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 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
596 static void port_run(struct ofport_dpif *);
597 static void port_run_fast(struct ofport_dpif *);
598 static void port_wait(struct ofport_dpif *);
599 static int set_cfm(struct ofport *, const struct cfm_settings *);
600 static void ofport_clear_priorities(struct ofport_dpif *);
601 static void run_fast_rl(void);
603 struct dpif_completion {
604 struct list list_node;
605 struct ofoperation *op;
608 /* Extra information about a classifier table.
609 * Currently used just for optimized flow revalidation. */
611 /* If either of these is nonnull, then this table has a form that allows
612 * flows to be tagged to avoid revalidating most flows for the most common
613 * kinds of flow table changes. */
614 struct cls_table *catchall_table; /* Table that wildcards all fields. */
615 struct cls_table *other_table; /* Table with any other wildcard set. */
616 uint32_t basis; /* Keeps each table's tags separate. */
619 /* Reasons that we might need to revalidate every facet, and corresponding
622 * A value of 0 means that there is no need to revalidate.
624 * It would be nice to have some cleaner way to integrate with coverage
625 * counters, but with only a few reasons I guess this is good enough for
627 enum revalidate_reason {
628 REV_RECONFIGURE = 1, /* Switch configuration changed. */
629 REV_STP, /* Spanning tree protocol port status change. */
630 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
631 REV_FLOW_TABLE, /* Flow table changed. */
632 REV_INCONSISTENCY /* Facet self-check failed. */
634 COVERAGE_DEFINE(rev_reconfigure);
635 COVERAGE_DEFINE(rev_stp);
636 COVERAGE_DEFINE(rev_port_toggled);
637 COVERAGE_DEFINE(rev_flow_table);
638 COVERAGE_DEFINE(rev_inconsistency);
640 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
641 * These are datapath flows which have no associated ofproto, if they did we
642 * would use facets. */
644 struct hmap_node hmap_node;
649 /* All datapaths of a given type share a single dpif backer instance. */
654 struct timer next_expiration;
655 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
657 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
659 /* Facet revalidation flags applying to facets which use this backer. */
660 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
661 struct tag_set revalidate_set; /* Revalidate only matching facets. */
663 struct hmap drop_keys; /* Set of dropped odp keys. */
666 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
667 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
669 static void drop_key_clear(struct dpif_backer *);
670 static struct ofport_dpif *
671 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
673 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
675 struct avg_subfacet_rates {
676 double add_rate; /* Moving average of new flows created per minute. */
677 double del_rate; /* Moving average of flows deleted per minute. */
679 static void show_dp_rates(struct ds *ds, const char *heading,
680 const struct avg_subfacet_rates *rates);
681 static void exp_mavg(double *avg, int base, double new);
683 struct ofproto_dpif {
684 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
686 struct dpif_backer *backer;
688 /* Special OpenFlow rules. */
689 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
690 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
696 struct netflow *netflow;
697 struct dpif_sflow *sflow;
698 struct dpif_ipfix *ipfix;
699 struct hmap bundles; /* Contains "struct ofbundle"s. */
700 struct mac_learning *ml;
701 struct ofmirror *mirrors[MAX_MIRRORS];
703 bool has_bonded_bundles;
707 struct hmap subfacets;
708 struct governor *governor;
709 long long int consistency_rl;
712 struct table_dpif tables[N_TABLES];
714 /* Support for debugging async flow mods. */
715 struct list completions;
717 bool has_bundle_action; /* True when the first bundle action appears. */
718 struct netdev_stats stats; /* To account packets generated and consumed in
723 long long int stp_last_tick;
725 /* VLAN splinters. */
726 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
727 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
730 struct sset ports; /* Set of standard port names. */
731 struct sset ghost_ports; /* Ports with no datapath port. */
732 struct sset port_poll_set; /* Queued names for port_poll() reply. */
733 int port_poll_errno; /* Last errno for port_poll() reply. */
735 /* Per ofproto's dpif stats. */
739 /* Subfacet statistics.
741 * These keep track of the total number of subfacets added and deleted and
742 * flow life span. They are useful for computing the flow rates stats
743 * exposed via "ovs-appctl dpif/show". The goal is to learn about
744 * traffic patterns in ways that we can use later to improve Open vSwitch
745 * performance in new situations. */
746 long long int created; /* Time when it is created. */
747 unsigned int max_n_subfacet; /* Maximum number of flows */
749 /* The average number of subfacets... */
750 struct avg_subfacet_rates hourly; /* ...over the last hour. */
751 struct avg_subfacet_rates daily; /* ...over the last day. */
752 long long int last_minute; /* Last time 'hourly' was updated. */
754 /* Number of subfacets added or deleted since 'last_minute'. */
755 unsigned int subfacet_add_count;
756 unsigned int subfacet_del_count;
758 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
759 unsigned long long int total_subfacet_add_count;
760 unsigned long long int total_subfacet_del_count;
762 /* Sum of the number of milliseconds that each subfacet existed,
763 * over the subfacets that have been added and then later deleted. */
764 unsigned long long int total_subfacet_life_span;
766 /* Incremented by the number of currently existing subfacets, each
767 * time we pull statistics from the kernel. */
768 unsigned long long int total_subfacet_count;
770 /* Number of times we pull statistics from the kernel. */
771 unsigned long long int n_update_stats;
773 static unsigned long long int avg_subfacet_life_span(
774 const struct ofproto_dpif *);
775 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
776 static void update_moving_averages(struct ofproto_dpif *ofproto);
777 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
779 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
781 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
782 * for debugging the asynchronous flow_mod implementation.) */
785 /* All existing ofproto_dpif instances, indexed by ->up.name. */
786 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
788 static void ofproto_dpif_unixctl_init(void);
790 static struct ofproto_dpif *
791 ofproto_dpif_cast(const struct ofproto *ofproto)
793 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
794 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
797 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
799 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
801 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
802 const struct ofpbuf *,
803 const struct initial_vals *, struct ds *);
805 /* Packet processing. */
806 static void update_learning_table(struct ofproto_dpif *,
807 const struct flow *, int vlan,
810 #define FLOW_MISS_MAX_BATCH 50
811 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
813 /* Flow expiration. */
814 static int expire(struct dpif_backer *);
817 static void send_netflow_active_timeouts(struct ofproto_dpif *);
820 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
821 static size_t compose_sflow_action(const struct ofproto_dpif *,
822 struct ofpbuf *odp_actions,
823 const struct flow *, uint32_t odp_port);
824 static void compose_ipfix_action(const struct ofproto_dpif *,
825 struct ofpbuf *odp_actions,
826 const struct flow *);
827 static void add_mirror_actions(struct action_xlate_ctx *ctx,
828 const struct flow *flow);
829 /* Global variables. */
830 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
832 /* Initial mappings of port to bridge mappings. */
833 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
835 /* Factory functions. */
838 init(const struct shash *iface_hints)
840 struct shash_node *node;
842 /* Make a local copy, since we don't own 'iface_hints' elements. */
843 SHASH_FOR_EACH(node, iface_hints) {
844 const struct iface_hint *orig_hint = node->data;
845 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
847 new_hint->br_name = xstrdup(orig_hint->br_name);
848 new_hint->br_type = xstrdup(orig_hint->br_type);
849 new_hint->ofp_port = orig_hint->ofp_port;
851 shash_add(&init_ofp_ports, node->name, new_hint);
856 enumerate_types(struct sset *types)
858 dp_enumerate_types(types);
862 enumerate_names(const char *type, struct sset *names)
864 struct ofproto_dpif *ofproto;
867 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
868 if (strcmp(type, ofproto->up.type)) {
871 sset_add(names, ofproto->up.name);
878 del(const char *type, const char *name)
883 error = dpif_open(name, type, &dpif);
885 error = dpif_delete(dpif);
892 port_open_type(const char *datapath_type, const char *port_type)
894 return dpif_port_open_type(datapath_type, port_type);
897 /* Type functions. */
899 static struct ofproto_dpif *
900 lookup_ofproto_dpif_by_port_name(const char *name)
902 struct ofproto_dpif *ofproto;
904 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
905 if (sset_contains(&ofproto->ports, name)) {
914 type_run(const char *type)
916 static long long int push_timer = LLONG_MIN;
917 struct dpif_backer *backer;
921 backer = shash_find_data(&all_dpif_backers, type);
923 /* This is not necessarily a problem, since backers are only
924 * created on demand. */
928 dpif_run(backer->dpif);
930 /* The most natural place to push facet statistics is when they're pulled
931 * from the datapath. However, when there are many flows in the datapath,
932 * this expensive operation can occur so frequently, that it reduces our
933 * ability to quickly set up flows. To reduce the cost, we push statistics
935 if (time_msec() > push_timer) {
936 push_timer = time_msec() + 2000;
940 if (backer->need_revalidate
941 || !tag_set_is_empty(&backer->revalidate_set)) {
942 struct tag_set revalidate_set = backer->revalidate_set;
943 bool need_revalidate = backer->need_revalidate;
944 struct ofproto_dpif *ofproto;
945 struct simap_node *node;
946 struct simap tmp_backers;
948 /* Handle tunnel garbage collection. */
949 simap_init(&tmp_backers);
950 simap_swap(&backer->tnl_backers, &tmp_backers);
952 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
953 struct ofport_dpif *iter;
955 if (backer != ofproto->backer) {
959 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
962 if (!iter->tnl_port) {
966 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
967 node = simap_find(&tmp_backers, dp_port);
969 simap_put(&backer->tnl_backers, dp_port, node->data);
970 simap_delete(&tmp_backers, node);
971 node = simap_find(&backer->tnl_backers, dp_port);
973 node = simap_find(&backer->tnl_backers, dp_port);
975 uint32_t odp_port = UINT32_MAX;
977 if (!dpif_port_add(backer->dpif, iter->up.netdev,
979 simap_put(&backer->tnl_backers, dp_port, odp_port);
980 node = simap_find(&backer->tnl_backers, dp_port);
985 iter->odp_port = node ? node->data : OVSP_NONE;
986 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
988 backer->need_revalidate = REV_RECONFIGURE;
993 SIMAP_FOR_EACH (node, &tmp_backers) {
994 dpif_port_del(backer->dpif, node->data);
996 simap_destroy(&tmp_backers);
998 switch (backer->need_revalidate) {
999 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1000 case REV_STP: COVERAGE_INC(rev_stp); break;
1001 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1002 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1003 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1006 if (backer->need_revalidate) {
1007 /* Clear the drop_keys in case we should now be accepting some
1008 * formerly dropped flows. */
1009 drop_key_clear(backer);
1012 /* Clear the revalidation flags. */
1013 tag_set_init(&backer->revalidate_set);
1014 backer->need_revalidate = 0;
1016 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1017 struct facet *facet, *next;
1019 if (ofproto->backer != backer) {
1023 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1025 || tag_set_intersects(&revalidate_set, facet->tags)) {
1026 facet_revalidate(facet);
1033 if (timer_expired(&backer->next_expiration)) {
1034 int delay = expire(backer);
1035 timer_set_duration(&backer->next_expiration, delay);
1038 /* Check for port changes in the dpif. */
1039 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1040 struct ofproto_dpif *ofproto;
1041 struct dpif_port port;
1043 /* Don't report on the datapath's device. */
1044 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1048 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1049 &all_ofproto_dpifs) {
1050 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1055 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1056 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1057 /* The port was removed. If we know the datapath,
1058 * report it through poll_set(). If we don't, it may be
1059 * notifying us of a removal we initiated, so ignore it.
1060 * If there's a pending ENOBUFS, let it stand, since
1061 * everything will be reevaluated. */
1062 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1063 sset_add(&ofproto->port_poll_set, devname);
1064 ofproto->port_poll_errno = 0;
1066 } else if (!ofproto) {
1067 /* The port was added, but we don't know with which
1068 * ofproto we should associate it. Delete it. */
1069 dpif_port_del(backer->dpif, port.port_no);
1071 dpif_port_destroy(&port);
1077 if (error != EAGAIN) {
1078 struct ofproto_dpif *ofproto;
1080 /* There was some sort of error, so propagate it to all
1081 * ofprotos that use this backer. */
1082 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1083 &all_ofproto_dpifs) {
1084 if (ofproto->backer == backer) {
1085 sset_clear(&ofproto->port_poll_set);
1086 ofproto->port_poll_errno = error;
1095 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1099 /* Handle one or more batches of upcalls, until there's nothing left to do
1100 * or until we do a fixed total amount of work.
1102 * We do work in batches because it can be much cheaper to set up a number
1103 * of flows and fire off their patches all at once. We do multiple batches
1104 * because in some cases handling a packet can cause another packet to be
1105 * queued almost immediately as part of the return flow. Both
1106 * optimizations can make major improvements on some benchmarks and
1107 * presumably for real traffic as well. */
1109 while (work < max_batch) {
1110 int retval = handle_upcalls(backer, max_batch - work);
1121 type_run_fast(const char *type)
1123 struct dpif_backer *backer;
1125 backer = shash_find_data(&all_dpif_backers, type);
1127 /* This is not necessarily a problem, since backers are only
1128 * created on demand. */
1132 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1138 static long long int port_rl = LLONG_MIN;
1139 static unsigned int backer_rl = 0;
1141 if (time_msec() >= port_rl) {
1142 struct ofproto_dpif *ofproto;
1143 struct ofport_dpif *ofport;
1145 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1147 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1148 port_run_fast(ofport);
1151 port_rl = time_msec() + 200;
1154 /* XXX: We have to be careful not to do too much work in this function. If
1155 * we call dpif_backer_run_fast() too often, or with too large a batch,
1156 * performance improves signifcantly, but at a cost. It's possible for the
1157 * number of flows in the datapath to increase without bound, and for poll
1158 * loops to take 10s of seconds. The correct solution to this problem,
1159 * long term, is to separate flow miss handling into it's own thread so it
1160 * isn't affected by revalidations, and expirations. Until then, this is
1161 * the best we can do. */
1162 if (++backer_rl >= 10) {
1163 struct shash_node *node;
1166 SHASH_FOR_EACH (node, &all_dpif_backers) {
1167 dpif_backer_run_fast(node->data, 1);
1173 type_wait(const char *type)
1175 struct dpif_backer *backer;
1177 backer = shash_find_data(&all_dpif_backers, type);
1179 /* This is not necessarily a problem, since backers are only
1180 * created on demand. */
1184 timer_wait(&backer->next_expiration);
1187 /* Basic life-cycle. */
1189 static int add_internal_flows(struct ofproto_dpif *);
1191 static struct ofproto *
1194 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1195 return &ofproto->up;
1199 dealloc(struct ofproto *ofproto_)
1201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1206 close_dpif_backer(struct dpif_backer *backer)
1208 struct shash_node *node;
1210 ovs_assert(backer->refcount > 0);
1212 if (--backer->refcount) {
1216 drop_key_clear(backer);
1217 hmap_destroy(&backer->drop_keys);
1219 simap_destroy(&backer->tnl_backers);
1220 hmap_destroy(&backer->odp_to_ofport_map);
1221 node = shash_find(&all_dpif_backers, backer->type);
1223 shash_delete(&all_dpif_backers, node);
1224 dpif_close(backer->dpif);
1229 /* Datapath port slated for removal from datapath. */
1230 struct odp_garbage {
1231 struct list list_node;
1236 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1238 struct dpif_backer *backer;
1239 struct dpif_port_dump port_dump;
1240 struct dpif_port port;
1241 struct shash_node *node;
1242 struct list garbage_list;
1243 struct odp_garbage *garbage, *next;
1249 backer = shash_find_data(&all_dpif_backers, type);
1256 backer_name = xasprintf("ovs-%s", type);
1258 /* Remove any existing datapaths, since we assume we're the only
1259 * userspace controlling the datapath. */
1261 dp_enumerate_names(type, &names);
1262 SSET_FOR_EACH(name, &names) {
1263 struct dpif *old_dpif;
1265 /* Don't remove our backer if it exists. */
1266 if (!strcmp(name, backer_name)) {
1270 if (dpif_open(name, type, &old_dpif)) {
1271 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1273 dpif_delete(old_dpif);
1274 dpif_close(old_dpif);
1277 sset_destroy(&names);
1279 backer = xmalloc(sizeof *backer);
1281 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1284 VLOG_ERR("failed to open datapath of type %s: %s", type,
1290 backer->type = xstrdup(type);
1291 backer->refcount = 1;
1292 hmap_init(&backer->odp_to_ofport_map);
1293 hmap_init(&backer->drop_keys);
1294 timer_set_duration(&backer->next_expiration, 1000);
1295 backer->need_revalidate = 0;
1296 simap_init(&backer->tnl_backers);
1297 tag_set_init(&backer->revalidate_set);
1300 dpif_flow_flush(backer->dpif);
1302 /* Loop through the ports already on the datapath and remove any
1303 * that we don't need anymore. */
1304 list_init(&garbage_list);
1305 dpif_port_dump_start(&port_dump, backer->dpif);
1306 while (dpif_port_dump_next(&port_dump, &port)) {
1307 node = shash_find(&init_ofp_ports, port.name);
1308 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1309 garbage = xmalloc(sizeof *garbage);
1310 garbage->odp_port = port.port_no;
1311 list_push_front(&garbage_list, &garbage->list_node);
1314 dpif_port_dump_done(&port_dump);
1316 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1317 dpif_port_del(backer->dpif, garbage->odp_port);
1318 list_remove(&garbage->list_node);
1322 shash_add(&all_dpif_backers, type, backer);
1324 error = dpif_recv_set(backer->dpif, true);
1326 VLOG_ERR("failed to listen on datapath of type %s: %s",
1327 type, strerror(error));
1328 close_dpif_backer(backer);
1336 construct(struct ofproto *ofproto_)
1338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1339 struct shash_node *node, *next;
1344 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1349 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1350 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1352 ofproto->n_matches = 0;
1354 ofproto->netflow = NULL;
1355 ofproto->sflow = NULL;
1356 ofproto->ipfix = NULL;
1357 ofproto->stp = NULL;
1358 hmap_init(&ofproto->bundles);
1359 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1360 for (i = 0; i < MAX_MIRRORS; i++) {
1361 ofproto->mirrors[i] = NULL;
1363 ofproto->has_bonded_bundles = false;
1365 hmap_init(&ofproto->facets);
1366 hmap_init(&ofproto->subfacets);
1367 ofproto->governor = NULL;
1368 ofproto->consistency_rl = LLONG_MIN;
1370 for (i = 0; i < N_TABLES; i++) {
1371 struct table_dpif *table = &ofproto->tables[i];
1373 table->catchall_table = NULL;
1374 table->other_table = NULL;
1375 table->basis = random_uint32();
1378 list_init(&ofproto->completions);
1380 ofproto_dpif_unixctl_init();
1382 ofproto->has_mirrors = false;
1383 ofproto->has_bundle_action = false;
1385 hmap_init(&ofproto->vlandev_map);
1386 hmap_init(&ofproto->realdev_vid_map);
1388 sset_init(&ofproto->ports);
1389 sset_init(&ofproto->ghost_ports);
1390 sset_init(&ofproto->port_poll_set);
1391 ofproto->port_poll_errno = 0;
1393 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1394 struct iface_hint *iface_hint = node->data;
1396 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1397 /* Check if the datapath already has this port. */
1398 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1399 sset_add(&ofproto->ports, node->name);
1402 free(iface_hint->br_name);
1403 free(iface_hint->br_type);
1405 shash_delete(&init_ofp_ports, node);
1409 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1410 hash_string(ofproto->up.name, 0));
1411 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1413 ofproto_init_tables(ofproto_, N_TABLES);
1414 error = add_internal_flows(ofproto);
1415 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1418 ofproto->n_missed = 0;
1420 ofproto->max_n_subfacet = 0;
1421 ofproto->created = time_msec();
1422 ofproto->last_minute = ofproto->created;
1423 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1424 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1425 ofproto->subfacet_add_count = 0;
1426 ofproto->subfacet_del_count = 0;
1427 ofproto->total_subfacet_add_count = 0;
1428 ofproto->total_subfacet_del_count = 0;
1429 ofproto->total_subfacet_life_span = 0;
1430 ofproto->total_subfacet_count = 0;
1431 ofproto->n_update_stats = 0;
1437 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1438 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1440 struct ofputil_flow_mod fm;
1443 match_init_catchall(&fm.match);
1445 match_set_reg(&fm.match, 0, id);
1446 fm.new_cookie = htonll(0);
1447 fm.cookie = htonll(0);
1448 fm.cookie_mask = htonll(0);
1449 fm.table_id = TBL_INTERNAL;
1450 fm.command = OFPFC_ADD;
1451 fm.idle_timeout = 0;
1452 fm.hard_timeout = 0;
1456 fm.ofpacts = ofpacts->data;
1457 fm.ofpacts_len = ofpacts->size;
1459 error = ofproto_flow_mod(&ofproto->up, &fm);
1461 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1462 id, ofperr_to_string(error));
1466 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1467 ovs_assert(*rulep != NULL);
1473 add_internal_flows(struct ofproto_dpif *ofproto)
1475 struct ofpact_controller *controller;
1476 uint64_t ofpacts_stub[128 / 8];
1477 struct ofpbuf ofpacts;
1481 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1484 controller = ofpact_put_CONTROLLER(&ofpacts);
1485 controller->max_len = UINT16_MAX;
1486 controller->controller_id = 0;
1487 controller->reason = OFPR_NO_MATCH;
1488 ofpact_pad(&ofpacts);
1490 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1495 ofpbuf_clear(&ofpacts);
1496 error = add_internal_flow(ofproto, id++, &ofpacts,
1497 &ofproto->no_packet_in_rule);
1502 complete_operations(struct ofproto_dpif *ofproto)
1504 struct dpif_completion *c, *next;
1506 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1507 ofoperation_complete(c->op, 0);
1508 list_remove(&c->list_node);
1514 destruct(struct ofproto *ofproto_)
1516 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1517 struct rule_dpif *rule, *next_rule;
1518 struct oftable *table;
1521 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1522 complete_operations(ofproto);
1524 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1525 struct cls_cursor cursor;
1527 cls_cursor_init(&cursor, &table->cls, NULL);
1528 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1529 ofproto_rule_destroy(&rule->up);
1533 for (i = 0; i < MAX_MIRRORS; i++) {
1534 mirror_destroy(ofproto->mirrors[i]);
1537 netflow_destroy(ofproto->netflow);
1538 dpif_sflow_destroy(ofproto->sflow);
1539 hmap_destroy(&ofproto->bundles);
1540 mac_learning_destroy(ofproto->ml);
1542 hmap_destroy(&ofproto->facets);
1543 hmap_destroy(&ofproto->subfacets);
1544 governor_destroy(ofproto->governor);
1546 hmap_destroy(&ofproto->vlandev_map);
1547 hmap_destroy(&ofproto->realdev_vid_map);
1549 sset_destroy(&ofproto->ports);
1550 sset_destroy(&ofproto->ghost_ports);
1551 sset_destroy(&ofproto->port_poll_set);
1553 close_dpif_backer(ofproto->backer);
1557 run_fast(struct ofproto *ofproto_)
1559 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1560 struct ofport_dpif *ofport;
1562 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1563 port_run_fast(ofport);
1570 run(struct ofproto *ofproto_)
1572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1573 struct ofport_dpif *ofport;
1574 struct ofbundle *bundle;
1578 complete_operations(ofproto);
1581 error = run_fast(ofproto_);
1586 if (ofproto->netflow) {
1587 if (netflow_run(ofproto->netflow)) {
1588 send_netflow_active_timeouts(ofproto);
1591 if (ofproto->sflow) {
1592 dpif_sflow_run(ofproto->sflow);
1595 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1598 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1603 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1605 /* Check the consistency of a random facet, to aid debugging. */
1606 if (time_msec() >= ofproto->consistency_rl
1607 && !hmap_is_empty(&ofproto->facets)
1608 && !ofproto->backer->need_revalidate) {
1609 struct facet *facet;
1611 ofproto->consistency_rl = time_msec() + 250;
1613 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1614 struct facet, hmap_node);
1615 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1617 if (!facet_check_consistency(facet)) {
1618 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1623 if (ofproto->governor) {
1626 governor_run(ofproto->governor);
1628 /* If the governor has shrunk to its minimum size and the number of
1629 * subfacets has dwindled, then drop the governor entirely.
1631 * For hysteresis, the number of subfacets to drop the governor is
1632 * smaller than the number needed to trigger its creation. */
1633 n_subfacets = hmap_count(&ofproto->subfacets);
1634 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1635 && governor_is_idle(ofproto->governor)) {
1636 governor_destroy(ofproto->governor);
1637 ofproto->governor = NULL;
1645 wait(struct ofproto *ofproto_)
1647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1648 struct ofport_dpif *ofport;
1649 struct ofbundle *bundle;
1651 if (!clogged && !list_is_empty(&ofproto->completions)) {
1652 poll_immediate_wake();
1655 dpif_wait(ofproto->backer->dpif);
1656 dpif_recv_wait(ofproto->backer->dpif);
1657 if (ofproto->sflow) {
1658 dpif_sflow_wait(ofproto->sflow);
1660 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1661 poll_immediate_wake();
1663 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1666 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1667 bundle_wait(bundle);
1669 if (ofproto->netflow) {
1670 netflow_wait(ofproto->netflow);
1672 mac_learning_wait(ofproto->ml);
1674 if (ofproto->backer->need_revalidate) {
1675 /* Shouldn't happen, but if it does just go around again. */
1676 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1677 poll_immediate_wake();
1679 if (ofproto->governor) {
1680 governor_wait(ofproto->governor);
1685 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1687 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1689 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1690 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1694 flush(struct ofproto *ofproto_)
1696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1697 struct subfacet *subfacet, *next_subfacet;
1698 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1702 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1703 &ofproto->subfacets) {
1704 if (subfacet->path != SF_NOT_INSTALLED) {
1705 batch[n_batch++] = subfacet;
1706 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1707 subfacet_destroy_batch(ofproto, batch, n_batch);
1711 subfacet_destroy(subfacet);
1716 subfacet_destroy_batch(ofproto, batch, n_batch);
1721 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1722 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1724 *arp_match_ip = true;
1725 *actions = (OFPUTIL_A_OUTPUT |
1726 OFPUTIL_A_SET_VLAN_VID |
1727 OFPUTIL_A_SET_VLAN_PCP |
1728 OFPUTIL_A_STRIP_VLAN |
1729 OFPUTIL_A_SET_DL_SRC |
1730 OFPUTIL_A_SET_DL_DST |
1731 OFPUTIL_A_SET_NW_SRC |
1732 OFPUTIL_A_SET_NW_DST |
1733 OFPUTIL_A_SET_NW_TOS |
1734 OFPUTIL_A_SET_TP_SRC |
1735 OFPUTIL_A_SET_TP_DST |
1740 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1743 struct dpif_dp_stats s;
1745 strcpy(ots->name, "classifier");
1747 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1749 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1750 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1753 static struct ofport *
1756 struct ofport_dpif *port = xmalloc(sizeof *port);
1761 port_dealloc(struct ofport *port_)
1763 struct ofport_dpif *port = ofport_dpif_cast(port_);
1768 port_construct(struct ofport *port_)
1770 struct ofport_dpif *port = ofport_dpif_cast(port_);
1771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1772 const struct netdev *netdev = port->up.netdev;
1773 struct dpif_port dpif_port;
1776 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1777 port->bundle = NULL;
1780 port->tag = tag_create_random();
1781 port->may_enable = true;
1782 port->stp_port = NULL;
1783 port->stp_state = STP_DISABLED;
1784 port->tnl_port = NULL;
1785 hmap_init(&port->priorities);
1786 port->realdev_ofp_port = 0;
1787 port->vlandev_vid = 0;
1788 port->carrier_seq = netdev_get_carrier_resets(netdev);
1790 if (netdev_vport_is_patch(netdev)) {
1791 /* By bailing out here, we don't submit the port to the sFlow module
1792 * to be considered for counter polling export. This is correct
1793 * because the patch port represents an interface that sFlow considers
1794 * to be "internal" to the switch as a whole, and therefore not an
1795 * candidate for counter polling. */
1796 port->odp_port = OVSP_NONE;
1800 error = dpif_port_query_by_name(ofproto->backer->dpif,
1801 netdev_vport_get_dpif_port(netdev),
1807 port->odp_port = dpif_port.port_no;
1809 if (netdev_get_tunnel_config(netdev)) {
1810 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1812 /* Sanity-check that a mapping doesn't already exist. This
1813 * shouldn't happen for non-tunnel ports. */
1814 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1815 VLOG_ERR("port %s already has an OpenFlow port number",
1817 dpif_port_destroy(&dpif_port);
1821 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1822 hash_int(port->odp_port, 0));
1824 dpif_port_destroy(&dpif_port);
1826 if (ofproto->sflow) {
1827 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1834 port_destruct(struct ofport *port_)
1836 struct ofport_dpif *port = ofport_dpif_cast(port_);
1837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1838 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1839 const char *devname = netdev_get_name(port->up.netdev);
1841 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1842 /* The underlying device is still there, so delete it. This
1843 * happens when the ofproto is being destroyed, since the caller
1844 * assumes that removal of attached ports will happen as part of
1846 if (!port->tnl_port) {
1847 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1849 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1852 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1853 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1856 tnl_port_del(port->tnl_port);
1857 sset_find_and_delete(&ofproto->ports, devname);
1858 sset_find_and_delete(&ofproto->ghost_ports, devname);
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860 bundle_remove(port_);
1861 set_cfm(port_, NULL);
1862 if (ofproto->sflow) {
1863 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1866 ofport_clear_priorities(port);
1867 hmap_destroy(&port->priorities);
1871 port_modified(struct ofport *port_)
1873 struct ofport_dpif *port = ofport_dpif_cast(port_);
1875 if (port->bundle && port->bundle->bond) {
1876 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1881 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1883 struct ofport_dpif *port = ofport_dpif_cast(port_);
1884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1885 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1887 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1888 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1889 OFPUTIL_PC_NO_PACKET_IN)) {
1890 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1893 bundle_update(port->bundle);
1899 set_sflow(struct ofproto *ofproto_,
1900 const struct ofproto_sflow_options *sflow_options)
1902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1903 struct dpif_sflow *ds = ofproto->sflow;
1905 if (sflow_options) {
1907 struct ofport_dpif *ofport;
1909 ds = ofproto->sflow = dpif_sflow_create();
1910 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1911 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1915 dpif_sflow_set_options(ds, sflow_options);
1918 dpif_sflow_destroy(ds);
1919 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1920 ofproto->sflow = NULL;
1928 struct ofproto *ofproto_,
1929 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1930 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1931 size_t n_flow_exporters_options)
1933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1934 struct dpif_ipfix *di = ofproto->ipfix;
1936 if (bridge_exporter_options || flow_exporters_options) {
1938 di = ofproto->ipfix = dpif_ipfix_create();
1940 dpif_ipfix_set_options(
1941 di, bridge_exporter_options, flow_exporters_options,
1942 n_flow_exporters_options);
1945 dpif_ipfix_destroy(di);
1946 ofproto->ipfix = NULL;
1953 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1955 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1962 struct ofproto_dpif *ofproto;
1964 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1965 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1969 if (cfm_configure(ofport->cfm, s)) {
1975 cfm_destroy(ofport->cfm);
1981 get_cfm_status(const struct ofport *ofport_,
1982 struct ofproto_cfm_status *status)
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987 status->faults = cfm_get_fault(ofport->cfm);
1988 status->remote_opstate = cfm_get_opup(ofport->cfm);
1989 status->health = cfm_get_health(ofport->cfm);
1990 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1998 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2001 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2005 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2006 if (ofport->bfd != old) {
2007 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2014 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2016 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2019 bfd_get_status(ofport->bfd, smap);
2026 /* Spanning Tree. */
2029 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2031 struct ofproto_dpif *ofproto = ofproto_;
2032 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2033 struct ofport_dpif *ofport;
2035 ofport = stp_port_get_aux(sp);
2037 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2038 ofproto->up.name, port_num);
2040 struct eth_header *eth = pkt->l2;
2042 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2043 if (eth_addr_is_zero(eth->eth_src)) {
2044 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2045 "with unknown MAC", ofproto->up.name, port_num);
2047 send_packet(ofport, pkt);
2053 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2055 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2059 /* Only revalidate flows if the configuration changed. */
2060 if (!s != !ofproto->stp) {
2061 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2065 if (!ofproto->stp) {
2066 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2067 send_bpdu_cb, ofproto);
2068 ofproto->stp_last_tick = time_msec();
2071 stp_set_bridge_id(ofproto->stp, s->system_id);
2072 stp_set_bridge_priority(ofproto->stp, s->priority);
2073 stp_set_hello_time(ofproto->stp, s->hello_time);
2074 stp_set_max_age(ofproto->stp, s->max_age);
2075 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2077 struct ofport *ofport;
2079 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2080 set_stp_port(ofport, NULL);
2083 stp_destroy(ofproto->stp);
2084 ofproto->stp = NULL;
2091 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2093 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2097 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2098 s->designated_root = stp_get_designated_root(ofproto->stp);
2099 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2108 update_stp_port_state(struct ofport_dpif *ofport)
2110 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2111 enum stp_state state;
2113 /* Figure out new state. */
2114 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2118 if (ofport->stp_state != state) {
2119 enum ofputil_port_state of_state;
2122 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2123 netdev_get_name(ofport->up.netdev),
2124 stp_state_name(ofport->stp_state),
2125 stp_state_name(state));
2126 if (stp_learn_in_state(ofport->stp_state)
2127 != stp_learn_in_state(state)) {
2128 /* xxx Learning action flows should also be flushed. */
2129 mac_learning_flush(ofproto->ml,
2130 &ofproto->backer->revalidate_set);
2132 fwd_change = stp_forward_in_state(ofport->stp_state)
2133 != stp_forward_in_state(state);
2135 ofproto->backer->need_revalidate = REV_STP;
2136 ofport->stp_state = state;
2137 ofport->stp_state_entered = time_msec();
2139 if (fwd_change && ofport->bundle) {
2140 bundle_update(ofport->bundle);
2143 /* Update the STP state bits in the OpenFlow port description. */
2144 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2145 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2146 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2147 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2148 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2150 ofproto_port_set_state(&ofport->up, of_state);
2154 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2155 * caller is responsible for assigning STP port numbers and ensuring
2156 * there are no duplicates. */
2158 set_stp_port(struct ofport *ofport_,
2159 const struct ofproto_port_stp_settings *s)
2161 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2163 struct stp_port *sp = ofport->stp_port;
2165 if (!s || !s->enable) {
2167 ofport->stp_port = NULL;
2168 stp_port_disable(sp);
2169 update_stp_port_state(ofport);
2172 } else if (sp && stp_port_no(sp) != s->port_num
2173 && ofport == stp_port_get_aux(sp)) {
2174 /* The port-id changed, so disable the old one if it's not
2175 * already in use by another port. */
2176 stp_port_disable(sp);
2179 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2180 stp_port_enable(sp);
2182 stp_port_set_aux(sp, ofport);
2183 stp_port_set_priority(sp, s->priority);
2184 stp_port_set_path_cost(sp, s->path_cost);
2186 update_stp_port_state(ofport);
2192 get_stp_port_status(struct ofport *ofport_,
2193 struct ofproto_port_stp_status *s)
2195 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2197 struct stp_port *sp = ofport->stp_port;
2199 if (!ofproto->stp || !sp) {
2205 s->port_id = stp_port_get_id(sp);
2206 s->state = stp_port_get_state(sp);
2207 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2208 s->role = stp_port_get_role(sp);
2209 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2215 stp_run(struct ofproto_dpif *ofproto)
2218 long long int now = time_msec();
2219 long long int elapsed = now - ofproto->stp_last_tick;
2220 struct stp_port *sp;
2223 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2224 ofproto->stp_last_tick = now;
2226 while (stp_get_changed_port(ofproto->stp, &sp)) {
2227 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2230 update_stp_port_state(ofport);
2234 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2235 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2241 stp_wait(struct ofproto_dpif *ofproto)
2244 poll_timer_wait(1000);
2248 /* Returns true if STP should process 'flow'. */
2250 stp_should_process_flow(const struct flow *flow)
2252 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2256 stp_process_packet(const struct ofport_dpif *ofport,
2257 const struct ofpbuf *packet)
2259 struct ofpbuf payload = *packet;
2260 struct eth_header *eth = payload.data;
2261 struct stp_port *sp = ofport->stp_port;
2263 /* Sink packets on ports that have STP disabled when the bridge has
2265 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2269 /* Trim off padding on payload. */
2270 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2271 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2274 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2275 stp_received_bpdu(sp, payload.data, payload.size);
2279 static struct priority_to_dscp *
2280 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2282 struct priority_to_dscp *pdscp;
2285 hash = hash_int(priority, 0);
2286 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2287 if (pdscp->priority == priority) {
2295 ofport_clear_priorities(struct ofport_dpif *ofport)
2297 struct priority_to_dscp *pdscp, *next;
2299 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2300 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2306 set_queues(struct ofport *ofport_,
2307 const struct ofproto_port_queue *qdscp_list,
2310 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2312 struct hmap new = HMAP_INITIALIZER(&new);
2315 for (i = 0; i < n_qdscp; i++) {
2316 struct priority_to_dscp *pdscp;
2320 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2321 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2326 pdscp = get_priority(ofport, priority);
2328 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2330 pdscp = xmalloc(sizeof *pdscp);
2331 pdscp->priority = priority;
2333 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2336 if (pdscp->dscp != dscp) {
2338 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2341 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2344 if (!hmap_is_empty(&ofport->priorities)) {
2345 ofport_clear_priorities(ofport);
2346 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2349 hmap_swap(&new, &ofport->priorities);
2357 /* Expires all MAC learning entries associated with 'bundle' and forces its
2358 * ofproto to revalidate every flow.
2360 * Normally MAC learning entries are removed only from the ofproto associated
2361 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2362 * are removed from every ofproto. When patch ports and SLB bonds are in use
2363 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2364 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2365 * with the host from which it migrated. */
2367 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2369 struct ofproto_dpif *ofproto = bundle->ofproto;
2370 struct mac_learning *ml = ofproto->ml;
2371 struct mac_entry *mac, *next_mac;
2373 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2374 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2375 if (mac->port.p == bundle) {
2377 struct ofproto_dpif *o;
2379 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2381 struct mac_entry *e;
2383 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2386 mac_learning_expire(o->ml, e);
2392 mac_learning_expire(ml, mac);
2397 static struct ofbundle *
2398 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2400 struct ofbundle *bundle;
2402 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2403 &ofproto->bundles) {
2404 if (bundle->aux == aux) {
2411 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2412 * ones that are found to 'bundles'. */
2414 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2415 void **auxes, size_t n_auxes,
2416 struct hmapx *bundles)
2420 hmapx_init(bundles);
2421 for (i = 0; i < n_auxes; i++) {
2422 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2424 hmapx_add(bundles, bundle);
2430 bundle_update(struct ofbundle *bundle)
2432 struct ofport_dpif *port;
2434 bundle->floodable = true;
2435 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2436 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2437 || !stp_forward_in_state(port->stp_state)) {
2438 bundle->floodable = false;
2445 bundle_del_port(struct ofport_dpif *port)
2447 struct ofbundle *bundle = port->bundle;
2449 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2451 list_remove(&port->bundle_node);
2452 port->bundle = NULL;
2455 lacp_slave_unregister(bundle->lacp, port);
2458 bond_slave_unregister(bundle->bond, port);
2461 bundle_update(bundle);
2465 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2466 struct lacp_slave_settings *lacp)
2468 struct ofport_dpif *port;
2470 port = get_ofp_port(bundle->ofproto, ofp_port);
2475 if (port->bundle != bundle) {
2476 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2478 bundle_del_port(port);
2481 port->bundle = bundle;
2482 list_push_back(&bundle->ports, &port->bundle_node);
2483 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2484 || !stp_forward_in_state(port->stp_state)) {
2485 bundle->floodable = false;
2489 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2490 lacp_slave_register(bundle->lacp, port, lacp);
2497 bundle_destroy(struct ofbundle *bundle)
2499 struct ofproto_dpif *ofproto;
2500 struct ofport_dpif *port, *next_port;
2507 ofproto = bundle->ofproto;
2508 for (i = 0; i < MAX_MIRRORS; i++) {
2509 struct ofmirror *m = ofproto->mirrors[i];
2511 if (m->out == bundle) {
2513 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2514 || hmapx_find_and_delete(&m->dsts, bundle)) {
2515 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2520 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2521 bundle_del_port(port);
2524 bundle_flush_macs(bundle, true);
2525 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2527 free(bundle->trunks);
2528 lacp_destroy(bundle->lacp);
2529 bond_destroy(bundle->bond);
2534 bundle_set(struct ofproto *ofproto_, void *aux,
2535 const struct ofproto_bundle_settings *s)
2537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2538 bool need_flush = false;
2539 struct ofport_dpif *port;
2540 struct ofbundle *bundle;
2541 unsigned long *trunks;
2547 bundle_destroy(bundle_lookup(ofproto, aux));
2551 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2552 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2554 bundle = bundle_lookup(ofproto, aux);
2556 bundle = xmalloc(sizeof *bundle);
2558 bundle->ofproto = ofproto;
2559 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2560 hash_pointer(aux, 0));
2562 bundle->name = NULL;
2564 list_init(&bundle->ports);
2565 bundle->vlan_mode = PORT_VLAN_TRUNK;
2567 bundle->trunks = NULL;
2568 bundle->use_priority_tags = s->use_priority_tags;
2569 bundle->lacp = NULL;
2570 bundle->bond = NULL;
2572 bundle->floodable = true;
2574 bundle->src_mirrors = 0;
2575 bundle->dst_mirrors = 0;
2576 bundle->mirror_out = 0;
2579 if (!bundle->name || strcmp(s->name, bundle->name)) {
2581 bundle->name = xstrdup(s->name);
2586 if (!bundle->lacp) {
2587 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2588 bundle->lacp = lacp_create();
2590 lacp_configure(bundle->lacp, s->lacp);
2592 lacp_destroy(bundle->lacp);
2593 bundle->lacp = NULL;
2596 /* Update set of ports. */
2598 for (i = 0; i < s->n_slaves; i++) {
2599 if (!bundle_add_port(bundle, s->slaves[i],
2600 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2604 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2605 struct ofport_dpif *next_port;
2607 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2608 for (i = 0; i < s->n_slaves; i++) {
2609 if (s->slaves[i] == port->up.ofp_port) {
2614 bundle_del_port(port);
2618 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2620 if (list_is_empty(&bundle->ports)) {
2621 bundle_destroy(bundle);
2625 /* Set VLAN tagging mode */
2626 if (s->vlan_mode != bundle->vlan_mode
2627 || s->use_priority_tags != bundle->use_priority_tags) {
2628 bundle->vlan_mode = s->vlan_mode;
2629 bundle->use_priority_tags = s->use_priority_tags;
2634 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2635 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2637 if (vlan != bundle->vlan) {
2638 bundle->vlan = vlan;
2642 /* Get trunked VLANs. */
2643 switch (s->vlan_mode) {
2644 case PORT_VLAN_ACCESS:
2648 case PORT_VLAN_TRUNK:
2649 trunks = CONST_CAST(unsigned long *, s->trunks);
2652 case PORT_VLAN_NATIVE_UNTAGGED:
2653 case PORT_VLAN_NATIVE_TAGGED:
2654 if (vlan != 0 && (!s->trunks
2655 || !bitmap_is_set(s->trunks, vlan)
2656 || bitmap_is_set(s->trunks, 0))) {
2657 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2659 trunks = bitmap_clone(s->trunks, 4096);
2661 trunks = bitmap_allocate1(4096);
2663 bitmap_set1(trunks, vlan);
2664 bitmap_set0(trunks, 0);
2666 trunks = CONST_CAST(unsigned long *, s->trunks);
2673 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2674 free(bundle->trunks);
2675 if (trunks == s->trunks) {
2676 bundle->trunks = vlan_bitmap_clone(trunks);
2678 bundle->trunks = trunks;
2683 if (trunks != s->trunks) {
2688 if (!list_is_short(&bundle->ports)) {
2689 bundle->ofproto->has_bonded_bundles = true;
2691 if (bond_reconfigure(bundle->bond, s->bond)) {
2692 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2695 bundle->bond = bond_create(s->bond);
2696 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2699 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2700 bond_slave_register(bundle->bond, port, port->up.netdev);
2703 bond_destroy(bundle->bond);
2704 bundle->bond = NULL;
2707 /* If we changed something that would affect MAC learning, un-learn
2708 * everything on this port and force flow revalidation. */
2710 bundle_flush_macs(bundle, false);
2717 bundle_remove(struct ofport *port_)
2719 struct ofport_dpif *port = ofport_dpif_cast(port_);
2720 struct ofbundle *bundle = port->bundle;
2723 bundle_del_port(port);
2724 if (list_is_empty(&bundle->ports)) {
2725 bundle_destroy(bundle);
2726 } else if (list_is_short(&bundle->ports)) {
2727 bond_destroy(bundle->bond);
2728 bundle->bond = NULL;
2734 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2736 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2737 struct ofport_dpif *port = port_;
2738 uint8_t ea[ETH_ADDR_LEN];
2741 error = netdev_get_etheraddr(port->up.netdev, ea);
2743 struct ofpbuf packet;
2746 ofpbuf_init(&packet, 0);
2747 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2749 memcpy(packet_pdu, pdu, pdu_size);
2751 send_packet(port, &packet);
2752 ofpbuf_uninit(&packet);
2754 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2755 "%s (%s)", port->bundle->name,
2756 netdev_get_name(port->up.netdev), strerror(error));
2761 bundle_send_learning_packets(struct ofbundle *bundle)
2763 struct ofproto_dpif *ofproto = bundle->ofproto;
2764 int error, n_packets, n_errors;
2765 struct mac_entry *e;
2767 error = n_packets = n_errors = 0;
2768 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2769 if (e->port.p != bundle) {
2770 struct ofpbuf *learning_packet;
2771 struct ofport_dpif *port;
2775 /* The assignment to "port" is unnecessary but makes "grep"ing for
2776 * struct ofport_dpif more effective. */
2777 learning_packet = bond_compose_learning_packet(bundle->bond,
2781 ret = send_packet(port, learning_packet);
2782 ofpbuf_delete(learning_packet);
2792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2793 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2794 "packets, last error was: %s",
2795 bundle->name, n_errors, n_packets, strerror(error));
2797 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2798 bundle->name, n_packets);
2803 bundle_run(struct ofbundle *bundle)
2806 lacp_run(bundle->lacp, send_pdu_cb);
2809 struct ofport_dpif *port;
2811 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2812 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2815 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2816 lacp_status(bundle->lacp));
2817 if (bond_should_send_learning_packets(bundle->bond)) {
2818 bundle_send_learning_packets(bundle);
2824 bundle_wait(struct ofbundle *bundle)
2827 lacp_wait(bundle->lacp);
2830 bond_wait(bundle->bond);
2837 mirror_scan(struct ofproto_dpif *ofproto)
2841 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2842 if (!ofproto->mirrors[idx]) {
2849 static struct ofmirror *
2850 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2854 for (i = 0; i < MAX_MIRRORS; i++) {
2855 struct ofmirror *mirror = ofproto->mirrors[i];
2856 if (mirror && mirror->aux == aux) {
2864 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2866 mirror_update_dups(struct ofproto_dpif *ofproto)
2870 for (i = 0; i < MAX_MIRRORS; i++) {
2871 struct ofmirror *m = ofproto->mirrors[i];
2874 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2878 for (i = 0; i < MAX_MIRRORS; i++) {
2879 struct ofmirror *m1 = ofproto->mirrors[i];
2886 for (j = i + 1; j < MAX_MIRRORS; j++) {
2887 struct ofmirror *m2 = ofproto->mirrors[j];
2889 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2890 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2891 m2->dup_mirrors |= m1->dup_mirrors;
2898 mirror_set(struct ofproto *ofproto_, void *aux,
2899 const struct ofproto_mirror_settings *s)
2901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2902 mirror_mask_t mirror_bit;
2903 struct ofbundle *bundle;
2904 struct ofmirror *mirror;
2905 struct ofbundle *out;
2906 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2907 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2910 mirror = mirror_lookup(ofproto, aux);
2912 mirror_destroy(mirror);
2918 idx = mirror_scan(ofproto);
2920 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2922 ofproto->up.name, MAX_MIRRORS, s->name);
2926 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2927 mirror->ofproto = ofproto;
2930 mirror->out_vlan = -1;
2931 mirror->name = NULL;
2934 if (!mirror->name || strcmp(s->name, mirror->name)) {
2936 mirror->name = xstrdup(s->name);
2939 /* Get the new configuration. */
2940 if (s->out_bundle) {
2941 out = bundle_lookup(ofproto, s->out_bundle);
2943 mirror_destroy(mirror);
2949 out_vlan = s->out_vlan;
2951 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2952 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2954 /* If the configuration has not changed, do nothing. */
2955 if (hmapx_equals(&srcs, &mirror->srcs)
2956 && hmapx_equals(&dsts, &mirror->dsts)
2957 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2958 && mirror->out == out
2959 && mirror->out_vlan == out_vlan)
2961 hmapx_destroy(&srcs);
2962 hmapx_destroy(&dsts);
2966 hmapx_swap(&srcs, &mirror->srcs);
2967 hmapx_destroy(&srcs);
2969 hmapx_swap(&dsts, &mirror->dsts);
2970 hmapx_destroy(&dsts);
2972 free(mirror->vlans);
2973 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2976 mirror->out_vlan = out_vlan;
2978 /* Update bundles. */
2979 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2980 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2981 if (hmapx_contains(&mirror->srcs, bundle)) {
2982 bundle->src_mirrors |= mirror_bit;
2984 bundle->src_mirrors &= ~mirror_bit;
2987 if (hmapx_contains(&mirror->dsts, bundle)) {
2988 bundle->dst_mirrors |= mirror_bit;
2990 bundle->dst_mirrors &= ~mirror_bit;
2993 if (mirror->out == bundle) {
2994 bundle->mirror_out |= mirror_bit;
2996 bundle->mirror_out &= ~mirror_bit;
3000 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3001 ofproto->has_mirrors = true;
3002 mac_learning_flush(ofproto->ml,
3003 &ofproto->backer->revalidate_set);
3004 mirror_update_dups(ofproto);
3010 mirror_destroy(struct ofmirror *mirror)
3012 struct ofproto_dpif *ofproto;
3013 mirror_mask_t mirror_bit;
3014 struct ofbundle *bundle;
3021 ofproto = mirror->ofproto;
3022 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3023 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3025 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3026 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3027 bundle->src_mirrors &= ~mirror_bit;
3028 bundle->dst_mirrors &= ~mirror_bit;
3029 bundle->mirror_out &= ~mirror_bit;
3032 hmapx_destroy(&mirror->srcs);
3033 hmapx_destroy(&mirror->dsts);
3034 free(mirror->vlans);
3036 ofproto->mirrors[mirror->idx] = NULL;
3040 mirror_update_dups(ofproto);
3042 ofproto->has_mirrors = false;
3043 for (i = 0; i < MAX_MIRRORS; i++) {
3044 if (ofproto->mirrors[i]) {
3045 ofproto->has_mirrors = true;
3052 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3053 uint64_t *packets, uint64_t *bytes)
3055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3056 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3059 *packets = *bytes = UINT64_MAX;
3065 *packets = mirror->packet_count;
3066 *bytes = mirror->byte_count;
3072 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3075 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3076 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3082 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3084 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3085 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3086 return bundle && bundle->mirror_out != 0;
3090 forward_bpdu_changed(struct ofproto *ofproto_)
3092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3093 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3097 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 mac_learning_set_idle_time(ofproto->ml, idle_time);
3102 mac_learning_set_max_entries(ofproto->ml, max_entries);
3107 static struct ofport_dpif *
3108 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3110 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3111 return ofport ? ofport_dpif_cast(ofport) : NULL;
3114 static struct ofport_dpif *
3115 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3117 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3118 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3122 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3123 struct ofproto_port *ofproto_port,
3124 struct dpif_port *dpif_port)
3126 ofproto_port->name = dpif_port->name;
3127 ofproto_port->type = dpif_port->type;
3128 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3131 static struct ofport_dpif *
3132 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3134 const struct ofproto_dpif *ofproto;
3137 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3142 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3143 struct ofport *ofport;
3145 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3146 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3147 return ofport_dpif_cast(ofport);
3154 port_run_fast(struct ofport_dpif *ofport)
3156 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3157 struct ofpbuf packet;
3159 ofpbuf_init(&packet, 0);
3160 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3161 send_packet(ofport, &packet);
3162 ofpbuf_uninit(&packet);
3165 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3166 struct ofpbuf packet;
3168 ofpbuf_init(&packet, 0);
3169 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3170 send_packet(ofport, &packet);
3171 ofpbuf_uninit(&packet);
3176 port_run(struct ofport_dpif *ofport)
3178 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3179 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3180 bool enable = netdev_get_carrier(ofport->up.netdev);
3182 ofport->carrier_seq = carrier_seq;
3184 port_run_fast(ofport);
3186 if (ofport->tnl_port
3187 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3188 &ofport->tnl_port)) {
3189 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3193 int cfm_opup = cfm_get_opup(ofport->cfm);
3195 cfm_run(ofport->cfm);
3196 enable = enable && !cfm_get_fault(ofport->cfm);
3198 if (cfm_opup >= 0) {
3199 enable = enable && cfm_opup;
3204 bfd_run(ofport->bfd);
3205 enable = enable && bfd_forwarding(ofport->bfd);
3208 if (ofport->bundle) {
3209 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3210 if (carrier_changed) {
3211 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3215 if (ofport->may_enable != enable) {
3216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3218 if (ofproto->has_bundle_action) {
3219 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3223 ofport->may_enable = enable;
3227 port_wait(struct ofport_dpif *ofport)
3230 cfm_wait(ofport->cfm);
3234 bfd_wait(ofport->bfd);
3239 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3240 struct ofproto_port *ofproto_port)
3242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3243 struct dpif_port dpif_port;
3246 if (sset_contains(&ofproto->ghost_ports, devname)) {
3247 const char *type = netdev_get_type_from_name(devname);
3249 /* We may be called before ofproto->up.port_by_name is populated with
3250 * the appropriate ofport. For this reason, we must get the name and
3251 * type from the netdev layer directly. */
3253 const struct ofport *ofport;
3255 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3256 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3257 ofproto_port->name = xstrdup(devname);
3258 ofproto_port->type = xstrdup(type);
3264 if (!sset_contains(&ofproto->ports, devname)) {
3267 error = dpif_port_query_by_name(ofproto->backer->dpif,
3268 devname, &dpif_port);
3270 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3276 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3278 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3279 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3280 const char *devname = netdev_get_name(netdev);
3282 if (netdev_vport_is_patch(netdev)) {
3283 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3287 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3288 uint32_t port_no = UINT32_MAX;
3291 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3295 if (netdev_get_tunnel_config(netdev)) {
3296 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3300 if (netdev_get_tunnel_config(netdev)) {
3301 sset_add(&ofproto->ghost_ports, devname);
3303 sset_add(&ofproto->ports, devname);
3309 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3312 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3319 sset_find_and_delete(&ofproto->ghost_ports,
3320 netdev_get_name(ofport->up.netdev));
3321 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3322 if (!ofport->tnl_port) {
3323 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3325 /* The caller is going to close ofport->up.netdev. If this is a
3326 * bonded port, then the bond is using that netdev, so remove it
3327 * from the bond. The client will need to reconfigure everything
3328 * after deleting ports, so then the slave will get re-added. */
3329 bundle_remove(&ofport->up);
3336 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3338 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3343 error = netdev_get_stats(ofport->up.netdev, stats);
3345 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3348 /* ofproto->stats.tx_packets represents packets that we created
3349 * internally and sent to some port (e.g. packets sent with
3350 * send_packet()). Account for them as if they had come from
3351 * OFPP_LOCAL and got forwarded. */
3353 if (stats->rx_packets != UINT64_MAX) {
3354 stats->rx_packets += ofproto->stats.tx_packets;
3357 if (stats->rx_bytes != UINT64_MAX) {
3358 stats->rx_bytes += ofproto->stats.tx_bytes;
3361 /* ofproto->stats.rx_packets represents packets that were received on
3362 * some port and we processed internally and dropped (e.g. STP).
3363 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3365 if (stats->tx_packets != UINT64_MAX) {
3366 stats->tx_packets += ofproto->stats.rx_packets;
3369 if (stats->tx_bytes != UINT64_MAX) {
3370 stats->tx_bytes += ofproto->stats.rx_bytes;
3377 /* Account packets for LOCAL port. */
3379 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3380 size_t tx_size, size_t rx_size)
3382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3385 ofproto->stats.rx_packets++;
3386 ofproto->stats.rx_bytes += rx_size;
3389 ofproto->stats.tx_packets++;
3390 ofproto->stats.tx_bytes += tx_size;
3394 struct port_dump_state {
3399 struct ofproto_port port;
3404 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3406 *statep = xzalloc(sizeof(struct port_dump_state));
3411 port_dump_next(const struct ofproto *ofproto_, void *state_,
3412 struct ofproto_port *port)
3414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3415 struct port_dump_state *state = state_;
3416 const struct sset *sset;
3417 struct sset_node *node;
3419 if (state->has_port) {
3420 ofproto_port_destroy(&state->port);
3421 state->has_port = false;
3423 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3424 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3427 error = port_query_by_name(ofproto_, node->name, &state->port);
3429 *port = state->port;
3430 state->has_port = true;
3432 } else if (error != ENODEV) {
3437 if (!state->ghost) {
3438 state->ghost = true;
3441 return port_dump_next(ofproto_, state_, port);
3448 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3450 struct port_dump_state *state = state_;
3452 if (state->has_port) {
3453 ofproto_port_destroy(&state->port);
3460 port_poll(const struct ofproto *ofproto_, char **devnamep)
3462 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3464 if (ofproto->port_poll_errno) {
3465 int error = ofproto->port_poll_errno;
3466 ofproto->port_poll_errno = 0;
3470 if (sset_is_empty(&ofproto->port_poll_set)) {
3474 *devnamep = sset_pop(&ofproto->port_poll_set);
3479 port_poll_wait(const struct ofproto *ofproto_)
3481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3482 dpif_port_poll_wait(ofproto->backer->dpif);
3486 port_is_lacp_current(const struct ofport *ofport_)
3488 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3489 return (ofport->bundle && ofport->bundle->lacp
3490 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3494 /* Upcall handling. */
3496 /* Flow miss batching.
3498 * Some dpifs implement operations faster when you hand them off in a batch.
3499 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3500 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3501 * more packets, plus possibly installing the flow in the dpif.
3503 * So far we only batch the operations that affect flow setup time the most.
3504 * It's possible to batch more than that, but the benefit might be minimal. */
3506 struct hmap_node hmap_node;
3507 struct ofproto_dpif *ofproto;
3509 enum odp_key_fitness key_fitness;
3510 const struct nlattr *key;
3512 struct initial_vals initial_vals;
3513 struct list packets;
3514 enum dpif_upcall_type upcall_type;
3515 uint32_t odp_in_port;
3518 struct flow_miss_op {
3519 struct dpif_op dpif_op;
3520 void *garbage; /* Pointer to pass to free(), NULL if none. */
3521 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3524 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3525 * OpenFlow controller as necessary according to their individual
3526 * configurations. */
3528 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3529 const struct flow *flow)
3531 struct ofputil_packet_in pin;
3533 pin.packet = packet->data;
3534 pin.packet_len = packet->size;
3535 pin.reason = OFPR_NO_MATCH;
3536 pin.controller_id = 0;
3541 pin.send_len = 0; /* not used for flow table misses */
3543 flow_get_metadata(flow, &pin.fmd);
3545 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3548 static enum slow_path_reason
3549 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3550 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3554 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3556 cfm_process_heartbeat(ofport->cfm, packet);
3559 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3561 bfd_process_packet(ofport->bfd, flow, packet);
3564 } else if (ofport->bundle && ofport->bundle->lacp
3565 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3567 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3570 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3572 stp_process_packet(ofport, packet);
3580 static struct flow_miss *
3581 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3582 const struct flow *flow, uint32_t hash)
3584 struct flow_miss *miss;
3586 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3587 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3595 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3596 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3597 * 'miss' is associated with a subfacet the caller must also initialize the
3598 * returned op->subfacet, and if anything needs to be freed after processing
3599 * the op, the caller must initialize op->garbage also. */
3601 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3602 struct flow_miss_op *op)
3604 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3605 /* This packet was received on a VLAN splinter port. We
3606 * added a VLAN to the packet to make the packet resemble
3607 * the flow, but the actions were composed assuming that
3608 * the packet contained no VLAN. So, we must remove the
3609 * VLAN header from the packet before trying to execute the
3611 eth_pop_vlan(packet);
3615 op->dpif_op.type = DPIF_OP_EXECUTE;
3616 op->dpif_op.u.execute.key = miss->key;
3617 op->dpif_op.u.execute.key_len = miss->key_len;
3618 op->dpif_op.u.execute.packet = packet;
3621 /* Helper for handle_flow_miss_without_facet() and
3622 * handle_flow_miss_with_facet(). */
3624 handle_flow_miss_common(struct rule_dpif *rule,
3625 struct ofpbuf *packet, const struct flow *flow)
3627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3629 ofproto->n_matches++;
3631 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3633 * Extra-special case for fail-open mode.
3635 * We are in fail-open mode and the packet matched the fail-open
3636 * rule, but we are connected to a controller too. We should send
3637 * the packet up to the controller in the hope that it will try to
3638 * set up a flow and thereby allow us to exit fail-open.
3640 * See the top-level comment in fail-open.c for more information.
3642 send_packet_in_miss(ofproto, packet, flow);
3646 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3647 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3648 * installing a datapath flow. The answer is usually "yes" (a return value of
3649 * true). However, for short flows the cost of bookkeeping is much higher than
3650 * the benefits, so when the datapath holds a large number of flows we impose
3651 * some heuristics to decide which flows are likely to be worth tracking. */
3653 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3654 struct flow_miss *miss, uint32_t hash)
3656 if (!ofproto->governor) {
3659 n_subfacets = hmap_count(&ofproto->subfacets);
3660 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3664 ofproto->governor = governor_create(ofproto->up.name);
3667 return governor_should_install_flow(ofproto->governor, hash,
3668 list_size(&miss->packets));
3671 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3672 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3673 * increment '*n_ops'. */
3675 handle_flow_miss_without_facet(struct flow_miss *miss,
3676 struct rule_dpif *rule,
3677 struct flow_miss_op *ops, size_t *n_ops)
3679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3680 long long int now = time_msec();
3681 struct action_xlate_ctx ctx;
3682 struct ofpbuf *packet;
3684 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3685 struct flow_miss_op *op = &ops[*n_ops];
3686 struct dpif_flow_stats stats;
3687 struct ofpbuf odp_actions;
3689 COVERAGE_INC(facet_suppress);
3691 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3693 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3694 rule_credit_stats(rule, &stats);
3696 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3697 &miss->initial_vals, rule, 0, packet);
3698 ctx.resubmit_stats = &stats;
3699 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3702 if (odp_actions.size) {
3703 struct dpif_execute *execute = &op->dpif_op.u.execute;
3705 init_flow_miss_execute_op(miss, packet, op);
3706 execute->actions = odp_actions.data;
3707 execute->actions_len = odp_actions.size;
3708 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3712 ofpbuf_uninit(&odp_actions);
3717 /* Handles 'miss', which matches 'facet'. May add any required datapath
3718 * operations to 'ops', incrementing '*n_ops' for each new op.
3720 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3721 * This is really important only for new facets: if we just called time_msec()
3722 * here, then the new subfacet or its packets could look (occasionally) as
3723 * though it was used some time after the facet was used. That can make a
3724 * one-packet flow look like it has a nonzero duration, which looks odd in
3725 * e.g. NetFlow statistics. */
3727 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3729 struct flow_miss_op *ops, size_t *n_ops)
3731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3732 enum subfacet_path want_path;
3733 struct subfacet *subfacet;
3734 struct ofpbuf *packet;
3736 subfacet = subfacet_create(facet, miss, now);
3738 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3739 struct flow_miss_op *op = &ops[*n_ops];
3740 struct dpif_flow_stats stats;
3741 struct ofpbuf odp_actions;
3743 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3745 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3746 if (!subfacet->actions || subfacet->slow) {
3747 subfacet_make_actions(subfacet, packet, &odp_actions);
3750 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3751 subfacet_update_stats(subfacet, &stats);
3753 if (subfacet->actions_len) {
3754 struct dpif_execute *execute = &op->dpif_op.u.execute;
3756 init_flow_miss_execute_op(miss, packet, op);
3757 if (!subfacet->slow) {
3758 execute->actions = subfacet->actions;
3759 execute->actions_len = subfacet->actions_len;
3760 ofpbuf_uninit(&odp_actions);
3762 execute->actions = odp_actions.data;
3763 execute->actions_len = odp_actions.size;
3764 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3769 ofpbuf_uninit(&odp_actions);
3773 want_path = subfacet_want_path(subfacet->slow);
3774 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3775 struct flow_miss_op *op = &ops[(*n_ops)++];
3776 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3778 subfacet->path = want_path;
3781 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3782 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3783 put->key = miss->key;
3784 put->key_len = miss->key_len;
3785 if (want_path == SF_FAST_PATH) {
3786 put->actions = subfacet->actions;
3787 put->actions_len = subfacet->actions_len;
3789 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3790 op->stub, sizeof op->stub,
3791 &put->actions, &put->actions_len);
3797 /* Handles flow miss 'miss'. May add any required datapath operations
3798 * to 'ops', incrementing '*n_ops' for each new op. */
3800 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3803 struct ofproto_dpif *ofproto = miss->ofproto;
3804 struct facet *facet;
3808 /* The caller must ensure that miss->hmap_node.hash contains
3809 * flow_hash(miss->flow, 0). */
3810 hash = miss->hmap_node.hash;
3812 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3814 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3816 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3817 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3821 facet = facet_create(rule, &miss->flow, hash);
3826 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3829 static struct drop_key *
3830 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3833 struct drop_key *drop_key;
3835 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3836 &backer->drop_keys) {
3837 if (drop_key->key_len == key_len
3838 && !memcmp(drop_key->key, key, key_len)) {
3846 drop_key_clear(struct dpif_backer *backer)
3848 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3849 struct drop_key *drop_key, *next;
3851 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3854 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3856 if (error && !VLOG_DROP_WARN(&rl)) {
3857 struct ds ds = DS_EMPTY_INITIALIZER;
3858 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3859 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3864 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3865 free(drop_key->key);
3870 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3871 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3872 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3873 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3874 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3875 * 'packet' ingressed.
3877 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3878 * 'flow''s in_port to OFPP_NONE.
3880 * This function does post-processing on data returned from
3881 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3882 * of the upcall processing logic. In particular, if the extracted in_port is
3883 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3884 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3885 * a VLAN header onto 'packet' (if it is nonnull).
3887 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3888 * to the VLAN TCI with which the packet was really received, that is, the
3889 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3890 * the value returned in flow->vlan_tci only for packets received on
3893 * Similarly, this function also includes some logic to help with tunnels. It
3894 * may modify 'flow' as necessary to make the tunneling implementation
3895 * transparent to the upcall processing logic.
3897 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3898 * or some other positive errno if there are other problems. */
3900 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3901 const struct nlattr *key, size_t key_len,
3902 struct flow *flow, enum odp_key_fitness *fitnessp,
3903 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3904 struct initial_vals *initial_vals)
3906 const struct ofport_dpif *port;
3907 enum odp_key_fitness fitness;
3910 fitness = odp_flow_key_to_flow(key, key_len, flow);
3911 if (fitness == ODP_FIT_ERROR) {
3917 initial_vals->vlan_tci = flow->vlan_tci;
3921 *odp_in_port = flow->in_port;
3924 port = (tnl_port_should_receive(flow)
3925 ? ofport_dpif_cast(tnl_port_receive(flow))
3926 : odp_port_to_ofport(backer, flow->in_port));
3927 flow->in_port = port ? port->up.ofp_port : OFPP_NONE;
3932 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3933 * it's theoretically possible that we'll receive an ofport belonging to an
3934 * entirely different datapath. In practice, this can't happen because no
3935 * platforms has two separate datapaths which each support tunneling. */
3936 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3938 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3940 /* Make the packet resemble the flow, so that it gets sent to
3941 * an OpenFlow controller properly, so that it looks correct
3942 * for sFlow, and so that flow_extract() will get the correct
3943 * vlan_tci if it is called on 'packet'.
3945 * The allocated space inside 'packet' probably also contains
3946 * 'key', that is, both 'packet' and 'key' are probably part of
3947 * a struct dpif_upcall (see the large comment on that
3948 * structure definition), so pushing data on 'packet' is in
3949 * general not a good idea since it could overwrite 'key' or
3950 * free it as a side effect. However, it's OK in this special
3951 * case because we know that 'packet' is inside a Netlink
3952 * attribute: pushing 4 bytes will just overwrite the 4-byte
3953 * "struct nlattr", which is fine since we don't need that
3954 * header anymore. */
3955 eth_push_vlan(packet, flow->vlan_tci);
3957 /* We can't reproduce 'key' from 'flow'. */
3958 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3963 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3968 *fitnessp = fitness;
3974 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3977 struct dpif_upcall *upcall;
3978 struct flow_miss *miss;
3979 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3980 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3981 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3991 /* Construct the to-do list.
3993 * This just amounts to extracting the flow from each packet and sticking
3994 * the packets that have the same flow in the same "flow_miss" structure so
3995 * that we can process them together. */
3998 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3999 struct flow_miss *miss = &misses[n_misses];
4000 struct flow_miss *existing_miss;
4001 struct ofproto_dpif *ofproto;
4002 uint32_t odp_in_port;
4007 error = ofproto_receive(backer, upcall->packet, upcall->key,
4008 upcall->key_len, &flow, &miss->key_fitness,
4009 &ofproto, &odp_in_port, &miss->initial_vals);
4010 if (error == ENODEV) {
4011 struct drop_key *drop_key;
4013 /* Received packet on port for which we couldn't associate
4014 * an ofproto. This can happen if a port is removed while
4015 * traffic is being received. Print a rate-limited message
4016 * in case it happens frequently. Install a drop flow so
4017 * that future packets of the flow are inexpensively dropped
4019 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4022 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4024 drop_key = xmalloc(sizeof *drop_key);
4025 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4026 drop_key->key_len = upcall->key_len;
4028 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4029 hash_bytes(drop_key->key, drop_key->key_len, 0));
4030 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4031 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4039 ofproto->n_missed++;
4040 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4041 &flow.tunnel, flow.in_port, &miss->flow);
4043 /* Add other packets to a to-do list. */
4044 hash = flow_hash(&miss->flow, 0);
4045 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4046 if (!existing_miss) {
4047 hmap_insert(&todo, &miss->hmap_node, hash);
4048 miss->ofproto = ofproto;
4049 miss->key = upcall->key;
4050 miss->key_len = upcall->key_len;
4051 miss->upcall_type = upcall->type;
4052 miss->odp_in_port = odp_in_port;
4053 list_init(&miss->packets);
4057 miss = existing_miss;
4059 list_push_back(&miss->packets, &upcall->packet->list_node);
4062 /* Process each element in the to-do list, constructing the set of
4063 * operations to batch. */
4065 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4066 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4068 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4070 /* Execute batch. */
4071 for (i = 0; i < n_ops; i++) {
4072 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4074 dpif_operate(backer->dpif, dpif_ops, n_ops);
4077 for (i = 0; i < n_ops; i++) {
4078 free(flow_miss_ops[i].garbage);
4080 hmap_destroy(&todo);
4083 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4085 classify_upcall(const struct dpif_upcall *upcall)
4087 size_t userdata_len;
4088 union user_action_cookie cookie;
4090 /* First look at the upcall type. */
4091 switch (upcall->type) {
4092 case DPIF_UC_ACTION:
4098 case DPIF_N_UC_TYPES:
4100 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4104 /* "action" upcalls need a closer look. */
4105 if (!upcall->userdata) {
4106 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4109 userdata_len = nl_attr_get_size(upcall->userdata);
4110 if (userdata_len < sizeof cookie.type
4111 || userdata_len > sizeof cookie) {
4112 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4116 memset(&cookie, 0, sizeof cookie);
4117 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4118 if (userdata_len == sizeof cookie.sflow
4119 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4120 return SFLOW_UPCALL;
4121 } else if (userdata_len == sizeof cookie.slow_path
4122 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4124 } else if (userdata_len == sizeof cookie.flow_sample
4125 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4126 return FLOW_SAMPLE_UPCALL;
4127 } else if (userdata_len == sizeof cookie.ipfix
4128 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4129 return IPFIX_UPCALL;
4131 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4132 " and size %zu", cookie.type, userdata_len);
4138 handle_sflow_upcall(struct dpif_backer *backer,
4139 const struct dpif_upcall *upcall)
4141 struct ofproto_dpif *ofproto;
4142 union user_action_cookie cookie;
4144 uint32_t odp_in_port;
4146 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4147 &flow, NULL, &ofproto, &odp_in_port, NULL)
4148 || !ofproto->sflow) {
4152 memset(&cookie, 0, sizeof cookie);
4153 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4154 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4155 odp_in_port, &cookie);
4159 handle_flow_sample_upcall(struct dpif_backer *backer,
4160 const struct dpif_upcall *upcall)
4162 struct ofproto_dpif *ofproto;
4163 union user_action_cookie cookie;
4166 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4167 &flow, NULL, &ofproto, NULL, NULL)
4168 || !ofproto->ipfix) {
4172 memset(&cookie, 0, sizeof cookie);
4173 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4175 /* The flow reflects exactly the contents of the packet. Sample
4176 * the packet using it. */
4177 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4178 cookie.flow_sample.collector_set_id,
4179 cookie.flow_sample.probability,
4180 cookie.flow_sample.obs_domain_id,
4181 cookie.flow_sample.obs_point_id);
4185 handle_ipfix_upcall(struct dpif_backer *backer,
4186 const struct dpif_upcall *upcall)
4188 struct ofproto_dpif *ofproto;
4191 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4192 &flow, NULL, &ofproto, NULL, NULL)
4193 || !ofproto->ipfix) {
4197 /* The flow reflects exactly the contents of the packet. Sample
4198 * the packet using it. */
4199 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4203 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4205 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4206 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4207 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4212 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4215 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4216 struct dpif_upcall *upcall = &misses[n_misses];
4217 struct ofpbuf *buf = &miss_bufs[n_misses];
4220 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4221 sizeof miss_buf_stubs[n_misses]);
4222 error = dpif_recv(backer->dpif, upcall, buf);
4228 switch (classify_upcall(upcall)) {
4230 /* Handle it later. */
4235 handle_sflow_upcall(backer, upcall);
4239 case FLOW_SAMPLE_UPCALL:
4240 handle_flow_sample_upcall(backer, upcall);
4245 handle_ipfix_upcall(backer, upcall);
4255 /* Handle deferred MISS_UPCALL processing. */
4256 handle_miss_upcalls(backer, misses, n_misses);
4257 for (i = 0; i < n_misses; i++) {
4258 ofpbuf_uninit(&miss_bufs[i]);
4264 /* Flow expiration. */
4266 static int subfacet_max_idle(const struct ofproto_dpif *);
4267 static void update_stats(struct dpif_backer *);
4268 static void rule_expire(struct rule_dpif *);
4269 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4271 /* This function is called periodically by run(). Its job is to collect
4272 * updates for the flows that have been installed into the datapath, most
4273 * importantly when they last were used, and then use that information to
4274 * expire flows that have not been used recently.
4276 * Returns the number of milliseconds after which it should be called again. */
4278 expire(struct dpif_backer *backer)
4280 struct ofproto_dpif *ofproto;
4281 int max_idle = INT32_MAX;
4283 /* Periodically clear out the drop keys in an effort to keep them
4284 * relatively few. */
4285 drop_key_clear(backer);
4287 /* Update stats for each flow in the backer. */
4288 update_stats(backer);
4290 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4291 struct rule *rule, *next_rule;
4294 if (ofproto->backer != backer) {
4298 /* Keep track of the max number of flows per ofproto_dpif. */
4299 update_max_subfacet_count(ofproto);
4301 /* Expire subfacets that have been idle too long. */
4302 dp_max_idle = subfacet_max_idle(ofproto);
4303 expire_subfacets(ofproto, dp_max_idle);
4305 max_idle = MIN(max_idle, dp_max_idle);
4307 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4309 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4310 &ofproto->up.expirable) {
4311 rule_expire(rule_dpif_cast(rule));
4314 /* All outstanding data in existing flows has been accounted, so it's a
4315 * good time to do bond rebalancing. */
4316 if (ofproto->has_bonded_bundles) {
4317 struct ofbundle *bundle;
4319 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4321 bond_rebalance(bundle->bond, &backer->revalidate_set);
4327 return MIN(max_idle, 1000);
4330 /* Updates flow table statistics given that the datapath just reported 'stats'
4331 * as 'subfacet''s statistics. */
4333 update_subfacet_stats(struct subfacet *subfacet,
4334 const struct dpif_flow_stats *stats)
4336 struct facet *facet = subfacet->facet;
4338 if (stats->n_packets >= subfacet->dp_packet_count) {
4339 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4340 facet->packet_count += extra;
4342 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4345 if (stats->n_bytes >= subfacet->dp_byte_count) {
4346 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4348 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4351 subfacet->dp_packet_count = stats->n_packets;
4352 subfacet->dp_byte_count = stats->n_bytes;
4354 facet->tcp_flags |= stats->tcp_flags;
4356 subfacet_update_time(subfacet, stats->used);
4357 if (facet->accounted_bytes < facet->byte_count) {
4359 facet_account(facet);
4360 facet->accounted_bytes = facet->byte_count;
4364 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4365 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4367 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4368 const struct nlattr *key, size_t key_len)
4370 if (!VLOG_DROP_WARN(&rl)) {
4374 odp_flow_key_format(key, key_len, &s);
4375 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4379 COVERAGE_INC(facet_unexpected);
4380 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4383 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4385 * This function also pushes statistics updates to rules which each facet
4386 * resubmits into. Generally these statistics will be accurate. However, if a
4387 * facet changes the rule it resubmits into at some time in between
4388 * update_stats() runs, it is possible that statistics accrued to the
4389 * old rule will be incorrectly attributed to the new rule. This could be
4390 * avoided by calling update_stats() whenever rules are created or
4391 * deleted. However, the performance impact of making so many calls to the
4392 * datapath do not justify the benefit of having perfectly accurate statistics.
4394 * In addition, this function maintains per ofproto flow hit counts. The patch
4395 * port is not treated specially. e.g. A packet ingress from br0 patched into
4396 * br1 will increase the hit count of br0 by 1, however, does not affect
4397 * the hit or miss counts of br1.
4400 update_stats(struct dpif_backer *backer)
4402 const struct dpif_flow_stats *stats;
4403 struct dpif_flow_dump dump;
4404 const struct nlattr *key;
4405 struct ofproto_dpif *ofproto;
4408 dpif_flow_dump_start(&dump, backer->dpif);
4409 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4411 struct subfacet *subfacet;
4412 struct ofport_dpif *ofport;
4415 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4420 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4421 ofproto->n_update_stats++;
4423 ofport = get_ofp_port(ofproto, flow.in_port);
4424 if (ofport && ofport->tnl_port) {
4425 netdev_vport_inc_rx(ofport->up.netdev, stats);
4428 key_hash = odp_flow_key_hash(key, key_len);
4429 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4430 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4432 /* Update ofproto_dpif's hit count. */
4433 if (stats->n_packets > subfacet->dp_packet_count) {
4434 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4435 dpif_stats_update_hit_count(ofproto, delta);
4438 update_subfacet_stats(subfacet, stats);
4442 /* Stats are updated per-packet. */
4445 case SF_NOT_INSTALLED:
4447 delete_unexpected_flow(ofproto, key, key_len);
4452 dpif_flow_dump_done(&dump);
4454 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4455 update_moving_averages(ofproto);
4460 /* Calculates and returns the number of milliseconds of idle time after which
4461 * subfacets should expire from the datapath. When a subfacet expires, we fold
4462 * its statistics into its facet, and when a facet's last subfacet expires, we
4463 * fold its statistic into its rule. */
4465 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4468 * Idle time histogram.
4470 * Most of the time a switch has a relatively small number of subfacets.
4471 * When this is the case we might as well keep statistics for all of them
4472 * in userspace and to cache them in the kernel datapath for performance as
4475 * As the number of subfacets increases, the memory required to maintain
4476 * statistics about them in userspace and in the kernel becomes
4477 * significant. However, with a large number of subfacets it is likely
4478 * that only a few of them are "heavy hitters" that consume a large amount
4479 * of bandwidth. At this point, only heavy hitters are worth caching in
4480 * the kernel and maintaining in userspaces; other subfacets we can
4483 * The technique used to compute the idle time is to build a histogram with
4484 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4485 * that is installed in the kernel gets dropped in the appropriate bucket.
4486 * After the histogram has been built, we compute the cutoff so that only
4487 * the most-recently-used 1% of subfacets (but at least
4488 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4489 * the most-recently-used bucket of subfacets is kept, so actually an
4490 * arbitrary number of subfacets can be kept in any given expiration run
4491 * (though the next run will delete most of those unless they receive
4494 * This requires a second pass through the subfacets, in addition to the
4495 * pass made by update_stats(), because the former function never looks at
4496 * uninstallable subfacets.
4498 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4499 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4500 int buckets[N_BUCKETS] = { 0 };
4501 int total, subtotal, bucket;
4502 struct subfacet *subfacet;
4506 total = hmap_count(&ofproto->subfacets);
4507 if (total <= ofproto->up.flow_eviction_threshold) {
4508 return N_BUCKETS * BUCKET_WIDTH;
4511 /* Build histogram. */
4513 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4514 long long int idle = now - subfacet->used;
4515 int bucket = (idle <= 0 ? 0
4516 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4517 : (unsigned int) idle / BUCKET_WIDTH);
4521 /* Find the first bucket whose flows should be expired. */
4522 subtotal = bucket = 0;
4524 subtotal += buckets[bucket++];
4525 } while (bucket < N_BUCKETS &&
4526 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4528 if (VLOG_IS_DBG_ENABLED()) {
4532 ds_put_cstr(&s, "keep");
4533 for (i = 0; i < N_BUCKETS; i++) {
4535 ds_put_cstr(&s, ", drop");
4538 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4541 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4545 return bucket * BUCKET_WIDTH;
4549 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4551 /* Cutoff time for most flows. */
4552 long long int normal_cutoff = time_msec() - dp_max_idle;
4554 /* We really want to keep flows for special protocols around, so use a more
4555 * conservative cutoff. */
4556 long long int special_cutoff = time_msec() - 10000;
4558 struct subfacet *subfacet, *next_subfacet;
4559 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4563 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4564 &ofproto->subfacets) {
4565 long long int cutoff;
4567 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
4570 if (subfacet->used < cutoff) {
4571 if (subfacet->path != SF_NOT_INSTALLED) {
4572 batch[n_batch++] = subfacet;
4573 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4574 subfacet_destroy_batch(ofproto, batch, n_batch);
4578 subfacet_destroy(subfacet);
4584 subfacet_destroy_batch(ofproto, batch, n_batch);
4588 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4589 * then delete it entirely. */
4591 rule_expire(struct rule_dpif *rule)
4593 struct facet *facet, *next_facet;
4597 if (rule->up.pending) {
4598 /* We'll have to expire it later. */
4602 /* Has 'rule' expired? */
4604 if (rule->up.hard_timeout
4605 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4606 reason = OFPRR_HARD_TIMEOUT;
4607 } else if (rule->up.idle_timeout
4608 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4609 reason = OFPRR_IDLE_TIMEOUT;
4614 COVERAGE_INC(ofproto_dpif_expired);
4616 /* Update stats. (This is a no-op if the rule expired due to an idle
4617 * timeout, because that only happens when the rule has no facets left.) */
4618 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4619 facet_remove(facet);
4622 /* Get rid of the rule. */
4623 ofproto_rule_expire(&rule->up, reason);
4628 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4630 * The caller must already have determined that no facet with an identical
4631 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4632 * the ofproto's classifier table.
4634 * 'hash' must be the return value of flow_hash(flow, 0).
4636 * The facet will initially have no subfacets. The caller should create (at
4637 * least) one subfacet with subfacet_create(). */
4638 static struct facet *
4639 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4642 struct facet *facet;
4644 facet = xzalloc(sizeof *facet);
4645 facet->used = time_msec();
4646 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4647 list_push_back(&rule->facets, &facet->list_node);
4649 facet->flow = *flow;
4650 list_init(&facet->subfacets);
4651 netflow_flow_init(&facet->nf_flow);
4652 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4654 facet->learn_rl = time_msec() + 500;
4660 facet_free(struct facet *facet)
4665 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4666 * 'packet', which arrived on 'in_port'. */
4668 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4669 const struct nlattr *odp_actions, size_t actions_len,
4670 struct ofpbuf *packet)
4672 struct odputil_keybuf keybuf;
4676 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4677 odp_flow_key_from_flow(&key, flow,
4678 ofp_port_to_odp_port(ofproto, flow->in_port));
4680 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4681 odp_actions, actions_len, packet);
4685 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4687 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4688 * rule's statistics, via subfacet_uninstall().
4690 * - Removes 'facet' from its rule and from ofproto->facets.
4693 facet_remove(struct facet *facet)
4695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4696 struct subfacet *subfacet, *next_subfacet;
4698 ovs_assert(!list_is_empty(&facet->subfacets));
4700 /* First uninstall all of the subfacets to get final statistics. */
4701 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4702 subfacet_uninstall(subfacet);
4705 /* Flush the final stats to the rule.
4707 * This might require us to have at least one subfacet around so that we
4708 * can use its actions for accounting in facet_account(), which is why we
4709 * have uninstalled but not yet destroyed the subfacets. */
4710 facet_flush_stats(facet);
4712 /* Now we're really all done so destroy everything. */
4713 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4714 &facet->subfacets) {
4715 subfacet_destroy__(subfacet);
4717 hmap_remove(&ofproto->facets, &facet->hmap_node);
4718 list_remove(&facet->list_node);
4722 /* Feed information from 'facet' back into the learning table to keep it in
4723 * sync with what is actually flowing through the datapath. */
4725 facet_learn(struct facet *facet)
4727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4728 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4729 struct subfacet, list_node);
4730 long long int now = time_msec();
4731 struct action_xlate_ctx ctx;
4733 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4737 facet->learn_rl = now + 500;
4739 if (!facet->has_learn
4740 && !facet->has_normal
4741 && (!facet->has_fin_timeout
4742 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4746 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4747 &subfacet->initial_vals,
4748 facet->rule, facet->tcp_flags, NULL);
4749 ctx.may_learn = true;
4750 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4751 facet->rule->up.ofpacts_len);
4755 facet_account(struct facet *facet)
4757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4758 struct subfacet *subfacet = facet_get_subfacet(facet);
4759 const struct nlattr *a;
4764 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4767 n_bytes = facet->byte_count - facet->accounted_bytes;
4769 /* This loop feeds byte counters to bond_account() for rebalancing to use
4770 * as a basis. We also need to track the actual VLAN on which the packet
4771 * is going to be sent to ensure that it matches the one passed to
4772 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4775 * We use the actions from an arbitrary subfacet because they should all
4776 * be equally valid for our purpose. */
4777 vlan_tci = facet->flow.vlan_tci;
4778 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4779 subfacet->actions, subfacet->actions_len) {
4780 const struct ovs_action_push_vlan *vlan;
4781 struct ofport_dpif *port;
4783 switch (nl_attr_type(a)) {
4784 case OVS_ACTION_ATTR_OUTPUT:
4785 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4786 if (port && port->bundle && port->bundle->bond) {
4787 bond_account(port->bundle->bond, &facet->flow,
4788 vlan_tci_to_vid(vlan_tci), n_bytes);
4792 case OVS_ACTION_ATTR_POP_VLAN:
4793 vlan_tci = htons(0);
4796 case OVS_ACTION_ATTR_PUSH_VLAN:
4797 vlan = nl_attr_get(a);
4798 vlan_tci = vlan->vlan_tci;
4804 /* Returns true if the only action for 'facet' is to send to the controller.
4805 * (We don't report NetFlow expiration messages for such facets because they
4806 * are just part of the control logic for the network, not real traffic). */
4808 facet_is_controller_flow(struct facet *facet)
4811 const struct rule *rule = &facet->rule->up;
4812 const struct ofpact *ofpacts = rule->ofpacts;
4813 size_t ofpacts_len = rule->ofpacts_len;
4815 if (ofpacts_len > 0 &&
4816 ofpacts->type == OFPACT_CONTROLLER &&
4817 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4824 /* Folds all of 'facet''s statistics into its rule. Also updates the
4825 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4826 * 'facet''s statistics in the datapath should have been zeroed and folded into
4827 * its packet and byte counts before this function is called. */
4829 facet_flush_stats(struct facet *facet)
4831 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4832 struct subfacet *subfacet;
4834 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4835 ovs_assert(!subfacet->dp_byte_count);
4836 ovs_assert(!subfacet->dp_packet_count);
4839 facet_push_stats(facet);
4840 if (facet->accounted_bytes < facet->byte_count) {
4841 facet_account(facet);
4842 facet->accounted_bytes = facet->byte_count;
4845 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4846 struct ofexpired expired;
4847 expired.flow = facet->flow;
4848 expired.packet_count = facet->packet_count;
4849 expired.byte_count = facet->byte_count;
4850 expired.used = facet->used;
4851 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4854 facet->rule->packet_count += facet->packet_count;
4855 facet->rule->byte_count += facet->byte_count;
4857 /* Reset counters to prevent double counting if 'facet' ever gets
4859 facet_reset_counters(facet);
4861 netflow_flow_clear(&facet->nf_flow);
4862 facet->tcp_flags = 0;
4865 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4866 * Returns it if found, otherwise a null pointer.
4868 * 'hash' must be the return value of flow_hash(flow, 0).
4870 * The returned facet might need revalidation; use facet_lookup_valid()
4871 * instead if that is important. */
4872 static struct facet *
4873 facet_find(struct ofproto_dpif *ofproto,
4874 const struct flow *flow, uint32_t hash)
4876 struct facet *facet;
4878 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4879 if (flow_equal(flow, &facet->flow)) {
4887 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4888 * Returns it if found, otherwise a null pointer.
4890 * 'hash' must be the return value of flow_hash(flow, 0).
4892 * The returned facet is guaranteed to be valid. */
4893 static struct facet *
4894 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4897 struct facet *facet;
4899 facet = facet_find(ofproto, flow, hash);
4901 && (ofproto->backer->need_revalidate
4902 || tag_set_intersects(&ofproto->backer->revalidate_set,
4904 facet_revalidate(facet);
4906 /* facet_revalidate() may have destroyed 'facet'. */
4907 facet = facet_find(ofproto, flow, hash);
4913 /* Return a subfacet from 'facet'. A facet consists of one or more
4914 * subfacets, and this function returns one of them. */
4915 static struct subfacet *facet_get_subfacet(struct facet *facet)
4917 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4922 subfacet_path_to_string(enum subfacet_path path)
4925 case SF_NOT_INSTALLED:
4926 return "not installed";
4928 return "in fast path";
4930 return "in slow path";
4936 /* Returns the path in which a subfacet should be installed if its 'slow'
4937 * member has the specified value. */
4938 static enum subfacet_path
4939 subfacet_want_path(enum slow_path_reason slow)
4941 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4944 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4945 * supposing that its actions have been recalculated as 'want_actions' and that
4946 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4948 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4949 const struct ofpbuf *want_actions)
4951 enum subfacet_path want_path = subfacet_want_path(slow);
4952 return (want_path != subfacet->path
4953 || (want_path == SF_FAST_PATH
4954 && (subfacet->actions_len != want_actions->size
4955 || memcmp(subfacet->actions, want_actions->data,
4956 subfacet->actions_len))));
4960 facet_check_consistency(struct facet *facet)
4962 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4966 uint64_t odp_actions_stub[1024 / 8];
4967 struct ofpbuf odp_actions;
4969 struct rule_dpif *rule;
4970 struct subfacet *subfacet;
4971 bool may_log = false;
4974 /* Check the rule for consistency. */
4975 rule = rule_dpif_lookup(ofproto, &facet->flow);
4976 ok = rule == facet->rule;
4978 may_log = !VLOG_DROP_WARN(&rl);
4983 flow_format(&s, &facet->flow);
4984 ds_put_format(&s, ": facet associated with wrong rule (was "
4985 "table=%"PRIu8",", facet->rule->up.table_id);
4986 cls_rule_format(&facet->rule->up.cr, &s);
4987 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4989 cls_rule_format(&rule->up.cr, &s);
4990 ds_put_char(&s, ')');
4992 VLOG_WARN("%s", ds_cstr(&s));
4997 /* Check the datapath actions for consistency. */
4998 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4999 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5000 enum subfacet_path want_path;
5001 struct action_xlate_ctx ctx;
5004 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5005 &subfacet->initial_vals, rule, 0, NULL);
5006 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
5009 if (subfacet->path == SF_NOT_INSTALLED) {
5010 /* This only happens if the datapath reported an error when we
5011 * tried to install the flow. Don't flag another error here. */
5015 want_path = subfacet_want_path(subfacet->slow);
5016 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
5017 /* The actions for slow-path flows may legitimately vary from one
5018 * packet to the next. We're done. */
5022 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
5026 /* Inconsistency! */
5028 may_log = !VLOG_DROP_WARN(&rl);
5032 /* Rate-limited, skip reporting. */
5037 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5039 ds_put_cstr(&s, ": inconsistency in subfacet");
5040 if (want_path != subfacet->path) {
5041 enum odp_key_fitness fitness = subfacet->key_fitness;
5043 ds_put_format(&s, " (%s, fitness=%s)",
5044 subfacet_path_to_string(subfacet->path),
5045 odp_key_fitness_to_string(fitness));
5046 ds_put_format(&s, " (should have been %s)",
5047 subfacet_path_to_string(want_path));
5048 } else if (want_path == SF_FAST_PATH) {
5049 ds_put_cstr(&s, " (actions were: ");
5050 format_odp_actions(&s, subfacet->actions,
5051 subfacet->actions_len);
5052 ds_put_cstr(&s, ") (correct actions: ");
5053 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5054 ds_put_char(&s, ')');
5056 ds_put_cstr(&s, " (actions: ");
5057 format_odp_actions(&s, subfacet->actions,
5058 subfacet->actions_len);
5059 ds_put_char(&s, ')');
5061 VLOG_WARN("%s", ds_cstr(&s));
5064 ofpbuf_uninit(&odp_actions);
5069 /* Re-searches the classifier for 'facet':
5071 * - If the rule found is different from 'facet''s current rule, moves
5072 * 'facet' to the new rule and recompiles its actions.
5074 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5075 * where it is and recompiles its actions anyway.
5077 * - If any of 'facet''s subfacets correspond to a new flow according to
5078 * ofproto_receive(), 'facet' is removed. */
5080 facet_revalidate(struct facet *facet)
5082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5084 struct nlattr *odp_actions;
5087 struct actions *new_actions;
5089 struct action_xlate_ctx ctx;
5090 uint64_t odp_actions_stub[1024 / 8];
5091 struct ofpbuf odp_actions;
5093 struct rule_dpif *new_rule;
5094 struct subfacet *subfacet;
5097 COVERAGE_INC(facet_revalidate);
5099 /* Check that child subfacets still correspond to this facet. Tunnel
5100 * configuration changes could cause a subfacet's OpenFlow in_port to
5102 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5103 struct ofproto_dpif *recv_ofproto;
5104 struct flow recv_flow;
5107 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5108 subfacet->key_len, &recv_flow, NULL,
5109 &recv_ofproto, NULL, NULL);
5111 || recv_ofproto != ofproto
5112 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5113 facet_remove(facet);
5118 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5120 /* Calculate new datapath actions.
5122 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5123 * emit a NetFlow expiration and, if so, we need to have the old state
5124 * around to properly compose it. */
5126 /* If the datapath actions changed or the installability changed,
5127 * then we need to talk to the datapath. */
5130 memset(&ctx, 0, sizeof ctx);
5131 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5132 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5133 enum slow_path_reason slow;
5135 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5136 &subfacet->initial_vals, new_rule, 0, NULL);
5137 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5140 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5141 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5142 struct dpif_flow_stats stats;
5144 subfacet_install(subfacet,
5145 odp_actions.data, odp_actions.size, &stats, slow);
5146 subfacet_update_stats(subfacet, &stats);
5149 new_actions = xcalloc(list_size(&facet->subfacets),
5150 sizeof *new_actions);
5152 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5154 new_actions[i].actions_len = odp_actions.size;
5159 ofpbuf_uninit(&odp_actions);
5162 facet_flush_stats(facet);
5165 /* Update 'facet' now that we've taken care of all the old state. */
5166 facet->tags = ctx.tags;
5167 facet->nf_flow.output_iface = ctx.nf_output_iface;
5168 facet->has_learn = ctx.has_learn;
5169 facet->has_normal = ctx.has_normal;
5170 facet->has_fin_timeout = ctx.has_fin_timeout;
5171 facet->mirrors = ctx.mirrors;
5174 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5175 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5177 if (new_actions && new_actions[i].odp_actions) {
5178 free(subfacet->actions);
5179 subfacet->actions = new_actions[i].odp_actions;
5180 subfacet->actions_len = new_actions[i].actions_len;
5186 if (facet->rule != new_rule) {
5187 COVERAGE_INC(facet_changed_rule);
5188 list_remove(&facet->list_node);
5189 list_push_back(&new_rule->facets, &facet->list_node);
5190 facet->rule = new_rule;
5191 facet->used = new_rule->up.created;
5192 facet->prev_used = facet->used;
5196 /* Updates 'facet''s used time. Caller is responsible for calling
5197 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5199 facet_update_time(struct facet *facet, long long int used)
5201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5202 if (used > facet->used) {
5204 ofproto_rule_update_used(&facet->rule->up, used);
5205 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5210 facet_reset_counters(struct facet *facet)
5212 facet->packet_count = 0;
5213 facet->byte_count = 0;
5214 facet->prev_packet_count = 0;
5215 facet->prev_byte_count = 0;
5216 facet->accounted_bytes = 0;
5220 facet_push_stats(struct facet *facet)
5222 struct dpif_flow_stats stats;
5224 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5225 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5226 ovs_assert(facet->used >= facet->prev_used);
5228 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5229 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5230 stats.used = facet->used;
5231 stats.tcp_flags = 0;
5233 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5234 facet->prev_packet_count = facet->packet_count;
5235 facet->prev_byte_count = facet->byte_count;
5236 facet->prev_used = facet->used;
5238 flow_push_stats(facet, &stats);
5240 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5241 facet->mirrors, stats.n_packets, stats.n_bytes);
5246 push_all_stats__(bool run_fast)
5248 static long long int rl = LLONG_MIN;
5249 struct ofproto_dpif *ofproto;
5251 if (time_msec() < rl) {
5255 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5256 struct facet *facet;
5258 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5259 facet_push_stats(facet);
5266 rl = time_msec() + 100;
5270 push_all_stats(void)
5272 push_all_stats__(true);
5276 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5278 rule->packet_count += stats->n_packets;
5279 rule->byte_count += stats->n_bytes;
5280 ofproto_rule_update_used(&rule->up, stats->used);
5283 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5284 * into given 'facet->rule''s actions and mirrors. */
5286 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5288 struct rule_dpif *rule = facet->rule;
5289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5290 struct subfacet *subfacet = facet_get_subfacet(facet);
5291 struct action_xlate_ctx ctx;
5293 ofproto_rule_update_used(&rule->up, stats->used);
5295 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5296 &subfacet->initial_vals, rule, 0, NULL);
5297 ctx.resubmit_stats = stats;
5298 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5299 rule->up.ofpacts_len);
5304 static struct subfacet *
5305 subfacet_find(struct ofproto_dpif *ofproto,
5306 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5308 struct subfacet *subfacet;
5310 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5311 &ofproto->subfacets) {
5312 if (subfacet->key_len == key_len
5313 && !memcmp(key, subfacet->key, key_len)) {
5321 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5322 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5323 * existing subfacet if there is one, otherwise creates and returns a
5326 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5327 * which case the caller must populate the actions with
5328 * subfacet_make_actions(). */
5329 static struct subfacet *
5330 subfacet_create(struct facet *facet, struct flow_miss *miss,
5333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5334 enum odp_key_fitness key_fitness = miss->key_fitness;
5335 const struct nlattr *key = miss->key;
5336 size_t key_len = miss->key_len;
5338 struct subfacet *subfacet;
5340 key_hash = odp_flow_key_hash(key, key_len);
5342 if (list_is_empty(&facet->subfacets)) {
5343 subfacet = &facet->one_subfacet;
5345 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5347 if (subfacet->facet == facet) {
5351 /* This shouldn't happen. */
5352 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5353 subfacet_destroy(subfacet);
5356 subfacet = xmalloc(sizeof *subfacet);
5359 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5360 list_push_back(&facet->subfacets, &subfacet->list_node);
5361 subfacet->facet = facet;
5362 subfacet->key_fitness = key_fitness;
5363 subfacet->key = xmemdup(key, key_len);
5364 subfacet->key_len = key_len;
5365 subfacet->used = now;
5366 subfacet->created = now;
5367 subfacet->dp_packet_count = 0;
5368 subfacet->dp_byte_count = 0;
5369 subfacet->actions_len = 0;
5370 subfacet->actions = NULL;
5371 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5374 subfacet->path = SF_NOT_INSTALLED;
5375 subfacet->initial_vals = miss->initial_vals;
5376 subfacet->odp_in_port = miss->odp_in_port;
5378 ofproto->subfacet_add_count++;
5382 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5383 * its facet within 'ofproto', and frees it. */
5385 subfacet_destroy__(struct subfacet *subfacet)
5387 struct facet *facet = subfacet->facet;
5388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5390 /* Update ofproto stats before uninstall the subfacet. */
5391 ofproto->subfacet_del_count++;
5392 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5394 subfacet_uninstall(subfacet);
5395 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5396 list_remove(&subfacet->list_node);
5397 free(subfacet->key);
5398 free(subfacet->actions);
5399 if (subfacet != &facet->one_subfacet) {
5404 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5405 * last remaining subfacet in its facet destroys the facet too. */
5407 subfacet_destroy(struct subfacet *subfacet)
5409 struct facet *facet = subfacet->facet;
5411 if (list_is_singleton(&facet->subfacets)) {
5412 /* facet_remove() needs at least one subfacet (it will remove it). */
5413 facet_remove(facet);
5415 subfacet_destroy__(subfacet);
5420 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5421 struct subfacet **subfacets, int n)
5423 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5424 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5425 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5428 for (i = 0; i < n; i++) {
5429 ops[i].type = DPIF_OP_FLOW_DEL;
5430 ops[i].u.flow_del.key = subfacets[i]->key;
5431 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5432 ops[i].u.flow_del.stats = &stats[i];
5436 dpif_operate(ofproto->backer->dpif, opsp, n);
5437 for (i = 0; i < n; i++) {
5438 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5439 subfacets[i]->path = SF_NOT_INSTALLED;
5440 subfacet_destroy(subfacets[i]);
5445 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5446 * Translates the actions into 'odp_actions', which the caller must have
5447 * initialized and is responsible for uninitializing. */
5449 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5450 struct ofpbuf *odp_actions)
5452 struct facet *facet = subfacet->facet;
5453 struct rule_dpif *rule = facet->rule;
5454 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5456 struct action_xlate_ctx ctx;
5458 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5459 &subfacet->initial_vals, rule, 0, packet);
5460 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5461 facet->tags = ctx.tags;
5462 facet->has_learn = ctx.has_learn;
5463 facet->has_normal = ctx.has_normal;
5464 facet->has_fin_timeout = ctx.has_fin_timeout;
5465 facet->nf_flow.output_iface = ctx.nf_output_iface;
5466 facet->mirrors = ctx.mirrors;
5468 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5469 if (subfacet->actions_len != odp_actions->size
5470 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5471 free(subfacet->actions);
5472 subfacet->actions_len = odp_actions->size;
5473 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5477 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5478 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5479 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5480 * since 'subfacet' was last updated.
5482 * Returns 0 if successful, otherwise a positive errno value. */
5484 subfacet_install(struct subfacet *subfacet,
5485 const struct nlattr *actions, size_t actions_len,
5486 struct dpif_flow_stats *stats,
5487 enum slow_path_reason slow)
5489 struct facet *facet = subfacet->facet;
5490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5491 enum subfacet_path path = subfacet_want_path(slow);
5492 uint64_t slow_path_stub[128 / 8];
5493 enum dpif_flow_put_flags flags;
5496 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5498 flags |= DPIF_FP_ZERO_STATS;
5501 if (path == SF_SLOW_PATH) {
5502 compose_slow_path(ofproto, &facet->flow, slow,
5503 slow_path_stub, sizeof slow_path_stub,
5504 &actions, &actions_len);
5507 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5508 subfacet->key_len, actions, actions_len, stats);
5511 subfacet_reset_dp_stats(subfacet, stats);
5515 subfacet->path = path;
5521 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5523 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5524 stats, subfacet->slow);
5527 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5529 subfacet_uninstall(struct subfacet *subfacet)
5531 if (subfacet->path != SF_NOT_INSTALLED) {
5532 struct rule_dpif *rule = subfacet->facet->rule;
5533 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5534 struct dpif_flow_stats stats;
5537 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5538 subfacet->key_len, &stats);
5539 subfacet_reset_dp_stats(subfacet, &stats);
5541 subfacet_update_stats(subfacet, &stats);
5543 subfacet->path = SF_NOT_INSTALLED;
5545 ovs_assert(subfacet->dp_packet_count == 0);
5546 ovs_assert(subfacet->dp_byte_count == 0);
5550 /* Resets 'subfacet''s datapath statistics counters. This should be called
5551 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5552 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5553 * was reset in the datapath. 'stats' will be modified to include only
5554 * statistics new since 'subfacet' was last updated. */
5556 subfacet_reset_dp_stats(struct subfacet *subfacet,
5557 struct dpif_flow_stats *stats)
5560 && subfacet->dp_packet_count <= stats->n_packets
5561 && subfacet->dp_byte_count <= stats->n_bytes) {
5562 stats->n_packets -= subfacet->dp_packet_count;
5563 stats->n_bytes -= subfacet->dp_byte_count;
5566 subfacet->dp_packet_count = 0;
5567 subfacet->dp_byte_count = 0;
5570 /* Updates 'subfacet''s used time. The caller is responsible for calling
5571 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5573 subfacet_update_time(struct subfacet *subfacet, long long int used)
5575 if (used > subfacet->used) {
5576 subfacet->used = used;
5577 facet_update_time(subfacet->facet, used);
5581 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5583 * Because of the meaning of a subfacet's counters, it only makes sense to do
5584 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5585 * represents a packet that was sent by hand or if it represents statistics
5586 * that have been cleared out of the datapath. */
5588 subfacet_update_stats(struct subfacet *subfacet,
5589 const struct dpif_flow_stats *stats)
5591 if (stats->n_packets || stats->used > subfacet->used) {
5592 struct facet *facet = subfacet->facet;
5594 subfacet_update_time(subfacet, stats->used);
5595 facet->packet_count += stats->n_packets;
5596 facet->byte_count += stats->n_bytes;
5597 facet->tcp_flags |= stats->tcp_flags;
5598 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5604 static struct rule_dpif *
5605 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5607 struct rule_dpif *rule;
5609 rule = rule_dpif_lookup__(ofproto, flow, 0);
5614 return rule_dpif_miss_rule(ofproto, flow);
5617 static struct rule_dpif *
5618 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5621 struct cls_rule *cls_rule;
5622 struct classifier *cls;
5624 if (table_id >= N_TABLES) {
5628 cls = &ofproto->up.tables[table_id].cls;
5629 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5630 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5631 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5632 * are unavailable. */
5633 struct flow ofpc_normal_flow = *flow;
5634 ofpc_normal_flow.tp_src = htons(0);
5635 ofpc_normal_flow.tp_dst = htons(0);
5636 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5638 cls_rule = classifier_lookup(cls, flow);
5640 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5643 static struct rule_dpif *
5644 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5646 struct ofport_dpif *port;
5648 port = get_ofp_port(ofproto, flow->in_port);
5650 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5651 return ofproto->miss_rule;
5654 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5655 return ofproto->no_packet_in_rule;
5657 return ofproto->miss_rule;
5661 complete_operation(struct rule_dpif *rule)
5663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5665 rule_invalidate(rule);
5667 struct dpif_completion *c = xmalloc(sizeof *c);
5668 c->op = rule->up.pending;
5669 list_push_back(&ofproto->completions, &c->list_node);
5671 ofoperation_complete(rule->up.pending, 0);
5675 static struct rule *
5678 struct rule_dpif *rule = xmalloc(sizeof *rule);
5683 rule_dealloc(struct rule *rule_)
5685 struct rule_dpif *rule = rule_dpif_cast(rule_);
5690 rule_construct(struct rule *rule_)
5692 struct rule_dpif *rule = rule_dpif_cast(rule_);
5693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5694 struct rule_dpif *victim;
5697 rule->packet_count = 0;
5698 rule->byte_count = 0;
5700 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5701 if (victim && !list_is_empty(&victim->facets)) {
5702 struct facet *facet;
5704 rule->facets = victim->facets;
5705 list_moved(&rule->facets);
5706 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5707 /* XXX: We're only clearing our local counters here. It's possible
5708 * that quite a few packets are unaccounted for in the datapath
5709 * statistics. These will be accounted to the new rule instead of
5710 * cleared as required. This could be fixed by clearing out the
5711 * datapath statistics for this facet, but currently it doesn't
5713 facet_reset_counters(facet);
5717 /* Must avoid list_moved() in this case. */
5718 list_init(&rule->facets);
5721 table_id = rule->up.table_id;
5723 rule->tag = victim->tag;
5724 } else if (table_id == 0) {
5729 miniflow_expand(&rule->up.cr.match.flow, &flow);
5730 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5731 ofproto->tables[table_id].basis);
5734 complete_operation(rule);
5739 rule_destruct(struct rule *rule_)
5741 struct rule_dpif *rule = rule_dpif_cast(rule_);
5742 struct facet *facet, *next_facet;
5744 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5745 facet_revalidate(facet);
5748 complete_operation(rule);
5752 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5754 struct rule_dpif *rule = rule_dpif_cast(rule_);
5755 struct facet *facet;
5757 /* push_all_stats() can handle flow misses which, when using the learn
5758 * action, can cause rules to be added and deleted. This can corrupt our
5759 * caller's datastructures which assume that rule_get_stats() doesn't have
5760 * an impact on the flow table. To be safe, we disable miss handling. */
5761 push_all_stats__(false);
5763 /* Start from historical data for 'rule' itself that are no longer tracked
5764 * in facets. This counts, for example, facets that have expired. */
5765 *packets = rule->packet_count;
5766 *bytes = rule->byte_count;
5768 /* Add any statistics that are tracked by facets. This includes
5769 * statistical data recently updated by ofproto_update_stats() as well as
5770 * stats for packets that were executed "by hand" via dpif_execute(). */
5771 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5772 *packets += facet->packet_count;
5773 *bytes += facet->byte_count;
5778 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5779 struct ofpbuf *packet)
5781 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5782 struct initial_vals initial_vals;
5783 struct dpif_flow_stats stats;
5784 struct action_xlate_ctx ctx;
5785 uint64_t odp_actions_stub[1024 / 8];
5786 struct ofpbuf odp_actions;
5788 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5789 rule_credit_stats(rule, &stats);
5791 initial_vals.vlan_tci = flow->vlan_tci;
5792 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5793 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5794 rule, stats.tcp_flags, packet);
5795 ctx.resubmit_stats = &stats;
5796 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5798 execute_odp_actions(ofproto, flow, odp_actions.data,
5799 odp_actions.size, packet);
5801 ofpbuf_uninit(&odp_actions);
5805 rule_execute(struct rule *rule, const struct flow *flow,
5806 struct ofpbuf *packet)
5808 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5809 ofpbuf_delete(packet);
5814 rule_modify_actions(struct rule *rule_)
5816 struct rule_dpif *rule = rule_dpif_cast(rule_);
5818 complete_operation(rule);
5821 /* Sends 'packet' out 'ofport'.
5822 * May modify 'packet'.
5823 * Returns 0 if successful, otherwise a positive errno value. */
5825 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5827 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5828 uint64_t odp_actions_stub[1024 / 8];
5829 struct ofpbuf key, odp_actions;
5830 struct odputil_keybuf keybuf;
5835 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5836 if (netdev_vport_is_patch(ofport->up.netdev)) {
5837 struct ofproto_dpif *peer_ofproto;
5838 struct dpif_flow_stats stats;
5839 struct ofport_dpif *peer;
5840 struct rule_dpif *rule;
5842 peer = ofport_get_peer(ofport);
5847 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5848 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5849 netdev_vport_inc_rx(peer->up.netdev, &stats);
5851 flow.in_port = peer->up.ofp_port;
5852 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5853 rule = rule_dpif_lookup(peer_ofproto, &flow);
5854 rule_dpif_execute(rule, &flow, packet);
5859 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5861 if (ofport->tnl_port) {
5862 struct dpif_flow_stats stats;
5864 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5865 if (odp_port == OVSP_NONE) {
5869 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5870 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5871 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5872 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5874 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5876 if (odp_port != ofport->odp_port) {
5877 eth_pop_vlan(packet);
5878 flow.vlan_tci = htons(0);
5882 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5883 odp_flow_key_from_flow(&key, &flow,
5884 ofp_port_to_odp_port(ofproto, flow.in_port));
5886 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5887 compose_ipfix_action(ofproto, &odp_actions, &flow);
5889 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5890 error = dpif_execute(ofproto->backer->dpif,
5892 odp_actions.data, odp_actions.size,
5894 ofpbuf_uninit(&odp_actions);
5897 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5898 ofproto->up.name, odp_port, strerror(error));
5900 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5904 /* OpenFlow to datapath action translation. */
5906 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5907 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5908 struct action_xlate_ctx *);
5909 static void xlate_normal(struct action_xlate_ctx *);
5911 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5912 * The action will state 'slow' as the reason that the action is in the slow
5913 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5914 * dump-flows" output to see why a flow is in the slow path.)
5916 * The 'stub_size' bytes in 'stub' will be used to store the action.
5917 * 'stub_size' must be large enough for the action.
5919 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5922 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5923 enum slow_path_reason slow,
5924 uint64_t *stub, size_t stub_size,
5925 const struct nlattr **actionsp, size_t *actions_lenp)
5927 union user_action_cookie cookie;
5930 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5931 cookie.slow_path.unused = 0;
5932 cookie.slow_path.reason = slow;
5934 ofpbuf_use_stack(&buf, stub, stub_size);
5935 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5936 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5937 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5939 put_userspace_action(ofproto, &buf, flow, &cookie,
5940 sizeof cookie.slow_path);
5942 *actionsp = buf.data;
5943 *actions_lenp = buf.size;
5947 put_userspace_action(const struct ofproto_dpif *ofproto,
5948 struct ofpbuf *odp_actions,
5949 const struct flow *flow,
5950 const union user_action_cookie *cookie,
5951 const size_t cookie_size)
5955 pid = dpif_port_get_pid(ofproto->backer->dpif,
5956 ofp_port_to_odp_port(ofproto, flow->in_port));
5958 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5961 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5962 * the number of packets out of UINT32_MAX to sample. The given
5963 * cookie is passed back in the callback for each sampled packet.
5966 compose_sample_action(const struct ofproto_dpif *ofproto,
5967 struct ofpbuf *odp_actions,
5968 const struct flow *flow,
5969 const uint32_t probability,
5970 const union user_action_cookie *cookie,
5971 const size_t cookie_size)
5973 size_t sample_offset, actions_offset;
5976 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5978 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5980 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5981 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5984 nl_msg_end_nested(odp_actions, actions_offset);
5985 nl_msg_end_nested(odp_actions, sample_offset);
5986 return cookie_offset;
5990 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5991 ovs_be16 vlan_tci, uint32_t odp_port,
5992 unsigned int n_outputs, union user_action_cookie *cookie)
5996 cookie->type = USER_ACTION_COOKIE_SFLOW;
5997 cookie->sflow.vlan_tci = vlan_tci;
5999 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
6000 * port information") for the interpretation of cookie->output. */
6001 switch (n_outputs) {
6003 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
6004 cookie->sflow.output = 0x40000000 | 256;
6008 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
6010 cookie->sflow.output = ifindex;
6015 /* 0x80000000 means "multiple output ports. */
6016 cookie->sflow.output = 0x80000000 | n_outputs;
6021 /* Compose SAMPLE action for sFlow bridge sampling. */
6023 compose_sflow_action(const struct ofproto_dpif *ofproto,
6024 struct ofpbuf *odp_actions,
6025 const struct flow *flow,
6028 uint32_t probability;
6029 union user_action_cookie cookie;
6031 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
6035 probability = dpif_sflow_get_probability(ofproto->sflow);
6036 compose_sflow_cookie(ofproto, htons(0), odp_port,
6037 odp_port == OVSP_NONE ? 0 : 1, &cookie);
6039 return compose_sample_action(ofproto, odp_actions, flow, probability,
6040 &cookie, sizeof cookie.sflow);
6044 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
6045 uint32_t obs_domain_id, uint32_t obs_point_id,
6046 union user_action_cookie *cookie)
6048 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6049 cookie->flow_sample.probability = probability;
6050 cookie->flow_sample.collector_set_id = collector_set_id;
6051 cookie->flow_sample.obs_domain_id = obs_domain_id;
6052 cookie->flow_sample.obs_point_id = obs_point_id;
6056 compose_ipfix_cookie(union user_action_cookie *cookie)
6058 cookie->type = USER_ACTION_COOKIE_IPFIX;
6061 /* Compose SAMPLE action for IPFIX bridge sampling. */
6063 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6064 struct ofpbuf *odp_actions,
6065 const struct flow *flow)
6067 uint32_t probability;
6068 union user_action_cookie cookie;
6070 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6074 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6075 compose_ipfix_cookie(&cookie);
6077 compose_sample_action(ofproto, odp_actions, flow, probability,
6078 &cookie, sizeof cookie.ipfix);
6081 /* SAMPLE action for sFlow must be first action in any given list of
6082 * actions. At this point we do not have all information required to
6083 * build it. So try to build sample action as complete as possible. */
6085 add_sflow_action(struct action_xlate_ctx *ctx)
6087 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6089 &ctx->flow, OVSP_NONE);
6090 ctx->sflow_odp_port = 0;
6091 ctx->sflow_n_outputs = 0;
6094 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6095 * of actions, eventually after the SAMPLE action for sFlow. */
6097 add_ipfix_action(struct action_xlate_ctx *ctx)
6099 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6102 /* Fix SAMPLE action according to data collected while composing ODP actions.
6103 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6104 * USERSPACE action's user-cookie which is required for sflow. */
6106 fix_sflow_action(struct action_xlate_ctx *ctx)
6108 const struct flow *base = &ctx->base_flow;
6109 union user_action_cookie *cookie;
6111 if (!ctx->user_cookie_offset) {
6115 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6116 sizeof cookie->sflow);
6117 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6119 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6120 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6124 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6127 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6128 ovs_be16 flow_vlan_tci;
6129 uint32_t flow_skb_mark;
6130 uint8_t flow_nw_tos;
6131 struct priority_to_dscp *pdscp;
6132 uint32_t out_port, odp_port;
6134 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6135 * before traversing a patch port. */
6136 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6139 xlate_report(ctx, "Nonexistent output port");
6141 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6142 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6144 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6145 xlate_report(ctx, "STP not in forwarding state, skipping output");
6149 if (netdev_vport_is_patch(ofport->up.netdev)) {
6150 struct ofport_dpif *peer = ofport_get_peer(ofport);
6151 struct flow old_flow = ctx->flow;
6152 const struct ofproto_dpif *peer_ofproto;
6153 enum slow_path_reason special;
6154 struct ofport_dpif *in_port;
6157 xlate_report(ctx, "Nonexistent patch port peer");
6161 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6162 if (peer_ofproto->backer != ctx->ofproto->backer) {
6163 xlate_report(ctx, "Patch port peer on a different datapath");
6167 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6168 ctx->flow.in_port = peer->up.ofp_port;
6169 ctx->flow.metadata = htonll(0);
6170 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6171 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6173 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6174 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6177 ctx->slow |= special;
6178 } else if (!in_port || may_receive(in_port, ctx)) {
6179 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6180 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6182 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6183 * learning action look at the packet, then drop it. */
6184 struct flow old_base_flow = ctx->base_flow;
6185 size_t old_size = ctx->odp_actions->size;
6186 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6187 ctx->base_flow = old_base_flow;
6188 ctx->odp_actions->size = old_size;
6192 ctx->flow = old_flow;
6193 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6195 if (ctx->resubmit_stats) {
6196 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6197 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6203 flow_vlan_tci = ctx->flow.vlan_tci;
6204 flow_skb_mark = ctx->flow.skb_mark;
6205 flow_nw_tos = ctx->flow.nw_tos;
6207 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6209 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6210 ctx->flow.nw_tos |= pdscp->dscp;
6213 if (ofport->tnl_port) {
6214 /* Save tunnel metadata so that changes made due to
6215 * the Logical (tunnel) Port are not visible for any further
6216 * matches, while explicit set actions on tunnel metadata are.
6218 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6219 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6220 if (odp_port == OVSP_NONE) {
6221 xlate_report(ctx, "Tunneling decided against output");
6222 goto out; /* restore flow_nw_tos */
6225 if (ctx->resubmit_stats) {
6226 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6228 out_port = odp_port;
6229 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6231 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6233 odp_port = ofport->odp_port;
6234 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6235 ctx->flow.vlan_tci);
6236 if (out_port != odp_port) {
6237 ctx->flow.vlan_tci = htons(0);
6239 ctx->flow.skb_mark &= ~IPSEC_MARK;
6241 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6242 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6244 ctx->sflow_odp_port = odp_port;
6245 ctx->sflow_n_outputs++;
6246 ctx->nf_output_iface = ofp_port;
6249 ctx->flow.vlan_tci = flow_vlan_tci;
6250 ctx->flow.skb_mark = flow_skb_mark;
6252 ctx->flow.nw_tos = flow_nw_tos;
6256 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6258 compose_output_action__(ctx, ofp_port, true);
6262 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6264 struct ofproto_dpif *ofproto = ctx->ofproto;
6265 uint8_t table_id = ctx->table_id;
6267 if (table_id > 0 && table_id < N_TABLES) {
6268 struct table_dpif *table = &ofproto->tables[table_id];
6269 if (table->other_table) {
6270 ctx->tags |= (rule && rule->tag
6272 : rule_calculate_tag(&ctx->flow,
6273 &table->other_table->mask,
6279 /* Common rule processing in one place to avoid duplicating code. */
6280 static struct rule_dpif *
6281 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6284 if (ctx->resubmit_hook) {
6285 ctx->resubmit_hook(ctx, rule);
6287 if (rule == NULL && may_packet_in) {
6289 * check if table configuration flags
6290 * OFPTC_TABLE_MISS_CONTROLLER, default.
6291 * OFPTC_TABLE_MISS_CONTINUE,
6292 * OFPTC_TABLE_MISS_DROP
6293 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6295 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6297 if (rule && ctx->resubmit_stats) {
6298 rule_credit_stats(rule, ctx->resubmit_stats);
6304 xlate_table_action(struct action_xlate_ctx *ctx,
6305 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6307 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6308 struct rule_dpif *rule;
6309 uint16_t old_in_port = ctx->flow.in_port;
6310 uint8_t old_table_id = ctx->table_id;
6312 ctx->table_id = table_id;
6314 /* Look up a flow with 'in_port' as the input port. */
6315 ctx->flow.in_port = in_port;
6316 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6318 tag_the_flow(ctx, rule);
6320 /* Restore the original input port. Otherwise OFPP_NORMAL and
6321 * OFPP_IN_PORT will have surprising behavior. */
6322 ctx->flow.in_port = old_in_port;
6324 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6327 struct rule_dpif *old_rule = ctx->rule;
6331 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6332 ctx->rule = old_rule;
6336 ctx->table_id = old_table_id;
6338 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6340 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6341 MAX_RESUBMIT_RECURSION);
6342 ctx->max_resubmit_trigger = true;
6347 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6348 const struct ofpact_resubmit *resubmit)
6353 in_port = resubmit->in_port;
6354 if (in_port == OFPP_IN_PORT) {
6355 in_port = ctx->flow.in_port;
6358 table_id = resubmit->table_id;
6359 if (table_id == 255) {
6360 table_id = ctx->table_id;
6363 xlate_table_action(ctx, in_port, table_id, false);
6367 flood_packets(struct action_xlate_ctx *ctx, bool all)
6369 struct ofport_dpif *ofport;
6371 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6372 uint16_t ofp_port = ofport->up.ofp_port;
6374 if (ofp_port == ctx->flow.in_port) {
6379 compose_output_action__(ctx, ofp_port, false);
6380 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6381 compose_output_action(ctx, ofp_port);
6385 ctx->nf_output_iface = NF_OUT_FLOOD;
6389 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6390 enum ofp_packet_in_reason reason,
6391 uint16_t controller_id)
6393 struct ofputil_packet_in pin;
6394 struct ofpbuf *packet;
6396 ctx->slow |= SLOW_CONTROLLER;
6401 packet = ofpbuf_clone(ctx->packet);
6403 if (packet->l2 && packet->l3) {
6404 struct eth_header *eh;
6405 uint16_t mpls_depth;
6407 eth_pop_vlan(packet);
6410 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6411 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6413 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6414 eth_push_vlan(packet, ctx->flow.vlan_tci);
6417 mpls_depth = eth_mpls_depth(packet);
6419 if (mpls_depth < ctx->flow.mpls_depth) {
6420 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6421 } else if (mpls_depth > ctx->flow.mpls_depth) {
6422 pop_mpls(packet, ctx->flow.dl_type);
6423 } else if (mpls_depth) {
6424 set_mpls_lse(packet, ctx->flow.mpls_lse);
6428 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6429 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6430 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6434 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6435 packet_set_tcp_port(packet, ctx->flow.tp_src,
6437 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6438 packet_set_udp_port(packet, ctx->flow.tp_src,
6445 pin.packet = packet->data;
6446 pin.packet_len = packet->size;
6447 pin.reason = reason;
6448 pin.controller_id = controller_id;
6449 pin.table_id = ctx->table_id;
6450 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6453 flow_get_metadata(&ctx->flow, &pin.fmd);
6455 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6456 ofpbuf_delete(packet);
6460 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6462 ovs_assert(eth_type_mpls(eth_type));
6464 if (ctx->base_flow.mpls_depth) {
6465 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6466 ctx->flow.mpls_depth++;
6471 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6472 label = htonl(0x2); /* IPV6 Explicit Null. */
6474 label = htonl(0x0); /* IPV4 Explicit Null. */
6476 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6477 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6478 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6479 ctx->flow.mpls_depth = 1;
6481 ctx->flow.dl_type = eth_type;
6485 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6487 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6488 ovs_assert(!eth_type_mpls(eth_type));
6490 if (ctx->flow.mpls_depth) {
6491 ctx->flow.mpls_depth--;
6492 ctx->flow.mpls_lse = htonl(0);
6493 if (!ctx->flow.mpls_depth) {
6494 ctx->flow.dl_type = eth_type;
6500 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6502 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6503 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6507 if (ctx->flow.nw_ttl > 1) {
6513 for (i = 0; i < ids->n_controllers; i++) {
6514 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6518 /* Stop processing for current table. */
6524 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6526 if (!eth_type_mpls(ctx->flow.dl_type)) {
6530 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6535 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6537 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6539 if (!eth_type_mpls(ctx->flow.dl_type)) {
6545 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6548 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6550 /* Stop processing for current table. */
6556 xlate_output_action(struct action_xlate_ctx *ctx,
6557 uint16_t port, uint16_t max_len, bool may_packet_in)
6559 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6561 ctx->nf_output_iface = NF_OUT_DROP;
6565 compose_output_action(ctx, ctx->flow.in_port);
6568 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6574 flood_packets(ctx, false);
6577 flood_packets(ctx, true);
6579 case OFPP_CONTROLLER:
6580 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6586 if (port != ctx->flow.in_port) {
6587 compose_output_action(ctx, port);
6589 xlate_report(ctx, "skipping output to input port");
6594 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6595 ctx->nf_output_iface = NF_OUT_FLOOD;
6596 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6597 ctx->nf_output_iface = prev_nf_output_iface;
6598 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6599 ctx->nf_output_iface != NF_OUT_FLOOD) {
6600 ctx->nf_output_iface = NF_OUT_MULTI;
6605 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6606 const struct ofpact_output_reg *or)
6608 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6609 if (port <= UINT16_MAX) {
6610 xlate_output_action(ctx, port, or->max_len, false);
6615 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6616 const struct ofpact_enqueue *enqueue)
6618 uint16_t ofp_port = enqueue->port;
6619 uint32_t queue_id = enqueue->queue;
6620 uint32_t flow_priority, priority;
6623 /* Translate queue to priority. */
6624 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6625 queue_id, &priority);
6627 /* Fall back to ordinary output action. */
6628 xlate_output_action(ctx, enqueue->port, 0, false);
6632 /* Check output port. */
6633 if (ofp_port == OFPP_IN_PORT) {
6634 ofp_port = ctx->flow.in_port;
6635 } else if (ofp_port == ctx->flow.in_port) {
6639 /* Add datapath actions. */
6640 flow_priority = ctx->flow.skb_priority;
6641 ctx->flow.skb_priority = priority;
6642 compose_output_action(ctx, ofp_port);
6643 ctx->flow.skb_priority = flow_priority;
6645 /* Update NetFlow output port. */
6646 if (ctx->nf_output_iface == NF_OUT_DROP) {
6647 ctx->nf_output_iface = ofp_port;
6648 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6649 ctx->nf_output_iface = NF_OUT_MULTI;
6654 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6656 uint32_t skb_priority;
6658 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6659 queue_id, &skb_priority)) {
6660 ctx->flow.skb_priority = skb_priority;
6662 /* Couldn't translate queue to a priority. Nothing to do. A warning
6663 * has already been logged. */
6668 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6670 struct ofproto_dpif *ofproto = ofproto_;
6671 struct ofport_dpif *port;
6681 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6684 port = get_ofp_port(ofproto, ofp_port);
6685 return port ? port->may_enable : false;
6690 xlate_bundle_action(struct action_xlate_ctx *ctx,
6691 const struct ofpact_bundle *bundle)
6695 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6696 if (bundle->dst.field) {
6697 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6699 xlate_output_action(ctx, port, 0, false);
6704 xlate_learn_action(struct action_xlate_ctx *ctx,
6705 const struct ofpact_learn *learn)
6707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6708 struct ofputil_flow_mod fm;
6709 uint64_t ofpacts_stub[1024 / 8];
6710 struct ofpbuf ofpacts;
6713 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6714 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6716 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6717 if (error && !VLOG_DROP_WARN(&rl)) {
6718 VLOG_WARN("learning action failed to modify flow table (%s)",
6719 ofperr_get_name(error));
6722 ofpbuf_uninit(&ofpacts);
6725 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6726 * means "infinite". */
6728 reduce_timeout(uint16_t max, uint16_t *timeout)
6730 if (max && (!*timeout || *timeout > max)) {
6736 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6737 const struct ofpact_fin_timeout *oft)
6739 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6740 struct rule_dpif *rule = ctx->rule;
6742 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6743 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6748 xlate_sample_action(struct action_xlate_ctx *ctx,
6749 const struct ofpact_sample *os)
6751 union user_action_cookie cookie;
6752 /* Scale the probability from 16-bit to 32-bit while representing
6753 * the same percentage. */
6754 uint32_t probability = (os->probability << 16) | os->probability;
6756 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6758 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6759 os->obs_domain_id, os->obs_point_id, &cookie);
6760 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6761 probability, &cookie, sizeof cookie.flow_sample);
6765 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6767 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6768 ? OFPUTIL_PC_NO_RECV_STP
6769 : OFPUTIL_PC_NO_RECV)) {
6773 /* Only drop packets here if both forwarding and learning are
6774 * disabled. If just learning is enabled, we need to have
6775 * OFPP_NORMAL and the learning action have a look at the packet
6776 * before we can drop it. */
6777 if (!stp_forward_in_state(port->stp_state)
6778 && !stp_learn_in_state(port->stp_state)) {
6786 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6788 if (is_ip_any(&ctx->base_flow)
6789 && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6790 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6791 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6792 " but is not ECN capable");
6795 /* Set the ECN CE value in the tunneled packet. */
6796 ctx->flow.nw_tos |= IP_ECN_CE;
6804 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6805 struct action_xlate_ctx *ctx)
6807 bool was_evictable = true;
6808 const struct ofpact *a;
6811 /* Don't let the rule we're working on get evicted underneath us. */
6812 was_evictable = ctx->rule->up.evictable;
6813 ctx->rule->up.evictable = false;
6816 do_xlate_actions_again:
6817 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6818 struct ofpact_controller *controller;
6819 const struct ofpact_metadata *metadata;
6827 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6828 ofpact_get_OUTPUT(a)->max_len, true);
6831 case OFPACT_CONTROLLER:
6832 controller = ofpact_get_CONTROLLER(a);
6833 execute_controller_action(ctx, controller->max_len,
6835 controller->controller_id);
6838 case OFPACT_ENQUEUE:
6839 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6842 case OFPACT_SET_VLAN_VID:
6843 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6844 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6848 case OFPACT_SET_VLAN_PCP:
6849 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6850 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6855 case OFPACT_STRIP_VLAN:
6856 ctx->flow.vlan_tci = htons(0);
6859 case OFPACT_PUSH_VLAN:
6860 /* XXX 802.1AD(QinQ) */
6861 ctx->flow.vlan_tci = htons(VLAN_CFI);
6864 case OFPACT_SET_ETH_SRC:
6865 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6869 case OFPACT_SET_ETH_DST:
6870 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6874 case OFPACT_SET_IPV4_SRC:
6875 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6876 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6880 case OFPACT_SET_IPV4_DST:
6881 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6882 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6886 case OFPACT_SET_IPV4_DSCP:
6887 /* OpenFlow 1.0 only supports IPv4. */
6888 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6889 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6890 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6894 case OFPACT_SET_L4_SRC_PORT:
6895 if (is_ip_any(&ctx->flow)) {
6896 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6900 case OFPACT_SET_L4_DST_PORT:
6901 if (is_ip_any(&ctx->flow)) {
6902 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6906 case OFPACT_RESUBMIT:
6907 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6910 case OFPACT_SET_TUNNEL:
6911 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6914 case OFPACT_SET_QUEUE:
6915 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6918 case OFPACT_POP_QUEUE:
6919 ctx->flow.skb_priority = ctx->orig_skb_priority;
6922 case OFPACT_REG_MOVE:
6923 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6926 case OFPACT_REG_LOAD:
6927 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6930 case OFPACT_STACK_PUSH:
6931 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6935 case OFPACT_STACK_POP:
6936 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6940 case OFPACT_PUSH_MPLS:
6941 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6944 case OFPACT_POP_MPLS:
6945 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6948 case OFPACT_SET_MPLS_TTL:
6949 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6954 case OFPACT_DEC_MPLS_TTL:
6955 if (execute_dec_mpls_ttl_action(ctx)) {
6960 case OFPACT_DEC_TTL:
6961 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6967 /* Nothing to do. */
6970 case OFPACT_MULTIPATH:
6971 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6975 ctx->ofproto->has_bundle_action = true;
6976 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6979 case OFPACT_OUTPUT_REG:
6980 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6984 ctx->has_learn = true;
6985 if (ctx->may_learn) {
6986 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6994 case OFPACT_FIN_TIMEOUT:
6995 ctx->has_fin_timeout = true;
6996 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6999 case OFPACT_CLEAR_ACTIONS:
7001 * Nothing to do because writa-actions is not supported for now.
7002 * When writa-actions is supported, clear-actions also must
7003 * be supported at the same time.
7007 case OFPACT_WRITE_METADATA:
7008 metadata = ofpact_get_WRITE_METADATA(a);
7009 ctx->flow.metadata &= ~metadata->mask;
7010 ctx->flow.metadata |= metadata->metadata & metadata->mask;
7013 case OFPACT_GOTO_TABLE: {
7014 /* It is assumed that goto-table is the last action. */
7015 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
7016 struct rule_dpif *rule;
7018 ovs_assert(ctx->table_id < ogt->table_id);
7020 ctx->table_id = ogt->table_id;
7022 /* Look up a flow from the new table. */
7023 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
7025 tag_the_flow(ctx, rule);
7027 rule = ctx_rule_hooks(ctx, rule, true);
7031 ctx->rule->up.evictable = was_evictable;
7034 was_evictable = rule->up.evictable;
7035 rule->up.evictable = false;
7037 /* Tail recursion removal. */
7038 ofpacts = rule->up.ofpacts;
7039 ofpacts_len = rule->up.ofpacts_len;
7040 goto do_xlate_actions_again;
7046 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7053 ctx->rule->up.evictable = was_evictable;
7058 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7059 struct ofproto_dpif *ofproto, const struct flow *flow,
7060 const struct initial_vals *initial_vals,
7061 struct rule_dpif *rule,
7062 uint8_t tcp_flags, const struct ofpbuf *packet)
7064 /* Flow initialization rules:
7065 * - 'base_flow' must match the kernel's view of the packet at the
7066 * time that action processing starts. 'flow' represents any
7067 * transformations we wish to make through actions.
7068 * - By default 'base_flow' and 'flow' are the same since the input
7069 * packet matches the output before any actions are applied.
7070 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7071 * of the received packet as seen by the kernel. If we later output
7072 * to another device without any modifications this will cause us to
7073 * insert a new tag since the original one was stripped off by the
7075 * - Tunnel metadata as received is retained in 'flow'. This allows
7076 * tunnel metadata matching also in later tables.
7077 * Since a kernel action for setting the tunnel metadata will only be
7078 * generated with actual tunnel output, changing the tunnel metadata
7079 * values in 'flow' (such as tun_id) will only have effect with a later
7080 * tunnel output action.
7081 * - Tunnel 'base_flow' is completely cleared since that is what the
7082 * kernel does. If we wish to maintain the original values an action
7083 * needs to be generated. */
7085 ctx->ofproto = ofproto;
7087 ctx->base_flow = ctx->flow;
7088 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
7089 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7091 ctx->packet = packet;
7092 ctx->may_learn = packet != NULL;
7093 ctx->tcp_flags = tcp_flags;
7094 ctx->resubmit_hook = NULL;
7095 ctx->report_hook = NULL;
7096 ctx->resubmit_stats = NULL;
7099 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7100 * into datapath actions in 'odp_actions', using 'ctx'. */
7102 xlate_actions(struct action_xlate_ctx *ctx,
7103 const struct ofpact *ofpacts, size_t ofpacts_len,
7104 struct ofpbuf *odp_actions)
7106 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7107 * that in the future we always keep a copy of the original flow for
7108 * tracing purposes. */
7109 static bool hit_resubmit_limit;
7111 enum slow_path_reason special;
7112 struct ofport_dpif *in_port;
7113 struct flow orig_flow;
7115 COVERAGE_INC(ofproto_dpif_xlate);
7117 ofpbuf_clear(odp_actions);
7118 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7120 ctx->odp_actions = odp_actions;
7123 ctx->has_learn = false;
7124 ctx->has_normal = false;
7125 ctx->has_fin_timeout = false;
7126 ctx->nf_output_iface = NF_OUT_DROP;
7129 ctx->max_resubmit_trigger = false;
7130 ctx->orig_skb_priority = ctx->flow.skb_priority;
7134 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7136 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7137 /* Do this conditionally because the copy is expensive enough that it
7138 * shows up in profiles. */
7139 orig_flow = ctx->flow;
7142 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7143 switch (ctx->ofproto->up.frag_handling) {
7144 case OFPC_FRAG_NORMAL:
7145 /* We must pretend that transport ports are unavailable. */
7146 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7147 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7150 case OFPC_FRAG_DROP:
7153 case OFPC_FRAG_REASM:
7156 case OFPC_FRAG_NX_MATCH:
7157 /* Nothing to do. */
7160 case OFPC_INVALID_TTL_TO_CONTROLLER:
7165 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7166 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7168 ctx->slow |= special;
7170 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7171 struct initial_vals initial_vals;
7172 size_t sample_actions_len;
7173 uint32_t local_odp_port;
7175 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7177 add_sflow_action(ctx);
7178 add_ipfix_action(ctx);
7179 sample_actions_len = ctx->odp_actions->size;
7181 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7182 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7184 /* We've let OFPP_NORMAL and the learning action look at the
7185 * packet, so drop it now if forwarding is disabled. */
7186 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7187 ctx->odp_actions->size = sample_actions_len;
7191 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7192 if (!hit_resubmit_limit) {
7193 /* We didn't record the original flow. Make sure we do from
7195 hit_resubmit_limit = true;
7196 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7197 struct ds ds = DS_EMPTY_INITIALIZER;
7199 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7200 &initial_vals, &ds);
7201 VLOG_ERR("Trace triggered by excessive resubmit "
7202 "recursion:\n%s", ds_cstr(&ds));
7207 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7208 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
7210 ctx->odp_actions->data,
7211 ctx->odp_actions->size)) {
7212 ctx->slow |= SLOW_IN_BAND;
7214 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
7216 compose_output_action(ctx, OFPP_LOCAL);
7219 if (ctx->ofproto->has_mirrors) {
7220 add_mirror_actions(ctx, &orig_flow);
7222 fix_sflow_action(ctx);
7225 ofpbuf_uninit(&ctx->stack);
7228 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7229 * into datapath actions, using 'ctx', and discards the datapath actions. */
7231 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7232 const struct ofpact *ofpacts,
7235 uint64_t odp_actions_stub[1024 / 8];
7236 struct ofpbuf odp_actions;
7238 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7239 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7240 ofpbuf_uninit(&odp_actions);
7244 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7246 if (ctx->report_hook) {
7247 ctx->report_hook(ctx, s);
7251 /* OFPP_NORMAL implementation. */
7253 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7255 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7256 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7257 * the bundle on which the packet was received, returns the VLAN to which the
7260 * Both 'vid' and the return value are in the range 0...4095. */
7262 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7264 switch (in_bundle->vlan_mode) {
7265 case PORT_VLAN_ACCESS:
7266 return in_bundle->vlan;
7269 case PORT_VLAN_TRUNK:
7272 case PORT_VLAN_NATIVE_UNTAGGED:
7273 case PORT_VLAN_NATIVE_TAGGED:
7274 return vid ? vid : in_bundle->vlan;
7281 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7282 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7285 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7286 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7289 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7291 /* Allow any VID on the OFPP_NONE port. */
7292 if (in_bundle == &ofpp_none_bundle) {
7296 switch (in_bundle->vlan_mode) {
7297 case PORT_VLAN_ACCESS:
7300 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7301 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7302 "packet received on port %s configured as VLAN "
7303 "%"PRIu16" access port",
7304 in_bundle->ofproto->up.name, vid,
7305 in_bundle->name, in_bundle->vlan);
7311 case PORT_VLAN_NATIVE_UNTAGGED:
7312 case PORT_VLAN_NATIVE_TAGGED:
7314 /* Port must always carry its native VLAN. */
7318 case PORT_VLAN_TRUNK:
7319 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7321 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7322 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7323 "received on port %s not configured for trunking "
7325 in_bundle->ofproto->up.name, vid,
7326 in_bundle->name, vid);
7338 /* Given 'vlan', the VLAN that a packet belongs to, and
7339 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7340 * that should be included in the 802.1Q header. (If the return value is 0,
7341 * then the 802.1Q header should only be included in the packet if there is a
7344 * Both 'vlan' and the return value are in the range 0...4095. */
7346 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7348 switch (out_bundle->vlan_mode) {
7349 case PORT_VLAN_ACCESS:
7352 case PORT_VLAN_TRUNK:
7353 case PORT_VLAN_NATIVE_TAGGED:
7356 case PORT_VLAN_NATIVE_UNTAGGED:
7357 return vlan == out_bundle->vlan ? 0 : vlan;
7365 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7368 struct ofport_dpif *port;
7370 ovs_be16 tci, old_tci;
7372 vid = output_vlan_to_vid(out_bundle, vlan);
7373 if (!out_bundle->bond) {
7374 port = ofbundle_get_a_port(out_bundle);
7376 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7379 /* No slaves enabled, so drop packet. */
7384 old_tci = ctx->flow.vlan_tci;
7386 if (tci || out_bundle->use_priority_tags) {
7387 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7389 tci |= htons(VLAN_CFI);
7392 ctx->flow.vlan_tci = tci;
7394 compose_output_action(ctx, port->up.ofp_port);
7395 ctx->flow.vlan_tci = old_tci;
7399 mirror_mask_ffs(mirror_mask_t mask)
7401 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7406 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7408 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7409 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7413 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7415 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7418 /* Returns an arbitrary interface within 'bundle'. */
7419 static struct ofport_dpif *
7420 ofbundle_get_a_port(const struct ofbundle *bundle)
7422 return CONTAINER_OF(list_front(&bundle->ports),
7423 struct ofport_dpif, bundle_node);
7427 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7429 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7433 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7435 struct ofproto_dpif *ofproto = ctx->ofproto;
7436 mirror_mask_t mirrors;
7437 struct ofbundle *in_bundle;
7440 const struct nlattr *a;
7443 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7444 ctx->packet != NULL, NULL);
7448 mirrors = in_bundle->src_mirrors;
7450 /* Drop frames on bundles reserved for mirroring. */
7451 if (in_bundle->mirror_out) {
7452 if (ctx->packet != NULL) {
7453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7454 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7455 "%s, which is reserved exclusively for mirroring",
7456 ctx->ofproto->up.name, in_bundle->name);
7462 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7463 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7466 vlan = input_vid_to_vlan(in_bundle, vid);
7468 /* Look at the output ports to check for destination selections. */
7470 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7471 ctx->odp_actions->size) {
7472 enum ovs_action_attr type = nl_attr_type(a);
7473 struct ofport_dpif *ofport;
7475 if (type != OVS_ACTION_ATTR_OUTPUT) {
7479 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7480 if (ofport && ofport->bundle) {
7481 mirrors |= ofport->bundle->dst_mirrors;
7489 /* Restore the original packet before adding the mirror actions. */
7490 ctx->flow = *orig_flow;
7495 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7497 if (!vlan_is_mirrored(m, vlan)) {
7498 mirrors = zero_rightmost_1bit(mirrors);
7502 mirrors &= ~m->dup_mirrors;
7503 ctx->mirrors |= m->dup_mirrors;
7505 output_normal(ctx, m->out, vlan);
7506 } else if (vlan != m->out_vlan
7507 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7508 struct ofbundle *bundle;
7510 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7511 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7512 && !bundle->mirror_out) {
7513 output_normal(ctx, bundle, m->out_vlan);
7521 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7522 uint64_t packets, uint64_t bytes)
7528 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7531 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7534 /* In normal circumstances 'm' will not be NULL. However,
7535 * if mirrors are reconfigured, we can temporarily get out
7536 * of sync in facet_revalidate(). We could "correct" the
7537 * mirror list before reaching here, but doing that would
7538 * not properly account the traffic stats we've currently
7539 * accumulated for previous mirror configuration. */
7543 m->packet_count += packets;
7544 m->byte_count += bytes;
7548 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7549 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7550 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7552 is_gratuitous_arp(const struct flow *flow)
7554 return (flow->dl_type == htons(ETH_TYPE_ARP)
7555 && eth_addr_is_broadcast(flow->dl_dst)
7556 && (flow->nw_proto == ARP_OP_REPLY
7557 || (flow->nw_proto == ARP_OP_REQUEST
7558 && flow->nw_src == flow->nw_dst)));
7562 update_learning_table(struct ofproto_dpif *ofproto,
7563 const struct flow *flow, int vlan,
7564 struct ofbundle *in_bundle)
7566 struct mac_entry *mac;
7568 /* Don't learn the OFPP_NONE port. */
7569 if (in_bundle == &ofpp_none_bundle) {
7573 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7577 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7578 if (is_gratuitous_arp(flow)) {
7579 /* We don't want to learn from gratuitous ARP packets that are
7580 * reflected back over bond slaves so we lock the learning table. */
7581 if (!in_bundle->bond) {
7582 mac_entry_set_grat_arp_lock(mac);
7583 } else if (mac_entry_is_grat_arp_locked(mac)) {
7588 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7589 /* The log messages here could actually be useful in debugging,
7590 * so keep the rate limit relatively high. */
7591 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7592 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7593 "on port %s in VLAN %d",
7594 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7595 in_bundle->name, vlan);
7597 mac->port.p = in_bundle;
7598 tag_set_add(&ofproto->backer->revalidate_set,
7599 mac_learning_changed(ofproto->ml, mac));
7603 static struct ofbundle *
7604 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7605 bool warn, struct ofport_dpif **in_ofportp)
7607 struct ofport_dpif *ofport;
7609 /* Find the port and bundle for the received packet. */
7610 ofport = get_ofp_port(ofproto, in_port);
7612 *in_ofportp = ofport;
7614 if (ofport && ofport->bundle) {
7615 return ofport->bundle;
7618 /* Special-case OFPP_NONE, which a controller may use as the ingress
7619 * port for traffic that it is sourcing. */
7620 if (in_port == OFPP_NONE) {
7621 return &ofpp_none_bundle;
7624 /* Odd. A few possible reasons here:
7626 * - We deleted a port but there are still a few packets queued up
7629 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7630 * we don't know about.
7632 * - The ofproto client didn't configure the port as part of a bundle.
7633 * This is particularly likely to happen if a packet was received on the
7634 * port after it was created, but before the client had a chance to
7635 * configure its bundle.
7638 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7640 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7641 "port %"PRIu16, ofproto->up.name, in_port);
7646 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7647 * dropped. Returns true if they may be forwarded, false if they should be
7650 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7651 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7653 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7654 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7655 * checked by input_vid_is_valid().
7657 * May also add tags to '*tags', although the current implementation only does
7658 * so in one special case.
7661 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7664 struct ofproto_dpif *ofproto = ctx->ofproto;
7665 struct flow *flow = &ctx->flow;
7666 struct ofbundle *in_bundle = in_port->bundle;
7668 /* Drop frames for reserved multicast addresses
7669 * only if forward_bpdu option is absent. */
7670 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7671 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7675 if (in_bundle->bond) {
7676 struct mac_entry *mac;
7678 switch (bond_check_admissibility(in_bundle->bond, in_port,
7679 flow->dl_dst, &ctx->tags)) {
7684 xlate_report(ctx, "bonding refused admissibility, dropping");
7687 case BV_DROP_IF_MOVED:
7688 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7689 if (mac && mac->port.p != in_bundle &&
7690 (!is_gratuitous_arp(flow)
7691 || mac_entry_is_grat_arp_locked(mac))) {
7692 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7704 xlate_normal(struct action_xlate_ctx *ctx)
7706 struct ofport_dpif *in_port;
7707 struct ofbundle *in_bundle;
7708 struct mac_entry *mac;
7712 ctx->has_normal = true;
7714 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7715 ctx->packet != NULL, &in_port);
7717 xlate_report(ctx, "no input bundle, dropping");
7721 /* Drop malformed frames. */
7722 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7723 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7724 if (ctx->packet != NULL) {
7725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7726 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7727 "VLAN tag received on port %s",
7728 ctx->ofproto->up.name, in_bundle->name);
7730 xlate_report(ctx, "partial VLAN tag, dropping");
7734 /* Drop frames on bundles reserved for mirroring. */
7735 if (in_bundle->mirror_out) {
7736 if (ctx->packet != NULL) {
7737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7738 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7739 "%s, which is reserved exclusively for mirroring",
7740 ctx->ofproto->up.name, in_bundle->name);
7742 xlate_report(ctx, "input port is mirror output port, dropping");
7747 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7748 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7749 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7752 vlan = input_vid_to_vlan(in_bundle, vid);
7754 /* Check other admissibility requirements. */
7755 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7759 /* Learn source MAC. */
7760 if (ctx->may_learn) {
7761 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7764 /* Determine output bundle. */
7765 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7768 if (mac->port.p != in_bundle) {
7769 xlate_report(ctx, "forwarding to learned port");
7770 output_normal(ctx, mac->port.p, vlan);
7772 xlate_report(ctx, "learned port is input port, dropping");
7775 struct ofbundle *bundle;
7777 xlate_report(ctx, "no learned MAC for destination, flooding");
7778 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7779 if (bundle != in_bundle
7780 && ofbundle_includes_vlan(bundle, vlan)
7781 && bundle->floodable
7782 && !bundle->mirror_out) {
7783 output_normal(ctx, bundle, vlan);
7786 ctx->nf_output_iface = NF_OUT_FLOOD;
7790 /* Optimized flow revalidation.
7792 * It's a difficult problem, in general, to tell which facets need to have
7793 * their actions recalculated whenever the OpenFlow flow table changes. We
7794 * don't try to solve that general problem: for most kinds of OpenFlow flow
7795 * table changes, we recalculate the actions for every facet. This is
7796 * relatively expensive, but it's good enough if the OpenFlow flow table
7797 * doesn't change very often.
7799 * However, we can expect one particular kind of OpenFlow flow table change to
7800 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7801 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7802 * table, we add a special case that applies to flow tables in which every rule
7803 * has the same form (that is, the same wildcards), except that the table is
7804 * also allowed to have a single "catch-all" flow that matches all packets. We
7805 * optimize this case by tagging all of the facets that resubmit into the table
7806 * and invalidating the same tag whenever a flow changes in that table. The
7807 * end result is that we revalidate just the facets that need it (and sometimes
7808 * a few more, but not all of the facets or even all of the facets that
7809 * resubmit to the table modified by MAC learning). */
7811 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7812 * into an OpenFlow table with the given 'basis'. */
7814 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7817 if (minimask_is_catchall(mask)) {
7820 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7821 return tag_create_deterministic(hash);
7825 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7826 * taggability of that table.
7828 * This function must be called after *each* change to a flow table. If you
7829 * skip calling it on some changes then the pointer comparisons at the end can
7830 * be invalid if you get unlucky. For example, if a flow removal causes a
7831 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7832 * different wildcards to be created with the same address, then this function
7833 * will incorrectly skip revalidation. */
7835 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7837 struct table_dpif *table = &ofproto->tables[table_id];
7838 const struct oftable *oftable = &ofproto->up.tables[table_id];
7839 struct cls_table *catchall, *other;
7840 struct cls_table *t;
7842 catchall = other = NULL;
7844 switch (hmap_count(&oftable->cls.tables)) {
7846 /* We could tag this OpenFlow table but it would make the logic a
7847 * little harder and it's a corner case that doesn't seem worth it
7853 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7854 if (cls_table_is_catchall(t)) {
7856 } else if (!other) {
7859 /* Indicate that we can't tag this by setting both tables to
7860 * NULL. (We know that 'catchall' is already NULL.) */
7867 /* Can't tag this table. */
7871 if (table->catchall_table != catchall || table->other_table != other) {
7872 table->catchall_table = catchall;
7873 table->other_table = other;
7874 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7878 /* Given 'rule' that has changed in some way (either it is a rule being
7879 * inserted, a rule being deleted, or a rule whose actions are being
7880 * modified), marks facets for revalidation to ensure that packets will be
7881 * forwarded correctly according to the new state of the flow table.
7883 * This function must be called after *each* change to a flow table. See
7884 * the comment on table_update_taggable() for more information. */
7886 rule_invalidate(const struct rule_dpif *rule)
7888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7890 table_update_taggable(ofproto, rule->up.table_id);
7892 if (!ofproto->backer->need_revalidate) {
7893 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7895 if (table->other_table && rule->tag) {
7896 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7898 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7904 set_frag_handling(struct ofproto *ofproto_,
7905 enum ofp_config_flags frag_handling)
7907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7908 if (frag_handling != OFPC_FRAG_REASM) {
7909 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7917 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7918 const struct flow *flow,
7919 const struct ofpact *ofpacts, size_t ofpacts_len)
7921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7922 struct initial_vals initial_vals;
7923 struct odputil_keybuf keybuf;
7924 struct dpif_flow_stats stats;
7928 struct action_xlate_ctx ctx;
7929 uint64_t odp_actions_stub[1024 / 8];
7930 struct ofpbuf odp_actions;
7932 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7933 odp_flow_key_from_flow(&key, flow,
7934 ofp_port_to_odp_port(ofproto, flow->in_port));
7936 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7938 initial_vals.vlan_tci = flow->vlan_tci;
7939 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7940 packet_get_tcp_flags(packet, flow), packet);
7941 ctx.resubmit_stats = &stats;
7943 ofpbuf_use_stub(&odp_actions,
7944 odp_actions_stub, sizeof odp_actions_stub);
7945 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7946 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7947 odp_actions.data, odp_actions.size, packet);
7948 ofpbuf_uninit(&odp_actions);
7956 set_netflow(struct ofproto *ofproto_,
7957 const struct netflow_options *netflow_options)
7959 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7961 if (netflow_options) {
7962 if (!ofproto->netflow) {
7963 ofproto->netflow = netflow_create();
7965 return netflow_set_options(ofproto->netflow, netflow_options);
7967 netflow_destroy(ofproto->netflow);
7968 ofproto->netflow = NULL;
7974 get_netflow_ids(const struct ofproto *ofproto_,
7975 uint8_t *engine_type, uint8_t *engine_id)
7977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7979 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7983 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7985 if (!facet_is_controller_flow(facet) &&
7986 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7987 struct subfacet *subfacet;
7988 struct ofexpired expired;
7990 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7991 if (subfacet->path == SF_FAST_PATH) {
7992 struct dpif_flow_stats stats;
7994 subfacet_reinstall(subfacet, &stats);
7995 subfacet_update_stats(subfacet, &stats);
7999 expired.flow = facet->flow;
8000 expired.packet_count = facet->packet_count;
8001 expired.byte_count = facet->byte_count;
8002 expired.used = facet->used;
8003 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8008 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8010 struct facet *facet;
8012 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8013 send_active_timeout(ofproto, facet);
8017 static struct ofproto_dpif *
8018 ofproto_dpif_lookup(const char *name)
8020 struct ofproto_dpif *ofproto;
8022 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8023 hash_string(name, 0), &all_ofproto_dpifs) {
8024 if (!strcmp(ofproto->up.name, name)) {
8032 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8033 const char *argv[], void *aux OVS_UNUSED)
8035 struct ofproto_dpif *ofproto;
8038 ofproto = ofproto_dpif_lookup(argv[1]);
8040 unixctl_command_reply_error(conn, "no such bridge");
8043 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8045 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8046 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8050 unixctl_command_reply(conn, "table successfully flushed");
8054 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8055 const char *argv[], void *aux OVS_UNUSED)
8057 struct ds ds = DS_EMPTY_INITIALIZER;
8058 const struct ofproto_dpif *ofproto;
8059 const struct mac_entry *e;
8061 ofproto = ofproto_dpif_lookup(argv[1]);
8063 unixctl_command_reply_error(conn, "no such bridge");
8067 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8068 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8069 struct ofbundle *bundle = e->port.p;
8070 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8071 ofbundle_get_a_port(bundle)->odp_port,
8072 e->vlan, ETH_ADDR_ARGS(e->mac),
8073 mac_entry_age(ofproto->ml, e));
8075 unixctl_command_reply(conn, ds_cstr(&ds));
8080 struct action_xlate_ctx ctx;
8086 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8087 const struct rule_dpif *rule)
8089 ds_put_char_multiple(result, '\t', level);
8091 ds_put_cstr(result, "No match\n");
8095 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8096 table_id, ntohll(rule->up.flow_cookie));
8097 cls_rule_format(&rule->up.cr, result);
8098 ds_put_char(result, '\n');
8100 ds_put_char_multiple(result, '\t', level);
8101 ds_put_cstr(result, "OpenFlow ");
8102 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8103 ds_put_char(result, '\n');
8107 trace_format_flow(struct ds *result, int level, const char *title,
8108 struct trace_ctx *trace)
8110 ds_put_char_multiple(result, '\t', level);
8111 ds_put_format(result, "%s: ", title);
8112 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8113 ds_put_cstr(result, "unchanged");
8115 flow_format(result, &trace->ctx.flow);
8116 trace->flow = trace->ctx.flow;
8118 ds_put_char(result, '\n');
8122 trace_format_regs(struct ds *result, int level, const char *title,
8123 struct trace_ctx *trace)
8127 ds_put_char_multiple(result, '\t', level);
8128 ds_put_format(result, "%s:", title);
8129 for (i = 0; i < FLOW_N_REGS; i++) {
8130 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8132 ds_put_char(result, '\n');
8136 trace_format_odp(struct ds *result, int level, const char *title,
8137 struct trace_ctx *trace)
8139 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8141 ds_put_char_multiple(result, '\t', level);
8142 ds_put_format(result, "%s: ", title);
8143 format_odp_actions(result, odp_actions->data, odp_actions->size);
8144 ds_put_char(result, '\n');
8148 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8150 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8151 struct ds *result = trace->result;
8153 ds_put_char(result, '\n');
8154 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8155 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8156 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8157 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8161 trace_report(struct action_xlate_ctx *ctx, const char *s)
8163 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8164 struct ds *result = trace->result;
8166 ds_put_char_multiple(result, '\t', ctx->recurse);
8167 ds_put_cstr(result, s);
8168 ds_put_char(result, '\n');
8172 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8173 void *aux OVS_UNUSED)
8175 const char *dpname = argv[1];
8176 struct ofproto_dpif *ofproto;
8177 struct ofpbuf odp_key;
8178 struct ofpbuf *packet;
8179 struct initial_vals initial_vals;
8185 ofpbuf_init(&odp_key, 0);
8188 ofproto = ofproto_dpif_lookup(dpname);
8190 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8194 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8195 /* ofproto/trace dpname flow [-generate] */
8196 const char *flow_s = argv[2];
8197 const char *generate_s = argv[3];
8199 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8200 * flow. We guess which type it is based on whether 'flow_s' contains
8201 * an '(', since a datapath flow always contains '(') but an
8202 * OpenFlow-like flow should not (in fact it's allowed but I believe
8203 * that's not documented anywhere).
8205 * An alternative would be to try to parse 'flow_s' both ways, but then
8206 * it would be tricky giving a sensible error message. After all, do
8207 * you just say "syntax error" or do you present both error messages?
8208 * Both choices seem lousy. */
8209 if (strchr(flow_s, '(')) {
8212 /* Convert string to datapath key. */
8213 ofpbuf_init(&odp_key, 0);
8214 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8216 unixctl_command_reply_error(conn, "Bad flow syntax");
8220 /* The user might have specified the wrong ofproto but within the
8221 * same backer. That's OK, ofproto_receive() can find the right
8223 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8224 odp_key.size, &flow, NULL, &ofproto, NULL,
8226 unixctl_command_reply_error(conn, "Invalid flow");
8229 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8233 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8235 unixctl_command_reply_error(conn, error_s);
8240 initial_vals.vlan_tci = flow.vlan_tci;
8243 /* Generate a packet, if requested. */
8245 packet = ofpbuf_new(0);
8246 flow_compose(packet, &flow);
8248 } else if (argc == 7) {
8249 /* ofproto/trace dpname priority tun_id in_port mark packet */
8250 const char *priority_s = argv[2];
8251 const char *tun_id_s = argv[3];
8252 const char *in_port_s = argv[4];
8253 const char *mark_s = argv[5];
8254 const char *packet_s = argv[6];
8255 uint32_t in_port = atoi(in_port_s);
8256 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8257 uint32_t priority = atoi(priority_s);
8258 uint32_t mark = atoi(mark_s);
8261 msg = eth_from_hex(packet_s, &packet);
8263 unixctl_command_reply_error(conn, msg);
8267 ds_put_cstr(&result, "Packet: ");
8268 s = ofp_packet_to_string(packet->data, packet->size);
8269 ds_put_cstr(&result, s);
8272 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8273 flow.tunnel.tun_id = tun_id;
8274 initial_vals.vlan_tci = flow.vlan_tci;
8276 unixctl_command_reply_error(conn, "Bad command syntax");
8280 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8281 unixctl_command_reply(conn, ds_cstr(&result));
8284 ds_destroy(&result);
8285 ofpbuf_delete(packet);
8286 ofpbuf_uninit(&odp_key);
8290 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8291 const struct ofpbuf *packet,
8292 const struct initial_vals *initial_vals, struct ds *ds)
8294 struct rule_dpif *rule;
8296 ds_put_cstr(ds, "Flow: ");
8297 flow_format(ds, flow);
8298 ds_put_char(ds, '\n');
8300 rule = rule_dpif_lookup(ofproto, flow);
8302 trace_format_rule(ds, 0, 0, rule);
8303 if (rule == ofproto->miss_rule) {
8304 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8305 } else if (rule == ofproto->no_packet_in_rule) {
8306 ds_put_cstr(ds, "\nNo match, packets dropped because "
8307 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8311 uint64_t odp_actions_stub[1024 / 8];
8312 struct ofpbuf odp_actions;
8314 struct trace_ctx trace;
8317 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8320 ofpbuf_use_stub(&odp_actions,
8321 odp_actions_stub, sizeof odp_actions_stub);
8322 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8323 rule, tcp_flags, packet);
8324 trace.ctx.resubmit_hook = trace_resubmit;
8325 trace.ctx.report_hook = trace_report;
8326 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8329 ds_put_char(ds, '\n');
8330 trace_format_flow(ds, 0, "Final flow", &trace);
8331 ds_put_cstr(ds, "Datapath actions: ");
8332 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8333 ofpbuf_uninit(&odp_actions);
8335 if (trace.ctx.slow) {
8336 enum slow_path_reason slow;
8338 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8339 "slow path because it:");
8340 for (slow = trace.ctx.slow; slow; ) {
8341 enum slow_path_reason bit = rightmost_1bit(slow);
8345 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8348 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8351 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8354 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8357 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8360 ds_put_cstr(ds, "\n\t (The datapath actions are "
8361 "incomplete--for complete actions, "
8362 "please supply a packet.)");
8365 case SLOW_CONTROLLER:
8366 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8367 "to the OpenFlow controller.");
8370 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8371 "than the datapath supports.");
8378 if (slow & ~SLOW_MATCH) {
8379 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8380 "the special slow-path processing.");
8387 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8388 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8391 unixctl_command_reply(conn, NULL);
8395 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8396 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8399 unixctl_command_reply(conn, NULL);
8402 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8403 * 'reply' describing the results. */
8405 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8407 struct facet *facet;
8411 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8412 if (!facet_check_consistency(facet)) {
8417 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8421 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8422 ofproto->up.name, errors);
8424 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8429 ofproto_dpif_self_check(struct unixctl_conn *conn,
8430 int argc, const char *argv[], void *aux OVS_UNUSED)
8432 struct ds reply = DS_EMPTY_INITIALIZER;
8433 struct ofproto_dpif *ofproto;
8436 ofproto = ofproto_dpif_lookup(argv[1]);
8438 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8439 "ofproto/list for help)");
8442 ofproto_dpif_self_check__(ofproto, &reply);
8444 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8445 ofproto_dpif_self_check__(ofproto, &reply);
8449 unixctl_command_reply(conn, ds_cstr(&reply));
8453 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8454 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8455 * to destroy 'ofproto_shash' and free the returned value. */
8456 static const struct shash_node **
8457 get_ofprotos(struct shash *ofproto_shash)
8459 const struct ofproto_dpif *ofproto;
8461 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8462 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8463 shash_add_nocopy(ofproto_shash, name, ofproto);
8466 return shash_sort(ofproto_shash);
8470 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8471 const char *argv[] OVS_UNUSED,
8472 void *aux OVS_UNUSED)
8474 struct ds ds = DS_EMPTY_INITIALIZER;
8475 struct shash ofproto_shash;
8476 const struct shash_node **sorted_ofprotos;
8479 shash_init(&ofproto_shash);
8480 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8481 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8482 const struct shash_node *node = sorted_ofprotos[i];
8483 ds_put_format(&ds, "%s\n", node->name);
8486 shash_destroy(&ofproto_shash);
8487 free(sorted_ofprotos);
8489 unixctl_command_reply(conn, ds_cstr(&ds));
8494 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8496 const struct shash_node **ports;
8498 struct avg_subfacet_rates lifetime;
8499 unsigned long long int minutes;
8500 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8502 minutes = (time_msec() - ofproto->created) / min_ms;
8505 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8507 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8510 lifetime.add_rate = 0.0;
8511 lifetime.del_rate = 0.0;
8514 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8515 dpif_name(ofproto->backer->dpif));
8517 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8518 ofproto->n_hit, ofproto->n_missed);
8519 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8520 " life span: %llu(ms)\n",
8521 hmap_count(&ofproto->subfacets),
8522 avg_subfacet_count(ofproto),
8523 ofproto->max_n_subfacet,
8524 avg_subfacet_life_span(ofproto));
8525 if (minutes >= 60) {
8526 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8528 if (minutes >= 60 * 24) {
8529 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8531 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8533 ports = shash_sort(&ofproto->up.port_by_name);
8534 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8535 const struct shash_node *node = ports[i];
8536 struct ofport *ofport = node->data;
8537 const char *name = netdev_get_name(ofport->netdev);
8538 const char *type = netdev_get_type(ofport->netdev);
8541 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8543 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8544 if (odp_port != OVSP_NONE) {
8545 ds_put_format(ds, "%"PRIu32":", odp_port);
8547 ds_put_cstr(ds, "none:");
8550 if (strcmp(type, "system")) {
8551 struct netdev *netdev;
8554 ds_put_format(ds, " (%s", type);
8556 error = netdev_open(name, type, &netdev);
8561 error = netdev_get_config(netdev, &config);
8563 const struct smap_node **nodes;
8566 nodes = smap_sort(&config);
8567 for (i = 0; i < smap_count(&config); i++) {
8568 const struct smap_node *node = nodes[i];
8569 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8570 node->key, node->value);
8574 smap_destroy(&config);
8576 netdev_close(netdev);
8578 ds_put_char(ds, ')');
8580 ds_put_char(ds, '\n');
8586 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8587 const char *argv[], void *aux OVS_UNUSED)
8589 struct ds ds = DS_EMPTY_INITIALIZER;
8590 const struct ofproto_dpif *ofproto;
8594 for (i = 1; i < argc; i++) {
8595 ofproto = ofproto_dpif_lookup(argv[i]);
8597 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8598 "for help)", argv[i]);
8599 unixctl_command_reply_error(conn, ds_cstr(&ds));
8602 show_dp_format(ofproto, &ds);
8605 struct shash ofproto_shash;
8606 const struct shash_node **sorted_ofprotos;
8609 shash_init(&ofproto_shash);
8610 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8611 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8612 const struct shash_node *node = sorted_ofprotos[i];
8613 show_dp_format(node->data, &ds);
8616 shash_destroy(&ofproto_shash);
8617 free(sorted_ofprotos);
8620 unixctl_command_reply(conn, ds_cstr(&ds));
8625 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8626 int argc OVS_UNUSED, const char *argv[],
8627 void *aux OVS_UNUSED)
8629 struct ds ds = DS_EMPTY_INITIALIZER;
8630 const struct ofproto_dpif *ofproto;
8631 struct subfacet *subfacet;
8633 ofproto = ofproto_dpif_lookup(argv[1]);
8635 unixctl_command_reply_error(conn, "no such bridge");
8639 update_stats(ofproto->backer);
8641 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8642 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8644 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8645 subfacet->dp_packet_count, subfacet->dp_byte_count);
8646 if (subfacet->used) {
8647 ds_put_format(&ds, "%.3fs",
8648 (time_msec() - subfacet->used) / 1000.0);
8650 ds_put_format(&ds, "never");
8652 if (subfacet->facet->tcp_flags) {
8653 ds_put_cstr(&ds, ", flags:");
8654 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8657 ds_put_cstr(&ds, ", actions:");
8658 if (subfacet->slow) {
8659 uint64_t slow_path_stub[128 / 8];
8660 const struct nlattr *actions;
8663 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8664 slow_path_stub, sizeof slow_path_stub,
8665 &actions, &actions_len);
8666 format_odp_actions(&ds, actions, actions_len);
8668 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8670 ds_put_char(&ds, '\n');
8673 unixctl_command_reply(conn, ds_cstr(&ds));
8678 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8679 int argc OVS_UNUSED, const char *argv[],
8680 void *aux OVS_UNUSED)
8682 struct ds ds = DS_EMPTY_INITIALIZER;
8683 struct ofproto_dpif *ofproto;
8685 ofproto = ofproto_dpif_lookup(argv[1]);
8687 unixctl_command_reply_error(conn, "no such bridge");
8691 flush(&ofproto->up);
8693 unixctl_command_reply(conn, ds_cstr(&ds));
8698 ofproto_dpif_unixctl_init(void)
8700 static bool registered;
8706 unixctl_command_register(
8708 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8709 2, 6, ofproto_unixctl_trace, NULL);
8710 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8711 ofproto_unixctl_fdb_flush, NULL);
8712 unixctl_command_register("fdb/show", "bridge", 1, 1,
8713 ofproto_unixctl_fdb_show, NULL);
8714 unixctl_command_register("ofproto/clog", "", 0, 0,
8715 ofproto_dpif_clog, NULL);
8716 unixctl_command_register("ofproto/unclog", "", 0, 0,
8717 ofproto_dpif_unclog, NULL);
8718 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8719 ofproto_dpif_self_check, NULL);
8720 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8721 ofproto_unixctl_dpif_dump_dps, NULL);
8722 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8723 ofproto_unixctl_dpif_show, NULL);
8724 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8725 ofproto_unixctl_dpif_dump_flows, NULL);
8726 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8727 ofproto_unixctl_dpif_del_flows, NULL);
8730 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8732 * This is deprecated. It is only for compatibility with broken device drivers
8733 * in old versions of Linux that do not properly support VLANs when VLAN
8734 * devices are not used. When broken device drivers are no longer in
8735 * widespread use, we will delete these interfaces. */
8738 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8741 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8743 if (realdev_ofp_port == ofport->realdev_ofp_port
8744 && vid == ofport->vlandev_vid) {
8748 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8750 if (ofport->realdev_ofp_port) {
8753 if (realdev_ofp_port && ofport->bundle) {
8754 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8755 * themselves be part of a bundle. */
8756 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8759 ofport->realdev_ofp_port = realdev_ofp_port;
8760 ofport->vlandev_vid = vid;
8762 if (realdev_ofp_port) {
8763 vsp_add(ofport, realdev_ofp_port, vid);
8770 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8772 return hash_2words(realdev_ofp_port, vid);
8775 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8776 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8777 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8778 * it would return the port number of eth0.9.
8780 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8781 * function just returns its 'realdev_odp_port' argument. */
8783 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8784 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8786 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8787 uint16_t realdev_ofp_port;
8788 int vid = vlan_tci_to_vid(vlan_tci);
8789 const struct vlan_splinter *vsp;
8791 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8792 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8793 hash_realdev_vid(realdev_ofp_port, vid),
8794 &ofproto->realdev_vid_map) {
8795 if (vsp->realdev_ofp_port == realdev_ofp_port
8796 && vsp->vid == vid) {
8797 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8801 return realdev_odp_port;
8804 static struct vlan_splinter *
8805 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8807 struct vlan_splinter *vsp;
8809 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8810 &ofproto->vlandev_map) {
8811 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8819 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8820 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8821 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8822 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8823 * eth0 and store 9 in '*vid'.
8825 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8826 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8829 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8830 uint16_t vlandev_ofp_port, int *vid)
8832 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8833 const struct vlan_splinter *vsp;
8835 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8840 return vsp->realdev_ofp_port;
8846 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8847 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8848 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8849 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8850 * always the case unless VLAN splinters are enabled), returns false without
8851 * making any changes. */
8853 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8858 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8863 /* Cause the flow to be processed as if it came in on the real device with
8864 * the VLAN device's VLAN ID. */
8865 flow->in_port = realdev;
8866 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8871 vsp_remove(struct ofport_dpif *port)
8873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8874 struct vlan_splinter *vsp;
8876 vsp = vlandev_find(ofproto, port->up.ofp_port);
8878 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8879 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8882 port->realdev_ofp_port = 0;
8884 VLOG_ERR("missing vlan device record");
8889 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8893 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8894 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8895 == realdev_ofp_port)) {
8896 struct vlan_splinter *vsp;
8898 vsp = xmalloc(sizeof *vsp);
8899 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8900 hash_int(port->up.ofp_port, 0));
8901 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8902 hash_realdev_vid(realdev_ofp_port, vid));
8903 vsp->realdev_ofp_port = realdev_ofp_port;
8904 vsp->vlandev_ofp_port = port->up.ofp_port;
8907 port->realdev_ofp_port = realdev_ofp_port;
8909 VLOG_ERR("duplicate vlan device record");
8914 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8916 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8917 return ofport ? ofport->odp_port : OVSP_NONE;
8920 static struct ofport_dpif *
8921 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8923 struct ofport_dpif *port;
8925 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8926 hash_int(odp_port, 0),
8927 &backer->odp_to_ofport_map) {
8928 if (port->odp_port == odp_port) {
8937 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8939 struct ofport_dpif *port;
8941 port = odp_port_to_ofport(ofproto->backer, odp_port);
8942 if (port && &ofproto->up == port->up.ofproto) {
8943 return port->up.ofp_port;
8948 static unsigned long long int
8949 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8951 unsigned long long int dc;
8952 unsigned long long int avg;
8954 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8955 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8961 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8965 if (ofproto->n_update_stats) {
8966 avg_c = (double)ofproto->total_subfacet_count
8967 / ofproto->n_update_stats;
8974 show_dp_rates(struct ds *ds, const char *heading,
8975 const struct avg_subfacet_rates *rates)
8977 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8978 heading, rates->add_rate, rates->del_rate);
8982 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8984 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8985 hmap_count(&ofproto->subfacets));
8988 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8989 * most heavily weighted element. 'base' designates the rate of decay: after
8990 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8993 exp_mavg(double *avg, int base, double new)
8995 *avg = (*avg * (base - 1) + new) / base;
8999 update_moving_averages(struct ofproto_dpif *ofproto)
9001 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9003 /* Update hourly averages on the minute boundaries. */
9004 if (time_msec() - ofproto->last_minute >= min_ms) {
9005 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9006 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9008 /* Update daily averages on the hour boundaries. */
9009 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9010 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9011 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9014 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9015 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9016 ofproto->subfacet_add_count = 0;
9017 ofproto->subfacet_del_count = 0;
9018 ofproto->last_minute += min_ms;
9023 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9025 ofproto->n_hit += delta;
9028 const struct ofproto_class ofproto_dpif_class = {
9063 port_is_lacp_current,
9064 NULL, /* rule_choose_table */
9071 rule_modify_actions,
9085 get_stp_port_status,
9092 is_mirror_output_bundle,
9093 forward_bpdu_changed,
9094 set_mac_table_config,