2 * Copyright (c) 2009, 2010, 2011, 2012 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"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
53 #include "unaligned.h"
55 #include "vlan-bitmap.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
60 COVERAGE_DEFINE(ofproto_dpif_expired);
61 COVERAGE_DEFINE(ofproto_dpif_xlate);
62 COVERAGE_DEFINE(facet_changed_rule);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 64
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
113 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
114 const struct flow *flow);
116 static void rule_credit_stats(struct rule_dpif *,
117 const struct dpif_flow_stats *);
118 static void flow_push_stats(struct rule_dpif *, const struct flow *,
119 const struct dpif_flow_stats *);
120 static tag_type rule_calculate_tag(const struct flow *,
121 const struct flow_wildcards *,
123 static void rule_invalidate(const struct rule_dpif *);
125 #define MAX_MIRRORS 32
126 typedef uint32_t mirror_mask_t;
127 #define MIRROR_MASK_C(X) UINT32_C(X)
128 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 size_t idx; /* In ofproto's "mirrors" array. */
132 void *aux; /* Key supplied by ofproto's client. */
133 char *name; /* Identifier for log messages. */
135 /* Selection criteria. */
136 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
137 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
138 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
140 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
141 struct ofbundle *out; /* Output port or NULL. */
142 int out_vlan; /* Output VLAN or -1. */
143 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
146 int64_t packet_count; /* Number of packets sent. */
147 int64_t byte_count; /* Number of bytes sent. */
150 static void mirror_destroy(struct ofmirror *);
151 static void update_mirror_stats(struct ofproto_dpif *ofproto,
152 mirror_mask_t mirrors,
153 uint64_t packets, uint64_t bytes);
156 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
157 struct ofproto_dpif *ofproto; /* Owning ofproto. */
158 void *aux; /* Key supplied by ofproto's client. */
159 char *name; /* Identifier for log messages. */
162 struct list ports; /* Contains "struct ofport"s. */
163 enum port_vlan_mode vlan_mode; /* VLAN mode */
164 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
165 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
166 * NULL if all VLANs are trunked. */
167 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
168 struct bond *bond; /* Nonnull iff more than one port. */
169 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
172 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
174 /* Port mirroring info. */
175 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
176 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
177 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
180 static void bundle_remove(struct ofport *);
181 static void bundle_update(struct ofbundle *);
182 static void bundle_destroy(struct ofbundle *);
183 static void bundle_del_port(struct ofport_dpif *);
184 static void bundle_run(struct ofbundle *);
185 static void bundle_wait(struct ofbundle *);
186 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
187 uint16_t in_port, bool warn,
188 struct ofport_dpif **in_ofportp);
190 /* A controller may use OFPP_NONE as the ingress port to indicate that
191 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
192 * when an input bundle is needed for validation (e.g., mirroring or
193 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
194 * any 'port' structs, so care must be taken when dealing with it. */
195 static struct ofbundle ofpp_none_bundle = {
197 .vlan_mode = PORT_VLAN_TRUNK
200 static void stp_run(struct ofproto_dpif *ofproto);
201 static void stp_wait(struct ofproto_dpif *ofproto);
202 static int set_stp_port(struct ofport *,
203 const struct ofproto_port_stp_settings *);
205 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
207 struct action_xlate_ctx {
208 /* action_xlate_ctx_init() initializes these members. */
211 struct ofproto_dpif *ofproto;
213 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
214 * this flow when actions change header fields. */
217 /* The packet corresponding to 'flow', or a null pointer if we are
218 * revalidating without a packet to refer to. */
219 const struct ofpbuf *packet;
221 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
222 * actions update the flow table?
224 * We want to update these tables if we are actually processing a packet,
225 * or if we are accounting for packets that the datapath has processed, but
226 * not if we are just revalidating. */
229 /* The rule that we are currently translating, or NULL. */
230 struct rule_dpif *rule;
232 /* Union of the set of TCP flags seen so far in this flow. (Used only by
233 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
237 /* If nonnull, flow translation calls this function just before executing a
238 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
239 * when the recursion depth is exceeded.
241 * 'rule' is the rule being submitted into. It will be null if the
242 * resubmit or OFPP_TABLE action didn't find a matching rule.
244 * This is normally null so the client has to set it manually after
245 * calling action_xlate_ctx_init(). */
246 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
248 /* If nonnull, flow translation calls this function to report some
249 * significant decision, e.g. to explain why OFPP_NORMAL translation
250 * dropped a packet. */
251 void (*report_hook)(struct action_xlate_ctx *, const char *s);
253 /* If nonnull, flow translation credits the specified statistics to each
254 * rule reached through a resubmit or OFPP_TABLE action.
256 * This is normally null so the client has to set it manually after
257 * calling action_xlate_ctx_init(). */
258 const struct dpif_flow_stats *resubmit_stats;
260 /* xlate_actions() initializes and uses these members. The client might want
261 * to look at them after it returns. */
263 struct ofpbuf *odp_actions; /* Datapath actions. */
264 tag_type tags; /* Tags associated with actions. */
265 enum slow_path_reason slow; /* 0 if fast path may be used. */
266 bool has_learn; /* Actions include NXAST_LEARN? */
267 bool has_normal; /* Actions output to OFPP_NORMAL? */
268 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
269 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
270 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
272 /* xlate_actions() initializes and uses these members, but the client has no
273 * reason to look at them. */
275 int recurse; /* Recursion level, via xlate_table_action. */
276 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
277 struct flow base_flow; /* Flow at the last commit. */
278 uint32_t orig_skb_priority; /* Priority when packet arrived. */
279 uint8_t table_id; /* OpenFlow table ID where flow was found. */
280 uint32_t sflow_n_outputs; /* Number of output ports. */
281 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
282 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
283 bool exit; /* No further actions should be processed. */
284 struct flow orig_flow; /* Copy of original flow. */
287 static void action_xlate_ctx_init(struct action_xlate_ctx *,
288 struct ofproto_dpif *, const struct flow *,
289 ovs_be16 initial_tci, struct rule_dpif *,
290 uint8_t tcp_flags, const struct ofpbuf *);
291 static void xlate_actions(struct action_xlate_ctx *,
292 const struct ofpact *ofpacts, size_t ofpacts_len,
293 struct ofpbuf *odp_actions);
294 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
295 const struct ofpact *ofpacts,
298 static size_t put_userspace_action(const struct ofproto_dpif *,
299 struct ofpbuf *odp_actions,
301 const union user_action_cookie *);
303 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
304 enum slow_path_reason,
305 uint64_t *stub, size_t stub_size,
306 const struct nlattr **actionsp,
307 size_t *actions_lenp);
309 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
311 /* A subfacet (see "struct subfacet" below) has three possible installation
314 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
315 * case just after the subfacet is created, just before the subfacet is
316 * destroyed, or if the datapath returns an error when we try to install a
319 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
321 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
322 * ofproto_dpif is installed in the datapath.
325 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
326 SF_FAST_PATH, /* Full actions are installed. */
327 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
330 static const char *subfacet_path_to_string(enum subfacet_path);
332 /* A dpif flow and actions associated with a facet.
334 * See also the large comment on struct facet. */
337 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
338 struct list list_node; /* In struct facet's 'facets' list. */
339 struct facet *facet; /* Owning facet. */
343 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
344 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
345 * regenerate the ODP flow key from ->facet->flow. */
346 enum odp_key_fitness key_fitness;
350 long long int used; /* Time last used; time created if not used. */
352 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
353 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
357 * These should be essentially identical for every subfacet in a facet, but
358 * may differ in trivial ways due to VLAN splinters. */
359 size_t actions_len; /* Number of bytes in actions[]. */
360 struct nlattr *actions; /* Datapath actions. */
362 enum slow_path_reason slow; /* 0 if fast path may be used. */
363 enum subfacet_path path; /* Installed in datapath? */
365 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
366 * splinters can cause it to differ. This value should be removed when
367 * the VLAN splinters feature is no longer needed. */
368 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
372 const struct nlattr *key,
373 size_t key_len, ovs_be16 initial_tci,
375 static struct subfacet *subfacet_find(struct ofproto_dpif *,
376 const struct nlattr *key, size_t key_len);
377 static void subfacet_destroy(struct subfacet *);
378 static void subfacet_destroy__(struct subfacet *);
379 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
381 static void subfacet_reset_dp_stats(struct subfacet *,
382 struct dpif_flow_stats *);
383 static void subfacet_update_time(struct subfacet *, long long int used);
384 static void subfacet_update_stats(struct subfacet *,
385 const struct dpif_flow_stats *);
386 static void subfacet_make_actions(struct subfacet *,
387 const struct ofpbuf *packet,
388 struct ofpbuf *odp_actions);
389 static int subfacet_install(struct subfacet *,
390 const struct nlattr *actions, size_t actions_len,
391 struct dpif_flow_stats *, enum slow_path_reason);
392 static void subfacet_uninstall(struct subfacet *);
394 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
396 /* An exact-match instantiation of an OpenFlow flow.
398 * A facet associates a "struct flow", which represents the Open vSwitch
399 * userspace idea of an exact-match flow, with one or more subfacets. Each
400 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
401 * the facet. When the kernel module (or other dpif implementation) and Open
402 * vSwitch userspace agree on the definition of a flow key, there is exactly
403 * one subfacet per facet. If the dpif implementation supports more-specific
404 * flow matching than userspace, however, a facet can have more than one
405 * subfacet, each of which corresponds to some distinction in flow that
406 * userspace simply doesn't understand.
408 * Flow expiration works in terms of subfacets, so a facet must have at least
409 * one subfacet or it will never expire, leaking memory. */
412 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
413 struct list list_node; /* In owning rule's 'facets' list. */
414 struct rule_dpif *rule; /* Owning rule. */
417 struct list subfacets;
418 long long int used; /* Time last used; time created if not used. */
425 * - Do include packets and bytes sent "by hand", e.g. with
428 * - Do include packets and bytes that were obtained from the datapath
429 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
430 * DPIF_FP_ZERO_STATS).
432 * - Do not include packets or bytes that can be obtained from the
433 * datapath for any existing subfacet.
435 uint64_t packet_count; /* Number of packets received. */
436 uint64_t byte_count; /* Number of bytes received. */
438 /* Resubmit statistics. */
439 uint64_t prev_packet_count; /* Number of packets from last stats push. */
440 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
441 long long int prev_used; /* Used time from last stats push. */
444 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
445 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
446 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
448 /* Properties of datapath actions.
450 * Every subfacet has its own actions because actions can differ slightly
451 * between splintered and non-splintered subfacets due to the VLAN tag
452 * being initially different (present vs. absent). All of them have these
453 * properties in common so we just store one copy of them here. */
454 bool has_learn; /* Actions include NXAST_LEARN? */
455 bool has_normal; /* Actions output to OFPP_NORMAL? */
456 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
457 tag_type tags; /* Tags that would require revalidation. */
458 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
460 /* Storage for a single subfacet, to reduce malloc() time and space
461 * overhead. (A facet always has at least one subfacet and in the common
462 * case has exactly one subfacet.) */
463 struct subfacet one_subfacet;
466 static struct facet *facet_create(struct rule_dpif *,
467 const struct flow *, uint32_t hash);
468 static void facet_remove(struct facet *);
469 static void facet_free(struct facet *);
471 static struct facet *facet_find(struct ofproto_dpif *,
472 const struct flow *, uint32_t hash);
473 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
474 const struct flow *, uint32_t hash);
475 static void facet_revalidate(struct facet *);
476 static bool facet_check_consistency(struct facet *);
478 static void facet_flush_stats(struct facet *);
480 static void facet_update_time(struct facet *, long long int used);
481 static void facet_reset_counters(struct facet *);
482 static void facet_push_stats(struct facet *);
483 static void facet_learn(struct facet *);
484 static void facet_account(struct facet *);
486 static bool facet_is_controller_flow(struct facet *);
492 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
493 struct list bundle_node; /* In struct ofbundle's "ports" list. */
494 struct cfm *cfm; /* Connectivity Fault Management, if any. */
495 tag_type tag; /* Tag associated with this port. */
496 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
497 bool may_enable; /* May be enabled in bonds. */
498 long long int carrier_seq; /* Carrier status changes. */
501 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
502 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
503 long long int stp_state_entered;
505 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
507 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
509 * This is deprecated. It is only for compatibility with broken device
510 * drivers in old versions of Linux that do not properly support VLANs when
511 * VLAN devices are not used. When broken device drivers are no longer in
512 * widespread use, we will delete these interfaces. */
513 uint16_t realdev_ofp_port;
517 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
518 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
519 * traffic egressing the 'ofport' with that priority should be marked with. */
520 struct priority_to_dscp {
521 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
522 uint32_t priority; /* Priority of this queue (see struct flow). */
524 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
527 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
529 * This is deprecated. It is only for compatibility with broken device drivers
530 * in old versions of Linux that do not properly support VLANs when VLAN
531 * devices are not used. When broken device drivers are no longer in
532 * widespread use, we will delete these interfaces. */
533 struct vlan_splinter {
534 struct hmap_node realdev_vid_node;
535 struct hmap_node vlandev_node;
536 uint16_t realdev_ofp_port;
537 uint16_t vlandev_ofp_port;
541 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
542 uint32_t realdev, ovs_be16 vlan_tci);
543 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
544 static void vsp_remove(struct ofport_dpif *);
545 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
547 static struct ofport_dpif *
548 ofport_dpif_cast(const struct ofport *ofport)
550 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
551 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
554 static void port_run(struct ofport_dpif *);
555 static void port_run_fast(struct ofport_dpif *);
556 static void port_wait(struct ofport_dpif *);
557 static int set_cfm(struct ofport *, const struct cfm_settings *);
558 static void ofport_clear_priorities(struct ofport_dpif *);
560 struct dpif_completion {
561 struct list list_node;
562 struct ofoperation *op;
565 /* Extra information about a classifier table.
566 * Currently used just for optimized flow revalidation. */
568 /* If either of these is nonnull, then this table has a form that allows
569 * flows to be tagged to avoid revalidating most flows for the most common
570 * kinds of flow table changes. */
571 struct cls_table *catchall_table; /* Table that wildcards all fields. */
572 struct cls_table *other_table; /* Table with any other wildcard set. */
573 uint32_t basis; /* Keeps each table's tags separate. */
576 /* Reasons that we might need to revalidate every facet, and corresponding
579 * A value of 0 means that there is no need to revalidate.
581 * It would be nice to have some cleaner way to integrate with coverage
582 * counters, but with only a few reasons I guess this is good enough for
584 enum revalidate_reason {
585 REV_RECONFIGURE = 1, /* Switch configuration changed. */
586 REV_STP, /* Spanning tree protocol port status change. */
587 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
588 REV_FLOW_TABLE, /* Flow table changed. */
589 REV_INCONSISTENCY /* Facet self-check failed. */
591 COVERAGE_DEFINE(rev_reconfigure);
592 COVERAGE_DEFINE(rev_stp);
593 COVERAGE_DEFINE(rev_port_toggled);
594 COVERAGE_DEFINE(rev_flow_table);
595 COVERAGE_DEFINE(rev_inconsistency);
597 struct ofproto_dpif {
598 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
602 /* Special OpenFlow rules. */
603 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
604 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
610 struct netflow *netflow;
611 struct dpif_sflow *sflow;
612 struct hmap bundles; /* Contains "struct ofbundle"s. */
613 struct mac_learning *ml;
614 struct ofmirror *mirrors[MAX_MIRRORS];
616 bool has_bonded_bundles;
619 struct timer next_expiration;
623 struct hmap subfacets;
624 struct governor *governor;
627 struct table_dpif tables[N_TABLES];
628 enum revalidate_reason need_revalidate;
629 struct tag_set revalidate_set;
631 /* Support for debugging async flow mods. */
632 struct list completions;
634 bool has_bundle_action; /* True when the first bundle action appears. */
635 struct netdev_stats stats; /* To account packets generated and consumed in
640 long long int stp_last_tick;
642 /* VLAN splinters. */
643 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
644 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
647 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
648 * for debugging the asynchronous flow_mod implementation.) */
651 /* All existing ofproto_dpif instances, indexed by ->up.name. */
652 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
654 static void ofproto_dpif_unixctl_init(void);
656 static struct ofproto_dpif *
657 ofproto_dpif_cast(const struct ofproto *ofproto)
659 assert(ofproto->ofproto_class == &ofproto_dpif_class);
660 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
663 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
665 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
667 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
668 const struct ofpbuf *, ovs_be16 initial_tci,
671 /* Packet processing. */
672 static void update_learning_table(struct ofproto_dpif *,
673 const struct flow *, int vlan,
676 #define FLOW_MISS_MAX_BATCH 50
677 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
679 /* Flow expiration. */
680 static int expire(struct ofproto_dpif *);
683 static void send_netflow_active_timeouts(struct ofproto_dpif *);
686 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
687 static size_t compose_sflow_action(const struct ofproto_dpif *,
688 struct ofpbuf *odp_actions,
689 const struct flow *, uint32_t odp_port);
690 static void add_mirror_actions(struct action_xlate_ctx *ctx,
691 const struct flow *flow);
692 /* Global variables. */
693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
695 /* Factory functions. */
698 enumerate_types(struct sset *types)
700 dp_enumerate_types(types);
704 enumerate_names(const char *type, struct sset *names)
706 return dp_enumerate_names(type, names);
710 del(const char *type, const char *name)
715 error = dpif_open(name, type, &dpif);
717 error = dpif_delete(dpif);
723 /* Basic life-cycle. */
725 static int add_internal_flows(struct ofproto_dpif *);
727 static struct ofproto *
730 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
735 dealloc(struct ofproto *ofproto_)
737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
742 construct(struct ofproto *ofproto_)
744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
745 const char *name = ofproto->up.name;
750 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
752 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
756 max_ports = dpif_get_max_ports(ofproto->dpif);
757 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
759 ofproto->n_matches = 0;
761 dpif_flow_flush(ofproto->dpif);
762 dpif_recv_purge(ofproto->dpif);
764 error = dpif_recv_set(ofproto->dpif, true);
766 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
767 dpif_close(ofproto->dpif);
771 ofproto->netflow = NULL;
772 ofproto->sflow = NULL;
774 hmap_init(&ofproto->bundles);
775 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
776 for (i = 0; i < MAX_MIRRORS; i++) {
777 ofproto->mirrors[i] = NULL;
779 ofproto->has_bonded_bundles = false;
781 timer_set_duration(&ofproto->next_expiration, 1000);
783 hmap_init(&ofproto->facets);
784 hmap_init(&ofproto->subfacets);
785 ofproto->governor = NULL;
787 for (i = 0; i < N_TABLES; i++) {
788 struct table_dpif *table = &ofproto->tables[i];
790 table->catchall_table = NULL;
791 table->other_table = NULL;
792 table->basis = random_uint32();
794 ofproto->need_revalidate = 0;
795 tag_set_init(&ofproto->revalidate_set);
797 list_init(&ofproto->completions);
799 ofproto_dpif_unixctl_init();
801 ofproto->has_mirrors = false;
802 ofproto->has_bundle_action = false;
804 hmap_init(&ofproto->vlandev_map);
805 hmap_init(&ofproto->realdev_vid_map);
807 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
808 hash_string(ofproto->up.name, 0));
809 memset(&ofproto->stats, 0, sizeof ofproto->stats);
811 ofproto_init_tables(ofproto_, N_TABLES);
812 error = add_internal_flows(ofproto);
813 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
819 add_internal_flow(struct ofproto_dpif *ofproto, int id,
820 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
822 struct ofputil_flow_mod fm;
825 match_init_catchall(&fm.match);
827 match_set_reg(&fm.match, 0, id);
828 fm.new_cookie = htonll(0);
829 fm.cookie = htonll(0);
830 fm.cookie_mask = htonll(0);
831 fm.table_id = TBL_INTERNAL;
832 fm.command = OFPFC_ADD;
838 fm.ofpacts = ofpacts->data;
839 fm.ofpacts_len = ofpacts->size;
841 error = ofproto_flow_mod(&ofproto->up, &fm);
843 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
844 id, ofperr_to_string(error));
848 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
849 assert(*rulep != NULL);
855 add_internal_flows(struct ofproto_dpif *ofproto)
857 struct ofpact_controller *controller;
858 uint64_t ofpacts_stub[128 / 8];
859 struct ofpbuf ofpacts;
863 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
866 controller = ofpact_put_CONTROLLER(&ofpacts);
867 controller->max_len = UINT16_MAX;
868 controller->controller_id = 0;
869 controller->reason = OFPR_NO_MATCH;
870 ofpact_pad(&ofpacts);
872 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
877 ofpbuf_clear(&ofpacts);
878 error = add_internal_flow(ofproto, id++, &ofpacts,
879 &ofproto->no_packet_in_rule);
884 complete_operations(struct ofproto_dpif *ofproto)
886 struct dpif_completion *c, *next;
888 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
889 ofoperation_complete(c->op, 0);
890 list_remove(&c->list_node);
896 destruct(struct ofproto *ofproto_)
898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
899 struct rule_dpif *rule, *next_rule;
900 struct oftable *table;
903 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
904 complete_operations(ofproto);
906 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
907 struct cls_cursor cursor;
909 cls_cursor_init(&cursor, &table->cls, NULL);
910 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
911 ofproto_rule_destroy(&rule->up);
915 for (i = 0; i < MAX_MIRRORS; i++) {
916 mirror_destroy(ofproto->mirrors[i]);
919 netflow_destroy(ofproto->netflow);
920 dpif_sflow_destroy(ofproto->sflow);
921 hmap_destroy(&ofproto->bundles);
922 mac_learning_destroy(ofproto->ml);
924 hmap_destroy(&ofproto->facets);
925 hmap_destroy(&ofproto->subfacets);
926 governor_destroy(ofproto->governor);
928 hmap_destroy(&ofproto->vlandev_map);
929 hmap_destroy(&ofproto->realdev_vid_map);
931 dpif_close(ofproto->dpif);
935 run_fast(struct ofproto *ofproto_)
937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
938 struct ofport_dpif *ofport;
941 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
942 port_run_fast(ofport);
945 /* Handle one or more batches of upcalls, until there's nothing left to do
946 * or until we do a fixed total amount of work.
948 * We do work in batches because it can be much cheaper to set up a number
949 * of flows and fire off their patches all at once. We do multiple batches
950 * because in some cases handling a packet can cause another packet to be
951 * queued almost immediately as part of the return flow. Both
952 * optimizations can make major improvements on some benchmarks and
953 * presumably for real traffic as well. */
955 while (work < FLOW_MISS_MAX_BATCH) {
956 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
966 run(struct ofproto *ofproto_)
968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
969 struct ofport_dpif *ofport;
970 struct ofbundle *bundle;
974 complete_operations(ofproto);
976 dpif_run(ofproto->dpif);
978 error = run_fast(ofproto_);
983 if (timer_expired(&ofproto->next_expiration)) {
984 int delay = expire(ofproto);
985 timer_set_duration(&ofproto->next_expiration, delay);
988 if (ofproto->netflow) {
989 if (netflow_run(ofproto->netflow)) {
990 send_netflow_active_timeouts(ofproto);
993 if (ofproto->sflow) {
994 dpif_sflow_run(ofproto->sflow);
997 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1000 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1005 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1007 /* Now revalidate if there's anything to do. */
1008 if (ofproto->need_revalidate
1009 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1010 struct tag_set revalidate_set = ofproto->revalidate_set;
1011 bool revalidate_all = ofproto->need_revalidate;
1012 struct facet *facet;
1014 switch (ofproto->need_revalidate) {
1015 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1016 case REV_STP: COVERAGE_INC(rev_stp); break;
1017 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1018 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1019 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1022 /* Clear the revalidation flags. */
1023 tag_set_init(&ofproto->revalidate_set);
1024 ofproto->need_revalidate = 0;
1026 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1028 || tag_set_intersects(&revalidate_set, facet->tags)) {
1029 facet_revalidate(facet);
1034 /* Check the consistency of a random facet, to aid debugging. */
1035 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1036 struct facet *facet;
1038 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1039 struct facet, hmap_node);
1040 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1041 if (!facet_check_consistency(facet)) {
1042 ofproto->need_revalidate = REV_INCONSISTENCY;
1047 if (ofproto->governor) {
1050 governor_run(ofproto->governor);
1052 /* If the governor has shrunk to its minimum size and the number of
1053 * subfacets has dwindled, then drop the governor entirely.
1055 * For hysteresis, the number of subfacets to drop the governor is
1056 * smaller than the number needed to trigger its creation. */
1057 n_subfacets = hmap_count(&ofproto->subfacets);
1058 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1059 && governor_is_idle(ofproto->governor)) {
1060 governor_destroy(ofproto->governor);
1061 ofproto->governor = NULL;
1069 wait(struct ofproto *ofproto_)
1071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1072 struct ofport_dpif *ofport;
1073 struct ofbundle *bundle;
1075 if (!clogged && !list_is_empty(&ofproto->completions)) {
1076 poll_immediate_wake();
1079 dpif_wait(ofproto->dpif);
1080 dpif_recv_wait(ofproto->dpif);
1081 if (ofproto->sflow) {
1082 dpif_sflow_wait(ofproto->sflow);
1084 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1085 poll_immediate_wake();
1087 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1090 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1091 bundle_wait(bundle);
1093 if (ofproto->netflow) {
1094 netflow_wait(ofproto->netflow);
1096 mac_learning_wait(ofproto->ml);
1098 if (ofproto->need_revalidate) {
1099 /* Shouldn't happen, but if it does just go around again. */
1100 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1101 poll_immediate_wake();
1103 timer_wait(&ofproto->next_expiration);
1105 if (ofproto->governor) {
1106 governor_wait(ofproto->governor);
1111 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1113 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1115 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1116 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1120 flush(struct ofproto *ofproto_)
1122 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1123 struct facet *facet, *next_facet;
1125 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1126 /* Mark the facet as not installed so that facet_remove() doesn't
1127 * bother trying to uninstall it. There is no point in uninstalling it
1128 * individually since we are about to blow away all the facets with
1129 * dpif_flow_flush(). */
1130 struct subfacet *subfacet;
1132 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1133 subfacet->path = SF_NOT_INSTALLED;
1134 subfacet->dp_packet_count = 0;
1135 subfacet->dp_byte_count = 0;
1137 facet_remove(facet);
1139 dpif_flow_flush(ofproto->dpif);
1143 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1144 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1146 *arp_match_ip = true;
1147 *actions = (OFPUTIL_A_OUTPUT |
1148 OFPUTIL_A_SET_VLAN_VID |
1149 OFPUTIL_A_SET_VLAN_PCP |
1150 OFPUTIL_A_STRIP_VLAN |
1151 OFPUTIL_A_SET_DL_SRC |
1152 OFPUTIL_A_SET_DL_DST |
1153 OFPUTIL_A_SET_NW_SRC |
1154 OFPUTIL_A_SET_NW_DST |
1155 OFPUTIL_A_SET_NW_TOS |
1156 OFPUTIL_A_SET_TP_SRC |
1157 OFPUTIL_A_SET_TP_DST |
1162 get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots)
1164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1165 struct dpif_dp_stats s;
1167 strcpy(ots->name, "classifier");
1169 dpif_get_dp_stats(ofproto->dpif, &s);
1170 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1171 put_32aligned_be64(&ots->matched_count,
1172 htonll(s.n_hit + ofproto->n_matches));
1175 static struct ofport *
1178 struct ofport_dpif *port = xmalloc(sizeof *port);
1183 port_dealloc(struct ofport *port_)
1185 struct ofport_dpif *port = ofport_dpif_cast(port_);
1190 port_construct(struct ofport *port_)
1192 struct ofport_dpif *port = ofport_dpif_cast(port_);
1193 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1195 ofproto->need_revalidate = REV_RECONFIGURE;
1196 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1197 port->bundle = NULL;
1199 port->tag = tag_create_random();
1200 port->may_enable = true;
1201 port->stp_port = NULL;
1202 port->stp_state = STP_DISABLED;
1203 hmap_init(&port->priorities);
1204 port->realdev_ofp_port = 0;
1205 port->vlandev_vid = 0;
1206 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1208 if (ofproto->sflow) {
1209 dpif_sflow_add_port(ofproto->sflow, port_);
1216 port_destruct(struct ofport *port_)
1218 struct ofport_dpif *port = ofport_dpif_cast(port_);
1219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1221 ofproto->need_revalidate = REV_RECONFIGURE;
1222 bundle_remove(port_);
1223 set_cfm(port_, NULL);
1224 if (ofproto->sflow) {
1225 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1228 ofport_clear_priorities(port);
1229 hmap_destroy(&port->priorities);
1233 port_modified(struct ofport *port_)
1235 struct ofport_dpif *port = ofport_dpif_cast(port_);
1237 if (port->bundle && port->bundle->bond) {
1238 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1243 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1245 struct ofport_dpif *port = ofport_dpif_cast(port_);
1246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1247 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1249 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1250 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1251 OFPUTIL_PC_NO_PACKET_IN)) {
1252 ofproto->need_revalidate = REV_RECONFIGURE;
1254 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1255 bundle_update(port->bundle);
1261 set_sflow(struct ofproto *ofproto_,
1262 const struct ofproto_sflow_options *sflow_options)
1264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1265 struct dpif_sflow *ds = ofproto->sflow;
1267 if (sflow_options) {
1269 struct ofport_dpif *ofport;
1271 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1272 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1273 dpif_sflow_add_port(ds, &ofport->up);
1275 ofproto->need_revalidate = REV_RECONFIGURE;
1277 dpif_sflow_set_options(ds, sflow_options);
1280 dpif_sflow_destroy(ds);
1281 ofproto->need_revalidate = REV_RECONFIGURE;
1282 ofproto->sflow = NULL;
1289 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1291 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1298 struct ofproto_dpif *ofproto;
1300 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1301 ofproto->need_revalidate = REV_RECONFIGURE;
1302 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1305 if (cfm_configure(ofport->cfm, s)) {
1311 cfm_destroy(ofport->cfm);
1317 get_cfm_fault(const struct ofport *ofport_)
1319 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1321 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1325 get_cfm_opup(const struct ofport *ofport_)
1327 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1329 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1333 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1336 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1339 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1347 get_cfm_health(const struct ofport *ofport_)
1349 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1351 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1354 /* Spanning Tree. */
1357 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1359 struct ofproto_dpif *ofproto = ofproto_;
1360 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1361 struct ofport_dpif *ofport;
1363 ofport = stp_port_get_aux(sp);
1365 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1366 ofproto->up.name, port_num);
1368 struct eth_header *eth = pkt->l2;
1370 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1371 if (eth_addr_is_zero(eth->eth_src)) {
1372 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1373 "with unknown MAC", ofproto->up.name, port_num);
1375 send_packet(ofport, pkt);
1381 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1383 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1385 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1387 /* Only revalidate flows if the configuration changed. */
1388 if (!s != !ofproto->stp) {
1389 ofproto->need_revalidate = REV_RECONFIGURE;
1393 if (!ofproto->stp) {
1394 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1395 send_bpdu_cb, ofproto);
1396 ofproto->stp_last_tick = time_msec();
1399 stp_set_bridge_id(ofproto->stp, s->system_id);
1400 stp_set_bridge_priority(ofproto->stp, s->priority);
1401 stp_set_hello_time(ofproto->stp, s->hello_time);
1402 stp_set_max_age(ofproto->stp, s->max_age);
1403 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1405 struct ofport *ofport;
1407 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1408 set_stp_port(ofport, NULL);
1411 stp_destroy(ofproto->stp);
1412 ofproto->stp = NULL;
1419 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1421 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1425 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1426 s->designated_root = stp_get_designated_root(ofproto->stp);
1427 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1436 update_stp_port_state(struct ofport_dpif *ofport)
1438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1439 enum stp_state state;
1441 /* Figure out new state. */
1442 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1446 if (ofport->stp_state != state) {
1447 enum ofputil_port_state of_state;
1450 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1451 netdev_get_name(ofport->up.netdev),
1452 stp_state_name(ofport->stp_state),
1453 stp_state_name(state));
1454 if (stp_learn_in_state(ofport->stp_state)
1455 != stp_learn_in_state(state)) {
1456 /* xxx Learning action flows should also be flushed. */
1457 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1459 fwd_change = stp_forward_in_state(ofport->stp_state)
1460 != stp_forward_in_state(state);
1462 ofproto->need_revalidate = REV_STP;
1463 ofport->stp_state = state;
1464 ofport->stp_state_entered = time_msec();
1466 if (fwd_change && ofport->bundle) {
1467 bundle_update(ofport->bundle);
1470 /* Update the STP state bits in the OpenFlow port description. */
1471 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1472 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1473 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1474 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1475 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1477 ofproto_port_set_state(&ofport->up, of_state);
1481 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1482 * caller is responsible for assigning STP port numbers and ensuring
1483 * there are no duplicates. */
1485 set_stp_port(struct ofport *ofport_,
1486 const struct ofproto_port_stp_settings *s)
1488 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1489 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1490 struct stp_port *sp = ofport->stp_port;
1492 if (!s || !s->enable) {
1494 ofport->stp_port = NULL;
1495 stp_port_disable(sp);
1496 update_stp_port_state(ofport);
1499 } else if (sp && stp_port_no(sp) != s->port_num
1500 && ofport == stp_port_get_aux(sp)) {
1501 /* The port-id changed, so disable the old one if it's not
1502 * already in use by another port. */
1503 stp_port_disable(sp);
1506 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1507 stp_port_enable(sp);
1509 stp_port_set_aux(sp, ofport);
1510 stp_port_set_priority(sp, s->priority);
1511 stp_port_set_path_cost(sp, s->path_cost);
1513 update_stp_port_state(ofport);
1519 get_stp_port_status(struct ofport *ofport_,
1520 struct ofproto_port_stp_status *s)
1522 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1524 struct stp_port *sp = ofport->stp_port;
1526 if (!ofproto->stp || !sp) {
1532 s->port_id = stp_port_get_id(sp);
1533 s->state = stp_port_get_state(sp);
1534 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1535 s->role = stp_port_get_role(sp);
1536 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1542 stp_run(struct ofproto_dpif *ofproto)
1545 long long int now = time_msec();
1546 long long int elapsed = now - ofproto->stp_last_tick;
1547 struct stp_port *sp;
1550 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1551 ofproto->stp_last_tick = now;
1553 while (stp_get_changed_port(ofproto->stp, &sp)) {
1554 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1557 update_stp_port_state(ofport);
1561 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1562 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1568 stp_wait(struct ofproto_dpif *ofproto)
1571 poll_timer_wait(1000);
1575 /* Returns true if STP should process 'flow'. */
1577 stp_should_process_flow(const struct flow *flow)
1579 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1583 stp_process_packet(const struct ofport_dpif *ofport,
1584 const struct ofpbuf *packet)
1586 struct ofpbuf payload = *packet;
1587 struct eth_header *eth = payload.data;
1588 struct stp_port *sp = ofport->stp_port;
1590 /* Sink packets on ports that have STP disabled when the bridge has
1592 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1596 /* Trim off padding on payload. */
1597 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1598 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1601 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1602 stp_received_bpdu(sp, payload.data, payload.size);
1606 static struct priority_to_dscp *
1607 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1609 struct priority_to_dscp *pdscp;
1612 hash = hash_int(priority, 0);
1613 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1614 if (pdscp->priority == priority) {
1622 ofport_clear_priorities(struct ofport_dpif *ofport)
1624 struct priority_to_dscp *pdscp, *next;
1626 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1627 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1633 set_queues(struct ofport *ofport_,
1634 const struct ofproto_port_queue *qdscp_list,
1637 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1639 struct hmap new = HMAP_INITIALIZER(&new);
1642 for (i = 0; i < n_qdscp; i++) {
1643 struct priority_to_dscp *pdscp;
1647 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1648 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1653 pdscp = get_priority(ofport, priority);
1655 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1657 pdscp = xmalloc(sizeof *pdscp);
1658 pdscp->priority = priority;
1660 ofproto->need_revalidate = REV_RECONFIGURE;
1663 if (pdscp->dscp != dscp) {
1665 ofproto->need_revalidate = REV_RECONFIGURE;
1668 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1671 if (!hmap_is_empty(&ofport->priorities)) {
1672 ofport_clear_priorities(ofport);
1673 ofproto->need_revalidate = REV_RECONFIGURE;
1676 hmap_swap(&new, &ofport->priorities);
1684 /* Expires all MAC learning entries associated with 'bundle' and forces its
1685 * ofproto to revalidate every flow.
1687 * Normally MAC learning entries are removed only from the ofproto associated
1688 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1689 * are removed from every ofproto. When patch ports and SLB bonds are in use
1690 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1691 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1692 * with the host from which it migrated. */
1694 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1696 struct ofproto_dpif *ofproto = bundle->ofproto;
1697 struct mac_learning *ml = ofproto->ml;
1698 struct mac_entry *mac, *next_mac;
1700 ofproto->need_revalidate = REV_RECONFIGURE;
1701 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1702 if (mac->port.p == bundle) {
1704 struct ofproto_dpif *o;
1706 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1708 struct mac_entry *e;
1710 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1713 tag_set_add(&o->revalidate_set, e->tag);
1714 mac_learning_expire(o->ml, e);
1720 mac_learning_expire(ml, mac);
1725 static struct ofbundle *
1726 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1728 struct ofbundle *bundle;
1730 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1731 &ofproto->bundles) {
1732 if (bundle->aux == aux) {
1739 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1740 * ones that are found to 'bundles'. */
1742 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1743 void **auxes, size_t n_auxes,
1744 struct hmapx *bundles)
1748 hmapx_init(bundles);
1749 for (i = 0; i < n_auxes; i++) {
1750 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1752 hmapx_add(bundles, bundle);
1758 bundle_update(struct ofbundle *bundle)
1760 struct ofport_dpif *port;
1762 bundle->floodable = true;
1763 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1764 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1765 || !stp_forward_in_state(port->stp_state)) {
1766 bundle->floodable = false;
1773 bundle_del_port(struct ofport_dpif *port)
1775 struct ofbundle *bundle = port->bundle;
1777 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1779 list_remove(&port->bundle_node);
1780 port->bundle = NULL;
1783 lacp_slave_unregister(bundle->lacp, port);
1786 bond_slave_unregister(bundle->bond, port);
1789 bundle_update(bundle);
1793 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1794 struct lacp_slave_settings *lacp,
1795 uint32_t bond_stable_id)
1797 struct ofport_dpif *port;
1799 port = get_ofp_port(bundle->ofproto, ofp_port);
1804 if (port->bundle != bundle) {
1805 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1807 bundle_del_port(port);
1810 port->bundle = bundle;
1811 list_push_back(&bundle->ports, &port->bundle_node);
1812 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1813 || !stp_forward_in_state(port->stp_state)) {
1814 bundle->floodable = false;
1818 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1819 lacp_slave_register(bundle->lacp, port, lacp);
1822 port->bond_stable_id = bond_stable_id;
1828 bundle_destroy(struct ofbundle *bundle)
1830 struct ofproto_dpif *ofproto;
1831 struct ofport_dpif *port, *next_port;
1838 ofproto = bundle->ofproto;
1839 for (i = 0; i < MAX_MIRRORS; i++) {
1840 struct ofmirror *m = ofproto->mirrors[i];
1842 if (m->out == bundle) {
1844 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1845 || hmapx_find_and_delete(&m->dsts, bundle)) {
1846 ofproto->need_revalidate = REV_RECONFIGURE;
1851 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1852 bundle_del_port(port);
1855 bundle_flush_macs(bundle, true);
1856 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1858 free(bundle->trunks);
1859 lacp_destroy(bundle->lacp);
1860 bond_destroy(bundle->bond);
1865 bundle_set(struct ofproto *ofproto_, void *aux,
1866 const struct ofproto_bundle_settings *s)
1868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1869 bool need_flush = false;
1870 struct ofport_dpif *port;
1871 struct ofbundle *bundle;
1872 unsigned long *trunks;
1878 bundle_destroy(bundle_lookup(ofproto, aux));
1882 assert(s->n_slaves == 1 || s->bond != NULL);
1883 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1885 bundle = bundle_lookup(ofproto, aux);
1887 bundle = xmalloc(sizeof *bundle);
1889 bundle->ofproto = ofproto;
1890 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1891 hash_pointer(aux, 0));
1893 bundle->name = NULL;
1895 list_init(&bundle->ports);
1896 bundle->vlan_mode = PORT_VLAN_TRUNK;
1898 bundle->trunks = NULL;
1899 bundle->use_priority_tags = s->use_priority_tags;
1900 bundle->lacp = NULL;
1901 bundle->bond = NULL;
1903 bundle->floodable = true;
1905 bundle->src_mirrors = 0;
1906 bundle->dst_mirrors = 0;
1907 bundle->mirror_out = 0;
1910 if (!bundle->name || strcmp(s->name, bundle->name)) {
1912 bundle->name = xstrdup(s->name);
1917 if (!bundle->lacp) {
1918 ofproto->need_revalidate = REV_RECONFIGURE;
1919 bundle->lacp = lacp_create();
1921 lacp_configure(bundle->lacp, s->lacp);
1923 lacp_destroy(bundle->lacp);
1924 bundle->lacp = NULL;
1927 /* Update set of ports. */
1929 for (i = 0; i < s->n_slaves; i++) {
1930 if (!bundle_add_port(bundle, s->slaves[i],
1931 s->lacp ? &s->lacp_slaves[i] : NULL,
1932 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1936 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1937 struct ofport_dpif *next_port;
1939 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1940 for (i = 0; i < s->n_slaves; i++) {
1941 if (s->slaves[i] == port->up.ofp_port) {
1946 bundle_del_port(port);
1950 assert(list_size(&bundle->ports) <= s->n_slaves);
1952 if (list_is_empty(&bundle->ports)) {
1953 bundle_destroy(bundle);
1957 /* Set VLAN tagging mode */
1958 if (s->vlan_mode != bundle->vlan_mode
1959 || s->use_priority_tags != bundle->use_priority_tags) {
1960 bundle->vlan_mode = s->vlan_mode;
1961 bundle->use_priority_tags = s->use_priority_tags;
1966 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1967 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1969 if (vlan != bundle->vlan) {
1970 bundle->vlan = vlan;
1974 /* Get trunked VLANs. */
1975 switch (s->vlan_mode) {
1976 case PORT_VLAN_ACCESS:
1980 case PORT_VLAN_TRUNK:
1981 trunks = CONST_CAST(unsigned long *, s->trunks);
1984 case PORT_VLAN_NATIVE_UNTAGGED:
1985 case PORT_VLAN_NATIVE_TAGGED:
1986 if (vlan != 0 && (!s->trunks
1987 || !bitmap_is_set(s->trunks, vlan)
1988 || bitmap_is_set(s->trunks, 0))) {
1989 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1991 trunks = bitmap_clone(s->trunks, 4096);
1993 trunks = bitmap_allocate1(4096);
1995 bitmap_set1(trunks, vlan);
1996 bitmap_set0(trunks, 0);
1998 trunks = CONST_CAST(unsigned long *, s->trunks);
2005 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2006 free(bundle->trunks);
2007 if (trunks == s->trunks) {
2008 bundle->trunks = vlan_bitmap_clone(trunks);
2010 bundle->trunks = trunks;
2015 if (trunks != s->trunks) {
2020 if (!list_is_short(&bundle->ports)) {
2021 bundle->ofproto->has_bonded_bundles = true;
2023 if (bond_reconfigure(bundle->bond, s->bond)) {
2024 ofproto->need_revalidate = REV_RECONFIGURE;
2027 bundle->bond = bond_create(s->bond);
2028 ofproto->need_revalidate = REV_RECONFIGURE;
2031 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2032 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2036 bond_destroy(bundle->bond);
2037 bundle->bond = NULL;
2040 /* If we changed something that would affect MAC learning, un-learn
2041 * everything on this port and force flow revalidation. */
2043 bundle_flush_macs(bundle, false);
2050 bundle_remove(struct ofport *port_)
2052 struct ofport_dpif *port = ofport_dpif_cast(port_);
2053 struct ofbundle *bundle = port->bundle;
2056 bundle_del_port(port);
2057 if (list_is_empty(&bundle->ports)) {
2058 bundle_destroy(bundle);
2059 } else if (list_is_short(&bundle->ports)) {
2060 bond_destroy(bundle->bond);
2061 bundle->bond = NULL;
2067 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2069 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2070 struct ofport_dpif *port = port_;
2071 uint8_t ea[ETH_ADDR_LEN];
2074 error = netdev_get_etheraddr(port->up.netdev, ea);
2076 struct ofpbuf packet;
2079 ofpbuf_init(&packet, 0);
2080 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2082 memcpy(packet_pdu, pdu, pdu_size);
2084 send_packet(port, &packet);
2085 ofpbuf_uninit(&packet);
2087 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2088 "%s (%s)", port->bundle->name,
2089 netdev_get_name(port->up.netdev), strerror(error));
2094 bundle_send_learning_packets(struct ofbundle *bundle)
2096 struct ofproto_dpif *ofproto = bundle->ofproto;
2097 int error, n_packets, n_errors;
2098 struct mac_entry *e;
2100 error = n_packets = n_errors = 0;
2101 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2102 if (e->port.p != bundle) {
2103 struct ofpbuf *learning_packet;
2104 struct ofport_dpif *port;
2108 /* The assignment to "port" is unnecessary but makes "grep"ing for
2109 * struct ofport_dpif more effective. */
2110 learning_packet = bond_compose_learning_packet(bundle->bond,
2114 ret = send_packet(port, learning_packet);
2115 ofpbuf_delete(learning_packet);
2125 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2126 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2127 "packets, last error was: %s",
2128 bundle->name, n_errors, n_packets, strerror(error));
2130 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2131 bundle->name, n_packets);
2136 bundle_run(struct ofbundle *bundle)
2139 lacp_run(bundle->lacp, send_pdu_cb);
2142 struct ofport_dpif *port;
2144 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2145 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2148 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2149 lacp_status(bundle->lacp));
2150 if (bond_should_send_learning_packets(bundle->bond)) {
2151 bundle_send_learning_packets(bundle);
2157 bundle_wait(struct ofbundle *bundle)
2160 lacp_wait(bundle->lacp);
2163 bond_wait(bundle->bond);
2170 mirror_scan(struct ofproto_dpif *ofproto)
2174 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2175 if (!ofproto->mirrors[idx]) {
2182 static struct ofmirror *
2183 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2187 for (i = 0; i < MAX_MIRRORS; i++) {
2188 struct ofmirror *mirror = ofproto->mirrors[i];
2189 if (mirror && mirror->aux == aux) {
2197 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2199 mirror_update_dups(struct ofproto_dpif *ofproto)
2203 for (i = 0; i < MAX_MIRRORS; i++) {
2204 struct ofmirror *m = ofproto->mirrors[i];
2207 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2211 for (i = 0; i < MAX_MIRRORS; i++) {
2212 struct ofmirror *m1 = ofproto->mirrors[i];
2219 for (j = i + 1; j < MAX_MIRRORS; j++) {
2220 struct ofmirror *m2 = ofproto->mirrors[j];
2222 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2223 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2224 m2->dup_mirrors |= m1->dup_mirrors;
2231 mirror_set(struct ofproto *ofproto_, void *aux,
2232 const struct ofproto_mirror_settings *s)
2234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2235 mirror_mask_t mirror_bit;
2236 struct ofbundle *bundle;
2237 struct ofmirror *mirror;
2238 struct ofbundle *out;
2239 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2240 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2243 mirror = mirror_lookup(ofproto, aux);
2245 mirror_destroy(mirror);
2251 idx = mirror_scan(ofproto);
2253 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2255 ofproto->up.name, MAX_MIRRORS, s->name);
2259 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2260 mirror->ofproto = ofproto;
2263 mirror->out_vlan = -1;
2264 mirror->name = NULL;
2267 if (!mirror->name || strcmp(s->name, mirror->name)) {
2269 mirror->name = xstrdup(s->name);
2272 /* Get the new configuration. */
2273 if (s->out_bundle) {
2274 out = bundle_lookup(ofproto, s->out_bundle);
2276 mirror_destroy(mirror);
2282 out_vlan = s->out_vlan;
2284 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2285 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2287 /* If the configuration has not changed, do nothing. */
2288 if (hmapx_equals(&srcs, &mirror->srcs)
2289 && hmapx_equals(&dsts, &mirror->dsts)
2290 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2291 && mirror->out == out
2292 && mirror->out_vlan == out_vlan)
2294 hmapx_destroy(&srcs);
2295 hmapx_destroy(&dsts);
2299 hmapx_swap(&srcs, &mirror->srcs);
2300 hmapx_destroy(&srcs);
2302 hmapx_swap(&dsts, &mirror->dsts);
2303 hmapx_destroy(&dsts);
2305 free(mirror->vlans);
2306 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2309 mirror->out_vlan = out_vlan;
2311 /* Update bundles. */
2312 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2313 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2314 if (hmapx_contains(&mirror->srcs, bundle)) {
2315 bundle->src_mirrors |= mirror_bit;
2317 bundle->src_mirrors &= ~mirror_bit;
2320 if (hmapx_contains(&mirror->dsts, bundle)) {
2321 bundle->dst_mirrors |= mirror_bit;
2323 bundle->dst_mirrors &= ~mirror_bit;
2326 if (mirror->out == bundle) {
2327 bundle->mirror_out |= mirror_bit;
2329 bundle->mirror_out &= ~mirror_bit;
2333 ofproto->need_revalidate = REV_RECONFIGURE;
2334 ofproto->has_mirrors = true;
2335 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2336 mirror_update_dups(ofproto);
2342 mirror_destroy(struct ofmirror *mirror)
2344 struct ofproto_dpif *ofproto;
2345 mirror_mask_t mirror_bit;
2346 struct ofbundle *bundle;
2353 ofproto = mirror->ofproto;
2354 ofproto->need_revalidate = REV_RECONFIGURE;
2355 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2357 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2358 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2359 bundle->src_mirrors &= ~mirror_bit;
2360 bundle->dst_mirrors &= ~mirror_bit;
2361 bundle->mirror_out &= ~mirror_bit;
2364 hmapx_destroy(&mirror->srcs);
2365 hmapx_destroy(&mirror->dsts);
2366 free(mirror->vlans);
2368 ofproto->mirrors[mirror->idx] = NULL;
2372 mirror_update_dups(ofproto);
2374 ofproto->has_mirrors = false;
2375 for (i = 0; i < MAX_MIRRORS; i++) {
2376 if (ofproto->mirrors[i]) {
2377 ofproto->has_mirrors = true;
2384 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2385 uint64_t *packets, uint64_t *bytes)
2387 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2388 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2391 *packets = *bytes = UINT64_MAX;
2395 *packets = mirror->packet_count;
2396 *bytes = mirror->byte_count;
2402 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2404 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2405 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2406 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2412 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2415 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2416 return bundle && bundle->mirror_out != 0;
2420 forward_bpdu_changed(struct ofproto *ofproto_)
2422 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2423 ofproto->need_revalidate = REV_RECONFIGURE;
2427 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2430 mac_learning_set_idle_time(ofproto->ml, idle_time);
2435 static struct ofport_dpif *
2436 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2438 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2439 return ofport ? ofport_dpif_cast(ofport) : NULL;
2442 static struct ofport_dpif *
2443 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2445 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2449 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2450 struct dpif_port *dpif_port)
2452 ofproto_port->name = dpif_port->name;
2453 ofproto_port->type = dpif_port->type;
2454 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2458 port_run_fast(struct ofport_dpif *ofport)
2460 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2461 struct ofpbuf packet;
2463 ofpbuf_init(&packet, 0);
2464 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2465 send_packet(ofport, &packet);
2466 ofpbuf_uninit(&packet);
2471 port_run(struct ofport_dpif *ofport)
2473 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2474 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2475 bool enable = netdev_get_carrier(ofport->up.netdev);
2477 ofport->carrier_seq = carrier_seq;
2479 port_run_fast(ofport);
2481 int cfm_opup = cfm_get_opup(ofport->cfm);
2483 cfm_run(ofport->cfm);
2484 enable = enable && !cfm_get_fault(ofport->cfm);
2486 if (cfm_opup >= 0) {
2487 enable = enable && cfm_opup;
2491 if (ofport->bundle) {
2492 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2493 if (carrier_changed) {
2494 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2498 if (ofport->may_enable != enable) {
2499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2501 if (ofproto->has_bundle_action) {
2502 ofproto->need_revalidate = REV_PORT_TOGGLED;
2506 ofport->may_enable = enable;
2510 port_wait(struct ofport_dpif *ofport)
2513 cfm_wait(ofport->cfm);
2518 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2519 struct ofproto_port *ofproto_port)
2521 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2522 struct dpif_port dpif_port;
2525 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2527 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2533 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2536 uint16_t odp_port = UINT16_MAX;
2539 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2541 *ofp_portp = odp_port_to_ofp_port(odp_port);
2547 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2549 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2552 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2554 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2556 /* The caller is going to close ofport->up.netdev. If this is a
2557 * bonded port, then the bond is using that netdev, so remove it
2558 * from the bond. The client will need to reconfigure everything
2559 * after deleting ports, so then the slave will get re-added. */
2560 bundle_remove(&ofport->up);
2567 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2569 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2572 error = netdev_get_stats(ofport->up.netdev, stats);
2574 if (!error && ofport->odp_port == OVSP_LOCAL) {
2575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2577 /* ofproto->stats.tx_packets represents packets that we created
2578 * internally and sent to some port (e.g. packets sent with
2579 * send_packet()). Account for them as if they had come from
2580 * OFPP_LOCAL and got forwarded. */
2582 if (stats->rx_packets != UINT64_MAX) {
2583 stats->rx_packets += ofproto->stats.tx_packets;
2586 if (stats->rx_bytes != UINT64_MAX) {
2587 stats->rx_bytes += ofproto->stats.tx_bytes;
2590 /* ofproto->stats.rx_packets represents packets that were received on
2591 * some port and we processed internally and dropped (e.g. STP).
2592 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2594 if (stats->tx_packets != UINT64_MAX) {
2595 stats->tx_packets += ofproto->stats.rx_packets;
2598 if (stats->tx_bytes != UINT64_MAX) {
2599 stats->tx_bytes += ofproto->stats.rx_bytes;
2606 /* Account packets for LOCAL port. */
2608 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2609 size_t tx_size, size_t rx_size)
2611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2614 ofproto->stats.rx_packets++;
2615 ofproto->stats.rx_bytes += rx_size;
2618 ofproto->stats.tx_packets++;
2619 ofproto->stats.tx_bytes += tx_size;
2623 struct port_dump_state {
2624 struct dpif_port_dump dump;
2629 port_dump_start(const struct ofproto *ofproto_, void **statep)
2631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2632 struct port_dump_state *state;
2634 *statep = state = xmalloc(sizeof *state);
2635 dpif_port_dump_start(&state->dump, ofproto->dpif);
2636 state->done = false;
2641 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2642 struct ofproto_port *port)
2644 struct port_dump_state *state = state_;
2645 struct dpif_port dpif_port;
2647 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2648 ofproto_port_from_dpif_port(port, &dpif_port);
2651 int error = dpif_port_dump_done(&state->dump);
2653 return error ? error : EOF;
2658 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2660 struct port_dump_state *state = state_;
2663 dpif_port_dump_done(&state->dump);
2670 port_poll(const struct ofproto *ofproto_, char **devnamep)
2672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2673 return dpif_port_poll(ofproto->dpif, devnamep);
2677 port_poll_wait(const struct ofproto *ofproto_)
2679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2680 dpif_port_poll_wait(ofproto->dpif);
2684 port_is_lacp_current(const struct ofport *ofport_)
2686 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2687 return (ofport->bundle && ofport->bundle->lacp
2688 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2692 /* Upcall handling. */
2694 /* Flow miss batching.
2696 * Some dpifs implement operations faster when you hand them off in a batch.
2697 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2698 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2699 * more packets, plus possibly installing the flow in the dpif.
2701 * So far we only batch the operations that affect flow setup time the most.
2702 * It's possible to batch more than that, but the benefit might be minimal. */
2704 struct hmap_node hmap_node;
2706 enum odp_key_fitness key_fitness;
2707 const struct nlattr *key;
2709 ovs_be16 initial_tci;
2710 struct list packets;
2711 enum dpif_upcall_type upcall_type;
2714 struct flow_miss_op {
2715 struct dpif_op dpif_op;
2716 struct subfacet *subfacet; /* Subfacet */
2717 void *garbage; /* Pointer to pass to free(), NULL if none. */
2718 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2721 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2722 * OpenFlow controller as necessary according to their individual
2723 * configurations. */
2725 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2726 const struct flow *flow)
2728 struct ofputil_packet_in pin;
2730 pin.packet = packet->data;
2731 pin.packet_len = packet->size;
2732 pin.reason = OFPR_NO_MATCH;
2733 pin.controller_id = 0;
2738 pin.send_len = 0; /* not used for flow table misses */
2740 flow_get_metadata(flow, &pin.fmd);
2742 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2745 static enum slow_path_reason
2746 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2747 const struct ofpbuf *packet)
2749 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2755 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2757 cfm_process_heartbeat(ofport->cfm, packet);
2760 } else if (ofport->bundle && ofport->bundle->lacp
2761 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2763 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2766 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2768 stp_process_packet(ofport, packet);
2775 static struct flow_miss *
2776 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2778 struct flow_miss *miss;
2780 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2781 if (flow_equal(&miss->flow, flow)) {
2789 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2790 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2791 * 'miss' is associated with a subfacet the caller must also initialize the
2792 * returned op->subfacet, and if anything needs to be freed after processing
2793 * the op, the caller must initialize op->garbage also. */
2795 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2796 struct flow_miss_op *op)
2798 if (miss->flow.vlan_tci != miss->initial_tci) {
2799 /* This packet was received on a VLAN splinter port. We
2800 * added a VLAN to the packet to make the packet resemble
2801 * the flow, but the actions were composed assuming that
2802 * the packet contained no VLAN. So, we must remove the
2803 * VLAN header from the packet before trying to execute the
2805 eth_pop_vlan(packet);
2808 op->subfacet = NULL;
2810 op->dpif_op.type = DPIF_OP_EXECUTE;
2811 op->dpif_op.u.execute.key = miss->key;
2812 op->dpif_op.u.execute.key_len = miss->key_len;
2813 op->dpif_op.u.execute.packet = packet;
2816 /* Helper for handle_flow_miss_without_facet() and
2817 * handle_flow_miss_with_facet(). */
2819 handle_flow_miss_common(struct rule_dpif *rule,
2820 struct ofpbuf *packet, const struct flow *flow)
2822 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2824 ofproto->n_matches++;
2826 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2828 * Extra-special case for fail-open mode.
2830 * We are in fail-open mode and the packet matched the fail-open
2831 * rule, but we are connected to a controller too. We should send
2832 * the packet up to the controller in the hope that it will try to
2833 * set up a flow and thereby allow us to exit fail-open.
2835 * See the top-level comment in fail-open.c for more information.
2837 send_packet_in_miss(ofproto, packet, flow);
2841 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2842 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2843 * installing a datapath flow. The answer is usually "yes" (a return value of
2844 * true). However, for short flows the cost of bookkeeping is much higher than
2845 * the benefits, so when the datapath holds a large number of flows we impose
2846 * some heuristics to decide which flows are likely to be worth tracking. */
2848 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2849 struct flow_miss *miss, uint32_t hash)
2851 if (!ofproto->governor) {
2854 n_subfacets = hmap_count(&ofproto->subfacets);
2855 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2859 ofproto->governor = governor_create(ofproto->up.name);
2862 return governor_should_install_flow(ofproto->governor, hash,
2863 list_size(&miss->packets));
2866 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2867 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2868 * increment '*n_ops'. */
2870 handle_flow_miss_without_facet(struct flow_miss *miss,
2871 struct rule_dpif *rule,
2872 struct flow_miss_op *ops, size_t *n_ops)
2874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2875 long long int now = time_msec();
2876 struct action_xlate_ctx ctx;
2877 struct ofpbuf *packet;
2879 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2880 struct flow_miss_op *op = &ops[*n_ops];
2881 struct dpif_flow_stats stats;
2882 struct ofpbuf odp_actions;
2884 COVERAGE_INC(facet_suppress);
2886 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2888 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
2889 rule_credit_stats(rule, &stats);
2891 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2893 ctx.resubmit_stats = &stats;
2894 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
2897 if (odp_actions.size) {
2898 struct dpif_execute *execute = &op->dpif_op.u.execute;
2900 init_flow_miss_execute_op(miss, packet, op);
2901 execute->actions = odp_actions.data;
2902 execute->actions_len = odp_actions.size;
2903 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2907 ofpbuf_uninit(&odp_actions);
2912 /* Handles 'miss', which matches 'facet'. May add any required datapath
2913 * operations to 'ops', incrementing '*n_ops' for each new op.
2915 * All of the packets in 'miss' are considered to have arrived at time 'now'.
2916 * This is really important only for new facets: if we just called time_msec()
2917 * here, then the new subfacet or its packets could look (occasionally) as
2918 * though it was used some time after the facet was used. That can make a
2919 * one-packet flow look like it has a nonzero duration, which looks odd in
2920 * e.g. NetFlow statistics. */
2922 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2924 struct flow_miss_op *ops, size_t *n_ops)
2926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2927 enum subfacet_path want_path;
2928 struct subfacet *subfacet;
2929 struct ofpbuf *packet;
2931 subfacet = subfacet_create(facet,
2932 miss->key_fitness, miss->key, miss->key_len,
2933 miss->initial_tci, now);
2935 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2936 struct flow_miss_op *op = &ops[*n_ops];
2937 struct dpif_flow_stats stats;
2938 struct ofpbuf odp_actions;
2940 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2942 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2943 if (!subfacet->actions || subfacet->slow) {
2944 subfacet_make_actions(subfacet, packet, &odp_actions);
2947 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
2948 subfacet_update_stats(subfacet, &stats);
2950 if (subfacet->actions_len) {
2951 struct dpif_execute *execute = &op->dpif_op.u.execute;
2953 init_flow_miss_execute_op(miss, packet, op);
2954 op->subfacet = subfacet;
2955 if (!subfacet->slow) {
2956 execute->actions = subfacet->actions;
2957 execute->actions_len = subfacet->actions_len;
2958 ofpbuf_uninit(&odp_actions);
2960 execute->actions = odp_actions.data;
2961 execute->actions_len = odp_actions.size;
2962 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2967 ofpbuf_uninit(&odp_actions);
2971 want_path = subfacet_want_path(subfacet->slow);
2972 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2973 struct flow_miss_op *op = &ops[(*n_ops)++];
2974 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2976 op->subfacet = subfacet;
2978 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2979 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2980 put->key = miss->key;
2981 put->key_len = miss->key_len;
2982 if (want_path == SF_FAST_PATH) {
2983 put->actions = subfacet->actions;
2984 put->actions_len = subfacet->actions_len;
2986 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2987 op->stub, sizeof op->stub,
2988 &put->actions, &put->actions_len);
2994 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2995 * operations to 'ops', incrementing '*n_ops' for each new op. */
2997 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2998 struct flow_miss_op *ops, size_t *n_ops)
3000 struct facet *facet;
3004 /* The caller must ensure that miss->hmap_node.hash contains
3005 * flow_hash(miss->flow, 0). */
3006 hash = miss->hmap_node.hash;
3008 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3010 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3012 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3013 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3017 facet = facet_create(rule, &miss->flow, hash);
3022 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3025 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
3026 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
3027 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
3028 * what a flow key should contain.
3030 * This function also includes some logic to help make VLAN splinters
3031 * transparent to the rest of the upcall processing logic. In particular, if
3032 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
3033 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
3034 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
3036 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3037 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3038 * (This differs from the value returned in flow->vlan_tci only for packets
3039 * received on VLAN splinters.)
3041 static enum odp_key_fitness
3042 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
3043 const struct nlattr *key, size_t key_len,
3044 struct flow *flow, ovs_be16 *initial_tci,
3045 struct ofpbuf *packet)
3047 enum odp_key_fitness fitness;
3049 fitness = odp_flow_key_to_flow(key, key_len, flow);
3050 if (fitness == ODP_FIT_ERROR) {
3053 *initial_tci = flow->vlan_tci;
3055 if (vsp_adjust_flow(ofproto, flow)) {
3057 /* Make the packet resemble the flow, so that it gets sent to an
3058 * OpenFlow controller properly, so that it looks correct for
3059 * sFlow, and so that flow_extract() will get the correct vlan_tci
3060 * if it is called on 'packet'.
3062 * The allocated space inside 'packet' probably also contains
3063 * 'key', that is, both 'packet' and 'key' are probably part of a
3064 * struct dpif_upcall (see the large comment on that structure
3065 * definition), so pushing data on 'packet' is in general not a
3066 * good idea since it could overwrite 'key' or free it as a side
3067 * effect. However, it's OK in this special case because we know
3068 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3069 * will just overwrite the 4-byte "struct nlattr", which is fine
3070 * since we don't need that header anymore. */
3071 eth_push_vlan(packet, flow->vlan_tci);
3074 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3075 if (fitness == ODP_FIT_PERFECT) {
3076 fitness = ODP_FIT_TOO_MUCH;
3084 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3087 struct dpif_upcall *upcall;
3088 struct flow_miss *miss;
3089 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3090 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3091 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3101 /* Construct the to-do list.
3103 * This just amounts to extracting the flow from each packet and sticking
3104 * the packets that have the same flow in the same "flow_miss" structure so
3105 * that we can process them together. */
3108 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3109 struct flow_miss *miss = &misses[n_misses];
3110 struct flow_miss *existing_miss;
3113 /* Obtain metadata and check userspace/kernel agreement on flow match,
3114 * then set 'flow''s header pointers. */
3115 miss->key_fitness = ofproto_dpif_extract_flow_key(
3116 ofproto, upcall->key, upcall->key_len,
3117 &miss->flow, &miss->initial_tci, upcall->packet);
3118 if (miss->key_fitness == ODP_FIT_ERROR) {
3121 flow_extract(upcall->packet, miss->flow.skb_priority,
3122 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3124 /* Add other packets to a to-do list. */
3125 hash = flow_hash(&miss->flow, 0);
3126 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3127 if (!existing_miss) {
3128 hmap_insert(&todo, &miss->hmap_node, hash);
3129 miss->key = upcall->key;
3130 miss->key_len = upcall->key_len;
3131 miss->upcall_type = upcall->type;
3132 list_init(&miss->packets);
3136 miss = existing_miss;
3138 list_push_back(&miss->packets, &upcall->packet->list_node);
3141 /* Process each element in the to-do list, constructing the set of
3142 * operations to batch. */
3144 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3145 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3147 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3149 /* Execute batch. */
3150 for (i = 0; i < n_ops; i++) {
3151 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3153 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3155 /* Free memory and update facets. */
3156 for (i = 0; i < n_ops; i++) {
3157 struct flow_miss_op *op = &flow_miss_ops[i];
3159 switch (op->dpif_op.type) {
3160 case DPIF_OP_EXECUTE:
3163 case DPIF_OP_FLOW_PUT:
3164 if (!op->dpif_op.error) {
3165 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3169 case DPIF_OP_FLOW_DEL:
3175 hmap_destroy(&todo);
3178 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3179 classify_upcall(const struct dpif_upcall *upcall)
3181 union user_action_cookie cookie;
3183 /* First look at the upcall type. */
3184 switch (upcall->type) {
3185 case DPIF_UC_ACTION:
3191 case DPIF_N_UC_TYPES:
3193 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3197 /* "action" upcalls need a closer look. */
3198 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3199 switch (cookie.type) {
3200 case USER_ACTION_COOKIE_SFLOW:
3201 return SFLOW_UPCALL;
3203 case USER_ACTION_COOKIE_SLOW_PATH:
3206 case USER_ACTION_COOKIE_UNSPEC:
3208 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3214 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3215 const struct dpif_upcall *upcall)
3217 union user_action_cookie cookie;
3218 enum odp_key_fitness fitness;
3219 ovs_be16 initial_tci;
3222 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3223 upcall->key_len, &flow,
3224 &initial_tci, upcall->packet);
3225 if (fitness == ODP_FIT_ERROR) {
3229 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3230 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3234 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3236 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3237 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3238 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3243 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3246 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3247 struct dpif_upcall *upcall = &misses[n_misses];
3248 struct ofpbuf *buf = &miss_bufs[n_misses];
3251 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3252 sizeof miss_buf_stubs[n_misses]);
3253 error = dpif_recv(ofproto->dpif, upcall, buf);
3259 switch (classify_upcall(upcall)) {
3261 /* Handle it later. */
3266 if (ofproto->sflow) {
3267 handle_sflow_upcall(ofproto, upcall);
3278 /* Handle deferred MISS_UPCALL processing. */
3279 handle_miss_upcalls(ofproto, misses, n_misses);
3280 for (i = 0; i < n_misses; i++) {
3281 ofpbuf_uninit(&miss_bufs[i]);
3287 /* Flow expiration. */
3289 static int subfacet_max_idle(const struct ofproto_dpif *);
3290 static void update_stats(struct ofproto_dpif *);
3291 static void rule_expire(struct rule_dpif *);
3292 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3294 /* This function is called periodically by run(). Its job is to collect
3295 * updates for the flows that have been installed into the datapath, most
3296 * importantly when they last were used, and then use that information to
3297 * expire flows that have not been used recently.
3299 * Returns the number of milliseconds after which it should be called again. */
3301 expire(struct ofproto_dpif *ofproto)
3303 struct rule_dpif *rule, *next_rule;
3304 struct oftable *table;
3307 /* Update stats for each flow in the datapath. */
3308 update_stats(ofproto);
3310 /* Expire subfacets that have been idle too long. */
3311 dp_max_idle = subfacet_max_idle(ofproto);
3312 expire_subfacets(ofproto, dp_max_idle);
3314 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3315 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3316 struct cls_cursor cursor;
3318 cls_cursor_init(&cursor, &table->cls, NULL);
3319 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3324 /* All outstanding data in existing flows has been accounted, so it's a
3325 * good time to do bond rebalancing. */
3326 if (ofproto->has_bonded_bundles) {
3327 struct ofbundle *bundle;
3329 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3331 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3336 return MIN(dp_max_idle, 1000);
3339 /* Updates flow table statistics given that the datapath just reported 'stats'
3340 * as 'subfacet''s statistics. */
3342 update_subfacet_stats(struct subfacet *subfacet,
3343 const struct dpif_flow_stats *stats)
3345 struct facet *facet = subfacet->facet;
3347 if (stats->n_packets >= subfacet->dp_packet_count) {
3348 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3349 facet->packet_count += extra;
3351 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3354 if (stats->n_bytes >= subfacet->dp_byte_count) {
3355 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3357 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3360 subfacet->dp_packet_count = stats->n_packets;
3361 subfacet->dp_byte_count = stats->n_bytes;
3363 facet->tcp_flags |= stats->tcp_flags;
3365 subfacet_update_time(subfacet, stats->used);
3366 if (facet->accounted_bytes < facet->byte_count) {
3368 facet_account(facet);
3369 facet->accounted_bytes = facet->byte_count;
3371 facet_push_stats(facet);
3374 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3375 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3377 delete_unexpected_flow(struct dpif *dpif,
3378 const struct nlattr *key, size_t key_len)
3380 if (!VLOG_DROP_WARN(&rl)) {
3384 odp_flow_key_format(key, key_len, &s);
3385 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3389 COVERAGE_INC(facet_unexpected);
3390 dpif_flow_del(dpif, key, key_len, NULL);
3393 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3395 * This function also pushes statistics updates to rules which each facet
3396 * resubmits into. Generally these statistics will be accurate. However, if a
3397 * facet changes the rule it resubmits into at some time in between
3398 * update_stats() runs, it is possible that statistics accrued to the
3399 * old rule will be incorrectly attributed to the new rule. This could be
3400 * avoided by calling update_stats() whenever rules are created or
3401 * deleted. However, the performance impact of making so many calls to the
3402 * datapath do not justify the benefit of having perfectly accurate statistics.
3405 update_stats(struct ofproto_dpif *p)
3407 const struct dpif_flow_stats *stats;
3408 struct dpif_flow_dump dump;
3409 const struct nlattr *key;
3412 dpif_flow_dump_start(&dump, p->dpif);
3413 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3414 struct subfacet *subfacet;
3416 subfacet = subfacet_find(p, key, key_len);
3417 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3419 update_subfacet_stats(subfacet, stats);
3423 /* Stats are updated per-packet. */
3426 case SF_NOT_INSTALLED:
3428 delete_unexpected_flow(p->dpif, key, key_len);
3432 dpif_flow_dump_done(&dump);
3435 /* Calculates and returns the number of milliseconds of idle time after which
3436 * subfacets should expire from the datapath. When a subfacet expires, we fold
3437 * its statistics into its facet, and when a facet's last subfacet expires, we
3438 * fold its statistic into its rule. */
3440 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3443 * Idle time histogram.
3445 * Most of the time a switch has a relatively small number of subfacets.
3446 * When this is the case we might as well keep statistics for all of them
3447 * in userspace and to cache them in the kernel datapath for performance as
3450 * As the number of subfacets increases, the memory required to maintain
3451 * statistics about them in userspace and in the kernel becomes
3452 * significant. However, with a large number of subfacets it is likely
3453 * that only a few of them are "heavy hitters" that consume a large amount
3454 * of bandwidth. At this point, only heavy hitters are worth caching in
3455 * the kernel and maintaining in userspaces; other subfacets we can
3458 * The technique used to compute the idle time is to build a histogram with
3459 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3460 * that is installed in the kernel gets dropped in the appropriate bucket.
3461 * After the histogram has been built, we compute the cutoff so that only
3462 * the most-recently-used 1% of subfacets (but at least
3463 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3464 * the most-recently-used bucket of subfacets is kept, so actually an
3465 * arbitrary number of subfacets can be kept in any given expiration run
3466 * (though the next run will delete most of those unless they receive
3469 * This requires a second pass through the subfacets, in addition to the
3470 * pass made by update_stats(), because the former function never looks at
3471 * uninstallable subfacets.
3473 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3474 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3475 int buckets[N_BUCKETS] = { 0 };
3476 int total, subtotal, bucket;
3477 struct subfacet *subfacet;
3481 total = hmap_count(&ofproto->subfacets);
3482 if (total <= ofproto->up.flow_eviction_threshold) {
3483 return N_BUCKETS * BUCKET_WIDTH;
3486 /* Build histogram. */
3488 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3489 long long int idle = now - subfacet->used;
3490 int bucket = (idle <= 0 ? 0
3491 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3492 : (unsigned int) idle / BUCKET_WIDTH);
3496 /* Find the first bucket whose flows should be expired. */
3497 subtotal = bucket = 0;
3499 subtotal += buckets[bucket++];
3500 } while (bucket < N_BUCKETS &&
3501 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3503 if (VLOG_IS_DBG_ENABLED()) {
3507 ds_put_cstr(&s, "keep");
3508 for (i = 0; i < N_BUCKETS; i++) {
3510 ds_put_cstr(&s, ", drop");
3513 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3516 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3520 return bucket * BUCKET_WIDTH;
3523 enum { EXPIRE_MAX_BATCH = 50 };
3526 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3528 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3529 struct dpif_op ops[EXPIRE_MAX_BATCH];
3530 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3531 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3532 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3535 for (i = 0; i < n; i++) {
3536 ops[i].type = DPIF_OP_FLOW_DEL;
3537 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3538 ops[i].u.flow_del.key = keys[i].data;
3539 ops[i].u.flow_del.key_len = keys[i].size;
3540 ops[i].u.flow_del.stats = &stats[i];
3544 dpif_operate(ofproto->dpif, opsp, n);
3545 for (i = 0; i < n; i++) {
3546 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3547 subfacets[i]->path = SF_NOT_INSTALLED;
3548 subfacet_destroy(subfacets[i]);
3553 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3555 /* Cutoff time for most flows. */
3556 long long int normal_cutoff = time_msec() - dp_max_idle;
3558 /* We really want to keep flows for special protocols around, so use a more
3559 * conservative cutoff. */
3560 long long int special_cutoff = time_msec() - 10000;
3562 struct subfacet *subfacet, *next_subfacet;
3563 struct subfacet *batch[EXPIRE_MAX_BATCH];
3567 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3568 &ofproto->subfacets) {
3569 long long int cutoff;
3571 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3574 if (subfacet->used < cutoff) {
3575 if (subfacet->path != SF_NOT_INSTALLED) {
3576 batch[n_batch++] = subfacet;
3577 if (n_batch >= EXPIRE_MAX_BATCH) {
3578 expire_batch(ofproto, batch, n_batch);
3582 subfacet_destroy(subfacet);
3588 expire_batch(ofproto, batch, n_batch);
3592 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3593 * then delete it entirely. */
3595 rule_expire(struct rule_dpif *rule)
3597 struct facet *facet, *next_facet;
3601 if (rule->up.pending) {
3602 /* We'll have to expire it later. */
3606 /* Has 'rule' expired? */
3608 if (rule->up.hard_timeout
3609 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3610 reason = OFPRR_HARD_TIMEOUT;
3611 } else if (rule->up.idle_timeout
3612 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3613 reason = OFPRR_IDLE_TIMEOUT;
3618 COVERAGE_INC(ofproto_dpif_expired);
3620 /* Update stats. (This is a no-op if the rule expired due to an idle
3621 * timeout, because that only happens when the rule has no facets left.) */
3622 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3623 facet_remove(facet);
3626 /* Get rid of the rule. */
3627 ofproto_rule_expire(&rule->up, reason);
3632 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3634 * The caller must already have determined that no facet with an identical
3635 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3636 * the ofproto's classifier table.
3638 * 'hash' must be the return value of flow_hash(flow, 0).
3640 * The facet will initially have no subfacets. The caller should create (at
3641 * least) one subfacet with subfacet_create(). */
3642 static struct facet *
3643 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3646 struct facet *facet;
3648 facet = xzalloc(sizeof *facet);
3649 facet->used = time_msec();
3650 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3651 list_push_back(&rule->facets, &facet->list_node);
3653 facet->flow = *flow;
3654 list_init(&facet->subfacets);
3655 netflow_flow_init(&facet->nf_flow);
3656 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3662 facet_free(struct facet *facet)
3667 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3668 * 'packet', which arrived on 'in_port'.
3670 * Takes ownership of 'packet'. */
3672 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3673 const struct nlattr *odp_actions, size_t actions_len,
3674 struct ofpbuf *packet)
3676 struct odputil_keybuf keybuf;
3680 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3681 odp_flow_key_from_flow(&key, flow);
3683 error = dpif_execute(ofproto->dpif, key.data, key.size,
3684 odp_actions, actions_len, packet);
3686 ofpbuf_delete(packet);
3690 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3692 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3693 * rule's statistics, via subfacet_uninstall().
3695 * - Removes 'facet' from its rule and from ofproto->facets.
3698 facet_remove(struct facet *facet)
3700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3701 struct subfacet *subfacet, *next_subfacet;
3703 assert(!list_is_empty(&facet->subfacets));
3705 /* First uninstall all of the subfacets to get final statistics. */
3706 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3707 subfacet_uninstall(subfacet);
3710 /* Flush the final stats to the rule.
3712 * This might require us to have at least one subfacet around so that we
3713 * can use its actions for accounting in facet_account(), which is why we
3714 * have uninstalled but not yet destroyed the subfacets. */
3715 facet_flush_stats(facet);
3717 /* Now we're really all done so destroy everything. */
3718 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3719 &facet->subfacets) {
3720 subfacet_destroy__(subfacet);
3722 hmap_remove(&ofproto->facets, &facet->hmap_node);
3723 list_remove(&facet->list_node);
3727 /* Feed information from 'facet' back into the learning table to keep it in
3728 * sync with what is actually flowing through the datapath. */
3730 facet_learn(struct facet *facet)
3732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3733 struct action_xlate_ctx ctx;
3735 if (!facet->has_learn
3736 && !facet->has_normal
3737 && (!facet->has_fin_timeout
3738 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3742 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3743 facet->flow.vlan_tci,
3744 facet->rule, facet->tcp_flags, NULL);
3745 ctx.may_learn = true;
3746 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3747 facet->rule->up.ofpacts_len);
3751 facet_account(struct facet *facet)
3753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3754 struct subfacet *subfacet;
3755 const struct nlattr *a;
3760 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3763 n_bytes = facet->byte_count - facet->accounted_bytes;
3765 /* This loop feeds byte counters to bond_account() for rebalancing to use
3766 * as a basis. We also need to track the actual VLAN on which the packet
3767 * is going to be sent to ensure that it matches the one passed to
3768 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3771 * We use the actions from an arbitrary subfacet because they should all
3772 * be equally valid for our purpose. */
3773 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3774 struct subfacet, list_node);
3775 vlan_tci = facet->flow.vlan_tci;
3776 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3777 subfacet->actions, subfacet->actions_len) {
3778 const struct ovs_action_push_vlan *vlan;
3779 struct ofport_dpif *port;
3781 switch (nl_attr_type(a)) {
3782 case OVS_ACTION_ATTR_OUTPUT:
3783 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3784 if (port && port->bundle && port->bundle->bond) {
3785 bond_account(port->bundle->bond, &facet->flow,
3786 vlan_tci_to_vid(vlan_tci), n_bytes);
3790 case OVS_ACTION_ATTR_POP_VLAN:
3791 vlan_tci = htons(0);
3794 case OVS_ACTION_ATTR_PUSH_VLAN:
3795 vlan = nl_attr_get(a);
3796 vlan_tci = vlan->vlan_tci;
3802 /* Returns true if the only action for 'facet' is to send to the controller.
3803 * (We don't report NetFlow expiration messages for such facets because they
3804 * are just part of the control logic for the network, not real traffic). */
3806 facet_is_controller_flow(struct facet *facet)
3809 const struct rule *rule = &facet->rule->up;
3810 const struct ofpact *ofpacts = rule->ofpacts;
3811 size_t ofpacts_len = rule->ofpacts_len;
3813 if (ofpacts_len > 0 &&
3814 ofpacts->type == OFPACT_CONTROLLER &&
3815 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
3822 /* Folds all of 'facet''s statistics into its rule. Also updates the
3823 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3824 * 'facet''s statistics in the datapath should have been zeroed and folded into
3825 * its packet and byte counts before this function is called. */
3827 facet_flush_stats(struct facet *facet)
3829 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3830 struct subfacet *subfacet;
3832 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3833 assert(!subfacet->dp_byte_count);
3834 assert(!subfacet->dp_packet_count);
3837 facet_push_stats(facet);
3838 if (facet->accounted_bytes < facet->byte_count) {
3839 facet_account(facet);
3840 facet->accounted_bytes = facet->byte_count;
3843 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3844 struct ofexpired expired;
3845 expired.flow = facet->flow;
3846 expired.packet_count = facet->packet_count;
3847 expired.byte_count = facet->byte_count;
3848 expired.used = facet->used;
3849 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3852 facet->rule->packet_count += facet->packet_count;
3853 facet->rule->byte_count += facet->byte_count;
3855 /* Reset counters to prevent double counting if 'facet' ever gets
3857 facet_reset_counters(facet);
3859 netflow_flow_clear(&facet->nf_flow);
3860 facet->tcp_flags = 0;
3863 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3864 * Returns it if found, otherwise a null pointer.
3866 * 'hash' must be the return value of flow_hash(flow, 0).
3868 * The returned facet might need revalidation; use facet_lookup_valid()
3869 * instead if that is important. */
3870 static struct facet *
3871 facet_find(struct ofproto_dpif *ofproto,
3872 const struct flow *flow, uint32_t hash)
3874 struct facet *facet;
3876 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3877 if (flow_equal(flow, &facet->flow)) {
3885 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3886 * Returns it if found, otherwise a null pointer.
3888 * 'hash' must be the return value of flow_hash(flow, 0).
3890 * The returned facet is guaranteed to be valid. */
3891 static struct facet *
3892 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3895 struct facet *facet;
3897 facet = facet_find(ofproto, flow, hash);
3899 && (ofproto->need_revalidate
3900 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3901 facet_revalidate(facet);
3908 subfacet_path_to_string(enum subfacet_path path)
3911 case SF_NOT_INSTALLED:
3912 return "not installed";
3914 return "in fast path";
3916 return "in slow path";
3922 /* Returns the path in which a subfacet should be installed if its 'slow'
3923 * member has the specified value. */
3924 static enum subfacet_path
3925 subfacet_want_path(enum slow_path_reason slow)
3927 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3930 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3931 * supposing that its actions have been recalculated as 'want_actions' and that
3932 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3934 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3935 const struct ofpbuf *want_actions)
3937 enum subfacet_path want_path = subfacet_want_path(slow);
3938 return (want_path != subfacet->path
3939 || (want_path == SF_FAST_PATH
3940 && (subfacet->actions_len != want_actions->size
3941 || memcmp(subfacet->actions, want_actions->data,
3942 subfacet->actions_len))));
3946 facet_check_consistency(struct facet *facet)
3948 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3950 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3952 uint64_t odp_actions_stub[1024 / 8];
3953 struct ofpbuf odp_actions;
3955 struct rule_dpif *rule;
3956 struct subfacet *subfacet;
3957 bool may_log = false;
3960 /* Check the rule for consistency. */
3961 rule = rule_dpif_lookup(ofproto, &facet->flow);
3962 ok = rule == facet->rule;
3964 may_log = !VLOG_DROP_WARN(&rl);
3969 flow_format(&s, &facet->flow);
3970 ds_put_format(&s, ": facet associated with wrong rule (was "
3971 "table=%"PRIu8",", facet->rule->up.table_id);
3972 cls_rule_format(&facet->rule->up.cr, &s);
3973 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3975 cls_rule_format(&rule->up.cr, &s);
3976 ds_put_char(&s, ')');
3978 VLOG_WARN("%s", ds_cstr(&s));
3983 /* Check the datapath actions for consistency. */
3984 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3985 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3986 enum subfacet_path want_path;
3987 struct odputil_keybuf keybuf;
3988 struct action_xlate_ctx ctx;
3992 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3993 subfacet->initial_tci, rule, 0, NULL);
3994 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3997 if (subfacet->path == SF_NOT_INSTALLED) {
3998 /* This only happens if the datapath reported an error when we
3999 * tried to install the flow. Don't flag another error here. */
4003 want_path = subfacet_want_path(subfacet->slow);
4004 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4005 /* The actions for slow-path flows may legitimately vary from one
4006 * packet to the next. We're done. */
4010 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4014 /* Inconsistency! */
4016 may_log = !VLOG_DROP_WARN(&rl);
4020 /* Rate-limited, skip reporting. */
4025 subfacet_get_key(subfacet, &keybuf, &key);
4026 odp_flow_key_format(key.data, key.size, &s);
4028 ds_put_cstr(&s, ": inconsistency in subfacet");
4029 if (want_path != subfacet->path) {
4030 enum odp_key_fitness fitness = subfacet->key_fitness;
4032 ds_put_format(&s, " (%s, fitness=%s)",
4033 subfacet_path_to_string(subfacet->path),
4034 odp_key_fitness_to_string(fitness));
4035 ds_put_format(&s, " (should have been %s)",
4036 subfacet_path_to_string(want_path));
4037 } else if (want_path == SF_FAST_PATH) {
4038 ds_put_cstr(&s, " (actions were: ");
4039 format_odp_actions(&s, subfacet->actions,
4040 subfacet->actions_len);
4041 ds_put_cstr(&s, ") (correct actions: ");
4042 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4043 ds_put_char(&s, ')');
4045 ds_put_cstr(&s, " (actions: ");
4046 format_odp_actions(&s, subfacet->actions,
4047 subfacet->actions_len);
4048 ds_put_char(&s, ')');
4050 VLOG_WARN("%s", ds_cstr(&s));
4053 ofpbuf_uninit(&odp_actions);
4058 /* Re-searches the classifier for 'facet':
4060 * - If the rule found is different from 'facet''s current rule, moves
4061 * 'facet' to the new rule and recompiles its actions.
4063 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4064 * where it is and recompiles its actions anyway. */
4066 facet_revalidate(struct facet *facet)
4068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4070 struct nlattr *odp_actions;
4073 struct actions *new_actions;
4075 struct action_xlate_ctx ctx;
4076 uint64_t odp_actions_stub[1024 / 8];
4077 struct ofpbuf odp_actions;
4079 struct rule_dpif *new_rule;
4080 struct subfacet *subfacet;
4083 COVERAGE_INC(facet_revalidate);
4085 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4087 /* Calculate new datapath actions.
4089 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4090 * emit a NetFlow expiration and, if so, we need to have the old state
4091 * around to properly compose it. */
4093 /* If the datapath actions changed or the installability changed,
4094 * then we need to talk to the datapath. */
4097 memset(&ctx, 0, sizeof ctx);
4098 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4099 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4100 enum slow_path_reason slow;
4102 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4103 subfacet->initial_tci, new_rule, 0, NULL);
4104 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4107 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4108 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4109 struct dpif_flow_stats stats;
4111 subfacet_install(subfacet,
4112 odp_actions.data, odp_actions.size, &stats, slow);
4113 subfacet_update_stats(subfacet, &stats);
4116 new_actions = xcalloc(list_size(&facet->subfacets),
4117 sizeof *new_actions);
4119 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4121 new_actions[i].actions_len = odp_actions.size;
4126 ofpbuf_uninit(&odp_actions);
4129 facet_flush_stats(facet);
4132 /* Update 'facet' now that we've taken care of all the old state. */
4133 facet->tags = ctx.tags;
4134 facet->nf_flow.output_iface = ctx.nf_output_iface;
4135 facet->has_learn = ctx.has_learn;
4136 facet->has_normal = ctx.has_normal;
4137 facet->has_fin_timeout = ctx.has_fin_timeout;
4138 facet->mirrors = ctx.mirrors;
4141 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4142 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4144 if (new_actions && new_actions[i].odp_actions) {
4145 free(subfacet->actions);
4146 subfacet->actions = new_actions[i].odp_actions;
4147 subfacet->actions_len = new_actions[i].actions_len;
4153 if (facet->rule != new_rule) {
4154 COVERAGE_INC(facet_changed_rule);
4155 list_remove(&facet->list_node);
4156 list_push_back(&new_rule->facets, &facet->list_node);
4157 facet->rule = new_rule;
4158 facet->used = new_rule->up.created;
4159 facet->prev_used = facet->used;
4163 /* Updates 'facet''s used time. Caller is responsible for calling
4164 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4166 facet_update_time(struct facet *facet, long long int used)
4168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4169 if (used > facet->used) {
4171 ofproto_rule_update_used(&facet->rule->up, used);
4172 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4177 facet_reset_counters(struct facet *facet)
4179 facet->packet_count = 0;
4180 facet->byte_count = 0;
4181 facet->prev_packet_count = 0;
4182 facet->prev_byte_count = 0;
4183 facet->accounted_bytes = 0;
4187 facet_push_stats(struct facet *facet)
4189 struct dpif_flow_stats stats;
4191 assert(facet->packet_count >= facet->prev_packet_count);
4192 assert(facet->byte_count >= facet->prev_byte_count);
4193 assert(facet->used >= facet->prev_used);
4195 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4196 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4197 stats.used = facet->used;
4198 stats.tcp_flags = 0;
4200 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4201 facet->prev_packet_count = facet->packet_count;
4202 facet->prev_byte_count = facet->byte_count;
4203 facet->prev_used = facet->used;
4205 flow_push_stats(facet->rule, &facet->flow, &stats);
4207 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4208 facet->mirrors, stats.n_packets, stats.n_bytes);
4213 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4215 rule->packet_count += stats->n_packets;
4216 rule->byte_count += stats->n_bytes;
4217 ofproto_rule_update_used(&rule->up, stats->used);
4220 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4221 * 'rule''s actions and mirrors. */
4223 flow_push_stats(struct rule_dpif *rule,
4224 const struct flow *flow, const struct dpif_flow_stats *stats)
4226 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4227 struct action_xlate_ctx ctx;
4229 ofproto_rule_update_used(&rule->up, stats->used);
4231 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4233 ctx.resubmit_stats = stats;
4234 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4235 rule->up.ofpacts_len);
4240 static struct subfacet *
4241 subfacet_find__(struct ofproto_dpif *ofproto,
4242 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4243 const struct flow *flow)
4245 struct subfacet *subfacet;
4247 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4248 &ofproto->subfacets) {
4250 ? (subfacet->key_len == key_len
4251 && !memcmp(key, subfacet->key, key_len))
4252 : flow_equal(flow, &subfacet->facet->flow)) {
4260 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4261 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4262 * there is one, otherwise creates and returns a new subfacet.
4264 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4265 * which case the caller must populate the actions with
4266 * subfacet_make_actions(). */
4267 static struct subfacet *
4268 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4269 const struct nlattr *key, size_t key_len,
4270 ovs_be16 initial_tci, long long int now)
4272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4273 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4274 struct subfacet *subfacet;
4276 if (list_is_empty(&facet->subfacets)) {
4277 subfacet = &facet->one_subfacet;
4279 subfacet = subfacet_find__(ofproto, key, key_len, key_hash,
4282 if (subfacet->facet == facet) {
4286 /* This shouldn't happen. */
4287 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4288 subfacet_destroy(subfacet);
4291 subfacet = xmalloc(sizeof *subfacet);
4294 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4295 list_push_back(&facet->subfacets, &subfacet->list_node);
4296 subfacet->facet = facet;
4297 subfacet->key_fitness = key_fitness;
4298 if (key_fitness != ODP_FIT_PERFECT) {
4299 subfacet->key = xmemdup(key, key_len);
4300 subfacet->key_len = key_len;
4302 subfacet->key = NULL;
4303 subfacet->key_len = 0;
4305 subfacet->used = now;
4306 subfacet->dp_packet_count = 0;
4307 subfacet->dp_byte_count = 0;
4308 subfacet->actions_len = 0;
4309 subfacet->actions = NULL;
4310 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4313 subfacet->path = SF_NOT_INSTALLED;
4314 subfacet->initial_tci = initial_tci;
4319 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4320 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4321 static struct subfacet *
4322 subfacet_find(struct ofproto_dpif *ofproto,
4323 const struct nlattr *key, size_t key_len)
4325 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4326 enum odp_key_fitness fitness;
4329 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4330 if (fitness == ODP_FIT_ERROR) {
4334 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4337 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4338 * its facet within 'ofproto', and frees it. */
4340 subfacet_destroy__(struct subfacet *subfacet)
4342 struct facet *facet = subfacet->facet;
4343 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4345 subfacet_uninstall(subfacet);
4346 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4347 list_remove(&subfacet->list_node);
4348 free(subfacet->key);
4349 free(subfacet->actions);
4350 if (subfacet != &facet->one_subfacet) {
4355 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4356 * last remaining subfacet in its facet destroys the facet too. */
4358 subfacet_destroy(struct subfacet *subfacet)
4360 struct facet *facet = subfacet->facet;
4362 if (list_is_singleton(&facet->subfacets)) {
4363 /* facet_remove() needs at least one subfacet (it will remove it). */
4364 facet_remove(facet);
4366 subfacet_destroy__(subfacet);
4370 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4371 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4372 * for use as temporary storage. */
4374 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4377 if (!subfacet->key) {
4378 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4379 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4381 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4385 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4386 * Translates the actions into 'odp_actions', which the caller must have
4387 * initialized and is responsible for uninitializing. */
4389 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4390 struct ofpbuf *odp_actions)
4392 struct facet *facet = subfacet->facet;
4393 struct rule_dpif *rule = facet->rule;
4394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4396 struct action_xlate_ctx ctx;
4398 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4400 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4401 facet->tags = ctx.tags;
4402 facet->has_learn = ctx.has_learn;
4403 facet->has_normal = ctx.has_normal;
4404 facet->has_fin_timeout = ctx.has_fin_timeout;
4405 facet->nf_flow.output_iface = ctx.nf_output_iface;
4406 facet->mirrors = ctx.mirrors;
4408 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4409 if (subfacet->actions_len != odp_actions->size
4410 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4411 free(subfacet->actions);
4412 subfacet->actions_len = odp_actions->size;
4413 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4417 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4418 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4419 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4420 * since 'subfacet' was last updated.
4422 * Returns 0 if successful, otherwise a positive errno value. */
4424 subfacet_install(struct subfacet *subfacet,
4425 const struct nlattr *actions, size_t actions_len,
4426 struct dpif_flow_stats *stats,
4427 enum slow_path_reason slow)
4429 struct facet *facet = subfacet->facet;
4430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4431 enum subfacet_path path = subfacet_want_path(slow);
4432 uint64_t slow_path_stub[128 / 8];
4433 struct odputil_keybuf keybuf;
4434 enum dpif_flow_put_flags flags;
4438 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4440 flags |= DPIF_FP_ZERO_STATS;
4443 if (path == SF_SLOW_PATH) {
4444 compose_slow_path(ofproto, &facet->flow, slow,
4445 slow_path_stub, sizeof slow_path_stub,
4446 &actions, &actions_len);
4449 subfacet_get_key(subfacet, &keybuf, &key);
4450 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4451 actions, actions_len, stats);
4454 subfacet_reset_dp_stats(subfacet, stats);
4458 subfacet->path = path;
4464 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4466 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4467 stats, subfacet->slow);
4470 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4472 subfacet_uninstall(struct subfacet *subfacet)
4474 if (subfacet->path != SF_NOT_INSTALLED) {
4475 struct rule_dpif *rule = subfacet->facet->rule;
4476 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4477 struct odputil_keybuf keybuf;
4478 struct dpif_flow_stats stats;
4482 subfacet_get_key(subfacet, &keybuf, &key);
4483 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4484 subfacet_reset_dp_stats(subfacet, &stats);
4486 subfacet_update_stats(subfacet, &stats);
4488 subfacet->path = SF_NOT_INSTALLED;
4490 assert(subfacet->dp_packet_count == 0);
4491 assert(subfacet->dp_byte_count == 0);
4495 /* Resets 'subfacet''s datapath statistics counters. This should be called
4496 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4497 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4498 * was reset in the datapath. 'stats' will be modified to include only
4499 * statistics new since 'subfacet' was last updated. */
4501 subfacet_reset_dp_stats(struct subfacet *subfacet,
4502 struct dpif_flow_stats *stats)
4505 && subfacet->dp_packet_count <= stats->n_packets
4506 && subfacet->dp_byte_count <= stats->n_bytes) {
4507 stats->n_packets -= subfacet->dp_packet_count;
4508 stats->n_bytes -= subfacet->dp_byte_count;
4511 subfacet->dp_packet_count = 0;
4512 subfacet->dp_byte_count = 0;
4515 /* Updates 'subfacet''s used time. The caller is responsible for calling
4516 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4518 subfacet_update_time(struct subfacet *subfacet, long long int used)
4520 if (used > subfacet->used) {
4521 subfacet->used = used;
4522 facet_update_time(subfacet->facet, used);
4526 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4528 * Because of the meaning of a subfacet's counters, it only makes sense to do
4529 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4530 * represents a packet that was sent by hand or if it represents statistics
4531 * that have been cleared out of the datapath. */
4533 subfacet_update_stats(struct subfacet *subfacet,
4534 const struct dpif_flow_stats *stats)
4536 if (stats->n_packets || stats->used > subfacet->used) {
4537 struct facet *facet = subfacet->facet;
4539 subfacet_update_time(subfacet, stats->used);
4540 facet->packet_count += stats->n_packets;
4541 facet->byte_count += stats->n_bytes;
4542 facet->tcp_flags |= stats->tcp_flags;
4543 facet_push_stats(facet);
4544 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4550 static struct rule_dpif *
4551 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4553 struct rule_dpif *rule;
4555 rule = rule_dpif_lookup__(ofproto, flow, 0);
4560 return rule_dpif_miss_rule(ofproto, flow);
4563 static struct rule_dpif *
4564 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4567 struct cls_rule *cls_rule;
4568 struct classifier *cls;
4570 if (table_id >= N_TABLES) {
4574 cls = &ofproto->up.tables[table_id].cls;
4575 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4576 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4577 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4578 * are unavailable. */
4579 struct flow ofpc_normal_flow = *flow;
4580 ofpc_normal_flow.tp_src = htons(0);
4581 ofpc_normal_flow.tp_dst = htons(0);
4582 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4584 cls_rule = classifier_lookup(cls, flow);
4586 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4589 static struct rule_dpif *
4590 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4592 struct ofport_dpif *port;
4594 port = get_ofp_port(ofproto, flow->in_port);
4596 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4597 return ofproto->miss_rule;
4600 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4601 return ofproto->no_packet_in_rule;
4603 return ofproto->miss_rule;
4607 complete_operation(struct rule_dpif *rule)
4609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4611 rule_invalidate(rule);
4613 struct dpif_completion *c = xmalloc(sizeof *c);
4614 c->op = rule->up.pending;
4615 list_push_back(&ofproto->completions, &c->list_node);
4617 ofoperation_complete(rule->up.pending, 0);
4621 static struct rule *
4624 struct rule_dpif *rule = xmalloc(sizeof *rule);
4629 rule_dealloc(struct rule *rule_)
4631 struct rule_dpif *rule = rule_dpif_cast(rule_);
4636 rule_construct(struct rule *rule_)
4638 struct rule_dpif *rule = rule_dpif_cast(rule_);
4639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4640 struct rule_dpif *victim;
4643 rule->packet_count = 0;
4644 rule->byte_count = 0;
4646 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4647 if (victim && !list_is_empty(&victim->facets)) {
4648 struct facet *facet;
4650 rule->facets = victim->facets;
4651 list_moved(&rule->facets);
4652 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4653 /* XXX: We're only clearing our local counters here. It's possible
4654 * that quite a few packets are unaccounted for in the datapath
4655 * statistics. These will be accounted to the new rule instead of
4656 * cleared as required. This could be fixed by clearing out the
4657 * datapath statistics for this facet, but currently it doesn't
4659 facet_reset_counters(facet);
4663 /* Must avoid list_moved() in this case. */
4664 list_init(&rule->facets);
4667 table_id = rule->up.table_id;
4668 rule->tag = (victim ? victim->tag
4670 : rule_calculate_tag(&rule->up.cr.match.flow,
4671 &rule->up.cr.match.wc,
4672 ofproto->tables[table_id].basis));
4674 complete_operation(rule);
4679 rule_destruct(struct rule *rule_)
4681 struct rule_dpif *rule = rule_dpif_cast(rule_);
4682 struct facet *facet, *next_facet;
4684 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4685 facet_revalidate(facet);
4688 complete_operation(rule);
4692 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4694 struct rule_dpif *rule = rule_dpif_cast(rule_);
4695 struct facet *facet;
4697 /* Start from historical data for 'rule' itself that are no longer tracked
4698 * in facets. This counts, for example, facets that have expired. */
4699 *packets = rule->packet_count;
4700 *bytes = rule->byte_count;
4702 /* Add any statistics that are tracked by facets. This includes
4703 * statistical data recently updated by ofproto_update_stats() as well as
4704 * stats for packets that were executed "by hand" via dpif_execute(). */
4705 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4706 *packets += facet->packet_count;
4707 *bytes += facet->byte_count;
4712 rule_execute(struct rule *rule_, const struct flow *flow,
4713 struct ofpbuf *packet)
4715 struct rule_dpif *rule = rule_dpif_cast(rule_);
4716 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4718 struct dpif_flow_stats stats;
4720 struct action_xlate_ctx ctx;
4721 uint64_t odp_actions_stub[1024 / 8];
4722 struct ofpbuf odp_actions;
4724 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4725 rule_credit_stats(rule, &stats);
4727 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4728 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4729 rule, stats.tcp_flags, packet);
4730 ctx.resubmit_stats = &stats;
4731 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4733 execute_odp_actions(ofproto, flow, odp_actions.data,
4734 odp_actions.size, packet);
4736 ofpbuf_uninit(&odp_actions);
4742 rule_modify_actions(struct rule *rule_)
4744 struct rule_dpif *rule = rule_dpif_cast(rule_);
4746 complete_operation(rule);
4749 /* Sends 'packet' out 'ofport'.
4750 * May modify 'packet'.
4751 * Returns 0 if successful, otherwise a positive errno value. */
4753 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4755 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4756 struct ofpbuf key, odp_actions;
4757 struct odputil_keybuf keybuf;
4762 flow_extract(packet, 0, 0, 0, &flow);
4763 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4765 if (odp_port != ofport->odp_port) {
4766 eth_pop_vlan(packet);
4767 flow.vlan_tci = htons(0);
4770 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4771 odp_flow_key_from_flow(&key, &flow);
4773 ofpbuf_init(&odp_actions, 32);
4774 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4776 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4777 error = dpif_execute(ofproto->dpif,
4779 odp_actions.data, odp_actions.size,
4781 ofpbuf_uninit(&odp_actions);
4784 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4785 ofproto->up.name, odp_port, strerror(error));
4787 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4791 /* OpenFlow to datapath action translation. */
4793 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
4794 struct action_xlate_ctx *);
4795 static void xlate_normal(struct action_xlate_ctx *);
4797 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4798 * The action will state 'slow' as the reason that the action is in the slow
4799 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4800 * dump-flows" output to see why a flow is in the slow path.)
4802 * The 'stub_size' bytes in 'stub' will be used to store the action.
4803 * 'stub_size' must be large enough for the action.
4805 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4808 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4809 enum slow_path_reason slow,
4810 uint64_t *stub, size_t stub_size,
4811 const struct nlattr **actionsp, size_t *actions_lenp)
4813 union user_action_cookie cookie;
4816 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4817 cookie.slow_path.unused = 0;
4818 cookie.slow_path.reason = slow;
4820 ofpbuf_use_stack(&buf, stub, stub_size);
4821 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4822 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4823 odp_put_userspace_action(pid, &cookie, &buf);
4825 put_userspace_action(ofproto, &buf, flow, &cookie);
4827 *actionsp = buf.data;
4828 *actions_lenp = buf.size;
4832 put_userspace_action(const struct ofproto_dpif *ofproto,
4833 struct ofpbuf *odp_actions,
4834 const struct flow *flow,
4835 const union user_action_cookie *cookie)
4839 pid = dpif_port_get_pid(ofproto->dpif,
4840 ofp_port_to_odp_port(flow->in_port));
4842 return odp_put_userspace_action(pid, cookie, odp_actions);
4846 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4847 ovs_be16 vlan_tci, uint32_t odp_port,
4848 unsigned int n_outputs, union user_action_cookie *cookie)
4852 cookie->type = USER_ACTION_COOKIE_SFLOW;
4853 cookie->sflow.vlan_tci = vlan_tci;
4855 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4856 * port information") for the interpretation of cookie->output. */
4857 switch (n_outputs) {
4859 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4860 cookie->sflow.output = 0x40000000 | 256;
4864 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4866 cookie->sflow.output = ifindex;
4871 /* 0x80000000 means "multiple output ports. */
4872 cookie->sflow.output = 0x80000000 | n_outputs;
4877 /* Compose SAMPLE action for sFlow. */
4879 compose_sflow_action(const struct ofproto_dpif *ofproto,
4880 struct ofpbuf *odp_actions,
4881 const struct flow *flow,
4884 uint32_t probability;
4885 union user_action_cookie cookie;
4886 size_t sample_offset, actions_offset;
4889 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4893 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4895 /* Number of packets out of UINT_MAX to sample. */
4896 probability = dpif_sflow_get_probability(ofproto->sflow);
4897 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4899 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4900 compose_sflow_cookie(ofproto, htons(0), odp_port,
4901 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4902 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4904 nl_msg_end_nested(odp_actions, actions_offset);
4905 nl_msg_end_nested(odp_actions, sample_offset);
4906 return cookie_offset;
4909 /* SAMPLE action must be first action in any given list of actions.
4910 * At this point we do not have all information required to build it. So try to
4911 * build sample action as complete as possible. */
4913 add_sflow_action(struct action_xlate_ctx *ctx)
4915 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4917 &ctx->flow, OVSP_NONE);
4918 ctx->sflow_odp_port = 0;
4919 ctx->sflow_n_outputs = 0;
4922 /* Fix SAMPLE action according to data collected while composing ODP actions.
4923 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4924 * USERSPACE action's user-cookie which is required for sflow. */
4926 fix_sflow_action(struct action_xlate_ctx *ctx)
4928 const struct flow *base = &ctx->base_flow;
4929 union user_action_cookie *cookie;
4931 if (!ctx->user_cookie_offset) {
4935 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4937 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4939 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4940 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4944 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4947 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4948 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4949 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4950 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4954 struct priority_to_dscp *pdscp;
4956 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4957 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4961 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4963 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4964 ctx->flow.nw_tos |= pdscp->dscp;
4967 /* We may not have an ofport record for this port, but it doesn't hurt
4968 * to allow forwarding to it anyhow. Maybe such a port will appear
4969 * later and we're pre-populating the flow table. */
4972 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4973 ctx->flow.vlan_tci);
4974 if (out_port != odp_port) {
4975 ctx->flow.vlan_tci = htons(0);
4977 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4978 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4980 ctx->sflow_odp_port = odp_port;
4981 ctx->sflow_n_outputs++;
4982 ctx->nf_output_iface = ofp_port;
4983 ctx->flow.vlan_tci = flow_vlan_tci;
4984 ctx->flow.nw_tos = flow_nw_tos;
4988 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4990 compose_output_action__(ctx, ofp_port, true);
4994 xlate_table_action(struct action_xlate_ctx *ctx,
4995 uint16_t in_port, uint8_t table_id, bool may_packet_in)
4997 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4998 struct ofproto_dpif *ofproto = ctx->ofproto;
4999 struct rule_dpif *rule;
5000 uint16_t old_in_port;
5001 uint8_t old_table_id;
5003 old_table_id = ctx->table_id;
5004 ctx->table_id = table_id;
5006 /* Look up a flow with 'in_port' as the input port. */
5007 old_in_port = ctx->flow.in_port;
5008 ctx->flow.in_port = in_port;
5009 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5012 if (table_id > 0 && table_id < N_TABLES) {
5013 struct table_dpif *table = &ofproto->tables[table_id];
5014 if (table->other_table) {
5015 ctx->tags |= (rule && rule->tag
5017 : rule_calculate_tag(&ctx->flow,
5018 &table->other_table->wc,
5023 /* Restore the original input port. Otherwise OFPP_NORMAL and
5024 * OFPP_IN_PORT will have surprising behavior. */
5025 ctx->flow.in_port = old_in_port;
5027 if (ctx->resubmit_hook) {
5028 ctx->resubmit_hook(ctx, rule);
5031 if (rule == NULL && may_packet_in) {
5033 * check if table configuration flags
5034 * OFPTC_TABLE_MISS_CONTROLLER, default.
5035 * OFPTC_TABLE_MISS_CONTINUE,
5036 * OFPTC_TABLE_MISS_DROP
5037 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5039 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5043 struct rule_dpif *old_rule = ctx->rule;
5045 if (ctx->resubmit_stats) {
5046 rule_credit_stats(rule, ctx->resubmit_stats);
5051 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5052 ctx->rule = old_rule;
5056 ctx->table_id = old_table_id;
5058 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5060 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5061 MAX_RESUBMIT_RECURSION);
5062 ctx->max_resubmit_trigger = true;
5067 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5068 const struct ofpact_resubmit *resubmit)
5073 in_port = resubmit->in_port;
5074 if (in_port == OFPP_IN_PORT) {
5075 in_port = ctx->flow.in_port;
5078 table_id = resubmit->table_id;
5079 if (table_id == 255) {
5080 table_id = ctx->table_id;
5083 xlate_table_action(ctx, in_port, table_id, false);
5087 flood_packets(struct action_xlate_ctx *ctx, bool all)
5089 struct ofport_dpif *ofport;
5091 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5092 uint16_t ofp_port = ofport->up.ofp_port;
5094 if (ofp_port == ctx->flow.in_port) {
5099 compose_output_action__(ctx, ofp_port, false);
5100 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5101 compose_output_action(ctx, ofp_port);
5105 ctx->nf_output_iface = NF_OUT_FLOOD;
5109 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5110 enum ofp_packet_in_reason reason,
5111 uint16_t controller_id)
5113 struct ofputil_packet_in pin;
5114 struct ofpbuf *packet;
5116 ctx->slow |= SLOW_CONTROLLER;
5121 packet = ofpbuf_clone(ctx->packet);
5123 if (packet->l2 && packet->l3) {
5124 struct eth_header *eh;
5126 eth_pop_vlan(packet);
5129 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5130 * LLC frame. Calculating the Ethernet type of these frames is more
5131 * trouble than seems appropriate for a simple assertion. */
5132 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5133 || eh->eth_type == ctx->flow.dl_type);
5135 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5136 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5138 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5139 eth_push_vlan(packet, ctx->flow.vlan_tci);
5143 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5144 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5145 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5149 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5150 packet_set_tcp_port(packet, ctx->flow.tp_src,
5152 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5153 packet_set_udp_port(packet, ctx->flow.tp_src,
5160 pin.packet = packet->data;
5161 pin.packet_len = packet->size;
5162 pin.reason = reason;
5163 pin.controller_id = controller_id;
5164 pin.table_id = ctx->table_id;
5165 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5168 flow_get_metadata(&ctx->flow, &pin.fmd);
5170 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5171 ofpbuf_delete(packet);
5175 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5177 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5178 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5182 if (ctx->flow.nw_ttl > 1) {
5188 for (i = 0; i < ids->n_controllers; i++) {
5189 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5193 /* Stop processing for current table. */
5199 xlate_output_action(struct action_xlate_ctx *ctx,
5200 uint16_t port, uint16_t max_len, bool may_packet_in)
5202 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5204 ctx->nf_output_iface = NF_OUT_DROP;
5208 compose_output_action(ctx, ctx->flow.in_port);
5211 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5217 flood_packets(ctx, false);
5220 flood_packets(ctx, true);
5222 case OFPP_CONTROLLER:
5223 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5229 if (port != ctx->flow.in_port) {
5230 compose_output_action(ctx, port);
5235 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5236 ctx->nf_output_iface = NF_OUT_FLOOD;
5237 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5238 ctx->nf_output_iface = prev_nf_output_iface;
5239 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5240 ctx->nf_output_iface != NF_OUT_FLOOD) {
5241 ctx->nf_output_iface = NF_OUT_MULTI;
5246 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5247 const struct ofpact_output_reg *or)
5249 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5250 if (port <= UINT16_MAX) {
5251 xlate_output_action(ctx, port, or->max_len, false);
5256 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5257 const struct ofpact_enqueue *enqueue)
5259 uint16_t ofp_port = enqueue->port;
5260 uint32_t queue_id = enqueue->queue;
5261 uint32_t flow_priority, priority;
5264 /* Translate queue to priority. */
5265 error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority);
5267 /* Fall back to ordinary output action. */
5268 xlate_output_action(ctx, enqueue->port, 0, false);
5272 /* Check output port. */
5273 if (ofp_port == OFPP_IN_PORT) {
5274 ofp_port = ctx->flow.in_port;
5275 } else if (ofp_port == ctx->flow.in_port) {
5279 /* Add datapath actions. */
5280 flow_priority = ctx->flow.skb_priority;
5281 ctx->flow.skb_priority = priority;
5282 compose_output_action(ctx, ofp_port);
5283 ctx->flow.skb_priority = flow_priority;
5285 /* Update NetFlow output port. */
5286 if (ctx->nf_output_iface == NF_OUT_DROP) {
5287 ctx->nf_output_iface = ofp_port;
5288 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5289 ctx->nf_output_iface = NF_OUT_MULTI;
5294 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5296 uint32_t skb_priority;
5298 if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) {
5299 ctx->flow.skb_priority = skb_priority;
5301 /* Couldn't translate queue to a priority. Nothing to do. A warning
5302 * has already been logged. */
5306 struct xlate_reg_state {
5312 xlate_autopath(struct action_xlate_ctx *ctx,
5313 const struct ofpact_autopath *ap)
5315 uint16_t ofp_port = ap->port;
5316 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5318 if (!port || !port->bundle) {
5319 ofp_port = OFPP_NONE;
5320 } else if (port->bundle->bond) {
5321 /* Autopath does not support VLAN hashing. */
5322 struct ofport_dpif *slave = bond_choose_output_slave(
5323 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5325 ofp_port = slave->up.ofp_port;
5328 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5332 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5334 struct ofproto_dpif *ofproto = ofproto_;
5335 struct ofport_dpif *port;
5345 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5348 port = get_ofp_port(ofproto, ofp_port);
5349 return port ? port->may_enable : false;
5354 xlate_bundle_action(struct action_xlate_ctx *ctx,
5355 const struct ofpact_bundle *bundle)
5359 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5360 if (bundle->dst.field) {
5361 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5363 xlate_output_action(ctx, port, 0, false);
5368 xlate_learn_action(struct action_xlate_ctx *ctx,
5369 const struct ofpact_learn *learn)
5371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5372 struct ofputil_flow_mod fm;
5373 uint64_t ofpacts_stub[1024 / 8];
5374 struct ofpbuf ofpacts;
5377 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5378 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5380 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5381 if (error && !VLOG_DROP_WARN(&rl)) {
5382 VLOG_WARN("learning action failed to modify flow table (%s)",
5383 ofperr_get_name(error));
5386 ofpbuf_uninit(&ofpacts);
5389 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5390 * means "infinite". */
5392 reduce_timeout(uint16_t max, uint16_t *timeout)
5394 if (max && (!*timeout || *timeout > max)) {
5400 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5401 const struct ofpact_fin_timeout *oft)
5403 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5404 struct rule_dpif *rule = ctx->rule;
5406 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5407 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5412 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5414 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5415 ? OFPUTIL_PC_NO_RECV_STP
5416 : OFPUTIL_PC_NO_RECV)) {
5420 /* Only drop packets here if both forwarding and learning are
5421 * disabled. If just learning is enabled, we need to have
5422 * OFPP_NORMAL and the learning action have a look at the packet
5423 * before we can drop it. */
5424 if (!stp_forward_in_state(port->stp_state)
5425 && !stp_learn_in_state(port->stp_state)) {
5433 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5434 struct action_xlate_ctx *ctx)
5436 const struct ofport_dpif *port;
5437 bool was_evictable = true;
5438 const struct ofpact *a;
5440 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5441 if (port && !may_receive(port, ctx)) {
5442 /* Drop this flow. */
5447 /* Don't let the rule we're working on get evicted underneath us. */
5448 was_evictable = ctx->rule->up.evictable;
5449 ctx->rule->up.evictable = false;
5451 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5452 struct ofpact_controller *controller;
5460 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5461 ofpact_get_OUTPUT(a)->max_len, true);
5464 case OFPACT_CONTROLLER:
5465 controller = ofpact_get_CONTROLLER(a);
5466 execute_controller_action(ctx, controller->max_len,
5468 controller->controller_id);
5471 case OFPACT_ENQUEUE:
5472 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5475 case OFPACT_SET_VLAN_VID:
5476 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5477 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5481 case OFPACT_SET_VLAN_PCP:
5482 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5483 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5488 case OFPACT_STRIP_VLAN:
5489 ctx->flow.vlan_tci = htons(0);
5492 case OFPACT_SET_ETH_SRC:
5493 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5497 case OFPACT_SET_ETH_DST:
5498 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5502 case OFPACT_SET_IPV4_SRC:
5503 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5506 case OFPACT_SET_IPV4_DST:
5507 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5510 case OFPACT_SET_IPV4_DSCP:
5511 /* OpenFlow 1.0 only supports IPv4. */
5512 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5513 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5514 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5518 case OFPACT_SET_L4_SRC_PORT:
5519 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5522 case OFPACT_SET_L4_DST_PORT:
5523 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5526 case OFPACT_RESUBMIT:
5527 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5530 case OFPACT_SET_TUNNEL:
5531 ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5534 case OFPACT_SET_QUEUE:
5535 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5538 case OFPACT_POP_QUEUE:
5539 ctx->flow.skb_priority = ctx->orig_skb_priority;
5542 case OFPACT_REG_MOVE:
5543 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5546 case OFPACT_REG_LOAD:
5547 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5550 case OFPACT_DEC_TTL:
5551 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5557 /* Nothing to do. */
5560 case OFPACT_MULTIPATH:
5561 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5564 case OFPACT_AUTOPATH:
5565 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5569 ctx->ofproto->has_bundle_action = true;
5570 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5573 case OFPACT_OUTPUT_REG:
5574 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
5578 ctx->has_learn = true;
5579 if (ctx->may_learn) {
5580 xlate_learn_action(ctx, ofpact_get_LEARN(a));
5588 case OFPACT_FIN_TIMEOUT:
5589 ctx->has_fin_timeout = true;
5590 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
5596 /* We've let OFPP_NORMAL and the learning action look at the packet,
5597 * so drop it now if forwarding is disabled. */
5598 if (port && !stp_forward_in_state(port->stp_state)) {
5599 ofpbuf_clear(ctx->odp_actions);
5600 add_sflow_action(ctx);
5603 ctx->rule->up.evictable = was_evictable;
5608 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5609 struct ofproto_dpif *ofproto, const struct flow *flow,
5610 ovs_be16 initial_tci, struct rule_dpif *rule,
5611 uint8_t tcp_flags, const struct ofpbuf *packet)
5613 ctx->ofproto = ofproto;
5615 ctx->base_flow = ctx->flow;
5616 ctx->base_flow.tun_id = 0;
5617 ctx->base_flow.vlan_tci = initial_tci;
5619 ctx->packet = packet;
5620 ctx->may_learn = packet != NULL;
5621 ctx->tcp_flags = tcp_flags;
5622 ctx->resubmit_hook = NULL;
5623 ctx->report_hook = NULL;
5624 ctx->resubmit_stats = NULL;
5627 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
5628 * into datapath actions in 'odp_actions', using 'ctx'. */
5630 xlate_actions(struct action_xlate_ctx *ctx,
5631 const struct ofpact *ofpacts, size_t ofpacts_len,
5632 struct ofpbuf *odp_actions)
5634 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5635 * that in the future we always keep a copy of the original flow for
5636 * tracing purposes. */
5637 static bool hit_resubmit_limit;
5639 enum slow_path_reason special;
5641 COVERAGE_INC(ofproto_dpif_xlate);
5643 ofpbuf_clear(odp_actions);
5644 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5646 ctx->odp_actions = odp_actions;
5649 ctx->has_learn = false;
5650 ctx->has_normal = false;
5651 ctx->has_fin_timeout = false;
5652 ctx->nf_output_iface = NF_OUT_DROP;
5655 ctx->max_resubmit_trigger = false;
5656 ctx->orig_skb_priority = ctx->flow.skb_priority;
5660 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5661 /* Do this conditionally because the copy is expensive enough that it
5662 * shows up in profiles.
5664 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5665 * believe that I wasn't using it without initializing it if I kept it
5666 * in a local variable. */
5667 ctx->orig_flow = ctx->flow;
5670 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5671 switch (ctx->ofproto->up.frag_handling) {
5672 case OFPC_FRAG_NORMAL:
5673 /* We must pretend that transport ports are unavailable. */
5674 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5675 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5678 case OFPC_FRAG_DROP:
5681 case OFPC_FRAG_REASM:
5684 case OFPC_FRAG_NX_MATCH:
5685 /* Nothing to do. */
5688 case OFPC_INVALID_TTL_TO_CONTROLLER:
5693 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5695 ctx->slow |= special;
5697 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5698 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5700 add_sflow_action(ctx);
5701 do_xlate_actions(ofpacts, ofpacts_len, ctx);
5703 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5704 if (!hit_resubmit_limit) {
5705 /* We didn't record the original flow. Make sure we do from
5707 hit_resubmit_limit = true;
5708 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5709 struct ds ds = DS_EMPTY_INITIALIZER;
5711 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5713 VLOG_ERR("Trace triggered by excessive resubmit "
5714 "recursion:\n%s", ds_cstr(&ds));
5719 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5720 ctx->odp_actions->data,
5721 ctx->odp_actions->size)) {
5722 ctx->slow |= SLOW_IN_BAND;
5724 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5726 compose_output_action(ctx, OFPP_LOCAL);
5729 if (ctx->ofproto->has_mirrors) {
5730 add_mirror_actions(ctx, &ctx->orig_flow);
5732 fix_sflow_action(ctx);
5736 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5737 * into datapath actions, using 'ctx', and discards the datapath actions. */
5739 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5740 const struct ofpact *ofpacts,
5743 uint64_t odp_actions_stub[1024 / 8];
5744 struct ofpbuf odp_actions;
5746 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5747 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
5748 ofpbuf_uninit(&odp_actions);
5752 xlate_report(struct action_xlate_ctx *ctx, const char *s)
5754 if (ctx->report_hook) {
5755 ctx->report_hook(ctx, s);
5759 /* OFPP_NORMAL implementation. */
5761 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5763 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5764 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5765 * the bundle on which the packet was received, returns the VLAN to which the
5768 * Both 'vid' and the return value are in the range 0...4095. */
5770 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5772 switch (in_bundle->vlan_mode) {
5773 case PORT_VLAN_ACCESS:
5774 return in_bundle->vlan;
5777 case PORT_VLAN_TRUNK:
5780 case PORT_VLAN_NATIVE_UNTAGGED:
5781 case PORT_VLAN_NATIVE_TAGGED:
5782 return vid ? vid : in_bundle->vlan;
5789 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5790 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5793 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5794 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5797 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5799 /* Allow any VID on the OFPP_NONE port. */
5800 if (in_bundle == &ofpp_none_bundle) {
5804 switch (in_bundle->vlan_mode) {
5805 case PORT_VLAN_ACCESS:
5808 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5809 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5810 "packet received on port %s configured as VLAN "
5811 "%"PRIu16" access port",
5812 in_bundle->ofproto->up.name, vid,
5813 in_bundle->name, in_bundle->vlan);
5819 case PORT_VLAN_NATIVE_UNTAGGED:
5820 case PORT_VLAN_NATIVE_TAGGED:
5822 /* Port must always carry its native VLAN. */
5826 case PORT_VLAN_TRUNK:
5827 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5829 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5830 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5831 "received on port %s not configured for trunking "
5833 in_bundle->ofproto->up.name, vid,
5834 in_bundle->name, vid);
5846 /* Given 'vlan', the VLAN that a packet belongs to, and
5847 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5848 * that should be included in the 802.1Q header. (If the return value is 0,
5849 * then the 802.1Q header should only be included in the packet if there is a
5852 * Both 'vlan' and the return value are in the range 0...4095. */
5854 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5856 switch (out_bundle->vlan_mode) {
5857 case PORT_VLAN_ACCESS:
5860 case PORT_VLAN_TRUNK:
5861 case PORT_VLAN_NATIVE_TAGGED:
5864 case PORT_VLAN_NATIVE_UNTAGGED:
5865 return vlan == out_bundle->vlan ? 0 : vlan;
5873 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5876 struct ofport_dpif *port;
5878 ovs_be16 tci, old_tci;
5880 vid = output_vlan_to_vid(out_bundle, vlan);
5881 if (!out_bundle->bond) {
5882 port = ofbundle_get_a_port(out_bundle);
5884 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5887 /* No slaves enabled, so drop packet. */
5892 old_tci = ctx->flow.vlan_tci;
5894 if (tci || out_bundle->use_priority_tags) {
5895 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5897 tci |= htons(VLAN_CFI);
5900 ctx->flow.vlan_tci = tci;
5902 compose_output_action(ctx, port->up.ofp_port);
5903 ctx->flow.vlan_tci = old_tci;
5907 mirror_mask_ffs(mirror_mask_t mask)
5909 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5914 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5916 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5917 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5921 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5923 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5926 /* Returns an arbitrary interface within 'bundle'. */
5927 static struct ofport_dpif *
5928 ofbundle_get_a_port(const struct ofbundle *bundle)
5930 return CONTAINER_OF(list_front(&bundle->ports),
5931 struct ofport_dpif, bundle_node);
5935 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5937 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5941 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5943 struct ofproto_dpif *ofproto = ctx->ofproto;
5944 mirror_mask_t mirrors;
5945 struct ofbundle *in_bundle;
5948 const struct nlattr *a;
5951 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5952 ctx->packet != NULL, NULL);
5956 mirrors = in_bundle->src_mirrors;
5958 /* Drop frames on bundles reserved for mirroring. */
5959 if (in_bundle->mirror_out) {
5960 if (ctx->packet != NULL) {
5961 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5962 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5963 "%s, which is reserved exclusively for mirroring",
5964 ctx->ofproto->up.name, in_bundle->name);
5970 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5971 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5974 vlan = input_vid_to_vlan(in_bundle, vid);
5976 /* Look at the output ports to check for destination selections. */
5978 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5979 ctx->odp_actions->size) {
5980 enum ovs_action_attr type = nl_attr_type(a);
5981 struct ofport_dpif *ofport;
5983 if (type != OVS_ACTION_ATTR_OUTPUT) {
5987 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5988 if (ofport && ofport->bundle) {
5989 mirrors |= ofport->bundle->dst_mirrors;
5997 /* Restore the original packet before adding the mirror actions. */
5998 ctx->flow = *orig_flow;
6003 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6005 if (!vlan_is_mirrored(m, vlan)) {
6006 mirrors = zero_rightmost_1bit(mirrors);
6010 mirrors &= ~m->dup_mirrors;
6011 ctx->mirrors |= m->dup_mirrors;
6013 output_normal(ctx, m->out, vlan);
6014 } else if (vlan != m->out_vlan
6015 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6016 struct ofbundle *bundle;
6018 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6019 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6020 && !bundle->mirror_out) {
6021 output_normal(ctx, bundle, m->out_vlan);
6029 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6030 uint64_t packets, uint64_t bytes)
6036 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6039 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6042 /* In normal circumstances 'm' will not be NULL. However,
6043 * if mirrors are reconfigured, we can temporarily get out
6044 * of sync in facet_revalidate(). We could "correct" the
6045 * mirror list before reaching here, but doing that would
6046 * not properly account the traffic stats we've currently
6047 * accumulated for previous mirror configuration. */
6051 m->packet_count += packets;
6052 m->byte_count += bytes;
6056 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6057 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6058 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6060 is_gratuitous_arp(const struct flow *flow)
6062 return (flow->dl_type == htons(ETH_TYPE_ARP)
6063 && eth_addr_is_broadcast(flow->dl_dst)
6064 && (flow->nw_proto == ARP_OP_REPLY
6065 || (flow->nw_proto == ARP_OP_REQUEST
6066 && flow->nw_src == flow->nw_dst)));
6070 update_learning_table(struct ofproto_dpif *ofproto,
6071 const struct flow *flow, int vlan,
6072 struct ofbundle *in_bundle)
6074 struct mac_entry *mac;
6076 /* Don't learn the OFPP_NONE port. */
6077 if (in_bundle == &ofpp_none_bundle) {
6081 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6085 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6086 if (is_gratuitous_arp(flow)) {
6087 /* We don't want to learn from gratuitous ARP packets that are
6088 * reflected back over bond slaves so we lock the learning table. */
6089 if (!in_bundle->bond) {
6090 mac_entry_set_grat_arp_lock(mac);
6091 } else if (mac_entry_is_grat_arp_locked(mac)) {
6096 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6097 /* The log messages here could actually be useful in debugging,
6098 * so keep the rate limit relatively high. */
6099 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6100 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6101 "on port %s in VLAN %d",
6102 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6103 in_bundle->name, vlan);
6105 mac->port.p = in_bundle;
6106 tag_set_add(&ofproto->revalidate_set,
6107 mac_learning_changed(ofproto->ml, mac));
6111 static struct ofbundle *
6112 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6113 bool warn, struct ofport_dpif **in_ofportp)
6115 struct ofport_dpif *ofport;
6117 /* Find the port and bundle for the received packet. */
6118 ofport = get_ofp_port(ofproto, in_port);
6120 *in_ofportp = ofport;
6122 if (ofport && ofport->bundle) {
6123 return ofport->bundle;
6126 /* Special-case OFPP_NONE, which a controller may use as the ingress
6127 * port for traffic that it is sourcing. */
6128 if (in_port == OFPP_NONE) {
6129 return &ofpp_none_bundle;
6132 /* Odd. A few possible reasons here:
6134 * - We deleted a port but there are still a few packets queued up
6137 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6138 * we don't know about.
6140 * - The ofproto client didn't configure the port as part of a bundle.
6141 * This is particularly likely to happen if a packet was received on the
6142 * port after it was created, but before the client had a chance to
6143 * configure its bundle.
6146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6148 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6149 "port %"PRIu16, ofproto->up.name, in_port);
6154 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6155 * dropped. Returns true if they may be forwarded, false if they should be
6158 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6159 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6161 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6162 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6163 * checked by input_vid_is_valid().
6165 * May also add tags to '*tags', although the current implementation only does
6166 * so in one special case.
6169 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6172 struct ofproto_dpif *ofproto = ctx->ofproto;
6173 struct flow *flow = &ctx->flow;
6174 struct ofbundle *in_bundle = in_port->bundle;
6176 /* Drop frames for reserved multicast addresses
6177 * only if forward_bpdu option is absent. */
6178 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6179 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6183 if (in_bundle->bond) {
6184 struct mac_entry *mac;
6186 switch (bond_check_admissibility(in_bundle->bond, in_port,
6187 flow->dl_dst, &ctx->tags)) {
6192 xlate_report(ctx, "bonding refused admissibility, dropping");
6195 case BV_DROP_IF_MOVED:
6196 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6197 if (mac && mac->port.p != in_bundle &&
6198 (!is_gratuitous_arp(flow)
6199 || mac_entry_is_grat_arp_locked(mac))) {
6200 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6212 xlate_normal(struct action_xlate_ctx *ctx)
6214 struct ofport_dpif *in_port;
6215 struct ofbundle *in_bundle;
6216 struct mac_entry *mac;
6220 ctx->has_normal = true;
6222 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6223 ctx->packet != NULL, &in_port);
6225 xlate_report(ctx, "no input bundle, dropping");
6229 /* Drop malformed frames. */
6230 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6231 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6232 if (ctx->packet != NULL) {
6233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6234 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6235 "VLAN tag received on port %s",
6236 ctx->ofproto->up.name, in_bundle->name);
6238 xlate_report(ctx, "partial VLAN tag, dropping");
6242 /* Drop frames on bundles reserved for mirroring. */
6243 if (in_bundle->mirror_out) {
6244 if (ctx->packet != NULL) {
6245 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6246 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6247 "%s, which is reserved exclusively for mirroring",
6248 ctx->ofproto->up.name, in_bundle->name);
6250 xlate_report(ctx, "input port is mirror output port, dropping");
6255 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6256 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6257 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6260 vlan = input_vid_to_vlan(in_bundle, vid);
6262 /* Check other admissibility requirements. */
6263 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6267 /* Learn source MAC. */
6268 if (ctx->may_learn) {
6269 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6272 /* Determine output bundle. */
6273 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6276 if (mac->port.p != in_bundle) {
6277 xlate_report(ctx, "forwarding to learned port");
6278 output_normal(ctx, mac->port.p, vlan);
6280 xlate_report(ctx, "learned port is input port, dropping");
6283 struct ofbundle *bundle;
6285 xlate_report(ctx, "no learned MAC for destination, flooding");
6286 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6287 if (bundle != in_bundle
6288 && ofbundle_includes_vlan(bundle, vlan)
6289 && bundle->floodable
6290 && !bundle->mirror_out) {
6291 output_normal(ctx, bundle, vlan);
6294 ctx->nf_output_iface = NF_OUT_FLOOD;
6298 /* Optimized flow revalidation.
6300 * It's a difficult problem, in general, to tell which facets need to have
6301 * their actions recalculated whenever the OpenFlow flow table changes. We
6302 * don't try to solve that general problem: for most kinds of OpenFlow flow
6303 * table changes, we recalculate the actions for every facet. This is
6304 * relatively expensive, but it's good enough if the OpenFlow flow table
6305 * doesn't change very often.
6307 * However, we can expect one particular kind of OpenFlow flow table change to
6308 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6309 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6310 * table, we add a special case that applies to flow tables in which every rule
6311 * has the same form (that is, the same wildcards), except that the table is
6312 * also allowed to have a single "catch-all" flow that matches all packets. We
6313 * optimize this case by tagging all of the facets that resubmit into the table
6314 * and invalidating the same tag whenever a flow changes in that table. The
6315 * end result is that we revalidate just the facets that need it (and sometimes
6316 * a few more, but not all of the facets or even all of the facets that
6317 * resubmit to the table modified by MAC learning). */
6319 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6320 * into an OpenFlow table with the given 'basis'. */
6322 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6325 if (flow_wildcards_is_catchall(wc)) {
6328 struct flow tag_flow = *flow;
6329 flow_zero_wildcards(&tag_flow, wc);
6330 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6334 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6335 * taggability of that table.
6337 * This function must be called after *each* change to a flow table. If you
6338 * skip calling it on some changes then the pointer comparisons at the end can
6339 * be invalid if you get unlucky. For example, if a flow removal causes a
6340 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6341 * different wildcards to be created with the same address, then this function
6342 * will incorrectly skip revalidation. */
6344 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6346 struct table_dpif *table = &ofproto->tables[table_id];
6347 const struct oftable *oftable = &ofproto->up.tables[table_id];
6348 struct cls_table *catchall, *other;
6349 struct cls_table *t;
6351 catchall = other = NULL;
6353 switch (hmap_count(&oftable->cls.tables)) {
6355 /* We could tag this OpenFlow table but it would make the logic a
6356 * little harder and it's a corner case that doesn't seem worth it
6362 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6363 if (cls_table_is_catchall(t)) {
6365 } else if (!other) {
6368 /* Indicate that we can't tag this by setting both tables to
6369 * NULL. (We know that 'catchall' is already NULL.) */
6376 /* Can't tag this table. */
6380 if (table->catchall_table != catchall || table->other_table != other) {
6381 table->catchall_table = catchall;
6382 table->other_table = other;
6383 ofproto->need_revalidate = REV_FLOW_TABLE;
6387 /* Given 'rule' that has changed in some way (either it is a rule being
6388 * inserted, a rule being deleted, or a rule whose actions are being
6389 * modified), marks facets for revalidation to ensure that packets will be
6390 * forwarded correctly according to the new state of the flow table.
6392 * This function must be called after *each* change to a flow table. See
6393 * the comment on table_update_taggable() for more information. */
6395 rule_invalidate(const struct rule_dpif *rule)
6397 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6399 table_update_taggable(ofproto, rule->up.table_id);
6401 if (!ofproto->need_revalidate) {
6402 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6404 if (table->other_table && rule->tag) {
6405 tag_set_add(&ofproto->revalidate_set, rule->tag);
6407 ofproto->need_revalidate = REV_FLOW_TABLE;
6413 set_frag_handling(struct ofproto *ofproto_,
6414 enum ofp_config_flags frag_handling)
6416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6418 if (frag_handling != OFPC_FRAG_REASM) {
6419 ofproto->need_revalidate = REV_RECONFIGURE;
6427 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6428 const struct flow *flow,
6429 const struct ofpact *ofpacts, size_t ofpacts_len)
6431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6432 struct odputil_keybuf keybuf;
6433 struct dpif_flow_stats stats;
6437 struct action_xlate_ctx ctx;
6438 uint64_t odp_actions_stub[1024 / 8];
6439 struct ofpbuf odp_actions;
6441 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6442 odp_flow_key_from_flow(&key, flow);
6444 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6446 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6447 packet_get_tcp_flags(packet, flow), packet);
6448 ctx.resubmit_stats = &stats;
6450 ofpbuf_use_stub(&odp_actions,
6451 odp_actions_stub, sizeof odp_actions_stub);
6452 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6453 dpif_execute(ofproto->dpif, key.data, key.size,
6454 odp_actions.data, odp_actions.size, packet);
6455 ofpbuf_uninit(&odp_actions);
6463 set_netflow(struct ofproto *ofproto_,
6464 const struct netflow_options *netflow_options)
6466 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6468 if (netflow_options) {
6469 if (!ofproto->netflow) {
6470 ofproto->netflow = netflow_create();
6472 return netflow_set_options(ofproto->netflow, netflow_options);
6474 netflow_destroy(ofproto->netflow);
6475 ofproto->netflow = NULL;
6481 get_netflow_ids(const struct ofproto *ofproto_,
6482 uint8_t *engine_type, uint8_t *engine_id)
6484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6486 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6490 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6492 if (!facet_is_controller_flow(facet) &&
6493 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6494 struct subfacet *subfacet;
6495 struct ofexpired expired;
6497 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6498 if (subfacet->path == SF_FAST_PATH) {
6499 struct dpif_flow_stats stats;
6501 subfacet_reinstall(subfacet, &stats);
6502 subfacet_update_stats(subfacet, &stats);
6506 expired.flow = facet->flow;
6507 expired.packet_count = facet->packet_count;
6508 expired.byte_count = facet->byte_count;
6509 expired.used = facet->used;
6510 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6515 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6517 struct facet *facet;
6519 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6520 send_active_timeout(ofproto, facet);
6524 static struct ofproto_dpif *
6525 ofproto_dpif_lookup(const char *name)
6527 struct ofproto_dpif *ofproto;
6529 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6530 hash_string(name, 0), &all_ofproto_dpifs) {
6531 if (!strcmp(ofproto->up.name, name)) {
6539 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6540 const char *argv[], void *aux OVS_UNUSED)
6542 struct ofproto_dpif *ofproto;
6545 ofproto = ofproto_dpif_lookup(argv[1]);
6547 unixctl_command_reply_error(conn, "no such bridge");
6550 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6552 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6553 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6557 unixctl_command_reply(conn, "table successfully flushed");
6561 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6562 const char *argv[], void *aux OVS_UNUSED)
6564 struct ds ds = DS_EMPTY_INITIALIZER;
6565 const struct ofproto_dpif *ofproto;
6566 const struct mac_entry *e;
6568 ofproto = ofproto_dpif_lookup(argv[1]);
6570 unixctl_command_reply_error(conn, "no such bridge");
6574 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6575 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6576 struct ofbundle *bundle = e->port.p;
6577 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6578 ofbundle_get_a_port(bundle)->odp_port,
6579 e->vlan, ETH_ADDR_ARGS(e->mac),
6580 mac_entry_age(ofproto->ml, e));
6582 unixctl_command_reply(conn, ds_cstr(&ds));
6587 struct action_xlate_ctx ctx;
6593 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6594 const struct rule_dpif *rule)
6596 ds_put_char_multiple(result, '\t', level);
6598 ds_put_cstr(result, "No match\n");
6602 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6603 table_id, ntohll(rule->up.flow_cookie));
6604 cls_rule_format(&rule->up.cr, result);
6605 ds_put_char(result, '\n');
6607 ds_put_char_multiple(result, '\t', level);
6608 ds_put_cstr(result, "OpenFlow ");
6609 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
6610 ds_put_char(result, '\n');
6614 trace_format_flow(struct ds *result, int level, const char *title,
6615 struct trace_ctx *trace)
6617 ds_put_char_multiple(result, '\t', level);
6618 ds_put_format(result, "%s: ", title);
6619 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6620 ds_put_cstr(result, "unchanged");
6622 flow_format(result, &trace->ctx.flow);
6623 trace->flow = trace->ctx.flow;
6625 ds_put_char(result, '\n');
6629 trace_format_regs(struct ds *result, int level, const char *title,
6630 struct trace_ctx *trace)
6634 ds_put_char_multiple(result, '\t', level);
6635 ds_put_format(result, "%s:", title);
6636 for (i = 0; i < FLOW_N_REGS; i++) {
6637 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6639 ds_put_char(result, '\n');
6643 trace_format_odp(struct ds *result, int level, const char *title,
6644 struct trace_ctx *trace)
6646 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6648 ds_put_char_multiple(result, '\t', level);
6649 ds_put_format(result, "%s: ", title);
6650 format_odp_actions(result, odp_actions->data, odp_actions->size);
6651 ds_put_char(result, '\n');
6655 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6657 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6658 struct ds *result = trace->result;
6660 ds_put_char(result, '\n');
6661 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6662 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6663 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6664 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6668 trace_report(struct action_xlate_ctx *ctx, const char *s)
6670 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6671 struct ds *result = trace->result;
6673 ds_put_char_multiple(result, '\t', ctx->recurse);
6674 ds_put_cstr(result, s);
6675 ds_put_char(result, '\n');
6679 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6680 void *aux OVS_UNUSED)
6682 const char *dpname = argv[1];
6683 struct ofproto_dpif *ofproto;
6684 struct ofpbuf odp_key;
6685 struct ofpbuf *packet;
6686 ovs_be16 initial_tci;
6692 ofpbuf_init(&odp_key, 0);
6695 ofproto = ofproto_dpif_lookup(dpname);
6697 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6701 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6702 /* ofproto/trace dpname flow [-generate] */
6703 const char *flow_s = argv[2];
6704 const char *generate_s = argv[3];
6706 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6707 * flow. We guess which type it is based on whether 'flow_s' contains
6708 * an '(', since a datapath flow always contains '(') but an
6709 * OpenFlow-like flow should not (in fact it's allowed but I believe
6710 * that's not documented anywhere).
6712 * An alternative would be to try to parse 'flow_s' both ways, but then
6713 * it would be tricky giving a sensible error message. After all, do
6714 * you just say "syntax error" or do you present both error messages?
6715 * Both choices seem lousy. */
6716 if (strchr(flow_s, '(')) {
6719 /* Convert string to datapath key. */
6720 ofpbuf_init(&odp_key, 0);
6721 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6723 unixctl_command_reply_error(conn, "Bad flow syntax");
6727 /* Convert odp_key to flow. */
6728 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6729 odp_key.size, &flow,
6730 &initial_tci, NULL);
6731 if (error == ODP_FIT_ERROR) {
6732 unixctl_command_reply_error(conn, "Invalid flow");
6738 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6740 unixctl_command_reply_error(conn, error_s);
6745 initial_tci = flow.vlan_tci;
6746 vsp_adjust_flow(ofproto, &flow);
6749 /* Generate a packet, if requested. */
6751 packet = ofpbuf_new(0);
6752 flow_compose(packet, &flow);
6754 } else if (argc == 6) {
6755 /* ofproto/trace dpname priority tun_id in_port packet */
6756 const char *priority_s = argv[2];
6757 const char *tun_id_s = argv[3];
6758 const char *in_port_s = argv[4];
6759 const char *packet_s = argv[5];
6760 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6761 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6762 uint32_t priority = atoi(priority_s);
6765 msg = eth_from_hex(packet_s, &packet);
6767 unixctl_command_reply_error(conn, msg);
6771 ds_put_cstr(&result, "Packet: ");
6772 s = ofp_packet_to_string(packet->data, packet->size);
6773 ds_put_cstr(&result, s);
6776 flow_extract(packet, priority, tun_id, in_port, &flow);
6777 initial_tci = flow.vlan_tci;
6779 unixctl_command_reply_error(conn, "Bad command syntax");
6783 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6784 unixctl_command_reply(conn, ds_cstr(&result));
6787 ds_destroy(&result);
6788 ofpbuf_delete(packet);
6789 ofpbuf_uninit(&odp_key);
6793 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6794 const struct ofpbuf *packet, ovs_be16 initial_tci,
6797 struct rule_dpif *rule;
6799 ds_put_cstr(ds, "Flow: ");
6800 flow_format(ds, flow);
6801 ds_put_char(ds, '\n');
6803 rule = rule_dpif_lookup(ofproto, flow);
6805 trace_format_rule(ds, 0, 0, rule);
6806 if (rule == ofproto->miss_rule) {
6807 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6808 } else if (rule == ofproto->no_packet_in_rule) {
6809 ds_put_cstr(ds, "\nNo match, packets dropped because "
6810 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6814 uint64_t odp_actions_stub[1024 / 8];
6815 struct ofpbuf odp_actions;
6817 struct trace_ctx trace;
6820 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6823 ofpbuf_use_stub(&odp_actions,
6824 odp_actions_stub, sizeof odp_actions_stub);
6825 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6826 rule, tcp_flags, packet);
6827 trace.ctx.resubmit_hook = trace_resubmit;
6828 trace.ctx.report_hook = trace_report;
6829 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
6832 ds_put_char(ds, '\n');
6833 trace_format_flow(ds, 0, "Final flow", &trace);
6834 ds_put_cstr(ds, "Datapath actions: ");
6835 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6836 ofpbuf_uninit(&odp_actions);
6838 if (trace.ctx.slow) {
6839 enum slow_path_reason slow;
6841 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6842 "slow path because it:");
6843 for (slow = trace.ctx.slow; slow; ) {
6844 enum slow_path_reason bit = rightmost_1bit(slow);
6848 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6851 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6854 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6857 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6860 ds_put_cstr(ds, "\n\t (The datapath actions are "
6861 "incomplete--for complete actions, "
6862 "please supply a packet.)");
6865 case SLOW_CONTROLLER:
6866 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6867 "to the OpenFlow controller.");
6870 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6871 "than the datapath supports.");
6878 if (slow & ~SLOW_MATCH) {
6879 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6880 "the special slow-path processing.");
6887 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6888 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6891 unixctl_command_reply(conn, NULL);
6895 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6896 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6899 unixctl_command_reply(conn, NULL);
6902 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6903 * 'reply' describing the results. */
6905 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6907 struct facet *facet;
6911 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6912 if (!facet_check_consistency(facet)) {
6917 ofproto->need_revalidate = REV_INCONSISTENCY;
6921 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6922 ofproto->up.name, errors);
6924 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6929 ofproto_dpif_self_check(struct unixctl_conn *conn,
6930 int argc, const char *argv[], void *aux OVS_UNUSED)
6932 struct ds reply = DS_EMPTY_INITIALIZER;
6933 struct ofproto_dpif *ofproto;
6936 ofproto = ofproto_dpif_lookup(argv[1]);
6938 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6939 "ofproto/list for help)");
6942 ofproto_dpif_self_check__(ofproto, &reply);
6944 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6945 ofproto_dpif_self_check__(ofproto, &reply);
6949 unixctl_command_reply(conn, ds_cstr(&reply));
6954 ofproto_dpif_unixctl_init(void)
6956 static bool registered;
6962 unixctl_command_register(
6964 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6965 2, 5, ofproto_unixctl_trace, NULL);
6966 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6967 ofproto_unixctl_fdb_flush, NULL);
6968 unixctl_command_register("fdb/show", "bridge", 1, 1,
6969 ofproto_unixctl_fdb_show, NULL);
6970 unixctl_command_register("ofproto/clog", "", 0, 0,
6971 ofproto_dpif_clog, NULL);
6972 unixctl_command_register("ofproto/unclog", "", 0, 0,
6973 ofproto_dpif_unclog, NULL);
6974 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6975 ofproto_dpif_self_check, NULL);
6978 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6980 * This is deprecated. It is only for compatibility with broken device drivers
6981 * in old versions of Linux that do not properly support VLANs when VLAN
6982 * devices are not used. When broken device drivers are no longer in
6983 * widespread use, we will delete these interfaces. */
6986 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6989 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6991 if (realdev_ofp_port == ofport->realdev_ofp_port
6992 && vid == ofport->vlandev_vid) {
6996 ofproto->need_revalidate = REV_RECONFIGURE;
6998 if (ofport->realdev_ofp_port) {
7001 if (realdev_ofp_port && ofport->bundle) {
7002 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7003 * themselves be part of a bundle. */
7004 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7007 ofport->realdev_ofp_port = realdev_ofp_port;
7008 ofport->vlandev_vid = vid;
7010 if (realdev_ofp_port) {
7011 vsp_add(ofport, realdev_ofp_port, vid);
7018 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7020 return hash_2words(realdev_ofp_port, vid);
7023 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7024 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7025 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7026 * it would return the port number of eth0.9.
7028 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7029 * function just returns its 'realdev_odp_port' argument. */
7031 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7032 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7034 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7035 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
7036 int vid = vlan_tci_to_vid(vlan_tci);
7037 const struct vlan_splinter *vsp;
7039 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7040 hash_realdev_vid(realdev_ofp_port, vid),
7041 &ofproto->realdev_vid_map) {
7042 if (vsp->realdev_ofp_port == realdev_ofp_port
7043 && vsp->vid == vid) {
7044 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
7048 return realdev_odp_port;
7051 static struct vlan_splinter *
7052 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7054 struct vlan_splinter *vsp;
7056 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7057 &ofproto->vlandev_map) {
7058 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7066 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7067 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7068 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7069 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7070 * eth0 and store 9 in '*vid'.
7072 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7073 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7076 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7077 uint16_t vlandev_ofp_port, int *vid)
7079 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7080 const struct vlan_splinter *vsp;
7082 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7087 return vsp->realdev_ofp_port;
7093 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7094 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7095 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7096 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7097 * always the case unless VLAN splinters are enabled), returns false without
7098 * making any changes. */
7100 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7105 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7110 /* Cause the flow to be processed as if it came in on the real device with
7111 * the VLAN device's VLAN ID. */
7112 flow->in_port = realdev;
7113 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7118 vsp_remove(struct ofport_dpif *port)
7120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7121 struct vlan_splinter *vsp;
7123 vsp = vlandev_find(ofproto, port->up.ofp_port);
7125 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7126 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7129 port->realdev_ofp_port = 0;
7131 VLOG_ERR("missing vlan device record");
7136 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7138 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7140 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7141 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7142 == realdev_ofp_port)) {
7143 struct vlan_splinter *vsp;
7145 vsp = xmalloc(sizeof *vsp);
7146 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7147 hash_int(port->up.ofp_port, 0));
7148 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7149 hash_realdev_vid(realdev_ofp_port, vid));
7150 vsp->realdev_ofp_port = realdev_ofp_port;
7151 vsp->vlandev_ofp_port = port->up.ofp_port;
7154 port->realdev_ofp_port = realdev_ofp_port;
7156 VLOG_ERR("duplicate vlan device record");
7160 const struct ofproto_class ofproto_dpif_class = {
7190 port_is_lacp_current,
7191 NULL, /* rule_choose_table */
7198 rule_modify_actions,
7207 get_cfm_remote_mpids,
7212 get_stp_port_status,
7219 is_mirror_output_bundle,
7220 forward_bpdu_changed,