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-sflow.h"
51 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
87 * - Do include packets and bytes from facets that have been deleted or
88 * whose own statistics have been folded into the rule.
90 * - Do include packets and bytes sent "by hand" that were accounted to
91 * the rule without any facet being involved (this is a rare corner
92 * case in rule_execute()).
94 * - Do not include packet or bytes that can be obtained from any facet's
95 * packet_count or byte_count member or that can be obtained from the
96 * datapath by, e.g., dpif_flow_get() for any subfacet.
98 uint64_t packet_count; /* Number of packets received. */
99 uint64_t byte_count; /* Number of bytes received. */
101 tag_type tag; /* Caches rule_calculate_tag() result. */
103 struct list facets; /* List of "struct facet"s. */
106 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
108 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
111 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
112 const struct flow *);
113 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
116 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
117 const struct flow *flow);
119 static void rule_credit_stats(struct rule_dpif *,
120 const struct dpif_flow_stats *);
121 static void flow_push_stats(struct rule_dpif *, const struct flow *,
122 const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* The packet corresponding to 'flow', or a null pointer if we are
220 * revalidating without a packet to refer to. */
221 const struct ofpbuf *packet;
223 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
224 * actions update the flow table?
226 * We want to update these tables if we are actually processing a packet,
227 * or if we are accounting for packets that the datapath has processed, but
228 * not if we are just revalidating. */
231 /* The rule that we are currently translating, or NULL. */
232 struct rule_dpif *rule;
234 /* Union of the set of TCP flags seen so far in this flow. (Used only by
235 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
239 /* If nonnull, flow translation calls this function just before executing a
240 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
241 * when the recursion depth is exceeded.
243 * 'rule' is the rule being submitted into. It will be null if the
244 * resubmit or OFPP_TABLE action didn't find a matching rule.
246 * This is normally null so the client has to set it manually after
247 * calling action_xlate_ctx_init(). */
248 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
250 /* If nonnull, flow translation calls this function to report some
251 * significant decision, e.g. to explain why OFPP_NORMAL translation
252 * dropped a packet. */
253 void (*report_hook)(struct action_xlate_ctx *, const char *s);
255 /* If nonnull, flow translation credits the specified statistics to each
256 * rule reached through a resubmit or OFPP_TABLE action.
258 * This is normally null so the client has to set it manually after
259 * calling action_xlate_ctx_init(). */
260 const struct dpif_flow_stats *resubmit_stats;
262 /* xlate_actions() initializes and uses these members. The client might want
263 * to look at them after it returns. */
265 struct ofpbuf *odp_actions; /* Datapath actions. */
266 tag_type tags; /* Tags associated with actions. */
267 enum slow_path_reason slow; /* 0 if fast path may be used. */
268 bool has_learn; /* Actions include NXAST_LEARN? */
269 bool has_normal; /* Actions output to OFPP_NORMAL? */
270 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
271 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
272 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
274 /* xlate_actions() initializes and uses these members, but the client has no
275 * reason to look at them. */
277 int recurse; /* Recursion level, via xlate_table_action. */
278 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
279 struct flow base_flow; /* Flow at the last commit. */
280 uint32_t orig_skb_priority; /* Priority when packet arrived. */
281 uint8_t table_id; /* OpenFlow table ID where flow was found. */
282 uint32_t sflow_n_outputs; /* Number of output ports. */
283 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
284 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
285 bool exit; /* No further actions should be processed. */
286 struct flow orig_flow; /* Copy of original flow. */
289 static void action_xlate_ctx_init(struct action_xlate_ctx *,
290 struct ofproto_dpif *, const struct flow *,
291 ovs_be16 initial_tci, struct rule_dpif *,
292 uint8_t tcp_flags, const struct ofpbuf *);
293 static void xlate_actions(struct action_xlate_ctx *,
294 const struct ofpact *ofpacts, size_t ofpacts_len,
295 struct ofpbuf *odp_actions);
296 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
297 const struct ofpact *ofpacts,
299 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
300 uint8_t table_id, bool may_packet_in);
302 static size_t put_userspace_action(const struct ofproto_dpif *,
303 struct ofpbuf *odp_actions,
305 const union user_action_cookie *);
307 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
308 enum slow_path_reason,
309 uint64_t *stub, size_t stub_size,
310 const struct nlattr **actionsp,
311 size_t *actions_lenp);
313 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
315 /* A subfacet (see "struct subfacet" below) has three possible installation
318 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
319 * case just after the subfacet is created, just before the subfacet is
320 * destroyed, or if the datapath returns an error when we try to install a
323 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
325 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
326 * ofproto_dpif is installed in the datapath.
329 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
330 SF_FAST_PATH, /* Full actions are installed. */
331 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
334 static const char *subfacet_path_to_string(enum subfacet_path);
336 /* A dpif flow and actions associated with a facet.
338 * See also the large comment on struct facet. */
341 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
342 struct list list_node; /* In struct facet's 'facets' list. */
343 struct facet *facet; /* Owning facet. */
347 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
348 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
349 * regenerate the ODP flow key from ->facet->flow. */
350 enum odp_key_fitness key_fitness;
354 long long int used; /* Time last used; time created if not used. */
356 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
357 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
361 * These should be essentially identical for every subfacet in a facet, but
362 * may differ in trivial ways due to VLAN splinters. */
363 size_t actions_len; /* Number of bytes in actions[]. */
364 struct nlattr *actions; /* Datapath actions. */
366 enum slow_path_reason slow; /* 0 if fast path may be used. */
367 enum subfacet_path path; /* Installed in datapath? */
369 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
370 * splinters can cause it to differ. This value should be removed when
371 * the VLAN splinters feature is no longer needed. */
372 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
374 /* Datapath port the packet arrived on. This is needed to remove
375 * flows for ports that are no longer part of the bridge. Since the
376 * flow definition only has the OpenFlow port number and the port is
377 * no longer part of the bridge, we can't determine the datapath port
378 * number needed to delete the flow from the datapath. */
379 uint32_t odp_in_port;
382 #define SUBFACET_DESTROY_MAX_BATCH 50
384 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
386 static struct subfacet *subfacet_find(struct ofproto_dpif *,
387 const struct nlattr *key, size_t key_len,
389 const struct flow *flow);
390 static void subfacet_destroy(struct subfacet *);
391 static void subfacet_destroy__(struct subfacet *);
392 static void subfacet_destroy_batch(struct ofproto_dpif *,
393 struct subfacet **, int n);
394 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
396 static void subfacet_reset_dp_stats(struct subfacet *,
397 struct dpif_flow_stats *);
398 static void subfacet_update_time(struct subfacet *, long long int used);
399 static void subfacet_update_stats(struct subfacet *,
400 const struct dpif_flow_stats *);
401 static void subfacet_make_actions(struct subfacet *,
402 const struct ofpbuf *packet,
403 struct ofpbuf *odp_actions);
404 static int subfacet_install(struct subfacet *,
405 const struct nlattr *actions, size_t actions_len,
406 struct dpif_flow_stats *, enum slow_path_reason);
407 static void subfacet_uninstall(struct subfacet *);
409 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
411 /* An exact-match instantiation of an OpenFlow flow.
413 * A facet associates a "struct flow", which represents the Open vSwitch
414 * userspace idea of an exact-match flow, with one or more subfacets. Each
415 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
416 * the facet. When the kernel module (or other dpif implementation) and Open
417 * vSwitch userspace agree on the definition of a flow key, there is exactly
418 * one subfacet per facet. If the dpif implementation supports more-specific
419 * flow matching than userspace, however, a facet can have more than one
420 * subfacet, each of which corresponds to some distinction in flow that
421 * userspace simply doesn't understand.
423 * Flow expiration works in terms of subfacets, so a facet must have at least
424 * one subfacet or it will never expire, leaking memory. */
427 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
428 struct list list_node; /* In owning rule's 'facets' list. */
429 struct rule_dpif *rule; /* Owning rule. */
432 struct list subfacets;
433 long long int used; /* Time last used; time created if not used. */
440 * - Do include packets and bytes sent "by hand", e.g. with
443 * - Do include packets and bytes that were obtained from the datapath
444 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
445 * DPIF_FP_ZERO_STATS).
447 * - Do not include packets or bytes that can be obtained from the
448 * datapath for any existing subfacet.
450 uint64_t packet_count; /* Number of packets received. */
451 uint64_t byte_count; /* Number of bytes received. */
453 /* Resubmit statistics. */
454 uint64_t prev_packet_count; /* Number of packets from last stats push. */
455 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
456 long long int prev_used; /* Used time from last stats push. */
459 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
460 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
461 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
463 /* Properties of datapath actions.
465 * Every subfacet has its own actions because actions can differ slightly
466 * between splintered and non-splintered subfacets due to the VLAN tag
467 * being initially different (present vs. absent). All of them have these
468 * properties in common so we just store one copy of them here. */
469 bool has_learn; /* Actions include NXAST_LEARN? */
470 bool has_normal; /* Actions output to OFPP_NORMAL? */
471 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
472 tag_type tags; /* Tags that would require revalidation. */
473 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
475 /* Storage for a single subfacet, to reduce malloc() time and space
476 * overhead. (A facet always has at least one subfacet and in the common
477 * case has exactly one subfacet.) */
478 struct subfacet one_subfacet;
481 static struct facet *facet_create(struct rule_dpif *,
482 const struct flow *, uint32_t hash);
483 static void facet_remove(struct facet *);
484 static void facet_free(struct facet *);
486 static struct facet *facet_find(struct ofproto_dpif *,
487 const struct flow *, uint32_t hash);
488 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
489 const struct flow *, uint32_t hash);
490 static void facet_revalidate(struct facet *);
491 static bool facet_check_consistency(struct facet *);
493 static void facet_flush_stats(struct facet *);
495 static void facet_update_time(struct facet *, long long int used);
496 static void facet_reset_counters(struct facet *);
497 static void facet_push_stats(struct facet *);
498 static void facet_learn(struct facet *);
499 static void facet_account(struct facet *);
501 static bool facet_is_controller_flow(struct facet *);
504 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
508 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
509 struct list bundle_node; /* In struct ofbundle's "ports" list. */
510 struct cfm *cfm; /* Connectivity Fault Management, if any. */
511 tag_type tag; /* Tag associated with this port. */
512 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
513 bool may_enable; /* May be enabled in bonds. */
514 long long int carrier_seq; /* Carrier status changes. */
517 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
518 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
519 long long int stp_state_entered;
521 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
523 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
525 * This is deprecated. It is only for compatibility with broken device
526 * drivers in old versions of Linux that do not properly support VLANs when
527 * VLAN devices are not used. When broken device drivers are no longer in
528 * widespread use, we will delete these interfaces. */
529 uint16_t realdev_ofp_port;
533 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
534 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
535 * traffic egressing the 'ofport' with that priority should be marked with. */
536 struct priority_to_dscp {
537 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
538 uint32_t priority; /* Priority of this queue (see struct flow). */
540 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
543 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
545 * This is deprecated. It is only for compatibility with broken device drivers
546 * in old versions of Linux that do not properly support VLANs when VLAN
547 * devices are not used. When broken device drivers are no longer in
548 * widespread use, we will delete these interfaces. */
549 struct vlan_splinter {
550 struct hmap_node realdev_vid_node;
551 struct hmap_node vlandev_node;
552 uint16_t realdev_ofp_port;
553 uint16_t vlandev_ofp_port;
557 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
558 uint32_t realdev, ovs_be16 vlan_tci);
559 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
560 static void vsp_remove(struct ofport_dpif *);
561 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
563 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
565 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
568 static struct ofport_dpif *
569 ofport_dpif_cast(const struct ofport *ofport)
571 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
572 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
575 static void port_run(struct ofport_dpif *);
576 static void port_run_fast(struct ofport_dpif *);
577 static void port_wait(struct ofport_dpif *);
578 static int set_cfm(struct ofport *, const struct cfm_settings *);
579 static void ofport_clear_priorities(struct ofport_dpif *);
581 struct dpif_completion {
582 struct list list_node;
583 struct ofoperation *op;
586 /* Extra information about a classifier table.
587 * Currently used just for optimized flow revalidation. */
589 /* If either of these is nonnull, then this table has a form that allows
590 * flows to be tagged to avoid revalidating most flows for the most common
591 * kinds of flow table changes. */
592 struct cls_table *catchall_table; /* Table that wildcards all fields. */
593 struct cls_table *other_table; /* Table with any other wildcard set. */
594 uint32_t basis; /* Keeps each table's tags separate. */
597 /* Reasons that we might need to revalidate every facet, and corresponding
600 * A value of 0 means that there is no need to revalidate.
602 * It would be nice to have some cleaner way to integrate with coverage
603 * counters, but with only a few reasons I guess this is good enough for
605 enum revalidate_reason {
606 REV_RECONFIGURE = 1, /* Switch configuration changed. */
607 REV_STP, /* Spanning tree protocol port status change. */
608 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
609 REV_FLOW_TABLE, /* Flow table changed. */
610 REV_INCONSISTENCY /* Facet self-check failed. */
612 COVERAGE_DEFINE(rev_reconfigure);
613 COVERAGE_DEFINE(rev_stp);
614 COVERAGE_DEFINE(rev_port_toggled);
615 COVERAGE_DEFINE(rev_flow_table);
616 COVERAGE_DEFINE(rev_inconsistency);
618 /* All datapaths of a given type share a single dpif backer instance. */
623 struct timer next_expiration;
624 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
626 /* Facet revalidation flags applying to facets which use this backer. */
627 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
628 struct tag_set revalidate_set; /* Revalidate only matching facets. */
631 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
632 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
634 static struct ofport_dpif *
635 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
637 struct ofproto_dpif {
638 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
640 struct dpif_backer *backer;
642 /* Special OpenFlow rules. */
643 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
644 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
650 struct netflow *netflow;
651 struct dpif_sflow *sflow;
652 struct hmap bundles; /* Contains "struct ofbundle"s. */
653 struct mac_learning *ml;
654 struct ofmirror *mirrors[MAX_MIRRORS];
656 bool has_bonded_bundles;
660 struct hmap subfacets;
661 struct governor *governor;
664 struct table_dpif tables[N_TABLES];
666 /* Support for debugging async flow mods. */
667 struct list completions;
669 bool has_bundle_action; /* True when the first bundle action appears. */
670 struct netdev_stats stats; /* To account packets generated and consumed in
675 long long int stp_last_tick;
677 /* VLAN splinters. */
678 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
679 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
682 struct sset ports; /* Set of standard port names. */
683 struct sset ghost_ports; /* Ports with no datapath port. */
684 struct sset port_poll_set; /* Queued names for port_poll() reply. */
685 int port_poll_errno; /* Last errno for port_poll() reply. */
688 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
689 * for debugging the asynchronous flow_mod implementation.) */
692 /* All existing ofproto_dpif instances, indexed by ->up.name. */
693 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
695 static void ofproto_dpif_unixctl_init(void);
697 static struct ofproto_dpif *
698 ofproto_dpif_cast(const struct ofproto *ofproto)
700 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
701 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
704 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
706 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
708 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
709 const struct ofpbuf *, ovs_be16 initial_tci,
712 /* Packet processing. */
713 static void update_learning_table(struct ofproto_dpif *,
714 const struct flow *, int vlan,
717 #define FLOW_MISS_MAX_BATCH 50
718 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
720 /* Flow expiration. */
721 static int expire(struct dpif_backer *);
724 static void send_netflow_active_timeouts(struct ofproto_dpif *);
727 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
728 static size_t compose_sflow_action(const struct ofproto_dpif *,
729 struct ofpbuf *odp_actions,
730 const struct flow *, uint32_t odp_port);
731 static void add_mirror_actions(struct action_xlate_ctx *ctx,
732 const struct flow *flow);
733 /* Global variables. */
734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
736 /* Initial mappings of port to bridge mappings. */
737 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
739 /* Factory functions. */
742 init(const struct shash *iface_hints)
744 struct shash_node *node;
746 /* Make a local copy, since we don't own 'iface_hints' elements. */
747 SHASH_FOR_EACH(node, iface_hints) {
748 const struct iface_hint *orig_hint = node->data;
749 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
751 new_hint->br_name = xstrdup(orig_hint->br_name);
752 new_hint->br_type = xstrdup(orig_hint->br_type);
753 new_hint->ofp_port = orig_hint->ofp_port;
755 shash_add(&init_ofp_ports, node->name, new_hint);
760 enumerate_types(struct sset *types)
762 dp_enumerate_types(types);
766 enumerate_names(const char *type, struct sset *names)
768 struct ofproto_dpif *ofproto;
771 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
772 if (strcmp(type, ofproto->up.type)) {
775 sset_add(names, ofproto->up.name);
782 del(const char *type, const char *name)
787 error = dpif_open(name, type, &dpif);
789 error = dpif_delete(dpif);
796 port_open_type(const char *datapath_type, const char *port_type)
798 return dpif_port_open_type(datapath_type, port_type);
801 /* Type functions. */
803 static struct ofproto_dpif *
804 lookup_ofproto_dpif_by_port_name(const char *name)
806 struct ofproto_dpif *ofproto;
808 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
809 if (sset_contains(&ofproto->ports, name)) {
818 type_run(const char *type)
820 struct dpif_backer *backer;
824 backer = shash_find_data(&all_dpif_backers, type);
826 /* This is not necessarily a problem, since backers are only
827 * created on demand. */
831 dpif_run(backer->dpif);
833 if (backer->need_revalidate
834 || !tag_set_is_empty(&backer->revalidate_set)) {
835 struct tag_set revalidate_set = backer->revalidate_set;
836 bool need_revalidate = backer->need_revalidate;
837 struct ofproto_dpif *ofproto;
839 switch (backer->need_revalidate) {
840 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
841 case REV_STP: COVERAGE_INC(rev_stp); break;
842 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
843 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
844 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
847 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
850 if (ofproto->backer != backer) {
854 /* Clear the revalidation flags. */
855 tag_set_init(&backer->revalidate_set);
856 backer->need_revalidate = 0;
858 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
860 || tag_set_intersects(&revalidate_set, facet->tags)) {
861 facet_revalidate(facet);
868 if (timer_expired(&backer->next_expiration)) {
869 int delay = expire(backer);
870 timer_set_duration(&backer->next_expiration, delay);
873 /* Check for port changes in the dpif. */
874 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
875 struct ofproto_dpif *ofproto;
876 struct dpif_port port;
878 /* Don't report on the datapath's device. */
879 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
883 ofproto = lookup_ofproto_dpif_by_port_name(devname);
884 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
885 /* The port was removed. If we know the datapath,
886 * report it through poll_set(). If we don't, it may be
887 * notifying us of a removal we initiated, so ignore it.
888 * If there's a pending ENOBUFS, let it stand, since
889 * everything will be reevaluated. */
890 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
891 sset_add(&ofproto->port_poll_set, devname);
892 ofproto->port_poll_errno = 0;
894 } else if (!ofproto) {
895 /* The port was added, but we don't know with which
896 * ofproto we should associate it. Delete it. */
897 dpif_port_del(backer->dpif, port.port_no);
899 dpif_port_destroy(&port);
905 if (error != EAGAIN) {
906 struct ofproto_dpif *ofproto;
908 /* There was some sort of error, so propagate it to all
909 * ofprotos that use this backer. */
910 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
911 &all_ofproto_dpifs) {
912 if (ofproto->backer == backer) {
913 sset_clear(&ofproto->port_poll_set);
914 ofproto->port_poll_errno = error;
923 type_run_fast(const char *type)
925 struct dpif_backer *backer;
928 backer = shash_find_data(&all_dpif_backers, type);
930 /* This is not necessarily a problem, since backers are only
931 * created on demand. */
935 /* Handle one or more batches of upcalls, until there's nothing left to do
936 * or until we do a fixed total amount of work.
938 * We do work in batches because it can be much cheaper to set up a number
939 * of flows and fire off their patches all at once. We do multiple batches
940 * because in some cases handling a packet can cause another packet to be
941 * queued almost immediately as part of the return flow. Both
942 * optimizations can make major improvements on some benchmarks and
943 * presumably for real traffic as well. */
945 while (work < FLOW_MISS_MAX_BATCH) {
946 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
957 type_wait(const char *type)
959 struct dpif_backer *backer;
961 backer = shash_find_data(&all_dpif_backers, type);
963 /* This is not necessarily a problem, since backers are only
964 * created on demand. */
968 timer_wait(&backer->next_expiration);
971 /* Basic life-cycle. */
973 static int add_internal_flows(struct ofproto_dpif *);
975 static struct ofproto *
978 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
983 dealloc(struct ofproto *ofproto_)
985 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
990 close_dpif_backer(struct dpif_backer *backer)
992 struct shash_node *node;
994 ovs_assert(backer->refcount > 0);
996 if (--backer->refcount) {
1000 hmap_destroy(&backer->odp_to_ofport_map);
1001 node = shash_find(&all_dpif_backers, backer->type);
1003 shash_delete(&all_dpif_backers, node);
1004 dpif_close(backer->dpif);
1009 /* Datapath port slated for removal from datapath. */
1010 struct odp_garbage {
1011 struct list list_node;
1016 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1018 struct dpif_backer *backer;
1019 struct dpif_port_dump port_dump;
1020 struct dpif_port port;
1021 struct shash_node *node;
1022 struct list garbage_list;
1023 struct odp_garbage *garbage, *next;
1029 backer = shash_find_data(&all_dpif_backers, type);
1036 backer_name = xasprintf("ovs-%s", type);
1038 /* Remove any existing datapaths, since we assume we're the only
1039 * userspace controlling the datapath. */
1041 dp_enumerate_names(type, &names);
1042 SSET_FOR_EACH(name, &names) {
1043 struct dpif *old_dpif;
1045 /* Don't remove our backer if it exists. */
1046 if (!strcmp(name, backer_name)) {
1050 if (dpif_open(name, type, &old_dpif)) {
1051 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1053 dpif_delete(old_dpif);
1054 dpif_close(old_dpif);
1057 sset_destroy(&names);
1059 backer = xmalloc(sizeof *backer);
1061 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1064 VLOG_ERR("failed to open datapath of type %s: %s", type,
1070 backer->type = xstrdup(type);
1071 backer->refcount = 1;
1072 hmap_init(&backer->odp_to_ofport_map);
1073 timer_set_duration(&backer->next_expiration, 1000);
1074 backer->need_revalidate = 0;
1075 tag_set_init(&backer->revalidate_set);
1078 dpif_flow_flush(backer->dpif);
1080 /* Loop through the ports already on the datapath and remove any
1081 * that we don't need anymore. */
1082 list_init(&garbage_list);
1083 dpif_port_dump_start(&port_dump, backer->dpif);
1084 while (dpif_port_dump_next(&port_dump, &port)) {
1085 node = shash_find(&init_ofp_ports, port.name);
1086 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1087 garbage = xmalloc(sizeof *garbage);
1088 garbage->odp_port = port.port_no;
1089 list_push_front(&garbage_list, &garbage->list_node);
1092 dpif_port_dump_done(&port_dump);
1094 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1095 dpif_port_del(backer->dpif, garbage->odp_port);
1096 list_remove(&garbage->list_node);
1100 shash_add(&all_dpif_backers, type, backer);
1102 error = dpif_recv_set(backer->dpif, true);
1104 VLOG_ERR("failed to listen on datapath of type %s: %s",
1105 type, strerror(error));
1106 close_dpif_backer(backer);
1114 construct(struct ofproto *ofproto_)
1116 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1117 struct shash_node *node, *next;
1122 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1127 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1128 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1130 ofproto->n_matches = 0;
1132 ofproto->netflow = NULL;
1133 ofproto->sflow = NULL;
1134 ofproto->stp = NULL;
1135 hmap_init(&ofproto->bundles);
1136 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1137 for (i = 0; i < MAX_MIRRORS; i++) {
1138 ofproto->mirrors[i] = NULL;
1140 ofproto->has_bonded_bundles = false;
1142 hmap_init(&ofproto->facets);
1143 hmap_init(&ofproto->subfacets);
1144 ofproto->governor = NULL;
1146 for (i = 0; i < N_TABLES; i++) {
1147 struct table_dpif *table = &ofproto->tables[i];
1149 table->catchall_table = NULL;
1150 table->other_table = NULL;
1151 table->basis = random_uint32();
1154 list_init(&ofproto->completions);
1156 ofproto_dpif_unixctl_init();
1158 ofproto->has_mirrors = false;
1159 ofproto->has_bundle_action = false;
1161 hmap_init(&ofproto->vlandev_map);
1162 hmap_init(&ofproto->realdev_vid_map);
1164 sset_init(&ofproto->ports);
1165 sset_init(&ofproto->ghost_ports);
1166 sset_init(&ofproto->port_poll_set);
1167 ofproto->port_poll_errno = 0;
1169 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1170 struct iface_hint *iface_hint = node->data;
1172 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1173 /* Check if the datapath already has this port. */
1174 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1175 sset_add(&ofproto->ports, node->name);
1178 free(iface_hint->br_name);
1179 free(iface_hint->br_type);
1181 shash_delete(&init_ofp_ports, node);
1185 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1186 hash_string(ofproto->up.name, 0));
1187 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1189 ofproto_init_tables(ofproto_, N_TABLES);
1190 error = add_internal_flows(ofproto);
1191 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1197 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1198 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1200 struct ofputil_flow_mod fm;
1203 match_init_catchall(&fm.match);
1205 match_set_reg(&fm.match, 0, id);
1206 fm.new_cookie = htonll(0);
1207 fm.cookie = htonll(0);
1208 fm.cookie_mask = htonll(0);
1209 fm.table_id = TBL_INTERNAL;
1210 fm.command = OFPFC_ADD;
1211 fm.idle_timeout = 0;
1212 fm.hard_timeout = 0;
1216 fm.ofpacts = ofpacts->data;
1217 fm.ofpacts_len = ofpacts->size;
1219 error = ofproto_flow_mod(&ofproto->up, &fm);
1221 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1222 id, ofperr_to_string(error));
1226 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1227 ovs_assert(*rulep != NULL);
1233 add_internal_flows(struct ofproto_dpif *ofproto)
1235 struct ofpact_controller *controller;
1236 uint64_t ofpacts_stub[128 / 8];
1237 struct ofpbuf ofpacts;
1241 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1244 controller = ofpact_put_CONTROLLER(&ofpacts);
1245 controller->max_len = UINT16_MAX;
1246 controller->controller_id = 0;
1247 controller->reason = OFPR_NO_MATCH;
1248 ofpact_pad(&ofpacts);
1250 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1255 ofpbuf_clear(&ofpacts);
1256 error = add_internal_flow(ofproto, id++, &ofpacts,
1257 &ofproto->no_packet_in_rule);
1262 complete_operations(struct ofproto_dpif *ofproto)
1264 struct dpif_completion *c, *next;
1266 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1267 ofoperation_complete(c->op, 0);
1268 list_remove(&c->list_node);
1274 destruct(struct ofproto *ofproto_)
1276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1277 struct rule_dpif *rule, *next_rule;
1278 struct oftable *table;
1281 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1282 complete_operations(ofproto);
1284 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1285 struct cls_cursor cursor;
1287 cls_cursor_init(&cursor, &table->cls, NULL);
1288 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1289 ofproto_rule_destroy(&rule->up);
1293 for (i = 0; i < MAX_MIRRORS; i++) {
1294 mirror_destroy(ofproto->mirrors[i]);
1297 netflow_destroy(ofproto->netflow);
1298 dpif_sflow_destroy(ofproto->sflow);
1299 hmap_destroy(&ofproto->bundles);
1300 mac_learning_destroy(ofproto->ml);
1302 hmap_destroy(&ofproto->facets);
1303 hmap_destroy(&ofproto->subfacets);
1304 governor_destroy(ofproto->governor);
1306 hmap_destroy(&ofproto->vlandev_map);
1307 hmap_destroy(&ofproto->realdev_vid_map);
1309 sset_destroy(&ofproto->ports);
1310 sset_destroy(&ofproto->ghost_ports);
1311 sset_destroy(&ofproto->port_poll_set);
1313 close_dpif_backer(ofproto->backer);
1317 run_fast(struct ofproto *ofproto_)
1319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1320 struct ofport_dpif *ofport;
1322 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1323 port_run_fast(ofport);
1330 run(struct ofproto *ofproto_)
1332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1333 struct ofport_dpif *ofport;
1334 struct ofbundle *bundle;
1338 complete_operations(ofproto);
1341 error = run_fast(ofproto_);
1346 if (ofproto->netflow) {
1347 if (netflow_run(ofproto->netflow)) {
1348 send_netflow_active_timeouts(ofproto);
1351 if (ofproto->sflow) {
1352 dpif_sflow_run(ofproto->sflow);
1355 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1358 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1363 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1365 /* Check the consistency of a random facet, to aid debugging. */
1366 if (!hmap_is_empty(&ofproto->facets)
1367 && !ofproto->backer->need_revalidate) {
1368 struct facet *facet;
1370 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1371 struct facet, hmap_node);
1372 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1374 if (!facet_check_consistency(facet)) {
1375 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1380 if (ofproto->governor) {
1383 governor_run(ofproto->governor);
1385 /* If the governor has shrunk to its minimum size and the number of
1386 * subfacets has dwindled, then drop the governor entirely.
1388 * For hysteresis, the number of subfacets to drop the governor is
1389 * smaller than the number needed to trigger its creation. */
1390 n_subfacets = hmap_count(&ofproto->subfacets);
1391 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1392 && governor_is_idle(ofproto->governor)) {
1393 governor_destroy(ofproto->governor);
1394 ofproto->governor = NULL;
1402 wait(struct ofproto *ofproto_)
1404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1405 struct ofport_dpif *ofport;
1406 struct ofbundle *bundle;
1408 if (!clogged && !list_is_empty(&ofproto->completions)) {
1409 poll_immediate_wake();
1412 dpif_wait(ofproto->backer->dpif);
1413 dpif_recv_wait(ofproto->backer->dpif);
1414 if (ofproto->sflow) {
1415 dpif_sflow_wait(ofproto->sflow);
1417 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1418 poll_immediate_wake();
1420 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1423 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1424 bundle_wait(bundle);
1426 if (ofproto->netflow) {
1427 netflow_wait(ofproto->netflow);
1429 mac_learning_wait(ofproto->ml);
1431 if (ofproto->backer->need_revalidate) {
1432 /* Shouldn't happen, but if it does just go around again. */
1433 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1434 poll_immediate_wake();
1436 if (ofproto->governor) {
1437 governor_wait(ofproto->governor);
1442 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1444 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1446 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1447 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1451 flush(struct ofproto *ofproto_)
1453 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1454 struct subfacet *subfacet, *next_subfacet;
1455 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1459 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1460 &ofproto->subfacets) {
1461 if (subfacet->path != SF_NOT_INSTALLED) {
1462 batch[n_batch++] = subfacet;
1463 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1464 subfacet_destroy_batch(ofproto, batch, n_batch);
1468 subfacet_destroy(subfacet);
1473 subfacet_destroy_batch(ofproto, batch, n_batch);
1478 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1479 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1481 *arp_match_ip = true;
1482 *actions = (OFPUTIL_A_OUTPUT |
1483 OFPUTIL_A_SET_VLAN_VID |
1484 OFPUTIL_A_SET_VLAN_PCP |
1485 OFPUTIL_A_STRIP_VLAN |
1486 OFPUTIL_A_SET_DL_SRC |
1487 OFPUTIL_A_SET_DL_DST |
1488 OFPUTIL_A_SET_NW_SRC |
1489 OFPUTIL_A_SET_NW_DST |
1490 OFPUTIL_A_SET_NW_TOS |
1491 OFPUTIL_A_SET_TP_SRC |
1492 OFPUTIL_A_SET_TP_DST |
1497 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 struct dpif_dp_stats s;
1502 strcpy(ots->name, "classifier");
1504 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1506 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1507 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1510 static struct ofport *
1513 struct ofport_dpif *port = xmalloc(sizeof *port);
1518 port_dealloc(struct ofport *port_)
1520 struct ofport_dpif *port = ofport_dpif_cast(port_);
1525 port_construct(struct ofport *port_)
1527 struct ofport_dpif *port = ofport_dpif_cast(port_);
1528 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1529 struct dpif_port dpif_port;
1532 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1533 port->bundle = NULL;
1535 port->tag = tag_create_random();
1536 port->may_enable = true;
1537 port->stp_port = NULL;
1538 port->stp_state = STP_DISABLED;
1539 hmap_init(&port->priorities);
1540 port->realdev_ofp_port = 0;
1541 port->vlandev_vid = 0;
1542 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1544 if (netdev_vport_is_patch(port->up.netdev)) {
1545 /* XXX By bailing out here, we don't do required sFlow work. */
1546 port->odp_port = OVSP_NONE;
1550 error = dpif_port_query_by_name(ofproto->backer->dpif,
1551 netdev_get_name(port->up.netdev),
1557 port->odp_port = dpif_port.port_no;
1559 /* Sanity-check that a mapping doesn't already exist. This
1560 * shouldn't happen. */
1561 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1562 VLOG_ERR("port %s already has an OpenFlow port number\n",
1567 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1568 hash_int(port->odp_port, 0));
1570 if (ofproto->sflow) {
1571 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1578 port_destruct(struct ofport *port_)
1580 struct ofport_dpif *port = ofport_dpif_cast(port_);
1581 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1582 const char *devname = netdev_get_name(port->up.netdev);
1584 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1585 /* The underlying device is still there, so delete it. This
1586 * happens when the ofproto is being destroyed, since the caller
1587 * assumes that removal of attached ports will happen as part of
1589 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1592 if (port->odp_port != OVSP_NONE) {
1593 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1596 sset_find_and_delete(&ofproto->ports, devname);
1597 sset_find_and_delete(&ofproto->ghost_ports, devname);
1598 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1599 bundle_remove(port_);
1600 set_cfm(port_, NULL);
1601 if (ofproto->sflow) {
1602 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1605 ofport_clear_priorities(port);
1606 hmap_destroy(&port->priorities);
1610 port_modified(struct ofport *port_)
1612 struct ofport_dpif *port = ofport_dpif_cast(port_);
1614 if (port->bundle && port->bundle->bond) {
1615 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1620 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1622 struct ofport_dpif *port = ofport_dpif_cast(port_);
1623 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1624 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1626 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1627 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1628 OFPUTIL_PC_NO_PACKET_IN)) {
1629 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1631 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1632 bundle_update(port->bundle);
1638 set_sflow(struct ofproto *ofproto_,
1639 const struct ofproto_sflow_options *sflow_options)
1641 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1642 struct dpif_sflow *ds = ofproto->sflow;
1644 if (sflow_options) {
1646 struct ofport_dpif *ofport;
1648 ds = ofproto->sflow = dpif_sflow_create();
1649 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1650 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1652 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1654 dpif_sflow_set_options(ds, sflow_options);
1657 dpif_sflow_destroy(ds);
1658 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1659 ofproto->sflow = NULL;
1666 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1668 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1675 struct ofproto_dpif *ofproto;
1677 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1678 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1679 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1682 if (cfm_configure(ofport->cfm, s)) {
1688 cfm_destroy(ofport->cfm);
1694 get_cfm_fault(const struct ofport *ofport_)
1696 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1698 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1702 get_cfm_opup(const struct ofport *ofport_)
1704 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1706 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1710 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1713 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1716 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1724 get_cfm_health(const struct ofport *ofport_)
1726 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1728 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1731 /* Spanning Tree. */
1734 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1736 struct ofproto_dpif *ofproto = ofproto_;
1737 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1738 struct ofport_dpif *ofport;
1740 ofport = stp_port_get_aux(sp);
1742 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1743 ofproto->up.name, port_num);
1745 struct eth_header *eth = pkt->l2;
1747 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1748 if (eth_addr_is_zero(eth->eth_src)) {
1749 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1750 "with unknown MAC", ofproto->up.name, port_num);
1752 send_packet(ofport, pkt);
1758 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1760 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1764 /* Only revalidate flows if the configuration changed. */
1765 if (!s != !ofproto->stp) {
1766 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1770 if (!ofproto->stp) {
1771 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1772 send_bpdu_cb, ofproto);
1773 ofproto->stp_last_tick = time_msec();
1776 stp_set_bridge_id(ofproto->stp, s->system_id);
1777 stp_set_bridge_priority(ofproto->stp, s->priority);
1778 stp_set_hello_time(ofproto->stp, s->hello_time);
1779 stp_set_max_age(ofproto->stp, s->max_age);
1780 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1782 struct ofport *ofport;
1784 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1785 set_stp_port(ofport, NULL);
1788 stp_destroy(ofproto->stp);
1789 ofproto->stp = NULL;
1796 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1802 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1803 s->designated_root = stp_get_designated_root(ofproto->stp);
1804 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1813 update_stp_port_state(struct ofport_dpif *ofport)
1815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1816 enum stp_state state;
1818 /* Figure out new state. */
1819 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1823 if (ofport->stp_state != state) {
1824 enum ofputil_port_state of_state;
1827 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1828 netdev_get_name(ofport->up.netdev),
1829 stp_state_name(ofport->stp_state),
1830 stp_state_name(state));
1831 if (stp_learn_in_state(ofport->stp_state)
1832 != stp_learn_in_state(state)) {
1833 /* xxx Learning action flows should also be flushed. */
1834 mac_learning_flush(ofproto->ml,
1835 &ofproto->backer->revalidate_set);
1837 fwd_change = stp_forward_in_state(ofport->stp_state)
1838 != stp_forward_in_state(state);
1840 ofproto->backer->need_revalidate = REV_STP;
1841 ofport->stp_state = state;
1842 ofport->stp_state_entered = time_msec();
1844 if (fwd_change && ofport->bundle) {
1845 bundle_update(ofport->bundle);
1848 /* Update the STP state bits in the OpenFlow port description. */
1849 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1850 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1851 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1852 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1853 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1855 ofproto_port_set_state(&ofport->up, of_state);
1859 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1860 * caller is responsible for assigning STP port numbers and ensuring
1861 * there are no duplicates. */
1863 set_stp_port(struct ofport *ofport_,
1864 const struct ofproto_port_stp_settings *s)
1866 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1868 struct stp_port *sp = ofport->stp_port;
1870 if (!s || !s->enable) {
1872 ofport->stp_port = NULL;
1873 stp_port_disable(sp);
1874 update_stp_port_state(ofport);
1877 } else if (sp && stp_port_no(sp) != s->port_num
1878 && ofport == stp_port_get_aux(sp)) {
1879 /* The port-id changed, so disable the old one if it's not
1880 * already in use by another port. */
1881 stp_port_disable(sp);
1884 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1885 stp_port_enable(sp);
1887 stp_port_set_aux(sp, ofport);
1888 stp_port_set_priority(sp, s->priority);
1889 stp_port_set_path_cost(sp, s->path_cost);
1891 update_stp_port_state(ofport);
1897 get_stp_port_status(struct ofport *ofport_,
1898 struct ofproto_port_stp_status *s)
1900 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1902 struct stp_port *sp = ofport->stp_port;
1904 if (!ofproto->stp || !sp) {
1910 s->port_id = stp_port_get_id(sp);
1911 s->state = stp_port_get_state(sp);
1912 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1913 s->role = stp_port_get_role(sp);
1914 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1920 stp_run(struct ofproto_dpif *ofproto)
1923 long long int now = time_msec();
1924 long long int elapsed = now - ofproto->stp_last_tick;
1925 struct stp_port *sp;
1928 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1929 ofproto->stp_last_tick = now;
1931 while (stp_get_changed_port(ofproto->stp, &sp)) {
1932 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1935 update_stp_port_state(ofport);
1939 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1940 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
1946 stp_wait(struct ofproto_dpif *ofproto)
1949 poll_timer_wait(1000);
1953 /* Returns true if STP should process 'flow'. */
1955 stp_should_process_flow(const struct flow *flow)
1957 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1961 stp_process_packet(const struct ofport_dpif *ofport,
1962 const struct ofpbuf *packet)
1964 struct ofpbuf payload = *packet;
1965 struct eth_header *eth = payload.data;
1966 struct stp_port *sp = ofport->stp_port;
1968 /* Sink packets on ports that have STP disabled when the bridge has
1970 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1974 /* Trim off padding on payload. */
1975 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1976 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1979 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1980 stp_received_bpdu(sp, payload.data, payload.size);
1984 static struct priority_to_dscp *
1985 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1987 struct priority_to_dscp *pdscp;
1990 hash = hash_int(priority, 0);
1991 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1992 if (pdscp->priority == priority) {
2000 ofport_clear_priorities(struct ofport_dpif *ofport)
2002 struct priority_to_dscp *pdscp, *next;
2004 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2005 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2011 set_queues(struct ofport *ofport_,
2012 const struct ofproto_port_queue *qdscp_list,
2015 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2017 struct hmap new = HMAP_INITIALIZER(&new);
2020 for (i = 0; i < n_qdscp; i++) {
2021 struct priority_to_dscp *pdscp;
2025 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2026 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2031 pdscp = get_priority(ofport, priority);
2033 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2035 pdscp = xmalloc(sizeof *pdscp);
2036 pdscp->priority = priority;
2038 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2041 if (pdscp->dscp != dscp) {
2043 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2046 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2049 if (!hmap_is_empty(&ofport->priorities)) {
2050 ofport_clear_priorities(ofport);
2051 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2054 hmap_swap(&new, &ofport->priorities);
2062 /* Expires all MAC learning entries associated with 'bundle' and forces its
2063 * ofproto to revalidate every flow.
2065 * Normally MAC learning entries are removed only from the ofproto associated
2066 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2067 * are removed from every ofproto. When patch ports and SLB bonds are in use
2068 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2069 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2070 * with the host from which it migrated. */
2072 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2074 struct ofproto_dpif *ofproto = bundle->ofproto;
2075 struct mac_learning *ml = ofproto->ml;
2076 struct mac_entry *mac, *next_mac;
2078 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2079 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2080 if (mac->port.p == bundle) {
2082 struct ofproto_dpif *o;
2084 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2086 struct mac_entry *e;
2088 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2091 mac_learning_expire(o->ml, e);
2097 mac_learning_expire(ml, mac);
2102 static struct ofbundle *
2103 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2105 struct ofbundle *bundle;
2107 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2108 &ofproto->bundles) {
2109 if (bundle->aux == aux) {
2116 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2117 * ones that are found to 'bundles'. */
2119 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2120 void **auxes, size_t n_auxes,
2121 struct hmapx *bundles)
2125 hmapx_init(bundles);
2126 for (i = 0; i < n_auxes; i++) {
2127 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2129 hmapx_add(bundles, bundle);
2135 bundle_update(struct ofbundle *bundle)
2137 struct ofport_dpif *port;
2139 bundle->floodable = true;
2140 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2141 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2142 || !stp_forward_in_state(port->stp_state)) {
2143 bundle->floodable = false;
2150 bundle_del_port(struct ofport_dpif *port)
2152 struct ofbundle *bundle = port->bundle;
2154 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2156 list_remove(&port->bundle_node);
2157 port->bundle = NULL;
2160 lacp_slave_unregister(bundle->lacp, port);
2163 bond_slave_unregister(bundle->bond, port);
2166 bundle_update(bundle);
2170 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2171 struct lacp_slave_settings *lacp,
2172 uint32_t bond_stable_id)
2174 struct ofport_dpif *port;
2176 port = get_ofp_port(bundle->ofproto, ofp_port);
2181 if (port->bundle != bundle) {
2182 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2184 bundle_del_port(port);
2187 port->bundle = bundle;
2188 list_push_back(&bundle->ports, &port->bundle_node);
2189 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2190 || !stp_forward_in_state(port->stp_state)) {
2191 bundle->floodable = false;
2195 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2196 lacp_slave_register(bundle->lacp, port, lacp);
2199 port->bond_stable_id = bond_stable_id;
2205 bundle_destroy(struct ofbundle *bundle)
2207 struct ofproto_dpif *ofproto;
2208 struct ofport_dpif *port, *next_port;
2215 ofproto = bundle->ofproto;
2216 for (i = 0; i < MAX_MIRRORS; i++) {
2217 struct ofmirror *m = ofproto->mirrors[i];
2219 if (m->out == bundle) {
2221 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2222 || hmapx_find_and_delete(&m->dsts, bundle)) {
2223 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2228 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2229 bundle_del_port(port);
2232 bundle_flush_macs(bundle, true);
2233 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2235 free(bundle->trunks);
2236 lacp_destroy(bundle->lacp);
2237 bond_destroy(bundle->bond);
2242 bundle_set(struct ofproto *ofproto_, void *aux,
2243 const struct ofproto_bundle_settings *s)
2245 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2246 bool need_flush = false;
2247 struct ofport_dpif *port;
2248 struct ofbundle *bundle;
2249 unsigned long *trunks;
2255 bundle_destroy(bundle_lookup(ofproto, aux));
2259 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2260 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2262 bundle = bundle_lookup(ofproto, aux);
2264 bundle = xmalloc(sizeof *bundle);
2266 bundle->ofproto = ofproto;
2267 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2268 hash_pointer(aux, 0));
2270 bundle->name = NULL;
2272 list_init(&bundle->ports);
2273 bundle->vlan_mode = PORT_VLAN_TRUNK;
2275 bundle->trunks = NULL;
2276 bundle->use_priority_tags = s->use_priority_tags;
2277 bundle->lacp = NULL;
2278 bundle->bond = NULL;
2280 bundle->floodable = true;
2282 bundle->src_mirrors = 0;
2283 bundle->dst_mirrors = 0;
2284 bundle->mirror_out = 0;
2287 if (!bundle->name || strcmp(s->name, bundle->name)) {
2289 bundle->name = xstrdup(s->name);
2294 if (!bundle->lacp) {
2295 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2296 bundle->lacp = lacp_create();
2298 lacp_configure(bundle->lacp, s->lacp);
2300 lacp_destroy(bundle->lacp);
2301 bundle->lacp = NULL;
2304 /* Update set of ports. */
2306 for (i = 0; i < s->n_slaves; i++) {
2307 if (!bundle_add_port(bundle, s->slaves[i],
2308 s->lacp ? &s->lacp_slaves[i] : NULL,
2309 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2313 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2314 struct ofport_dpif *next_port;
2316 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2317 for (i = 0; i < s->n_slaves; i++) {
2318 if (s->slaves[i] == port->up.ofp_port) {
2323 bundle_del_port(port);
2327 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2329 if (list_is_empty(&bundle->ports)) {
2330 bundle_destroy(bundle);
2334 /* Set VLAN tagging mode */
2335 if (s->vlan_mode != bundle->vlan_mode
2336 || s->use_priority_tags != bundle->use_priority_tags) {
2337 bundle->vlan_mode = s->vlan_mode;
2338 bundle->use_priority_tags = s->use_priority_tags;
2343 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2344 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2346 if (vlan != bundle->vlan) {
2347 bundle->vlan = vlan;
2351 /* Get trunked VLANs. */
2352 switch (s->vlan_mode) {
2353 case PORT_VLAN_ACCESS:
2357 case PORT_VLAN_TRUNK:
2358 trunks = CONST_CAST(unsigned long *, s->trunks);
2361 case PORT_VLAN_NATIVE_UNTAGGED:
2362 case PORT_VLAN_NATIVE_TAGGED:
2363 if (vlan != 0 && (!s->trunks
2364 || !bitmap_is_set(s->trunks, vlan)
2365 || bitmap_is_set(s->trunks, 0))) {
2366 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2368 trunks = bitmap_clone(s->trunks, 4096);
2370 trunks = bitmap_allocate1(4096);
2372 bitmap_set1(trunks, vlan);
2373 bitmap_set0(trunks, 0);
2375 trunks = CONST_CAST(unsigned long *, s->trunks);
2382 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2383 free(bundle->trunks);
2384 if (trunks == s->trunks) {
2385 bundle->trunks = vlan_bitmap_clone(trunks);
2387 bundle->trunks = trunks;
2392 if (trunks != s->trunks) {
2397 if (!list_is_short(&bundle->ports)) {
2398 bundle->ofproto->has_bonded_bundles = true;
2400 if (bond_reconfigure(bundle->bond, s->bond)) {
2401 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2404 bundle->bond = bond_create(s->bond);
2405 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2408 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2409 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2413 bond_destroy(bundle->bond);
2414 bundle->bond = NULL;
2417 /* If we changed something that would affect MAC learning, un-learn
2418 * everything on this port and force flow revalidation. */
2420 bundle_flush_macs(bundle, false);
2427 bundle_remove(struct ofport *port_)
2429 struct ofport_dpif *port = ofport_dpif_cast(port_);
2430 struct ofbundle *bundle = port->bundle;
2433 bundle_del_port(port);
2434 if (list_is_empty(&bundle->ports)) {
2435 bundle_destroy(bundle);
2436 } else if (list_is_short(&bundle->ports)) {
2437 bond_destroy(bundle->bond);
2438 bundle->bond = NULL;
2444 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2446 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2447 struct ofport_dpif *port = port_;
2448 uint8_t ea[ETH_ADDR_LEN];
2451 error = netdev_get_etheraddr(port->up.netdev, ea);
2453 struct ofpbuf packet;
2456 ofpbuf_init(&packet, 0);
2457 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2459 memcpy(packet_pdu, pdu, pdu_size);
2461 send_packet(port, &packet);
2462 ofpbuf_uninit(&packet);
2464 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2465 "%s (%s)", port->bundle->name,
2466 netdev_get_name(port->up.netdev), strerror(error));
2471 bundle_send_learning_packets(struct ofbundle *bundle)
2473 struct ofproto_dpif *ofproto = bundle->ofproto;
2474 int error, n_packets, n_errors;
2475 struct mac_entry *e;
2477 error = n_packets = n_errors = 0;
2478 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2479 if (e->port.p != bundle) {
2480 struct ofpbuf *learning_packet;
2481 struct ofport_dpif *port;
2485 /* The assignment to "port" is unnecessary but makes "grep"ing for
2486 * struct ofport_dpif more effective. */
2487 learning_packet = bond_compose_learning_packet(bundle->bond,
2491 ret = send_packet(port, learning_packet);
2492 ofpbuf_delete(learning_packet);
2502 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2503 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2504 "packets, last error was: %s",
2505 bundle->name, n_errors, n_packets, strerror(error));
2507 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2508 bundle->name, n_packets);
2513 bundle_run(struct ofbundle *bundle)
2516 lacp_run(bundle->lacp, send_pdu_cb);
2519 struct ofport_dpif *port;
2521 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2522 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2525 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2526 lacp_status(bundle->lacp));
2527 if (bond_should_send_learning_packets(bundle->bond)) {
2528 bundle_send_learning_packets(bundle);
2534 bundle_wait(struct ofbundle *bundle)
2537 lacp_wait(bundle->lacp);
2540 bond_wait(bundle->bond);
2547 mirror_scan(struct ofproto_dpif *ofproto)
2551 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2552 if (!ofproto->mirrors[idx]) {
2559 static struct ofmirror *
2560 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2564 for (i = 0; i < MAX_MIRRORS; i++) {
2565 struct ofmirror *mirror = ofproto->mirrors[i];
2566 if (mirror && mirror->aux == aux) {
2574 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2576 mirror_update_dups(struct ofproto_dpif *ofproto)
2580 for (i = 0; i < MAX_MIRRORS; i++) {
2581 struct ofmirror *m = ofproto->mirrors[i];
2584 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2588 for (i = 0; i < MAX_MIRRORS; i++) {
2589 struct ofmirror *m1 = ofproto->mirrors[i];
2596 for (j = i + 1; j < MAX_MIRRORS; j++) {
2597 struct ofmirror *m2 = ofproto->mirrors[j];
2599 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2600 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2601 m2->dup_mirrors |= m1->dup_mirrors;
2608 mirror_set(struct ofproto *ofproto_, void *aux,
2609 const struct ofproto_mirror_settings *s)
2611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2612 mirror_mask_t mirror_bit;
2613 struct ofbundle *bundle;
2614 struct ofmirror *mirror;
2615 struct ofbundle *out;
2616 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2617 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2620 mirror = mirror_lookup(ofproto, aux);
2622 mirror_destroy(mirror);
2628 idx = mirror_scan(ofproto);
2630 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2632 ofproto->up.name, MAX_MIRRORS, s->name);
2636 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2637 mirror->ofproto = ofproto;
2640 mirror->out_vlan = -1;
2641 mirror->name = NULL;
2644 if (!mirror->name || strcmp(s->name, mirror->name)) {
2646 mirror->name = xstrdup(s->name);
2649 /* Get the new configuration. */
2650 if (s->out_bundle) {
2651 out = bundle_lookup(ofproto, s->out_bundle);
2653 mirror_destroy(mirror);
2659 out_vlan = s->out_vlan;
2661 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2662 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2664 /* If the configuration has not changed, do nothing. */
2665 if (hmapx_equals(&srcs, &mirror->srcs)
2666 && hmapx_equals(&dsts, &mirror->dsts)
2667 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2668 && mirror->out == out
2669 && mirror->out_vlan == out_vlan)
2671 hmapx_destroy(&srcs);
2672 hmapx_destroy(&dsts);
2676 hmapx_swap(&srcs, &mirror->srcs);
2677 hmapx_destroy(&srcs);
2679 hmapx_swap(&dsts, &mirror->dsts);
2680 hmapx_destroy(&dsts);
2682 free(mirror->vlans);
2683 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2686 mirror->out_vlan = out_vlan;
2688 /* Update bundles. */
2689 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2690 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2691 if (hmapx_contains(&mirror->srcs, bundle)) {
2692 bundle->src_mirrors |= mirror_bit;
2694 bundle->src_mirrors &= ~mirror_bit;
2697 if (hmapx_contains(&mirror->dsts, bundle)) {
2698 bundle->dst_mirrors |= mirror_bit;
2700 bundle->dst_mirrors &= ~mirror_bit;
2703 if (mirror->out == bundle) {
2704 bundle->mirror_out |= mirror_bit;
2706 bundle->mirror_out &= ~mirror_bit;
2710 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2711 ofproto->has_mirrors = true;
2712 mac_learning_flush(ofproto->ml,
2713 &ofproto->backer->revalidate_set);
2714 mirror_update_dups(ofproto);
2720 mirror_destroy(struct ofmirror *mirror)
2722 struct ofproto_dpif *ofproto;
2723 mirror_mask_t mirror_bit;
2724 struct ofbundle *bundle;
2731 ofproto = mirror->ofproto;
2732 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2733 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2735 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2736 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2737 bundle->src_mirrors &= ~mirror_bit;
2738 bundle->dst_mirrors &= ~mirror_bit;
2739 bundle->mirror_out &= ~mirror_bit;
2742 hmapx_destroy(&mirror->srcs);
2743 hmapx_destroy(&mirror->dsts);
2744 free(mirror->vlans);
2746 ofproto->mirrors[mirror->idx] = NULL;
2750 mirror_update_dups(ofproto);
2752 ofproto->has_mirrors = false;
2753 for (i = 0; i < MAX_MIRRORS; i++) {
2754 if (ofproto->mirrors[i]) {
2755 ofproto->has_mirrors = true;
2762 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2763 uint64_t *packets, uint64_t *bytes)
2765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2766 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2769 *packets = *bytes = UINT64_MAX;
2773 *packets = mirror->packet_count;
2774 *bytes = mirror->byte_count;
2780 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2782 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2783 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2784 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2790 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2793 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2794 return bundle && bundle->mirror_out != 0;
2798 forward_bpdu_changed(struct ofproto *ofproto_)
2800 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2805 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2809 mac_learning_set_idle_time(ofproto->ml, idle_time);
2810 mac_learning_set_max_entries(ofproto->ml, max_entries);
2815 static struct ofport_dpif *
2816 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2818 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2819 return ofport ? ofport_dpif_cast(ofport) : NULL;
2822 static struct ofport_dpif *
2823 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2825 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2826 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2830 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2831 struct ofproto_port *ofproto_port,
2832 struct dpif_port *dpif_port)
2834 ofproto_port->name = dpif_port->name;
2835 ofproto_port->type = dpif_port->type;
2836 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2839 static struct ofport_dpif *
2840 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2842 const struct ofproto_dpif *ofproto;
2845 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2850 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2851 struct ofport *ofport;
2853 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2854 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2855 return ofport_dpif_cast(ofport);
2862 port_run_fast(struct ofport_dpif *ofport)
2864 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2865 struct ofpbuf packet;
2867 ofpbuf_init(&packet, 0);
2868 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2869 send_packet(ofport, &packet);
2870 ofpbuf_uninit(&packet);
2875 port_run(struct ofport_dpif *ofport)
2877 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2878 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2879 bool enable = netdev_get_carrier(ofport->up.netdev);
2881 ofport->carrier_seq = carrier_seq;
2883 port_run_fast(ofport);
2885 int cfm_opup = cfm_get_opup(ofport->cfm);
2887 cfm_run(ofport->cfm);
2888 enable = enable && !cfm_get_fault(ofport->cfm);
2890 if (cfm_opup >= 0) {
2891 enable = enable && cfm_opup;
2895 if (ofport->bundle) {
2896 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2897 if (carrier_changed) {
2898 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2902 if (ofport->may_enable != enable) {
2903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2905 if (ofproto->has_bundle_action) {
2906 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2910 ofport->may_enable = enable;
2914 port_wait(struct ofport_dpif *ofport)
2917 cfm_wait(ofport->cfm);
2922 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2923 struct ofproto_port *ofproto_port)
2925 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2926 struct dpif_port dpif_port;
2929 if (sset_contains(&ofproto->ghost_ports, devname)) {
2930 const char *type = netdev_get_type_from_name(devname);
2932 /* We may be called before ofproto->up.port_by_name is populated with
2933 * the appropriate ofport. For this reason, we must get the name and
2934 * type from the netdev layer directly. */
2936 const struct ofport *ofport;
2938 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2939 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2940 ofproto_port->name = xstrdup(devname);
2941 ofproto_port->type = xstrdup(type);
2947 if (!sset_contains(&ofproto->ports, devname)) {
2950 error = dpif_port_query_by_name(ofproto->backer->dpif,
2951 devname, &dpif_port);
2953 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2959 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2961 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2962 uint32_t odp_port = UINT32_MAX;
2965 if (netdev_vport_is_patch(netdev)) {
2966 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2970 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2972 sset_add(&ofproto->ports, netdev_get_name(netdev));
2978 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2981 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2984 if (odp_port != OFPP_NONE) {
2985 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2988 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2990 /* The caller is going to close ofport->up.netdev. If this is a
2991 * bonded port, then the bond is using that netdev, so remove it
2992 * from the bond. The client will need to reconfigure everything
2993 * after deleting ports, so then the slave will get re-added. */
2994 bundle_remove(&ofport->up);
3001 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3003 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3006 error = netdev_get_stats(ofport->up.netdev, stats);
3008 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3011 /* ofproto->stats.tx_packets represents packets that we created
3012 * internally and sent to some port (e.g. packets sent with
3013 * send_packet()). Account for them as if they had come from
3014 * OFPP_LOCAL and got forwarded. */
3016 if (stats->rx_packets != UINT64_MAX) {
3017 stats->rx_packets += ofproto->stats.tx_packets;
3020 if (stats->rx_bytes != UINT64_MAX) {
3021 stats->rx_bytes += ofproto->stats.tx_bytes;
3024 /* ofproto->stats.rx_packets represents packets that were received on
3025 * some port and we processed internally and dropped (e.g. STP).
3026 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3028 if (stats->tx_packets != UINT64_MAX) {
3029 stats->tx_packets += ofproto->stats.rx_packets;
3032 if (stats->tx_bytes != UINT64_MAX) {
3033 stats->tx_bytes += ofproto->stats.rx_bytes;
3040 /* Account packets for LOCAL port. */
3042 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3043 size_t tx_size, size_t rx_size)
3045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3048 ofproto->stats.rx_packets++;
3049 ofproto->stats.rx_bytes += rx_size;
3052 ofproto->stats.tx_packets++;
3053 ofproto->stats.tx_bytes += tx_size;
3057 struct port_dump_state {
3064 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3066 *statep = xzalloc(sizeof(struct port_dump_state));
3071 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
3072 struct ofproto_port *port)
3074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3075 struct port_dump_state *state = state_;
3076 const struct sset *sset;
3077 struct sset_node *node;
3079 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3080 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3083 error = port_query_by_name(ofproto_, node->name, port);
3084 if (error != ENODEV) {
3089 if (!state->ghost) {
3090 state->ghost = true;
3093 return port_dump_next(ofproto_, state_, port);
3100 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3102 struct port_dump_state *state = state_;
3109 port_poll(const struct ofproto *ofproto_, char **devnamep)
3111 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3113 if (ofproto->port_poll_errno) {
3114 int error = ofproto->port_poll_errno;
3115 ofproto->port_poll_errno = 0;
3119 if (sset_is_empty(&ofproto->port_poll_set)) {
3123 *devnamep = sset_pop(&ofproto->port_poll_set);
3128 port_poll_wait(const struct ofproto *ofproto_)
3130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3131 dpif_port_poll_wait(ofproto->backer->dpif);
3135 port_is_lacp_current(const struct ofport *ofport_)
3137 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3138 return (ofport->bundle && ofport->bundle->lacp
3139 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3143 /* Upcall handling. */
3145 /* Flow miss batching.
3147 * Some dpifs implement operations faster when you hand them off in a batch.
3148 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3149 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3150 * more packets, plus possibly installing the flow in the dpif.
3152 * So far we only batch the operations that affect flow setup time the most.
3153 * It's possible to batch more than that, but the benefit might be minimal. */
3155 struct hmap_node hmap_node;
3156 struct ofproto_dpif *ofproto;
3158 enum odp_key_fitness key_fitness;
3159 const struct nlattr *key;
3161 ovs_be16 initial_tci;
3162 struct list packets;
3163 enum dpif_upcall_type upcall_type;
3164 uint32_t odp_in_port;
3167 struct flow_miss_op {
3168 struct dpif_op dpif_op;
3169 struct subfacet *subfacet; /* Subfacet */
3170 void *garbage; /* Pointer to pass to free(), NULL if none. */
3171 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3174 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3175 * OpenFlow controller as necessary according to their individual
3176 * configurations. */
3178 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3179 const struct flow *flow)
3181 struct ofputil_packet_in pin;
3183 pin.packet = packet->data;
3184 pin.packet_len = packet->size;
3185 pin.reason = OFPR_NO_MATCH;
3186 pin.controller_id = 0;
3191 pin.send_len = 0; /* not used for flow table misses */
3193 flow_get_metadata(flow, &pin.fmd);
3195 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3198 static enum slow_path_reason
3199 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3200 const struct ofpbuf *packet)
3202 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3208 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3210 cfm_process_heartbeat(ofport->cfm, packet);
3213 } else if (ofport->bundle && ofport->bundle->lacp
3214 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3216 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3219 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3221 stp_process_packet(ofport, packet);
3228 static struct flow_miss *
3229 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3231 struct flow_miss *miss;
3233 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3234 if (flow_equal(&miss->flow, flow)) {
3242 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3243 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3244 * 'miss' is associated with a subfacet the caller must also initialize the
3245 * returned op->subfacet, and if anything needs to be freed after processing
3246 * the op, the caller must initialize op->garbage also. */
3248 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3249 struct flow_miss_op *op)
3251 if (miss->flow.vlan_tci != miss->initial_tci) {
3252 /* This packet was received on a VLAN splinter port. We
3253 * added a VLAN to the packet to make the packet resemble
3254 * the flow, but the actions were composed assuming that
3255 * the packet contained no VLAN. So, we must remove the
3256 * VLAN header from the packet before trying to execute the
3258 eth_pop_vlan(packet);
3261 op->subfacet = NULL;
3263 op->dpif_op.type = DPIF_OP_EXECUTE;
3264 op->dpif_op.u.execute.key = miss->key;
3265 op->dpif_op.u.execute.key_len = miss->key_len;
3266 op->dpif_op.u.execute.packet = packet;
3269 /* Helper for handle_flow_miss_without_facet() and
3270 * handle_flow_miss_with_facet(). */
3272 handle_flow_miss_common(struct rule_dpif *rule,
3273 struct ofpbuf *packet, const struct flow *flow)
3275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3277 ofproto->n_matches++;
3279 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3281 * Extra-special case for fail-open mode.
3283 * We are in fail-open mode and the packet matched the fail-open
3284 * rule, but we are connected to a controller too. We should send
3285 * the packet up to the controller in the hope that it will try to
3286 * set up a flow and thereby allow us to exit fail-open.
3288 * See the top-level comment in fail-open.c for more information.
3290 send_packet_in_miss(ofproto, packet, flow);
3294 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3295 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3296 * installing a datapath flow. The answer is usually "yes" (a return value of
3297 * true). However, for short flows the cost of bookkeeping is much higher than
3298 * the benefits, so when the datapath holds a large number of flows we impose
3299 * some heuristics to decide which flows are likely to be worth tracking. */
3301 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3302 struct flow_miss *miss, uint32_t hash)
3304 if (!ofproto->governor) {
3307 n_subfacets = hmap_count(&ofproto->subfacets);
3308 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3312 ofproto->governor = governor_create(ofproto->up.name);
3315 return governor_should_install_flow(ofproto->governor, hash,
3316 list_size(&miss->packets));
3319 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3320 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3321 * increment '*n_ops'. */
3323 handle_flow_miss_without_facet(struct flow_miss *miss,
3324 struct rule_dpif *rule,
3325 struct flow_miss_op *ops, size_t *n_ops)
3327 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3328 long long int now = time_msec();
3329 struct action_xlate_ctx ctx;
3330 struct ofpbuf *packet;
3332 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3333 struct flow_miss_op *op = &ops[*n_ops];
3334 struct dpif_flow_stats stats;
3335 struct ofpbuf odp_actions;
3337 COVERAGE_INC(facet_suppress);
3339 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3341 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3342 rule_credit_stats(rule, &stats);
3344 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3346 ctx.resubmit_stats = &stats;
3347 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3350 if (odp_actions.size) {
3351 struct dpif_execute *execute = &op->dpif_op.u.execute;
3353 init_flow_miss_execute_op(miss, packet, op);
3354 execute->actions = odp_actions.data;
3355 execute->actions_len = odp_actions.size;
3356 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3360 ofpbuf_uninit(&odp_actions);
3365 /* Handles 'miss', which matches 'facet'. May add any required datapath
3366 * operations to 'ops', incrementing '*n_ops' for each new op.
3368 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3369 * This is really important only for new facets: if we just called time_msec()
3370 * here, then the new subfacet or its packets could look (occasionally) as
3371 * though it was used some time after the facet was used. That can make a
3372 * one-packet flow look like it has a nonzero duration, which looks odd in
3373 * e.g. NetFlow statistics. */
3375 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3377 struct flow_miss_op *ops, size_t *n_ops)
3379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3380 enum subfacet_path want_path;
3381 struct subfacet *subfacet;
3382 struct ofpbuf *packet;
3384 subfacet = subfacet_create(facet, miss, now);
3386 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3387 struct flow_miss_op *op = &ops[*n_ops];
3388 struct dpif_flow_stats stats;
3389 struct ofpbuf odp_actions;
3391 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3393 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3394 if (!subfacet->actions || subfacet->slow) {
3395 subfacet_make_actions(subfacet, packet, &odp_actions);
3398 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3399 subfacet_update_stats(subfacet, &stats);
3401 if (subfacet->actions_len) {
3402 struct dpif_execute *execute = &op->dpif_op.u.execute;
3404 init_flow_miss_execute_op(miss, packet, op);
3405 op->subfacet = subfacet;
3406 if (!subfacet->slow) {
3407 execute->actions = subfacet->actions;
3408 execute->actions_len = subfacet->actions_len;
3409 ofpbuf_uninit(&odp_actions);
3411 execute->actions = odp_actions.data;
3412 execute->actions_len = odp_actions.size;
3413 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3418 ofpbuf_uninit(&odp_actions);
3422 want_path = subfacet_want_path(subfacet->slow);
3423 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3424 struct flow_miss_op *op = &ops[(*n_ops)++];
3425 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3427 op->subfacet = subfacet;
3429 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3430 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3431 put->key = miss->key;
3432 put->key_len = miss->key_len;
3433 if (want_path == SF_FAST_PATH) {
3434 put->actions = subfacet->actions;
3435 put->actions_len = subfacet->actions_len;
3437 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3438 op->stub, sizeof op->stub,
3439 &put->actions, &put->actions_len);
3445 /* Handles flow miss 'miss'. May add any required datapath operations
3446 * to 'ops', incrementing '*n_ops' for each new op. */
3448 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3451 struct ofproto_dpif *ofproto = miss->ofproto;
3452 struct facet *facet;
3456 /* The caller must ensure that miss->hmap_node.hash contains
3457 * flow_hash(miss->flow, 0). */
3458 hash = miss->hmap_node.hash;
3460 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3462 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3464 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3465 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3469 facet = facet_create(rule, &miss->flow, hash);
3474 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3477 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3478 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3479 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3480 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3481 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3482 * 'packet' ingressed.
3484 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3485 * 'flow''s in_port to OFPP_NONE.
3487 * This function does post-processing on data returned from
3488 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3489 * of the upcall processing logic. In particular, if the extracted in_port is
3490 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3491 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3492 * a VLAN header onto 'packet' (if it is nonnull).
3494 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3495 * packet was really received, that is, the actual VLAN TCI extracted by
3496 * odp_flow_key_to_flow(). (This differs from the value returned in
3497 * flow->vlan_tci only for packets received on VLAN splinters.)
3499 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3500 * or some other positive errno if there are other problems. */
3502 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3503 const struct nlattr *key, size_t key_len,
3504 struct flow *flow, enum odp_key_fitness *fitnessp,
3505 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3506 ovs_be16 *initial_tci)
3508 const struct ofport_dpif *port;
3509 enum odp_key_fitness fitness;
3512 fitness = odp_flow_key_to_flow(key, key_len, flow);
3513 if (fitness == ODP_FIT_ERROR) {
3519 *initial_tci = flow->vlan_tci;
3523 *odp_in_port = flow->in_port;
3526 port = odp_port_to_ofport(backer, flow->in_port);
3528 flow->in_port = OFPP_NONE;
3534 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3537 flow->in_port = port->up.ofp_port;
3538 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3540 /* Make the packet resemble the flow, so that it gets sent to an
3541 * OpenFlow controller properly, so that it looks correct for
3542 * sFlow, and so that flow_extract() will get the correct vlan_tci
3543 * if it is called on 'packet'.
3545 * The allocated space inside 'packet' probably also contains
3546 * 'key', that is, both 'packet' and 'key' are probably part of a
3547 * struct dpif_upcall (see the large comment on that structure
3548 * definition), so pushing data on 'packet' is in general not a
3549 * good idea since it could overwrite 'key' or free it as a side
3550 * effect. However, it's OK in this special case because we know
3551 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3552 * will just overwrite the 4-byte "struct nlattr", which is fine
3553 * since we don't need that header anymore. */
3554 eth_push_vlan(packet, flow->vlan_tci);
3557 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3558 if (fitness == ODP_FIT_PERFECT) {
3559 fitness = ODP_FIT_TOO_MUCH;
3566 *fitnessp = fitness;
3572 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3575 struct dpif_upcall *upcall;
3576 struct flow_miss *miss;
3577 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3578 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3579 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3589 /* Construct the to-do list.
3591 * This just amounts to extracting the flow from each packet and sticking
3592 * the packets that have the same flow in the same "flow_miss" structure so
3593 * that we can process them together. */
3596 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3597 struct flow_miss *miss = &misses[n_misses];
3598 struct flow_miss *existing_miss;
3599 struct ofproto_dpif *ofproto;
3600 uint32_t odp_in_port;
3605 error = ofproto_receive(backer, upcall->packet, upcall->key,
3606 upcall->key_len, &flow, &miss->key_fitness,
3607 &ofproto, &odp_in_port, &miss->initial_tci);
3608 if (error == ENODEV) {
3609 /* Received packet on port for which we couldn't associate
3610 * an ofproto. This can happen if a port is removed while
3611 * traffic is being received. Print a rate-limited message
3612 * in case it happens frequently. */
3613 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3619 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3620 &flow.tunnel, flow.in_port, &miss->flow);
3622 /* Add other packets to a to-do list. */
3623 hash = flow_hash(&miss->flow, 0);
3624 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3625 if (!existing_miss) {
3626 hmap_insert(&todo, &miss->hmap_node, hash);
3627 miss->ofproto = ofproto;
3628 miss->key = upcall->key;
3629 miss->key_len = upcall->key_len;
3630 miss->upcall_type = upcall->type;
3631 miss->odp_in_port = odp_in_port;
3632 list_init(&miss->packets);
3636 miss = existing_miss;
3638 list_push_back(&miss->packets, &upcall->packet->list_node);
3641 /* Process each element in the to-do list, constructing the set of
3642 * operations to batch. */
3644 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3645 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3647 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3649 /* Execute batch. */
3650 for (i = 0; i < n_ops; i++) {
3651 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3653 dpif_operate(backer->dpif, dpif_ops, n_ops);
3655 /* Free memory and update facets. */
3656 for (i = 0; i < n_ops; i++) {
3657 struct flow_miss_op *op = &flow_miss_ops[i];
3659 switch (op->dpif_op.type) {
3660 case DPIF_OP_EXECUTE:
3663 case DPIF_OP_FLOW_PUT:
3664 if (!op->dpif_op.error) {
3665 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3669 case DPIF_OP_FLOW_DEL:
3675 hmap_destroy(&todo);
3678 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3679 classify_upcall(const struct dpif_upcall *upcall)
3681 union user_action_cookie cookie;
3683 /* First look at the upcall type. */
3684 switch (upcall->type) {
3685 case DPIF_UC_ACTION:
3691 case DPIF_N_UC_TYPES:
3693 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3697 /* "action" upcalls need a closer look. */
3698 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3699 switch (cookie.type) {
3700 case USER_ACTION_COOKIE_SFLOW:
3701 return SFLOW_UPCALL;
3703 case USER_ACTION_COOKIE_SLOW_PATH:
3706 case USER_ACTION_COOKIE_UNSPEC:
3708 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3714 handle_sflow_upcall(struct dpif_backer *backer,
3715 const struct dpif_upcall *upcall)
3717 struct ofproto_dpif *ofproto;
3718 union user_action_cookie cookie;
3720 uint32_t odp_in_port;
3722 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3723 &flow, NULL, &ofproto, &odp_in_port, NULL)
3724 || !ofproto->sflow) {
3728 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3729 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3730 odp_in_port, &cookie);
3734 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3736 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3737 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3738 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3743 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3746 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3747 struct dpif_upcall *upcall = &misses[n_misses];
3748 struct ofpbuf *buf = &miss_bufs[n_misses];
3751 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3752 sizeof miss_buf_stubs[n_misses]);
3753 error = dpif_recv(backer->dpif, upcall, buf);
3759 switch (classify_upcall(upcall)) {
3761 /* Handle it later. */
3766 handle_sflow_upcall(backer, upcall);
3776 /* Handle deferred MISS_UPCALL processing. */
3777 handle_miss_upcalls(backer, misses, n_misses);
3778 for (i = 0; i < n_misses; i++) {
3779 ofpbuf_uninit(&miss_bufs[i]);
3785 /* Flow expiration. */
3787 static int subfacet_max_idle(const struct ofproto_dpif *);
3788 static void update_stats(struct dpif_backer *);
3789 static void rule_expire(struct rule_dpif *);
3790 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3792 /* This function is called periodically by run(). Its job is to collect
3793 * updates for the flows that have been installed into the datapath, most
3794 * importantly when they last were used, and then use that information to
3795 * expire flows that have not been used recently.
3797 * Returns the number of milliseconds after which it should be called again. */
3799 expire(struct dpif_backer *backer)
3801 struct ofproto_dpif *ofproto;
3802 int max_idle = INT32_MAX;
3804 /* Update stats for each flow in the backer. */
3805 update_stats(backer);
3807 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3808 struct rule *rule, *next_rule;
3811 if (ofproto->backer != backer) {
3815 /* Expire subfacets that have been idle too long. */
3816 dp_max_idle = subfacet_max_idle(ofproto);
3817 expire_subfacets(ofproto, dp_max_idle);
3819 max_idle = MIN(max_idle, dp_max_idle);
3821 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3823 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3824 &ofproto->up.expirable) {
3825 rule_expire(rule_dpif_cast(rule));
3828 /* All outstanding data in existing flows has been accounted, so it's a
3829 * good time to do bond rebalancing. */
3830 if (ofproto->has_bonded_bundles) {
3831 struct ofbundle *bundle;
3833 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3835 bond_rebalance(bundle->bond, &backer->revalidate_set);
3841 return MIN(max_idle, 1000);
3844 /* Updates flow table statistics given that the datapath just reported 'stats'
3845 * as 'subfacet''s statistics. */
3847 update_subfacet_stats(struct subfacet *subfacet,
3848 const struct dpif_flow_stats *stats)
3850 struct facet *facet = subfacet->facet;
3852 if (stats->n_packets >= subfacet->dp_packet_count) {
3853 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3854 facet->packet_count += extra;
3856 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3859 if (stats->n_bytes >= subfacet->dp_byte_count) {
3860 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3862 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3865 subfacet->dp_packet_count = stats->n_packets;
3866 subfacet->dp_byte_count = stats->n_bytes;
3868 facet->tcp_flags |= stats->tcp_flags;
3870 subfacet_update_time(subfacet, stats->used);
3871 if (facet->accounted_bytes < facet->byte_count) {
3873 facet_account(facet);
3874 facet->accounted_bytes = facet->byte_count;
3876 facet_push_stats(facet);
3879 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3880 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3882 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3883 const struct nlattr *key, size_t key_len)
3885 if (!VLOG_DROP_WARN(&rl)) {
3889 odp_flow_key_format(key, key_len, &s);
3890 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3894 COVERAGE_INC(facet_unexpected);
3895 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3898 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3900 * This function also pushes statistics updates to rules which each facet
3901 * resubmits into. Generally these statistics will be accurate. However, if a
3902 * facet changes the rule it resubmits into at some time in between
3903 * update_stats() runs, it is possible that statistics accrued to the
3904 * old rule will be incorrectly attributed to the new rule. This could be
3905 * avoided by calling update_stats() whenever rules are created or
3906 * deleted. However, the performance impact of making so many calls to the
3907 * datapath do not justify the benefit of having perfectly accurate statistics.
3910 update_stats(struct dpif_backer *backer)
3912 const struct dpif_flow_stats *stats;
3913 struct dpif_flow_dump dump;
3914 const struct nlattr *key;
3917 dpif_flow_dump_start(&dump, backer->dpif);
3918 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3920 struct subfacet *subfacet;
3921 struct ofproto_dpif *ofproto;
3924 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
3929 key_hash = odp_flow_key_hash(key, key_len);
3930 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3931 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3933 update_subfacet_stats(subfacet, stats);
3937 /* Stats are updated per-packet. */
3940 case SF_NOT_INSTALLED:
3942 delete_unexpected_flow(ofproto, key, key_len);
3946 dpif_flow_dump_done(&dump);
3949 /* Calculates and returns the number of milliseconds of idle time after which
3950 * subfacets should expire from the datapath. When a subfacet expires, we fold
3951 * its statistics into its facet, and when a facet's last subfacet expires, we
3952 * fold its statistic into its rule. */
3954 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3957 * Idle time histogram.
3959 * Most of the time a switch has a relatively small number of subfacets.
3960 * When this is the case we might as well keep statistics for all of them
3961 * in userspace and to cache them in the kernel datapath for performance as
3964 * As the number of subfacets increases, the memory required to maintain
3965 * statistics about them in userspace and in the kernel becomes
3966 * significant. However, with a large number of subfacets it is likely
3967 * that only a few of them are "heavy hitters" that consume a large amount
3968 * of bandwidth. At this point, only heavy hitters are worth caching in
3969 * the kernel and maintaining in userspaces; other subfacets we can
3972 * The technique used to compute the idle time is to build a histogram with
3973 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3974 * that is installed in the kernel gets dropped in the appropriate bucket.
3975 * After the histogram has been built, we compute the cutoff so that only
3976 * the most-recently-used 1% of subfacets (but at least
3977 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3978 * the most-recently-used bucket of subfacets is kept, so actually an
3979 * arbitrary number of subfacets can be kept in any given expiration run
3980 * (though the next run will delete most of those unless they receive
3983 * This requires a second pass through the subfacets, in addition to the
3984 * pass made by update_stats(), because the former function never looks at
3985 * uninstallable subfacets.
3987 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3988 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3989 int buckets[N_BUCKETS] = { 0 };
3990 int total, subtotal, bucket;
3991 struct subfacet *subfacet;
3995 total = hmap_count(&ofproto->subfacets);
3996 if (total <= ofproto->up.flow_eviction_threshold) {
3997 return N_BUCKETS * BUCKET_WIDTH;
4000 /* Build histogram. */
4002 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4003 long long int idle = now - subfacet->used;
4004 int bucket = (idle <= 0 ? 0
4005 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4006 : (unsigned int) idle / BUCKET_WIDTH);
4010 /* Find the first bucket whose flows should be expired. */
4011 subtotal = bucket = 0;
4013 subtotal += buckets[bucket++];
4014 } while (bucket < N_BUCKETS &&
4015 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4017 if (VLOG_IS_DBG_ENABLED()) {
4021 ds_put_cstr(&s, "keep");
4022 for (i = 0; i < N_BUCKETS; i++) {
4024 ds_put_cstr(&s, ", drop");
4027 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4030 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4034 return bucket * BUCKET_WIDTH;
4038 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4040 /* Cutoff time for most flows. */
4041 long long int normal_cutoff = time_msec() - dp_max_idle;
4043 /* We really want to keep flows for special protocols around, so use a more
4044 * conservative cutoff. */
4045 long long int special_cutoff = time_msec() - 10000;
4047 struct subfacet *subfacet, *next_subfacet;
4048 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4052 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4053 &ofproto->subfacets) {
4054 long long int cutoff;
4056 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4059 if (subfacet->used < cutoff) {
4060 if (subfacet->path != SF_NOT_INSTALLED) {
4061 batch[n_batch++] = subfacet;
4062 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4063 subfacet_destroy_batch(ofproto, batch, n_batch);
4067 subfacet_destroy(subfacet);
4073 subfacet_destroy_batch(ofproto, batch, n_batch);
4077 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4078 * then delete it entirely. */
4080 rule_expire(struct rule_dpif *rule)
4082 struct facet *facet, *next_facet;
4086 if (rule->up.pending) {
4087 /* We'll have to expire it later. */
4091 /* Has 'rule' expired? */
4093 if (rule->up.hard_timeout
4094 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4095 reason = OFPRR_HARD_TIMEOUT;
4096 } else if (rule->up.idle_timeout
4097 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4098 reason = OFPRR_IDLE_TIMEOUT;
4103 COVERAGE_INC(ofproto_dpif_expired);
4105 /* Update stats. (This is a no-op if the rule expired due to an idle
4106 * timeout, because that only happens when the rule has no facets left.) */
4107 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4108 facet_remove(facet);
4111 /* Get rid of the rule. */
4112 ofproto_rule_expire(&rule->up, reason);
4117 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4119 * The caller must already have determined that no facet with an identical
4120 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4121 * the ofproto's classifier table.
4123 * 'hash' must be the return value of flow_hash(flow, 0).
4125 * The facet will initially have no subfacets. The caller should create (at
4126 * least) one subfacet with subfacet_create(). */
4127 static struct facet *
4128 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4130 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4131 struct facet *facet;
4133 facet = xzalloc(sizeof *facet);
4134 facet->used = time_msec();
4135 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4136 list_push_back(&rule->facets, &facet->list_node);
4138 facet->flow = *flow;
4139 list_init(&facet->subfacets);
4140 netflow_flow_init(&facet->nf_flow);
4141 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4147 facet_free(struct facet *facet)
4152 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4153 * 'packet', which arrived on 'in_port'. */
4155 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4156 const struct nlattr *odp_actions, size_t actions_len,
4157 struct ofpbuf *packet)
4159 struct odputil_keybuf keybuf;
4163 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4164 odp_flow_key_from_flow(&key, flow,
4165 ofp_port_to_odp_port(ofproto, flow->in_port));
4167 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4168 odp_actions, actions_len, packet);
4172 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4174 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4175 * rule's statistics, via subfacet_uninstall().
4177 * - Removes 'facet' from its rule and from ofproto->facets.
4180 facet_remove(struct facet *facet)
4182 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4183 struct subfacet *subfacet, *next_subfacet;
4185 ovs_assert(!list_is_empty(&facet->subfacets));
4187 /* First uninstall all of the subfacets to get final statistics. */
4188 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4189 subfacet_uninstall(subfacet);
4192 /* Flush the final stats to the rule.
4194 * This might require us to have at least one subfacet around so that we
4195 * can use its actions for accounting in facet_account(), which is why we
4196 * have uninstalled but not yet destroyed the subfacets. */
4197 facet_flush_stats(facet);
4199 /* Now we're really all done so destroy everything. */
4200 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4201 &facet->subfacets) {
4202 subfacet_destroy__(subfacet);
4204 hmap_remove(&ofproto->facets, &facet->hmap_node);
4205 list_remove(&facet->list_node);
4209 /* Feed information from 'facet' back into the learning table to keep it in
4210 * sync with what is actually flowing through the datapath. */
4212 facet_learn(struct facet *facet)
4214 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4215 struct action_xlate_ctx ctx;
4217 if (!facet->has_learn
4218 && !facet->has_normal
4219 && (!facet->has_fin_timeout
4220 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4224 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4225 facet->flow.vlan_tci,
4226 facet->rule, facet->tcp_flags, NULL);
4227 ctx.may_learn = true;
4228 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4229 facet->rule->up.ofpacts_len);
4233 facet_account(struct facet *facet)
4235 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4236 struct subfacet *subfacet;
4237 const struct nlattr *a;
4242 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4245 n_bytes = facet->byte_count - facet->accounted_bytes;
4247 /* This loop feeds byte counters to bond_account() for rebalancing to use
4248 * as a basis. We also need to track the actual VLAN on which the packet
4249 * is going to be sent to ensure that it matches the one passed to
4250 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4253 * We use the actions from an arbitrary subfacet because they should all
4254 * be equally valid for our purpose. */
4255 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4256 struct subfacet, list_node);
4257 vlan_tci = facet->flow.vlan_tci;
4258 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4259 subfacet->actions, subfacet->actions_len) {
4260 const struct ovs_action_push_vlan *vlan;
4261 struct ofport_dpif *port;
4263 switch (nl_attr_type(a)) {
4264 case OVS_ACTION_ATTR_OUTPUT:
4265 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4266 if (port && port->bundle && port->bundle->bond) {
4267 bond_account(port->bundle->bond, &facet->flow,
4268 vlan_tci_to_vid(vlan_tci), n_bytes);
4272 case OVS_ACTION_ATTR_POP_VLAN:
4273 vlan_tci = htons(0);
4276 case OVS_ACTION_ATTR_PUSH_VLAN:
4277 vlan = nl_attr_get(a);
4278 vlan_tci = vlan->vlan_tci;
4284 /* Returns true if the only action for 'facet' is to send to the controller.
4285 * (We don't report NetFlow expiration messages for such facets because they
4286 * are just part of the control logic for the network, not real traffic). */
4288 facet_is_controller_flow(struct facet *facet)
4291 const struct rule *rule = &facet->rule->up;
4292 const struct ofpact *ofpacts = rule->ofpacts;
4293 size_t ofpacts_len = rule->ofpacts_len;
4295 if (ofpacts_len > 0 &&
4296 ofpacts->type == OFPACT_CONTROLLER &&
4297 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4304 /* Folds all of 'facet''s statistics into its rule. Also updates the
4305 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4306 * 'facet''s statistics in the datapath should have been zeroed and folded into
4307 * its packet and byte counts before this function is called. */
4309 facet_flush_stats(struct facet *facet)
4311 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4312 struct subfacet *subfacet;
4314 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4315 ovs_assert(!subfacet->dp_byte_count);
4316 ovs_assert(!subfacet->dp_packet_count);
4319 facet_push_stats(facet);
4320 if (facet->accounted_bytes < facet->byte_count) {
4321 facet_account(facet);
4322 facet->accounted_bytes = facet->byte_count;
4325 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4326 struct ofexpired expired;
4327 expired.flow = facet->flow;
4328 expired.packet_count = facet->packet_count;
4329 expired.byte_count = facet->byte_count;
4330 expired.used = facet->used;
4331 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4334 facet->rule->packet_count += facet->packet_count;
4335 facet->rule->byte_count += facet->byte_count;
4337 /* Reset counters to prevent double counting if 'facet' ever gets
4339 facet_reset_counters(facet);
4341 netflow_flow_clear(&facet->nf_flow);
4342 facet->tcp_flags = 0;
4345 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4346 * Returns it if found, otherwise a null pointer.
4348 * 'hash' must be the return value of flow_hash(flow, 0).
4350 * The returned facet might need revalidation; use facet_lookup_valid()
4351 * instead if that is important. */
4352 static struct facet *
4353 facet_find(struct ofproto_dpif *ofproto,
4354 const struct flow *flow, uint32_t hash)
4356 struct facet *facet;
4358 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4359 if (flow_equal(flow, &facet->flow)) {
4367 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4368 * Returns it if found, otherwise a null pointer.
4370 * 'hash' must be the return value of flow_hash(flow, 0).
4372 * The returned facet is guaranteed to be valid. */
4373 static struct facet *
4374 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4377 struct facet *facet;
4379 facet = facet_find(ofproto, flow, hash);
4381 && (ofproto->backer->need_revalidate
4382 || tag_set_intersects(&ofproto->backer->revalidate_set,
4384 facet_revalidate(facet);
4391 subfacet_path_to_string(enum subfacet_path path)
4394 case SF_NOT_INSTALLED:
4395 return "not installed";
4397 return "in fast path";
4399 return "in slow path";
4405 /* Returns the path in which a subfacet should be installed if its 'slow'
4406 * member has the specified value. */
4407 static enum subfacet_path
4408 subfacet_want_path(enum slow_path_reason slow)
4410 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4413 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4414 * supposing that its actions have been recalculated as 'want_actions' and that
4415 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4417 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4418 const struct ofpbuf *want_actions)
4420 enum subfacet_path want_path = subfacet_want_path(slow);
4421 return (want_path != subfacet->path
4422 || (want_path == SF_FAST_PATH
4423 && (subfacet->actions_len != want_actions->size
4424 || memcmp(subfacet->actions, want_actions->data,
4425 subfacet->actions_len))));
4429 facet_check_consistency(struct facet *facet)
4431 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4435 uint64_t odp_actions_stub[1024 / 8];
4436 struct ofpbuf odp_actions;
4438 struct rule_dpif *rule;
4439 struct subfacet *subfacet;
4440 bool may_log = false;
4443 /* Check the rule for consistency. */
4444 rule = rule_dpif_lookup(ofproto, &facet->flow);
4445 ok = rule == facet->rule;
4447 may_log = !VLOG_DROP_WARN(&rl);
4452 flow_format(&s, &facet->flow);
4453 ds_put_format(&s, ": facet associated with wrong rule (was "
4454 "table=%"PRIu8",", facet->rule->up.table_id);
4455 cls_rule_format(&facet->rule->up.cr, &s);
4456 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4458 cls_rule_format(&rule->up.cr, &s);
4459 ds_put_char(&s, ')');
4461 VLOG_WARN("%s", ds_cstr(&s));
4466 /* Check the datapath actions for consistency. */
4467 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4468 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4469 enum subfacet_path want_path;
4470 struct odputil_keybuf keybuf;
4471 struct action_xlate_ctx ctx;
4475 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4476 subfacet->initial_tci, rule, 0, NULL);
4477 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4480 if (subfacet->path == SF_NOT_INSTALLED) {
4481 /* This only happens if the datapath reported an error when we
4482 * tried to install the flow. Don't flag another error here. */
4486 want_path = subfacet_want_path(subfacet->slow);
4487 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4488 /* The actions for slow-path flows may legitimately vary from one
4489 * packet to the next. We're done. */
4493 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4497 /* Inconsistency! */
4499 may_log = !VLOG_DROP_WARN(&rl);
4503 /* Rate-limited, skip reporting. */
4508 subfacet_get_key(subfacet, &keybuf, &key);
4509 odp_flow_key_format(key.data, key.size, &s);
4511 ds_put_cstr(&s, ": inconsistency in subfacet");
4512 if (want_path != subfacet->path) {
4513 enum odp_key_fitness fitness = subfacet->key_fitness;
4515 ds_put_format(&s, " (%s, fitness=%s)",
4516 subfacet_path_to_string(subfacet->path),
4517 odp_key_fitness_to_string(fitness));
4518 ds_put_format(&s, " (should have been %s)",
4519 subfacet_path_to_string(want_path));
4520 } else if (want_path == SF_FAST_PATH) {
4521 ds_put_cstr(&s, " (actions were: ");
4522 format_odp_actions(&s, subfacet->actions,
4523 subfacet->actions_len);
4524 ds_put_cstr(&s, ") (correct actions: ");
4525 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4526 ds_put_char(&s, ')');
4528 ds_put_cstr(&s, " (actions: ");
4529 format_odp_actions(&s, subfacet->actions,
4530 subfacet->actions_len);
4531 ds_put_char(&s, ')');
4533 VLOG_WARN("%s", ds_cstr(&s));
4536 ofpbuf_uninit(&odp_actions);
4541 /* Re-searches the classifier for 'facet':
4543 * - If the rule found is different from 'facet''s current rule, moves
4544 * 'facet' to the new rule and recompiles its actions.
4546 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4547 * where it is and recompiles its actions anyway. */
4549 facet_revalidate(struct facet *facet)
4551 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4553 struct nlattr *odp_actions;
4556 struct actions *new_actions;
4558 struct action_xlate_ctx ctx;
4559 uint64_t odp_actions_stub[1024 / 8];
4560 struct ofpbuf odp_actions;
4562 struct rule_dpif *new_rule;
4563 struct subfacet *subfacet;
4566 COVERAGE_INC(facet_revalidate);
4568 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4570 /* Calculate new datapath actions.
4572 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4573 * emit a NetFlow expiration and, if so, we need to have the old state
4574 * around to properly compose it. */
4576 /* If the datapath actions changed or the installability changed,
4577 * then we need to talk to the datapath. */
4580 memset(&ctx, 0, sizeof ctx);
4581 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4582 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4583 enum slow_path_reason slow;
4585 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4586 subfacet->initial_tci, new_rule, 0, NULL);
4587 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4590 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4591 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4592 struct dpif_flow_stats stats;
4594 subfacet_install(subfacet,
4595 odp_actions.data, odp_actions.size, &stats, slow);
4596 subfacet_update_stats(subfacet, &stats);
4599 new_actions = xcalloc(list_size(&facet->subfacets),
4600 sizeof *new_actions);
4602 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4604 new_actions[i].actions_len = odp_actions.size;
4609 ofpbuf_uninit(&odp_actions);
4612 facet_flush_stats(facet);
4615 /* Update 'facet' now that we've taken care of all the old state. */
4616 facet->tags = ctx.tags;
4617 facet->nf_flow.output_iface = ctx.nf_output_iface;
4618 facet->has_learn = ctx.has_learn;
4619 facet->has_normal = ctx.has_normal;
4620 facet->has_fin_timeout = ctx.has_fin_timeout;
4621 facet->mirrors = ctx.mirrors;
4624 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4625 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4627 if (new_actions && new_actions[i].odp_actions) {
4628 free(subfacet->actions);
4629 subfacet->actions = new_actions[i].odp_actions;
4630 subfacet->actions_len = new_actions[i].actions_len;
4636 if (facet->rule != new_rule) {
4637 COVERAGE_INC(facet_changed_rule);
4638 list_remove(&facet->list_node);
4639 list_push_back(&new_rule->facets, &facet->list_node);
4640 facet->rule = new_rule;
4641 facet->used = new_rule->up.created;
4642 facet->prev_used = facet->used;
4646 /* Updates 'facet''s used time. Caller is responsible for calling
4647 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4649 facet_update_time(struct facet *facet, long long int used)
4651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4652 if (used > facet->used) {
4654 ofproto_rule_update_used(&facet->rule->up, used);
4655 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4660 facet_reset_counters(struct facet *facet)
4662 facet->packet_count = 0;
4663 facet->byte_count = 0;
4664 facet->prev_packet_count = 0;
4665 facet->prev_byte_count = 0;
4666 facet->accounted_bytes = 0;
4670 facet_push_stats(struct facet *facet)
4672 struct dpif_flow_stats stats;
4674 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4675 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4676 ovs_assert(facet->used >= facet->prev_used);
4678 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4679 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4680 stats.used = facet->used;
4681 stats.tcp_flags = 0;
4683 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4684 facet->prev_packet_count = facet->packet_count;
4685 facet->prev_byte_count = facet->byte_count;
4686 facet->prev_used = facet->used;
4688 flow_push_stats(facet->rule, &facet->flow, &stats);
4690 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4691 facet->mirrors, stats.n_packets, stats.n_bytes);
4696 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4698 rule->packet_count += stats->n_packets;
4699 rule->byte_count += stats->n_bytes;
4700 ofproto_rule_update_used(&rule->up, stats->used);
4703 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4704 * 'rule''s actions and mirrors. */
4706 flow_push_stats(struct rule_dpif *rule,
4707 const struct flow *flow, const struct dpif_flow_stats *stats)
4709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4710 struct action_xlate_ctx ctx;
4712 ofproto_rule_update_used(&rule->up, stats->used);
4714 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4716 ctx.resubmit_stats = stats;
4717 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4718 rule->up.ofpacts_len);
4723 static struct subfacet *
4724 subfacet_find(struct ofproto_dpif *ofproto,
4725 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4726 const struct flow *flow)
4728 struct subfacet *subfacet;
4730 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4731 &ofproto->subfacets) {
4733 ? (subfacet->key_len == key_len
4734 && !memcmp(key, subfacet->key, key_len))
4735 : flow_equal(flow, &subfacet->facet->flow)) {
4743 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4744 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4745 * existing subfacet if there is one, otherwise creates and returns a
4748 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4749 * which case the caller must populate the actions with
4750 * subfacet_make_actions(). */
4751 static struct subfacet *
4752 subfacet_create(struct facet *facet, struct flow_miss *miss,
4755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4756 enum odp_key_fitness key_fitness = miss->key_fitness;
4757 const struct nlattr *key = miss->key;
4758 size_t key_len = miss->key_len;
4760 struct subfacet *subfacet;
4762 key_hash = odp_flow_key_hash(key, key_len);
4764 if (list_is_empty(&facet->subfacets)) {
4765 subfacet = &facet->one_subfacet;
4767 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4770 if (subfacet->facet == facet) {
4774 /* This shouldn't happen. */
4775 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4776 subfacet_destroy(subfacet);
4779 subfacet = xmalloc(sizeof *subfacet);
4782 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4783 list_push_back(&facet->subfacets, &subfacet->list_node);
4784 subfacet->facet = facet;
4785 subfacet->key_fitness = key_fitness;
4786 if (key_fitness != ODP_FIT_PERFECT) {
4787 subfacet->key = xmemdup(key, key_len);
4788 subfacet->key_len = key_len;
4790 subfacet->key = NULL;
4791 subfacet->key_len = 0;
4793 subfacet->used = now;
4794 subfacet->dp_packet_count = 0;
4795 subfacet->dp_byte_count = 0;
4796 subfacet->actions_len = 0;
4797 subfacet->actions = NULL;
4798 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4801 subfacet->path = SF_NOT_INSTALLED;
4802 subfacet->initial_tci = miss->initial_tci;
4803 subfacet->odp_in_port = miss->odp_in_port;
4808 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4809 * its facet within 'ofproto', and frees it. */
4811 subfacet_destroy__(struct subfacet *subfacet)
4813 struct facet *facet = subfacet->facet;
4814 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4816 subfacet_uninstall(subfacet);
4817 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4818 list_remove(&subfacet->list_node);
4819 free(subfacet->key);
4820 free(subfacet->actions);
4821 if (subfacet != &facet->one_subfacet) {
4826 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4827 * last remaining subfacet in its facet destroys the facet too. */
4829 subfacet_destroy(struct subfacet *subfacet)
4831 struct facet *facet = subfacet->facet;
4833 if (list_is_singleton(&facet->subfacets)) {
4834 /* facet_remove() needs at least one subfacet (it will remove it). */
4835 facet_remove(facet);
4837 subfacet_destroy__(subfacet);
4842 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4843 struct subfacet **subfacets, int n)
4845 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4846 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4847 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4848 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4849 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4852 for (i = 0; i < n; i++) {
4853 ops[i].type = DPIF_OP_FLOW_DEL;
4854 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4855 ops[i].u.flow_del.key = keys[i].data;
4856 ops[i].u.flow_del.key_len = keys[i].size;
4857 ops[i].u.flow_del.stats = &stats[i];
4861 dpif_operate(ofproto->backer->dpif, opsp, n);
4862 for (i = 0; i < n; i++) {
4863 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4864 subfacets[i]->path = SF_NOT_INSTALLED;
4865 subfacet_destroy(subfacets[i]);
4869 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4870 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4871 * for use as temporary storage. */
4873 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4877 if (!subfacet->key) {
4878 struct flow *flow = &subfacet->facet->flow;
4880 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4881 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4883 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4887 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4888 * Translates the actions into 'odp_actions', which the caller must have
4889 * initialized and is responsible for uninitializing. */
4891 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4892 struct ofpbuf *odp_actions)
4894 struct facet *facet = subfacet->facet;
4895 struct rule_dpif *rule = facet->rule;
4896 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4898 struct action_xlate_ctx ctx;
4900 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4902 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4903 facet->tags = ctx.tags;
4904 facet->has_learn = ctx.has_learn;
4905 facet->has_normal = ctx.has_normal;
4906 facet->has_fin_timeout = ctx.has_fin_timeout;
4907 facet->nf_flow.output_iface = ctx.nf_output_iface;
4908 facet->mirrors = ctx.mirrors;
4910 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4911 if (subfacet->actions_len != odp_actions->size
4912 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4913 free(subfacet->actions);
4914 subfacet->actions_len = odp_actions->size;
4915 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4919 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4920 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4921 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4922 * since 'subfacet' was last updated.
4924 * Returns 0 if successful, otherwise a positive errno value. */
4926 subfacet_install(struct subfacet *subfacet,
4927 const struct nlattr *actions, size_t actions_len,
4928 struct dpif_flow_stats *stats,
4929 enum slow_path_reason slow)
4931 struct facet *facet = subfacet->facet;
4932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4933 enum subfacet_path path = subfacet_want_path(slow);
4934 uint64_t slow_path_stub[128 / 8];
4935 struct odputil_keybuf keybuf;
4936 enum dpif_flow_put_flags flags;
4940 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4942 flags |= DPIF_FP_ZERO_STATS;
4945 if (path == SF_SLOW_PATH) {
4946 compose_slow_path(ofproto, &facet->flow, slow,
4947 slow_path_stub, sizeof slow_path_stub,
4948 &actions, &actions_len);
4951 subfacet_get_key(subfacet, &keybuf, &key);
4952 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4953 actions, actions_len, stats);
4956 subfacet_reset_dp_stats(subfacet, stats);
4960 subfacet->path = path;
4966 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4968 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4969 stats, subfacet->slow);
4972 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4974 subfacet_uninstall(struct subfacet *subfacet)
4976 if (subfacet->path != SF_NOT_INSTALLED) {
4977 struct rule_dpif *rule = subfacet->facet->rule;
4978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4979 struct odputil_keybuf keybuf;
4980 struct dpif_flow_stats stats;
4984 subfacet_get_key(subfacet, &keybuf, &key);
4985 error = dpif_flow_del(ofproto->backer->dpif,
4986 key.data, key.size, &stats);
4987 subfacet_reset_dp_stats(subfacet, &stats);
4989 subfacet_update_stats(subfacet, &stats);
4991 subfacet->path = SF_NOT_INSTALLED;
4993 ovs_assert(subfacet->dp_packet_count == 0);
4994 ovs_assert(subfacet->dp_byte_count == 0);
4998 /* Resets 'subfacet''s datapath statistics counters. This should be called
4999 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5000 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5001 * was reset in the datapath. 'stats' will be modified to include only
5002 * statistics new since 'subfacet' was last updated. */
5004 subfacet_reset_dp_stats(struct subfacet *subfacet,
5005 struct dpif_flow_stats *stats)
5008 && subfacet->dp_packet_count <= stats->n_packets
5009 && subfacet->dp_byte_count <= stats->n_bytes) {
5010 stats->n_packets -= subfacet->dp_packet_count;
5011 stats->n_bytes -= subfacet->dp_byte_count;
5014 subfacet->dp_packet_count = 0;
5015 subfacet->dp_byte_count = 0;
5018 /* Updates 'subfacet''s used time. The caller is responsible for calling
5019 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5021 subfacet_update_time(struct subfacet *subfacet, long long int used)
5023 if (used > subfacet->used) {
5024 subfacet->used = used;
5025 facet_update_time(subfacet->facet, used);
5029 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5031 * Because of the meaning of a subfacet's counters, it only makes sense to do
5032 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5033 * represents a packet that was sent by hand or if it represents statistics
5034 * that have been cleared out of the datapath. */
5036 subfacet_update_stats(struct subfacet *subfacet,
5037 const struct dpif_flow_stats *stats)
5039 if (stats->n_packets || stats->used > subfacet->used) {
5040 struct facet *facet = subfacet->facet;
5042 subfacet_update_time(subfacet, stats->used);
5043 facet->packet_count += stats->n_packets;
5044 facet->byte_count += stats->n_bytes;
5045 facet->tcp_flags |= stats->tcp_flags;
5046 facet_push_stats(facet);
5047 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5053 static struct rule_dpif *
5054 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5056 struct rule_dpif *rule;
5058 rule = rule_dpif_lookup__(ofproto, flow, 0);
5063 return rule_dpif_miss_rule(ofproto, flow);
5066 static struct rule_dpif *
5067 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5070 struct cls_rule *cls_rule;
5071 struct classifier *cls;
5073 if (table_id >= N_TABLES) {
5077 cls = &ofproto->up.tables[table_id].cls;
5078 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5079 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5080 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5081 * are unavailable. */
5082 struct flow ofpc_normal_flow = *flow;
5083 ofpc_normal_flow.tp_src = htons(0);
5084 ofpc_normal_flow.tp_dst = htons(0);
5085 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5087 cls_rule = classifier_lookup(cls, flow);
5089 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5092 static struct rule_dpif *
5093 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5095 struct ofport_dpif *port;
5097 port = get_ofp_port(ofproto, flow->in_port);
5099 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5100 return ofproto->miss_rule;
5103 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5104 return ofproto->no_packet_in_rule;
5106 return ofproto->miss_rule;
5110 complete_operation(struct rule_dpif *rule)
5112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5114 rule_invalidate(rule);
5116 struct dpif_completion *c = xmalloc(sizeof *c);
5117 c->op = rule->up.pending;
5118 list_push_back(&ofproto->completions, &c->list_node);
5120 ofoperation_complete(rule->up.pending, 0);
5124 static struct rule *
5127 struct rule_dpif *rule = xmalloc(sizeof *rule);
5132 rule_dealloc(struct rule *rule_)
5134 struct rule_dpif *rule = rule_dpif_cast(rule_);
5139 rule_construct(struct rule *rule_)
5141 struct rule_dpif *rule = rule_dpif_cast(rule_);
5142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5143 struct rule_dpif *victim;
5146 rule->packet_count = 0;
5147 rule->byte_count = 0;
5149 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5150 if (victim && !list_is_empty(&victim->facets)) {
5151 struct facet *facet;
5153 rule->facets = victim->facets;
5154 list_moved(&rule->facets);
5155 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5156 /* XXX: We're only clearing our local counters here. It's possible
5157 * that quite a few packets are unaccounted for in the datapath
5158 * statistics. These will be accounted to the new rule instead of
5159 * cleared as required. This could be fixed by clearing out the
5160 * datapath statistics for this facet, but currently it doesn't
5162 facet_reset_counters(facet);
5166 /* Must avoid list_moved() in this case. */
5167 list_init(&rule->facets);
5170 table_id = rule->up.table_id;
5172 rule->tag = victim->tag;
5173 } else if (table_id == 0) {
5178 miniflow_expand(&rule->up.cr.match.flow, &flow);
5179 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5180 ofproto->tables[table_id].basis);
5183 complete_operation(rule);
5188 rule_destruct(struct rule *rule_)
5190 struct rule_dpif *rule = rule_dpif_cast(rule_);
5191 struct facet *facet, *next_facet;
5193 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5194 facet_revalidate(facet);
5197 complete_operation(rule);
5201 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5203 struct rule_dpif *rule = rule_dpif_cast(rule_);
5204 struct facet *facet;
5206 /* Start from historical data for 'rule' itself that are no longer tracked
5207 * in facets. This counts, for example, facets that have expired. */
5208 *packets = rule->packet_count;
5209 *bytes = rule->byte_count;
5211 /* Add any statistics that are tracked by facets. This includes
5212 * statistical data recently updated by ofproto_update_stats() as well as
5213 * stats for packets that were executed "by hand" via dpif_execute(). */
5214 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5215 *packets += facet->packet_count;
5216 *bytes += facet->byte_count;
5221 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5222 struct ofpbuf *packet)
5224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5226 struct dpif_flow_stats stats;
5228 struct action_xlate_ctx ctx;
5229 uint64_t odp_actions_stub[1024 / 8];
5230 struct ofpbuf odp_actions;
5232 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5233 rule_credit_stats(rule, &stats);
5235 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5236 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5237 rule, stats.tcp_flags, packet);
5238 ctx.resubmit_stats = &stats;
5239 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5241 execute_odp_actions(ofproto, flow, odp_actions.data,
5242 odp_actions.size, packet);
5244 ofpbuf_uninit(&odp_actions);
5248 rule_execute(struct rule *rule, const struct flow *flow,
5249 struct ofpbuf *packet)
5251 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5252 ofpbuf_delete(packet);
5257 rule_modify_actions(struct rule *rule_)
5259 struct rule_dpif *rule = rule_dpif_cast(rule_);
5261 complete_operation(rule);
5264 /* Sends 'packet' out 'ofport'.
5265 * May modify 'packet'.
5266 * Returns 0 if successful, otherwise a positive errno value. */
5268 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5270 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5271 struct ofpbuf key, odp_actions;
5272 struct odputil_keybuf keybuf;
5277 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5278 if (netdev_vport_is_patch(ofport->up.netdev)) {
5279 struct ofproto_dpif *peer_ofproto;
5280 struct dpif_flow_stats stats;
5281 struct ofport_dpif *peer;
5282 struct rule_dpif *rule;
5284 peer = ofport_get_peer(ofport);
5289 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5290 netdev_vport_patch_inc_tx(ofport->up.netdev, &stats);
5291 netdev_vport_patch_inc_rx(peer->up.netdev, &stats);
5293 flow.in_port = peer->up.ofp_port;
5294 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5295 rule = rule_dpif_lookup(peer_ofproto, &flow);
5296 rule_dpif_execute(rule, &flow, packet);
5301 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5303 if (odp_port != ofport->odp_port) {
5304 eth_pop_vlan(packet);
5305 flow.vlan_tci = htons(0);
5308 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5309 odp_flow_key_from_flow(&key, &flow,
5310 ofp_port_to_odp_port(ofproto, flow.in_port));
5312 ofpbuf_init(&odp_actions, 32);
5313 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5315 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5316 error = dpif_execute(ofproto->backer->dpif,
5318 odp_actions.data, odp_actions.size,
5320 ofpbuf_uninit(&odp_actions);
5323 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5324 ofproto->up.name, odp_port, strerror(error));
5326 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5330 /* OpenFlow to datapath action translation. */
5332 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5333 struct action_xlate_ctx *);
5334 static void xlate_normal(struct action_xlate_ctx *);
5336 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5337 * The action will state 'slow' as the reason that the action is in the slow
5338 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5339 * dump-flows" output to see why a flow is in the slow path.)
5341 * The 'stub_size' bytes in 'stub' will be used to store the action.
5342 * 'stub_size' must be large enough for the action.
5344 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5347 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5348 enum slow_path_reason slow,
5349 uint64_t *stub, size_t stub_size,
5350 const struct nlattr **actionsp, size_t *actions_lenp)
5352 union user_action_cookie cookie;
5355 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5356 cookie.slow_path.unused = 0;
5357 cookie.slow_path.reason = slow;
5359 ofpbuf_use_stack(&buf, stub, stub_size);
5360 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5361 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5362 odp_put_userspace_action(pid, &cookie, &buf);
5364 put_userspace_action(ofproto, &buf, flow, &cookie);
5366 *actionsp = buf.data;
5367 *actions_lenp = buf.size;
5371 put_userspace_action(const struct ofproto_dpif *ofproto,
5372 struct ofpbuf *odp_actions,
5373 const struct flow *flow,
5374 const union user_action_cookie *cookie)
5378 pid = dpif_port_get_pid(ofproto->backer->dpif,
5379 ofp_port_to_odp_port(ofproto, flow->in_port));
5381 return odp_put_userspace_action(pid, cookie, odp_actions);
5385 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5386 ovs_be16 vlan_tci, uint32_t odp_port,
5387 unsigned int n_outputs, union user_action_cookie *cookie)
5391 cookie->type = USER_ACTION_COOKIE_SFLOW;
5392 cookie->sflow.vlan_tci = vlan_tci;
5394 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5395 * port information") for the interpretation of cookie->output. */
5396 switch (n_outputs) {
5398 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5399 cookie->sflow.output = 0x40000000 | 256;
5403 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5405 cookie->sflow.output = ifindex;
5410 /* 0x80000000 means "multiple output ports. */
5411 cookie->sflow.output = 0x80000000 | n_outputs;
5416 /* Compose SAMPLE action for sFlow. */
5418 compose_sflow_action(const struct ofproto_dpif *ofproto,
5419 struct ofpbuf *odp_actions,
5420 const struct flow *flow,
5423 uint32_t probability;
5424 union user_action_cookie cookie;
5425 size_t sample_offset, actions_offset;
5428 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5432 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5434 /* Number of packets out of UINT_MAX to sample. */
5435 probability = dpif_sflow_get_probability(ofproto->sflow);
5436 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5438 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5439 compose_sflow_cookie(ofproto, htons(0), odp_port,
5440 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5441 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5443 nl_msg_end_nested(odp_actions, actions_offset);
5444 nl_msg_end_nested(odp_actions, sample_offset);
5445 return cookie_offset;
5448 /* SAMPLE action must be first action in any given list of actions.
5449 * At this point we do not have all information required to build it. So try to
5450 * build sample action as complete as possible. */
5452 add_sflow_action(struct action_xlate_ctx *ctx)
5454 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5456 &ctx->flow, OVSP_NONE);
5457 ctx->sflow_odp_port = 0;
5458 ctx->sflow_n_outputs = 0;
5461 /* Fix SAMPLE action according to data collected while composing ODP actions.
5462 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5463 * USERSPACE action's user-cookie which is required for sflow. */
5465 fix_sflow_action(struct action_xlate_ctx *ctx)
5467 const struct flow *base = &ctx->base_flow;
5468 union user_action_cookie *cookie;
5470 if (!ctx->user_cookie_offset) {
5474 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5476 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5478 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5479 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5483 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5486 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5487 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5488 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5489 struct priority_to_dscp *pdscp;
5490 uint32_t out_port, odp_port;
5492 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5493 * before traversing a patch port. */
5494 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5497 xlate_report(ctx, "Nonexistent output port");
5499 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5500 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5502 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5503 xlate_report(ctx, "STP not in forwarding state, skipping output");
5507 if (netdev_vport_is_patch(ofport->up.netdev)) {
5508 struct ofport_dpif *peer = ofport_get_peer(ofport);
5509 struct flow old_flow = ctx->flow;
5510 const struct ofproto_dpif *peer_ofproto;
5513 xlate_report(ctx, "Nonexistent patch port peer");
5517 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5518 if (peer_ofproto->backer != ctx->ofproto->backer) {
5519 xlate_report(ctx, "Patch port peer on a different datapath");
5523 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5524 ctx->flow.in_port = peer->up.ofp_port;
5525 ctx->flow.metadata = htonll(0);
5526 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5527 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5528 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5529 ctx->flow = old_flow;
5530 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5532 if (ctx->resubmit_stats) {
5533 netdev_vport_patch_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5534 netdev_vport_patch_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5540 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5542 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5543 ctx->flow.nw_tos |= pdscp->dscp;
5546 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5547 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5548 ctx->flow.vlan_tci);
5549 if (out_port != odp_port) {
5550 ctx->flow.vlan_tci = htons(0);
5552 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5553 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5555 ctx->sflow_odp_port = odp_port;
5556 ctx->sflow_n_outputs++;
5557 ctx->nf_output_iface = ofp_port;
5558 ctx->flow.vlan_tci = flow_vlan_tci;
5559 ctx->flow.nw_tos = flow_nw_tos;
5563 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5565 compose_output_action__(ctx, ofp_port, true);
5569 xlate_table_action(struct action_xlate_ctx *ctx,
5570 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5572 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5573 struct ofproto_dpif *ofproto = ctx->ofproto;
5574 struct rule_dpif *rule;
5575 uint16_t old_in_port;
5576 uint8_t old_table_id;
5578 old_table_id = ctx->table_id;
5579 ctx->table_id = table_id;
5581 /* Look up a flow with 'in_port' as the input port. */
5582 old_in_port = ctx->flow.in_port;
5583 ctx->flow.in_port = in_port;
5584 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5587 if (table_id > 0 && table_id < N_TABLES) {
5588 struct table_dpif *table = &ofproto->tables[table_id];
5589 if (table->other_table) {
5590 ctx->tags |= (rule && rule->tag
5592 : rule_calculate_tag(&ctx->flow,
5593 &table->other_table->mask,
5598 /* Restore the original input port. Otherwise OFPP_NORMAL and
5599 * OFPP_IN_PORT will have surprising behavior. */
5600 ctx->flow.in_port = old_in_port;
5602 if (ctx->resubmit_hook) {
5603 ctx->resubmit_hook(ctx, rule);
5606 if (rule == NULL && may_packet_in) {
5608 * check if table configuration flags
5609 * OFPTC_TABLE_MISS_CONTROLLER, default.
5610 * OFPTC_TABLE_MISS_CONTINUE,
5611 * OFPTC_TABLE_MISS_DROP
5612 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5614 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5618 struct rule_dpif *old_rule = ctx->rule;
5620 if (ctx->resubmit_stats) {
5621 rule_credit_stats(rule, ctx->resubmit_stats);
5626 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5627 ctx->rule = old_rule;
5631 ctx->table_id = old_table_id;
5633 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5635 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5636 MAX_RESUBMIT_RECURSION);
5637 ctx->max_resubmit_trigger = true;
5642 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5643 const struct ofpact_resubmit *resubmit)
5648 in_port = resubmit->in_port;
5649 if (in_port == OFPP_IN_PORT) {
5650 in_port = ctx->flow.in_port;
5653 table_id = resubmit->table_id;
5654 if (table_id == 255) {
5655 table_id = ctx->table_id;
5658 xlate_table_action(ctx, in_port, table_id, false);
5662 flood_packets(struct action_xlate_ctx *ctx, bool all)
5664 struct ofport_dpif *ofport;
5666 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5667 uint16_t ofp_port = ofport->up.ofp_port;
5669 if (ofp_port == ctx->flow.in_port) {
5674 compose_output_action__(ctx, ofp_port, false);
5675 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5676 compose_output_action(ctx, ofp_port);
5680 ctx->nf_output_iface = NF_OUT_FLOOD;
5684 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5685 enum ofp_packet_in_reason reason,
5686 uint16_t controller_id)
5688 struct ofputil_packet_in pin;
5689 struct ofpbuf *packet;
5691 ctx->slow |= SLOW_CONTROLLER;
5696 packet = ofpbuf_clone(ctx->packet);
5698 if (packet->l2 && packet->l3) {
5699 struct eth_header *eh;
5701 eth_pop_vlan(packet);
5704 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5705 * LLC frame. Calculating the Ethernet type of these frames is more
5706 * trouble than seems appropriate for a simple assertion. */
5707 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5708 || eh->eth_type == ctx->flow.dl_type);
5710 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5711 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5713 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5714 eth_push_vlan(packet, ctx->flow.vlan_tci);
5718 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5719 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5720 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5724 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5725 packet_set_tcp_port(packet, ctx->flow.tp_src,
5727 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5728 packet_set_udp_port(packet, ctx->flow.tp_src,
5735 pin.packet = packet->data;
5736 pin.packet_len = packet->size;
5737 pin.reason = reason;
5738 pin.controller_id = controller_id;
5739 pin.table_id = ctx->table_id;
5740 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5743 flow_get_metadata(&ctx->flow, &pin.fmd);
5745 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5746 ofpbuf_delete(packet);
5750 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5752 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5753 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5757 if (ctx->flow.nw_ttl > 1) {
5763 for (i = 0; i < ids->n_controllers; i++) {
5764 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5768 /* Stop processing for current table. */
5774 xlate_output_action(struct action_xlate_ctx *ctx,
5775 uint16_t port, uint16_t max_len, bool may_packet_in)
5777 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5779 ctx->nf_output_iface = NF_OUT_DROP;
5783 compose_output_action(ctx, ctx->flow.in_port);
5786 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5792 flood_packets(ctx, false);
5795 flood_packets(ctx, true);
5797 case OFPP_CONTROLLER:
5798 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5804 if (port != ctx->flow.in_port) {
5805 compose_output_action(ctx, port);
5807 xlate_report(ctx, "skipping output to input port");
5812 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5813 ctx->nf_output_iface = NF_OUT_FLOOD;
5814 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5815 ctx->nf_output_iface = prev_nf_output_iface;
5816 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5817 ctx->nf_output_iface != NF_OUT_FLOOD) {
5818 ctx->nf_output_iface = NF_OUT_MULTI;
5823 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5824 const struct ofpact_output_reg *or)
5826 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5827 if (port <= UINT16_MAX) {
5828 xlate_output_action(ctx, port, or->max_len, false);
5833 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5834 const struct ofpact_enqueue *enqueue)
5836 uint16_t ofp_port = enqueue->port;
5837 uint32_t queue_id = enqueue->queue;
5838 uint32_t flow_priority, priority;
5841 /* Translate queue to priority. */
5842 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5843 queue_id, &priority);
5845 /* Fall back to ordinary output action. */
5846 xlate_output_action(ctx, enqueue->port, 0, false);
5850 /* Check output port. */
5851 if (ofp_port == OFPP_IN_PORT) {
5852 ofp_port = ctx->flow.in_port;
5853 } else if (ofp_port == ctx->flow.in_port) {
5857 /* Add datapath actions. */
5858 flow_priority = ctx->flow.skb_priority;
5859 ctx->flow.skb_priority = priority;
5860 compose_output_action(ctx, ofp_port);
5861 ctx->flow.skb_priority = flow_priority;
5863 /* Update NetFlow output port. */
5864 if (ctx->nf_output_iface == NF_OUT_DROP) {
5865 ctx->nf_output_iface = ofp_port;
5866 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5867 ctx->nf_output_iface = NF_OUT_MULTI;
5872 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5874 uint32_t skb_priority;
5876 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5877 queue_id, &skb_priority)) {
5878 ctx->flow.skb_priority = skb_priority;
5880 /* Couldn't translate queue to a priority. Nothing to do. A warning
5881 * has already been logged. */
5885 struct xlate_reg_state {
5891 xlate_autopath(struct action_xlate_ctx *ctx,
5892 const struct ofpact_autopath *ap)
5894 uint16_t ofp_port = ap->port;
5895 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5897 if (!port || !port->bundle) {
5898 ofp_port = OFPP_NONE;
5899 } else if (port->bundle->bond) {
5900 /* Autopath does not support VLAN hashing. */
5901 struct ofport_dpif *slave = bond_choose_output_slave(
5902 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5904 ofp_port = slave->up.ofp_port;
5907 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5911 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5913 struct ofproto_dpif *ofproto = ofproto_;
5914 struct ofport_dpif *port;
5924 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5927 port = get_ofp_port(ofproto, ofp_port);
5928 return port ? port->may_enable : false;
5933 xlate_bundle_action(struct action_xlate_ctx *ctx,
5934 const struct ofpact_bundle *bundle)
5938 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5939 if (bundle->dst.field) {
5940 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5942 xlate_output_action(ctx, port, 0, false);
5947 xlate_learn_action(struct action_xlate_ctx *ctx,
5948 const struct ofpact_learn *learn)
5950 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5951 struct ofputil_flow_mod fm;
5952 uint64_t ofpacts_stub[1024 / 8];
5953 struct ofpbuf ofpacts;
5956 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5957 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5959 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5960 if (error && !VLOG_DROP_WARN(&rl)) {
5961 VLOG_WARN("learning action failed to modify flow table (%s)",
5962 ofperr_get_name(error));
5965 ofpbuf_uninit(&ofpacts);
5968 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5969 * means "infinite". */
5971 reduce_timeout(uint16_t max, uint16_t *timeout)
5973 if (max && (!*timeout || *timeout > max)) {
5979 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5980 const struct ofpact_fin_timeout *oft)
5982 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5983 struct rule_dpif *rule = ctx->rule;
5985 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5986 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5991 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5993 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5994 ? OFPUTIL_PC_NO_RECV_STP
5995 : OFPUTIL_PC_NO_RECV)) {
5999 /* Only drop packets here if both forwarding and learning are
6000 * disabled. If just learning is enabled, we need to have
6001 * OFPP_NORMAL and the learning action have a look at the packet
6002 * before we can drop it. */
6003 if (!stp_forward_in_state(port->stp_state)
6004 && !stp_learn_in_state(port->stp_state)) {
6012 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6013 struct action_xlate_ctx *ctx)
6015 const struct ofport_dpif *port;
6016 bool was_evictable = true;
6017 const struct ofpact *a;
6019 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6020 if (port && !may_receive(port, ctx)) {
6021 /* Drop this flow. */
6026 /* Don't let the rule we're working on get evicted underneath us. */
6027 was_evictable = ctx->rule->up.evictable;
6028 ctx->rule->up.evictable = false;
6030 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6031 struct ofpact_controller *controller;
6032 const struct ofpact_metadata *metadata;
6040 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6041 ofpact_get_OUTPUT(a)->max_len, true);
6044 case OFPACT_CONTROLLER:
6045 controller = ofpact_get_CONTROLLER(a);
6046 execute_controller_action(ctx, controller->max_len,
6048 controller->controller_id);
6051 case OFPACT_ENQUEUE:
6052 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6055 case OFPACT_SET_VLAN_VID:
6056 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6057 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6061 case OFPACT_SET_VLAN_PCP:
6062 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6063 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6068 case OFPACT_STRIP_VLAN:
6069 ctx->flow.vlan_tci = htons(0);
6072 case OFPACT_PUSH_VLAN:
6073 /* XXX 802.1AD(QinQ) */
6074 ctx->flow.vlan_tci = htons(VLAN_CFI);
6077 case OFPACT_SET_ETH_SRC:
6078 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6082 case OFPACT_SET_ETH_DST:
6083 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6087 case OFPACT_SET_IPV4_SRC:
6088 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6091 case OFPACT_SET_IPV4_DST:
6092 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6095 case OFPACT_SET_IPV4_DSCP:
6096 /* OpenFlow 1.0 only supports IPv4. */
6097 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6098 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6099 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6103 case OFPACT_SET_L4_SRC_PORT:
6104 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6107 case OFPACT_SET_L4_DST_PORT:
6108 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6111 case OFPACT_RESUBMIT:
6112 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6115 case OFPACT_SET_TUNNEL:
6116 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6119 case OFPACT_SET_QUEUE:
6120 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6123 case OFPACT_POP_QUEUE:
6124 ctx->flow.skb_priority = ctx->orig_skb_priority;
6127 case OFPACT_REG_MOVE:
6128 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6131 case OFPACT_REG_LOAD:
6132 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6135 case OFPACT_DEC_TTL:
6136 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6142 /* Nothing to do. */
6145 case OFPACT_MULTIPATH:
6146 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6149 case OFPACT_AUTOPATH:
6150 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6154 ctx->ofproto->has_bundle_action = true;
6155 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6158 case OFPACT_OUTPUT_REG:
6159 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6163 ctx->has_learn = true;
6164 if (ctx->may_learn) {
6165 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6173 case OFPACT_FIN_TIMEOUT:
6174 ctx->has_fin_timeout = true;
6175 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6178 case OFPACT_CLEAR_ACTIONS:
6180 * Nothing to do because writa-actions is not supported for now.
6181 * When writa-actions is supported, clear-actions also must
6182 * be supported at the same time.
6186 case OFPACT_WRITE_METADATA:
6187 metadata = ofpact_get_WRITE_METADATA(a);
6188 ctx->flow.metadata &= ~metadata->mask;
6189 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6192 case OFPACT_GOTO_TABLE: {
6193 /* XXX remove recursion */
6194 /* It is assumed that goto-table is last action */
6195 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6196 ovs_assert(ctx->table_id < ogt->table_id);
6197 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6204 /* We've let OFPP_NORMAL and the learning action look at the packet,
6205 * so drop it now if forwarding is disabled. */
6206 if (port && !stp_forward_in_state(port->stp_state)) {
6207 ofpbuf_clear(ctx->odp_actions);
6208 add_sflow_action(ctx);
6211 ctx->rule->up.evictable = was_evictable;
6216 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6217 struct ofproto_dpif *ofproto, const struct flow *flow,
6218 ovs_be16 initial_tci, struct rule_dpif *rule,
6219 uint8_t tcp_flags, const struct ofpbuf *packet)
6221 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6223 /* Flow initialization rules:
6224 * - 'base_flow' must match the kernel's view of the packet at the
6225 * time that action processing starts. 'flow' represents any
6226 * transformations we wish to make through actions.
6227 * - By default 'base_flow' and 'flow' are the same since the input
6228 * packet matches the output before any actions are applied.
6229 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6230 * of the received packet as seen by the kernel. If we later output
6231 * to another device without any modifications this will cause us to
6232 * insert a new tag since the original one was stripped off by the
6234 * - Tunnel 'flow' is largely cleared when transitioning between
6235 * the input and output stages since it does not make sense to output
6236 * a packet with the exact headers that it was received with (i.e.
6237 * the destination IP is us). The one exception is the tun_id, which
6238 * is preserved to allow use in later resubmit lookups and loads into
6240 * - Tunnel 'base_flow' is completely cleared since that is what the
6241 * kernel does. If we wish to maintain the original values an action
6242 * needs to be generated. */
6244 ctx->ofproto = ofproto;
6246 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6247 ctx->base_flow = ctx->flow;
6248 ctx->base_flow.vlan_tci = initial_tci;
6249 ctx->flow.tunnel.tun_id = initial_tun_id;
6251 ctx->packet = packet;
6252 ctx->may_learn = packet != NULL;
6253 ctx->tcp_flags = tcp_flags;
6254 ctx->resubmit_hook = NULL;
6255 ctx->report_hook = NULL;
6256 ctx->resubmit_stats = NULL;
6259 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6260 * into datapath actions in 'odp_actions', using 'ctx'. */
6262 xlate_actions(struct action_xlate_ctx *ctx,
6263 const struct ofpact *ofpacts, size_t ofpacts_len,
6264 struct ofpbuf *odp_actions)
6266 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6267 * that in the future we always keep a copy of the original flow for
6268 * tracing purposes. */
6269 static bool hit_resubmit_limit;
6271 enum slow_path_reason special;
6273 COVERAGE_INC(ofproto_dpif_xlate);
6275 ofpbuf_clear(odp_actions);
6276 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6278 ctx->odp_actions = odp_actions;
6281 ctx->has_learn = false;
6282 ctx->has_normal = false;
6283 ctx->has_fin_timeout = false;
6284 ctx->nf_output_iface = NF_OUT_DROP;
6287 ctx->max_resubmit_trigger = false;
6288 ctx->orig_skb_priority = ctx->flow.skb_priority;
6292 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6293 /* Do this conditionally because the copy is expensive enough that it
6294 * shows up in profiles.
6296 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6297 * believe that I wasn't using it without initializing it if I kept it
6298 * in a local variable. */
6299 ctx->orig_flow = ctx->flow;
6302 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6303 switch (ctx->ofproto->up.frag_handling) {
6304 case OFPC_FRAG_NORMAL:
6305 /* We must pretend that transport ports are unavailable. */
6306 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6307 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6310 case OFPC_FRAG_DROP:
6313 case OFPC_FRAG_REASM:
6316 case OFPC_FRAG_NX_MATCH:
6317 /* Nothing to do. */
6320 case OFPC_INVALID_TTL_TO_CONTROLLER:
6325 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6327 ctx->slow |= special;
6329 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6330 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6331 uint32_t local_odp_port;
6333 add_sflow_action(ctx);
6334 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6336 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6337 if (!hit_resubmit_limit) {
6338 /* We didn't record the original flow. Make sure we do from
6340 hit_resubmit_limit = true;
6341 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6342 struct ds ds = DS_EMPTY_INITIALIZER;
6344 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6346 VLOG_ERR("Trace triggered by excessive resubmit "
6347 "recursion:\n%s", ds_cstr(&ds));
6352 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6353 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6355 ctx->odp_actions->data,
6356 ctx->odp_actions->size)) {
6357 ctx->slow |= SLOW_IN_BAND;
6359 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6361 compose_output_action(ctx, OFPP_LOCAL);
6364 if (ctx->ofproto->has_mirrors) {
6365 add_mirror_actions(ctx, &ctx->orig_flow);
6367 fix_sflow_action(ctx);
6371 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6372 * into datapath actions, using 'ctx', and discards the datapath actions. */
6374 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6375 const struct ofpact *ofpacts,
6378 uint64_t odp_actions_stub[1024 / 8];
6379 struct ofpbuf odp_actions;
6381 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6382 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6383 ofpbuf_uninit(&odp_actions);
6387 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6389 if (ctx->report_hook) {
6390 ctx->report_hook(ctx, s);
6394 /* OFPP_NORMAL implementation. */
6396 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6398 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6399 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6400 * the bundle on which the packet was received, returns the VLAN to which the
6403 * Both 'vid' and the return value are in the range 0...4095. */
6405 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6407 switch (in_bundle->vlan_mode) {
6408 case PORT_VLAN_ACCESS:
6409 return in_bundle->vlan;
6412 case PORT_VLAN_TRUNK:
6415 case PORT_VLAN_NATIVE_UNTAGGED:
6416 case PORT_VLAN_NATIVE_TAGGED:
6417 return vid ? vid : in_bundle->vlan;
6424 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6425 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6428 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6429 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6432 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6434 /* Allow any VID on the OFPP_NONE port. */
6435 if (in_bundle == &ofpp_none_bundle) {
6439 switch (in_bundle->vlan_mode) {
6440 case PORT_VLAN_ACCESS:
6443 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6444 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6445 "packet received on port %s configured as VLAN "
6446 "%"PRIu16" access port",
6447 in_bundle->ofproto->up.name, vid,
6448 in_bundle->name, in_bundle->vlan);
6454 case PORT_VLAN_NATIVE_UNTAGGED:
6455 case PORT_VLAN_NATIVE_TAGGED:
6457 /* Port must always carry its native VLAN. */
6461 case PORT_VLAN_TRUNK:
6462 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6464 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6465 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6466 "received on port %s not configured for trunking "
6468 in_bundle->ofproto->up.name, vid,
6469 in_bundle->name, vid);
6481 /* Given 'vlan', the VLAN that a packet belongs to, and
6482 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6483 * that should be included in the 802.1Q header. (If the return value is 0,
6484 * then the 802.1Q header should only be included in the packet if there is a
6487 * Both 'vlan' and the return value are in the range 0...4095. */
6489 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6491 switch (out_bundle->vlan_mode) {
6492 case PORT_VLAN_ACCESS:
6495 case PORT_VLAN_TRUNK:
6496 case PORT_VLAN_NATIVE_TAGGED:
6499 case PORT_VLAN_NATIVE_UNTAGGED:
6500 return vlan == out_bundle->vlan ? 0 : vlan;
6508 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6511 struct ofport_dpif *port;
6513 ovs_be16 tci, old_tci;
6515 vid = output_vlan_to_vid(out_bundle, vlan);
6516 if (!out_bundle->bond) {
6517 port = ofbundle_get_a_port(out_bundle);
6519 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6522 /* No slaves enabled, so drop packet. */
6527 old_tci = ctx->flow.vlan_tci;
6529 if (tci || out_bundle->use_priority_tags) {
6530 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6532 tci |= htons(VLAN_CFI);
6535 ctx->flow.vlan_tci = tci;
6537 compose_output_action(ctx, port->up.ofp_port);
6538 ctx->flow.vlan_tci = old_tci;
6542 mirror_mask_ffs(mirror_mask_t mask)
6544 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6549 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6551 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6552 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6556 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6558 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6561 /* Returns an arbitrary interface within 'bundle'. */
6562 static struct ofport_dpif *
6563 ofbundle_get_a_port(const struct ofbundle *bundle)
6565 return CONTAINER_OF(list_front(&bundle->ports),
6566 struct ofport_dpif, bundle_node);
6570 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6572 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6576 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6578 struct ofproto_dpif *ofproto = ctx->ofproto;
6579 mirror_mask_t mirrors;
6580 struct ofbundle *in_bundle;
6583 const struct nlattr *a;
6586 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6587 ctx->packet != NULL, NULL);
6591 mirrors = in_bundle->src_mirrors;
6593 /* Drop frames on bundles reserved for mirroring. */
6594 if (in_bundle->mirror_out) {
6595 if (ctx->packet != NULL) {
6596 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6597 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6598 "%s, which is reserved exclusively for mirroring",
6599 ctx->ofproto->up.name, in_bundle->name);
6605 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6606 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6609 vlan = input_vid_to_vlan(in_bundle, vid);
6611 /* Look at the output ports to check for destination selections. */
6613 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6614 ctx->odp_actions->size) {
6615 enum ovs_action_attr type = nl_attr_type(a);
6616 struct ofport_dpif *ofport;
6618 if (type != OVS_ACTION_ATTR_OUTPUT) {
6622 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6623 if (ofport && ofport->bundle) {
6624 mirrors |= ofport->bundle->dst_mirrors;
6632 /* Restore the original packet before adding the mirror actions. */
6633 ctx->flow = *orig_flow;
6638 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6640 if (!vlan_is_mirrored(m, vlan)) {
6641 mirrors = zero_rightmost_1bit(mirrors);
6645 mirrors &= ~m->dup_mirrors;
6646 ctx->mirrors |= m->dup_mirrors;
6648 output_normal(ctx, m->out, vlan);
6649 } else if (vlan != m->out_vlan
6650 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6651 struct ofbundle *bundle;
6653 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6654 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6655 && !bundle->mirror_out) {
6656 output_normal(ctx, bundle, m->out_vlan);
6664 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6665 uint64_t packets, uint64_t bytes)
6671 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6674 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6677 /* In normal circumstances 'm' will not be NULL. However,
6678 * if mirrors are reconfigured, we can temporarily get out
6679 * of sync in facet_revalidate(). We could "correct" the
6680 * mirror list before reaching here, but doing that would
6681 * not properly account the traffic stats we've currently
6682 * accumulated for previous mirror configuration. */
6686 m->packet_count += packets;
6687 m->byte_count += bytes;
6691 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6692 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6693 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6695 is_gratuitous_arp(const struct flow *flow)
6697 return (flow->dl_type == htons(ETH_TYPE_ARP)
6698 && eth_addr_is_broadcast(flow->dl_dst)
6699 && (flow->nw_proto == ARP_OP_REPLY
6700 || (flow->nw_proto == ARP_OP_REQUEST
6701 && flow->nw_src == flow->nw_dst)));
6705 update_learning_table(struct ofproto_dpif *ofproto,
6706 const struct flow *flow, int vlan,
6707 struct ofbundle *in_bundle)
6709 struct mac_entry *mac;
6711 /* Don't learn the OFPP_NONE port. */
6712 if (in_bundle == &ofpp_none_bundle) {
6716 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6720 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6721 if (is_gratuitous_arp(flow)) {
6722 /* We don't want to learn from gratuitous ARP packets that are
6723 * reflected back over bond slaves so we lock the learning table. */
6724 if (!in_bundle->bond) {
6725 mac_entry_set_grat_arp_lock(mac);
6726 } else if (mac_entry_is_grat_arp_locked(mac)) {
6731 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6732 /* The log messages here could actually be useful in debugging,
6733 * so keep the rate limit relatively high. */
6734 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6735 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6736 "on port %s in VLAN %d",
6737 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6738 in_bundle->name, vlan);
6740 mac->port.p = in_bundle;
6741 tag_set_add(&ofproto->backer->revalidate_set,
6742 mac_learning_changed(ofproto->ml, mac));
6746 static struct ofbundle *
6747 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6748 bool warn, struct ofport_dpif **in_ofportp)
6750 struct ofport_dpif *ofport;
6752 /* Find the port and bundle for the received packet. */
6753 ofport = get_ofp_port(ofproto, in_port);
6755 *in_ofportp = ofport;
6757 if (ofport && ofport->bundle) {
6758 return ofport->bundle;
6761 /* Special-case OFPP_NONE, which a controller may use as the ingress
6762 * port for traffic that it is sourcing. */
6763 if (in_port == OFPP_NONE) {
6764 return &ofpp_none_bundle;
6767 /* Odd. A few possible reasons here:
6769 * - We deleted a port but there are still a few packets queued up
6772 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6773 * we don't know about.
6775 * - The ofproto client didn't configure the port as part of a bundle.
6776 * This is particularly likely to happen if a packet was received on the
6777 * port after it was created, but before the client had a chance to
6778 * configure its bundle.
6781 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6783 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6784 "port %"PRIu16, ofproto->up.name, in_port);
6789 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6790 * dropped. Returns true if they may be forwarded, false if they should be
6793 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6794 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6796 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6797 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6798 * checked by input_vid_is_valid().
6800 * May also add tags to '*tags', although the current implementation only does
6801 * so in one special case.
6804 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6807 struct ofproto_dpif *ofproto = ctx->ofproto;
6808 struct flow *flow = &ctx->flow;
6809 struct ofbundle *in_bundle = in_port->bundle;
6811 /* Drop frames for reserved multicast addresses
6812 * only if forward_bpdu option is absent. */
6813 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6814 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6818 if (in_bundle->bond) {
6819 struct mac_entry *mac;
6821 switch (bond_check_admissibility(in_bundle->bond, in_port,
6822 flow->dl_dst, &ctx->tags)) {
6827 xlate_report(ctx, "bonding refused admissibility, dropping");
6830 case BV_DROP_IF_MOVED:
6831 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6832 if (mac && mac->port.p != in_bundle &&
6833 (!is_gratuitous_arp(flow)
6834 || mac_entry_is_grat_arp_locked(mac))) {
6835 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6847 xlate_normal(struct action_xlate_ctx *ctx)
6849 struct ofport_dpif *in_port;
6850 struct ofbundle *in_bundle;
6851 struct mac_entry *mac;
6855 ctx->has_normal = true;
6857 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6858 ctx->packet != NULL, &in_port);
6860 xlate_report(ctx, "no input bundle, dropping");
6864 /* Drop malformed frames. */
6865 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6866 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6867 if (ctx->packet != NULL) {
6868 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6869 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6870 "VLAN tag received on port %s",
6871 ctx->ofproto->up.name, in_bundle->name);
6873 xlate_report(ctx, "partial VLAN tag, dropping");
6877 /* Drop frames on bundles reserved for mirroring. */
6878 if (in_bundle->mirror_out) {
6879 if (ctx->packet != NULL) {
6880 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6881 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6882 "%s, which is reserved exclusively for mirroring",
6883 ctx->ofproto->up.name, in_bundle->name);
6885 xlate_report(ctx, "input port is mirror output port, dropping");
6890 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6891 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6892 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6895 vlan = input_vid_to_vlan(in_bundle, vid);
6897 /* Check other admissibility requirements. */
6898 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6902 /* Learn source MAC. */
6903 if (ctx->may_learn) {
6904 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6907 /* Determine output bundle. */
6908 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6911 if (mac->port.p != in_bundle) {
6912 xlate_report(ctx, "forwarding to learned port");
6913 output_normal(ctx, mac->port.p, vlan);
6915 xlate_report(ctx, "learned port is input port, dropping");
6918 struct ofbundle *bundle;
6920 xlate_report(ctx, "no learned MAC for destination, flooding");
6921 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6922 if (bundle != in_bundle
6923 && ofbundle_includes_vlan(bundle, vlan)
6924 && bundle->floodable
6925 && !bundle->mirror_out) {
6926 output_normal(ctx, bundle, vlan);
6929 ctx->nf_output_iface = NF_OUT_FLOOD;
6933 /* Optimized flow revalidation.
6935 * It's a difficult problem, in general, to tell which facets need to have
6936 * their actions recalculated whenever the OpenFlow flow table changes. We
6937 * don't try to solve that general problem: for most kinds of OpenFlow flow
6938 * table changes, we recalculate the actions for every facet. This is
6939 * relatively expensive, but it's good enough if the OpenFlow flow table
6940 * doesn't change very often.
6942 * However, we can expect one particular kind of OpenFlow flow table change to
6943 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6944 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6945 * table, we add a special case that applies to flow tables in which every rule
6946 * has the same form (that is, the same wildcards), except that the table is
6947 * also allowed to have a single "catch-all" flow that matches all packets. We
6948 * optimize this case by tagging all of the facets that resubmit into the table
6949 * and invalidating the same tag whenever a flow changes in that table. The
6950 * end result is that we revalidate just the facets that need it (and sometimes
6951 * a few more, but not all of the facets or even all of the facets that
6952 * resubmit to the table modified by MAC learning). */
6954 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6955 * into an OpenFlow table with the given 'basis'. */
6957 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6960 if (minimask_is_catchall(mask)) {
6963 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6964 return tag_create_deterministic(hash);
6968 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6969 * taggability of that table.
6971 * This function must be called after *each* change to a flow table. If you
6972 * skip calling it on some changes then the pointer comparisons at the end can
6973 * be invalid if you get unlucky. For example, if a flow removal causes a
6974 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6975 * different wildcards to be created with the same address, then this function
6976 * will incorrectly skip revalidation. */
6978 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6980 struct table_dpif *table = &ofproto->tables[table_id];
6981 const struct oftable *oftable = &ofproto->up.tables[table_id];
6982 struct cls_table *catchall, *other;
6983 struct cls_table *t;
6985 catchall = other = NULL;
6987 switch (hmap_count(&oftable->cls.tables)) {
6989 /* We could tag this OpenFlow table but it would make the logic a
6990 * little harder and it's a corner case that doesn't seem worth it
6996 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6997 if (cls_table_is_catchall(t)) {
6999 } else if (!other) {
7002 /* Indicate that we can't tag this by setting both tables to
7003 * NULL. (We know that 'catchall' is already NULL.) */
7010 /* Can't tag this table. */
7014 if (table->catchall_table != catchall || table->other_table != other) {
7015 table->catchall_table = catchall;
7016 table->other_table = other;
7017 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7021 /* Given 'rule' that has changed in some way (either it is a rule being
7022 * inserted, a rule being deleted, or a rule whose actions are being
7023 * modified), marks facets for revalidation to ensure that packets will be
7024 * forwarded correctly according to the new state of the flow table.
7026 * This function must be called after *each* change to a flow table. See
7027 * the comment on table_update_taggable() for more information. */
7029 rule_invalidate(const struct rule_dpif *rule)
7031 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7033 table_update_taggable(ofproto, rule->up.table_id);
7035 if (!ofproto->backer->need_revalidate) {
7036 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7038 if (table->other_table && rule->tag) {
7039 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7041 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7047 set_frag_handling(struct ofproto *ofproto_,
7048 enum ofp_config_flags frag_handling)
7050 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7051 if (frag_handling != OFPC_FRAG_REASM) {
7052 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7060 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7061 const struct flow *flow,
7062 const struct ofpact *ofpacts, size_t ofpacts_len)
7064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7065 struct odputil_keybuf keybuf;
7066 struct dpif_flow_stats stats;
7070 struct action_xlate_ctx ctx;
7071 uint64_t odp_actions_stub[1024 / 8];
7072 struct ofpbuf odp_actions;
7074 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7075 odp_flow_key_from_flow(&key, flow,
7076 ofp_port_to_odp_port(ofproto, flow->in_port));
7078 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7080 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7081 packet_get_tcp_flags(packet, flow), packet);
7082 ctx.resubmit_stats = &stats;
7084 ofpbuf_use_stub(&odp_actions,
7085 odp_actions_stub, sizeof odp_actions_stub);
7086 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7087 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7088 odp_actions.data, odp_actions.size, packet);
7089 ofpbuf_uninit(&odp_actions);
7097 set_netflow(struct ofproto *ofproto_,
7098 const struct netflow_options *netflow_options)
7100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7102 if (netflow_options) {
7103 if (!ofproto->netflow) {
7104 ofproto->netflow = netflow_create();
7106 return netflow_set_options(ofproto->netflow, netflow_options);
7108 netflow_destroy(ofproto->netflow);
7109 ofproto->netflow = NULL;
7115 get_netflow_ids(const struct ofproto *ofproto_,
7116 uint8_t *engine_type, uint8_t *engine_id)
7118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7120 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7124 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7126 if (!facet_is_controller_flow(facet) &&
7127 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7128 struct subfacet *subfacet;
7129 struct ofexpired expired;
7131 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7132 if (subfacet->path == SF_FAST_PATH) {
7133 struct dpif_flow_stats stats;
7135 subfacet_reinstall(subfacet, &stats);
7136 subfacet_update_stats(subfacet, &stats);
7140 expired.flow = facet->flow;
7141 expired.packet_count = facet->packet_count;
7142 expired.byte_count = facet->byte_count;
7143 expired.used = facet->used;
7144 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7149 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7151 struct facet *facet;
7153 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7154 send_active_timeout(ofproto, facet);
7158 static struct ofproto_dpif *
7159 ofproto_dpif_lookup(const char *name)
7161 struct ofproto_dpif *ofproto;
7163 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7164 hash_string(name, 0), &all_ofproto_dpifs) {
7165 if (!strcmp(ofproto->up.name, name)) {
7173 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7174 const char *argv[], void *aux OVS_UNUSED)
7176 struct ofproto_dpif *ofproto;
7179 ofproto = ofproto_dpif_lookup(argv[1]);
7181 unixctl_command_reply_error(conn, "no such bridge");
7184 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7186 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7187 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7191 unixctl_command_reply(conn, "table successfully flushed");
7195 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7196 const char *argv[], void *aux OVS_UNUSED)
7198 struct ds ds = DS_EMPTY_INITIALIZER;
7199 const struct ofproto_dpif *ofproto;
7200 const struct mac_entry *e;
7202 ofproto = ofproto_dpif_lookup(argv[1]);
7204 unixctl_command_reply_error(conn, "no such bridge");
7208 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7209 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7210 struct ofbundle *bundle = e->port.p;
7211 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7212 ofbundle_get_a_port(bundle)->odp_port,
7213 e->vlan, ETH_ADDR_ARGS(e->mac),
7214 mac_entry_age(ofproto->ml, e));
7216 unixctl_command_reply(conn, ds_cstr(&ds));
7221 struct action_xlate_ctx ctx;
7227 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7228 const struct rule_dpif *rule)
7230 ds_put_char_multiple(result, '\t', level);
7232 ds_put_cstr(result, "No match\n");
7236 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7237 table_id, ntohll(rule->up.flow_cookie));
7238 cls_rule_format(&rule->up.cr, result);
7239 ds_put_char(result, '\n');
7241 ds_put_char_multiple(result, '\t', level);
7242 ds_put_cstr(result, "OpenFlow ");
7243 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7244 ds_put_char(result, '\n');
7248 trace_format_flow(struct ds *result, int level, const char *title,
7249 struct trace_ctx *trace)
7251 ds_put_char_multiple(result, '\t', level);
7252 ds_put_format(result, "%s: ", title);
7253 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7254 ds_put_cstr(result, "unchanged");
7256 flow_format(result, &trace->ctx.flow);
7257 trace->flow = trace->ctx.flow;
7259 ds_put_char(result, '\n');
7263 trace_format_regs(struct ds *result, int level, const char *title,
7264 struct trace_ctx *trace)
7268 ds_put_char_multiple(result, '\t', level);
7269 ds_put_format(result, "%s:", title);
7270 for (i = 0; i < FLOW_N_REGS; i++) {
7271 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7273 ds_put_char(result, '\n');
7277 trace_format_odp(struct ds *result, int level, const char *title,
7278 struct trace_ctx *trace)
7280 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7282 ds_put_char_multiple(result, '\t', level);
7283 ds_put_format(result, "%s: ", title);
7284 format_odp_actions(result, odp_actions->data, odp_actions->size);
7285 ds_put_char(result, '\n');
7289 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7291 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7292 struct ds *result = trace->result;
7294 ds_put_char(result, '\n');
7295 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7296 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7297 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7298 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7302 trace_report(struct action_xlate_ctx *ctx, const char *s)
7304 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7305 struct ds *result = trace->result;
7307 ds_put_char_multiple(result, '\t', ctx->recurse);
7308 ds_put_cstr(result, s);
7309 ds_put_char(result, '\n');
7313 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7314 void *aux OVS_UNUSED)
7316 const char *dpname = argv[1];
7317 struct ofproto_dpif *ofproto;
7318 struct ofpbuf odp_key;
7319 struct ofpbuf *packet;
7320 ovs_be16 initial_tci;
7326 ofpbuf_init(&odp_key, 0);
7329 ofproto = ofproto_dpif_lookup(dpname);
7331 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7335 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7336 /* ofproto/trace dpname flow [-generate] */
7337 const char *flow_s = argv[2];
7338 const char *generate_s = argv[3];
7340 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7341 * flow. We guess which type it is based on whether 'flow_s' contains
7342 * an '(', since a datapath flow always contains '(') but an
7343 * OpenFlow-like flow should not (in fact it's allowed but I believe
7344 * that's not documented anywhere).
7346 * An alternative would be to try to parse 'flow_s' both ways, but then
7347 * it would be tricky giving a sensible error message. After all, do
7348 * you just say "syntax error" or do you present both error messages?
7349 * Both choices seem lousy. */
7350 if (strchr(flow_s, '(')) {
7353 /* Convert string to datapath key. */
7354 ofpbuf_init(&odp_key, 0);
7355 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7357 unixctl_command_reply_error(conn, "Bad flow syntax");
7361 /* XXX: Since we allow the user to specify an ofproto, it's
7362 * possible they will specify a different ofproto than the one the
7363 * port actually belongs too. Ideally we should simply remove the
7364 * ability to specify the ofproto. */
7365 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7366 odp_key.size, &flow, NULL, NULL, NULL,
7368 unixctl_command_reply_error(conn, "Invalid flow");
7374 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7376 unixctl_command_reply_error(conn, error_s);
7381 initial_tci = flow.vlan_tci;
7384 /* Generate a packet, if requested. */
7386 packet = ofpbuf_new(0);
7387 flow_compose(packet, &flow);
7389 } else if (argc == 7) {
7390 /* ofproto/trace dpname priority tun_id in_port mark packet */
7391 const char *priority_s = argv[2];
7392 const char *tun_id_s = argv[3];
7393 const char *in_port_s = argv[4];
7394 const char *mark_s = argv[5];
7395 const char *packet_s = argv[6];
7396 uint32_t in_port = atoi(in_port_s);
7397 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7398 uint32_t priority = atoi(priority_s);
7399 uint32_t mark = atoi(mark_s);
7402 msg = eth_from_hex(packet_s, &packet);
7404 unixctl_command_reply_error(conn, msg);
7408 ds_put_cstr(&result, "Packet: ");
7409 s = ofp_packet_to_string(packet->data, packet->size);
7410 ds_put_cstr(&result, s);
7413 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7414 flow.tunnel.tun_id = tun_id;
7415 initial_tci = flow.vlan_tci;
7417 unixctl_command_reply_error(conn, "Bad command syntax");
7421 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7422 unixctl_command_reply(conn, ds_cstr(&result));
7425 ds_destroy(&result);
7426 ofpbuf_delete(packet);
7427 ofpbuf_uninit(&odp_key);
7431 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7432 const struct ofpbuf *packet, ovs_be16 initial_tci,
7435 struct rule_dpif *rule;
7437 ds_put_cstr(ds, "Flow: ");
7438 flow_format(ds, flow);
7439 ds_put_char(ds, '\n');
7441 rule = rule_dpif_lookup(ofproto, flow);
7443 trace_format_rule(ds, 0, 0, rule);
7444 if (rule == ofproto->miss_rule) {
7445 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7446 } else if (rule == ofproto->no_packet_in_rule) {
7447 ds_put_cstr(ds, "\nNo match, packets dropped because "
7448 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7452 uint64_t odp_actions_stub[1024 / 8];
7453 struct ofpbuf odp_actions;
7455 struct trace_ctx trace;
7458 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7461 ofpbuf_use_stub(&odp_actions,
7462 odp_actions_stub, sizeof odp_actions_stub);
7463 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7464 rule, tcp_flags, packet);
7465 trace.ctx.resubmit_hook = trace_resubmit;
7466 trace.ctx.report_hook = trace_report;
7467 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7470 ds_put_char(ds, '\n');
7471 trace_format_flow(ds, 0, "Final flow", &trace);
7472 ds_put_cstr(ds, "Datapath actions: ");
7473 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7474 ofpbuf_uninit(&odp_actions);
7476 if (trace.ctx.slow) {
7477 enum slow_path_reason slow;
7479 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7480 "slow path because it:");
7481 for (slow = trace.ctx.slow; slow; ) {
7482 enum slow_path_reason bit = rightmost_1bit(slow);
7486 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7489 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7492 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7495 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7498 ds_put_cstr(ds, "\n\t (The datapath actions are "
7499 "incomplete--for complete actions, "
7500 "please supply a packet.)");
7503 case SLOW_CONTROLLER:
7504 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7505 "to the OpenFlow controller.");
7508 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7509 "than the datapath supports.");
7516 if (slow & ~SLOW_MATCH) {
7517 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7518 "the special slow-path processing.");
7525 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7526 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7529 unixctl_command_reply(conn, NULL);
7533 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7534 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7537 unixctl_command_reply(conn, NULL);
7540 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7541 * 'reply' describing the results. */
7543 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7545 struct facet *facet;
7549 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7550 if (!facet_check_consistency(facet)) {
7555 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7559 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7560 ofproto->up.name, errors);
7562 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7567 ofproto_dpif_self_check(struct unixctl_conn *conn,
7568 int argc, const char *argv[], void *aux OVS_UNUSED)
7570 struct ds reply = DS_EMPTY_INITIALIZER;
7571 struct ofproto_dpif *ofproto;
7574 ofproto = ofproto_dpif_lookup(argv[1]);
7576 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7577 "ofproto/list for help)");
7580 ofproto_dpif_self_check__(ofproto, &reply);
7582 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7583 ofproto_dpif_self_check__(ofproto, &reply);
7587 unixctl_command_reply(conn, ds_cstr(&reply));
7591 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7592 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7593 * to destroy 'ofproto_shash' and free the returned value. */
7594 static const struct shash_node **
7595 get_ofprotos(struct shash *ofproto_shash)
7597 const struct ofproto_dpif *ofproto;
7599 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7600 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7601 shash_add_nocopy(ofproto_shash, name, ofproto);
7604 return shash_sort(ofproto_shash);
7608 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7609 const char *argv[] OVS_UNUSED,
7610 void *aux OVS_UNUSED)
7612 struct ds ds = DS_EMPTY_INITIALIZER;
7613 struct shash ofproto_shash;
7614 const struct shash_node **sorted_ofprotos;
7617 shash_init(&ofproto_shash);
7618 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7619 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7620 const struct shash_node *node = sorted_ofprotos[i];
7621 ds_put_format(&ds, "%s\n", node->name);
7624 shash_destroy(&ofproto_shash);
7625 free(sorted_ofprotos);
7627 unixctl_command_reply(conn, ds_cstr(&ds));
7632 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7634 struct dpif_dp_stats s;
7635 const struct shash_node **ports;
7638 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7640 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7641 dpif_name(ofproto->backer->dpif));
7642 /* xxx It would be better to show bridge-specific stats instead
7643 * xxx of dp ones. */
7645 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7646 s.n_hit, s.n_missed, s.n_lost);
7647 ds_put_format(ds, "\tflows: %zu\n",
7648 hmap_count(&ofproto->subfacets));
7650 ports = shash_sort(&ofproto->up.port_by_name);
7651 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7652 const struct shash_node *node = ports[i];
7653 struct ofport *ofport = node->data;
7654 const char *name = netdev_get_name(ofport->netdev);
7655 const char *type = netdev_get_type(ofport->netdev);
7658 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7660 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7661 if (odp_port != OVSP_NONE) {
7662 ds_put_format(ds, "%"PRIu32":", odp_port);
7664 ds_put_cstr(ds, "none:");
7667 if (strcmp(type, "system")) {
7668 struct netdev *netdev;
7671 ds_put_format(ds, " (%s", type);
7673 error = netdev_open(name, type, &netdev);
7678 error = netdev_get_config(netdev, &config);
7680 const struct smap_node **nodes;
7683 nodes = smap_sort(&config);
7684 for (i = 0; i < smap_count(&config); i++) {
7685 const struct smap_node *node = nodes[i];
7686 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7687 node->key, node->value);
7691 smap_destroy(&config);
7693 netdev_close(netdev);
7695 ds_put_char(ds, ')');
7697 ds_put_char(ds, '\n');
7703 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7704 const char *argv[], void *aux OVS_UNUSED)
7706 struct ds ds = DS_EMPTY_INITIALIZER;
7707 const struct ofproto_dpif *ofproto;
7711 for (i = 1; i < argc; i++) {
7712 ofproto = ofproto_dpif_lookup(argv[i]);
7714 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7715 "for help)", argv[i]);
7716 unixctl_command_reply_error(conn, ds_cstr(&ds));
7719 show_dp_format(ofproto, &ds);
7722 struct shash ofproto_shash;
7723 const struct shash_node **sorted_ofprotos;
7726 shash_init(&ofproto_shash);
7727 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7728 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7729 const struct shash_node *node = sorted_ofprotos[i];
7730 show_dp_format(node->data, &ds);
7733 shash_destroy(&ofproto_shash);
7734 free(sorted_ofprotos);
7737 unixctl_command_reply(conn, ds_cstr(&ds));
7742 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7743 int argc OVS_UNUSED, const char *argv[],
7744 void *aux OVS_UNUSED)
7746 struct ds ds = DS_EMPTY_INITIALIZER;
7747 const struct ofproto_dpif *ofproto;
7748 struct subfacet *subfacet;
7750 ofproto = ofproto_dpif_lookup(argv[1]);
7752 unixctl_command_reply_error(conn, "no such bridge");
7756 update_stats(ofproto->backer);
7758 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7759 struct odputil_keybuf keybuf;
7762 subfacet_get_key(subfacet, &keybuf, &key);
7763 odp_flow_key_format(key.data, key.size, &ds);
7765 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7766 subfacet->dp_packet_count, subfacet->dp_byte_count);
7767 if (subfacet->used) {
7768 ds_put_format(&ds, "%.3fs",
7769 (time_msec() - subfacet->used) / 1000.0);
7771 ds_put_format(&ds, "never");
7773 if (subfacet->facet->tcp_flags) {
7774 ds_put_cstr(&ds, ", flags:");
7775 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7778 ds_put_cstr(&ds, ", actions:");
7779 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7780 ds_put_char(&ds, '\n');
7783 unixctl_command_reply(conn, ds_cstr(&ds));
7788 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7789 int argc OVS_UNUSED, const char *argv[],
7790 void *aux OVS_UNUSED)
7792 struct ds ds = DS_EMPTY_INITIALIZER;
7793 struct ofproto_dpif *ofproto;
7795 ofproto = ofproto_dpif_lookup(argv[1]);
7797 unixctl_command_reply_error(conn, "no such bridge");
7801 flush(&ofproto->up);
7803 unixctl_command_reply(conn, ds_cstr(&ds));
7808 ofproto_dpif_unixctl_init(void)
7810 static bool registered;
7816 unixctl_command_register(
7818 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7819 2, 6, ofproto_unixctl_trace, NULL);
7820 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7821 ofproto_unixctl_fdb_flush, NULL);
7822 unixctl_command_register("fdb/show", "bridge", 1, 1,
7823 ofproto_unixctl_fdb_show, NULL);
7824 unixctl_command_register("ofproto/clog", "", 0, 0,
7825 ofproto_dpif_clog, NULL);
7826 unixctl_command_register("ofproto/unclog", "", 0, 0,
7827 ofproto_dpif_unclog, NULL);
7828 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7829 ofproto_dpif_self_check, NULL);
7830 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7831 ofproto_unixctl_dpif_dump_dps, NULL);
7832 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7833 ofproto_unixctl_dpif_show, NULL);
7834 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7835 ofproto_unixctl_dpif_dump_flows, NULL);
7836 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7837 ofproto_unixctl_dpif_del_flows, NULL);
7840 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7842 * This is deprecated. It is only for compatibility with broken device drivers
7843 * in old versions of Linux that do not properly support VLANs when VLAN
7844 * devices are not used. When broken device drivers are no longer in
7845 * widespread use, we will delete these interfaces. */
7848 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7851 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7853 if (realdev_ofp_port == ofport->realdev_ofp_port
7854 && vid == ofport->vlandev_vid) {
7858 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7860 if (ofport->realdev_ofp_port) {
7863 if (realdev_ofp_port && ofport->bundle) {
7864 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7865 * themselves be part of a bundle. */
7866 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7869 ofport->realdev_ofp_port = realdev_ofp_port;
7870 ofport->vlandev_vid = vid;
7872 if (realdev_ofp_port) {
7873 vsp_add(ofport, realdev_ofp_port, vid);
7880 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7882 return hash_2words(realdev_ofp_port, vid);
7885 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7886 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7887 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7888 * it would return the port number of eth0.9.
7890 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7891 * function just returns its 'realdev_odp_port' argument. */
7893 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7894 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7896 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7897 uint16_t realdev_ofp_port;
7898 int vid = vlan_tci_to_vid(vlan_tci);
7899 const struct vlan_splinter *vsp;
7901 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7902 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7903 hash_realdev_vid(realdev_ofp_port, vid),
7904 &ofproto->realdev_vid_map) {
7905 if (vsp->realdev_ofp_port == realdev_ofp_port
7906 && vsp->vid == vid) {
7907 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7911 return realdev_odp_port;
7914 static struct vlan_splinter *
7915 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7917 struct vlan_splinter *vsp;
7919 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7920 &ofproto->vlandev_map) {
7921 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7929 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7930 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7931 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7932 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7933 * eth0 and store 9 in '*vid'.
7935 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7936 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7939 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7940 uint16_t vlandev_ofp_port, int *vid)
7942 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7943 const struct vlan_splinter *vsp;
7945 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7950 return vsp->realdev_ofp_port;
7956 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7957 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7958 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7959 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7960 * always the case unless VLAN splinters are enabled), returns false without
7961 * making any changes. */
7963 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7968 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7973 /* Cause the flow to be processed as if it came in on the real device with
7974 * the VLAN device's VLAN ID. */
7975 flow->in_port = realdev;
7976 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7981 vsp_remove(struct ofport_dpif *port)
7983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7984 struct vlan_splinter *vsp;
7986 vsp = vlandev_find(ofproto, port->up.ofp_port);
7988 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7989 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7992 port->realdev_ofp_port = 0;
7994 VLOG_ERR("missing vlan device record");
7999 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8003 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8004 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8005 == realdev_ofp_port)) {
8006 struct vlan_splinter *vsp;
8008 vsp = xmalloc(sizeof *vsp);
8009 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8010 hash_int(port->up.ofp_port, 0));
8011 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8012 hash_realdev_vid(realdev_ofp_port, vid));
8013 vsp->realdev_ofp_port = realdev_ofp_port;
8014 vsp->vlandev_ofp_port = port->up.ofp_port;
8017 port->realdev_ofp_port = realdev_ofp_port;
8019 VLOG_ERR("duplicate vlan device record");
8024 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8026 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8027 return ofport ? ofport->odp_port : OVSP_NONE;
8030 static struct ofport_dpif *
8031 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8033 struct ofport_dpif *port;
8035 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8036 hash_int(odp_port, 0),
8037 &backer->odp_to_ofport_map) {
8038 if (port->odp_port == odp_port) {
8047 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8049 struct ofport_dpif *port;
8051 port = odp_port_to_ofport(ofproto->backer, odp_port);
8052 if (port && &ofproto->up == port->up.ofproto) {
8053 return port->up.ofp_port;
8059 const struct ofproto_class ofproto_dpif_class = {
8094 port_is_lacp_current,
8095 NULL, /* rule_choose_table */
8102 rule_modify_actions,
8111 get_cfm_remote_mpids,
8116 get_stp_port_status,
8123 is_mirror_output_bundle,
8124 forward_bpdu_changed,
8125 set_mac_table_config,