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-parse.h"
46 #include "ofp-print.h"
47 #include "ofproto-dpif-governor.h"
48 #include "ofproto-dpif-sflow.h"
49 #include "poll-loop.h"
52 #include "unaligned.h"
54 #include "vlan-bitmap.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
59 COVERAGE_DEFINE(ofproto_dpif_expired);
60 COVERAGE_DEFINE(ofproto_dpif_xlate);
61 COVERAGE_DEFINE(facet_changed_rule);
62 COVERAGE_DEFINE(facet_revalidate);
63 COVERAGE_DEFINE(facet_unexpected);
64 COVERAGE_DEFINE(facet_suppress);
66 /* Maximum depth of flow table recursion (due to resubmit actions) in a
67 * flow translation. */
68 #define MAX_RESUBMIT_RECURSION 32
70 /* Number of implemented OpenFlow tables. */
71 enum { N_TABLES = 255 };
72 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
73 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
83 * - Do include packets and bytes from facets that have been deleted or
84 * whose own statistics have been folded into the rule.
86 * - Do include packets and bytes sent "by hand" that were accounted to
87 * the rule without any facet being involved (this is a rare corner
88 * case in rule_execute()).
90 * - Do not include packet or bytes that can be obtained from any facet's
91 * packet_count or byte_count member or that can be obtained from the
92 * datapath by, e.g., dpif_flow_get() for any subfacet.
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
97 tag_type tag; /* Caches rule_calculate_tag() result. */
99 struct list facets; /* List of "struct facet"s. */
102 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
104 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
107 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
108 const struct flow *);
109 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
113 static void rule_credit_stats(struct rule_dpif *,
114 const struct dpif_flow_stats *);
115 static void flow_push_stats(struct rule_dpif *, const struct flow *,
116 const struct dpif_flow_stats *);
117 static tag_type rule_calculate_tag(const struct flow *,
118 const struct flow_wildcards *,
120 static void rule_invalidate(const struct rule_dpif *);
122 #define MAX_MIRRORS 32
123 typedef uint32_t mirror_mask_t;
124 #define MIRROR_MASK_C(X) UINT32_C(X)
125 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
127 struct ofproto_dpif *ofproto; /* Owning ofproto. */
128 size_t idx; /* In ofproto's "mirrors" array. */
129 void *aux; /* Key supplied by ofproto's client. */
130 char *name; /* Identifier for log messages. */
132 /* Selection criteria. */
133 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
134 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
135 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
137 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
138 struct ofbundle *out; /* Output port or NULL. */
139 int out_vlan; /* Output VLAN or -1. */
140 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
143 int64_t packet_count; /* Number of packets sent. */
144 int64_t byte_count; /* Number of bytes sent. */
147 static void mirror_destroy(struct ofmirror *);
148 static void update_mirror_stats(struct ofproto_dpif *ofproto,
149 mirror_mask_t mirrors,
150 uint64_t packets, uint64_t bytes);
153 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
154 struct ofproto_dpif *ofproto; /* Owning ofproto. */
155 void *aux; /* Key supplied by ofproto's client. */
156 char *name; /* Identifier for log messages. */
159 struct list ports; /* Contains "struct ofport"s. */
160 enum port_vlan_mode vlan_mode; /* VLAN mode */
161 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
162 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
163 * NULL if all VLANs are trunked. */
164 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
165 struct bond *bond; /* Nonnull iff more than one port. */
166 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
169 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
171 /* Port mirroring info. */
172 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
173 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
174 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
177 static void bundle_remove(struct ofport *);
178 static void bundle_update(struct ofbundle *);
179 static void bundle_destroy(struct ofbundle *);
180 static void bundle_del_port(struct ofport_dpif *);
181 static void bundle_run(struct ofbundle *);
182 static void bundle_wait(struct ofbundle *);
183 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
184 uint16_t in_port, bool warn,
185 struct ofport_dpif **in_ofportp);
187 /* A controller may use OFPP_NONE as the ingress port to indicate that
188 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
189 * when an input bundle is needed for validation (e.g., mirroring or
190 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
191 * any 'port' structs, so care must be taken when dealing with it. */
192 static struct ofbundle ofpp_none_bundle = {
194 .vlan_mode = PORT_VLAN_TRUNK
197 static void stp_run(struct ofproto_dpif *ofproto);
198 static void stp_wait(struct ofproto_dpif *ofproto);
199 static int set_stp_port(struct ofport *,
200 const struct ofproto_port_stp_settings *);
202 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
204 struct action_xlate_ctx {
205 /* action_xlate_ctx_init() initializes these members. */
208 struct ofproto_dpif *ofproto;
210 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
211 * this flow when actions change header fields. */
214 /* The packet corresponding to 'flow', or a null pointer if we are
215 * revalidating without a packet to refer to. */
216 const struct ofpbuf *packet;
218 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
219 * actions update the flow table?
221 * We want to update these tables if we are actually processing a packet,
222 * or if we are accounting for packets that the datapath has processed, but
223 * not if we are just revalidating. */
226 /* The rule that we are currently translating, or NULL. */
227 struct rule_dpif *rule;
229 /* Union of the set of TCP flags seen so far in this flow. (Used only by
230 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
234 /* If nonnull, flow translation calls this function just before executing a
235 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
236 * when the recursion depth is exceeded.
238 * 'rule' is the rule being submitted into. It will be null if the
239 * resubmit or OFPP_TABLE action didn't find a matching rule.
241 * This is normally null so the client has to set it manually after
242 * calling action_xlate_ctx_init(). */
243 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
245 /* If nonnull, flow translation credits the specified statistics to each
246 * rule reached through a resubmit or OFPP_TABLE action.
248 * This is normally null so the client has to set it manually after
249 * calling action_xlate_ctx_init(). */
250 const struct dpif_flow_stats *resubmit_stats;
252 /* xlate_actions() initializes and uses these members. The client might want
253 * to look at them after it returns. */
255 struct ofpbuf *odp_actions; /* Datapath actions. */
256 tag_type tags; /* Tags associated with actions. */
257 enum slow_path_reason slow; /* 0 if fast path may be used. */
258 bool has_learn; /* Actions include NXAST_LEARN? */
259 bool has_normal; /* Actions output to OFPP_NORMAL? */
260 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
261 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
262 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
264 /* xlate_actions() initializes and uses these members, but the client has no
265 * reason to look at them. */
267 int recurse; /* Recursion level, via xlate_table_action. */
268 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
269 struct flow base_flow; /* Flow at the last commit. */
270 uint32_t orig_skb_priority; /* Priority when packet arrived. */
271 uint8_t table_id; /* OpenFlow table ID where flow was found. */
272 uint32_t sflow_n_outputs; /* Number of output ports. */
273 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
274 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
275 bool exit; /* No further actions should be processed. */
276 struct flow orig_flow; /* Copy of original flow. */
279 static void action_xlate_ctx_init(struct action_xlate_ctx *,
280 struct ofproto_dpif *, const struct flow *,
281 ovs_be16 initial_tci, struct rule_dpif *,
282 uint8_t tcp_flags, const struct ofpbuf *);
283 static void xlate_actions(struct action_xlate_ctx *,
284 const union ofp_action *in, size_t n_in,
285 struct ofpbuf *odp_actions);
286 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
287 const union ofp_action *in,
290 static size_t put_userspace_action(const struct ofproto_dpif *,
291 struct ofpbuf *odp_actions,
293 const union user_action_cookie *);
295 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
296 enum slow_path_reason,
297 uint64_t *stub, size_t stub_size,
298 const struct nlattr **actionsp,
299 size_t *actions_lenp);
301 /* A subfacet (see "struct subfacet" below) has three possible installation
304 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
305 * case just after the subfacet is created, just before the subfacet is
306 * destroyed, or if the datapath returns an error when we try to install a
309 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
311 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
312 * ofproto_dpif is installed in the datapath.
315 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
316 SF_FAST_PATH, /* Full actions are installed. */
317 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
320 static const char *subfacet_path_to_string(enum subfacet_path);
322 /* A dpif flow and actions associated with a facet.
324 * See also the large comment on struct facet. */
327 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
328 struct list list_node; /* In struct facet's 'facets' list. */
329 struct facet *facet; /* Owning facet. */
333 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
334 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
335 * regenerate the ODP flow key from ->facet->flow. */
336 enum odp_key_fitness key_fitness;
340 long long int used; /* Time last used; time created if not used. */
342 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
343 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
347 * These should be essentially identical for every subfacet in a facet, but
348 * may differ in trivial ways due to VLAN splinters. */
349 size_t actions_len; /* Number of bytes in actions[]. */
350 struct nlattr *actions; /* Datapath actions. */
352 enum slow_path_reason slow; /* 0 if fast path may be used. */
353 enum subfacet_path path; /* Installed in datapath? */
355 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
356 * splinters can cause it to differ. This value should be removed when
357 * the VLAN splinters feature is no longer needed. */
358 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
361 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
362 const struct nlattr *key,
363 size_t key_len, ovs_be16 initial_tci);
364 static struct subfacet *subfacet_find(struct ofproto_dpif *,
365 const struct nlattr *key, size_t key_len);
366 static void subfacet_destroy(struct subfacet *);
367 static void subfacet_destroy__(struct subfacet *);
368 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
370 static void subfacet_reset_dp_stats(struct subfacet *,
371 struct dpif_flow_stats *);
372 static void subfacet_update_time(struct subfacet *, long long int used);
373 static void subfacet_update_stats(struct subfacet *,
374 const struct dpif_flow_stats *);
375 static void subfacet_make_actions(struct subfacet *,
376 const struct ofpbuf *packet,
377 struct ofpbuf *odp_actions);
378 static int subfacet_install(struct subfacet *,
379 const struct nlattr *actions, size_t actions_len,
380 struct dpif_flow_stats *, enum slow_path_reason);
381 static void subfacet_uninstall(struct subfacet *);
383 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
385 /* An exact-match instantiation of an OpenFlow flow.
387 * A facet associates a "struct flow", which represents the Open vSwitch
388 * userspace idea of an exact-match flow, with one or more subfacets. Each
389 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
390 * the facet. When the kernel module (or other dpif implementation) and Open
391 * vSwitch userspace agree on the definition of a flow key, there is exactly
392 * one subfacet per facet. If the dpif implementation supports more-specific
393 * flow matching than userspace, however, a facet can have more than one
394 * subfacet, each of which corresponds to some distinction in flow that
395 * userspace simply doesn't understand.
397 * Flow expiration works in terms of subfacets, so a facet must have at least
398 * one subfacet or it will never expire, leaking memory. */
401 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
402 struct list list_node; /* In owning rule's 'facets' list. */
403 struct rule_dpif *rule; /* Owning rule. */
406 struct list subfacets;
407 long long int used; /* Time last used; time created if not used. */
414 * - Do include packets and bytes sent "by hand", e.g. with
417 * - Do include packets and bytes that were obtained from the datapath
418 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
419 * DPIF_FP_ZERO_STATS).
421 * - Do not include packets or bytes that can be obtained from the
422 * datapath for any existing subfacet.
424 uint64_t packet_count; /* Number of packets received. */
425 uint64_t byte_count; /* Number of bytes received. */
427 /* Resubmit statistics. */
428 uint64_t prev_packet_count; /* Number of packets from last stats push. */
429 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
430 long long int prev_used; /* Used time from last stats push. */
433 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
434 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
435 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
437 /* Properties of datapath actions.
439 * Every subfacet has its own actions because actions can differ slightly
440 * between splintered and non-splintered subfacets due to the VLAN tag
441 * being initially different (present vs. absent). All of them have these
442 * properties in common so we just store one copy of them here. */
443 bool has_learn; /* Actions include NXAST_LEARN? */
444 bool has_normal; /* Actions output to OFPP_NORMAL? */
445 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
446 tag_type tags; /* Tags that would require revalidation. */
447 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
449 /* Storage for a single subfacet, to reduce malloc() time and space
450 * overhead. (A facet always has at least one subfacet and in the common
451 * case has exactly one subfacet.) */
452 struct subfacet one_subfacet;
455 static struct facet *facet_create(struct rule_dpif *,
456 const struct flow *, uint32_t hash);
457 static void facet_remove(struct facet *);
458 static void facet_free(struct facet *);
460 static struct facet *facet_find(struct ofproto_dpif *,
461 const struct flow *, uint32_t hash);
462 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
463 const struct flow *, uint32_t hash);
464 static void facet_revalidate(struct facet *);
465 static bool facet_check_consistency(struct facet *);
467 static void facet_flush_stats(struct facet *);
469 static void facet_update_time(struct facet *, long long int used);
470 static void facet_reset_counters(struct facet *);
471 static void facet_push_stats(struct facet *);
472 static void facet_learn(struct facet *);
473 static void facet_account(struct facet *);
475 static bool facet_is_controller_flow(struct facet *);
481 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
482 struct list bundle_node; /* In struct ofbundle's "ports" list. */
483 struct cfm *cfm; /* Connectivity Fault Management, if any. */
484 tag_type tag; /* Tag associated with this port. */
485 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
486 bool may_enable; /* May be enabled in bonds. */
487 long long int carrier_seq; /* Carrier status changes. */
490 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
491 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
492 long long int stp_state_entered;
494 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
496 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
498 * This is deprecated. It is only for compatibility with broken device
499 * drivers in old versions of Linux that do not properly support VLANs when
500 * VLAN devices are not used. When broken device drivers are no longer in
501 * widespread use, we will delete these interfaces. */
502 uint16_t realdev_ofp_port;
506 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
507 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
508 * traffic egressing the 'ofport' with that priority should be marked with. */
509 struct priority_to_dscp {
510 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
511 uint32_t priority; /* Priority of this queue (see struct flow). */
513 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
516 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
518 * This is deprecated. It is only for compatibility with broken device drivers
519 * in old versions of Linux that do not properly support VLANs when VLAN
520 * devices are not used. When broken device drivers are no longer in
521 * widespread use, we will delete these interfaces. */
522 struct vlan_splinter {
523 struct hmap_node realdev_vid_node;
524 struct hmap_node vlandev_node;
525 uint16_t realdev_ofp_port;
526 uint16_t vlandev_ofp_port;
530 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
531 uint32_t realdev, ovs_be16 vlan_tci);
532 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
533 static void vsp_remove(struct ofport_dpif *);
534 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
536 static struct ofport_dpif *
537 ofport_dpif_cast(const struct ofport *ofport)
539 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
540 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
543 static void port_run(struct ofport_dpif *);
544 static void port_wait(struct ofport_dpif *);
545 static int set_cfm(struct ofport *, const struct cfm_settings *);
546 static void ofport_clear_priorities(struct ofport_dpif *);
548 struct dpif_completion {
549 struct list list_node;
550 struct ofoperation *op;
553 /* Extra information about a classifier table.
554 * Currently used just for optimized flow revalidation. */
556 /* If either of these is nonnull, then this table has a form that allows
557 * flows to be tagged to avoid revalidating most flows for the most common
558 * kinds of flow table changes. */
559 struct cls_table *catchall_table; /* Table that wildcards all fields. */
560 struct cls_table *other_table; /* Table with any other wildcard set. */
561 uint32_t basis; /* Keeps each table's tags separate. */
564 /* Reasons that we might need to revalidate every facet, and corresponding
567 * A value of 0 means that there is no need to revalidate.
569 * It would be nice to have some cleaner way to integrate with coverage
570 * counters, but with only a few reasons I guess this is good enough for
572 enum revalidate_reason {
573 REV_RECONFIGURE = 1, /* Switch configuration changed. */
574 REV_STP, /* Spanning tree protocol port status change. */
575 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
576 REV_FLOW_TABLE, /* Flow table changed. */
577 REV_INCONSISTENCY /* Facet self-check failed. */
579 COVERAGE_DEFINE(rev_reconfigure);
580 COVERAGE_DEFINE(rev_stp);
581 COVERAGE_DEFINE(rev_port_toggled);
582 COVERAGE_DEFINE(rev_flow_table);
583 COVERAGE_DEFINE(rev_inconsistency);
585 struct ofproto_dpif {
586 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
591 /* Special OpenFlow rules. */
592 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
593 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
599 struct netflow *netflow;
600 struct dpif_sflow *sflow;
601 struct hmap bundles; /* Contains "struct ofbundle"s. */
602 struct mac_learning *ml;
603 struct ofmirror *mirrors[MAX_MIRRORS];
605 bool has_bonded_bundles;
608 struct timer next_expiration;
612 struct hmap subfacets;
613 struct governor *governor;
616 struct table_dpif tables[N_TABLES];
617 enum revalidate_reason need_revalidate;
618 struct tag_set revalidate_set;
620 /* Support for debugging async flow mods. */
621 struct list completions;
623 bool has_bundle_action; /* True when the first bundle action appears. */
624 struct netdev_stats stats; /* To account packets generated and consumed in
629 long long int stp_last_tick;
631 /* VLAN splinters. */
632 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
633 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
636 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
637 * for debugging the asynchronous flow_mod implementation.) */
640 /* All existing ofproto_dpif instances, indexed by ->up.name. */
641 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
643 static void ofproto_dpif_unixctl_init(void);
645 static struct ofproto_dpif *
646 ofproto_dpif_cast(const struct ofproto *ofproto)
648 assert(ofproto->ofproto_class == &ofproto_dpif_class);
649 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
652 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
654 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
656 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
657 const struct ofpbuf *, ovs_be16 initial_tci,
660 /* Packet processing. */
661 static void update_learning_table(struct ofproto_dpif *,
662 const struct flow *, int vlan,
665 #define FLOW_MISS_MAX_BATCH 50
666 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
668 /* Flow expiration. */
669 static int expire(struct ofproto_dpif *);
672 static void send_netflow_active_timeouts(struct ofproto_dpif *);
675 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
676 static size_t compose_sflow_action(const struct ofproto_dpif *,
677 struct ofpbuf *odp_actions,
678 const struct flow *, uint32_t odp_port);
679 static void add_mirror_actions(struct action_xlate_ctx *ctx,
680 const struct flow *flow);
681 /* Global variables. */
682 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
684 /* Factory functions. */
687 enumerate_types(struct sset *types)
689 dp_enumerate_types(types);
693 enumerate_names(const char *type, struct sset *names)
695 return dp_enumerate_names(type, names);
699 del(const char *type, const char *name)
704 error = dpif_open(name, type, &dpif);
706 error = dpif_delete(dpif);
712 /* Basic life-cycle. */
714 static int add_internal_flows(struct ofproto_dpif *);
716 static struct ofproto *
719 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
724 dealloc(struct ofproto *ofproto_)
726 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
731 construct(struct ofproto *ofproto_)
733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
734 const char *name = ofproto->up.name;
738 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
740 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
744 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
745 ofproto->n_matches = 0;
747 dpif_flow_flush(ofproto->dpif);
748 dpif_recv_purge(ofproto->dpif);
750 error = dpif_recv_set(ofproto->dpif, true);
752 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
753 dpif_close(ofproto->dpif);
757 ofproto->netflow = NULL;
758 ofproto->sflow = NULL;
760 hmap_init(&ofproto->bundles);
761 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
762 for (i = 0; i < MAX_MIRRORS; i++) {
763 ofproto->mirrors[i] = NULL;
765 ofproto->has_bonded_bundles = false;
767 timer_set_duration(&ofproto->next_expiration, 1000);
769 hmap_init(&ofproto->facets);
770 hmap_init(&ofproto->subfacets);
771 ofproto->governor = NULL;
773 for (i = 0; i < N_TABLES; i++) {
774 struct table_dpif *table = &ofproto->tables[i];
776 table->catchall_table = NULL;
777 table->other_table = NULL;
778 table->basis = random_uint32();
780 ofproto->need_revalidate = 0;
781 tag_set_init(&ofproto->revalidate_set);
783 list_init(&ofproto->completions);
785 ofproto_dpif_unixctl_init();
787 ofproto->has_mirrors = false;
788 ofproto->has_bundle_action = false;
790 hmap_init(&ofproto->vlandev_map);
791 hmap_init(&ofproto->realdev_vid_map);
793 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
794 hash_string(ofproto->up.name, 0));
795 memset(&ofproto->stats, 0, sizeof ofproto->stats);
797 ofproto_init_tables(ofproto_, N_TABLES);
798 error = add_internal_flows(ofproto);
799 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
805 add_internal_flow(struct ofproto_dpif *ofproto, int id,
806 const struct ofpbuf *actions, struct rule_dpif **rulep)
808 struct ofputil_flow_mod fm;
811 cls_rule_init_catchall(&fm.cr, 0);
812 cls_rule_set_reg(&fm.cr, 0, id);
813 fm.new_cookie = htonll(0);
814 fm.cookie = htonll(0);
815 fm.cookie_mask = htonll(0);
816 fm.table_id = TBL_INTERNAL;
817 fm.command = OFPFC_ADD;
823 fm.actions = actions->data;
824 fm.n_actions = actions->size / sizeof(union ofp_action);
826 error = ofproto_flow_mod(&ofproto->up, &fm);
828 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
829 id, ofperr_to_string(error));
833 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
834 assert(*rulep != NULL);
840 add_internal_flows(struct ofproto_dpif *ofproto)
842 struct nx_action_controller *nac;
843 uint64_t actions_stub[128 / 8];
844 struct ofpbuf actions;
848 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
851 nac = ofputil_put_NXAST_CONTROLLER(&actions);
852 nac->max_len = htons(UINT16_MAX);
853 nac->controller_id = htons(0);
854 nac->reason = OFPR_NO_MATCH;
855 error = add_internal_flow(ofproto, id++, &actions, &ofproto->miss_rule);
860 ofpbuf_clear(&actions);
861 error = add_internal_flow(ofproto, id++, &actions,
862 &ofproto->no_packet_in_rule);
867 complete_operations(struct ofproto_dpif *ofproto)
869 struct dpif_completion *c, *next;
871 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
872 ofoperation_complete(c->op, 0);
873 list_remove(&c->list_node);
879 destruct(struct ofproto *ofproto_)
881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
882 struct rule_dpif *rule, *next_rule;
883 struct oftable *table;
886 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
887 complete_operations(ofproto);
889 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
890 struct cls_cursor cursor;
892 cls_cursor_init(&cursor, &table->cls, NULL);
893 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
894 ofproto_rule_destroy(&rule->up);
898 for (i = 0; i < MAX_MIRRORS; i++) {
899 mirror_destroy(ofproto->mirrors[i]);
902 netflow_destroy(ofproto->netflow);
903 dpif_sflow_destroy(ofproto->sflow);
904 hmap_destroy(&ofproto->bundles);
905 mac_learning_destroy(ofproto->ml);
907 hmap_destroy(&ofproto->facets);
908 hmap_destroy(&ofproto->subfacets);
909 governor_destroy(ofproto->governor);
911 hmap_destroy(&ofproto->vlandev_map);
912 hmap_destroy(&ofproto->realdev_vid_map);
914 dpif_close(ofproto->dpif);
918 run_fast(struct ofproto *ofproto_)
920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
923 /* Handle one or more batches of upcalls, until there's nothing left to do
924 * or until we do a fixed total amount of work.
926 * We do work in batches because it can be much cheaper to set up a number
927 * of flows and fire off their patches all at once. We do multiple batches
928 * because in some cases handling a packet can cause another packet to be
929 * queued almost immediately as part of the return flow. Both
930 * optimizations can make major improvements on some benchmarks and
931 * presumably for real traffic as well. */
933 while (work < FLOW_MISS_MAX_BATCH) {
934 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
944 run(struct ofproto *ofproto_)
946 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
947 struct ofport_dpif *ofport;
948 struct ofbundle *bundle;
952 complete_operations(ofproto);
954 dpif_run(ofproto->dpif);
956 error = run_fast(ofproto_);
961 if (timer_expired(&ofproto->next_expiration)) {
962 int delay = expire(ofproto);
963 timer_set_duration(&ofproto->next_expiration, delay);
966 if (ofproto->netflow) {
967 if (netflow_run(ofproto->netflow)) {
968 send_netflow_active_timeouts(ofproto);
971 if (ofproto->sflow) {
972 dpif_sflow_run(ofproto->sflow);
975 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
978 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
983 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
985 /* Now revalidate if there's anything to do. */
986 if (ofproto->need_revalidate
987 || !tag_set_is_empty(&ofproto->revalidate_set)) {
988 struct tag_set revalidate_set = ofproto->revalidate_set;
989 bool revalidate_all = ofproto->need_revalidate;
992 switch (ofproto->need_revalidate) {
993 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
994 case REV_STP: COVERAGE_INC(rev_stp); break;
995 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
996 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
997 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1000 /* Clear the revalidation flags. */
1001 tag_set_init(&ofproto->revalidate_set);
1002 ofproto->need_revalidate = 0;
1004 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1006 || tag_set_intersects(&revalidate_set, facet->tags)) {
1007 facet_revalidate(facet);
1012 /* Check the consistency of a random facet, to aid debugging. */
1013 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1014 struct facet *facet;
1016 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1017 struct facet, hmap_node);
1018 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1019 if (!facet_check_consistency(facet)) {
1020 ofproto->need_revalidate = REV_INCONSISTENCY;
1025 if (ofproto->governor) {
1028 governor_run(ofproto->governor);
1030 /* If the governor has shrunk to its minimum size and the number of
1031 * subfacets has dwindled, then drop the governor entirely.
1033 * For hysteresis, the number of subfacets to drop the governor is
1034 * smaller than the number needed to trigger its creation. */
1035 n_subfacets = hmap_count(&ofproto->subfacets);
1036 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1037 && governor_is_idle(ofproto->governor)) {
1038 governor_destroy(ofproto->governor);
1039 ofproto->governor = NULL;
1047 wait(struct ofproto *ofproto_)
1049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1050 struct ofport_dpif *ofport;
1051 struct ofbundle *bundle;
1053 if (!clogged && !list_is_empty(&ofproto->completions)) {
1054 poll_immediate_wake();
1057 dpif_wait(ofproto->dpif);
1058 dpif_recv_wait(ofproto->dpif);
1059 if (ofproto->sflow) {
1060 dpif_sflow_wait(ofproto->sflow);
1062 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1063 poll_immediate_wake();
1065 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1068 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1069 bundle_wait(bundle);
1071 if (ofproto->netflow) {
1072 netflow_wait(ofproto->netflow);
1074 mac_learning_wait(ofproto->ml);
1076 if (ofproto->need_revalidate) {
1077 /* Shouldn't happen, but if it does just go around again. */
1078 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1079 poll_immediate_wake();
1081 timer_wait(&ofproto->next_expiration);
1083 if (ofproto->governor) {
1084 governor_wait(ofproto->governor);
1089 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1091 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1093 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1094 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1098 flush(struct ofproto *ofproto_)
1100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1101 struct facet *facet, *next_facet;
1103 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1104 /* Mark the facet as not installed so that facet_remove() doesn't
1105 * bother trying to uninstall it. There is no point in uninstalling it
1106 * individually since we are about to blow away all the facets with
1107 * dpif_flow_flush(). */
1108 struct subfacet *subfacet;
1110 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1111 subfacet->path = SF_NOT_INSTALLED;
1112 subfacet->dp_packet_count = 0;
1113 subfacet->dp_byte_count = 0;
1115 facet_remove(facet);
1117 dpif_flow_flush(ofproto->dpif);
1121 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1122 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1124 *arp_match_ip = true;
1125 *actions = (OFPUTIL_A_OUTPUT |
1126 OFPUTIL_A_SET_VLAN_VID |
1127 OFPUTIL_A_SET_VLAN_PCP |
1128 OFPUTIL_A_STRIP_VLAN |
1129 OFPUTIL_A_SET_DL_SRC |
1130 OFPUTIL_A_SET_DL_DST |
1131 OFPUTIL_A_SET_NW_SRC |
1132 OFPUTIL_A_SET_NW_DST |
1133 OFPUTIL_A_SET_NW_TOS |
1134 OFPUTIL_A_SET_TP_SRC |
1135 OFPUTIL_A_SET_TP_DST |
1140 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
1142 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1143 struct dpif_dp_stats s;
1145 strcpy(ots->name, "classifier");
1147 dpif_get_dp_stats(ofproto->dpif, &s);
1148 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1149 put_32aligned_be64(&ots->matched_count,
1150 htonll(s.n_hit + ofproto->n_matches));
1153 static struct ofport *
1156 struct ofport_dpif *port = xmalloc(sizeof *port);
1161 port_dealloc(struct ofport *port_)
1163 struct ofport_dpif *port = ofport_dpif_cast(port_);
1168 port_construct(struct ofport *port_)
1170 struct ofport_dpif *port = ofport_dpif_cast(port_);
1171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1173 ofproto->need_revalidate = REV_RECONFIGURE;
1174 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1175 port->bundle = NULL;
1177 port->tag = tag_create_random();
1178 port->may_enable = true;
1179 port->stp_port = NULL;
1180 port->stp_state = STP_DISABLED;
1181 hmap_init(&port->priorities);
1182 port->realdev_ofp_port = 0;
1183 port->vlandev_vid = 0;
1184 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1186 if (ofproto->sflow) {
1187 dpif_sflow_add_port(ofproto->sflow, port_);
1194 port_destruct(struct ofport *port_)
1196 struct ofport_dpif *port = ofport_dpif_cast(port_);
1197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1199 ofproto->need_revalidate = REV_RECONFIGURE;
1200 bundle_remove(port_);
1201 set_cfm(port_, NULL);
1202 if (ofproto->sflow) {
1203 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1206 ofport_clear_priorities(port);
1207 hmap_destroy(&port->priorities);
1211 port_modified(struct ofport *port_)
1213 struct ofport_dpif *port = ofport_dpif_cast(port_);
1215 if (port->bundle && port->bundle->bond) {
1216 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1221 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1223 struct ofport_dpif *port = ofport_dpif_cast(port_);
1224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1225 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1227 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1228 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1229 OFPUTIL_PC_NO_PACKET_IN)) {
1230 ofproto->need_revalidate = REV_RECONFIGURE;
1232 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1233 bundle_update(port->bundle);
1239 set_sflow(struct ofproto *ofproto_,
1240 const struct ofproto_sflow_options *sflow_options)
1242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1243 struct dpif_sflow *ds = ofproto->sflow;
1245 if (sflow_options) {
1247 struct ofport_dpif *ofport;
1249 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1250 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1251 dpif_sflow_add_port(ds, &ofport->up);
1253 ofproto->need_revalidate = REV_RECONFIGURE;
1255 dpif_sflow_set_options(ds, sflow_options);
1258 dpif_sflow_destroy(ds);
1259 ofproto->need_revalidate = REV_RECONFIGURE;
1260 ofproto->sflow = NULL;
1267 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1269 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1276 struct ofproto_dpif *ofproto;
1278 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1279 ofproto->need_revalidate = REV_RECONFIGURE;
1280 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1283 if (cfm_configure(ofport->cfm, s)) {
1289 cfm_destroy(ofport->cfm);
1295 get_cfm_fault(const struct ofport *ofport_)
1297 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1299 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1303 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1306 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1309 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1317 get_cfm_health(const struct ofport *ofport_)
1319 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1321 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1324 /* Spanning Tree. */
1327 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1329 struct ofproto_dpif *ofproto = ofproto_;
1330 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1331 struct ofport_dpif *ofport;
1333 ofport = stp_port_get_aux(sp);
1335 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1336 ofproto->up.name, port_num);
1338 struct eth_header *eth = pkt->l2;
1340 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1341 if (eth_addr_is_zero(eth->eth_src)) {
1342 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1343 "with unknown MAC", ofproto->up.name, port_num);
1345 send_packet(ofport, pkt);
1351 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1353 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1355 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1357 /* Only revalidate flows if the configuration changed. */
1358 if (!s != !ofproto->stp) {
1359 ofproto->need_revalidate = REV_RECONFIGURE;
1363 if (!ofproto->stp) {
1364 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1365 send_bpdu_cb, ofproto);
1366 ofproto->stp_last_tick = time_msec();
1369 stp_set_bridge_id(ofproto->stp, s->system_id);
1370 stp_set_bridge_priority(ofproto->stp, s->priority);
1371 stp_set_hello_time(ofproto->stp, s->hello_time);
1372 stp_set_max_age(ofproto->stp, s->max_age);
1373 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1375 struct ofport *ofport;
1377 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1378 set_stp_port(ofport, NULL);
1381 stp_destroy(ofproto->stp);
1382 ofproto->stp = NULL;
1389 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1395 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1396 s->designated_root = stp_get_designated_root(ofproto->stp);
1397 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1406 update_stp_port_state(struct ofport_dpif *ofport)
1408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1409 enum stp_state state;
1411 /* Figure out new state. */
1412 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1416 if (ofport->stp_state != state) {
1417 enum ofputil_port_state of_state;
1420 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1421 netdev_get_name(ofport->up.netdev),
1422 stp_state_name(ofport->stp_state),
1423 stp_state_name(state));
1424 if (stp_learn_in_state(ofport->stp_state)
1425 != stp_learn_in_state(state)) {
1426 /* xxx Learning action flows should also be flushed. */
1427 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1429 fwd_change = stp_forward_in_state(ofport->stp_state)
1430 != stp_forward_in_state(state);
1432 ofproto->need_revalidate = REV_STP;
1433 ofport->stp_state = state;
1434 ofport->stp_state_entered = time_msec();
1436 if (fwd_change && ofport->bundle) {
1437 bundle_update(ofport->bundle);
1440 /* Update the STP state bits in the OpenFlow port description. */
1441 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1442 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1443 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1444 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1445 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1447 ofproto_port_set_state(&ofport->up, of_state);
1451 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1452 * caller is responsible for assigning STP port numbers and ensuring
1453 * there are no duplicates. */
1455 set_stp_port(struct ofport *ofport_,
1456 const struct ofproto_port_stp_settings *s)
1458 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1460 struct stp_port *sp = ofport->stp_port;
1462 if (!s || !s->enable) {
1464 ofport->stp_port = NULL;
1465 stp_port_disable(sp);
1466 update_stp_port_state(ofport);
1469 } else if (sp && stp_port_no(sp) != s->port_num
1470 && ofport == stp_port_get_aux(sp)) {
1471 /* The port-id changed, so disable the old one if it's not
1472 * already in use by another port. */
1473 stp_port_disable(sp);
1476 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1477 stp_port_enable(sp);
1479 stp_port_set_aux(sp, ofport);
1480 stp_port_set_priority(sp, s->priority);
1481 stp_port_set_path_cost(sp, s->path_cost);
1483 update_stp_port_state(ofport);
1489 get_stp_port_status(struct ofport *ofport_,
1490 struct ofproto_port_stp_status *s)
1492 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1494 struct stp_port *sp = ofport->stp_port;
1496 if (!ofproto->stp || !sp) {
1502 s->port_id = stp_port_get_id(sp);
1503 s->state = stp_port_get_state(sp);
1504 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1505 s->role = stp_port_get_role(sp);
1506 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1512 stp_run(struct ofproto_dpif *ofproto)
1515 long long int now = time_msec();
1516 long long int elapsed = now - ofproto->stp_last_tick;
1517 struct stp_port *sp;
1520 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1521 ofproto->stp_last_tick = now;
1523 while (stp_get_changed_port(ofproto->stp, &sp)) {
1524 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1527 update_stp_port_state(ofport);
1531 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1532 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1538 stp_wait(struct ofproto_dpif *ofproto)
1541 poll_timer_wait(1000);
1545 /* Returns true if STP should process 'flow'. */
1547 stp_should_process_flow(const struct flow *flow)
1549 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1553 stp_process_packet(const struct ofport_dpif *ofport,
1554 const struct ofpbuf *packet)
1556 struct ofpbuf payload = *packet;
1557 struct eth_header *eth = payload.data;
1558 struct stp_port *sp = ofport->stp_port;
1560 /* Sink packets on ports that have STP disabled when the bridge has
1562 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1566 /* Trim off padding on payload. */
1567 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1568 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1571 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1572 stp_received_bpdu(sp, payload.data, payload.size);
1576 static struct priority_to_dscp *
1577 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1579 struct priority_to_dscp *pdscp;
1582 hash = hash_int(priority, 0);
1583 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1584 if (pdscp->priority == priority) {
1592 ofport_clear_priorities(struct ofport_dpif *ofport)
1594 struct priority_to_dscp *pdscp, *next;
1596 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1597 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1603 set_queues(struct ofport *ofport_,
1604 const struct ofproto_port_queue *qdscp_list,
1607 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1609 struct hmap new = HMAP_INITIALIZER(&new);
1612 for (i = 0; i < n_qdscp; i++) {
1613 struct priority_to_dscp *pdscp;
1617 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1618 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1623 pdscp = get_priority(ofport, priority);
1625 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1627 pdscp = xmalloc(sizeof *pdscp);
1628 pdscp->priority = priority;
1630 ofproto->need_revalidate = REV_RECONFIGURE;
1633 if (pdscp->dscp != dscp) {
1635 ofproto->need_revalidate = REV_RECONFIGURE;
1638 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1641 if (!hmap_is_empty(&ofport->priorities)) {
1642 ofport_clear_priorities(ofport);
1643 ofproto->need_revalidate = REV_RECONFIGURE;
1646 hmap_swap(&new, &ofport->priorities);
1654 /* Expires all MAC learning entries associated with 'bundle' and forces its
1655 * ofproto to revalidate every flow.
1657 * Normally MAC learning entries are removed only from the ofproto associated
1658 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1659 * are removed from every ofproto. When patch ports and SLB bonds are in use
1660 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1661 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1662 * with the host from which it migrated. */
1664 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1666 struct ofproto_dpif *ofproto = bundle->ofproto;
1667 struct mac_learning *ml = ofproto->ml;
1668 struct mac_entry *mac, *next_mac;
1670 ofproto->need_revalidate = REV_RECONFIGURE;
1671 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1672 if (mac->port.p == bundle) {
1674 struct ofproto_dpif *o;
1676 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1678 struct mac_entry *e;
1680 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1683 tag_set_add(&o->revalidate_set, e->tag);
1684 mac_learning_expire(o->ml, e);
1690 mac_learning_expire(ml, mac);
1695 static struct ofbundle *
1696 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1698 struct ofbundle *bundle;
1700 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1701 &ofproto->bundles) {
1702 if (bundle->aux == aux) {
1709 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1710 * ones that are found to 'bundles'. */
1712 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1713 void **auxes, size_t n_auxes,
1714 struct hmapx *bundles)
1718 hmapx_init(bundles);
1719 for (i = 0; i < n_auxes; i++) {
1720 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1722 hmapx_add(bundles, bundle);
1728 bundle_update(struct ofbundle *bundle)
1730 struct ofport_dpif *port;
1732 bundle->floodable = true;
1733 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1734 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1735 || !stp_forward_in_state(port->stp_state)) {
1736 bundle->floodable = false;
1743 bundle_del_port(struct ofport_dpif *port)
1745 struct ofbundle *bundle = port->bundle;
1747 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1749 list_remove(&port->bundle_node);
1750 port->bundle = NULL;
1753 lacp_slave_unregister(bundle->lacp, port);
1756 bond_slave_unregister(bundle->bond, port);
1759 bundle_update(bundle);
1763 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1764 struct lacp_slave_settings *lacp,
1765 uint32_t bond_stable_id)
1767 struct ofport_dpif *port;
1769 port = get_ofp_port(bundle->ofproto, ofp_port);
1774 if (port->bundle != bundle) {
1775 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1777 bundle_del_port(port);
1780 port->bundle = bundle;
1781 list_push_back(&bundle->ports, &port->bundle_node);
1782 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1783 || !stp_forward_in_state(port->stp_state)) {
1784 bundle->floodable = false;
1788 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1789 lacp_slave_register(bundle->lacp, port, lacp);
1792 port->bond_stable_id = bond_stable_id;
1798 bundle_destroy(struct ofbundle *bundle)
1800 struct ofproto_dpif *ofproto;
1801 struct ofport_dpif *port, *next_port;
1808 ofproto = bundle->ofproto;
1809 for (i = 0; i < MAX_MIRRORS; i++) {
1810 struct ofmirror *m = ofproto->mirrors[i];
1812 if (m->out == bundle) {
1814 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1815 || hmapx_find_and_delete(&m->dsts, bundle)) {
1816 ofproto->need_revalidate = REV_RECONFIGURE;
1821 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1822 bundle_del_port(port);
1825 bundle_flush_macs(bundle, true);
1826 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1828 free(bundle->trunks);
1829 lacp_destroy(bundle->lacp);
1830 bond_destroy(bundle->bond);
1835 bundle_set(struct ofproto *ofproto_, void *aux,
1836 const struct ofproto_bundle_settings *s)
1838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1839 bool need_flush = false;
1840 struct ofport_dpif *port;
1841 struct ofbundle *bundle;
1842 unsigned long *trunks;
1848 bundle_destroy(bundle_lookup(ofproto, aux));
1852 assert(s->n_slaves == 1 || s->bond != NULL);
1853 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1855 bundle = bundle_lookup(ofproto, aux);
1857 bundle = xmalloc(sizeof *bundle);
1859 bundle->ofproto = ofproto;
1860 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1861 hash_pointer(aux, 0));
1863 bundle->name = NULL;
1865 list_init(&bundle->ports);
1866 bundle->vlan_mode = PORT_VLAN_TRUNK;
1868 bundle->trunks = NULL;
1869 bundle->use_priority_tags = s->use_priority_tags;
1870 bundle->lacp = NULL;
1871 bundle->bond = NULL;
1873 bundle->floodable = true;
1875 bundle->src_mirrors = 0;
1876 bundle->dst_mirrors = 0;
1877 bundle->mirror_out = 0;
1880 if (!bundle->name || strcmp(s->name, bundle->name)) {
1882 bundle->name = xstrdup(s->name);
1887 if (!bundle->lacp) {
1888 ofproto->need_revalidate = REV_RECONFIGURE;
1889 bundle->lacp = lacp_create();
1891 lacp_configure(bundle->lacp, s->lacp);
1893 lacp_destroy(bundle->lacp);
1894 bundle->lacp = NULL;
1897 /* Update set of ports. */
1899 for (i = 0; i < s->n_slaves; i++) {
1900 if (!bundle_add_port(bundle, s->slaves[i],
1901 s->lacp ? &s->lacp_slaves[i] : NULL,
1902 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1906 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1907 struct ofport_dpif *next_port;
1909 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1910 for (i = 0; i < s->n_slaves; i++) {
1911 if (s->slaves[i] == port->up.ofp_port) {
1916 bundle_del_port(port);
1920 assert(list_size(&bundle->ports) <= s->n_slaves);
1922 if (list_is_empty(&bundle->ports)) {
1923 bundle_destroy(bundle);
1927 /* Set VLAN tagging mode */
1928 if (s->vlan_mode != bundle->vlan_mode
1929 || s->use_priority_tags != bundle->use_priority_tags) {
1930 bundle->vlan_mode = s->vlan_mode;
1931 bundle->use_priority_tags = s->use_priority_tags;
1936 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1937 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1939 if (vlan != bundle->vlan) {
1940 bundle->vlan = vlan;
1944 /* Get trunked VLANs. */
1945 switch (s->vlan_mode) {
1946 case PORT_VLAN_ACCESS:
1950 case PORT_VLAN_TRUNK:
1951 trunks = (unsigned long *) s->trunks;
1954 case PORT_VLAN_NATIVE_UNTAGGED:
1955 case PORT_VLAN_NATIVE_TAGGED:
1956 if (vlan != 0 && (!s->trunks
1957 || !bitmap_is_set(s->trunks, vlan)
1958 || bitmap_is_set(s->trunks, 0))) {
1959 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1961 trunks = bitmap_clone(s->trunks, 4096);
1963 trunks = bitmap_allocate1(4096);
1965 bitmap_set1(trunks, vlan);
1966 bitmap_set0(trunks, 0);
1968 trunks = (unsigned long *) s->trunks;
1975 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1976 free(bundle->trunks);
1977 if (trunks == s->trunks) {
1978 bundle->trunks = vlan_bitmap_clone(trunks);
1980 bundle->trunks = trunks;
1985 if (trunks != s->trunks) {
1990 if (!list_is_short(&bundle->ports)) {
1991 bundle->ofproto->has_bonded_bundles = true;
1993 if (bond_reconfigure(bundle->bond, s->bond)) {
1994 ofproto->need_revalidate = REV_RECONFIGURE;
1997 bundle->bond = bond_create(s->bond);
1998 ofproto->need_revalidate = REV_RECONFIGURE;
2001 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2002 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2006 bond_destroy(bundle->bond);
2007 bundle->bond = NULL;
2010 /* If we changed something that would affect MAC learning, un-learn
2011 * everything on this port and force flow revalidation. */
2013 bundle_flush_macs(bundle, false);
2020 bundle_remove(struct ofport *port_)
2022 struct ofport_dpif *port = ofport_dpif_cast(port_);
2023 struct ofbundle *bundle = port->bundle;
2026 bundle_del_port(port);
2027 if (list_is_empty(&bundle->ports)) {
2028 bundle_destroy(bundle);
2029 } else if (list_is_short(&bundle->ports)) {
2030 bond_destroy(bundle->bond);
2031 bundle->bond = NULL;
2037 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2040 struct ofport_dpif *port = port_;
2041 uint8_t ea[ETH_ADDR_LEN];
2044 error = netdev_get_etheraddr(port->up.netdev, ea);
2046 struct ofpbuf packet;
2049 ofpbuf_init(&packet, 0);
2050 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2052 memcpy(packet_pdu, pdu, pdu_size);
2054 send_packet(port, &packet);
2055 ofpbuf_uninit(&packet);
2057 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2058 "%s (%s)", port->bundle->name,
2059 netdev_get_name(port->up.netdev), strerror(error));
2064 bundle_send_learning_packets(struct ofbundle *bundle)
2066 struct ofproto_dpif *ofproto = bundle->ofproto;
2067 int error, n_packets, n_errors;
2068 struct mac_entry *e;
2070 error = n_packets = n_errors = 0;
2071 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2072 if (e->port.p != bundle) {
2073 struct ofpbuf *learning_packet;
2074 struct ofport_dpif *port;
2078 /* The assignment to "port" is unnecessary but makes "grep"ing for
2079 * struct ofport_dpif more effective. */
2080 learning_packet = bond_compose_learning_packet(bundle->bond,
2084 ret = send_packet(port, learning_packet);
2085 ofpbuf_delete(learning_packet);
2095 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2096 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2097 "packets, last error was: %s",
2098 bundle->name, n_errors, n_packets, strerror(error));
2100 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2101 bundle->name, n_packets);
2106 bundle_run(struct ofbundle *bundle)
2109 lacp_run(bundle->lacp, send_pdu_cb);
2112 struct ofport_dpif *port;
2114 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2115 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2118 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2119 lacp_status(bundle->lacp));
2120 if (bond_should_send_learning_packets(bundle->bond)) {
2121 bundle_send_learning_packets(bundle);
2127 bundle_wait(struct ofbundle *bundle)
2130 lacp_wait(bundle->lacp);
2133 bond_wait(bundle->bond);
2140 mirror_scan(struct ofproto_dpif *ofproto)
2144 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2145 if (!ofproto->mirrors[idx]) {
2152 static struct ofmirror *
2153 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2157 for (i = 0; i < MAX_MIRRORS; i++) {
2158 struct ofmirror *mirror = ofproto->mirrors[i];
2159 if (mirror && mirror->aux == aux) {
2167 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2169 mirror_update_dups(struct ofproto_dpif *ofproto)
2173 for (i = 0; i < MAX_MIRRORS; i++) {
2174 struct ofmirror *m = ofproto->mirrors[i];
2177 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2181 for (i = 0; i < MAX_MIRRORS; i++) {
2182 struct ofmirror *m1 = ofproto->mirrors[i];
2189 for (j = i + 1; j < MAX_MIRRORS; j++) {
2190 struct ofmirror *m2 = ofproto->mirrors[j];
2192 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2193 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2194 m2->dup_mirrors |= m1->dup_mirrors;
2201 mirror_set(struct ofproto *ofproto_, void *aux,
2202 const struct ofproto_mirror_settings *s)
2204 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2205 mirror_mask_t mirror_bit;
2206 struct ofbundle *bundle;
2207 struct ofmirror *mirror;
2208 struct ofbundle *out;
2209 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2210 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2213 mirror = mirror_lookup(ofproto, aux);
2215 mirror_destroy(mirror);
2221 idx = mirror_scan(ofproto);
2223 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2225 ofproto->up.name, MAX_MIRRORS, s->name);
2229 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2230 mirror->ofproto = ofproto;
2233 mirror->out_vlan = -1;
2234 mirror->name = NULL;
2237 if (!mirror->name || strcmp(s->name, mirror->name)) {
2239 mirror->name = xstrdup(s->name);
2242 /* Get the new configuration. */
2243 if (s->out_bundle) {
2244 out = bundle_lookup(ofproto, s->out_bundle);
2246 mirror_destroy(mirror);
2252 out_vlan = s->out_vlan;
2254 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2255 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2257 /* If the configuration has not changed, do nothing. */
2258 if (hmapx_equals(&srcs, &mirror->srcs)
2259 && hmapx_equals(&dsts, &mirror->dsts)
2260 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2261 && mirror->out == out
2262 && mirror->out_vlan == out_vlan)
2264 hmapx_destroy(&srcs);
2265 hmapx_destroy(&dsts);
2269 hmapx_swap(&srcs, &mirror->srcs);
2270 hmapx_destroy(&srcs);
2272 hmapx_swap(&dsts, &mirror->dsts);
2273 hmapx_destroy(&dsts);
2275 free(mirror->vlans);
2276 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2279 mirror->out_vlan = out_vlan;
2281 /* Update bundles. */
2282 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2283 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2284 if (hmapx_contains(&mirror->srcs, bundle)) {
2285 bundle->src_mirrors |= mirror_bit;
2287 bundle->src_mirrors &= ~mirror_bit;
2290 if (hmapx_contains(&mirror->dsts, bundle)) {
2291 bundle->dst_mirrors |= mirror_bit;
2293 bundle->dst_mirrors &= ~mirror_bit;
2296 if (mirror->out == bundle) {
2297 bundle->mirror_out |= mirror_bit;
2299 bundle->mirror_out &= ~mirror_bit;
2303 ofproto->need_revalidate = REV_RECONFIGURE;
2304 ofproto->has_mirrors = true;
2305 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2306 mirror_update_dups(ofproto);
2312 mirror_destroy(struct ofmirror *mirror)
2314 struct ofproto_dpif *ofproto;
2315 mirror_mask_t mirror_bit;
2316 struct ofbundle *bundle;
2323 ofproto = mirror->ofproto;
2324 ofproto->need_revalidate = REV_RECONFIGURE;
2325 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2327 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2328 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2329 bundle->src_mirrors &= ~mirror_bit;
2330 bundle->dst_mirrors &= ~mirror_bit;
2331 bundle->mirror_out &= ~mirror_bit;
2334 hmapx_destroy(&mirror->srcs);
2335 hmapx_destroy(&mirror->dsts);
2336 free(mirror->vlans);
2338 ofproto->mirrors[mirror->idx] = NULL;
2342 mirror_update_dups(ofproto);
2344 ofproto->has_mirrors = false;
2345 for (i = 0; i < MAX_MIRRORS; i++) {
2346 if (ofproto->mirrors[i]) {
2347 ofproto->has_mirrors = true;
2354 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2355 uint64_t *packets, uint64_t *bytes)
2357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2358 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2361 *packets = *bytes = UINT64_MAX;
2365 *packets = mirror->packet_count;
2366 *bytes = mirror->byte_count;
2372 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2374 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2375 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2376 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2382 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2385 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2386 return bundle && bundle->mirror_out != 0;
2390 forward_bpdu_changed(struct ofproto *ofproto_)
2392 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2393 ofproto->need_revalidate = REV_RECONFIGURE;
2397 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2400 mac_learning_set_idle_time(ofproto->ml, idle_time);
2405 static struct ofport_dpif *
2406 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2408 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2409 return ofport ? ofport_dpif_cast(ofport) : NULL;
2412 static struct ofport_dpif *
2413 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2415 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2419 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2420 struct dpif_port *dpif_port)
2422 ofproto_port->name = dpif_port->name;
2423 ofproto_port->type = dpif_port->type;
2424 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2428 port_run(struct ofport_dpif *ofport)
2430 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2431 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2432 bool enable = netdev_get_carrier(ofport->up.netdev);
2434 ofport->carrier_seq = carrier_seq;
2437 cfm_run(ofport->cfm);
2439 if (cfm_should_send_ccm(ofport->cfm)) {
2440 struct ofpbuf packet;
2442 ofpbuf_init(&packet, 0);
2443 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2444 send_packet(ofport, &packet);
2445 ofpbuf_uninit(&packet);
2448 enable = enable && !cfm_get_fault(ofport->cfm)
2449 && cfm_get_opup(ofport->cfm);
2452 if (ofport->bundle) {
2453 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2454 if (carrier_changed) {
2455 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2459 if (ofport->may_enable != enable) {
2460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2462 if (ofproto->has_bundle_action) {
2463 ofproto->need_revalidate = REV_PORT_TOGGLED;
2467 ofport->may_enable = enable;
2471 port_wait(struct ofport_dpif *ofport)
2474 cfm_wait(ofport->cfm);
2479 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2480 struct ofproto_port *ofproto_port)
2482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2483 struct dpif_port dpif_port;
2486 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2488 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2494 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2496 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2500 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2502 *ofp_portp = odp_port_to_ofp_port(odp_port);
2508 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2513 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2515 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2517 /* The caller is going to close ofport->up.netdev. If this is a
2518 * bonded port, then the bond is using that netdev, so remove it
2519 * from the bond. The client will need to reconfigure everything
2520 * after deleting ports, so then the slave will get re-added. */
2521 bundle_remove(&ofport->up);
2528 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2530 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2533 error = netdev_get_stats(ofport->up.netdev, stats);
2535 if (!error && ofport->odp_port == OVSP_LOCAL) {
2536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2538 /* ofproto->stats.tx_packets represents packets that we created
2539 * internally and sent to some port (e.g. packets sent with
2540 * send_packet()). Account for them as if they had come from
2541 * OFPP_LOCAL and got forwarded. */
2543 if (stats->rx_packets != UINT64_MAX) {
2544 stats->rx_packets += ofproto->stats.tx_packets;
2547 if (stats->rx_bytes != UINT64_MAX) {
2548 stats->rx_bytes += ofproto->stats.tx_bytes;
2551 /* ofproto->stats.rx_packets represents packets that were received on
2552 * some port and we processed internally and dropped (e.g. STP).
2553 * Account fro them as if they had been forwarded to OFPP_LOCAL. */
2555 if (stats->tx_packets != UINT64_MAX) {
2556 stats->tx_packets += ofproto->stats.rx_packets;
2559 if (stats->tx_bytes != UINT64_MAX) {
2560 stats->tx_bytes += ofproto->stats.rx_bytes;
2567 /* Account packets for LOCAL port. */
2569 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2570 size_t tx_size, size_t rx_size)
2572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2575 ofproto->stats.rx_packets++;
2576 ofproto->stats.rx_bytes += rx_size;
2579 ofproto->stats.tx_packets++;
2580 ofproto->stats.tx_bytes += tx_size;
2584 struct port_dump_state {
2585 struct dpif_port_dump dump;
2590 port_dump_start(const struct ofproto *ofproto_, void **statep)
2592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2593 struct port_dump_state *state;
2595 *statep = state = xmalloc(sizeof *state);
2596 dpif_port_dump_start(&state->dump, ofproto->dpif);
2597 state->done = false;
2602 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2603 struct ofproto_port *port)
2605 struct port_dump_state *state = state_;
2606 struct dpif_port dpif_port;
2608 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2609 ofproto_port_from_dpif_port(port, &dpif_port);
2612 int error = dpif_port_dump_done(&state->dump);
2614 return error ? error : EOF;
2619 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2621 struct port_dump_state *state = state_;
2624 dpif_port_dump_done(&state->dump);
2631 port_poll(const struct ofproto *ofproto_, char **devnamep)
2633 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2634 return dpif_port_poll(ofproto->dpif, devnamep);
2638 port_poll_wait(const struct ofproto *ofproto_)
2640 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2641 dpif_port_poll_wait(ofproto->dpif);
2645 port_is_lacp_current(const struct ofport *ofport_)
2647 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2648 return (ofport->bundle && ofport->bundle->lacp
2649 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2653 /* Upcall handling. */
2655 /* Flow miss batching.
2657 * Some dpifs implement operations faster when you hand them off in a batch.
2658 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2659 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2660 * more packets, plus possibly installing the flow in the dpif.
2662 * So far we only batch the operations that affect flow setup time the most.
2663 * It's possible to batch more than that, but the benefit might be minimal. */
2665 struct hmap_node hmap_node;
2667 enum odp_key_fitness key_fitness;
2668 const struct nlattr *key;
2670 ovs_be16 initial_tci;
2671 struct list packets;
2672 enum dpif_upcall_type upcall_type;
2675 struct flow_miss_op {
2676 struct dpif_op dpif_op;
2677 struct subfacet *subfacet; /* Subfacet */
2678 void *garbage; /* Pointer to pass to free(), NULL if none. */
2679 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2682 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2683 * OpenFlow controller as necessary according to their individual
2684 * configurations. */
2686 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2687 const struct flow *flow)
2689 struct ofputil_packet_in pin;
2691 pin.packet = packet->data;
2692 pin.packet_len = packet->size;
2693 pin.reason = OFPR_NO_MATCH;
2694 pin.controller_id = 0;
2699 pin.send_len = 0; /* not used for flow table misses */
2701 flow_get_metadata(flow, &pin.fmd);
2703 /* Registers aren't meaningful on a miss. */
2704 memset(pin.fmd.reg_masks, 0, sizeof pin.fmd.reg_masks);
2706 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2709 static enum slow_path_reason
2710 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2711 const struct ofpbuf *packet)
2713 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2719 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2721 cfm_process_heartbeat(ofport->cfm, packet);
2724 } else if (ofport->bundle && ofport->bundle->lacp
2725 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2727 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2730 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2732 stp_process_packet(ofport, packet);
2739 static struct flow_miss *
2740 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2742 struct flow_miss *miss;
2744 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2745 if (flow_equal(&miss->flow, flow)) {
2753 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2754 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2755 * 'miss' is associated with a subfacet the caller must also initialize the
2756 * returned op->subfacet, and if anything needs to be freed after processing
2757 * the op, the caller must initialize op->garbage also. */
2759 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2760 struct flow_miss_op *op)
2762 if (miss->flow.vlan_tci != miss->initial_tci) {
2763 /* This packet was received on a VLAN splinter port. We
2764 * added a VLAN to the packet to make the packet resemble
2765 * the flow, but the actions were composed assuming that
2766 * the packet contained no VLAN. So, we must remove the
2767 * VLAN header from the packet before trying to execute the
2769 eth_pop_vlan(packet);
2772 op->subfacet = NULL;
2774 op->dpif_op.type = DPIF_OP_EXECUTE;
2775 op->dpif_op.u.execute.key = miss->key;
2776 op->dpif_op.u.execute.key_len = miss->key_len;
2777 op->dpif_op.u.execute.packet = packet;
2780 /* Helper for handle_flow_miss_without_facet() and
2781 * handle_flow_miss_with_facet(). */
2783 handle_flow_miss_common(struct rule_dpif *rule,
2784 struct ofpbuf *packet, const struct flow *flow)
2786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2788 ofproto->n_matches++;
2790 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2792 * Extra-special case for fail-open mode.
2794 * We are in fail-open mode and the packet matched the fail-open
2795 * rule, but we are connected to a controller too. We should send
2796 * the packet up to the controller in the hope that it will try to
2797 * set up a flow and thereby allow us to exit fail-open.
2799 * See the top-level comment in fail-open.c for more information.
2801 send_packet_in_miss(ofproto, packet, flow);
2805 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2806 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2807 * installing a datapath flow. The answer is usually "yes" (a return value of
2808 * true). However, for short flows the cost of bookkeeping is much higher than
2809 * the benefits, so when the datapath holds a large number of flows we impose
2810 * some heuristics to decide which flows are likely to be worth tracking. */
2812 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2813 struct flow_miss *miss, uint32_t hash)
2815 if (!ofproto->governor) {
2818 n_subfacets = hmap_count(&ofproto->subfacets);
2819 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2823 ofproto->governor = governor_create(ofproto->up.name);
2826 return governor_should_install_flow(ofproto->governor, hash,
2827 list_size(&miss->packets));
2830 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2831 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2832 * increment '*n_ops'. */
2834 handle_flow_miss_without_facet(struct flow_miss *miss,
2835 struct rule_dpif *rule,
2836 struct flow_miss_op *ops, size_t *n_ops)
2838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2839 struct action_xlate_ctx ctx;
2840 struct ofpbuf *packet;
2842 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2843 struct flow_miss_op *op = &ops[*n_ops];
2844 struct dpif_flow_stats stats;
2845 struct ofpbuf odp_actions;
2847 COVERAGE_INC(facet_suppress);
2849 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2851 dpif_flow_stats_extract(&miss->flow, packet, &stats);
2852 rule_credit_stats(rule, &stats);
2854 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2856 ctx.resubmit_stats = &stats;
2857 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
2860 if (odp_actions.size) {
2861 struct dpif_execute *execute = &op->dpif_op.u.execute;
2863 init_flow_miss_execute_op(miss, packet, op);
2864 execute->actions = odp_actions.data;
2865 execute->actions_len = odp_actions.size;
2866 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2870 ofpbuf_uninit(&odp_actions);
2875 /* Handles 'miss', which matches 'facet'. May add any required datapath
2876 * operations to 'ops', incrementing '*n_ops' for each new op. */
2878 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2879 struct flow_miss_op *ops, size_t *n_ops)
2881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2882 enum subfacet_path want_path;
2883 struct subfacet *subfacet;
2884 struct ofpbuf *packet;
2886 subfacet = subfacet_create(facet,
2887 miss->key_fitness, miss->key, miss->key_len,
2890 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2891 struct flow_miss_op *op = &ops[*n_ops];
2892 struct dpif_flow_stats stats;
2893 struct ofpbuf odp_actions;
2895 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2897 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2898 if (!subfacet->actions || subfacet->slow) {
2899 subfacet_make_actions(subfacet, packet, &odp_actions);
2902 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2903 subfacet_update_stats(subfacet, &stats);
2905 if (subfacet->actions_len) {
2906 struct dpif_execute *execute = &op->dpif_op.u.execute;
2908 init_flow_miss_execute_op(miss, packet, op);
2909 op->subfacet = subfacet;
2910 if (!subfacet->slow) {
2911 execute->actions = subfacet->actions;
2912 execute->actions_len = subfacet->actions_len;
2913 ofpbuf_uninit(&odp_actions);
2915 execute->actions = odp_actions.data;
2916 execute->actions_len = odp_actions.size;
2917 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2922 ofpbuf_uninit(&odp_actions);
2926 want_path = subfacet_want_path(subfacet->slow);
2927 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2928 struct flow_miss_op *op = &ops[(*n_ops)++];
2929 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2931 op->subfacet = subfacet;
2933 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2934 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2935 put->key = miss->key;
2936 put->key_len = miss->key_len;
2937 if (want_path == SF_FAST_PATH) {
2938 put->actions = subfacet->actions;
2939 put->actions_len = subfacet->actions_len;
2941 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2942 op->stub, sizeof op->stub,
2943 &put->actions, &put->actions_len);
2949 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2950 * operations to 'ops', incrementing '*n_ops' for each new op. */
2952 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2953 struct flow_miss_op *ops, size_t *n_ops)
2955 struct facet *facet;
2958 /* The caller must ensure that miss->hmap_node.hash contains
2959 * flow_hash(miss->flow, 0). */
2960 hash = miss->hmap_node.hash;
2962 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
2964 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
2966 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
2967 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
2971 facet = facet_create(rule, &miss->flow, hash);
2973 handle_flow_miss_with_facet(miss, facet, ops, n_ops);
2976 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
2977 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
2978 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
2979 * what a flow key should contain.
2981 * This function also includes some logic to help make VLAN splinters
2982 * transparent to the rest of the upcall processing logic. In particular, if
2983 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
2984 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
2985 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
2987 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
2988 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
2989 * (This differs from the value returned in flow->vlan_tci only for packets
2990 * received on VLAN splinters.)
2992 static enum odp_key_fitness
2993 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
2994 const struct nlattr *key, size_t key_len,
2995 struct flow *flow, ovs_be16 *initial_tci,
2996 struct ofpbuf *packet)
2998 enum odp_key_fitness fitness;
3000 fitness = odp_flow_key_to_flow(key, key_len, flow);
3001 if (fitness == ODP_FIT_ERROR) {
3004 *initial_tci = flow->vlan_tci;
3006 if (vsp_adjust_flow(ofproto, flow)) {
3008 /* Make the packet resemble the flow, so that it gets sent to an
3009 * OpenFlow controller properly, so that it looks correct for
3010 * sFlow, and so that flow_extract() will get the correct vlan_tci
3011 * if it is called on 'packet'.
3013 * The allocated space inside 'packet' probably also contains
3014 * 'key', that is, both 'packet' and 'key' are probably part of a
3015 * struct dpif_upcall (see the large comment on that structure
3016 * definition), so pushing data on 'packet' is in general not a
3017 * good idea since it could overwrite 'key' or free it as a side
3018 * effect. However, it's OK in this special case because we know
3019 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3020 * will just overwrite the 4-byte "struct nlattr", which is fine
3021 * since we don't need that header anymore. */
3022 eth_push_vlan(packet, flow->vlan_tci);
3025 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3026 if (fitness == ODP_FIT_PERFECT) {
3027 fitness = ODP_FIT_TOO_MUCH;
3035 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3038 struct dpif_upcall *upcall;
3039 struct flow_miss *miss;
3040 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3041 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3042 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3052 /* Construct the to-do list.
3054 * This just amounts to extracting the flow from each packet and sticking
3055 * the packets that have the same flow in the same "flow_miss" structure so
3056 * that we can process them together. */
3059 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3060 struct flow_miss *miss = &misses[n_misses];
3061 struct flow_miss *existing_miss;
3064 /* Obtain metadata and check userspace/kernel agreement on flow match,
3065 * then set 'flow''s header pointers. */
3066 miss->key_fitness = ofproto_dpif_extract_flow_key(
3067 ofproto, upcall->key, upcall->key_len,
3068 &miss->flow, &miss->initial_tci, upcall->packet);
3069 if (miss->key_fitness == ODP_FIT_ERROR) {
3072 flow_extract(upcall->packet, miss->flow.skb_priority,
3073 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3075 /* Add other packets to a to-do list. */
3076 hash = flow_hash(&miss->flow, 0);
3077 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3078 if (!existing_miss) {
3079 hmap_insert(&todo, &miss->hmap_node, hash);
3080 miss->key = upcall->key;
3081 miss->key_len = upcall->key_len;
3082 miss->upcall_type = upcall->type;
3083 list_init(&miss->packets);
3087 miss = existing_miss;
3089 list_push_back(&miss->packets, &upcall->packet->list_node);
3092 /* Process each element in the to-do list, constructing the set of
3093 * operations to batch. */
3095 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3096 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3098 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3100 /* Execute batch. */
3101 for (i = 0; i < n_ops; i++) {
3102 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3104 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3106 /* Free memory and update facets. */
3107 for (i = 0; i < n_ops; i++) {
3108 struct flow_miss_op *op = &flow_miss_ops[i];
3110 switch (op->dpif_op.type) {
3111 case DPIF_OP_EXECUTE:
3114 case DPIF_OP_FLOW_PUT:
3115 if (!op->dpif_op.error) {
3116 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3120 case DPIF_OP_FLOW_DEL:
3126 hmap_destroy(&todo);
3129 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3130 classify_upcall(const struct dpif_upcall *upcall)
3132 union user_action_cookie cookie;
3134 /* First look at the upcall type. */
3135 switch (upcall->type) {
3136 case DPIF_UC_ACTION:
3142 case DPIF_N_UC_TYPES:
3144 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3148 /* "action" upcalls need a closer look. */
3149 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3150 switch (cookie.type) {
3151 case USER_ACTION_COOKIE_SFLOW:
3152 return SFLOW_UPCALL;
3154 case USER_ACTION_COOKIE_SLOW_PATH:
3157 case USER_ACTION_COOKIE_UNSPEC:
3159 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3165 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3166 const struct dpif_upcall *upcall)
3168 union user_action_cookie cookie;
3169 enum odp_key_fitness fitness;
3170 ovs_be16 initial_tci;
3173 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3174 upcall->key_len, &flow,
3175 &initial_tci, upcall->packet);
3176 if (fitness == ODP_FIT_ERROR) {
3180 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3181 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3185 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3187 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3188 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3189 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3194 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3197 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3198 struct dpif_upcall *upcall = &misses[n_misses];
3199 struct ofpbuf *buf = &miss_bufs[n_misses];
3202 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3203 sizeof miss_buf_stubs[n_misses]);
3204 error = dpif_recv(ofproto->dpif, upcall, buf);
3210 switch (classify_upcall(upcall)) {
3212 /* Handle it later. */
3217 if (ofproto->sflow) {
3218 handle_sflow_upcall(ofproto, upcall);
3229 /* Handle deferred MISS_UPCALL processing. */
3230 handle_miss_upcalls(ofproto, misses, n_misses);
3231 for (i = 0; i < n_misses; i++) {
3232 ofpbuf_uninit(&miss_bufs[i]);
3238 /* Flow expiration. */
3240 static int subfacet_max_idle(const struct ofproto_dpif *);
3241 static void update_stats(struct ofproto_dpif *);
3242 static void rule_expire(struct rule_dpif *);
3243 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3245 /* This function is called periodically by run(). Its job is to collect
3246 * updates for the flows that have been installed into the datapath, most
3247 * importantly when they last were used, and then use that information to
3248 * expire flows that have not been used recently.
3250 * Returns the number of milliseconds after which it should be called again. */
3252 expire(struct ofproto_dpif *ofproto)
3254 struct rule_dpif *rule, *next_rule;
3255 struct oftable *table;
3258 /* Update stats for each flow in the datapath. */
3259 update_stats(ofproto);
3261 /* Expire subfacets that have been idle too long. */
3262 dp_max_idle = subfacet_max_idle(ofproto);
3263 expire_subfacets(ofproto, dp_max_idle);
3265 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3266 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3267 struct cls_cursor cursor;
3269 cls_cursor_init(&cursor, &table->cls, NULL);
3270 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3275 /* All outstanding data in existing flows has been accounted, so it's a
3276 * good time to do bond rebalancing. */
3277 if (ofproto->has_bonded_bundles) {
3278 struct ofbundle *bundle;
3280 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3282 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3287 return MIN(dp_max_idle, 1000);
3290 /* Updates flow table statistics given that the datapath just reported 'stats'
3291 * as 'subfacet''s statistics. */
3293 update_subfacet_stats(struct subfacet *subfacet,
3294 const struct dpif_flow_stats *stats)
3296 struct facet *facet = subfacet->facet;
3298 if (stats->n_packets >= subfacet->dp_packet_count) {
3299 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3300 facet->packet_count += extra;
3302 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3305 if (stats->n_bytes >= subfacet->dp_byte_count) {
3306 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3308 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3311 subfacet->dp_packet_count = stats->n_packets;
3312 subfacet->dp_byte_count = stats->n_bytes;
3314 facet->tcp_flags |= stats->tcp_flags;
3316 subfacet_update_time(subfacet, stats->used);
3317 if (facet->accounted_bytes < facet->byte_count) {
3319 facet_account(facet);
3320 facet->accounted_bytes = facet->byte_count;
3322 facet_push_stats(facet);
3325 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3326 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3328 delete_unexpected_flow(struct dpif *dpif,
3329 const struct nlattr *key, size_t key_len)
3331 if (!VLOG_DROP_WARN(&rl)) {
3335 odp_flow_key_format(key, key_len, &s);
3336 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3340 COVERAGE_INC(facet_unexpected);
3341 dpif_flow_del(dpif, key, key_len, NULL);
3344 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3346 * This function also pushes statistics updates to rules which each facet
3347 * resubmits into. Generally these statistics will be accurate. However, if a
3348 * facet changes the rule it resubmits into at some time in between
3349 * update_stats() runs, it is possible that statistics accrued to the
3350 * old rule will be incorrectly attributed to the new rule. This could be
3351 * avoided by calling update_stats() whenever rules are created or
3352 * deleted. However, the performance impact of making so many calls to the
3353 * datapath do not justify the benefit of having perfectly accurate statistics.
3356 update_stats(struct ofproto_dpif *p)
3358 const struct dpif_flow_stats *stats;
3359 struct dpif_flow_dump dump;
3360 const struct nlattr *key;
3363 dpif_flow_dump_start(&dump, p->dpif);
3364 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3365 struct subfacet *subfacet;
3367 subfacet = subfacet_find(p, key, key_len);
3368 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3370 update_subfacet_stats(subfacet, stats);
3374 /* Stats are updated per-packet. */
3377 case SF_NOT_INSTALLED:
3379 delete_unexpected_flow(p->dpif, key, key_len);
3383 dpif_flow_dump_done(&dump);
3386 /* Calculates and returns the number of milliseconds of idle time after which
3387 * subfacets should expire from the datapath. When a subfacet expires, we fold
3388 * its statistics into its facet, and when a facet's last subfacet expires, we
3389 * fold its statistic into its rule. */
3391 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3394 * Idle time histogram.
3396 * Most of the time a switch has a relatively small number of subfacets.
3397 * When this is the case we might as well keep statistics for all of them
3398 * in userspace and to cache them in the kernel datapath for performance as
3401 * As the number of subfacets increases, the memory required to maintain
3402 * statistics about them in userspace and in the kernel becomes
3403 * significant. However, with a large number of subfacets it is likely
3404 * that only a few of them are "heavy hitters" that consume a large amount
3405 * of bandwidth. At this point, only heavy hitters are worth caching in
3406 * the kernel and maintaining in userspaces; other subfacets we can
3409 * The technique used to compute the idle time is to build a histogram with
3410 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3411 * that is installed in the kernel gets dropped in the appropriate bucket.
3412 * After the histogram has been built, we compute the cutoff so that only
3413 * the most-recently-used 1% of subfacets (but at least
3414 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3415 * the most-recently-used bucket of subfacets is kept, so actually an
3416 * arbitrary number of subfacets can be kept in any given expiration run
3417 * (though the next run will delete most of those unless they receive
3420 * This requires a second pass through the subfacets, in addition to the
3421 * pass made by update_stats(), because the former function never looks at
3422 * uninstallable subfacets.
3424 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3425 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3426 int buckets[N_BUCKETS] = { 0 };
3427 int total, subtotal, bucket;
3428 struct subfacet *subfacet;
3432 total = hmap_count(&ofproto->subfacets);
3433 if (total <= ofproto->up.flow_eviction_threshold) {
3434 return N_BUCKETS * BUCKET_WIDTH;
3437 /* Build histogram. */
3439 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3440 long long int idle = now - subfacet->used;
3441 int bucket = (idle <= 0 ? 0
3442 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3443 : (unsigned int) idle / BUCKET_WIDTH);
3447 /* Find the first bucket whose flows should be expired. */
3448 subtotal = bucket = 0;
3450 subtotal += buckets[bucket++];
3451 } while (bucket < N_BUCKETS &&
3452 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3454 if (VLOG_IS_DBG_ENABLED()) {
3458 ds_put_cstr(&s, "keep");
3459 for (i = 0; i < N_BUCKETS; i++) {
3461 ds_put_cstr(&s, ", drop");
3464 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3467 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3471 return bucket * BUCKET_WIDTH;
3474 enum { EXPIRE_MAX_BATCH = 50 };
3477 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3479 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3480 struct dpif_op ops[EXPIRE_MAX_BATCH];
3481 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3482 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3483 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3486 for (i = 0; i < n; i++) {
3487 ops[i].type = DPIF_OP_FLOW_DEL;
3488 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3489 ops[i].u.flow_del.key = keys[i].data;
3490 ops[i].u.flow_del.key_len = keys[i].size;
3491 ops[i].u.flow_del.stats = &stats[i];
3495 dpif_operate(ofproto->dpif, opsp, n);
3496 for (i = 0; i < n; i++) {
3497 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3498 subfacets[i]->path = SF_NOT_INSTALLED;
3499 subfacet_destroy(subfacets[i]);
3504 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3506 /* Cutoff time for most flows. */
3507 long long int normal_cutoff = time_msec() - dp_max_idle;
3509 /* We really want to keep flows for special protocols around, so use a more
3510 * conservative cutoff. */
3511 long long int special_cutoff = time_msec() - 10000;
3513 struct subfacet *subfacet, *next_subfacet;
3514 struct subfacet *batch[EXPIRE_MAX_BATCH];
3518 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3519 &ofproto->subfacets) {
3520 long long int cutoff;
3522 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3525 if (subfacet->used < cutoff) {
3526 if (subfacet->path != SF_NOT_INSTALLED) {
3527 batch[n_batch++] = subfacet;
3528 if (n_batch >= EXPIRE_MAX_BATCH) {
3529 expire_batch(ofproto, batch, n_batch);
3533 subfacet_destroy(subfacet);
3539 expire_batch(ofproto, batch, n_batch);
3543 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3544 * then delete it entirely. */
3546 rule_expire(struct rule_dpif *rule)
3548 struct facet *facet, *next_facet;
3552 /* Has 'rule' expired? */
3554 if (rule->up.hard_timeout
3555 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3556 reason = OFPRR_HARD_TIMEOUT;
3557 } else if (rule->up.idle_timeout
3558 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3559 reason = OFPRR_IDLE_TIMEOUT;
3564 COVERAGE_INC(ofproto_dpif_expired);
3566 /* Update stats. (This is a no-op if the rule expired due to an idle
3567 * timeout, because that only happens when the rule has no facets left.) */
3568 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3569 facet_remove(facet);
3572 /* Get rid of the rule. */
3573 ofproto_rule_expire(&rule->up, reason);
3578 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3580 * The caller must already have determined that no facet with an identical
3581 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3582 * the ofproto's classifier table.
3584 * 'hash' must be the return value of flow_hash(flow, 0).
3586 * The facet will initially have no subfacets. The caller should create (at
3587 * least) one subfacet with subfacet_create(). */
3588 static struct facet *
3589 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3592 struct facet *facet;
3594 facet = xzalloc(sizeof *facet);
3595 facet->used = time_msec();
3596 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3597 list_push_back(&rule->facets, &facet->list_node);
3599 facet->flow = *flow;
3600 list_init(&facet->subfacets);
3601 netflow_flow_init(&facet->nf_flow);
3602 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3608 facet_free(struct facet *facet)
3613 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3614 * 'packet', which arrived on 'in_port'.
3616 * Takes ownership of 'packet'. */
3618 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3619 const struct nlattr *odp_actions, size_t actions_len,
3620 struct ofpbuf *packet)
3622 struct odputil_keybuf keybuf;
3626 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3627 odp_flow_key_from_flow(&key, flow);
3629 error = dpif_execute(ofproto->dpif, key.data, key.size,
3630 odp_actions, actions_len, packet);
3632 ofpbuf_delete(packet);
3636 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3638 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3639 * rule's statistics, via subfacet_uninstall().
3641 * - Removes 'facet' from its rule and from ofproto->facets.
3644 facet_remove(struct facet *facet)
3646 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3647 struct subfacet *subfacet, *next_subfacet;
3649 assert(!list_is_empty(&facet->subfacets));
3651 /* First uninstall all of the subfacets to get final statistics. */
3652 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3653 subfacet_uninstall(subfacet);
3656 /* Flush the final stats to the rule.
3658 * This might require us to have at least one subfacet around so that we
3659 * can use its actions for accounting in facet_account(), which is why we
3660 * have uninstalled but not yet destroyed the subfacets. */
3661 facet_flush_stats(facet);
3663 /* Now we're really all done so destroy everything. */
3664 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3665 &facet->subfacets) {
3666 subfacet_destroy__(subfacet);
3668 hmap_remove(&ofproto->facets, &facet->hmap_node);
3669 list_remove(&facet->list_node);
3673 /* Feed information from 'facet' back into the learning table to keep it in
3674 * sync with what is actually flowing through the datapath. */
3676 facet_learn(struct facet *facet)
3678 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3679 struct action_xlate_ctx ctx;
3681 if (!facet->has_learn
3682 && !facet->has_normal
3683 && (!facet->has_fin_timeout
3684 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3688 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3689 facet->flow.vlan_tci,
3690 facet->rule, facet->tcp_flags, NULL);
3691 ctx.may_learn = true;
3692 xlate_actions_for_side_effects(&ctx, facet->rule->up.actions,
3693 facet->rule->up.n_actions);
3697 facet_account(struct facet *facet)
3699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3700 struct subfacet *subfacet;
3701 const struct nlattr *a;
3706 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3709 n_bytes = facet->byte_count - facet->accounted_bytes;
3711 /* This loop feeds byte counters to bond_account() for rebalancing to use
3712 * as a basis. We also need to track the actual VLAN on which the packet
3713 * is going to be sent to ensure that it matches the one passed to
3714 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3717 * We use the actions from an arbitrary subfacet because they should all
3718 * be equally valid for our purpose. */
3719 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3720 struct subfacet, list_node);
3721 vlan_tci = facet->flow.vlan_tci;
3722 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3723 subfacet->actions, subfacet->actions_len) {
3724 const struct ovs_action_push_vlan *vlan;
3725 struct ofport_dpif *port;
3727 switch (nl_attr_type(a)) {
3728 case OVS_ACTION_ATTR_OUTPUT:
3729 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3730 if (port && port->bundle && port->bundle->bond) {
3731 bond_account(port->bundle->bond, &facet->flow,
3732 vlan_tci_to_vid(vlan_tci), n_bytes);
3736 case OVS_ACTION_ATTR_POP_VLAN:
3737 vlan_tci = htons(0);
3740 case OVS_ACTION_ATTR_PUSH_VLAN:
3741 vlan = nl_attr_get(a);
3742 vlan_tci = vlan->vlan_tci;
3748 /* Returns true if the only action for 'facet' is to send to the controller.
3749 * (We don't report NetFlow expiration messages for such facets because they
3750 * are just part of the control logic for the network, not real traffic). */
3752 facet_is_controller_flow(struct facet *facet)
3755 && facet->rule->up.n_actions == 1
3756 && action_outputs_to_port(&facet->rule->up.actions[0],
3757 htons(OFPP_CONTROLLER)));
3760 /* Folds all of 'facet''s statistics into its rule. Also updates the
3761 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3762 * 'facet''s statistics in the datapath should have been zeroed and folded into
3763 * its packet and byte counts before this function is called. */
3765 facet_flush_stats(struct facet *facet)
3767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3768 struct subfacet *subfacet;
3770 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3771 assert(!subfacet->dp_byte_count);
3772 assert(!subfacet->dp_packet_count);
3775 facet_push_stats(facet);
3776 if (facet->accounted_bytes < facet->byte_count) {
3777 facet_account(facet);
3778 facet->accounted_bytes = facet->byte_count;
3781 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3782 struct ofexpired expired;
3783 expired.flow = facet->flow;
3784 expired.packet_count = facet->packet_count;
3785 expired.byte_count = facet->byte_count;
3786 expired.used = facet->used;
3787 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3790 facet->rule->packet_count += facet->packet_count;
3791 facet->rule->byte_count += facet->byte_count;
3793 /* Reset counters to prevent double counting if 'facet' ever gets
3795 facet_reset_counters(facet);
3797 netflow_flow_clear(&facet->nf_flow);
3798 facet->tcp_flags = 0;
3801 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3802 * Returns it if found, otherwise a null pointer.
3804 * 'hash' must be the return value of flow_hash(flow, 0).
3806 * The returned facet might need revalidation; use facet_lookup_valid()
3807 * instead if that is important. */
3808 static struct facet *
3809 facet_find(struct ofproto_dpif *ofproto,
3810 const struct flow *flow, uint32_t hash)
3812 struct facet *facet;
3814 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3815 if (flow_equal(flow, &facet->flow)) {
3823 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3824 * Returns it if found, otherwise a null pointer.
3826 * 'hash' must be the return value of flow_hash(flow, 0).
3828 * The returned facet is guaranteed to be valid. */
3829 static struct facet *
3830 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3833 struct facet *facet;
3835 facet = facet_find(ofproto, flow, hash);
3837 && (ofproto->need_revalidate
3838 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3839 facet_revalidate(facet);
3846 subfacet_path_to_string(enum subfacet_path path)
3849 case SF_NOT_INSTALLED:
3850 return "not installed";
3852 return "in fast path";
3854 return "in slow path";
3860 /* Returns the path in which a subfacet should be installed if its 'slow'
3861 * member has the specified value. */
3862 static enum subfacet_path
3863 subfacet_want_path(enum slow_path_reason slow)
3865 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3868 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3869 * supposing that its actions have been recalculated as 'want_actions' and that
3870 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3872 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3873 const struct ofpbuf *want_actions)
3875 enum subfacet_path want_path = subfacet_want_path(slow);
3876 return (want_path != subfacet->path
3877 || (want_path == SF_FAST_PATH
3878 && (subfacet->actions_len != want_actions->size
3879 || memcmp(subfacet->actions, want_actions->data,
3880 subfacet->actions_len))));
3884 facet_check_consistency(struct facet *facet)
3886 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3890 uint64_t odp_actions_stub[1024 / 8];
3891 struct ofpbuf odp_actions;
3893 struct rule_dpif *rule;
3894 struct subfacet *subfacet;
3895 bool may_log = false;
3898 /* Check the rule for consistency. */
3899 rule = rule_dpif_lookup(ofproto, &facet->flow);
3900 ok = rule == facet->rule;
3902 may_log = !VLOG_DROP_WARN(&rl);
3907 flow_format(&s, &facet->flow);
3908 ds_put_format(&s, ": facet associated with wrong rule (was "
3909 "table=%"PRIu8",", facet->rule->up.table_id);
3910 cls_rule_format(&facet->rule->up.cr, &s);
3911 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3913 cls_rule_format(&rule->up.cr, &s);
3914 ds_put_char(&s, ')');
3916 VLOG_WARN("%s", ds_cstr(&s));
3921 /* Check the datapath actions for consistency. */
3922 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3923 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3924 enum subfacet_path want_path;
3925 struct odputil_keybuf keybuf;
3926 struct action_xlate_ctx ctx;
3930 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3931 subfacet->initial_tci, rule, 0, NULL);
3932 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions,
3935 if (subfacet->path == SF_NOT_INSTALLED) {
3936 /* This only happens if the datapath reported an error when we
3937 * tried to install the flow. Don't flag another error here. */
3941 want_path = subfacet_want_path(subfacet->slow);
3942 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
3943 /* The actions for slow-path flows may legitimately vary from one
3944 * packet to the next. We're done. */
3948 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
3952 /* Inconsistency! */
3954 may_log = !VLOG_DROP_WARN(&rl);
3958 /* Rate-limited, skip reporting. */
3963 subfacet_get_key(subfacet, &keybuf, &key);
3964 odp_flow_key_format(key.data, key.size, &s);
3966 ds_put_cstr(&s, ": inconsistency in subfacet");
3967 if (want_path != subfacet->path) {
3968 enum odp_key_fitness fitness = subfacet->key_fitness;
3970 ds_put_format(&s, " (%s, fitness=%s)",
3971 subfacet_path_to_string(subfacet->path),
3972 odp_key_fitness_to_string(fitness));
3973 ds_put_format(&s, " (should have been %s)",
3974 subfacet_path_to_string(want_path));
3975 } else if (want_path == SF_FAST_PATH) {
3976 ds_put_cstr(&s, " (actions were: ");
3977 format_odp_actions(&s, subfacet->actions,
3978 subfacet->actions_len);
3979 ds_put_cstr(&s, ") (correct actions: ");
3980 format_odp_actions(&s, odp_actions.data, odp_actions.size);
3981 ds_put_char(&s, ')');
3983 ds_put_cstr(&s, " (actions: ");
3984 format_odp_actions(&s, subfacet->actions,
3985 subfacet->actions_len);
3986 ds_put_char(&s, ')');
3988 VLOG_WARN("%s", ds_cstr(&s));
3991 ofpbuf_uninit(&odp_actions);
3996 /* Re-searches the classifier for 'facet':
3998 * - If the rule found is different from 'facet''s current rule, moves
3999 * 'facet' to the new rule and recompiles its actions.
4001 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4002 * where it is and recompiles its actions anyway. */
4004 facet_revalidate(struct facet *facet)
4006 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4008 struct nlattr *odp_actions;
4011 struct actions *new_actions;
4013 struct action_xlate_ctx ctx;
4014 uint64_t odp_actions_stub[1024 / 8];
4015 struct ofpbuf odp_actions;
4017 struct rule_dpif *new_rule;
4018 struct subfacet *subfacet;
4021 COVERAGE_INC(facet_revalidate);
4023 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4025 /* Calculate new datapath actions.
4027 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4028 * emit a NetFlow expiration and, if so, we need to have the old state
4029 * around to properly compose it. */
4031 /* If the datapath actions changed or the installability changed,
4032 * then we need to talk to the datapath. */
4035 memset(&ctx, 0, sizeof ctx);
4036 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4037 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4038 enum slow_path_reason slow;
4040 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4041 subfacet->initial_tci, new_rule, 0, NULL);
4042 xlate_actions(&ctx, new_rule->up.actions, new_rule->up.n_actions,
4045 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4046 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4047 struct dpif_flow_stats stats;
4049 subfacet_install(subfacet,
4050 odp_actions.data, odp_actions.size, &stats, slow);
4051 subfacet_update_stats(subfacet, &stats);
4054 new_actions = xcalloc(list_size(&facet->subfacets),
4055 sizeof *new_actions);
4057 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4059 new_actions[i].actions_len = odp_actions.size;
4064 ofpbuf_uninit(&odp_actions);
4067 facet_flush_stats(facet);
4070 /* Update 'facet' now that we've taken care of all the old state. */
4071 facet->tags = ctx.tags;
4072 facet->nf_flow.output_iface = ctx.nf_output_iface;
4073 facet->has_learn = ctx.has_learn;
4074 facet->has_normal = ctx.has_normal;
4075 facet->has_fin_timeout = ctx.has_fin_timeout;
4076 facet->mirrors = ctx.mirrors;
4079 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4080 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4082 if (new_actions && new_actions[i].odp_actions) {
4083 free(subfacet->actions);
4084 subfacet->actions = new_actions[i].odp_actions;
4085 subfacet->actions_len = new_actions[i].actions_len;
4091 if (facet->rule != new_rule) {
4092 COVERAGE_INC(facet_changed_rule);
4093 list_remove(&facet->list_node);
4094 list_push_back(&new_rule->facets, &facet->list_node);
4095 facet->rule = new_rule;
4096 facet->used = new_rule->up.created;
4097 facet->prev_used = facet->used;
4101 /* Updates 'facet''s used time. Caller is responsible for calling
4102 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4104 facet_update_time(struct facet *facet, long long int used)
4106 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4107 if (used > facet->used) {
4109 ofproto_rule_update_used(&facet->rule->up, used);
4110 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4115 facet_reset_counters(struct facet *facet)
4117 facet->packet_count = 0;
4118 facet->byte_count = 0;
4119 facet->prev_packet_count = 0;
4120 facet->prev_byte_count = 0;
4121 facet->accounted_bytes = 0;
4125 facet_push_stats(struct facet *facet)
4127 struct dpif_flow_stats stats;
4129 assert(facet->packet_count >= facet->prev_packet_count);
4130 assert(facet->byte_count >= facet->prev_byte_count);
4131 assert(facet->used >= facet->prev_used);
4133 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4134 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4135 stats.used = facet->used;
4136 stats.tcp_flags = 0;
4138 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4139 facet->prev_packet_count = facet->packet_count;
4140 facet->prev_byte_count = facet->byte_count;
4141 facet->prev_used = facet->used;
4143 flow_push_stats(facet->rule, &facet->flow, &stats);
4145 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4146 facet->mirrors, stats.n_packets, stats.n_bytes);
4151 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4153 rule->packet_count += stats->n_packets;
4154 rule->byte_count += stats->n_bytes;
4155 ofproto_rule_update_used(&rule->up, stats->used);
4158 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4159 * 'rule''s actions and mirrors. */
4161 flow_push_stats(struct rule_dpif *rule,
4162 const struct flow *flow, const struct dpif_flow_stats *stats)
4164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4165 struct action_xlate_ctx ctx;
4167 ofproto_rule_update_used(&rule->up, stats->used);
4169 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4171 ctx.resubmit_stats = stats;
4172 xlate_actions_for_side_effects(&ctx, rule->up.actions, rule->up.n_actions);
4177 static struct subfacet *
4178 subfacet_find__(struct ofproto_dpif *ofproto,
4179 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4180 const struct flow *flow)
4182 struct subfacet *subfacet;
4184 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4185 &ofproto->subfacets) {
4187 ? (subfacet->key_len == key_len
4188 && !memcmp(key, subfacet->key, key_len))
4189 : flow_equal(flow, &subfacet->facet->flow)) {
4197 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4198 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4199 * there is one, otherwise creates and returns a new subfacet.
4201 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4202 * which case the caller must populate the actions with
4203 * subfacet_make_actions(). */
4204 static struct subfacet *
4205 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4206 const struct nlattr *key, size_t key_len, ovs_be16 initial_tci)
4208 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4209 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4210 struct subfacet *subfacet;
4212 subfacet = subfacet_find__(ofproto, key, key_len, key_hash, &facet->flow);
4214 if (subfacet->facet == facet) {
4218 /* This shouldn't happen. */
4219 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4220 subfacet_destroy(subfacet);
4223 subfacet = (list_is_empty(&facet->subfacets)
4224 ? &facet->one_subfacet
4225 : xmalloc(sizeof *subfacet));
4226 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4227 list_push_back(&facet->subfacets, &subfacet->list_node);
4228 subfacet->facet = facet;
4229 subfacet->key_fitness = key_fitness;
4230 if (key_fitness != ODP_FIT_PERFECT) {
4231 subfacet->key = xmemdup(key, key_len);
4232 subfacet->key_len = key_len;
4234 subfacet->key = NULL;
4235 subfacet->key_len = 0;
4237 subfacet->used = time_msec();
4238 subfacet->dp_packet_count = 0;
4239 subfacet->dp_byte_count = 0;
4240 subfacet->actions_len = 0;
4241 subfacet->actions = NULL;
4242 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4245 subfacet->path = SF_NOT_INSTALLED;
4246 subfacet->initial_tci = initial_tci;
4251 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4252 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4253 static struct subfacet *
4254 subfacet_find(struct ofproto_dpif *ofproto,
4255 const struct nlattr *key, size_t key_len)
4257 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4258 enum odp_key_fitness fitness;
4261 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4262 if (fitness == ODP_FIT_ERROR) {
4266 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4269 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4270 * its facet within 'ofproto', and frees it. */
4272 subfacet_destroy__(struct subfacet *subfacet)
4274 struct facet *facet = subfacet->facet;
4275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4277 subfacet_uninstall(subfacet);
4278 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4279 list_remove(&subfacet->list_node);
4280 free(subfacet->key);
4281 free(subfacet->actions);
4282 if (subfacet != &facet->one_subfacet) {
4287 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4288 * last remaining subfacet in its facet destroys the facet too. */
4290 subfacet_destroy(struct subfacet *subfacet)
4292 struct facet *facet = subfacet->facet;
4294 if (list_is_singleton(&facet->subfacets)) {
4295 /* facet_remove() needs at least one subfacet (it will remove it). */
4296 facet_remove(facet);
4298 subfacet_destroy__(subfacet);
4302 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4303 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4304 * for use as temporary storage. */
4306 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4309 if (!subfacet->key) {
4310 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4311 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4313 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4317 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4318 * Translates the actions into 'odp_actions', which the caller must have
4319 * initialized and is responsible for uninitializing. */
4321 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4322 struct ofpbuf *odp_actions)
4324 struct facet *facet = subfacet->facet;
4325 struct rule_dpif *rule = facet->rule;
4326 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4328 struct action_xlate_ctx ctx;
4330 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4332 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, odp_actions);
4333 facet->tags = ctx.tags;
4334 facet->has_learn = ctx.has_learn;
4335 facet->has_normal = ctx.has_normal;
4336 facet->has_fin_timeout = ctx.has_fin_timeout;
4337 facet->nf_flow.output_iface = ctx.nf_output_iface;
4338 facet->mirrors = ctx.mirrors;
4340 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4341 if (subfacet->actions_len != odp_actions->size
4342 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4343 free(subfacet->actions);
4344 subfacet->actions_len = odp_actions->size;
4345 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4349 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4350 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4351 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4352 * since 'subfacet' was last updated.
4354 * Returns 0 if successful, otherwise a positive errno value. */
4356 subfacet_install(struct subfacet *subfacet,
4357 const struct nlattr *actions, size_t actions_len,
4358 struct dpif_flow_stats *stats,
4359 enum slow_path_reason slow)
4361 struct facet *facet = subfacet->facet;
4362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4363 enum subfacet_path path = subfacet_want_path(slow);
4364 uint64_t slow_path_stub[128 / 8];
4365 struct odputil_keybuf keybuf;
4366 enum dpif_flow_put_flags flags;
4370 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4372 flags |= DPIF_FP_ZERO_STATS;
4375 if (path == SF_SLOW_PATH) {
4376 compose_slow_path(ofproto, &facet->flow, slow,
4377 slow_path_stub, sizeof slow_path_stub,
4378 &actions, &actions_len);
4381 subfacet_get_key(subfacet, &keybuf, &key);
4382 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4383 actions, actions_len, stats);
4386 subfacet_reset_dp_stats(subfacet, stats);
4390 subfacet->path = path;
4396 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4398 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4399 stats, subfacet->slow);
4402 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4404 subfacet_uninstall(struct subfacet *subfacet)
4406 if (subfacet->path != SF_NOT_INSTALLED) {
4407 struct rule_dpif *rule = subfacet->facet->rule;
4408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4409 struct odputil_keybuf keybuf;
4410 struct dpif_flow_stats stats;
4414 subfacet_get_key(subfacet, &keybuf, &key);
4415 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4416 subfacet_reset_dp_stats(subfacet, &stats);
4418 subfacet_update_stats(subfacet, &stats);
4420 subfacet->path = SF_NOT_INSTALLED;
4422 assert(subfacet->dp_packet_count == 0);
4423 assert(subfacet->dp_byte_count == 0);
4427 /* Resets 'subfacet''s datapath statistics counters. This should be called
4428 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4429 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4430 * was reset in the datapath. 'stats' will be modified to include only
4431 * statistics new since 'subfacet' was last updated. */
4433 subfacet_reset_dp_stats(struct subfacet *subfacet,
4434 struct dpif_flow_stats *stats)
4437 && subfacet->dp_packet_count <= stats->n_packets
4438 && subfacet->dp_byte_count <= stats->n_bytes) {
4439 stats->n_packets -= subfacet->dp_packet_count;
4440 stats->n_bytes -= subfacet->dp_byte_count;
4443 subfacet->dp_packet_count = 0;
4444 subfacet->dp_byte_count = 0;
4447 /* Updates 'subfacet''s used time. The caller is responsible for calling
4448 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4450 subfacet_update_time(struct subfacet *subfacet, long long int used)
4452 if (used > subfacet->used) {
4453 subfacet->used = used;
4454 facet_update_time(subfacet->facet, used);
4458 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4460 * Because of the meaning of a subfacet's counters, it only makes sense to do
4461 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4462 * represents a packet that was sent by hand or if it represents statistics
4463 * that have been cleared out of the datapath. */
4465 subfacet_update_stats(struct subfacet *subfacet,
4466 const struct dpif_flow_stats *stats)
4468 if (stats->n_packets || stats->used > subfacet->used) {
4469 struct facet *facet = subfacet->facet;
4471 subfacet_update_time(subfacet, stats->used);
4472 facet->packet_count += stats->n_packets;
4473 facet->byte_count += stats->n_bytes;
4474 facet->tcp_flags |= stats->tcp_flags;
4475 facet_push_stats(facet);
4476 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4482 static struct rule_dpif *
4483 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4485 struct ofport_dpif *port;
4486 struct rule_dpif *rule;
4488 rule = rule_dpif_lookup__(ofproto, flow, 0);
4493 port = get_ofp_port(ofproto, flow->in_port);
4495 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4496 return ofproto->miss_rule;
4499 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4500 return ofproto->no_packet_in_rule;
4502 return ofproto->miss_rule;
4505 static struct rule_dpif *
4506 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4509 struct cls_rule *cls_rule;
4510 struct classifier *cls;
4512 if (table_id >= N_TABLES) {
4516 cls = &ofproto->up.tables[table_id].cls;
4517 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4518 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4519 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4520 * are unavailable. */
4521 struct flow ofpc_normal_flow = *flow;
4522 ofpc_normal_flow.tp_src = htons(0);
4523 ofpc_normal_flow.tp_dst = htons(0);
4524 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4526 cls_rule = classifier_lookup(cls, flow);
4528 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4532 complete_operation(struct rule_dpif *rule)
4534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4536 rule_invalidate(rule);
4538 struct dpif_completion *c = xmalloc(sizeof *c);
4539 c->op = rule->up.pending;
4540 list_push_back(&ofproto->completions, &c->list_node);
4542 ofoperation_complete(rule->up.pending, 0);
4546 static struct rule *
4549 struct rule_dpif *rule = xmalloc(sizeof *rule);
4554 rule_dealloc(struct rule *rule_)
4556 struct rule_dpif *rule = rule_dpif_cast(rule_);
4561 rule_construct(struct rule *rule_)
4563 struct rule_dpif *rule = rule_dpif_cast(rule_);
4564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4565 struct rule_dpif *victim;
4569 error = validate_actions(rule->up.actions, rule->up.n_actions,
4570 &rule->up.cr.flow, ofproto->max_ports);
4575 rule->packet_count = 0;
4576 rule->byte_count = 0;
4578 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4579 if (victim && !list_is_empty(&victim->facets)) {
4580 struct facet *facet;
4582 rule->facets = victim->facets;
4583 list_moved(&rule->facets);
4584 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4585 /* XXX: We're only clearing our local counters here. It's possible
4586 * that quite a few packets are unaccounted for in the datapath
4587 * statistics. These will be accounted to the new rule instead of
4588 * cleared as required. This could be fixed by clearing out the
4589 * datapath statistics for this facet, but currently it doesn't
4591 facet_reset_counters(facet);
4595 /* Must avoid list_moved() in this case. */
4596 list_init(&rule->facets);
4599 table_id = rule->up.table_id;
4600 rule->tag = (victim ? victim->tag
4602 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4603 ofproto->tables[table_id].basis));
4605 complete_operation(rule);
4610 rule_destruct(struct rule *rule_)
4612 struct rule_dpif *rule = rule_dpif_cast(rule_);
4613 struct facet *facet, *next_facet;
4615 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4616 facet_revalidate(facet);
4619 complete_operation(rule);
4623 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4625 struct rule_dpif *rule = rule_dpif_cast(rule_);
4626 struct facet *facet;
4628 /* Start from historical data for 'rule' itself that are no longer tracked
4629 * in facets. This counts, for example, facets that have expired. */
4630 *packets = rule->packet_count;
4631 *bytes = rule->byte_count;
4633 /* Add any statistics that are tracked by facets. This includes
4634 * statistical data recently updated by ofproto_update_stats() as well as
4635 * stats for packets that were executed "by hand" via dpif_execute(). */
4636 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4637 *packets += facet->packet_count;
4638 *bytes += facet->byte_count;
4643 rule_execute(struct rule *rule_, const struct flow *flow,
4644 struct ofpbuf *packet)
4646 struct rule_dpif *rule = rule_dpif_cast(rule_);
4647 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4649 struct dpif_flow_stats stats;
4651 struct action_xlate_ctx ctx;
4652 uint64_t odp_actions_stub[1024 / 8];
4653 struct ofpbuf odp_actions;
4655 dpif_flow_stats_extract(flow, packet, &stats);
4656 rule_credit_stats(rule, &stats);
4658 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4659 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4660 rule, stats.tcp_flags, packet);
4661 ctx.resubmit_stats = &stats;
4662 xlate_actions(&ctx, rule->up.actions, rule->up.n_actions, &odp_actions);
4664 execute_odp_actions(ofproto, flow, odp_actions.data,
4665 odp_actions.size, packet);
4667 ofpbuf_uninit(&odp_actions);
4673 rule_modify_actions(struct rule *rule_)
4675 struct rule_dpif *rule = rule_dpif_cast(rule_);
4676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4679 error = validate_actions(rule->up.actions, rule->up.n_actions,
4680 &rule->up.cr.flow, ofproto->max_ports);
4682 ofoperation_complete(rule->up.pending, error);
4686 complete_operation(rule);
4689 /* Sends 'packet' out 'ofport'.
4690 * May modify 'packet'.
4691 * Returns 0 if successful, otherwise a positive errno value. */
4693 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4695 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4696 struct ofpbuf key, odp_actions;
4697 struct odputil_keybuf keybuf;
4702 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
4703 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4705 if (odp_port != ofport->odp_port) {
4706 eth_pop_vlan(packet);
4707 flow.vlan_tci = htons(0);
4710 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4711 odp_flow_key_from_flow(&key, &flow);
4713 ofpbuf_init(&odp_actions, 32);
4714 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4716 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4717 error = dpif_execute(ofproto->dpif,
4719 odp_actions.data, odp_actions.size,
4721 ofpbuf_uninit(&odp_actions);
4724 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4725 ofproto->up.name, odp_port, strerror(error));
4727 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4731 /* OpenFlow to datapath action translation. */
4733 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
4734 struct action_xlate_ctx *ctx);
4735 static void xlate_normal(struct action_xlate_ctx *);
4737 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4738 * The action will state 'slow' as the reason that the action is in the slow
4739 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4740 * dump-flows" output to see why a flow is in the slow path.)
4742 * The 'stub_size' bytes in 'stub' will be used to store the action.
4743 * 'stub_size' must be large enough for the action.
4745 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4748 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4749 enum slow_path_reason slow,
4750 uint64_t *stub, size_t stub_size,
4751 const struct nlattr **actionsp, size_t *actions_lenp)
4753 union user_action_cookie cookie;
4756 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4757 cookie.slow_path.unused = 0;
4758 cookie.slow_path.reason = slow;
4760 ofpbuf_use_stack(&buf, stub, stub_size);
4761 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4762 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4763 odp_put_userspace_action(pid, &cookie, &buf);
4765 put_userspace_action(ofproto, &buf, flow, &cookie);
4767 *actionsp = buf.data;
4768 *actions_lenp = buf.size;
4772 put_userspace_action(const struct ofproto_dpif *ofproto,
4773 struct ofpbuf *odp_actions,
4774 const struct flow *flow,
4775 const union user_action_cookie *cookie)
4779 pid = dpif_port_get_pid(ofproto->dpif,
4780 ofp_port_to_odp_port(flow->in_port));
4782 return odp_put_userspace_action(pid, cookie, odp_actions);
4786 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4787 ovs_be16 vlan_tci, uint32_t odp_port,
4788 unsigned int n_outputs, union user_action_cookie *cookie)
4792 cookie->type = USER_ACTION_COOKIE_SFLOW;
4793 cookie->sflow.vlan_tci = vlan_tci;
4795 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4796 * port information") for the interpretation of cookie->output. */
4797 switch (n_outputs) {
4799 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4800 cookie->sflow.output = 0x40000000 | 256;
4804 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4806 cookie->sflow.output = ifindex;
4811 /* 0x80000000 means "multiple output ports. */
4812 cookie->sflow.output = 0x80000000 | n_outputs;
4817 /* Compose SAMPLE action for sFlow. */
4819 compose_sflow_action(const struct ofproto_dpif *ofproto,
4820 struct ofpbuf *odp_actions,
4821 const struct flow *flow,
4824 uint32_t probability;
4825 union user_action_cookie cookie;
4826 size_t sample_offset, actions_offset;
4829 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4833 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4835 /* Number of packets out of UINT_MAX to sample. */
4836 probability = dpif_sflow_get_probability(ofproto->sflow);
4837 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4839 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4840 compose_sflow_cookie(ofproto, htons(0), odp_port,
4841 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4842 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4844 nl_msg_end_nested(odp_actions, actions_offset);
4845 nl_msg_end_nested(odp_actions, sample_offset);
4846 return cookie_offset;
4849 /* SAMPLE action must be first action in any given list of actions.
4850 * At this point we do not have all information required to build it. So try to
4851 * build sample action as complete as possible. */
4853 add_sflow_action(struct action_xlate_ctx *ctx)
4855 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4857 &ctx->flow, OVSP_NONE);
4858 ctx->sflow_odp_port = 0;
4859 ctx->sflow_n_outputs = 0;
4862 /* Fix SAMPLE action according to data collected while composing ODP actions.
4863 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4864 * USERSPACE action's user-cookie which is required for sflow. */
4866 fix_sflow_action(struct action_xlate_ctx *ctx)
4868 const struct flow *base = &ctx->base_flow;
4869 union user_action_cookie *cookie;
4871 if (!ctx->user_cookie_offset) {
4875 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4877 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4879 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4880 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4884 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4887 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4888 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4889 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4890 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4894 struct priority_to_dscp *pdscp;
4896 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4897 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4901 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4903 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4904 ctx->flow.nw_tos |= pdscp->dscp;
4907 /* We may not have an ofport record for this port, but it doesn't hurt
4908 * to allow forwarding to it anyhow. Maybe such a port will appear
4909 * later and we're pre-populating the flow table. */
4912 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4913 ctx->flow.vlan_tci);
4914 if (out_port != odp_port) {
4915 ctx->flow.vlan_tci = htons(0);
4917 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4918 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4920 ctx->sflow_odp_port = odp_port;
4921 ctx->sflow_n_outputs++;
4922 ctx->nf_output_iface = ofp_port;
4923 ctx->flow.vlan_tci = flow_vlan_tci;
4924 ctx->flow.nw_tos = flow_nw_tos;
4928 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
4930 compose_output_action__(ctx, ofp_port, true);
4934 xlate_table_action(struct action_xlate_ctx *ctx,
4935 uint16_t in_port, uint8_t table_id)
4937 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
4938 struct ofproto_dpif *ofproto = ctx->ofproto;
4939 struct rule_dpif *rule;
4940 uint16_t old_in_port;
4941 uint8_t old_table_id;
4943 old_table_id = ctx->table_id;
4944 ctx->table_id = table_id;
4946 /* Look up a flow with 'in_port' as the input port. */
4947 old_in_port = ctx->flow.in_port;
4948 ctx->flow.in_port = in_port;
4949 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
4952 if (table_id > 0 && table_id < N_TABLES) {
4953 struct table_dpif *table = &ofproto->tables[table_id];
4954 if (table->other_table) {
4955 ctx->tags |= (rule && rule->tag
4957 : rule_calculate_tag(&ctx->flow,
4958 &table->other_table->wc,
4963 /* Restore the original input port. Otherwise OFPP_NORMAL and
4964 * OFPP_IN_PORT will have surprising behavior. */
4965 ctx->flow.in_port = old_in_port;
4967 if (ctx->resubmit_hook) {
4968 ctx->resubmit_hook(ctx, rule);
4972 struct rule_dpif *old_rule = ctx->rule;
4974 if (ctx->resubmit_stats) {
4975 rule_credit_stats(rule, ctx->resubmit_stats);
4980 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
4981 ctx->rule = old_rule;
4985 ctx->table_id = old_table_id;
4987 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4989 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
4990 MAX_RESUBMIT_RECURSION);
4991 ctx->max_resubmit_trigger = true;
4996 xlate_resubmit_table(struct action_xlate_ctx *ctx,
4997 const struct nx_action_resubmit *nar)
5002 in_port = (nar->in_port == htons(OFPP_IN_PORT)
5004 : ntohs(nar->in_port));
5005 table_id = nar->table == 255 ? ctx->table_id : nar->table;
5007 xlate_table_action(ctx, in_port, table_id);
5011 flood_packets(struct action_xlate_ctx *ctx, bool all)
5013 struct ofport_dpif *ofport;
5015 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5016 uint16_t ofp_port = ofport->up.ofp_port;
5018 if (ofp_port == ctx->flow.in_port) {
5023 compose_output_action__(ctx, ofp_port, false);
5024 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5025 compose_output_action(ctx, ofp_port);
5029 ctx->nf_output_iface = NF_OUT_FLOOD;
5033 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5034 enum ofp_packet_in_reason reason,
5035 uint16_t controller_id)
5037 struct ofputil_packet_in pin;
5038 struct ofpbuf *packet;
5040 ctx->slow |= SLOW_CONTROLLER;
5045 packet = ofpbuf_clone(ctx->packet);
5047 if (packet->l2 && packet->l3) {
5048 struct eth_header *eh;
5050 eth_pop_vlan(packet);
5053 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5054 * LLC frame. Calculating the Ethernet type of these frames is more
5055 * trouble than seems appropriate for a simple assertion. */
5056 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5057 || eh->eth_type == ctx->flow.dl_type);
5059 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5060 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5062 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5063 eth_push_vlan(packet, ctx->flow.vlan_tci);
5067 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5068 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5069 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5073 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5074 packet_set_tcp_port(packet, ctx->flow.tp_src,
5076 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5077 packet_set_udp_port(packet, ctx->flow.tp_src,
5084 pin.packet = packet->data;
5085 pin.packet_len = packet->size;
5086 pin.reason = reason;
5087 pin.controller_id = controller_id;
5088 pin.table_id = ctx->table_id;
5089 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5092 flow_get_metadata(&ctx->flow, &pin.fmd);
5094 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5095 ofpbuf_delete(packet);
5099 compose_dec_ttl(struct action_xlate_ctx *ctx)
5101 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5102 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5106 if (ctx->flow.nw_ttl > 1) {
5110 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
5112 /* Stop processing for current table. */
5118 xlate_output_action__(struct action_xlate_ctx *ctx,
5119 uint16_t port, uint16_t max_len)
5121 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5123 ctx->nf_output_iface = NF_OUT_DROP;
5127 compose_output_action(ctx, ctx->flow.in_port);
5130 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
5136 flood_packets(ctx, false);
5139 flood_packets(ctx, true);
5141 case OFPP_CONTROLLER:
5142 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5148 if (port != ctx->flow.in_port) {
5149 compose_output_action(ctx, port);
5154 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5155 ctx->nf_output_iface = NF_OUT_FLOOD;
5156 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5157 ctx->nf_output_iface = prev_nf_output_iface;
5158 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5159 ctx->nf_output_iface != NF_OUT_FLOOD) {
5160 ctx->nf_output_iface = NF_OUT_MULTI;
5165 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5166 const struct nx_action_output_reg *naor)
5168 struct mf_subfield src;
5171 nxm_decode(&src, naor->src, naor->ofs_nbits);
5172 ofp_port = mf_get_subfield(&src, &ctx->flow);
5174 if (ofp_port <= UINT16_MAX) {
5175 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
5180 xlate_output_action(struct action_xlate_ctx *ctx,
5181 const struct ofp_action_output *oao)
5183 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
5187 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5188 const struct ofp_action_enqueue *oae)
5191 uint32_t flow_priority, priority;
5194 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
5197 /* Fall back to ordinary output action. */
5198 xlate_output_action__(ctx, ntohs(oae->port), 0);
5202 /* Figure out datapath output port. */
5203 ofp_port = ntohs(oae->port);
5204 if (ofp_port == OFPP_IN_PORT) {
5205 ofp_port = ctx->flow.in_port;
5206 } else if (ofp_port == ctx->flow.in_port) {
5210 /* Add datapath actions. */
5211 flow_priority = ctx->flow.skb_priority;
5212 ctx->flow.skb_priority = priority;
5213 compose_output_action(ctx, ofp_port);
5214 ctx->flow.skb_priority = flow_priority;
5216 /* Update NetFlow output port. */
5217 if (ctx->nf_output_iface == NF_OUT_DROP) {
5218 ctx->nf_output_iface = ofp_port;
5219 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5220 ctx->nf_output_iface = NF_OUT_MULTI;
5225 xlate_set_queue_action(struct action_xlate_ctx *ctx,
5226 const struct nx_action_set_queue *nasq)
5231 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
5234 /* Couldn't translate queue to a priority, so ignore. A warning
5235 * has already been logged. */
5239 ctx->flow.skb_priority = priority;
5242 struct xlate_reg_state {
5248 xlate_autopath(struct action_xlate_ctx *ctx,
5249 const struct nx_action_autopath *naa)
5251 uint16_t ofp_port = ntohl(naa->id);
5252 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5254 if (!port || !port->bundle) {
5255 ofp_port = OFPP_NONE;
5256 } else if (port->bundle->bond) {
5257 /* Autopath does not support VLAN hashing. */
5258 struct ofport_dpif *slave = bond_choose_output_slave(
5259 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5261 ofp_port = slave->up.ofp_port;
5264 autopath_execute(naa, &ctx->flow, ofp_port);
5268 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5270 struct ofproto_dpif *ofproto = ofproto_;
5271 struct ofport_dpif *port;
5281 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5284 port = get_ofp_port(ofproto, ofp_port);
5285 return port ? port->may_enable : false;
5290 xlate_learn_action(struct action_xlate_ctx *ctx,
5291 const struct nx_action_learn *learn)
5293 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5294 struct ofputil_flow_mod fm;
5297 learn_execute(learn, &ctx->flow, &fm);
5299 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5300 if (error && !VLOG_DROP_WARN(&rl)) {
5301 VLOG_WARN("learning action failed to modify flow table (%s)",
5302 ofperr_get_name(error));
5308 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5309 * means "infinite". */
5311 reduce_timeout(uint16_t max, uint16_t *timeout)
5313 if (max && (!*timeout || *timeout > max)) {
5319 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5320 const struct nx_action_fin_timeout *naft)
5322 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5323 struct rule_dpif *rule = ctx->rule;
5325 reduce_timeout(ntohs(naft->fin_idle_timeout), &rule->up.idle_timeout);
5326 reduce_timeout(ntohs(naft->fin_hard_timeout), &rule->up.hard_timeout);
5331 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5333 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5334 ? OFPUTIL_PC_NO_RECV_STP
5335 : OFPUTIL_PC_NO_RECV)) {
5339 /* Only drop packets here if both forwarding and learning are
5340 * disabled. If just learning is enabled, we need to have
5341 * OFPP_NORMAL and the learning action have a look at the packet
5342 * before we can drop it. */
5343 if (!stp_forward_in_state(port->stp_state)
5344 && !stp_learn_in_state(port->stp_state)) {
5352 do_xlate_actions(const union ofp_action *in, size_t n_in,
5353 struct action_xlate_ctx *ctx)
5355 const struct ofport_dpif *port;
5356 const union ofp_action *ia;
5357 bool was_evictable = true;
5360 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5361 if (port && !may_receive(port, ctx)) {
5362 /* Drop this flow. */
5367 /* Don't let the rule we're working on get evicted underneath us. */
5368 was_evictable = ctx->rule->up.evictable;
5369 ctx->rule->up.evictable = false;
5371 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
5372 const struct ofp_action_dl_addr *oada;
5373 const struct nx_action_resubmit *nar;
5374 const struct nx_action_set_tunnel *nast;
5375 const struct nx_action_set_queue *nasq;
5376 const struct nx_action_multipath *nam;
5377 const struct nx_action_autopath *naa;
5378 const struct nx_action_bundle *nab;
5379 const struct nx_action_output_reg *naor;
5380 const struct nx_action_controller *nac;
5381 enum ofputil_action_code code;
5388 code = ofputil_decode_action_unsafe(ia);
5390 case OFPUTIL_OFPAT10_OUTPUT:
5391 xlate_output_action(ctx, &ia->output);
5394 case OFPUTIL_OFPAT10_SET_VLAN_VID:
5395 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5396 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
5399 case OFPUTIL_OFPAT10_SET_VLAN_PCP:
5400 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5401 ctx->flow.vlan_tci |= htons(
5402 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
5405 case OFPUTIL_OFPAT10_STRIP_VLAN:
5406 ctx->flow.vlan_tci = htons(0);
5409 case OFPUTIL_OFPAT10_SET_DL_SRC:
5410 oada = ((struct ofp_action_dl_addr *) ia);
5411 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
5414 case OFPUTIL_OFPAT10_SET_DL_DST:
5415 oada = ((struct ofp_action_dl_addr *) ia);
5416 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
5419 case OFPUTIL_OFPAT10_SET_NW_SRC:
5420 ctx->flow.nw_src = ia->nw_addr.nw_addr;
5423 case OFPUTIL_OFPAT10_SET_NW_DST:
5424 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
5427 case OFPUTIL_OFPAT10_SET_NW_TOS:
5428 /* OpenFlow 1.0 only supports IPv4. */
5429 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5430 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5431 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
5435 case OFPUTIL_OFPAT10_SET_TP_SRC:
5436 ctx->flow.tp_src = ia->tp_port.tp_port;
5439 case OFPUTIL_OFPAT10_SET_TP_DST:
5440 ctx->flow.tp_dst = ia->tp_port.tp_port;
5443 case OFPUTIL_OFPAT10_ENQUEUE:
5444 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
5447 case OFPUTIL_NXAST_RESUBMIT:
5448 nar = (const struct nx_action_resubmit *) ia;
5449 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
5452 case OFPUTIL_NXAST_RESUBMIT_TABLE:
5453 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
5456 case OFPUTIL_NXAST_SET_TUNNEL:
5457 nast = (const struct nx_action_set_tunnel *) ia;
5458 tun_id = htonll(ntohl(nast->tun_id));
5459 ctx->flow.tun_id = tun_id;
5462 case OFPUTIL_NXAST_SET_QUEUE:
5463 nasq = (const struct nx_action_set_queue *) ia;
5464 xlate_set_queue_action(ctx, nasq);
5467 case OFPUTIL_NXAST_POP_QUEUE:
5468 ctx->flow.skb_priority = ctx->orig_skb_priority;
5471 case OFPUTIL_NXAST_REG_MOVE:
5472 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
5476 case OFPUTIL_NXAST_REG_LOAD:
5477 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
5481 case OFPUTIL_NXAST_NOTE:
5482 /* Nothing to do. */
5485 case OFPUTIL_NXAST_SET_TUNNEL64:
5486 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
5487 ctx->flow.tun_id = tun_id;
5490 case OFPUTIL_NXAST_MULTIPATH:
5491 nam = (const struct nx_action_multipath *) ia;
5492 multipath_execute(nam, &ctx->flow);
5495 case OFPUTIL_NXAST_AUTOPATH:
5496 naa = (const struct nx_action_autopath *) ia;
5497 xlate_autopath(ctx, naa);
5500 case OFPUTIL_NXAST_BUNDLE:
5501 ctx->ofproto->has_bundle_action = true;
5502 nab = (const struct nx_action_bundle *) ia;
5503 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
5508 case OFPUTIL_NXAST_BUNDLE_LOAD:
5509 ctx->ofproto->has_bundle_action = true;
5510 nab = (const struct nx_action_bundle *) ia;
5511 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
5515 case OFPUTIL_NXAST_OUTPUT_REG:
5516 naor = (const struct nx_action_output_reg *) ia;
5517 xlate_output_reg_action(ctx, naor);
5520 case OFPUTIL_NXAST_LEARN:
5521 ctx->has_learn = true;
5522 if (ctx->may_learn) {
5523 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
5527 case OFPUTIL_NXAST_DEC_TTL:
5528 if (compose_dec_ttl(ctx)) {
5533 case OFPUTIL_NXAST_EXIT:
5537 case OFPUTIL_NXAST_FIN_TIMEOUT:
5538 ctx->has_fin_timeout = true;
5539 xlate_fin_timeout(ctx, (const struct nx_action_fin_timeout *) ia);
5542 case OFPUTIL_NXAST_CONTROLLER:
5543 nac = (const struct nx_action_controller *) ia;
5544 execute_controller_action(ctx, ntohs(nac->max_len), nac->reason,
5545 ntohs(nac->controller_id));
5551 /* We've let OFPP_NORMAL and the learning action look at the packet,
5552 * so drop it now if forwarding is disabled. */
5553 if (port && !stp_forward_in_state(port->stp_state)) {
5554 ofpbuf_clear(ctx->odp_actions);
5555 add_sflow_action(ctx);
5558 ctx->rule->up.evictable = was_evictable;
5563 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5564 struct ofproto_dpif *ofproto, const struct flow *flow,
5565 ovs_be16 initial_tci, struct rule_dpif *rule,
5566 uint8_t tcp_flags, const struct ofpbuf *packet)
5568 ctx->ofproto = ofproto;
5570 ctx->base_flow = ctx->flow;
5571 ctx->base_flow.tun_id = 0;
5572 ctx->base_flow.vlan_tci = initial_tci;
5574 ctx->packet = packet;
5575 ctx->may_learn = packet != NULL;
5576 ctx->tcp_flags = tcp_flags;
5577 ctx->resubmit_hook = NULL;
5578 ctx->resubmit_stats = NULL;
5581 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions in
5582 * 'odp_actions', using 'ctx'. */
5584 xlate_actions(struct action_xlate_ctx *ctx,
5585 const union ofp_action *in, size_t n_in,
5586 struct ofpbuf *odp_actions)
5588 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5589 * that in the future we always keep a copy of the original flow for
5590 * tracing purposes. */
5591 static bool hit_resubmit_limit;
5593 enum slow_path_reason special;
5595 COVERAGE_INC(ofproto_dpif_xlate);
5597 ofpbuf_clear(odp_actions);
5598 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5600 ctx->odp_actions = odp_actions;
5603 ctx->has_learn = false;
5604 ctx->has_normal = false;
5605 ctx->has_fin_timeout = false;
5606 ctx->nf_output_iface = NF_OUT_DROP;
5609 ctx->max_resubmit_trigger = false;
5610 ctx->orig_skb_priority = ctx->flow.skb_priority;
5614 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5615 /* Do this conditionally because the copy is expensive enough that it
5616 * shows up in profiles.
5618 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5619 * believe that I wasn't using it without initializing it if I kept it
5620 * in a local variable. */
5621 ctx->orig_flow = ctx->flow;
5624 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5625 switch (ctx->ofproto->up.frag_handling) {
5626 case OFPC_FRAG_NORMAL:
5627 /* We must pretend that transport ports are unavailable. */
5628 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5629 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5632 case OFPC_FRAG_DROP:
5635 case OFPC_FRAG_REASM:
5638 case OFPC_FRAG_NX_MATCH:
5639 /* Nothing to do. */
5642 case OFPC_INVALID_TTL_TO_CONTROLLER:
5647 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5649 ctx->slow |= special;
5651 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5652 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5654 add_sflow_action(ctx);
5655 do_xlate_actions(in, n_in, ctx);
5657 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5658 if (!hit_resubmit_limit) {
5659 /* We didn't record the original flow. Make sure we do from
5661 hit_resubmit_limit = true;
5662 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5663 struct ds ds = DS_EMPTY_INITIALIZER;
5665 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5667 VLOG_ERR("Trace triggered by excessive resubmit "
5668 "recursion:\n%s", ds_cstr(&ds));
5673 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5674 ctx->odp_actions->data,
5675 ctx->odp_actions->size)) {
5676 ctx->slow |= SLOW_IN_BAND;
5678 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5680 compose_output_action(ctx, OFPP_LOCAL);
5683 if (ctx->ofproto->has_mirrors) {
5684 add_mirror_actions(ctx, &ctx->orig_flow);
5686 fix_sflow_action(ctx);
5690 /* Translates the 'n_in' "union ofp_action"s in 'in' into datapath actions,
5691 * using 'ctx', and discards the datapath actions. */
5693 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5694 const union ofp_action *in, size_t n_in)
5696 uint64_t odp_actions_stub[1024 / 8];
5697 struct ofpbuf odp_actions;
5699 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5700 xlate_actions(ctx, in, n_in, &odp_actions);
5701 ofpbuf_uninit(&odp_actions);
5704 /* OFPP_NORMAL implementation. */
5706 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5708 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5709 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5710 * the bundle on which the packet was received, returns the VLAN to which the
5713 * Both 'vid' and the return value are in the range 0...4095. */
5715 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5717 switch (in_bundle->vlan_mode) {
5718 case PORT_VLAN_ACCESS:
5719 return in_bundle->vlan;
5722 case PORT_VLAN_TRUNK:
5725 case PORT_VLAN_NATIVE_UNTAGGED:
5726 case PORT_VLAN_NATIVE_TAGGED:
5727 return vid ? vid : in_bundle->vlan;
5734 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5735 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5738 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5739 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5742 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5744 /* Allow any VID on the OFPP_NONE port. */
5745 if (in_bundle == &ofpp_none_bundle) {
5749 switch (in_bundle->vlan_mode) {
5750 case PORT_VLAN_ACCESS:
5753 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5754 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5755 "packet received on port %s configured as VLAN "
5756 "%"PRIu16" access port",
5757 in_bundle->ofproto->up.name, vid,
5758 in_bundle->name, in_bundle->vlan);
5764 case PORT_VLAN_NATIVE_UNTAGGED:
5765 case PORT_VLAN_NATIVE_TAGGED:
5767 /* Port must always carry its native VLAN. */
5771 case PORT_VLAN_TRUNK:
5772 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5774 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5775 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5776 "received on port %s not configured for trunking "
5778 in_bundle->ofproto->up.name, vid,
5779 in_bundle->name, vid);
5791 /* Given 'vlan', the VLAN that a packet belongs to, and
5792 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5793 * that should be included in the 802.1Q header. (If the return value is 0,
5794 * then the 802.1Q header should only be included in the packet if there is a
5797 * Both 'vlan' and the return value are in the range 0...4095. */
5799 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5801 switch (out_bundle->vlan_mode) {
5802 case PORT_VLAN_ACCESS:
5805 case PORT_VLAN_TRUNK:
5806 case PORT_VLAN_NATIVE_TAGGED:
5809 case PORT_VLAN_NATIVE_UNTAGGED:
5810 return vlan == out_bundle->vlan ? 0 : vlan;
5818 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5821 struct ofport_dpif *port;
5823 ovs_be16 tci, old_tci;
5825 vid = output_vlan_to_vid(out_bundle, vlan);
5826 if (!out_bundle->bond) {
5827 port = ofbundle_get_a_port(out_bundle);
5829 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5832 /* No slaves enabled, so drop packet. */
5837 old_tci = ctx->flow.vlan_tci;
5839 if (tci || out_bundle->use_priority_tags) {
5840 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5842 tci |= htons(VLAN_CFI);
5845 ctx->flow.vlan_tci = tci;
5847 compose_output_action(ctx, port->up.ofp_port);
5848 ctx->flow.vlan_tci = old_tci;
5852 mirror_mask_ffs(mirror_mask_t mask)
5854 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5859 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5861 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5862 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5866 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5868 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5871 /* Returns an arbitrary interface within 'bundle'. */
5872 static struct ofport_dpif *
5873 ofbundle_get_a_port(const struct ofbundle *bundle)
5875 return CONTAINER_OF(list_front(&bundle->ports),
5876 struct ofport_dpif, bundle_node);
5880 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5882 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5886 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5888 struct ofproto_dpif *ofproto = ctx->ofproto;
5889 mirror_mask_t mirrors;
5890 struct ofbundle *in_bundle;
5893 const struct nlattr *a;
5896 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5897 ctx->packet != NULL, NULL);
5901 mirrors = in_bundle->src_mirrors;
5903 /* Drop frames on bundles reserved for mirroring. */
5904 if (in_bundle->mirror_out) {
5905 if (ctx->packet != NULL) {
5906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5907 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5908 "%s, which is reserved exclusively for mirroring",
5909 ctx->ofproto->up.name, in_bundle->name);
5915 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5916 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5919 vlan = input_vid_to_vlan(in_bundle, vid);
5921 /* Look at the output ports to check for destination selections. */
5923 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5924 ctx->odp_actions->size) {
5925 enum ovs_action_attr type = nl_attr_type(a);
5926 struct ofport_dpif *ofport;
5928 if (type != OVS_ACTION_ATTR_OUTPUT) {
5932 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
5933 if (ofport && ofport->bundle) {
5934 mirrors |= ofport->bundle->dst_mirrors;
5942 /* Restore the original packet before adding the mirror actions. */
5943 ctx->flow = *orig_flow;
5948 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5950 if (!vlan_is_mirrored(m, vlan)) {
5951 mirrors &= mirrors - 1;
5955 mirrors &= ~m->dup_mirrors;
5956 ctx->mirrors |= m->dup_mirrors;
5958 output_normal(ctx, m->out, vlan);
5959 } else if (vlan != m->out_vlan
5960 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
5961 struct ofbundle *bundle;
5963 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
5964 if (ofbundle_includes_vlan(bundle, m->out_vlan)
5965 && !bundle->mirror_out) {
5966 output_normal(ctx, bundle, m->out_vlan);
5974 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
5975 uint64_t packets, uint64_t bytes)
5981 for (; mirrors; mirrors &= mirrors - 1) {
5984 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
5987 /* In normal circumstances 'm' will not be NULL. However,
5988 * if mirrors are reconfigured, we can temporarily get out
5989 * of sync in facet_revalidate(). We could "correct" the
5990 * mirror list before reaching here, but doing that would
5991 * not properly account the traffic stats we've currently
5992 * accumulated for previous mirror configuration. */
5996 m->packet_count += packets;
5997 m->byte_count += bytes;
6001 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6002 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6003 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6005 is_gratuitous_arp(const struct flow *flow)
6007 return (flow->dl_type == htons(ETH_TYPE_ARP)
6008 && eth_addr_is_broadcast(flow->dl_dst)
6009 && (flow->nw_proto == ARP_OP_REPLY
6010 || (flow->nw_proto == ARP_OP_REQUEST
6011 && flow->nw_src == flow->nw_dst)));
6015 update_learning_table(struct ofproto_dpif *ofproto,
6016 const struct flow *flow, int vlan,
6017 struct ofbundle *in_bundle)
6019 struct mac_entry *mac;
6021 /* Don't learn the OFPP_NONE port. */
6022 if (in_bundle == &ofpp_none_bundle) {
6026 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6030 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6031 if (is_gratuitous_arp(flow)) {
6032 /* We don't want to learn from gratuitous ARP packets that are
6033 * reflected back over bond slaves so we lock the learning table. */
6034 if (!in_bundle->bond) {
6035 mac_entry_set_grat_arp_lock(mac);
6036 } else if (mac_entry_is_grat_arp_locked(mac)) {
6041 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6042 /* The log messages here could actually be useful in debugging,
6043 * so keep the rate limit relatively high. */
6044 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6045 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6046 "on port %s in VLAN %d",
6047 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6048 in_bundle->name, vlan);
6050 mac->port.p = in_bundle;
6051 tag_set_add(&ofproto->revalidate_set,
6052 mac_learning_changed(ofproto->ml, mac));
6056 static struct ofbundle *
6057 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6058 bool warn, struct ofport_dpif **in_ofportp)
6060 struct ofport_dpif *ofport;
6062 /* Find the port and bundle for the received packet. */
6063 ofport = get_ofp_port(ofproto, in_port);
6065 *in_ofportp = ofport;
6067 if (ofport && ofport->bundle) {
6068 return ofport->bundle;
6071 /* Special-case OFPP_NONE, which a controller may use as the ingress
6072 * port for traffic that it is sourcing. */
6073 if (in_port == OFPP_NONE) {
6074 return &ofpp_none_bundle;
6077 /* Odd. A few possible reasons here:
6079 * - We deleted a port but there are still a few packets queued up
6082 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6083 * we don't know about.
6085 * - The ofproto client didn't configure the port as part of a bundle.
6086 * This is particularly likely to happen if a packet was received on the
6087 * port after it was created, but before the client had a chance to
6088 * configure its bundle.
6091 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6093 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6094 "port %"PRIu16, ofproto->up.name, in_port);
6099 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6100 * dropped. Returns true if they may be forwarded, false if they should be
6103 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6104 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6106 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6107 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6108 * checked by input_vid_is_valid().
6110 * May also add tags to '*tags', although the current implementation only does
6111 * so in one special case.
6114 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
6115 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
6117 struct ofbundle *in_bundle = in_port->bundle;
6119 /* Drop frames for reserved multicast addresses
6120 * only if forward_bpdu option is absent. */
6121 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6125 if (in_bundle->bond) {
6126 struct mac_entry *mac;
6128 switch (bond_check_admissibility(in_bundle->bond, in_port,
6129 flow->dl_dst, tags)) {
6136 case BV_DROP_IF_MOVED:
6137 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6138 if (mac && mac->port.p != in_bundle &&
6139 (!is_gratuitous_arp(flow)
6140 || mac_entry_is_grat_arp_locked(mac))) {
6151 xlate_normal(struct action_xlate_ctx *ctx)
6153 struct ofport_dpif *in_port;
6154 struct ofbundle *in_bundle;
6155 struct mac_entry *mac;
6159 ctx->has_normal = true;
6161 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6162 ctx->packet != NULL, &in_port);
6167 /* Drop malformed frames. */
6168 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6169 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6170 if (ctx->packet != NULL) {
6171 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6172 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6173 "VLAN tag received on port %s",
6174 ctx->ofproto->up.name, in_bundle->name);
6179 /* Drop frames on bundles reserved for mirroring. */
6180 if (in_bundle->mirror_out) {
6181 if (ctx->packet != NULL) {
6182 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6183 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6184 "%s, which is reserved exclusively for mirroring",
6185 ctx->ofproto->up.name, in_bundle->name);
6191 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6192 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6195 vlan = input_vid_to_vlan(in_bundle, vid);
6197 /* Check other admissibility requirements. */
6199 !is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
6203 /* Learn source MAC. */
6204 if (ctx->may_learn) {
6205 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6208 /* Determine output bundle. */
6209 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6212 if (mac->port.p != in_bundle) {
6213 output_normal(ctx, mac->port.p, vlan);
6216 struct ofbundle *bundle;
6218 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6219 if (bundle != in_bundle
6220 && ofbundle_includes_vlan(bundle, vlan)
6221 && bundle->floodable
6222 && !bundle->mirror_out) {
6223 output_normal(ctx, bundle, vlan);
6226 ctx->nf_output_iface = NF_OUT_FLOOD;
6230 /* Optimized flow revalidation.
6232 * It's a difficult problem, in general, to tell which facets need to have
6233 * their actions recalculated whenever the OpenFlow flow table changes. We
6234 * don't try to solve that general problem: for most kinds of OpenFlow flow
6235 * table changes, we recalculate the actions for every facet. This is
6236 * relatively expensive, but it's good enough if the OpenFlow flow table
6237 * doesn't change very often.
6239 * However, we can expect one particular kind of OpenFlow flow table change to
6240 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6241 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6242 * table, we add a special case that applies to flow tables in which every rule
6243 * has the same form (that is, the same wildcards), except that the table is
6244 * also allowed to have a single "catch-all" flow that matches all packets. We
6245 * optimize this case by tagging all of the facets that resubmit into the table
6246 * and invalidating the same tag whenever a flow changes in that table. The
6247 * end result is that we revalidate just the facets that need it (and sometimes
6248 * a few more, but not all of the facets or even all of the facets that
6249 * resubmit to the table modified by MAC learning). */
6251 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6252 * into an OpenFlow table with the given 'basis'. */
6254 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6257 if (flow_wildcards_is_catchall(wc)) {
6260 struct flow tag_flow = *flow;
6261 flow_zero_wildcards(&tag_flow, wc);
6262 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6266 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6267 * taggability of that table.
6269 * This function must be called after *each* change to a flow table. If you
6270 * skip calling it on some changes then the pointer comparisons at the end can
6271 * be invalid if you get unlucky. For example, if a flow removal causes a
6272 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6273 * different wildcards to be created with the same address, then this function
6274 * will incorrectly skip revalidation. */
6276 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6278 struct table_dpif *table = &ofproto->tables[table_id];
6279 const struct oftable *oftable = &ofproto->up.tables[table_id];
6280 struct cls_table *catchall, *other;
6281 struct cls_table *t;
6283 catchall = other = NULL;
6285 switch (hmap_count(&oftable->cls.tables)) {
6287 /* We could tag this OpenFlow table but it would make the logic a
6288 * little harder and it's a corner case that doesn't seem worth it
6294 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6295 if (cls_table_is_catchall(t)) {
6297 } else if (!other) {
6300 /* Indicate that we can't tag this by setting both tables to
6301 * NULL. (We know that 'catchall' is already NULL.) */
6308 /* Can't tag this table. */
6312 if (table->catchall_table != catchall || table->other_table != other) {
6313 table->catchall_table = catchall;
6314 table->other_table = other;
6315 ofproto->need_revalidate = REV_FLOW_TABLE;
6319 /* Given 'rule' that has changed in some way (either it is a rule being
6320 * inserted, a rule being deleted, or a rule whose actions are being
6321 * modified), marks facets for revalidation to ensure that packets will be
6322 * forwarded correctly according to the new state of the flow table.
6324 * This function must be called after *each* change to a flow table. See
6325 * the comment on table_update_taggable() for more information. */
6327 rule_invalidate(const struct rule_dpif *rule)
6329 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6331 table_update_taggable(ofproto, rule->up.table_id);
6333 if (!ofproto->need_revalidate) {
6334 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6336 if (table->other_table && rule->tag) {
6337 tag_set_add(&ofproto->revalidate_set, rule->tag);
6339 ofproto->need_revalidate = REV_FLOW_TABLE;
6345 set_frag_handling(struct ofproto *ofproto_,
6346 enum ofp_config_flags frag_handling)
6348 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6350 if (frag_handling != OFPC_FRAG_REASM) {
6351 ofproto->need_revalidate = REV_RECONFIGURE;
6359 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6360 const struct flow *flow,
6361 const union ofp_action *ofp_actions, size_t n_ofp_actions)
6363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6366 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6367 return OFPERR_NXBRC_BAD_IN_PORT;
6370 error = validate_actions(ofp_actions, n_ofp_actions, flow,
6371 ofproto->max_ports);
6373 struct odputil_keybuf keybuf;
6374 struct dpif_flow_stats stats;
6378 struct action_xlate_ctx ctx;
6379 uint64_t odp_actions_stub[1024 / 8];
6380 struct ofpbuf odp_actions;
6382 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6383 odp_flow_key_from_flow(&key, flow);
6385 dpif_flow_stats_extract(flow, packet, &stats);
6387 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6388 packet_get_tcp_flags(packet, flow), packet);
6389 ctx.resubmit_stats = &stats;
6391 ofpbuf_use_stub(&odp_actions,
6392 odp_actions_stub, sizeof odp_actions_stub);
6393 xlate_actions(&ctx, ofp_actions, n_ofp_actions, &odp_actions);
6394 dpif_execute(ofproto->dpif, key.data, key.size,
6395 odp_actions.data, odp_actions.size, packet);
6396 ofpbuf_uninit(&odp_actions);
6404 set_netflow(struct ofproto *ofproto_,
6405 const struct netflow_options *netflow_options)
6407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6409 if (netflow_options) {
6410 if (!ofproto->netflow) {
6411 ofproto->netflow = netflow_create();
6413 return netflow_set_options(ofproto->netflow, netflow_options);
6415 netflow_destroy(ofproto->netflow);
6416 ofproto->netflow = NULL;
6422 get_netflow_ids(const struct ofproto *ofproto_,
6423 uint8_t *engine_type, uint8_t *engine_id)
6425 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6427 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6431 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6433 if (!facet_is_controller_flow(facet) &&
6434 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6435 struct subfacet *subfacet;
6436 struct ofexpired expired;
6438 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6439 if (subfacet->path == SF_FAST_PATH) {
6440 struct dpif_flow_stats stats;
6442 subfacet_reinstall(subfacet, &stats);
6443 subfacet_update_stats(subfacet, &stats);
6447 expired.flow = facet->flow;
6448 expired.packet_count = facet->packet_count;
6449 expired.byte_count = facet->byte_count;
6450 expired.used = facet->used;
6451 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6456 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6458 struct facet *facet;
6460 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6461 send_active_timeout(ofproto, facet);
6465 static struct ofproto_dpif *
6466 ofproto_dpif_lookup(const char *name)
6468 struct ofproto_dpif *ofproto;
6470 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6471 hash_string(name, 0), &all_ofproto_dpifs) {
6472 if (!strcmp(ofproto->up.name, name)) {
6480 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6481 const char *argv[], void *aux OVS_UNUSED)
6483 struct ofproto_dpif *ofproto;
6486 ofproto = ofproto_dpif_lookup(argv[1]);
6488 unixctl_command_reply_error(conn, "no such bridge");
6491 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6493 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6494 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6498 unixctl_command_reply(conn, "table successfully flushed");
6502 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6503 const char *argv[], void *aux OVS_UNUSED)
6505 struct ds ds = DS_EMPTY_INITIALIZER;
6506 const struct ofproto_dpif *ofproto;
6507 const struct mac_entry *e;
6509 ofproto = ofproto_dpif_lookup(argv[1]);
6511 unixctl_command_reply_error(conn, "no such bridge");
6515 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6516 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6517 struct ofbundle *bundle = e->port.p;
6518 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6519 ofbundle_get_a_port(bundle)->odp_port,
6520 e->vlan, ETH_ADDR_ARGS(e->mac),
6521 mac_entry_age(ofproto->ml, e));
6523 unixctl_command_reply(conn, ds_cstr(&ds));
6528 struct action_xlate_ctx ctx;
6534 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6535 const struct rule_dpif *rule)
6537 ds_put_char_multiple(result, '\t', level);
6539 ds_put_cstr(result, "No match\n");
6543 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6544 table_id, ntohll(rule->up.flow_cookie));
6545 cls_rule_format(&rule->up.cr, result);
6546 ds_put_char(result, '\n');
6548 ds_put_char_multiple(result, '\t', level);
6549 ds_put_cstr(result, "OpenFlow ");
6550 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
6551 ds_put_char(result, '\n');
6555 trace_format_flow(struct ds *result, int level, const char *title,
6556 struct trace_ctx *trace)
6558 ds_put_char_multiple(result, '\t', level);
6559 ds_put_format(result, "%s: ", title);
6560 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6561 ds_put_cstr(result, "unchanged");
6563 flow_format(result, &trace->ctx.flow);
6564 trace->flow = trace->ctx.flow;
6566 ds_put_char(result, '\n');
6570 trace_format_regs(struct ds *result, int level, const char *title,
6571 struct trace_ctx *trace)
6575 ds_put_char_multiple(result, '\t', level);
6576 ds_put_format(result, "%s:", title);
6577 for (i = 0; i < FLOW_N_REGS; i++) {
6578 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6580 ds_put_char(result, '\n');
6584 trace_format_odp(struct ds *result, int level, const char *title,
6585 struct trace_ctx *trace)
6587 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6589 ds_put_char_multiple(result, '\t', level);
6590 ds_put_format(result, "%s: ", title);
6591 format_odp_actions(result, odp_actions->data, odp_actions->size);
6592 ds_put_char(result, '\n');
6596 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6598 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6599 struct ds *result = trace->result;
6601 ds_put_char(result, '\n');
6602 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6603 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6604 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6605 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6609 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6610 void *aux OVS_UNUSED)
6612 const char *dpname = argv[1];
6613 struct ofproto_dpif *ofproto;
6614 struct ofpbuf odp_key;
6615 struct ofpbuf *packet;
6616 ovs_be16 initial_tci;
6622 ofpbuf_init(&odp_key, 0);
6625 ofproto = ofproto_dpif_lookup(dpname);
6627 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6631 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6632 /* ofproto/trace dpname flow [-generate] */
6633 const char *flow_s = argv[2];
6634 const char *generate_s = argv[3];
6636 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6637 * flow. We guess which type it is based on whether 'flow_s' contains
6638 * an '(', since a datapath flow always contains '(') but an
6639 * OpenFlow-like flow should not (in fact it's allowed but I believe
6640 * that's not documented anywhere).
6642 * An alternative would be to try to parse 'flow_s' both ways, but then
6643 * it would be tricky giving a sensible error message. After all, do
6644 * you just say "syntax error" or do you present both error messages?
6645 * Both choices seem lousy. */
6646 if (strchr(flow_s, '(')) {
6649 /* Convert string to datapath key. */
6650 ofpbuf_init(&odp_key, 0);
6651 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6653 unixctl_command_reply_error(conn, "Bad flow syntax");
6657 /* Convert odp_key to flow. */
6658 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6659 odp_key.size, &flow,
6660 &initial_tci, NULL);
6661 if (error == ODP_FIT_ERROR) {
6662 unixctl_command_reply_error(conn, "Invalid flow");
6668 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6670 unixctl_command_reply_error(conn, error_s);
6675 initial_tci = flow.vlan_tci;
6676 vsp_adjust_flow(ofproto, &flow);
6679 /* Generate a packet, if requested. */
6681 packet = ofpbuf_new(0);
6682 flow_compose(packet, &flow);
6684 } else if (argc == 6) {
6685 /* ofproto/trace dpname priority tun_id in_port packet */
6686 const char *priority_s = argv[2];
6687 const char *tun_id_s = argv[3];
6688 const char *in_port_s = argv[4];
6689 const char *packet_s = argv[5];
6690 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6691 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6692 uint32_t priority = atoi(priority_s);
6695 msg = eth_from_hex(packet_s, &packet);
6697 unixctl_command_reply_error(conn, msg);
6701 ds_put_cstr(&result, "Packet: ");
6702 s = ofp_packet_to_string(packet->data, packet->size);
6703 ds_put_cstr(&result, s);
6706 flow_extract(packet, priority, tun_id, in_port, &flow);
6707 initial_tci = flow.vlan_tci;
6709 unixctl_command_reply_error(conn, "Bad command syntax");
6713 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6714 unixctl_command_reply(conn, ds_cstr(&result));
6717 ds_destroy(&result);
6718 ofpbuf_delete(packet);
6719 ofpbuf_uninit(&odp_key);
6723 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6724 const struct ofpbuf *packet, ovs_be16 initial_tci,
6727 struct rule_dpif *rule;
6729 ds_put_cstr(ds, "Flow: ");
6730 flow_format(ds, flow);
6731 ds_put_char(ds, '\n');
6733 rule = rule_dpif_lookup(ofproto, flow);
6735 trace_format_rule(ds, 0, 0, rule);
6736 if (rule == ofproto->miss_rule) {
6737 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6738 } else if (rule == ofproto->no_packet_in_rule) {
6739 ds_put_cstr(ds, "\nNo match, packets dropped because "
6740 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6744 uint64_t odp_actions_stub[1024 / 8];
6745 struct ofpbuf odp_actions;
6747 struct trace_ctx trace;
6750 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6753 ofpbuf_use_stub(&odp_actions,
6754 odp_actions_stub, sizeof odp_actions_stub);
6755 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6756 rule, tcp_flags, packet);
6757 trace.ctx.resubmit_hook = trace_resubmit;
6758 xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions,
6761 ds_put_char(ds, '\n');
6762 trace_format_flow(ds, 0, "Final flow", &trace);
6763 ds_put_cstr(ds, "Datapath actions: ");
6764 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6765 ofpbuf_uninit(&odp_actions);
6767 if (trace.ctx.slow) {
6768 enum slow_path_reason slow;
6770 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6771 "slow path because it:");
6772 for (slow = trace.ctx.slow; slow; ) {
6773 enum slow_path_reason bit = rightmost_1bit(slow);
6777 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6780 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6783 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6786 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6789 ds_put_cstr(ds, "\n\t (The datapath actions are "
6790 "incomplete--for complete actions, "
6791 "please supply a packet.)");
6794 case SLOW_CONTROLLER:
6795 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6796 "to the OpenFlow controller.");
6799 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6800 "than the datapath supports.");
6807 if (slow & ~SLOW_MATCH) {
6808 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6809 "the special slow-path processing.");
6816 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6817 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6820 unixctl_command_reply(conn, NULL);
6824 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6825 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6828 unixctl_command_reply(conn, NULL);
6831 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6832 * 'reply' describing the results. */
6834 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6836 struct facet *facet;
6840 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6841 if (!facet_check_consistency(facet)) {
6846 ofproto->need_revalidate = REV_INCONSISTENCY;
6850 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6851 ofproto->up.name, errors);
6853 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6858 ofproto_dpif_self_check(struct unixctl_conn *conn,
6859 int argc, const char *argv[], void *aux OVS_UNUSED)
6861 struct ds reply = DS_EMPTY_INITIALIZER;
6862 struct ofproto_dpif *ofproto;
6865 ofproto = ofproto_dpif_lookup(argv[1]);
6867 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6868 "ofproto/list for help)");
6871 ofproto_dpif_self_check__(ofproto, &reply);
6873 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6874 ofproto_dpif_self_check__(ofproto, &reply);
6878 unixctl_command_reply(conn, ds_cstr(&reply));
6883 ofproto_dpif_unixctl_init(void)
6885 static bool registered;
6891 unixctl_command_register(
6893 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6894 2, 5, ofproto_unixctl_trace, NULL);
6895 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6896 ofproto_unixctl_fdb_flush, NULL);
6897 unixctl_command_register("fdb/show", "bridge", 1, 1,
6898 ofproto_unixctl_fdb_show, NULL);
6899 unixctl_command_register("ofproto/clog", "", 0, 0,
6900 ofproto_dpif_clog, NULL);
6901 unixctl_command_register("ofproto/unclog", "", 0, 0,
6902 ofproto_dpif_unclog, NULL);
6903 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6904 ofproto_dpif_self_check, NULL);
6907 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6909 * This is deprecated. It is only for compatibility with broken device drivers
6910 * in old versions of Linux that do not properly support VLANs when VLAN
6911 * devices are not used. When broken device drivers are no longer in
6912 * widespread use, we will delete these interfaces. */
6915 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
6917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6918 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6920 if (realdev_ofp_port == ofport->realdev_ofp_port
6921 && vid == ofport->vlandev_vid) {
6925 ofproto->need_revalidate = REV_RECONFIGURE;
6927 if (ofport->realdev_ofp_port) {
6930 if (realdev_ofp_port && ofport->bundle) {
6931 /* vlandevs are enslaved to their realdevs, so they are not allowed to
6932 * themselves be part of a bundle. */
6933 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6936 ofport->realdev_ofp_port = realdev_ofp_port;
6937 ofport->vlandev_vid = vid;
6939 if (realdev_ofp_port) {
6940 vsp_add(ofport, realdev_ofp_port, vid);
6947 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
6949 return hash_2words(realdev_ofp_port, vid);
6952 /* Returns the ODP port number of the Linux VLAN device that corresponds to
6953 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
6954 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
6955 * it would return the port number of eth0.9.
6957 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
6958 * function just returns its 'realdev_odp_port' argument. */
6960 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6961 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
6963 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6964 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
6965 int vid = vlan_tci_to_vid(vlan_tci);
6966 const struct vlan_splinter *vsp;
6968 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6969 hash_realdev_vid(realdev_ofp_port, vid),
6970 &ofproto->realdev_vid_map) {
6971 if (vsp->realdev_ofp_port == realdev_ofp_port
6972 && vsp->vid == vid) {
6973 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
6977 return realdev_odp_port;
6980 static struct vlan_splinter *
6981 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
6983 struct vlan_splinter *vsp;
6985 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
6986 &ofproto->vlandev_map) {
6987 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6995 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6996 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6997 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
6998 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6999 * eth0 and store 9 in '*vid'.
7001 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7002 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7005 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7006 uint16_t vlandev_ofp_port, int *vid)
7008 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7009 const struct vlan_splinter *vsp;
7011 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7016 return vsp->realdev_ofp_port;
7022 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7023 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7024 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7025 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7026 * always the case unless VLAN splinters are enabled), returns false without
7027 * making any changes. */
7029 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7034 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7039 /* Cause the flow to be processed as if it came in on the real device with
7040 * the VLAN device's VLAN ID. */
7041 flow->in_port = realdev;
7042 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7047 vsp_remove(struct ofport_dpif *port)
7049 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7050 struct vlan_splinter *vsp;
7052 vsp = vlandev_find(ofproto, port->up.ofp_port);
7054 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7055 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7058 port->realdev_ofp_port = 0;
7060 VLOG_ERR("missing vlan device record");
7065 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7069 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7070 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7071 == realdev_ofp_port)) {
7072 struct vlan_splinter *vsp;
7074 vsp = xmalloc(sizeof *vsp);
7075 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7076 hash_int(port->up.ofp_port, 0));
7077 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7078 hash_realdev_vid(realdev_ofp_port, vid));
7079 vsp->realdev_ofp_port = realdev_ofp_port;
7080 vsp->vlandev_ofp_port = port->up.ofp_port;
7083 port->realdev_ofp_port = realdev_ofp_port;
7085 VLOG_ERR("duplicate vlan device record");
7089 const struct ofproto_class ofproto_dpif_class = {
7119 port_is_lacp_current,
7120 NULL, /* rule_choose_table */
7127 rule_modify_actions,
7135 get_cfm_remote_mpids,
7140 get_stp_port_status,
7147 is_mirror_output_bundle,
7148 forward_bpdu_changed,