp_forward_in_state(port->stp_state)) {
2439 bundle->floodable = false;
2446 bundle_del_port(struct ofport_dpif *port)
2448 struct ofbundle *bundle = port->bundle;
2450 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2452 list_remove(&port->bundle_node);
2453 port->bundle = NULL;
2456 lacp_slave_unregister(bundle->lacp, port);
2459 bond_slave_unregister(bundle->bond, port);
2462 bundle_update(bundle);
2466 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2467 struct lacp_slave_settings *lacp)
2469 struct ofport_dpif *port;
2471 port = get_ofp_port(bundle->ofproto, ofp_port);
2476 if (port->bundle != bundle) {
2477 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2479 bundle_del_port(port);
2482 port->bundle = bundle;
2483 list_push_back(&bundle->ports, &port->bundle_node);
2484 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2485 || !stp_forward_in_state(port->stp_state)) {
2486 bundle->floodable = false;
2490 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2491 lacp_slave_register(bundle->lacp, port, lacp);
2498 bundle_destroy(struct ofbundle *bundle)
2500 struct ofproto_dpif *ofproto;
2501 struct ofport_dpif *port, *next_port;
2508 ofproto = bundle->ofproto;
2509 for (i = 0; i < MAX_MIRRORS; i++) {
2510 struct ofmirror *m = ofproto->mirrors[i];
2512 if (m->out == bundle) {
2514 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2515 || hmapx_find_and_delete(&m->dsts, bundle)) {
2516 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2521 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2522 bundle_del_port(port);
2525 bundle_flush_macs(bundle, true);
2526 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2528 free(bundle->trunks);
2529 lacp_destroy(bundle->lacp);
2530 bond_destroy(bundle->bond);
2535 bundle_set(struct ofproto *ofproto_, void *aux,
2536 const struct ofproto_bundle_settings *s)
2538 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2539 bool need_flush = false;
2540 struct ofport_dpif *port;
2541 struct ofbundle *bundle;
2542 unsigned long *trunks;
2548 bundle_destroy(bundle_lookup(ofproto, aux));
2552 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2553 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2555 bundle = bundle_lookup(ofproto, aux);
2557 bundle = xmalloc(sizeof *bundle);
2559 bundle->ofproto = ofproto;
2560 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2561 hash_pointer(aux, 0));
2563 bundle->name = NULL;
2565 list_init(&bundle->ports);
2566 bundle->vlan_mode = PORT_VLAN_TRUNK;
2568 bundle->trunks = NULL;
2569 bundle->use_priority_tags = s->use_priority_tags;
2570 bundle->lacp = NULL;
2571 bundle->bond = NULL;
2573 bundle->floodable = true;
2575 bundle->src_mirrors = 0;
2576 bundle->dst_mirrors = 0;
2577 bundle->mirror_out = 0;
2580 if (!bundle->name || strcmp(s->name, bundle->name)) {
2582 bundle->name = xstrdup(s->name);
2587 if (!bundle->lacp) {
2588 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2589 bundle->lacp = lacp_create();
2591 lacp_configure(bundle->lacp, s->lacp);
2593 lacp_destroy(bundle->lacp);
2594 bundle->lacp = NULL;
2597 /* Update set of ports. */
2599 for (i = 0; i < s->n_slaves; i++) {
2600 if (!bundle_add_port(bundle, s->slaves[i],
2601 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2605 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2606 struct ofport_dpif *next_port;
2608 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2609 for (i = 0; i < s->n_slaves; i++) {
2610 if (s->slaves[i] == port->up.ofp_port) {
2615 bundle_del_port(port);
2619 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2621 if (list_is_empty(&bundle->ports)) {
2622 bundle_destroy(bundle);
2626 /* Set VLAN tagging mode */
2627 if (s->vlan_mode != bundle->vlan_mode
2628 || s->use_priority_tags != bundle->use_priority_tags) {
2629 bundle->vlan_mode = s->vlan_mode;
2630 bundle->use_priority_tags = s->use_priority_tags;
2635 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2636 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2638 if (vlan != bundle->vlan) {
2639 bundle->vlan = vlan;
2643 /* Get trunked VLANs. */
2644 switch (s->vlan_mode) {
2645 case PORT_VLAN_ACCESS:
2649 case PORT_VLAN_TRUNK:
2650 trunks = CONST_CAST(unsigned long *, s->trunks);
2653 case PORT_VLAN_NATIVE_UNTAGGED:
2654 case PORT_VLAN_NATIVE_TAGGED:
2655 if (vlan != 0 && (!s->trunks
2656 || !bitmap_is_set(s->trunks, vlan)
2657 || bitmap_is_set(s->trunks, 0))) {
2658 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2660 trunks = bitmap_clone(s->trunks, 4096);
2662 trunks = bitmap_allocate1(4096);
2664 bitmap_set1(trunks, vlan);
2665 bitmap_set0(trunks, 0);
2667 trunks = CONST_CAST(unsigned long *, s->trunks);
2674 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2675 free(bundle->trunks);
2676 if (trunks == s->trunks) {
2677 bundle->trunks = vlan_bitmap_clone(trunks);
2679 bundle->trunks = trunks;
2684 if (trunks != s->trunks) {
2689 if (!list_is_short(&bundle->ports)) {
2690 bundle->ofproto->has_bonded_bundles = true;
2692 if (bond_reconfigure(bundle->bond, s->bond)) {
2693 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2696 bundle->bond = bond_create(s->bond);
2697 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2700 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2701 bond_slave_register(bundle->bond, port, port->up.netdev);
2704 bond_destroy(bundle->bond);
2705 bundle->bond = NULL;
2708 /* If we changed something that would affect MAC learning, un-learn
2709 * everything on this port and force flow revalidation. */
2711 bundle_flush_macs(bundle, false);
2718 bundle_remove(struct ofport *port_)
2720 struct ofport_dpif *port = ofport_dpif_cast(port_);
2721 struct ofbundle *bundle = port->bundle;
2724 bundle_del_port(port);
2725 if (list_is_empty(&bundle->ports)) {
2726 bundle_destroy(bundle);
2727 } else if (list_is_short(&bundle->ports)) {
2728 bond_destroy(bundle->bond);
2729 bundle->bond = NULL;
2735 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2738 struct ofport_dpif *port = port_;
2739 uint8_t ea[ETH_ADDR_LEN];
2742 error = netdev_get_etheraddr(port->up.netdev, ea);
2744 struct ofpbuf packet;
2747 ofpbuf_init(&packet, 0);
2748 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2750 memcpy(packet_pdu, pdu, pdu_size);
2752 send_packet(port, &packet);
2753 ofpbuf_uninit(&packet);
2755 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2756 "%s (%s)", port->bundle->name,
2757 netdev_get_name(port->up.netdev), strerror(error));
2762 bundle_send_learning_packets(struct ofbundle *bundle)
2764 struct ofproto_dpif *ofproto = bundle->ofproto;
2765 int error, n_packets, n_errors;
2766 struct mac_entry *e;
2768 error = n_packets = n_errors = 0;
2769 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2770 if (e->port.p != bundle) {
2771 struct ofpbuf *learning_packet;
2772 struct ofport_dpif *port;
2776 /* The assignment to "port" is unnecessary but makes "grep"ing for
2777 * struct ofport_dpif more effective. */
2778 learning_packet = bond_compose_learning_packet(bundle->bond,
2782 ret = send_packet(port, learning_packet);
2783 ofpbuf_delete(learning_packet);
2793 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2794 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2795 "packets, last error was: %s",
2796 bundle->name, n_errors, n_packets, strerror(error));
2798 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2799 bundle->name, n_packets);
2804 bundle_run(struct ofbundle *bundle)
2807 lacp_run(bundle->lacp, send_pdu_cb);
2810 struct ofport_dpif *port;
2812 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2813 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2816 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2817 lacp_status(bundle->lacp));
2818 if (bond_should_send_learning_packets(bundle->bond)) {
2819 bundle_send_learning_packets(bundle);
2825 bundle_wait(struct ofbundle *bundle)
2828 lacp_wait(bundle->lacp);
2831 bond_wait(bundle->bond);
2838 mirror_scan(struct ofproto_dpif *ofproto)
2842 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2843 if (!ofproto->mirrors[idx]) {
2850 static struct ofmirror *
2851 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2855 for (i = 0; i < MAX_MIRRORS; i++) {
2856 struct ofmirror *mirror = ofproto->mirrors[i];
2857 if (mirror && mirror->aux == aux) {
2865 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2867 mirror_update_dups(struct ofproto_dpif *ofproto)
2871 for (i = 0; i < MAX_MIRRORS; i++) {
2872 struct ofmirror *m = ofproto->mirrors[i];
2875 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2879 for (i = 0; i < MAX_MIRRORS; i++) {
2880 struct ofmirror *m1 = ofproto->mirrors[i];
2887 for (j = i + 1; j < MAX_MIRRORS; j++) {
2888 struct ofmirror *m2 = ofproto->mirrors[j];
2890 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2891 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2892 m2->dup_mirrors |= m1->dup_mirrors;
2899 mirror_set(struct ofproto *ofproto_, void *aux,
2900 const struct ofproto_mirror_settings *s)
2902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2903 mirror_mask_t mirror_bit;
2904 struct ofbundle *bundle;
2905 struct ofmirror *mirror;
2906 struct ofbundle *out;
2907 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2908 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2911 mirror = mirror_lookup(ofproto, aux);
2913 mirror_destroy(mirror);
2919 idx = mirror_scan(ofproto);
2921 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2923 ofproto->up.name, MAX_MIRRORS, s->name);
2927 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2928 mirror->ofproto = ofproto;
2931 mirror->out_vlan = -1;
2932 mirror->name = NULL;
2935 if (!mirror->name || strcmp(s->name, mirror->name)) {
2937 mirror->name = xstrdup(s->name);
2940 /* Get the new configuration. */
2941 if (s->out_bundle) {
2942 out = bundle_lookup(ofproto, s->out_bundle);
2944 mirror_destroy(mirror);
2950 out_vlan = s->out_vlan;
2952 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2953 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2955 /* If the configuration has not changed, do nothing. */
2956 if (hmapx_equals(&srcs, &mirror->srcs)
2957 && hmapx_equals(&dsts, &mirror->dsts)
2958 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2959 && mirror->out == out
2960 && mirror->out_vlan == out_vlan)
2962 hmapx_destroy(&srcs);
2963 hmapx_destroy(&dsts);
2967 hmapx_swap(&srcs, &mirror->srcs);
2968 hmapx_destroy(&srcs);
2970 hmapx_swap(&dsts, &mirror->dsts);
2971 hmapx_destroy(&dsts);
2973 free(mirror->vlans);
2974 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2977 mirror->out_vlan = out_vlan;
2979 /* Update bundles. */
2980 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2981 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2982 if (hmapx_contains(&mirror->srcs, bundle)) {
2983 bundle->src_mirrors |= mirror_bit;
2985 bundle->src_mirrors &= ~mirror_bit;
2988 if (hmapx_contains(&mirror->dsts, bundle)) {
2989 bundle->dst_mirrors |= mirror_bit;
2991 bundle->dst_mirrors &= ~mirror_bit;
2994 if (mirror->out == bundle) {
2995 bundle->mirror_out |= mirror_bit;
2997 bundle->mirror_out &= ~mirror_bit;
3001 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3002 ofproto->has_mirrors = true;
3003 mac_learning_flush(ofproto->ml,
3004 &ofproto->backer->revalidate_set);
3005 mirror_update_dups(ofproto);
3011 mirror_destroy(struct ofmirror *mirror)
3013 struct ofproto_dpif *ofproto;
3014 mirror_mask_t mirror_bit;
3015 struct ofbundle *bundle;
3022 ofproto = mirror->ofproto;
3023 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3024 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3026 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3027 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3028 bundle->src_mirrors &= ~mirror_bit;
3029 bundle->dst_mirrors &= ~mirror_bit;
3030 bundle->mirror_out &= ~mirror_bit;
3033 hmapx_destroy(&mirror->srcs);
3034 hmapx_destroy(&mirror->dsts);
3035 free(mirror->vlans);
3037 ofproto->mirrors[mirror->idx] = NULL;
3041 mirror_update_dups(ofproto);
3043 ofproto->has_mirrors = false;
3044 for (i = 0; i < MAX_MIRRORS; i++) {
3045 if (ofproto->mirrors[i]) {
3046 ofproto->has_mirrors = true;
3053 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3054 uint64_t *packets, uint64_t *bytes)
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3057 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3060 *packets = *bytes = UINT64_MAX;
3066 *packets = mirror->packet_count;
3067 *bytes = mirror->byte_count;
3073 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3076 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3077 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3083 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3086 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3087 return bundle && bundle->mirror_out != 0;
3091 forward_bpdu_changed(struct ofproto *ofproto_)
3093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3094 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3098 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3101 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102 mac_learning_set_idle_time(ofproto->ml, idle_time);
3103 mac_learning_set_max_entries(ofproto->ml, max_entries);
3108 static struct ofport_dpif *
3109 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3111 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3112 return ofport ? ofport_dpif_cast(ofport) : NULL;
3115 static struct ofport_dpif *
3116 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3118 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3119 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3123 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3124 struct ofproto_port *ofproto_port,
3125 struct dpif_port *dpif_port)
3127 ofproto_port->name = dpif_port->name;
3128 ofproto_port->type = dpif_port->type;
3129 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3132 static struct ofport_dpif *
3133 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3135 const struct ofproto_dpif *ofproto;
3138 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3143 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3144 struct ofport *ofport;
3146 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3147 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3148 return ofport_dpif_cast(ofport);
3155 port_run_fast(struct ofport_dpif *ofport)
3157 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3158 struct ofpbuf packet;
3160 ofpbuf_init(&packet, 0);
3161 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3162 send_packet(ofport, &packet);
3163 ofpbuf_uninit(&packet);
3166 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3167 struct ofpbuf packet;
3169 ofpbuf_init(&packet, 0);
3170 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3171 send_packet(ofport, &packet);
3172 ofpbuf_uninit(&packet);
3177 port_run(struct ofport_dpif *ofport)
3179 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3180 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3181 bool enable = netdev_get_carrier(ofport->up.netdev);
3183 ofport->carrier_seq = carrier_seq;
3185 port_run_fast(ofport);
3187 if (ofport->tnl_port
3188 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3189 &ofport->tnl_port)) {
3190 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3194 int cfm_opup = cfm_get_opup(ofport->cfm);
3196 cfm_run(ofport->cfm);
3197 enable = enable && !cfm_get_fault(ofport->cfm);
3199 if (cfm_opup >= 0) {
3200 enable = enable && cfm_opup;
3205 bfd_run(ofport->bfd);
3206 enable = enable && bfd_forwarding(ofport->bfd);
3209 if (ofport->bundle) {
3210 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3211 if (carrier_changed) {
3212 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3216 if (ofport->may_enable != enable) {
3217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3219 if (ofproto->has_bundle_action) {
3220 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3224 ofport->may_enable = enable;
3228 port_wait(struct ofport_dpif *ofport)
3231 cfm_wait(ofport->cfm);
3235 bfd_wait(ofport->bfd);
3240 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3241 struct ofproto_port *ofproto_port)
3243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3244 struct dpif_port dpif_port;
3247 if (sset_contains(&ofproto->ghost_ports, devname)) {
3248 const char *type = netdev_get_type_from_name(devname);
3250 /* We may be called before ofproto->up.port_by_name is populated with
3251 * the appropriate ofport. For this reason, we must get the name and
3252 * type from the netdev layer directly. */
3254 const struct ofport *ofport;
3256 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3257 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3258 ofproto_port->name = xstrdup(devname);
3259 ofproto_port->type = xstrdup(type);
3265 if (!sset_contains(&ofproto->ports, devname)) {
3268 error = dpif_port_query_by_name(ofproto->backer->dpif,
3269 devname, &dpif_port);
3271 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3277 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3279 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3280 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3281 const char *devname = netdev_get_name(netdev);
3283 if (netdev_vport_is_patch(netdev)) {
3284 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3288 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3289 uint32_t port_no = UINT32_MAX;
3292 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3296 if (netdev_get_tunnel_config(netdev)) {
3297 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3301 if (netdev_get_tunnel_config(netdev)) {
3302 sset_add(&ofproto->ghost_ports, devname);
3304 sset_add(&ofproto->ports, devname);
3310 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3312 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3313 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3320 sset_find_and_delete(&ofproto->ghost_ports,
3321 netdev_get_name(ofport->up.netdev));
3322 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3323 if (!ofport->tnl_port) {
3324 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3326 /* The caller is going to close ofport->up.netdev. If this is a
3327 * bonded port, then the bond is using that netdev, so remove it
3328 * from the bond. The client will need to reconfigure everything
3329 * after deleting ports, so then the slave will get re-added. */
3330 bundle_remove(&ofport->up);
3337 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3339 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3344 error = netdev_get_stats(ofport->up.netdev, stats);
3346 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3349 /* ofproto->stats.tx_packets represents packets that we created
3350 * internally and sent to some port (e.g. packets sent with
3351 * send_packet()). Account for them as if they had come from
3352 * OFPP_LOCAL and got forwarded. */
3354 if (stats->rx_packets != UINT64_MAX) {
3355 stats->rx_packets += ofproto->stats.tx_packets;
3358 if (stats->rx_bytes != UINT64_MAX) {
3359 stats->rx_bytes += ofproto->stats.tx_bytes;
3362 /* ofproto->stats.rx_packets represents packets that were received on
3363 * some port and we processed internally and dropped (e.g. STP).
3364 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3366 if (stats->tx_packets != UINT64_MAX) {
3367 stats->tx_packets += ofproto->stats.rx_packets;
3370 if (stats->tx_bytes != UINT64_MAX) {
3371 stats->tx_bytes += ofproto->stats.rx_bytes;
3378 /* Account packets for LOCAL port. */
3380 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3381 size_t tx_size, size_t rx_size)
3383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3386 ofproto->stats.rx_packets++;
3387 ofproto->stats.rx_bytes += rx_size;
3390 ofproto->stats.tx_packets++;
3391 ofproto->stats.tx_bytes += tx_size;
3395 struct port_dump_state {
3400 struct ofproto_port port;
3405 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3407 *statep = xzalloc(sizeof(struct port_dump_state));
3412 port_dump_next(const struct ofproto *ofproto_, void *state_,
3413 struct ofproto_port *port)
3415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3416 struct port_dump_state *state = state_;
3417 const struct sset *sset;
3418 struct sset_node *node;
3420 if (state->has_port) {
3421 ofproto_port_destroy(&state->port);
3422 state->has_port = false;
3424 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3425 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3428 error = port_query_by_name(ofproto_, node->name, &state->port);
3430 *port = state->port;
3431 state->has_port = true;
3433 } else if (error != ENODEV) {
3438 if (!state->ghost) {
3439 state->ghost = true;
3442 return port_dump_next(ofproto_, state_, port);
3449 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3451 struct port_dump_state *state = state_;
3453 if (state->has_port) {
3454 ofproto_port_destroy(&state->port);
3461 port_poll(const struct ofproto *ofproto_, char **devnamep)
3463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3465 if (ofproto->port_poll_errno) {
3466 int error = ofproto->port_poll_errno;
3467 ofproto->port_poll_errno = 0;
3471 if (sset_is_empty(&ofproto->port_poll_set)) {
3475 *devnamep = sset_pop(&ofproto->port_poll_set);
3480 port_poll_wait(const struct ofproto *ofproto_)
3482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3483 dpif_port_poll_wait(ofproto->backer->dpif);
3487 port_is_lacp_current(const struct ofport *ofport_)
3489 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3490 return (ofport->bundle && ofport->bundle->lacp
3491 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3495 /* Upcall handling. */
3497 /* Flow miss batching.
3499 * Some dpifs implement operations faster when you hand them off in a batch.
3500 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3501 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3502 * more packets, plus possibly installing the flow in the dpif.
3504 * So far we only batch the operations that affect flow setup time the most.
3505 * It's possible to batch more than that, but the benefit might be minimal. */
3507 struct hmap_node hmap_node;
3508 struct ofproto_dpif *ofproto;
3510 enum odp_key_fitness key_fitness;
3511 const struct nlattr *key;
3513 struct initial_vals initial_vals;
3514 struct list packets;
3515 enum dpif_upcall_type upcall_type;
3516 uint32_t odp_in_port;
3519 struct flow_miss_op {
3520 struct dpif_op dpif_op;
3521 void *garbage; /* Pointer to pass to free(), NULL if none. */
3522 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3525 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3526 * OpenFlow controller as necessary according to their individual
3527 * configurations. */
3529 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3530 const struct flow *flow)
3532 struct ofputil_packet_in pin;
3534 pin.packet = packet->data;
3535 pin.packet_len = packet->size;
3536 pin.reason = OFPR_NO_MATCH;
3537 pin.controller_id = 0;
3542 pin.send_len = 0; /* not used for flow table misses */
3544 flow_get_metadata(flow, &pin.fmd);
3546 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3549 static enum slow_path_reason
3550 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3551 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3555 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3557 cfm_process_heartbeat(ofport->cfm, packet);
3560 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3562 bfd_process_packet(ofport->bfd, flow, packet);
3565 } else if (ofport->bundle && ofport->bundle->lacp
3566 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3568 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3571 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3573 stp_process_packet(ofport, packet);
3581 static struct flow_miss *
3582 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3583 const struct flow *flow, uint32_t hash)
3585 struct flow_miss *miss;
3587 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3588 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3596 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3597 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3598 * 'miss' is associated with a subfacet the caller must also initialize the
3599 * returned op->subfacet, and if anything needs to be freed after processing
3600 * the op, the caller must initialize op->garbage also. */
3602 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3603 struct flow_miss_op *op)
3605 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3606 /* This packet was received on a VLAN splinter port. We
3607 * added a VLAN to the packet to make the packet resemble
3608 * the flow, but the actions were composed assuming that
3609 * the packet contained no VLAN. So, we must remove the
3610 * VLAN header from the packet before trying to execute the
3612 eth_pop_vlan(packet);
3616 op->dpif_op.type = DPIF_OP_EXECUTE;
3617 op->dpif_op.u.execute.key = miss->key;
3618 op->dpif_op.u.execute.key_len = miss->key_len;
3619 op->dpif_op.u.execute.packet = packet;
3622 /* Helper for handle_flow_miss_without_facet() and
3623 * handle_flow_miss_with_facet(). */
3625 handle_flow_miss_common(struct rule_dpif *rule,
3626 struct ofpbuf *packet, const struct flow *flow)
3628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3630 ofproto->n_matches++;
3632 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3634 * Extra-special case for fail-open mode.
3636 * We are in fail-open mode and the packet matched the fail-open
3637 * rule, but we are connected to a controller too. We should send
3638 * the packet up to the controller in the hope that it will try to
3639 * set up a flow and thereby allow us to exit fail-open.
3641 * See the top-level comment in fail-open.c for more information.
3643 send_packet_in_miss(ofproto, packet, flow);
3647 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3648 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3649 * installing a datapath flow. The answer is usually "yes" (a return value of
3650 * true). However, for short flows the cost of bookkeeping is much higher than
3651 * the benefits, so when the datapath holds a large number of flows we impose
3652 * some heuristics to decide which flows are likely to be worth tracking. */
3654 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3655 struct flow_miss *miss, uint32_t hash)
3657 if (!ofproto->governor) {
3660 n_subfacets = hmap_count(&ofproto->subfacets);
3661 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3665 ofproto->governor = governor_create(ofproto->up.name);
3668 return governor_should_install_flow(ofproto->governor, hash,
3669 list_size(&miss->packets));
3672 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3673 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3674 * increment '*n_ops'. */
3676 handle_flow_miss_without_facet(struct flow_miss *miss,
3677 struct rule_dpif *rule,
3678 struct flow_miss_op *ops, size_t *n_ops)
3680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3681 long long int now = time_msec();
3682 struct action_xlate_ctx ctx;
3683 struct ofpbuf *packet;
3685 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3686 struct flow_miss_op *op = &ops[*n_ops];
3687 struct dpif_flow_stats stats;
3688 struct ofpbuf odp_actions;
3690 COVERAGE_INC(facet_suppress);
3692 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3694 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3695 rule_credit_stats(rule, &stats);
3697 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3698 &miss->initial_vals, rule, 0, packet);
3699 ctx.resubmit_stats = &stats;
3700 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3703 if (odp_actions.size) {
3704 struct dpif_execute *execute = &op->dpif_op.u.execute;
3706 init_flow_miss_execute_op(miss, packet, op);
3707 execute->actions = odp_actions.data;
3708 execute->actions_len = odp_actions.size;
3709 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3713 ofpbuf_uninit(&odp_actions);
3718 /* Handles 'miss', which matches 'facet'. May add any required datapath
3719 * operations to 'ops', incrementing '*n_ops' for each new op.
3721 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3722 * This is really important only for new facets: if we just called time_msec()
3723 * here, then the new subfacet or its packets could look (occasionally) as
3724 * though it was used some time after the facet was used. That can make a
3725 * one-packet flow look like it has a nonzero duration, which looks odd in
3726 * e.g. NetFlow statistics. */
3728 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3730 struct flow_miss_op *ops, size_t *n_ops)
3732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3733 enum subfacet_path want_path;
3734 struct subfacet *subfacet;
3735 struct ofpbuf *packet;
3737 subfacet = subfacet_create(facet, miss, now);
3739 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3740 struct flow_miss_op *op = &ops[*n_ops];
3741 struct dpif_flow_stats stats;
3742 struct ofpbuf odp_actions;
3744 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3746 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3747 if (!subfacet->actions || subfacet->slow) {
3748 subfacet_make_actions(subfacet, packet, &odp_actions);
3751 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3752 subfacet_update_stats(subfacet, &stats);
3754 if (subfacet->actions_len) {
3755 struct dpif_execute *execute = &op->dpif_op.u.execute;
3757 init_flow_miss_execute_op(miss, packet, op);
3758 if (!subfacet->slow) {
3759 execute->actions = subfacet->actions;
3760 execute->actions_len = subfacet->actions_len;
3761 ofpbuf_uninit(&odp_actions);
3763 execute->actions = odp_actions.data;
3764 execute->actions_len = odp_actions.size;
3765 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3770 ofpbuf_uninit(&odp_actions);
3774 want_path = subfacet_want_path(subfacet->slow);
3775 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3776 struct flow_miss_op *op = &ops[(*n_ops)++];
3777 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3779 subfacet->path = want_path;
3782 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3783 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3784 put->key = miss->key;
3785 put->key_len = miss->key_len;
3786 if (want_path == SF_FAST_PATH) {
3787 put->actions = subfacet->actions;
3788 put->actions_len = subfacet->actions_len;
3790 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3791 op->stub, sizeof op->stub,
3792 &put->actions, &put->actions_len);
3798 /* Handles flow miss 'miss'. May add any required datapath operations
3799 * to 'ops', incrementing '*n_ops' for each new op. */
3801 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3804 struct ofproto_dpif *ofproto = miss->ofproto;
3805 struct facet *facet;
3809 /* The caller must ensure that miss->hmap_node.hash contains
3810 * flow_hash(miss->flow, 0). */
3811 hash = miss->hmap_node.hash;
3813 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3815 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3817 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3818 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3822 facet = facet_create(rule, &miss->flow, hash);
3827 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3830 static struct drop_key *
3831 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3834 struct drop_key *drop_key;
3836 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3837 &backer->drop_keys) {
3838 if (drop_key->key_len == key_len
3839 && !memcmp(drop_key->key, key, key_len)) {
3847 drop_key_clear(struct dpif_backer *backer)
3849 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3850 struct drop_key *drop_key, *next;
3852 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3855 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3857 if (error && !VLOG_DROP_WARN(&rl)) {
3858 struct ds ds = DS_EMPTY_INITIALIZER;
3859 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3860 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3865 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3866 free(drop_key->key);
3871 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3872 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3873 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3874 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3875 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3876 * 'packet' ingressed.
3878 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3879 * 'flow''s in_port to OFPP_NONE.
3881 * This function does post-processing on data returned from
3882 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3883 * of the upcall processing logic. In particular, if the extracted in_port is
3884 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3885 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3886 * a VLAN header onto 'packet' (if it is nonnull).
3888 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3889 * to the VLAN TCI with which the packet was really received, that is, the
3890 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3891 * the value returned in flow->vlan_tci only for packets received on
3894 * Similarly, this function also includes some logic to help with tunnels. It
3895 * may modify 'flow' as necessary to make the tunneling implementation
3896 * transparent to the upcall processing logic.
3898 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3899 * or some other positive errno if there are other problems. */
3901 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3902 const struct nlattr *key, size_t key_len,
3903 struct flow *flow, enum odp_key_fitness *fitnessp,
3904 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3905 struct initial_vals *initial_vals)
3907 const struct ofport_dpif *port;
3908 enum odp_key_fitness fitness;
3911 fitness = odp_flow_key_to_flow(key, key_len, flow);
3912 if (fitness == ODP_FIT_ERROR) {
3918 initial_vals->vlan_tci = flow->vlan_tci;
3922 *odp_in_port = flow->in_port;
3925 if (tnl_port_should_receive(flow)) {
3926 const struct ofport *ofport = tnl_port_receive(flow);
3928 flow->in_port = OFPP_NONE;
3931 port = ofport_dpif_cast(ofport);
3933 /* We can't reproduce 'key' from 'flow'. */
3934 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3936 /* XXX: Since the tunnel module is not scoped per backer, it's
3937 * theoretically possible that we'll receive an ofport belonging to an
3938 * entirely different datapath. In practice, this can't happen because
3939 * no platforms has two separate datapaths which each support
3941 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3943 port = odp_port_to_ofport(backer, flow->in_port);
3945 flow->in_port = OFPP_NONE;
3949 flow->in_port = port->up.ofp_port;
3950 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3952 /* Make the packet resemble the flow, so that it gets sent to
3953 * an OpenFlow controller properly, so that it looks correct
3954 * for sFlow, and so that flow_extract() will get the correct
3955 * vlan_tci if it is called on 'packet'.
3957 * The allocated space inside 'packet' probably also contains
3958 * 'key', that is, both 'packet' and 'key' are probably part of
3959 * a struct dpif_upcall (see the large comment on that
3960 * structure definition), so pushing data on 'packet' is in
3961 * general not a good idea since it could overwrite 'key' or
3962 * free it as a side effect. However, it's OK in this special
3963 * case because we know that 'packet' is inside a Netlink
3964 * attribute: pushing 4 bytes will just overwrite the 4-byte
3965 * "struct nlattr", which is fine since we don't need that
3966 * header anymore. */
3967 eth_push_vlan(packet, flow->vlan_tci);
3969 /* We can't reproduce 'key' from 'flow'. */
3970 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3976 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3981 *fitnessp = fitness;
3987 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3990 struct dpif_upcall *upcall;
3991 struct flow_miss *miss;
3992 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3993 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3994 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4004 /* Construct the to-do list.
4006 * This just amounts to extracting the flow from each packet and sticking
4007 * the packets that have the same flow in the same "flow_miss" structure so
4008 * that we can process them together. */
4011 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4012 struct flow_miss *miss = &misses[n_misses];
4013 struct flow_miss *existing_miss;
4014 struct ofproto_dpif *ofproto;
4015 uint32_t odp_in_port;
4020 error = ofproto_receive(backer, upcall->packet, upcall->key,
4021 upcall->key_len, &flow, &miss->key_fitness,
4022 &ofproto, &odp_in_port, &miss->initial_vals);
4023 if (error == ENODEV) {
4024 struct drop_key *drop_key;
4026 /* Received packet on port for which we couldn't associate
4027 * an ofproto. This can happen if a port is removed while
4028 * traffic is being received. Print a rate-limited message
4029 * in case it happens frequently. Install a drop flow so
4030 * that future packets of the flow are inexpensively dropped
4032 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4035 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4037 drop_key = xmalloc(sizeof *drop_key);
4038 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4039 drop_key->key_len = upcall->key_len;
4041 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4042 hash_bytes(drop_key->key, drop_key->key_len, 0));
4043 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4044 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4052 ofproto->n_missed++;
4053 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4054 &flow.tunnel, flow.in_port, &miss->flow);
4056 /* Add other packets to a to-do list. */
4057 hash = flow_hash(&miss->flow, 0);
4058 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4059 if (!existing_miss) {
4060 hmap_insert(&todo, &miss->hmap_node, hash);
4061 miss->ofproto = ofproto;
4062 miss->key = upcall->key;
4063 miss->key_len = upcall->key_len;
4064 miss->upcall_type = upcall->type;
4065 miss->odp_in_port = odp_in_port;
4066 list_init(&miss->packets);
4070 miss = existing_miss;
4072 list_push_back(&miss->packets, &upcall->packet->list_node);
4075 /* Process each element in the to-do list, constructing the set of
4076 * operations to batch. */
4078 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4079 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4081 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4083 /* Execute batch. */
4084 for (i = 0; i < n_ops; i++) {
4085 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4087 dpif_operate(backer->dpif, dpif_ops, n_ops);
4090 for (i = 0; i < n_ops; i++) {
4091 free(flow_miss_ops[i].garbage);
4093 hmap_destroy(&todo);
4096 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4098 classify_upcall(const struct dpif_upcall *upcall)
4100 size_t userdata_len;
4101 union user_action_cookie cookie;
4103 /* First look at the upcall type. */
4104 switch (upcall->type) {
4105 case DPIF_UC_ACTION:
4111 case DPIF_N_UC_TYPES:
4113 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4117 /* "action" upcalls need a closer look. */
4118 if (!upcall->userdata) {
4119 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4122 userdata_len = nl_attr_get_size(upcall->userdata);
4123 if (userdata_len < sizeof cookie.type
4124 || userdata_len > sizeof cookie) {
4125 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4129 memset(&cookie, 0, sizeof cookie);
4130 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4131 if (userdata_len == sizeof cookie.sflow
4132 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4133 return SFLOW_UPCALL;
4134 } else if (userdata_len == sizeof cookie.slow_path
4135 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4137 } else if (userdata_len == sizeof cookie.flow_sample
4138 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4139 return FLOW_SAMPLE_UPCALL;
4140 } else if (userdata_len == sizeof cookie.ipfix
4141 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4142 return IPFIX_UPCALL;
4144 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4145 " and size %zu", cookie.type, userdata_len);
4151 handle_sflow_upcall(struct dpif_backer *backer,
4152 const struct dpif_upcall *upcall)
4154 struct ofproto_dpif *ofproto;
4155 union user_action_cookie cookie;
4157 uint32_t odp_in_port;
4159 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4160 &flow, NULL, &ofproto, &odp_in_port, NULL)
4161 || !ofproto->sflow) {
4165 memset(&cookie, 0, sizeof cookie);
4166 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4167 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4168 odp_in_port, &cookie);
4172 handle_flow_sample_upcall(struct dpif_backer *backer,
4173 const struct dpif_upcall *upcall)
4175 struct ofproto_dpif *ofproto;
4176 union user_action_cookie cookie;
4179 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4180 &flow, NULL, &ofproto, NULL, NULL)
4181 || !ofproto->ipfix) {
4185 memset(&cookie, 0, sizeof cookie);
4186 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4188 /* The flow reflects exactly the contents of the packet. Sample
4189 * the packet using it. */
4190 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4191 cookie.flow_sample.collector_set_id,
4192 cookie.flow_sample.probability,
4193 cookie.flow_sample.obs_domain_id,
4194 cookie.flow_sample.obs_point_id);
4198 handle_ipfix_upcall(struct dpif_backer *backer,
4199 const struct dpif_upcall *upcall)
4201 struct ofproto_dpif *ofproto;
4204 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4205 &flow, NULL, &ofproto, NULL, NULL)
4206 || !ofproto->ipfix) {
4210 /* The flow reflects exactly the contents of the packet. Sample
4211 * the packet using it. */
4212 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4216 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4218 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4219 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4220 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4225 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4228 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4229 struct dpif_upcall *upcall = &misses[n_misses];
4230 struct ofpbuf *buf = &miss_bufs[n_misses];
4233 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4234 sizeof miss_buf_stubs[n_misses]);
4235 error = dpif_recv(backer->dpif, upcall, buf);
4241 switch (classify_upcall(upcall)) {
4243 /* Handle it later. */
4248 handle_sflow_upcall(backer, upcall);
4252 case FLOW_SAMPLE_UPCALL:
4253 handle_flow_sample_upcall(backer, upcall);
4258 handle_ipfix_upcall(backer, upcall);
4268 /* Handle deferred MISS_UPCALL processing. */
4269 handle_miss_upcalls(backer, misses, n_misses);
4270 for (i = 0; i < n_misses; i++) {
4271 ofpbuf_uninit(&miss_bufs[i]);
4277 /* Flow expiration. */
4279 static int subfacet_max_idle(const struct ofproto_dpif *);
4280 static void update_stats(struct dpif_backer *);
4281 static void rule_expire(struct rule_dpif *);
4282 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4284 /* This function is called periodically by run(). Its job is to collect
4285 * updates for the flows that have been installed into the datapath, most
4286 * importantly when they last were used, and then use that information to
4287 * expire flows that have not been used recently.
4289 * Returns the number of milliseconds after which it should be called again. */
4291 expire(struct dpif_backer *backer)
4293 struct ofproto_dpif *ofproto;
4294 int max_idle = INT32_MAX;
4296 /* Periodically clear out the drop keys in an effort to keep them
4297 * relatively few. */
4298 drop_key_clear(backer);
4300 /* Update stats for each flow in the backer. */
4301 update_stats(backer);
4303 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4304 struct rule *rule, *next_rule;
4307 if (ofproto->backer != backer) {
4311 /* Keep track of the max number of flows per ofproto_dpif. */
4312 update_max_subfacet_count(ofproto);
4314 /* Expire subfacets that have been idle too long. */
4315 dp_max_idle = subfacet_max_idle(ofproto);
4316 expire_subfacets(ofproto, dp_max_idle);
4318 max_idle = MIN(max_idle, dp_max_idle);
4320 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4322 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4323 &ofproto->up.expirable) {
4324 rule_expire(rule_dpif_cast(rule));
4327 /* All outstanding data in existing flows has been accounted, so it's a
4328 * good time to do bond rebalancing. */
4329 if (ofproto->has_bonded_bundles) {
4330 struct ofbundle *bundle;
4332 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4334 bond_rebalance(bundle->bond, &backer->revalidate_set);
4340 return MIN(max_idle, 1000);
4343 /* Updates flow table statistics given that the datapath just reported 'stats'
4344 * as 'subfacet''s statistics. */
4346 update_subfacet_stats(struct subfacet *subfacet,
4347 const struct dpif_flow_stats *stats)
4349 struct facet *facet = subfacet->facet;
4351 if (stats->n_packets >= subfacet->dp_packet_count) {
4352 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4353 facet->packet_count += extra;
4355 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4358 if (stats->n_bytes >= subfacet->dp_byte_count) {
4359 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4361 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4364 subfacet->dp_packet_count = stats->n_packets;
4365 subfacet->dp_byte_count = stats->n_bytes;
4367 facet->tcp_flags |= stats->tcp_flags;
4369 subfacet_update_time(subfacet, stats->used);
4370 if (facet->accounted_bytes < facet->byte_count) {
4372 facet_account(facet);
4373 facet->accounted_bytes = facet->byte_count;
4377 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4378 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4380 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4381 const struct nlattr *key, size_t key_len)
4383 if (!VLOG_DROP_WARN(&rl)) {
4387 odp_flow_key_format(key, key_len, &s);
4388 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4392 COVERAGE_INC(facet_unexpected);
4393 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4396 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4398 * This function also pushes statistics updates to rules which each facet
4399 * resubmits into. Generally these statistics will be accurate. However, if a
4400 * facet changes the rule it resubmits into at some time in between
4401 * update_stats() runs, it is possible that statistics accrued to the
4402 * old rule will be incorrectly attributed to the new rule. This could be
4403 * avoided by calling update_stats() whenever rules are created or
4404 * deleted. However, the performance impact of making so many calls to the
4405 * datapath do not justify the benefit of having perfectly accurate statistics.
4407 * In addition, this function maintains per ofproto flow hit counts. The patch
4408 * port is not treated specially. e.g. A packet ingress from br0 patched into
4409 * br1 will increase the hit count of br0 by 1, however, does not affect
4410 * the hit or miss counts of br1.
4413 update_stats(struct dpif_backer *backer)
4415 const struct dpif_flow_stats *stats;
4416 struct dpif_flow_dump dump;
4417 const struct nlattr *key;
4418 struct ofproto_dpif *ofproto;
4421 dpif_flow_dump_start(&dump, backer->dpif);
4422 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4424 struct subfacet *subfacet;
4425 struct ofport_dpif *ofport;
4428 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4433 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4434 ofproto->n_update_stats++;
4436 ofport = get_ofp_port(ofproto, flow.in_port);
4437 if (ofport && ofport->tnl_port) {
4438 netdev_vport_inc_rx(ofport->up.netdev, stats);
4441 key_hash = odp_flow_key_hash(key, key_len);
4442 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4443 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4445 /* Update ofproto_dpif's hit count. */
4446 if (stats->n_packets > subfacet->dp_packet_count) {
4447 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4448 dpif_stats_update_hit_count(ofproto, delta);
4451 update_subfacet_stats(subfacet, stats);
4455 /* Stats are updated per-packet. */
4458 case SF_NOT_INSTALLED:
4460 delete_unexpected_flow(ofproto, key, key_len);
4465 dpif_flow_dump_done(&dump);
4467 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4468 update_moving_averages(ofproto);
4473 /* Calculates and returns the number of milliseconds of idle time after which
4474 * subfacets should expire from the datapath. When a subfacet expires, we fold
4475 * its statistics into its facet, and when a facet's last subfacet expires, we
4476 * fold its statistic into its rule. */
4478 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4481 * Idle time histogram.
4483 * Most of the time a switch has a relatively small number of subfacets.
4484 * When this is the case we might as well keep statistics for all of them
4485 * in userspace and to cache them in the kernel datapath for performance as
4488 * As the number of subfacets increases, the memory required to maintain
4489 * statistics about them in userspace and in the kernel becomes
4490 * significant. However, with a large number of subfacets it is likely
4491 * that only a few of them are "heavy hitters" that consume a large amount
4492 * of bandwidth. At this point, only heavy hitters are worth caching in
4493 * the kernel and maintaining in userspaces; other subfacets we can
4496 * The technique used to compute the idle time is to build a histogram with
4497 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4498 * that is installed in the kernel gets dropped in the appropriate bucket.
4499 * After the histogram has been built, we compute the cutoff so that only
4500 * the most-recently-used 1% of subfacets (but at least
4501 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4502 * the most-recently-used bucket of subfacets is kept, so actually an
4503 * arbitrary number of subfacets can be kept in any given expiration run
4504 * (though the next run will delete most of those unless they receive
4507 * This requires a second pass through the subfacets, in addition to the
4508 * pass made by update_stats(), because the former function never looks at
4509 * uninstallable subfacets.
4511 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4512 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4513 int buckets[N_BUCKETS] = { 0 };
4514 int total, subtotal, bucket;
4515 struct subfacet *subfacet;
4519 total = hmap_count(&ofproto->subfacets);
4520 if (total <= ofproto->up.flow_eviction_threshold) {
4521 return N_BUCKETS * BUCKET_WIDTH;
4524 /* Build histogram. */
4526 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4527 long long int idle = now - subfacet->used;
4528 int bucket = (idle <= 0 ? 0
4529 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4530 : (unsigned int) idle / BUCKET_WIDTH);
4534 /* Find the first bucket whose flows should be expired. */
4535 subtotal = bucket = 0;
4537 subtotal += buckets[bucket++];
4538 } while (bucket < N_BUCKETS &&
4539 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4541 if (VLOG_IS_DBG_ENABLED()) {
4545 ds_put_cstr(&s, "keep");
4546 for (i = 0; i < N_BUCKETS; i++) {
4548 ds_put_cstr(&s, ", drop");
4551 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4554 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4558 return bucket * BUCKET_WIDTH;
4562 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4564 /* Cutoff time for most flows. */
4565 long long int normal_cutoff = time_msec() - dp_max_idle;
4567 /* We really want to keep flows for special protocols around, so use a more
4568 * conservative cutoff. */
4569 long long int special_cutoff = time_msec() - 10000;
4571 struct subfacet *subfacet, *next_subfacet;
4572 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4576 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4577 &ofproto->subfacets) {
4578 long long int cutoff;
4580 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
4583 if (subfacet->used < cutoff) {
4584 if (subfacet->path != SF_NOT_INSTALLED) {
4585 batch[n_batch++] = subfacet;
4586 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4587 subfacet_destroy_batch(ofproto, batch, n_batch);
4591 subfacet_destroy(subfacet);
4597 subfacet_destroy_batch(ofproto, batch, n_batch);
4601 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4602 * then delete it entirely. */
4604 rule_expire(struct rule_dpif *rule)
4606 struct facet *facet, *next_facet;
4610 if (rule->up.pending) {
4611 /* We'll have to expire it later. */
4615 /* Has 'rule' expired? */
4617 if (rule->up.hard_timeout
4618 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4619 reason = OFPRR_HARD_TIMEOUT;
4620 } else if (rule->up.idle_timeout
4621 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4622 reason = OFPRR_IDLE_TIMEOUT;
4627 COVERAGE_INC(ofproto_dpif_expired);
4629 /* Update stats. (This is a no-op if the rule expired due to an idle
4630 * timeout, because that only happens when the rule has no facets left.) */
4631 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4632 facet_remove(facet);
4635 /* Get rid of the rule. */
4636 ofproto_rule_expire(&rule->up, reason);
4641 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4643 * The caller must already have determined that no facet with an identical
4644 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4645 * the ofproto's classifier table.
4647 * 'hash' must be the return value of flow_hash(flow, 0).
4649 * The facet will initially have no subfacets. The caller should create (at
4650 * least) one subfacet with subfacet_create(). */
4651 static struct facet *
4652 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4655 struct facet *facet;
4657 facet = xzalloc(sizeof *facet);
4658 facet->used = time_msec();
4659 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4660 list_push_back(&rule->facets, &facet->list_node);
4662 facet->flow = *flow;
4663 list_init(&facet->subfacets);
4664 netflow_flow_init(&facet->nf_flow);
4665 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4667 facet->learn_rl = time_msec() + 500;
4673 facet_free(struct facet *facet)
4678 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4679 * 'packet', which arrived on 'in_port'. */
4681 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4682 const struct nlattr *odp_actions, size_t actions_len,
4683 struct ofpbuf *packet)
4685 struct odputil_keybuf keybuf;
4689 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4690 odp_flow_key_from_flow(&key, flow,
4691 ofp_port_to_odp_port(ofproto, flow->in_port));
4693 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4694 odp_actions, actions_len, packet);
4698 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4700 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4701 * rule's statistics, via subfacet_uninstall().
4703 * - Removes 'facet' from its rule and from ofproto->facets.
4706 facet_remove(struct facet *facet)
4708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4709 struct subfacet *subfacet, *next_subfacet;
4711 ovs_assert(!list_is_empty(&facet->subfacets));
4713 /* First uninstall all of the subfacets to get final statistics. */
4714 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4715 subfacet_uninstall(subfacet);
4718 /* Flush the final stats to the rule.
4720 * This might require us to have at least one subfacet around so that we
4721 * can use its actions for accounting in facet_account(), which is why we
4722 * have uninstalled but not yet destroyed the subfacets. */
4723 facet_flush_stats(facet);
4725 /* Now we're really all done so destroy everything. */
4726 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4727 &facet->subfacets) {
4728 subfacet_destroy__(subfacet);
4730 hmap_remove(&ofproto->facets, &facet->hmap_node);
4731 list_remove(&facet->list_node);
4735 /* Feed information from 'facet' back into the learning table to keep it in
4736 * sync with what is actually flowing through the datapath. */
4738 facet_learn(struct facet *facet)
4740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4741 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4742 struct subfacet, list_node);
4743 long long int now = time_msec();
4744 struct action_xlate_ctx ctx;
4746 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4750 facet->learn_rl = now + 500;
4752 if (!facet->has_learn
4753 && !facet->has_normal
4754 && (!facet->has_fin_timeout
4755 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4759 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4760 &subfacet->initial_vals,
4761 facet->rule, facet->tcp_flags, NULL);
4762 ctx.may_learn = true;
4763 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4764 facet->rule->up.ofpacts_len);
4768 facet_account(struct facet *facet)
4770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4771 struct subfacet *subfacet = facet_get_subfacet(facet);
4772 const struct nlattr *a;
4777 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4780 n_bytes = facet->byte_count - facet->accounted_bytes;
4782 /* This loop feeds byte counters to bond_account() for rebalancing to use
4783 * as a basis. We also need to track the actual VLAN on which the packet
4784 * is going to be sent to ensure that it matches the one passed to
4785 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4788 * We use the actions from an arbitrary subfacet because they should all
4789 * be equally valid for our purpose. */
4790 vlan_tci = facet->flow.vlan_tci;
4791 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4792 subfacet->actions, subfacet->actions_len) {
4793 const struct ovs_action_push_vlan *vlan;
4794 struct ofport_dpif *port;
4796 switch (nl_attr_type(a)) {
4797 case OVS_ACTION_ATTR_OUTPUT:
4798 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4799 if (port && port->bundle && port->bundle->bond) {
4800 bond_account(port->bundle->bond, &facet->flow,
4801 vlan_tci_to_vid(vlan_tci), n_bytes);
4805 case OVS_ACTION_ATTR_POP_VLAN:
4806 vlan_tci = htons(0);
4809 case OVS_ACTION_ATTR_PUSH_VLAN:
4810 vlan = nl_attr_get(a);
4811 vlan_tci = vlan->vlan_tci;
4817 /* Returns true if the only action for 'facet' is to send to the controller.
4818 * (We don't report NetFlow expiration messages for such facets because they
4819 * are just part of the control logic for the network, not real traffic). */
4821 facet_is_controller_flow(struct facet *facet)
4824 const struct rule *rule = &facet->rule->up;
4825 const struct ofpact *ofpacts = rule->ofpacts;
4826 size_t ofpacts_len = rule->ofpacts_len;
4828 if (ofpacts_len > 0 &&
4829 ofpacts->type == OFPACT_CONTROLLER &&
4830 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4837 /* Folds all of 'facet''s statistics into its rule. Also updates the
4838 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4839 * 'facet''s statistics in the datapath should have been zeroed and folded into
4840 * its packet and byte counts before this function is called. */
4842 facet_flush_stats(struct facet *facet)
4844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4845 struct subfacet *subfacet;
4847 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4848 ovs_assert(!subfacet->dp_byte_count);
4849 ovs_assert(!subfacet->dp_packet_count);
4852 facet_push_stats(facet);
4853 if (facet->accounted_bytes < facet->byte_count) {
4854 facet_account(facet);
4855 facet->accounted_bytes = facet->byte_count;
4858 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4859 struct ofexpired expired;
4860 expired.flow = facet->flow;
4861 expired.packet_count = facet->packet_count;
4862 expired.byte_count = facet->byte_count;
4863 expired.used = facet->used;
4864 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4867 facet->rule->packet_count += facet->packet_count;
4868 facet->rule->byte_count += facet->byte_count;
4870 /* Reset counters to prevent double counting if 'facet' ever gets
4872 facet_reset_counters(facet);
4874 netflow_flow_clear(&facet->nf_flow);
4875 facet->tcp_flags = 0;
4878 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4879 * Returns it if found, otherwise a null pointer.
4881 * 'hash' must be the return value of flow_hash(flow, 0).
4883 * The returned facet might need revalidation; use facet_lookup_valid()
4884 * instead if that is important. */
4885 static struct facet *
4886 facet_find(struct ofproto_dpif *ofproto,
4887 const struct flow *flow, uint32_t hash)
4889 struct facet *facet;
4891 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4892 if (flow_equal(flow, &facet->flow)) {
4900 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4901 * Returns it if found, otherwise a null pointer.
4903 * 'hash' must be the return value of flow_hash(flow, 0).
4905 * The returned facet is guaranteed to be valid. */
4906 static struct facet *
4907 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4910 struct facet *facet;
4912 facet = facet_find(ofproto, flow, hash);
4914 && (ofproto->backer->need_revalidate
4915 || tag_set_intersects(&ofproto->backer->revalidate_set,
4917 facet_revalidate(facet);
4919 /* facet_revalidate() may have destroyed 'facet'. */
4920 facet = facet_find(ofproto, flow, hash);
4926 /* Return a subfacet from 'facet'. A facet consists of one or more
4927 * subfacets, and this function returns one of them. */
4928 static struct subfacet *facet_get_subfacet(struct facet *facet)
4930 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4935 subfacet_path_to_string(enum subfacet_path path)
4938 case SF_NOT_INSTALLED:
4939 return "not installed";
4941 return "in fast path";
4943 return "in slow path";
4949 /* Returns the path in which a subfacet should be installed if its 'slow'
4950 * member has the specified value. */
4951 static enum subfacet_path
4952 subfacet_want_path(enum slow_path_reason slow)
4954 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4957 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4958 * supposing that its actions have been recalculated as 'want_actions' and that
4959 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4961 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4962 const struct ofpbuf *want_actions)
4964 enum subfacet_path want_path = subfacet_want_path(slow);
4965 return (want_path != subfacet->path
4966 || (want_path == SF_FAST_PATH
4967 && (subfacet->actions_len != want_actions->size
4968 || memcmp(subfacet->actions, want_actions->data,
4969 subfacet->actions_len))));
4973 facet_check_consistency(struct facet *facet)
4975 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4979 uint64_t odp_actions_stub[1024 / 8];
4980 struct ofpbuf odp_actions;
4982 struct rule_dpif *rule;
4983 struct subfacet *subfacet;
4984 bool may_log = false;
4987 /* Check the rule for consistency. */
4988 rule = rule_dpif_lookup(ofproto, &facet->flow);
4989 ok = rule == facet->rule;
4991 may_log = !VLOG_DROP_WARN(&rl);
4996 flow_format(&s, &facet->flow);
4997 ds_put_format(&s, ": facet associated with wrong rule (was "
4998 "table=%"PRIu8",", facet->rule->up.table_id);
4999 cls_rule_format(&facet->rule->up.cr, &s);
5000 ds_put_format(&s, ") (should have been table=%"PRIu8",",
5002 cls_rule_format(&rule->up.cr, &s);
5003 ds_put_char(&s, ')');
5005 VLOG_WARN("%s", ds_cstr(&s));
5010 /* Check the datapath actions for consistency. */
5011 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5012 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5013 enum subfacet_path want_path;
5014 struct action_xlate_ctx ctx;
5017 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5018 &subfacet->initial_vals, rule, 0, NULL);
5019 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
5022 if (subfacet->path == SF_NOT_INSTALLED) {
5023 /* This only happens if the datapath reported an error when we
5024 * tried to install the flow. Don't flag another error here. */
5028 want_path = subfacet_want_path(subfacet->slow);
5029 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
5030 /* The actions for slow-path flows may legitimately vary from one
5031 * packet to the next. We're done. */
5035 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
5039 /* Inconsistency! */
5041 may_log = !VLOG_DROP_WARN(&rl);
5045 /* Rate-limited, skip reporting. */
5050 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5052 ds_put_cstr(&s, ": inconsistency in subfacet");
5053 if (want_path != subfacet->path) {
5054 enum odp_key_fitness fitness = subfacet->key_fitness;
5056 ds_put_format(&s, " (%s, fitness=%s)",
5057 subfacet_path_to_string(subfacet->path),
5058 odp_key_fitness_to_string(fitness));
5059 ds_put_format(&s, " (should have been %s)",
5060 subfacet_path_to_string(want_path));
5061 } else if (want_path == SF_FAST_PATH) {
5062 ds_put_cstr(&s, " (actions were: ");
5063 format_odp_actions(&s, subfacet->actions,
5064 subfacet->actions_len);
5065 ds_put_cstr(&s, ") (correct actions: ");
5066 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5067 ds_put_char(&s, ')');
5069 ds_put_cstr(&s, " (actions: ");
5070 format_odp_actions(&s, subfacet->actions,
5071 subfacet->actions_len);
5072 ds_put_char(&s, ')');
5074 VLOG_WARN("%s", ds_cstr(&s));
5077 ofpbuf_uninit(&odp_actions);
5082 /* Re-searches the classifier for 'facet':
5084 * - If the rule found is different from 'facet''s current rule, moves
5085 * 'facet' to the new rule and recompiles its actions.
5087 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5088 * where it is and recompiles its actions anyway.
5090 * - If any of 'facet''s subfacets correspond to a new flow according to
5091 * ofproto_receive(), 'facet' is removed. */
5093 facet_revalidate(struct facet *facet)
5095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5097 struct nlattr *odp_actions;
5100 struct actions *new_actions;
5102 struct action_xlate_ctx ctx;
5103 uint64_t odp_actions_stub[1024 / 8];
5104 struct ofpbuf odp_actions;
5106 struct rule_dpif *new_rule;
5107 struct subfacet *subfacet;
5110 COVERAGE_INC(facet_revalidate);
5112 /* Check that child subfacets still correspond to this facet. Tunnel
5113 * configuration changes could cause a subfacet's OpenFlow in_port to
5115 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5116 struct ofproto_dpif *recv_ofproto;
5117 struct flow recv_flow;
5120 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5121 subfacet->key_len, &recv_flow, NULL,
5122 &recv_ofproto, NULL, NULL);
5124 || recv_ofproto != ofproto
5125 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5126 facet_remove(facet);
5131 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5133 /* Calculate new datapath actions.
5135 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5136 * emit a NetFlow expiration and, if so, we need to have the old state
5137 * around to properly compose it. */
5139 /* If the datapath actions changed or the installability changed,
5140 * then we need to talk to the datapath. */
5143 memset(&ctx, 0, sizeof ctx);
5144 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5145 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5146 enum slow_path_reason slow;
5148 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5149 &subfacet->initial_vals, new_rule, 0, NULL);
5150 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5153 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5154 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5155 struct dpif_flow_stats stats;
5157 subfacet_install(subfacet,
5158 odp_actions.data, odp_actions.size, &stats, slow);
5159 subfacet_update_stats(subfacet, &stats);
5162 new_actions = xcalloc(list_size(&facet->subfacets),
5163 sizeof *new_actions);
5165 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5167 new_actions[i].actions_len = odp_actions.size;
5172 ofpbuf_uninit(&odp_actions);
5175 facet_flush_stats(facet);
5178 /* Update 'facet' now that we've taken care of all the old state. */
5179 facet->tags = ctx.tags;
5180 facet->nf_flow.output_iface = ctx.nf_output_iface;
5181 facet->has_learn = ctx.has_learn;
5182 facet->has_normal = ctx.has_normal;
5183 facet->has_fin_timeout = ctx.has_fin_timeout;
5184 facet->mirrors = ctx.mirrors;
5187 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5188 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5190 if (new_actions && new_actions[i].odp_actions) {
5191 free(subfacet->actions);
5192 subfacet->actions = new_actions[i].odp_actions;
5193 subfacet->actions_len = new_actions[i].actions_len;
5199 if (facet->rule != new_rule) {
5200 COVERAGE_INC(facet_changed_rule);
5201 list_remove(&facet->list_node);
5202 list_push_back(&new_rule->facets, &facet->list_node);
5203 facet->rule = new_rule;
5204 facet->used = new_rule->up.created;
5205 facet->prev_used = facet->used;
5209 /* Updates 'facet''s used time. Caller is responsible for calling
5210 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5212 facet_update_time(struct facet *facet, long long int used)
5214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5215 if (used > facet->used) {
5217 ofproto_rule_update_used(&facet->rule->up, used);
5218 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5223 facet_reset_counters(struct facet *facet)
5225 facet->packet_count = 0;
5226 facet->byte_count = 0;
5227 facet->prev_packet_count = 0;
5228 facet->prev_byte_count = 0;
5229 facet->accounted_bytes = 0;
5233 facet_push_stats(struct facet *facet)
5235 struct dpif_flow_stats stats;
5237 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5238 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5239 ovs_assert(facet->used >= facet->prev_used);
5241 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5242 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5243 stats.used = facet->used;
5244 stats.tcp_flags = 0;
5246 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5247 facet->prev_packet_count = facet->packet_count;
5248 facet->prev_byte_count = facet->byte_count;
5249 facet->prev_used = facet->used;
5251 flow_push_stats(facet, &stats);
5253 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5254 facet->mirrors, stats.n_packets, stats.n_bytes);
5259 push_all_stats__(bool run_fast)
5261 static long long int rl = LLONG_MIN;
5262 struct ofproto_dpif *ofproto;
5264 if (time_msec() < rl) {
5268 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5269 struct facet *facet;
5271 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5272 facet_push_stats(facet);
5279 rl = time_msec() + 100;
5283 push_all_stats(void)
5285 push_all_stats__(true);
5289 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5291 rule->packet_count += stats->n_packets;
5292 rule->byte_count += stats->n_bytes;
5293 ofproto_rule_update_used(&rule->up, stats->used);
5296 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5297 * into given 'facet->rule''s actions and mirrors. */
5299 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5301 struct rule_dpif *rule = facet->rule;
5302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5303 struct subfacet *subfacet = facet_get_subfacet(facet);
5304 struct action_xlate_ctx ctx;
5306 ofproto_rule_update_used(&rule->up, stats->used);
5308 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5309 &subfacet->initial_vals, rule, 0, NULL);
5310 ctx.resubmit_stats = stats;
5311 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5312 rule->up.ofpacts_len);
5317 static struct subfacet *
5318 subfacet_find(struct ofproto_dpif *ofproto,
5319 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5321 struct subfacet *subfacet;
5323 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5324 &ofproto->subfacets) {
5325 if (subfacet->key_len == key_len
5326 && !memcmp(key, subfacet->key, key_len)) {
5334 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5335 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5336 * existing subfacet if there is one, otherwise creates and returns a
5339 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5340 * which case the caller must populate the actions with
5341 * subfacet_make_actions(). */
5342 static struct subfacet *
5343 subfacet_create(struct facet *facet, struct flow_miss *miss,
5346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5347 enum odp_key_fitness key_fitness = miss->key_fitness;
5348 const struct nlattr *key = miss->key;
5349 size_t key_len = miss->key_len;
5351 struct subfacet *subfacet;
5353 key_hash = odp_flow_key_hash(key, key_len);
5355 if (list_is_empty(&facet->subfacets)) {
5356 subfacet = &facet->one_subfacet;
5358 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5360 if (subfacet->facet == facet) {
5364 /* This shouldn't happen. */
5365 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5366 subfacet_destroy(subfacet);
5369 subfacet = xmalloc(sizeof *subfacet);
5372 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5373 list_push_back(&facet->subfacets, &subfacet->list_node);
5374 subfacet->facet = facet;
5375 subfacet->key_fitness = key_fitness;
5376 subfacet->key = xmemdup(key, key_len);
5377 subfacet->key_len = key_len;
5378 subfacet->used = now;
5379 subfacet->created = now;
5380 subfacet->dp_packet_count = 0;
5381 subfacet->dp_byte_count = 0;
5382 subfacet->actions_len = 0;
5383 subfacet->actions = NULL;
5384 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5387 subfacet->path = SF_NOT_INSTALLED;
5388 subfacet->initial_vals = miss->initial_vals;
5389 subfacet->odp_in_port = miss->odp_in_port;
5391 ofproto->subfacet_add_count++;
5395 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5396 * its facet within 'ofproto', and frees it. */
5398 subfacet_destroy__(struct subfacet *subfacet)
5400 struct facet *facet = subfacet->facet;
5401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5403 /* Update ofproto stats before uninstall the subfacet. */
5404 ofproto->subfacet_del_count++;
5405 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5407 subfacet_uninstall(subfacet);
5408 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5409 list_remove(&subfacet->list_node);
5410 free(subfacet->key);
5411 free(subfacet->actions);
5412 if (subfacet != &facet->one_subfacet) {
5417 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5418 * last remaining subfacet in its facet destroys the facet too. */
5420 subfacet_destroy(struct subfacet *subfacet)
5422 struct facet *facet = subfacet->facet;
5424 if (list_is_singleton(&facet->subfacets)) {
5425 /* facet_remove() needs at least one subfacet (it will remove it). */
5426 facet_remove(facet);
5428 subfacet_destroy__(subfacet);
5433 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5434 struct subfacet **subfacets, int n)
5436 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5437 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5438 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5441 for (i = 0; i < n; i++) {
5442 ops[i].type = DPIF_OP_FLOW_DEL;
5443 ops[i].u.flow_del.key = subfacets[i]->key;
5444 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5445 ops[i].u.flow_del.stats = &stats[i];
5449 dpif_operate(ofproto->backer->dpif, opsp, n);
5450 for (i = 0; i < n; i++) {
5451 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5452 subfacets[i]->path = SF_NOT_INSTALLED;
5453 subfacet_destroy(subfacets[i]);
5458 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5459 * Translates the actions into 'odp_actions', which the caller must have
5460 * initialized and is responsible for uninitializing. */
5462 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5463 struct ofpbuf *odp_actions)
5465 struct facet *facet = subfacet->facet;
5466 struct rule_dpif *rule = facet->rule;
5467 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5469 struct action_xlate_ctx ctx;
5471 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5472 &subfacet->initial_vals, rule, 0, packet);
5473 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5474 facet->tags = ctx.tags;
5475 facet->has_learn = ctx.has_learn;
5476 facet->has_normal = ctx.has_normal;
5477 facet->has_fin_timeout = ctx.has_fin_timeout;
5478 facet->nf_flow.output_iface = ctx.nf_output_iface;
5479 facet->mirrors = ctx.mirrors;
5481 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5482 if (subfacet->actions_len != odp_actions->size
5483 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5484 free(subfacet->actions);
5485 subfacet->actions_len = odp_actions->size;
5486 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5490 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5491 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5492 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5493 * since 'subfacet' was last updated.
5495 * Returns 0 if successful, otherwise a positive errno value. */
5497 subfacet_install(struct subfacet *subfacet,
5498 const struct nlattr *actions, size_t actions_len,
5499 struct dpif_flow_stats *stats,
5500 enum slow_path_reason slow)
5502 struct facet *facet = subfacet->facet;
5503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5504 enum subfacet_path path = subfacet_want_path(slow);
5505 uint64_t slow_path_stub[128 / 8];
5506 enum dpif_flow_put_flags flags;
5509 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5511 flags |= DPIF_FP_ZERO_STATS;
5514 if (path == SF_SLOW_PATH) {
5515 compose_slow_path(ofproto, &facet->flow, slow,
5516 slow_path_stub, sizeof slow_path_stub,
5517 &actions, &actions_len);
5520 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5521 subfacet->key_len, actions, actions_len, stats);
5524 subfacet_reset_dp_stats(subfacet, stats);
5528 subfacet->path = path;
5534 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5536 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5537 stats, subfacet->slow);
5540 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5542 subfacet_uninstall(struct subfacet *subfacet)
5544 if (subfacet->path != SF_NOT_INSTALLED) {
5545 struct rule_dpif *rule = subfacet->facet->rule;
5546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5547 struct dpif_flow_stats stats;
5550 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5551 subfacet->key_len, &stats);
5552 subfacet_reset_dp_stats(subfacet, &stats);
5554 subfacet_update_stats(subfacet, &stats);
5556 subfacet->path = SF_NOT_INSTALLED;
5558 ovs_assert(subfacet->dp_packet_count == 0);
5559 ovs_assert(subfacet->dp_byte_count == 0);
5563 /* Resets 'subfacet''s datapath statistics counters. This should be called
5564 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5565 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5566 * was reset in the datapath. 'stats' will be modified to include only
5567 * statistics new since 'subfacet' was last updated. */
5569 subfacet_reset_dp_stats(struct subfacet *subfacet,
5570 struct dpif_flow_stats *stats)
5573 && subfacet->dp_packet_count <= stats->n_packets
5574 && subfacet->dp_byte_count <= stats->n_bytes) {
5575 stats->n_packets -= subfacet->dp_packet_count;
5576 stats->n_bytes -= subfacet->dp_byte_count;
5579 subfacet->dp_packet_count = 0;
5580 subfacet->dp_byte_count = 0;
5583 /* Updates 'subfacet''s used time. The caller is responsible for calling
5584 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5586 subfacet_update_time(struct subfacet *subfacet, long long int used)
5588 if (used > subfacet->used) {
5589 subfacet->used = used;
5590 facet_update_time(subfacet->facet, used);
5594 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5596 * Because of the meaning of a subfacet's counters, it only makes sense to do
5597 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5598 * represents a packet that was sent by hand or if it represents statistics
5599 * that have been cleared out of the datapath. */
5601 subfacet_update_stats(struct subfacet *subfacet,
5602 const struct dpif_flow_stats *stats)
5604 if (stats->n_packets || stats->used > subfacet->used) {
5605 struct facet *facet = subfacet->facet;
5607 subfacet_update_time(subfacet, stats->used);
5608 facet->packet_count += stats->n_packets;
5609 facet->byte_count += stats->n_bytes;
5610 facet->tcp_flags |= stats->tcp_flags;
5611 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5617 static struct rule_dpif *
5618 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5620 struct rule_dpif *rule;
5622 rule = rule_dpif_lookup__(ofproto, flow, 0);
5627 return rule_dpif_miss_rule(ofproto, flow);
5630 static struct rule_dpif *
5631 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5634 struct cls_rule *cls_rule;
5635 struct classifier *cls;
5637 if (table_id >= N_TABLES) {
5641 cls = &ofproto->up.tables[table_id].cls;
5642 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5643 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5644 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5645 * are unavailable. */
5646 struct flow ofpc_normal_flow = *flow;
5647 ofpc_normal_flow.tp_src = htons(0);
5648 ofpc_normal_flow.tp_dst = htons(0);
5649 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5651 cls_rule = classifier_lookup(cls, flow);
5653 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5656 static struct rule_dpif *
5657 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5659 struct ofport_dpif *port;
5661 port = get_ofp_port(ofproto, flow->in_port);
5663 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5664 return ofproto->miss_rule;
5667 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5668 return ofproto->no_packet_in_rule;
5670 return ofproto->miss_rule;
5674 complete_operation(struct rule_dpif *rule)
5676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5678 rule_invalidate(rule);
5680 struct dpif_completion *c = xmalloc(sizeof *c);
5681 c->op = rule->up.pending;
5682 list_push_back(&ofproto->completions, &c->list_node);
5684 ofoperation_complete(rule->up.pending, 0);
5688 static struct rule *
5691 struct rule_dpif *rule = xmalloc(sizeof *rule);
5696 rule_dealloc(struct rule *rule_)
5698 struct rule_dpif *rule = rule_dpif_cast(rule_);
5703 rule_construct(struct rule *rule_)
5705 struct rule_dpif *rule = rule_dpif_cast(rule_);
5706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5707 struct rule_dpif *victim;
5710 rule->packet_count = 0;
5711 rule->byte_count = 0;
5713 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5714 if (victim && !list_is_empty(&victim->facets)) {
5715 struct facet *facet;
5717 rule->facets = victim->facets;
5718 list_moved(&rule->facets);
5719 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5720 /* XXX: We're only clearing our local counters here. It's possible
5721 * that quite a few packets are unaccounted for in the datapath
5722 * statistics. These will be accounted to the new rule instead of
5723 * cleared as required. This could be fixed by clearing out the
5724 * datapath statistics for this facet, but currently it doesn't
5726 facet_reset_counters(facet);
5730 /* Must avoid list_moved() in this case. */
5731 list_init(&rule->facets);
5734 table_id = rule->up.table_id;
5736 rule->tag = victim->tag;
5737 } else if (table_id == 0) {
5742 miniflow_expand(&rule->up.cr.match.flow, &flow);
5743 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5744 ofproto->tables[table_id].basis);
5747 complete_operation(rule);
5752 rule_destruct(struct rule *rule_)
5754 struct rule_dpif *rule = rule_dpif_cast(rule_);
5755 struct facet *facet, *next_facet;
5757 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5758 facet_revalidate(facet);
5761 complete_operation(rule);
5765 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5767 struct rule_dpif *rule = rule_dpif_cast(rule_);
5768 struct facet *facet;
5770 /* push_all_stats() can handle flow misses which, when using the learn
5771 * action, can cause rules to be added and deleted. This can corrupt our
5772 * caller's datastructures which assume that rule_get_stats() doesn't have
5773 * an impact on the flow table. To be safe, we disable miss handling. */
5774 push_all_stats__(false);
5776 /* Start from historical data for 'rule' itself that are no longer tracked
5777 * in facets. This counts, for example, facets that have expired. */
5778 *packets = rule->packet_count;
5779 *bytes = rule->byte_count;
5781 /* Add any statistics that are tracked by facets. This includes
5782 * statistical data recently updated by ofproto_update_stats() as well as
5783 * stats for packets that were executed "by hand" via dpif_execute(). */
5784 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5785 *packets += facet->packet_count;
5786 *bytes += facet->byte_count;
5791 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5792 struct ofpbuf *packet)
5794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5795 struct initial_vals initial_vals;
5796 struct dpif_flow_stats stats;
5797 struct action_xlate_ctx ctx;
5798 uint64_t odp_actions_stub[1024 / 8];
5799 struct ofpbuf odp_actions;
5801 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5802 rule_credit_stats(rule, &stats);
5804 initial_vals.vlan_tci = flow->vlan_tci;
5805 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5806 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5807 rule, stats.tcp_flags, packet);
5808 ctx.resubmit_stats = &stats;
5809 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5811 execute_odp_actions(ofproto, flow, odp_actions.data,
5812 odp_actions.size, packet);
5814 ofpbuf_uninit(&odp_actions);
5818 rule_execute(struct rule *rule, const struct flow *flow,
5819 struct ofpbuf *packet)
5821 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5822 ofpbuf_delete(packet);
5827 rule_modify_actions(struct rule *rule_)
5829 struct rule_dpif *rule = rule_dpif_cast(rule_);
5831 complete_operation(rule);
5834 /* Sends 'packet' out 'ofport'.
5835 * May modify 'packet'.
5836 * Returns 0 if successful, otherwise a positive errno value. */
5838 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5840 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5841 uint64_t odp_actions_stub[1024 / 8];
5842 struct ofpbuf key, odp_actions;
5843 struct odputil_keybuf keybuf;
5848 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5849 if (netdev_vport_is_patch(ofport->up.netdev)) {
5850 struct ofproto_dpif *peer_ofproto;
5851 struct dpif_flow_stats stats;
5852 struct ofport_dpif *peer;
5853 struct rule_dpif *rule;
5855 peer = ofport_get_peer(ofport);
5860 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5861 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5862 netdev_vport_inc_rx(peer->up.netdev, &stats);
5864 flow.in_port = peer->up.ofp_port;
5865 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5866 rule = rule_dpif_lookup(peer_ofproto, &flow);
5867 rule_dpif_execute(rule, &flow, packet);
5872 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5874 if (ofport->tnl_port) {
5875 struct dpif_flow_stats stats;
5877 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5878 if (odp_port == OVSP_NONE) {
5882 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5883 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5884 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5885 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5887 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5889 if (odp_port != ofport->odp_port) {
5890 eth_pop_vlan(packet);
5891 flow.vlan_tci = htons(0);
5895 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5896 odp_flow_key_from_flow(&key, &flow,
5897 ofp_port_to_odp_port(ofproto, flow.in_port));
5899 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5900 compose_ipfix_action(ofproto, &odp_actions, &flow);
5902 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5903 error = dpif_execute(ofproto->backer->dpif,
5905 odp_actions.data, odp_actions.size,
5907 ofpbuf_uninit(&odp_actions);
5910 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5911 ofproto->up.name, odp_port, strerror(error));
5913 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5917 /* OpenFlow to datapath action translation. */
5919 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5920 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5921 struct action_xlate_ctx *);
5922 static void xlate_normal(struct action_xlate_ctx *);
5924 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5925 * The action will state 'slow' as the reason that the action is in the slow
5926 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5927 * dump-flows" output to see why a flow is in the slow path.)
5929 * The 'stub_size' bytes in 'stub' will be used to store the action.
5930 * 'stub_size' must be large enough for the action.
5932 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5935 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5936 enum slow_path_reason slow,
5937 uint64_t *stub, size_t stub_size,
5938 const struct nlattr **actionsp, size_t *actions_lenp)
5940 union user_action_cookie cookie;
5943 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5944 cookie.slow_path.unused = 0;
5945 cookie.slow_path.reason = slow;
5947 ofpbuf_use_stack(&buf, stub, stub_size);
5948 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5949 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5950 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5952 put_userspace_action(ofproto, &buf, flow, &cookie,
5953 sizeof cookie.slow_path);
5955 *actionsp = buf.data;
5956 *actions_lenp = buf.size;
5960 put_userspace_action(const struct ofproto_dpif *ofproto,
5961 struct ofpbuf *odp_actions,
5962 const struct flow *flow,
5963 const union user_action_cookie *cookie,
5964 const size_t cookie_size)
5968 pid = dpif_port_get_pid(ofproto->backer->dpif,
5969 ofp_port_to_odp_port(ofproto, flow->in_port));
5971 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5974 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5975 * the number of packets out of UINT32_MAX to sample. The given
5976 * cookie is passed back in the callback for each sampled packet.
5979 compose_sample_action(const struct ofproto_dpif *ofproto,
5980 struct ofpbuf *odp_actions,
5981 const struct flow *flow,
5982 const uint32_t probability,
5983 const union user_action_cookie *cookie,
5984 const size_t cookie_size)
5986 size_t sample_offset, actions_offset;
5989 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5991 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5993 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5994 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5997 nl_msg_end_nested(odp_actions, actions_offset);
5998 nl_msg_end_nested(odp_actions, sample_offset);
5999 return cookie_offset;
6003 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
6004 ovs_be16 vlan_tci, uint32_t odp_port,
6005 unsigned int n_outputs, union user_action_cookie *cookie)
6009 cookie->type = USER_ACTION_COOKIE_SFLOW;
6010 cookie->sflow.vlan_tci = vlan_tci;
6012 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
6013 * port information") for the interpretation of cookie->output. */
6014 switch (n_outputs) {
6016 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
6017 cookie->sflow.output = 0x40000000 | 256;
6021 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
6023 cookie->sflow.output = ifindex;
6028 /* 0x80000000 means "multiple output ports. */
6029 cookie->sflow.output = 0x80000000 | n_outputs;
6034 /* Compose SAMPLE action for sFlow bridge sampling. */
6036 compose_sflow_action(const struct ofproto_dpif *ofproto,
6037 struct ofpbuf *odp_actions,
6038 const struct flow *flow,
6041 uint32_t probability;
6042 union user_action_cookie cookie;
6044 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
6048 probability = dpif_sflow_get_probability(ofproto->sflow);
6049 compose_sflow_cookie(ofproto, htons(0), odp_port,
6050 odp_port == OVSP_NONE ? 0 : 1, &cookie);
6052 return compose_sample_action(ofproto, odp_actions, flow, probability,
6053 &cookie, sizeof cookie.sflow);
6057 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
6058 uint32_t obs_domain_id, uint32_t obs_point_id,
6059 union user_action_cookie *cookie)
6061 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6062 cookie->flow_sample.probability = probability;
6063 cookie->flow_sample.collector_set_id = collector_set_id;
6064 cookie->flow_sample.obs_domain_id = obs_domain_id;
6065 cookie->flow_sample.obs_point_id = obs_point_id;
6069 compose_ipfix_cookie(union user_action_cookie *cookie)
6071 cookie->type = USER_ACTION_COOKIE_IPFIX;
6074 /* Compose SAMPLE action for IPFIX bridge sampling. */
6076 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6077 struct ofpbuf *odp_actions,
6078 const struct flow *flow)
6080 uint32_t probability;
6081 union user_action_cookie cookie;
6083 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6087 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6088 compose_ipfix_cookie(&cookie);
6090 compose_sample_action(ofproto, odp_actions, flow, probability,
6091 &cookie, sizeof cookie.ipfix);
6094 /* SAMPLE action for sFlow must be first action in any given list of
6095 * actions. At this point we do not have all information required to
6096 * build it. So try to build sample action as complete as possible. */
6098 add_sflow_action(struct action_xlate_ctx *ctx)
6100 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6102 &ctx->flow, OVSP_NONE);
6103 ctx->sflow_odp_port = 0;
6104 ctx->sflow_n_outputs = 0;
6107 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6108 * of actions, eventually after the SAMPLE action for sFlow. */
6110 add_ipfix_action(struct action_xlate_ctx *ctx)
6112 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6115 /* Fix SAMPLE action according to data collected while composing ODP actions.
6116 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6117 * USERSPACE action's user-cookie which is required for sflow. */
6119 fix_sflow_action(struct action_xlate_ctx *ctx)
6121 const struct flow *base = &ctx->base_flow;
6122 union user_action_cookie *cookie;
6124 if (!ctx->user_cookie_offset) {
6128 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6129 sizeof cookie->sflow);
6130 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6132 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6133 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6137 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6140 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6141 ovs_be16 flow_vlan_tci;
6142 uint32_t flow_skb_mark;
6143 uint8_t flow_nw_tos;
6144 struct priority_to_dscp *pdscp;
6145 uint32_t out_port, odp_port;
6147 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6148 * before traversing a patch port. */
6149 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6152 xlate_report(ctx, "Nonexistent output port");
6154 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6155 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6157 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6158 xlate_report(ctx, "STP not in forwarding state, skipping output");
6162 if (netdev_vport_is_patch(ofport->up.netdev)) {
6163 struct ofport_dpif *peer = ofport_get_peer(ofport);
6164 struct flow old_flow = ctx->flow;
6165 const struct ofproto_dpif *peer_ofproto;
6166 enum slow_path_reason special;
6167 struct ofport_dpif *in_port;
6170 xlate_report(ctx, "Nonexistent patch port peer");
6174 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6175 if (peer_ofproto->backer != ctx->ofproto->backer) {
6176 xlate_report(ctx, "Patch port peer on a different datapath");
6180 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6181 ctx->flow.in_port = peer->up.ofp_port;
6182 ctx->flow.metadata = htonll(0);
6183 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6184 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6186 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6187 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6190 ctx->slow |= special;
6191 } else if (!in_port || may_receive(in_port, ctx)) {
6192 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6193 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6195 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6196 * learning action look at the packet, then drop it. */
6197 struct flow old_base_flow = ctx->base_flow;
6198 size_t old_size = ctx->odp_actions->size;
6199 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6200 ctx->base_flow = old_base_flow;
6201 ctx->odp_actions->size = old_size;
6205 ctx->flow = old_flow;
6206 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6208 if (ctx->resubmit_stats) {
6209 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6210 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6216 flow_vlan_tci = ctx->flow.vlan_tci;
6217 flow_skb_mark = ctx->flow.skb_mark;
6218 flow_nw_tos = ctx->flow.nw_tos;
6220 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6222 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6223 ctx->flow.nw_tos |= pdscp->dscp;
6226 if (ofport->tnl_port) {
6227 /* Save tunnel metadata so that changes made due to
6228 * the Logical (tunnel) Port are not visible for any further
6229 * matches, while explicit set actions on tunnel metadata are.
6231 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6232 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6233 if (odp_port == OVSP_NONE) {
6234 xlate_report(ctx, "Tunneling decided against output");
6235 goto out; /* restore flow_nw_tos */
6238 if (ctx->resubmit_stats) {
6239 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6241 out_port = odp_port;
6242 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6244 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6246 odp_port = ofport->odp_port;
6247 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6248 ctx->flow.vlan_tci);
6249 if (out_port != odp_port) {
6250 ctx->flow.vlan_tci = htons(0);
6252 ctx->flow.skb_mark &= ~IPSEC_MARK;
6254 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6255 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6257 ctx->sflow_odp_port = odp_port;
6258 ctx->sflow_n_outputs++;
6259 ctx->nf_output_iface = ofp_port;
6262 ctx->flow.vlan_tci = flow_vlan_tci;
6263 ctx->flow.skb_mark = flow_skb_mark;
6265 ctx->flow.nw_tos = flow_nw_tos;
6269 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6271 compose_output_action__(ctx, ofp_port, true);
6275 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6277 struct ofproto_dpif *ofproto = ctx->ofproto;
6278 uint8_t table_id = ctx->table_id;
6280 if (table_id > 0 && table_id < N_TABLES) {
6281 struct table_dpif *table = &ofproto->tables[table_id];
6282 if (table->other_table) {
6283 ctx->tags |= (rule && rule->tag
6285 : rule_calculate_tag(&ctx->flow,
6286 &table->other_table->mask,
6292 /* Common rule processing in one place to avoid duplicating code. */
6293 static struct rule_dpif *
6294 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6297 if (ctx->resubmit_hook) {
6298 ctx->resubmit_hook(ctx, rule);
6300 if (rule == NULL && may_packet_in) {
6302 * check if table configuration flags
6303 * OFPTC_TABLE_MISS_CONTROLLER, default.
6304 * OFPTC_TABLE_MISS_CONTINUE,
6305 * OFPTC_TABLE_MISS_DROP
6306 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6308 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6310 if (rule && ctx->resubmit_stats) {
6311 rule_credit_stats(rule, ctx->resubmit_stats);
6317 xlate_table_action(struct action_xlate_ctx *ctx,
6318 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6320 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6321 struct rule_dpif *rule;
6322 uint16_t old_in_port = ctx->flow.in_port;
6323 uint8_t old_table_id = ctx->table_id;
6325 ctx->table_id = table_id;
6327 /* Look up a flow with 'in_port' as the input port. */
6328 ctx->flow.in_port = in_port;
6329 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6331 tag_the_flow(ctx, rule);
6333 /* Restore the original input port. Otherwise OFPP_NORMAL and
6334 * OFPP_IN_PORT will have surprising behavior. */
6335 ctx->flow.in_port = old_in_port;
6337 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6340 struct rule_dpif *old_rule = ctx->rule;
6344 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6345 ctx->rule = old_rule;
6349 ctx->table_id = old_table_id;
6351 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6353 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6354 MAX_RESUBMIT_RECURSION);
6355 ctx->max_resubmit_trigger = true;
6360 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6361 const struct ofpact_resubmit *resubmit)
6366 in_port = resubmit->in_port;
6367 if (in_port == OFPP_IN_PORT) {
6368 in_port = ctx->flow.in_port;
6371 table_id = resubmit->table_id;
6372 if (table_id == 255) {
6373 table_id = ctx->table_id;
6376 xlate_table_action(ctx, in_port, table_id, false);
6380 flood_packets(struct action_xlate_ctx *ctx, bool all)
6382 struct ofport_dpif *ofport;
6384 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6385 uint16_t ofp_port = ofport->up.ofp_port;
6387 if (ofp_port == ctx->flow.in_port) {
6392 compose_output_action__(ctx, ofp_port, false);
6393 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6394 compose_output_action(ctx, ofp_port);
6398 ctx->nf_output_iface = NF_OUT_FLOOD;
6402 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6403 enum ofp_packet_in_reason reason,
6404 uint16_t controller_id)
6406 struct ofputil_packet_in pin;
6407 struct ofpbuf *packet;
6409 ctx->slow |= SLOW_CONTROLLER;
6414 packet = ofpbuf_clone(ctx->packet);
6416 if (packet->l2 && packet->l3) {
6417 struct eth_header *eh;
6418 uint16_t mpls_depth;
6420 eth_pop_vlan(packet);
6423 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6424 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6426 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6427 eth_push_vlan(packet, ctx->flow.vlan_tci);
6430 mpls_depth = eth_mpls_depth(packet);
6432 if (mpls_depth < ctx->flow.mpls_depth) {
6433 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6434 } else if (mpls_depth > ctx->flow.mpls_depth) {
6435 pop_mpls(packet, ctx->flow.dl_type);
6436 } else if (mpls_depth) {
6437 set_mpls_lse(packet, ctx->flow.mpls_lse);
6441 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6442 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6443 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6447 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6448 packet_set_tcp_port(packet, ctx->flow.tp_src,
6450 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6451 packet_set_udp_port(packet, ctx->flow.tp_src,
6458 pin.packet = packet->data;
6459 pin.packet_len = packet->size;
6460 pin.reason = reason;
6461 pin.controller_id = controller_id;
6462 pin.table_id = ctx->table_id;
6463 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6466 flow_get_metadata(&ctx->flow, &pin.fmd);
6468 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6469 ofpbuf_delete(packet);
6473 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6475 ovs_assert(eth_type_mpls(eth_type));
6477 if (ctx->base_flow.mpls_depth) {
6478 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6479 ctx->flow.mpls_depth++;
6484 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6485 label = htonl(0x2); /* IPV6 Explicit Null. */
6487 label = htonl(0x0); /* IPV4 Explicit Null. */
6489 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6490 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6491 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6492 ctx->flow.mpls_depth = 1;
6494 ctx->flow.dl_type = eth_type;
6498 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6500 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6501 ovs_assert(!eth_type_mpls(eth_type));
6503 if (ctx->flow.mpls_depth) {
6504 ctx->flow.mpls_depth--;
6505 ctx->flow.mpls_lse = htonl(0);
6506 if (!ctx->flow.mpls_depth) {
6507 ctx->flow.dl_type = eth_type;
6513 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6515 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6516 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6520 if (ctx->flow.nw_ttl > 1) {
6526 for (i = 0; i < ids->n_controllers; i++) {
6527 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6531 /* Stop processing for current table. */
6537 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6539 if (!eth_type_mpls(ctx->flow.dl_type)) {
6543 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6548 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6550 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6552 if (!eth_type_mpls(ctx->flow.dl_type)) {
6558 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6561 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6563 /* Stop processing for current table. */
6569 xlate_output_action(struct action_xlate_ctx *ctx,
6570 uint16_t port, uint16_t max_len, bool may_packet_in)
6572 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6574 ctx->nf_output_iface = NF_OUT_DROP;
6578 compose_output_action(ctx, ctx->flow.in_port);
6581 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6587 flood_packets(ctx, false);
6590 flood_packets(ctx, true);
6592 case OFPP_CONTROLLER:
6593 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6599 if (port != ctx->flow.in_port) {
6600 compose_output_action(ctx, port);
6602 xlate_report(ctx, "skipping output to input port");
6607 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6608 ctx->nf_output_iface = NF_OUT_FLOOD;
6609 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6610 ctx->nf_output_iface = prev_nf_output_iface;
6611 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6612 ctx->nf_output_iface != NF_OUT_FLOOD) {
6613 ctx->nf_output_iface = NF_OUT_MULTI;
6618 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6619 const struct ofpact_output_reg *or)
6621 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6622 if (port <= UINT16_MAX) {
6623 xlate_output_action(ctx, port, or->max_len, false);
6628 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6629 const struct ofpact_enqueue *enqueue)
6631 uint16_t ofp_port = enqueue->port;
6632 uint32_t queue_id = enqueue->queue;
6633 uint32_t flow_priority, priority;
6636 /* Translate queue to priority. */
6637 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6638 queue_id, &priority);
6640 /* Fall back to ordinary output action. */
6641 xlate_output_action(ctx, enqueue->port, 0, false);
6645 /* Check output port. */
6646 if (ofp_port == OFPP_IN_PORT) {
6647 ofp_port = ctx->flow.in_port;
6648 } else if (ofp_port == ctx->flow.in_port) {
6652 /* Add datapath actions. */
6653 flow_priority = ctx->flow.skb_priority;
6654 ctx->flow.skb_priority = priority;
6655 compose_output_action(ctx, ofp_port);
6656 ctx->flow.skb_priority = flow_priority;
6658 /* Update NetFlow output port. */
6659 if (ctx->nf_output_iface == NF_OUT_DROP) {
6660 ctx->nf_output_iface = ofp_port;
6661 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6662 ctx->nf_output_iface = NF_OUT_MULTI;
6667 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6669 uint32_t skb_priority;
6671 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6672 queue_id, &skb_priority)) {
6673 ctx->flow.skb_priority = skb_priority;
6675 /* Couldn't translate queue to a priority. Nothing to do. A warning
6676 * has already been logged. */
6681 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6683 struct ofproto_dpif *ofproto = ofproto_;
6684 struct ofport_dpif *port;
6694 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6697 port = get_ofp_port(ofproto, ofp_port);
6698 return port ? port->may_enable : false;
6703 xlate_bundle_action(struct action_xlate_ctx *ctx,
6704 const struct ofpact_bundle *bundle)
6708 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6709 if (bundle->dst.field) {
6710 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6712 xlate_output_action(ctx, port, 0, false);
6717 xlate_learn_action(struct action_xlate_ctx *ctx,
6718 const struct ofpact_learn *learn)
6720 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6721 struct ofputil_flow_mod fm;
6722 uint64_t ofpacts_stub[1024 / 8];
6723 struct ofpbuf ofpacts;
6726 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6727 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6729 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6730 if (error && !VLOG_DROP_WARN(&rl)) {
6731 VLOG_WARN("learning action failed to modify flow table (%s)",
6732 ofperr_get_name(error));
6735 ofpbuf_uninit(&ofpacts);
6738 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6739 * means "infinite". */
6741 reduce_timeout(uint16_t max, uint16_t *timeout)
6743 if (max && (!*timeout || *timeout > max)) {
6749 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6750 const struct ofpact_fin_timeout *oft)
6752 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6753 struct rule_dpif *rule = ctx->rule;
6755 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6756 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6761 xlate_sample_action(struct action_xlate_ctx *ctx,
6762 const struct ofpact_sample *os)
6764 union user_action_cookie cookie;
6765 /* Scale the probability from 16-bit to 32-bit while representing
6766 * the same percentage. */
6767 uint32_t probability = (os->probability << 16) | os->probability;
6769 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6771 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6772 os->obs_domain_id, os->obs_point_id, &cookie);
6773 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6774 probability, &cookie, sizeof cookie.flow_sample);
6778 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6780 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6781 ? OFPUTIL_PC_NO_RECV_STP
6782 : OFPUTIL_PC_NO_RECV)) {
6786 /* Only drop packets here if both forwarding and learning are
6787 * disabled. If just learning is enabled, we need to have
6788 * OFPP_NORMAL and the learning action have a look at the packet
6789 * before we can drop it. */
6790 if (!stp_forward_in_state(port->stp_state)
6791 && !stp_learn_in_state(port->stp_state)) {
6799 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6801 if (is_ip_any(&ctx->base_flow)
6802 && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6803 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6804 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6805 " but is not ECN capable");
6808 /* Set the ECN CE value in the tunneled packet. */
6809 ctx->flow.nw_tos |= IP_ECN_CE;
6817 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6818 struct action_xlate_ctx *ctx)
6820 bool was_evictable = true;
6821 const struct ofpact *a;
6824 /* Don't let the rule we're working on get evicted underneath us. */
6825 was_evictable = ctx->rule->up.evictable;
6826 ctx->rule->up.evictable = false;
6829 do_xlate_actions_again:
6830 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6831 struct ofpact_controller *controller;
6832 const struct ofpact_metadata *metadata;
6840 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6841 ofpact_get_OUTPUT(a)->max_len, true);
6844 case OFPACT_CONTROLLER:
6845 controller = ofpact_get_CONTROLLER(a);
6846 execute_controller_action(ctx, controller->max_len,
6848 controller->controller_id);
6851 case OFPACT_ENQUEUE:
6852 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6855 case OFPACT_SET_VLAN_VID:
6856 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6857 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6861 case OFPACT_SET_VLAN_PCP:
6862 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6863 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6868 case OFPACT_STRIP_VLAN:
6869 ctx->flow.vlan_tci = htons(0);
6872 case OFPACT_PUSH_VLAN:
6873 /* XXX 802.1AD(QinQ) */
6874 ctx->flow.vlan_tci = htons(VLAN_CFI);
6877 case OFPACT_SET_ETH_SRC:
6878 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6882 case OFPACT_SET_ETH_DST:
6883 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6887 case OFPACT_SET_IPV4_SRC:
6888 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6889 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6893 case OFPACT_SET_IPV4_DST:
6894 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6895 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6899 case OFPACT_SET_IPV4_DSCP:
6900 /* OpenFlow 1.0 only supports IPv4. */
6901 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6902 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6903 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6907 case OFPACT_SET_L4_SRC_PORT:
6908 if (is_ip_any(&ctx->flow)) {
6909 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6913 case OFPACT_SET_L4_DST_PORT:
6914 if (is_ip_any(&ctx->flow)) {
6915 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6919 case OFPACT_RESUBMIT:
6920 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6923 case OFPACT_SET_TUNNEL:
6924 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6927 case OFPACT_SET_QUEUE:
6928 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6931 case OFPACT_POP_QUEUE:
6932 ctx->flow.skb_priority = ctx->orig_skb_priority;
6935 case OFPACT_REG_MOVE:
6936 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6939 case OFPACT_REG_LOAD:
6940 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6943 case OFPACT_STACK_PUSH:
6944 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6948 case OFPACT_STACK_POP:
6949 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6953 case OFPACT_PUSH_MPLS:
6954 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6957 case OFPACT_POP_MPLS:
6958 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6961 case OFPACT_SET_MPLS_TTL:
6962 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6967 case OFPACT_DEC_MPLS_TTL:
6968 if (execute_dec_mpls_ttl_action(ctx)) {
6973 case OFPACT_DEC_TTL:
6974 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6980 /* Nothing to do. */
6983 case OFPACT_MULTIPATH:
6984 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6988 ctx->ofproto->has_bundle_action = true;
6989 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6992 case OFPACT_OUTPUT_REG:
6993 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6997 ctx->has_learn = true;
6998 if (ctx->may_learn) {
6999 xlate_learn_action(ctx, ofpact_get_LEARN(a));
7007 case OFPACT_FIN_TIMEOUT:
7008 ctx->has_fin_timeout = true;
7009 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
7012 case OFPACT_CLEAR_ACTIONS:
7014 * Nothing to do because writa-actions is not supported for now.
7015 * When writa-actions is supported, clear-actions also must
7016 * be supported at the same time.
7020 case OFPACT_WRITE_METADATA:
7021 metadata = ofpact_get_WRITE_METADATA(a);
7022 ctx->flow.metadata &= ~metadata->mask;
7023 ctx->flow.metadata |= metadata->metadata & metadata->mask;
7026 case OFPACT_GOTO_TABLE: {
7027 /* It is assumed that goto-table is the last action. */
7028 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
7029 struct rule_dpif *rule;
7031 ovs_assert(ctx->table_id < ogt->table_id);
7033 ctx->table_id = ogt->table_id;
7035 /* Look up a flow from the new table. */
7036 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
7038 tag_the_flow(ctx, rule);
7040 rule = ctx_rule_hooks(ctx, rule, true);
7044 ctx->rule->up.evictable = was_evictable;
7047 was_evictable = rule->up.evictable;
7048 rule->up.evictable = false;
7050 /* Tail recursion removal. */
7051 ofpacts = rule->up.ofpacts;
7052 ofpacts_len = rule->up.ofpacts_len;
7053 goto do_xlate_actions_again;
7059 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7066 ctx->rule->up.evictable = was_evictable;
7071 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7072 struct ofproto_dpif *ofproto, const struct flow *flow,
7073 const struct initial_vals *initial_vals,
7074 struct rule_dpif *rule,
7075 uint8_t tcp_flags, const struct ofpbuf *packet)
7077 /* Flow initialization rules:
7078 * - 'base_flow' must match the kernel's view of the packet at the
7079 * time that action processing starts. 'flow' represents any
7080 * transformations we wish to make through actions.
7081 * - By default 'base_flow' and 'flow' are the same since the input
7082 * packet matches the output before any actions are applied.
7083 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7084 * of the received packet as seen by the kernel. If we later output
7085 * to another device without any modifications this will cause us to
7086 * insert a new tag since the original one was stripped off by the
7088 * - Tunnel metadata as received is retained in 'flow'. This allows
7089 * tunnel metadata matching also in later tables.
7090 * Since a kernel action for setting the tunnel metadata will only be
7091 * generated with actual tunnel output, changing the tunnel metadata
7092 * values in 'flow' (such as tun_id) will only have effect with a later
7093 * tunnel output action.
7094 * - Tunnel 'base_flow' is completely cleared since that is what the
7095 * kernel does. If we wish to maintain the original values an action
7096 * needs to be generated. */
7098 ctx->ofproto = ofproto;
7100 ctx->base_flow = ctx->flow;
7101 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
7102 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7104 ctx->packet = packet;
7105 ctx->may_learn = packet != NULL;
7106 ctx->tcp_flags = tcp_flags;
7107 ctx->resubmit_hook = NULL;
7108 ctx->report_hook = NULL;
7109 ctx->resubmit_stats = NULL;
7112 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7113 * into datapath actions in 'odp_actions', using 'ctx'. */
7115 xlate_actions(struct action_xlate_ctx *ctx,
7116 const struct ofpact *ofpacts, size_t ofpacts_len,
7117 struct ofpbuf *odp_actions)
7119 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7120 * that in the future we always keep a copy of the original flow for
7121 * tracing purposes. */
7122 static bool hit_resubmit_limit;
7124 enum slow_path_reason special;
7125 struct ofport_dpif *in_port;
7126 struct flow orig_flow;
7128 COVERAGE_INC(ofproto_dpif_xlate);
7130 ofpbuf_clear(odp_actions);
7131 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7133 ctx->odp_actions = odp_actions;
7136 ctx->has_learn = false;
7137 ctx->has_normal = false;
7138 ctx->has_fin_timeout = false;
7139 ctx->nf_output_iface = NF_OUT_DROP;
7142 ctx->max_resubmit_trigger = false;
7143 ctx->orig_skb_priority = ctx->flow.skb_priority;
7147 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7149 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7150 /* Do this conditionally because the copy is expensive enough that it
7151 * shows up in profiles. */
7152 orig_flow = ctx->flow;
7155 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7156 switch (ctx->ofproto->up.frag_handling) {
7157 case OFPC_FRAG_NORMAL:
7158 /* We must pretend that transport ports are unavailable. */
7159 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7160 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7163 case OFPC_FRAG_DROP:
7166 case OFPC_FRAG_REASM:
7169 case OFPC_FRAG_NX_MATCH:
7170 /* Nothing to do. */
7173 case OFPC_INVALID_TTL_TO_CONTROLLER:
7178 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7179 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7181 ctx->slow |= special;
7183 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7184 struct initial_vals initial_vals;
7185 size_t sample_actions_len;
7186 uint32_t local_odp_port;
7188 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7190 add_sflow_action(ctx);
7191 add_ipfix_action(ctx);
7192 sample_actions_len = ctx->odp_actions->size;
7194 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7195 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7197 /* We've let OFPP_NORMAL and the learning action look at the
7198 * packet, so drop it now if forwarding is disabled. */
7199 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7200 ctx->odp_actions->size = sample_actions_len;
7204 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7205 if (!hit_resubmit_limit) {
7206 /* We didn't record the original flow. Make sure we do from
7208 hit_resubmit_limit = true;
7209 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7210 struct ds ds = DS_EMPTY_INITIALIZER;
7212 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7213 &initial_vals, &ds);
7214 VLOG_ERR("Trace triggered by excessive resubmit "
7215 "recursion:\n%s", ds_cstr(&ds));
7220 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7221 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
7223 ctx->odp_actions->data,
7224 ctx->odp_actions->size)) {
7225 ctx->slow |= SLOW_IN_BAND;
7227 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
7229 compose_output_action(ctx, OFPP_LOCAL);
7232 if (ctx->ofproto->has_mirrors) {
7233 add_mirror_actions(ctx, &orig_flow);
7235 fix_sflow_action(ctx);
7238 ofpbuf_uninit(&ctx->stack);
7241 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7242 * into datapath actions, using 'ctx', and discards the datapath actions. */
7244 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7245 const struct ofpact *ofpacts,
7248 uint64_t odp_actions_stub[1024 / 8];
7249 struct ofpbuf odp_actions;
7251 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7252 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7253 ofpbuf_uninit(&odp_actions);
7257 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7259 if (ctx->report_hook) {
7260 ctx->report_hook(ctx, s);
7264 /* OFPP_NORMAL implementation. */
7266 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7268 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7269 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7270 * the bundle on which the packet was received, returns the VLAN to which the
7273 * Both 'vid' and the return value are in the range 0...4095. */
7275 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7277 switch (in_bundle->vlan_mode) {
7278 case PORT_VLAN_ACCESS:
7279 return in_bundle->vlan;
7282 case PORT_VLAN_TRUNK:
7285 case PORT_VLAN_NATIVE_UNTAGGED:
7286 case PORT_VLAN_NATIVE_TAGGED:
7287 return vid ? vid : in_bundle->vlan;
7294 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7295 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7298 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7299 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7302 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7304 /* Allow any VID on the OFPP_NONE port. */
7305 if (in_bundle == &ofpp_none_bundle) {
7309 switch (in_bundle->vlan_mode) {
7310 case PORT_VLAN_ACCESS:
7313 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7314 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7315 "packet received on port %s configured as VLAN "
7316 "%"PRIu16" access port",
7317 in_bundle->ofproto->up.name, vid,
7318 in_bundle->name, in_bundle->vlan);
7324 case PORT_VLAN_NATIVE_UNTAGGED:
7325 case PORT_VLAN_NATIVE_TAGGED:
7327 /* Port must always carry its native VLAN. */
7331 case PORT_VLAN_TRUNK:
7332 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7334 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7335 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7336 "received on port %s not configured for trunking "
7338 in_bundle->ofproto->up.name, vid,
7339 in_bundle->name, vid);
7351 /* Given 'vlan', the VLAN that a packet belongs to, and
7352 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7353 * that should be included in the 802.1Q header. (If the return value is 0,
7354 * then the 802.1Q header should only be included in the packet if there is a
7357 * Both 'vlan' and the return value are in the range 0...4095. */
7359 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7361 switch (out_bundle->vlan_mode) {
7362 case PORT_VLAN_ACCESS:
7365 case PORT_VLAN_TRUNK:
7366 case PORT_VLAN_NATIVE_TAGGED:
7369 case PORT_VLAN_NATIVE_UNTAGGED:
7370 return vlan == out_bundle->vlan ? 0 : vlan;
7378 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7381 struct ofport_dpif *port;
7383 ovs_be16 tci, old_tci;
7385 vid = output_vlan_to_vid(out_bundle, vlan);
7386 if (!out_bundle->bond) {
7387 port = ofbundle_get_a_port(out_bundle);
7389 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7392 /* No slaves enabled, so drop packet. */
7397 old_tci = ctx->flow.vlan_tci;
7399 if (tci || out_bundle->use_priority_tags) {
7400 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7402 tci |= htons(VLAN_CFI);
7405 ctx->flow.vlan_tci = tci;
7407 compose_output_action(ctx, port->up.ofp_port);
7408 ctx->flow.vlan_tci = old_tci;
7412 mirror_mask_ffs(mirror_mask_t mask)
7414 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7419 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7421 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7422 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7426 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7428 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7431 /* Returns an arbitrary interface within 'bundle'. */
7432 static struct ofport_dpif *
7433 ofbundle_get_a_port(const struct ofbundle *bundle)
7435 return CONTAINER_OF(list_front(&bundle->ports),
7436 struct ofport_dpif, bundle_node);
7440 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7442 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7446 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7448 struct ofproto_dpif *ofproto = ctx->ofproto;
7449 mirror_mask_t mirrors;
7450 struct ofbundle *in_bundle;
7453 const struct nlattr *a;
7456 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7457 ctx->packet != NULL, NULL);
7461 mirrors = in_bundle->src_mirrors;
7463 /* Drop frames on bundles reserved for mirroring. */
7464 if (in_bundle->mirror_out) {
7465 if (ctx->packet != NULL) {
7466 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7467 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7468 "%s, which is reserved exclusively for mirroring",
7469 ctx->ofproto->up.name, in_bundle->name);
7475 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7476 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7479 vlan = input_vid_to_vlan(in_bundle, vid);
7481 /* Look at the output ports to check for destination selections. */
7483 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7484 ctx->odp_actions->size) {
7485 enum ovs_action_attr type = nl_attr_type(a);
7486 struct ofport_dpif *ofport;
7488 if (type != OVS_ACTION_ATTR_OUTPUT) {
7492 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7493 if (ofport && ofport->bundle) {
7494 mirrors |= ofport->bundle->dst_mirrors;
7502 /* Restore the original packet before adding the mirror actions. */
7503 ctx->flow = *orig_flow;
7508 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7510 if (!vlan_is_mirrored(m, vlan)) {
7511 mirrors = zero_rightmost_1bit(mirrors);
7515 mirrors &= ~m->dup_mirrors;
7516 ctx->mirrors |= m->dup_mirrors;
7518 output_normal(ctx, m->out, vlan);
7519 } else if (vlan != m->out_vlan
7520 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7521 struct ofbundle *bundle;
7523 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7524 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7525 && !bundle->mirror_out) {
7526 output_normal(ctx, bundle, m->out_vlan);
7534 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7535 uint64_t packets, uint64_t bytes)
7541 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7544 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7547 /* In normal circumstances 'm' will not be NULL. However,
7548 * if mirrors are reconfigured, we can temporarily get out
7549 * of sync in facet_revalidate(). We could "correct" the
7550 * mirror list before reaching here, but doing that would
7551 * not properly account the traffic stats we've currently
7552 * accumulated for previous mirror configuration. */
7556 m->packet_count += packets;
7557 m->byte_count += bytes;
7561 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7562 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7563 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7565 is_gratuitous_arp(const struct flow *flow)
7567 return (flow->dl_type == htons(ETH_TYPE_ARP)
7568 && eth_addr_is_broadcast(flow->dl_dst)
7569 && (flow->nw_proto == ARP_OP_REPLY
7570 || (flow->nw_proto == ARP_OP_REQUEST
7571 && flow->nw_src == flow->nw_dst)));
7575 update_learning_table(struct ofproto_dpif *ofproto,
7576 const struct flow *flow, int vlan,
7577 struct ofbundle *in_bundle)
7579 struct mac_entry *mac;
7581 /* Don't learn the OFPP_NONE port. */
7582 if (in_bundle == &ofpp_none_bundle) {
7586 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7590 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7591 if (is_gratuitous_arp(flow)) {
7592 /* We don't want to learn from gratuitous ARP packets that are
7593 * reflected back over bond slaves so we lock the learning table. */
7594 if (!in_bundle->bond) {
7595 mac_entry_set_grat_arp_lock(mac);
7596 } else if (mac_entry_is_grat_arp_locked(mac)) {
7601 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7602 /* The log messages here could actually be useful in debugging,
7603 * so keep the rate limit relatively high. */
7604 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7605 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7606 "on port %s in VLAN %d",
7607 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7608 in_bundle->name, vlan);
7610 mac->port.p = in_bundle;
7611 tag_set_add(&ofproto->backer->revalidate_set,
7612 mac_learning_changed(ofproto->ml, mac));
7616 static struct ofbundle *
7617 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7618 bool warn, struct ofport_dpif **in_ofportp)
7620 struct ofport_dpif *ofport;
7622 /* Find the port and bundle for the received packet. */
7623 ofport = get_ofp_port(ofproto, in_port);
7625 *in_ofportp = ofport;
7627 if (ofport && ofport->bundle) {
7628 return ofport->bundle;
7631 /* Special-case OFPP_NONE, which a controller may use as the ingress
7632 * port for traffic that it is sourcing. */
7633 if (in_port == OFPP_NONE) {
7634 return &ofpp_none_bundle;
7637 /* Odd. A few possible reasons here:
7639 * - We deleted a port but there are still a few packets queued up
7642 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7643 * we don't know about.
7645 * - The ofproto client didn't configure the port as part of a bundle.
7646 * This is particularly likely to happen if a packet was received on the
7647 * port after it was created, but before the client had a chance to
7648 * configure its bundle.
7651 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7653 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7654 "port %"PRIu16, ofproto->up.name, in_port);
7659 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7660 * dropped. Returns true if they may be forwarded, false if they should be
7663 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7664 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7666 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7667 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7668 * checked by input_vid_is_valid().
7670 * May also add tags to '*tags', although the current implementation only does
7671 * so in one special case.
7674 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7677 struct ofproto_dpif *ofproto = ctx->ofproto;
7678 struct flow *flow = &ctx->flow;
7679 struct ofbundle *in_bundle = in_port->bundle;
7681 /* Drop frames for reserved multicast addresses
7682 * only if forward_bpdu option is absent. */
7683 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7684 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7688 if (in_bundle->bond) {
7689 struct mac_entry *mac;
7691 switch (bond_check_admissibility(in_bundle->bond, in_port,
7692 flow->dl_dst, &ctx->tags)) {
7697 xlate_report(ctx, "bonding refused admissibility, dropping");
7700 case BV_DROP_IF_MOVED:
7701 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7702 if (mac && mac->port.p != in_bundle &&
7703 (!is_gratuitous_arp(flow)
7704 || mac_entry_is_grat_arp_locked(mac))) {
7705 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7717 xlate_normal(struct action_xlate_ctx *ctx)
7719 struct ofport_dpif *in_port;
7720 struct ofbundle *in_bundle;
7721 struct mac_entry *mac;
7725 ctx->has_normal = true;
7727 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7728 ctx->packet != NULL, &in_port);
7730 xlate_report(ctx, "no input bundle, dropping");
7734 /* Drop malformed frames. */
7735 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7736 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7737 if (ctx->packet != NULL) {
7738 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7739 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7740 "VLAN tag received on port %s",
7741 ctx->ofproto->up.name, in_bundle->name);
7743 xlate_report(ctx, "partial VLAN tag, dropping");
7747 /* Drop frames on bundles reserved for mirroring. */
7748 if (in_bundle->mirror_out) {
7749 if (ctx->packet != NULL) {
7750 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7751 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7752 "%s, which is reserved exclusively for mirroring",
7753 ctx->ofproto->up.name, in_bundle->name);
7755 xlate_report(ctx, "input port is mirror output port, dropping");
7760 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7761 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7762 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7765 vlan = input_vid_to_vlan(in_bundle, vid);
7767 /* Check other admissibility requirements. */
7768 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7772 /* Learn source MAC. */
7773 if (ctx->may_learn) {
7774 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7777 /* Determine output bundle. */
7778 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7781 if (mac->port.p != in_bundle) {
7782 xlate_report(ctx, "forwarding to learned port");
7783 output_normal(ctx, mac->port.p, vlan);
7785 xlate_report(ctx, "learned port is input port, dropping");
7788 struct ofbundle *bundle;
7790 xlate_report(ctx, "no learned MAC for destination, flooding");
7791 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7792 if (bundle != in_bundle
7793 && ofbundle_includes_vlan(bundle, vlan)
7794 && bundle->floodable
7795 && !bundle->mirror_out) {
7796 output_normal(ctx, bundle, vlan);
7799 ctx->nf_output_iface = NF_OUT_FLOOD;
7803 /* Optimized flow revalidation.
7805 * It's a difficult problem, in general, to tell which facets need to have
7806 * their actions recalculated whenever the OpenFlow flow table changes. We
7807 * don't try to solve that general problem: for most kinds of OpenFlow flow
7808 * table changes, we recalculate the actions for every facet. This is
7809 * relatively expensive, but it's good enough if the OpenFlow flow table
7810 * doesn't change very often.
7812 * However, we can expect one particular kind of OpenFlow flow table change to
7813 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7814 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7815 * table, we add a special case that applies to flow tables in which every rule
7816 * has the same form (that is, the same wildcards), except that the table is
7817 * also allowed to have a single "catch-all" flow that matches all packets. We
7818 * optimize this case by tagging all of the facets that resubmit into the table
7819 * and invalidating the same tag whenever a flow changes in that table. The
7820 * end result is that we revalidate just the facets that need it (and sometimes
7821 * a few more, but not all of the facets or even all of the facets that
7822 * resubmit to the table modified by MAC learning). */
7824 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7825 * into an OpenFlow table with the given 'basis'. */
7827 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7830 if (minimask_is_catchall(mask)) {
7833 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7834 return tag_create_deterministic(hash);
7838 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7839 * taggability of that table.
7841 * This function must be called after *each* change to a flow table. If you
7842 * skip calling it on some changes then the pointer comparisons at the end can
7843 * be invalid if you get unlucky. For example, if a flow removal causes a
7844 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7845 * different wildcards to be created with the same address, then this function
7846 * will incorrectly skip revalidation. */
7848 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7850 struct table_dpif *table = &ofproto->tables[table_id];
7851 const struct oftable *oftable = &ofproto->up.tables[table_id];
7852 struct cls_table *catchall, *other;
7853 struct cls_table *t;
7855 catchall = other = NULL;
7857 switch (hmap_count(&oftable->cls.tables)) {
7859 /* We could tag this OpenFlow table but it would make the logic a
7860 * little harder and it's a corner case that doesn't seem worth it
7866 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7867 if (cls_table_is_catchall(t)) {
7869 } else if (!other) {
7872 /* Indicate that we can't tag this by setting both tables to
7873 * NULL. (We know that 'catchall' is already NULL.) */
7880 /* Can't tag this table. */
7884 if (table->catchall_table != catchall || table->other_table != other) {
7885 table->catchall_table = catchall;
7886 table->other_table = other;
7887 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7891 /* Given 'rule' that has changed in some way (either it is a rule being
7892 * inserted, a rule being deleted, or a rule whose actions are being
7893 * modified), marks facets for revalidation to ensure that packets will be
7894 * forwarded correctly according to the new state of the flow table.
7896 * This function must be called after *each* change to a flow table. See
7897 * the comment on table_update_taggable() for more information. */
7899 rule_invalidate(const struct rule_dpif *rule)
7901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7903 table_update_taggable(ofproto, rule->up.table_id);
7905 if (!ofproto->backer->need_revalidate) {
7906 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7908 if (table->other_table && rule->tag) {
7909 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7911 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7917 set_frag_handling(struct ofproto *ofproto_,
7918 enum ofp_config_flags frag_handling)
7920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7921 if (frag_handling != OFPC_FRAG_REASM) {
7922 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7930 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7931 const struct flow *flow,
7932 const struct ofpact *ofpacts, size_t ofpacts_len)
7934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7935 struct initial_vals initial_vals;
7936 struct odputil_keybuf keybuf;
7937 struct dpif_flow_stats stats;
7941 struct action_xlate_ctx ctx;
7942 uint64_t odp_actions_stub[1024 / 8];
7943 struct ofpbuf odp_actions;
7945 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7946 odp_flow_key_from_flow(&key, flow,
7947 ofp_port_to_odp_port(ofproto, flow->in_port));
7949 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7951 initial_vals.vlan_tci = flow->vlan_tci;
7952 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7953 packet_get_tcp_flags(packet, flow), packet);
7954 ctx.resubmit_stats = &stats;
7956 ofpbuf_use_stub(&odp_actions,
7957 odp_actions_stub, sizeof odp_actions_stub);
7958 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7959 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7960 odp_actions.data, odp_actions.size, packet);
7961 ofpbuf_uninit(&odp_actions);
7969 set_netflow(struct ofproto *ofproto_,
7970 const struct netflow_options *netflow_options)
7972 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7974 if (netflow_options) {
7975 if (!ofproto->netflow) {
7976 ofproto->netflow = netflow_create();
7978 return netflow_set_options(ofproto->netflow, netflow_options);
7980 netflow_destroy(ofproto->netflow);
7981 ofproto->netflow = NULL;
7987 get_netflow_ids(const struct ofproto *ofproto_,
7988 uint8_t *engine_type, uint8_t *engine_id)
7990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7992 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7996 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7998 if (!facet_is_controller_flow(facet) &&
7999 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
8000 struct subfacet *subfacet;
8001 struct ofexpired expired;
8003 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
8004 if (subfacet->path == SF_FAST_PATH) {
8005 struct dpif_flow_stats stats;
8007 subfacet_reinstall(subfacet, &stats);
8008 subfacet_update_stats(subfacet, &stats);
8012 expired.flow = facet->flow;
8013 expired.packet_count = facet->packet_count;
8014 expired.byte_count = facet->byte_count;
8015 expired.used = facet->used;
8016 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8021 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8023 struct facet *facet;
8025 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8026 send_active_timeout(ofproto, facet);
8030 static struct ofproto_dpif *
8031 ofproto_dpif_lookup(const char *name)
8033 struct ofproto_dpif *ofproto;
8035 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8036 hash_string(name, 0), &all_ofproto_dpifs) {
8037 if (!strcmp(ofproto->up.name, name)) {
8045 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8046 const char *argv[], void *aux OVS_UNUSED)
8048 struct ofproto_dpif *ofproto;
8051 ofproto = ofproto_dpif_lookup(argv[1]);
8053 unixctl_command_reply_error(conn, "no such bridge");
8056 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8058 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8059 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8063 unixctl_command_reply(conn, "table successfully flushed");
8067 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8068 const char *argv[], void *aux OVS_UNUSED)
8070 struct ds ds = DS_EMPTY_INITIALIZER;
8071 const struct ofproto_dpif *ofproto;
8072 const struct mac_entry *e;
8074 ofproto = ofproto_dpif_lookup(argv[1]);
8076 unixctl_command_reply_error(conn, "no such bridge");
8080 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8081 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8082 struct ofbundle *bundle = e->port.p;
8083 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8084 ofbundle_get_a_port(bundle)->odp_port,
8085 e->vlan, ETH_ADDR_ARGS(e->mac),
8086 mac_entry_age(ofproto->ml, e));
8088 unixctl_command_reply(conn, ds_cstr(&ds));
8093 struct action_xlate_ctx ctx;
8099 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8100 const struct rule_dpif *rule)
8102 ds_put_char_multiple(result, '\t', level);
8104 ds_put_cstr(result, "No match\n");
8108 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8109 table_id, ntohll(rule->up.flow_cookie));
8110 cls_rule_format(&rule->up.cr, result);
8111 ds_put_char(result, '\n');
8113 ds_put_char_multiple(result, '\t', level);
8114 ds_put_cstr(result, "OpenFlow ");
8115 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8116 ds_put_char(result, '\n');
8120 trace_format_flow(struct ds *result, int level, const char *title,
8121 struct trace_ctx *trace)
8123 ds_put_char_multiple(result, '\t', level);
8124 ds_put_format(result, "%s: ", title);
8125 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8126 ds_put_cstr(result, "unchanged");
8128 flow_format(result, &trace->ctx.flow);
8129 trace->flow = trace->ctx.flow;
8131 ds_put_char(result, '\n');
8135 trace_format_regs(struct ds *result, int level, const char *title,
8136 struct trace_ctx *trace)
8140 ds_put_char_multiple(result, '\t', level);
8141 ds_put_format(result, "%s:", title);
8142 for (i = 0; i < FLOW_N_REGS; i++) {
8143 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8145 ds_put_char(result, '\n');
8149 trace_format_odp(struct ds *result, int level, const char *title,
8150 struct trace_ctx *trace)
8152 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8154 ds_put_char_multiple(result, '\t', level);
8155 ds_put_format(result, "%s: ", title);
8156 format_odp_actions(result, odp_actions->data, odp_actions->size);
8157 ds_put_char(result, '\n');
8161 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8163 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8164 struct ds *result = trace->result;
8166 ds_put_char(result, '\n');
8167 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8168 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8169 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8170 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8174 trace_report(struct action_xlate_ctx *ctx, const char *s)
8176 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8177 struct ds *result = trace->result;
8179 ds_put_char_multiple(result, '\t', ctx->recurse);
8180 ds_put_cstr(result, s);
8181 ds_put_char(result, '\n');
8185 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8186 void *aux OVS_UNUSED)
8188 const char *dpname = argv[1];
8189 struct ofproto_dpif *ofproto;
8190 struct ofpbuf odp_key;
8191 struct ofpbuf *packet;
8192 struct initial_vals initial_vals;
8198 ofpbuf_init(&odp_key, 0);
8201 ofproto = ofproto_dpif_lookup(dpname);
8203 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8207 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8208 /* ofproto/trace dpname flow [-generate] */
8209 const char *flow_s = argv[2];
8210 const char *generate_s = argv[3];
8212 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8213 * flow. We guess which type it is based on whether 'flow_s' contains
8214 * an '(', since a datapath flow always contains '(') but an
8215 * OpenFlow-like flow should not (in fact it's allowed but I believe
8216 * that's not documented anywhere).
8218 * An alternative would be to try to parse 'flow_s' both ways, but then
8219 * it would be tricky giving a sensible error message. After all, do
8220 * you just say "syntax error" or do you present both error messages?
8221 * Both choices seem lousy. */
8222 if (strchr(flow_s, '(')) {
8225 /* Convert string to datapath key. */
8226 ofpbuf_init(&odp_key, 0);
8227 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8229 unixctl_command_reply_error(conn, "Bad flow syntax");
8233 /* The user might have specified the wrong ofproto but within the
8234 * same backer. That's OK, ofproto_receive() can find the right
8236 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8237 odp_key.size, &flow, NULL, &ofproto, NULL,
8239 unixctl_command_reply_error(conn, "Invalid flow");
8242 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8246 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8248 unixctl_command_reply_error(conn, error_s);
8253 initial_vals.vlan_tci = flow.vlan_tci;
8256 /* Generate a packet, if requested. */
8258 packet = ofpbuf_new(0);
8259 flow_compose(packet, &flow);
8261 } else if (argc == 7) {
8262 /* ofproto/trace dpname priority tun_id in_port mark packet */
8263 const char *priority_s = argv[2];
8264 const char *tun_id_s = argv[3];
8265 const char *in_port_s = argv[4];
8266 const char *mark_s = argv[5];
8267 const char *packet_s = argv[6];
8268 uint32_t in_port = atoi(in_port_s);
8269 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8270 uint32_t priority = atoi(priority_s);
8271 uint32_t mark = atoi(mark_s);
8274 msg = eth_from_hex(packet_s, &packet);
8276 unixctl_command_reply_error(conn, msg);
8280 ds_put_cstr(&result, "Packet: ");
8281 s = ofp_packet_to_string(packet->data, packet->size);
8282 ds_put_cstr(&result, s);
8285 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8286 flow.tunnel.tun_id = tun_id;
8287 initial_vals.vlan_tci = flow.vlan_tci;
8289 unixctl_command_reply_error(conn, "Bad command syntax");
8293 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8294 unixctl_command_reply(conn, ds_cstr(&result));
8297 ds_destroy(&result);
8298 ofpbuf_delete(packet);
8299 ofpbuf_uninit(&odp_key);
8303 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8304 const struct ofpbuf *packet,
8305 const struct initial_vals *initial_vals, struct ds *ds)
8307 struct rule_dpif *rule;
8309 ds_put_cstr(ds, "Flow: ");
8310 flow_format(ds, flow);
8311 ds_put_char(ds, '\n');
8313 rule = rule_dpif_lookup(ofproto, flow);
8315 trace_format_rule(ds, 0, 0, rule);
8316 if (rule == ofproto->miss_rule) {
8317 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8318 } else if (rule == ofproto->no_packet_in_rule) {
8319 ds_put_cstr(ds, "\nNo match, packets dropped because "
8320 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8324 uint64_t odp_actions_stub[1024 / 8];
8325 struct ofpbuf odp_actions;
8327 struct trace_ctx trace;
8330 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8333 ofpbuf_use_stub(&odp_actions,
8334 odp_actions_stub, sizeof odp_actions_stub);
8335 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8336 rule, tcp_flags, packet);
8337 trace.ctx.resubmit_hook = trace_resubmit;
8338 trace.ctx.report_hook = trace_report;
8339 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8342 ds_put_char(ds, '\n');
8343 trace_format_flow(ds, 0, "Final flow", &trace);
8344 ds_put_cstr(ds, "Datapath actions: ");
8345 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8346 ofpbuf_uninit(&odp_actions);
8348 if (trace.ctx.slow) {
8349 enum slow_path_reason slow;
8351 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8352 "slow path because it:");
8353 for (slow = trace.ctx.slow; slow; ) {
8354 enum slow_path_reason bit = rightmost_1bit(slow);
8358 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8361 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8364 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8367 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8370 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8373 ds_put_cstr(ds, "\n\t (The datapath actions are "
8374 "incomplete--for complete actions, "
8375 "please supply a packet.)");
8378 case SLOW_CONTROLLER:
8379 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8380 "to the OpenFlow controller.");
8383 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8384 "than the datapath supports.");
8391 if (slow & ~SLOW_MATCH) {
8392 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8393 "the special slow-path processing.");
8400 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8401 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8404 unixctl_command_reply(conn, NULL);
8408 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8409 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8412 unixctl_command_reply(conn, NULL);
8415 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8416 * 'reply' describing the results. */
8418 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8420 struct facet *facet;
8424 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8425 if (!facet_check_consistency(facet)) {
8430 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8434 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8435 ofproto->up.name, errors);
8437 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8442 ofproto_dpif_self_check(struct unixctl_conn *conn,
8443 int argc, const char *argv[], void *aux OVS_UNUSED)
8445 struct ds reply = DS_EMPTY_INITIALIZER;
8446 struct ofproto_dpif *ofproto;
8449 ofproto = ofproto_dpif_lookup(argv[1]);
8451 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8452 "ofproto/list for help)");
8455 ofproto_dpif_self_check__(ofproto, &reply);
8457 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8458 ofproto_dpif_self_check__(ofproto, &reply);
8462 unixctl_command_reply(conn, ds_cstr(&reply));
8466 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8467 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8468 * to destroy 'ofproto_shash' and free the returned value. */
8469 static const struct shash_node **
8470 get_ofprotos(struct shash *ofproto_shash)
8472 const struct ofproto_dpif *ofproto;
8474 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8475 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8476 shash_add_nocopy(ofproto_shash, name, ofproto);
8479 return shash_sort(ofproto_shash);
8483 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8484 const char *argv[] OVS_UNUSED,
8485 void *aux OVS_UNUSED)
8487 struct ds ds = DS_EMPTY_INITIALIZER;
8488 struct shash ofproto_shash;
8489 const struct shash_node **sorted_ofprotos;
8492 shash_init(&ofproto_shash);
8493 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8494 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8495 const struct shash_node *node = sorted_ofprotos[i];
8496 ds_put_format(&ds, "%s\n", node->name);
8499 shash_destroy(&ofproto_shash);
8500 free(sorted_ofprotos);
8502 unixctl_command_reply(conn, ds_cstr(&ds));
8507 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8509 const struct shash_node **ports;
8511 struct avg_subfacet_rates lifetime;
8512 unsigned long long int minutes;
8513 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8515 minutes = (time_msec() - ofproto->created) / min_ms;
8518 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8520 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8523 lifetime.add_rate = 0.0;
8524 lifetime.del_rate = 0.0;
8527 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8528 dpif_name(ofproto->backer->dpif));
8530 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8531 ofproto->n_hit, ofproto->n_missed);
8532 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8533 " life span: %llu(ms)\n",
8534 hmap_count(&ofproto->subfacets),
8535 avg_subfacet_count(ofproto),
8536 ofproto->max_n_subfacet,
8537 avg_subfacet_life_span(ofproto));
8538 if (minutes >= 60) {
8539 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8541 if (minutes >= 60 * 24) {
8542 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8544 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8546 ports = shash_sort(&ofproto->up.port_by_name);
8547 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8548 const struct shash_node *node = ports[i];
8549 struct ofport *ofport = node->data;
8550 const char *name = netdev_get_name(ofport->netdev);
8551 const char *type = netdev_get_type(ofport->netdev);
8554 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8556 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8557 if (odp_port != OVSP_NONE) {
8558 ds_put_format(ds, "%"PRIu32":", odp_port);
8560 ds_put_cstr(ds, "none:");
8563 if (strcmp(type, "system")) {
8564 struct netdev *netdev;
8567 ds_put_format(ds, " (%s", type);
8569 error = netdev_open(name, type, &netdev);
8574 error = netdev_get_config(netdev, &config);
8576 const struct smap_node **nodes;
8579 nodes = smap_sort(&config);
8580 for (i = 0; i < smap_count(&config); i++) {
8581 const struct smap_node *node = nodes[i];
8582 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8583 node->key, node->value);
8587 smap_destroy(&config);
8589 netdev_close(netdev);
8591 ds_put_char(ds, ')');
8593 ds_put_char(ds, '\n');
8599 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8600 const char *argv[], void *aux OVS_UNUSED)
8602 struct ds ds = DS_EMPTY_INITIALIZER;
8603 const struct ofproto_dpif *ofproto;
8607 for (i = 1; i < argc; i++) {
8608 ofproto = ofproto_dpif_lookup(argv[i]);
8610 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8611 "for help)", argv[i]);
8612 unixctl_command_reply_error(conn, ds_cstr(&ds));
8615 show_dp_format(ofproto, &ds);
8618 struct shash ofproto_shash;
8619 const struct shash_node **sorted_ofprotos;
8622 shash_init(&ofproto_shash);
8623 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8624 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8625 const struct shash_node *node = sorted_ofprotos[i];
8626 show_dp_format(node->data, &ds);
8629 shash_destroy(&ofproto_shash);
8630 free(sorted_ofprotos);
8633 unixctl_command_reply(conn, ds_cstr(&ds));
8638 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8639 int argc OVS_UNUSED, const char *argv[],
8640 void *aux OVS_UNUSED)
8642 struct ds ds = DS_EMPTY_INITIALIZER;
8643 const struct ofproto_dpif *ofproto;
8644 struct subfacet *subfacet;
8646 ofproto = ofproto_dpif_lookup(argv[1]);
8648 unixctl_command_reply_error(conn, "no such bridge");
8652 update_stats(ofproto->backer);
8654 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8655 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8657 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8658 subfacet->dp_packet_count, subfacet->dp_byte_count);
8659 if (subfacet->used) {
8660 ds_put_format(&ds, "%.3fs",
8661 (time_msec() - subfacet->used) / 1000.0);
8663 ds_put_format(&ds, "never");
8665 if (subfacet->facet->tcp_flags) {
8666 ds_put_cstr(&ds, ", flags:");
8667 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8670 ds_put_cstr(&ds, ", actions:");
8671 if (subfacet->slow) {
8672 uint64_t slow_path_stub[128 / 8];
8673 const struct nlattr *actions;
8676 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8677 slow_path_stub, sizeof slow_path_stub,
8678 &actions, &actions_len);
8679 format_odp_actions(&ds, actions, actions_len);
8681 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8683 ds_put_char(&ds, '\n');
8686 unixctl_command_reply(conn, ds_cstr(&ds));
8691 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8692 int argc OVS_UNUSED, const char *argv[],
8693 void *aux OVS_UNUSED)
8695 struct ds ds = DS_EMPTY_INITIALIZER;
8696 struct ofproto_dpif *ofproto;
8698 ofproto = ofproto_dpif_lookup(argv[1]);
8700 unixctl_command_reply_error(conn, "no such bridge");
8704 flush(&ofproto->up);
8706 unixctl_command_reply(conn, ds_cstr(&ds));
8711 ofproto_dpif_unixctl_init(void)
8713 static bool registered;
8719 unixctl_command_register(
8721 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8722 2, 6, ofproto_unixctl_trace, NULL);
8723 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8724 ofproto_unixctl_fdb_flush, NULL);
8725 unixctl_command_register("fdb/show", "bridge", 1, 1,
8726 ofproto_unixctl_fdb_show, NULL);
8727 unixctl_command_register("ofproto/clog", "", 0, 0,
8728 ofproto_dpif_clog, NULL);
8729 unixctl_command_register("ofproto/unclog", "", 0, 0,
8730 ofproto_dpif_unclog, NULL);
8731 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8732 ofproto_dpif_self_check, NULL);
8733 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8734 ofproto_unixctl_dpif_dump_dps, NULL);
8735 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8736 ofproto_unixctl_dpif_show, NULL);
8737 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8738 ofproto_unixctl_dpif_dump_flows, NULL);
8739 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8740 ofproto_unixctl_dpif_del_flows, NULL);
8743 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8745 * This is deprecated. It is only for compatibility with broken device drivers
8746 * in old versions of Linux that do not properly support VLANs when VLAN
8747 * devices are not used. When broken device drivers are no longer in
8748 * widespread use, we will delete these interfaces. */
8751 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8754 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8756 if (realdev_ofp_port == ofport->realdev_ofp_port
8757 && vid == ofport->vlandev_vid) {
8761 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8763 if (ofport->realdev_ofp_port) {
8766 if (realdev_ofp_port && ofport->bundle) {
8767 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8768 * themselves be part of a bundle. */
8769 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8772 ofport->realdev_ofp_port = realdev_ofp_port;
8773 ofport->vlandev_vid = vid;
8775 if (realdev_ofp_port) {
8776 vsp_add(ofport, realdev_ofp_port, vid);
8783 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8785 return hash_2words(realdev_ofp_port, vid);
8788 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8789 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8790 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8791 * it would return the port number of eth0.9.
8793 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8794 * function just returns its 'realdev_odp_port' argument. */
8796 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8797 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8799 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8800 uint16_t realdev_ofp_port;
8801 int vid = vlan_tci_to_vid(vlan_tci);
8802 const struct vlan_splinter *vsp;
8804 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8805 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8806 hash_realdev_vid(realdev_ofp_port, vid),
8807 &ofproto->realdev_vid_map) {
8808 if (vsp->realdev_ofp_port == realdev_ofp_port
8809 && vsp->vid == vid) {
8810 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8814 return realdev_odp_port;
8817 static struct vlan_splinter *
8818 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8820 struct vlan_splinter *vsp;
8822 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8823 &ofproto->vlandev_map) {
8824 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8832 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8833 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8834 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8835 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8836 * eth0 and store 9 in '*vid'.
8838 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8839 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8842 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8843 uint16_t vlandev_ofp_port, int *vid)
8845 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8846 const struct vlan_splinter *vsp;
8848 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8853 return vsp->realdev_ofp_port;
8859 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8860 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8861 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8862 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8863 * always the case unless VLAN splinters are enabled), returns false without
8864 * making any changes. */
8866 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8871 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8876 /* Cause the flow to be processed as if it came in on the real device with
8877 * the VLAN device's VLAN ID. */
8878 flow->in_port = realdev;
8879 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8884 vsp_remove(struct ofport_dpif *port)
8886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8887 struct vlan_splinter *vsp;
8889 vsp = vlandev_find(ofproto, port->up.ofp_port);
8891 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8892 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8895 port->realdev_ofp_port = 0;
8897 VLOG_ERR("missing vlan device record");
8902 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8906 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8907 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8908 == realdev_ofp_port)) {
8909 struct vlan_splinter *vsp;
8911 vsp = xmalloc(sizeof *vsp);
8912 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8913 hash_int(port->up.ofp_port, 0));
8914 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8915 hash_realdev_vid(realdev_ofp_port, vid));
8916 vsp->realdev_ofp_port = realdev_ofp_port;
8917 vsp->vlandev_ofp_port = port->up.ofp_port;
8920 port->realdev_ofp_port = realdev_ofp_port;
8922 VLOG_ERR("duplicate vlan device record");
8927 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8929 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8930 return ofport ? ofport->odp_port : OVSP_NONE;
8933 static struct ofport_dpif *
8934 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8936 struct ofport_dpif *port;
8938 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8939 hash_int(odp_port, 0),
8940 &backer->odp_to_ofport_map) {
8941 if (port->odp_port == odp_port) {
8950 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8952 struct ofport_dpif *port;
8954 port = odp_port_to_ofport(ofproto->backer, odp_port);
8955 if (port && &ofproto->up == port->up.ofproto) {
8956 return port->up.ofp_port;
8961 static unsigned long long int
8962 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8964 unsigned long long int dc;
8965 unsigned long long int avg;
8967 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8968 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8974 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8978 if (ofproto->n_update_stats) {
8979 avg_c = (double)ofproto->total_subfacet_count
8980 / ofproto->n_update_stats;
8987 show_dp_rates(struct ds *ds, const char *heading,
8988 const struct avg_subfacet_rates *rates)
8990 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8991 heading, rates->add_rate, rates->del_rate);
8995 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8997 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8998 hmap_count(&ofproto->subfacets));
9001 /* Compute exponentially weighted moving average, adding 'new' as the newest,
9002 * most heavily weighted element. 'base' designates the rate of decay: after
9003 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
9006 exp_mavg(double *avg, int base, double new)
9008 *avg = (*avg * (base - 1) + new) / base;
9012 update_moving_averages(struct ofproto_dpif *ofproto)
9014 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9016 /* Update hourly averages on the minute boundaries. */
9017 if (time_msec() - ofproto->last_minute >= min_ms) {
9018 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9019 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9021 /* Update daily averages on the hour boundaries. */
9022 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9023 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9024 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9027 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9028 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9029 ofproto->subfacet_add_count = 0;
9030 ofproto->subfacet_del_count = 0;
9031 ofproto->last_minute += min_ms;
9036 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9038 ofproto->n_hit += delta;
9041 const struct ofproto_class ofproto_dpif_class = {
9076 port_is_lacp_current,
9077 NULL, /* rule_choose_table */
9084 rule_modify_actions,
9098 get_stp_port_status,
9105 is_mirror_output_bundle,
9106 forward_bpdu_changed,
9107 set_mac_table_config,