2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count; /* Number of packets received. */
98 uint64_t byte_count; /* Number of bytes received. */
100 tag_type tag; /* Caches rule_calculate_tag() result. */
102 struct list facets; /* List of "struct facet"s. */
105 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
107 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
110 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
111 const struct flow *);
112 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
115 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
116 const struct flow *flow);
118 static void rule_credit_stats(struct rule_dpif *,
119 const struct dpif_flow_stats *);
120 static void flow_push_stats(struct rule_dpif *, const struct flow *,
121 const struct dpif_flow_stats *);
122 static tag_type rule_calculate_tag(const struct flow *,
123 const struct minimask *, uint32_t basis);
124 static void rule_invalidate(const struct rule_dpif *);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
131 struct ofproto_dpif *ofproto; /* Owning ofproto. */
132 size_t idx; /* In ofproto's "mirrors" array. */
133 void *aux; /* Key supplied by ofproto's client. */
134 char *name; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle *out; /* Output port or NULL. */
143 int out_vlan; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count; /* Number of packets sent. */
148 int64_t byte_count; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror *);
152 static void update_mirror_stats(struct ofproto_dpif *ofproto,
153 mirror_mask_t mirrors,
154 uint64_t packets, uint64_t bytes);
157 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif *ofproto; /* Owning ofproto. */
159 void *aux; /* Key supplied by ofproto's client. */
160 char *name; /* Identifier for log messages. */
163 struct list ports; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode; /* VLAN mode */
165 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond *bond; /* Nonnull iff more than one port. */
170 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport *);
182 static void bundle_update(struct ofbundle *);
183 static void bundle_destroy(struct ofbundle *);
184 static void bundle_del_port(struct ofport_dpif *);
185 static void bundle_run(struct ofbundle *);
186 static void bundle_wait(struct ofbundle *);
187 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
188 uint16_t in_port, bool warn,
189 struct ofport_dpif **in_ofportp);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle = {
198 .vlan_mode = PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif *ofproto);
202 static void stp_wait(struct ofproto_dpif *ofproto);
203 static int set_stp_port(struct ofport *,
204 const struct ofproto_port_stp_settings *);
206 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
208 struct action_xlate_ctx {
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif *ofproto;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf *packet;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif *rule;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook)(struct action_xlate_ctx *, const char *s);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats *resubmit_stats;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf *odp_actions; /* Datapath actions. */
265 tag_type tags; /* Tags associated with actions. */
266 enum slow_path_reason slow; /* 0 if fast path may be used. */
267 bool has_learn; /* Actions include NXAST_LEARN? */
268 bool has_normal; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
278 struct flow base_flow; /* Flow at the last commit. */
279 uint32_t orig_skb_priority; /* Priority when packet arrived. */
280 uint8_t table_id; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs; /* Number of output ports. */
282 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
284 bool exit; /* No further actions should be processed. */
285 struct flow orig_flow; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx *,
289 struct ofproto_dpif *, const struct flow *,
290 ovs_be16 initial_tci, struct rule_dpif *,
291 uint8_t tcp_flags, const struct ofpbuf *);
292 static void xlate_actions(struct action_xlate_ctx *,
293 const struct ofpact *ofpacts, size_t ofpacts_len,
294 struct ofpbuf *odp_actions);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
296 const struct ofpact *ofpacts,
299 static size_t put_userspace_action(const struct ofproto_dpif *,
300 struct ofpbuf *odp_actions,
302 const union user_action_cookie *);
304 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
305 enum slow_path_reason,
306 uint64_t *stub, size_t stub_size,
307 const struct nlattr **actionsp,
308 size_t *actions_lenp);
310 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH, /* Full actions are installed. */
328 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node; /* In struct facet's 'facets' list. */
340 struct facet *facet; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness;
351 long long int used; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len; /* Number of bytes in actions[]. */
361 struct nlattr *actions; /* Datapath actions. */
363 enum slow_path_reason slow; /* 0 if fast path may be used. */
364 enum subfacet_path path; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
383 static struct subfacet *subfacet_find(struct ofproto_dpif *,
384 const struct nlattr *key, size_t key_len,
386 const struct flow *flow);
387 static void subfacet_destroy(struct subfacet *);
388 static void subfacet_destroy__(struct subfacet *);
389 static void subfacet_destroy_batch(struct ofproto_dpif *,
390 struct subfacet **, int n);
391 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
393 static void subfacet_reset_dp_stats(struct subfacet *,
394 struct dpif_flow_stats *);
395 static void subfacet_update_time(struct subfacet *, long long int used);
396 static void subfacet_update_stats(struct subfacet *,
397 const struct dpif_flow_stats *);
398 static void subfacet_make_actions(struct subfacet *,
399 const struct ofpbuf *packet,
400 struct ofpbuf *odp_actions);
401 static int subfacet_install(struct subfacet *,
402 const struct nlattr *actions, size_t actions_len,
403 struct dpif_flow_stats *, enum slow_path_reason);
404 static void subfacet_uninstall(struct subfacet *);
406 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node; /* In owning rule's 'facets' list. */
426 struct rule_dpif *rule; /* Owning rule. */
429 struct list subfacets;
430 long long int used; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count; /* Number of packets received. */
448 uint64_t byte_count; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
453 long long int prev_used; /* Used time from last stats push. */
456 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn; /* Actions include NXAST_LEARN? */
467 bool has_normal; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet;
478 static struct facet *facet_create(struct rule_dpif *,
479 const struct flow *, uint32_t hash);
480 static void facet_remove(struct facet *);
481 static void facet_free(struct facet *);
483 static struct facet *facet_find(struct ofproto_dpif *,
484 const struct flow *, uint32_t hash);
485 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
486 const struct flow *, uint32_t hash);
487 static void facet_revalidate(struct facet *);
488 static bool facet_check_consistency(struct facet *);
490 static void facet_flush_stats(struct facet *);
492 static void facet_update_time(struct facet *, long long int used);
493 static void facet_reset_counters(struct facet *);
494 static void facet_push_stats(struct facet *);
495 static void facet_learn(struct facet *);
496 static void facet_account(struct facet *);
498 static bool facet_is_controller_flow(struct facet *);
501 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
506 struct list bundle_node; /* In struct ofbundle's "ports" list. */
507 struct cfm *cfm; /* Connectivity Fault Management, if any. */
508 tag_type tag; /* Tag associated with this port. */
509 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
510 bool may_enable; /* May be enabled in bonds. */
511 long long int carrier_seq; /* Carrier status changes. */
514 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered;
518 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp {
534 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority; /* Priority of this queue (see struct flow). */
537 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter {
547 struct hmap_node realdev_vid_node;
548 struct hmap_node vlandev_node;
549 uint16_t realdev_ofp_port;
550 uint16_t vlandev_ofp_port;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
555 uint32_t realdev, ovs_be16 vlan_tci);
556 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
557 static void vsp_remove(struct ofport_dpif *);
558 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
565 static struct ofport_dpif *
566 ofport_dpif_cast(const struct ofport *ofport)
568 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
569 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
572 static void port_run(struct ofport_dpif *);
573 static void port_run_fast(struct ofport_dpif *);
574 static void port_wait(struct ofport_dpif *);
575 static int set_cfm(struct ofport *, const struct cfm_settings *);
576 static void ofport_clear_priorities(struct ofport_dpif *);
578 struct dpif_completion {
579 struct list list_node;
580 struct ofoperation *op;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table *catchall_table; /* Table that wildcards all fields. */
590 struct cls_table *other_table; /* Table with any other wildcard set. */
591 uint32_t basis; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason {
603 REV_RECONFIGURE = 1, /* Switch configuration changed. */
604 REV_STP, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE, /* Flow table changed. */
607 REV_INCONSISTENCY /* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure);
610 COVERAGE_DEFINE(rev_stp);
611 COVERAGE_DEFINE(rev_port_toggled);
612 COVERAGE_DEFINE(rev_flow_table);
613 COVERAGE_DEFINE(rev_inconsistency);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration;
621 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
624 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
625 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
627 static struct ofport_dpif *
628 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
630 struct ofproto_dpif {
631 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
633 struct dpif_backer *backer;
635 /* Special OpenFlow rules. */
636 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
637 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
643 struct netflow *netflow;
644 struct dpif_sflow *sflow;
645 struct hmap bundles; /* Contains "struct ofbundle"s. */
646 struct mac_learning *ml;
647 struct ofmirror *mirrors[MAX_MIRRORS];
649 bool has_bonded_bundles;
653 struct hmap subfacets;
654 struct governor *governor;
657 struct table_dpif tables[N_TABLES];
658 enum revalidate_reason need_revalidate;
659 struct tag_set revalidate_set;
661 /* Support for debugging async flow mods. */
662 struct list completions;
664 bool has_bundle_action; /* True when the first bundle action appears. */
665 struct netdev_stats stats; /* To account packets generated and consumed in
670 long long int stp_last_tick;
672 /* VLAN splinters. */
673 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
674 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
677 struct sset ports; /* Set of port names. */
678 struct sset port_poll_set; /* Queued names for port_poll() reply. */
679 int port_poll_errno; /* Last errno for port_poll() reply. */
682 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
683 * for debugging the asynchronous flow_mod implementation.) */
686 /* All existing ofproto_dpif instances, indexed by ->up.name. */
687 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
689 static void ofproto_dpif_unixctl_init(void);
691 static struct ofproto_dpif *
692 ofproto_dpif_cast(const struct ofproto *ofproto)
694 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
695 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
698 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
700 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
702 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
703 const struct ofpbuf *, ovs_be16 initial_tci,
706 /* Packet processing. */
707 static void update_learning_table(struct ofproto_dpif *,
708 const struct flow *, int vlan,
711 #define FLOW_MISS_MAX_BATCH 50
712 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
714 /* Flow expiration. */
715 static int expire(struct dpif_backer *);
718 static void send_netflow_active_timeouts(struct ofproto_dpif *);
721 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
722 static size_t compose_sflow_action(const struct ofproto_dpif *,
723 struct ofpbuf *odp_actions,
724 const struct flow *, uint32_t odp_port);
725 static void add_mirror_actions(struct action_xlate_ctx *ctx,
726 const struct flow *flow);
727 /* Global variables. */
728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
730 /* Initial mappings of port to bridge mappings. */
731 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
733 /* Factory functions. */
736 init(const struct shash *iface_hints)
738 struct shash_node *node;
740 /* Make a local copy, since we don't own 'iface_hints' elements. */
741 SHASH_FOR_EACH(node, iface_hints) {
742 const struct iface_hint *orig_hint = node->data;
743 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
745 new_hint->br_name = xstrdup(orig_hint->br_name);
746 new_hint->br_type = xstrdup(orig_hint->br_type);
747 new_hint->ofp_port = orig_hint->ofp_port;
749 shash_add(&init_ofp_ports, node->name, new_hint);
754 enumerate_types(struct sset *types)
756 dp_enumerate_types(types);
760 enumerate_names(const char *type, struct sset *names)
762 struct ofproto_dpif *ofproto;
765 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
766 if (strcmp(type, ofproto->up.type)) {
769 sset_add(names, ofproto->up.name);
776 del(const char *type, const char *name)
781 error = dpif_open(name, type, &dpif);
783 error = dpif_delete(dpif);
790 port_open_type(const char *datapath_type, const char *port_type)
792 return dpif_port_open_type(datapath_type, port_type);
795 /* Type functions. */
797 static struct ofproto_dpif *
798 lookup_ofproto_dpif_by_port_name(const char *name)
800 struct ofproto_dpif *ofproto;
802 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
803 if (sset_contains(&ofproto->ports, name)) {
812 type_run(const char *type)
814 struct dpif_backer *backer;
818 backer = shash_find_data(&all_dpif_backers, type);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 dpif_run(backer->dpif);
827 if (timer_expired(&backer->next_expiration)) {
828 int delay = expire(backer);
829 timer_set_duration(&backer->next_expiration, delay);
832 /* Check for port changes in the dpif. */
833 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
834 struct ofproto_dpif *ofproto;
835 struct dpif_port port;
837 /* Don't report on the datapath's device. */
838 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
842 ofproto = lookup_ofproto_dpif_by_port_name(devname);
843 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
844 /* The port was removed. If we know the datapath,
845 * report it through poll_set(). If we don't, it may be
846 * notifying us of a removal we initiated, so ignore it.
847 * If there's a pending ENOBUFS, let it stand, since
848 * everything will be reevaluated. */
849 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
850 sset_add(&ofproto->port_poll_set, devname);
851 ofproto->port_poll_errno = 0;
853 } else if (!ofproto) {
854 /* The port was added, but we don't know with which
855 * ofproto we should associate it. Delete it. */
856 dpif_port_del(backer->dpif, port.port_no);
858 dpif_port_destroy(&port);
864 if (error != EAGAIN) {
865 struct ofproto_dpif *ofproto;
867 /* There was some sort of error, so propagate it to all
868 * ofprotos that use this backer. */
869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
870 &all_ofproto_dpifs) {
871 if (ofproto->backer == backer) {
872 sset_clear(&ofproto->port_poll_set);
873 ofproto->port_poll_errno = error;
882 type_run_fast(const char *type)
884 struct dpif_backer *backer;
887 backer = shash_find_data(&all_dpif_backers, type);
889 /* This is not necessarily a problem, since backers are only
890 * created on demand. */
894 /* Handle one or more batches of upcalls, until there's nothing left to do
895 * or until we do a fixed total amount of work.
897 * We do work in batches because it can be much cheaper to set up a number
898 * of flows and fire off their patches all at once. We do multiple batches
899 * because in some cases handling a packet can cause another packet to be
900 * queued almost immediately as part of the return flow. Both
901 * optimizations can make major improvements on some benchmarks and
902 * presumably for real traffic as well. */
904 while (work < FLOW_MISS_MAX_BATCH) {
905 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
916 type_wait(const char *type)
918 struct dpif_backer *backer;
920 backer = shash_find_data(&all_dpif_backers, type);
922 /* This is not necessarily a problem, since backers are only
923 * created on demand. */
927 timer_wait(&backer->next_expiration);
930 /* Basic life-cycle. */
932 static int add_internal_flows(struct ofproto_dpif *);
934 static struct ofproto *
937 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
942 dealloc(struct ofproto *ofproto_)
944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
949 close_dpif_backer(struct dpif_backer *backer)
951 struct shash_node *node;
953 ovs_assert(backer->refcount > 0);
955 if (--backer->refcount) {
959 hmap_destroy(&backer->odp_to_ofport_map);
960 node = shash_find(&all_dpif_backers, backer->type);
962 shash_delete(&all_dpif_backers, node);
963 dpif_close(backer->dpif);
968 /* Datapath port slated for removal from datapath. */
970 struct list list_node;
975 open_dpif_backer(const char *type, struct dpif_backer **backerp)
977 struct dpif_backer *backer;
978 struct dpif_port_dump port_dump;
979 struct dpif_port port;
980 struct shash_node *node;
981 struct list garbage_list;
982 struct odp_garbage *garbage, *next;
988 backer = shash_find_data(&all_dpif_backers, type);
995 backer_name = xasprintf("ovs-%s", type);
997 /* Remove any existing datapaths, since we assume we're the only
998 * userspace controlling the datapath. */
1000 dp_enumerate_names(type, &names);
1001 SSET_FOR_EACH(name, &names) {
1002 struct dpif *old_dpif;
1004 /* Don't remove our backer if it exists. */
1005 if (!strcmp(name, backer_name)) {
1009 if (dpif_open(name, type, &old_dpif)) {
1010 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1012 dpif_delete(old_dpif);
1013 dpif_close(old_dpif);
1016 sset_destroy(&names);
1018 backer = xmalloc(sizeof *backer);
1020 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1023 VLOG_ERR("failed to open datapath of type %s: %s", type,
1029 backer->type = xstrdup(type);
1030 backer->refcount = 1;
1031 hmap_init(&backer->odp_to_ofport_map);
1032 timer_set_duration(&backer->next_expiration, 1000);
1035 dpif_flow_flush(backer->dpif);
1037 /* Loop through the ports already on the datapath and remove any
1038 * that we don't need anymore. */
1039 list_init(&garbage_list);
1040 dpif_port_dump_start(&port_dump, backer->dpif);
1041 while (dpif_port_dump_next(&port_dump, &port)) {
1042 node = shash_find(&init_ofp_ports, port.name);
1043 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1044 garbage = xmalloc(sizeof *garbage);
1045 garbage->odp_port = port.port_no;
1046 list_push_front(&garbage_list, &garbage->list_node);
1049 dpif_port_dump_done(&port_dump);
1051 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1052 dpif_port_del(backer->dpif, garbage->odp_port);
1053 list_remove(&garbage->list_node);
1057 shash_add(&all_dpif_backers, type, backer);
1059 error = dpif_recv_set(backer->dpif, true);
1061 VLOG_ERR("failed to listen on datapath of type %s: %s",
1062 type, strerror(error));
1063 close_dpif_backer(backer);
1071 construct(struct ofproto *ofproto_)
1073 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1074 struct shash_node *node, *next;
1079 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1084 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1085 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1087 ofproto->n_matches = 0;
1089 ofproto->netflow = NULL;
1090 ofproto->sflow = NULL;
1091 ofproto->stp = NULL;
1092 hmap_init(&ofproto->bundles);
1093 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1094 for (i = 0; i < MAX_MIRRORS; i++) {
1095 ofproto->mirrors[i] = NULL;
1097 ofproto->has_bonded_bundles = false;
1099 hmap_init(&ofproto->facets);
1100 hmap_init(&ofproto->subfacets);
1101 ofproto->governor = NULL;
1103 for (i = 0; i < N_TABLES; i++) {
1104 struct table_dpif *table = &ofproto->tables[i];
1106 table->catchall_table = NULL;
1107 table->other_table = NULL;
1108 table->basis = random_uint32();
1110 ofproto->need_revalidate = 0;
1111 tag_set_init(&ofproto->revalidate_set);
1113 list_init(&ofproto->completions);
1115 ofproto_dpif_unixctl_init();
1117 ofproto->has_mirrors = false;
1118 ofproto->has_bundle_action = false;
1120 hmap_init(&ofproto->vlandev_map);
1121 hmap_init(&ofproto->realdev_vid_map);
1123 sset_init(&ofproto->ports);
1124 sset_init(&ofproto->port_poll_set);
1125 ofproto->port_poll_errno = 0;
1127 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1128 struct iface_hint *iface_hint = node->data;
1130 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1131 /* Check if the datapath already has this port. */
1132 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1133 sset_add(&ofproto->ports, node->name);
1136 free(iface_hint->br_name);
1137 free(iface_hint->br_type);
1139 shash_delete(&init_ofp_ports, node);
1143 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1144 hash_string(ofproto->up.name, 0));
1145 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1147 ofproto_init_tables(ofproto_, N_TABLES);
1148 error = add_internal_flows(ofproto);
1149 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1155 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1156 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1158 struct ofputil_flow_mod fm;
1161 match_init_catchall(&fm.match);
1163 match_set_reg(&fm.match, 0, id);
1164 fm.new_cookie = htonll(0);
1165 fm.cookie = htonll(0);
1166 fm.cookie_mask = htonll(0);
1167 fm.table_id = TBL_INTERNAL;
1168 fm.command = OFPFC_ADD;
1169 fm.idle_timeout = 0;
1170 fm.hard_timeout = 0;
1174 fm.ofpacts = ofpacts->data;
1175 fm.ofpacts_len = ofpacts->size;
1177 error = ofproto_flow_mod(&ofproto->up, &fm);
1179 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1180 id, ofperr_to_string(error));
1184 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1185 ovs_assert(*rulep != NULL);
1191 add_internal_flows(struct ofproto_dpif *ofproto)
1193 struct ofpact_controller *controller;
1194 uint64_t ofpacts_stub[128 / 8];
1195 struct ofpbuf ofpacts;
1199 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1202 controller = ofpact_put_CONTROLLER(&ofpacts);
1203 controller->max_len = UINT16_MAX;
1204 controller->controller_id = 0;
1205 controller->reason = OFPR_NO_MATCH;
1206 ofpact_pad(&ofpacts);
1208 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1213 ofpbuf_clear(&ofpacts);
1214 error = add_internal_flow(ofproto, id++, &ofpacts,
1215 &ofproto->no_packet_in_rule);
1220 complete_operations(struct ofproto_dpif *ofproto)
1222 struct dpif_completion *c, *next;
1224 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1225 ofoperation_complete(c->op, 0);
1226 list_remove(&c->list_node);
1232 destruct(struct ofproto *ofproto_)
1234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1235 struct rule_dpif *rule, *next_rule;
1236 struct oftable *table;
1239 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1240 complete_operations(ofproto);
1242 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1243 struct cls_cursor cursor;
1245 cls_cursor_init(&cursor, &table->cls, NULL);
1246 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1247 ofproto_rule_destroy(&rule->up);
1251 for (i = 0; i < MAX_MIRRORS; i++) {
1252 mirror_destroy(ofproto->mirrors[i]);
1255 netflow_destroy(ofproto->netflow);
1256 dpif_sflow_destroy(ofproto->sflow);
1257 hmap_destroy(&ofproto->bundles);
1258 mac_learning_destroy(ofproto->ml);
1260 hmap_destroy(&ofproto->facets);
1261 hmap_destroy(&ofproto->subfacets);
1262 governor_destroy(ofproto->governor);
1264 hmap_destroy(&ofproto->vlandev_map);
1265 hmap_destroy(&ofproto->realdev_vid_map);
1267 sset_destroy(&ofproto->ports);
1268 sset_destroy(&ofproto->port_poll_set);
1270 close_dpif_backer(ofproto->backer);
1274 run_fast(struct ofproto *ofproto_)
1276 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1277 struct ofport_dpif *ofport;
1279 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1280 port_run_fast(ofport);
1287 run(struct ofproto *ofproto_)
1289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1290 struct ofport_dpif *ofport;
1291 struct ofbundle *bundle;
1295 complete_operations(ofproto);
1298 error = run_fast(ofproto_);
1303 if (ofproto->netflow) {
1304 if (netflow_run(ofproto->netflow)) {
1305 send_netflow_active_timeouts(ofproto);
1308 if (ofproto->sflow) {
1309 dpif_sflow_run(ofproto->sflow);
1312 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1315 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1320 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1322 /* Now revalidate if there's anything to do. */
1323 if (ofproto->need_revalidate
1324 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1325 struct tag_set revalidate_set = ofproto->revalidate_set;
1326 bool revalidate_all = ofproto->need_revalidate;
1327 struct facet *facet;
1329 switch (ofproto->need_revalidate) {
1330 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1331 case REV_STP: COVERAGE_INC(rev_stp); break;
1332 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1333 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1334 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1337 /* Clear the revalidation flags. */
1338 tag_set_init(&ofproto->revalidate_set);
1339 ofproto->need_revalidate = 0;
1341 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1343 || tag_set_intersects(&revalidate_set, facet->tags)) {
1344 facet_revalidate(facet);
1349 /* Check the consistency of a random facet, to aid debugging. */
1350 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1351 struct facet *facet;
1353 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1354 struct facet, hmap_node);
1355 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1356 if (!facet_check_consistency(facet)) {
1357 ofproto->need_revalidate = REV_INCONSISTENCY;
1362 if (ofproto->governor) {
1365 governor_run(ofproto->governor);
1367 /* If the governor has shrunk to its minimum size and the number of
1368 * subfacets has dwindled, then drop the governor entirely.
1370 * For hysteresis, the number of subfacets to drop the governor is
1371 * smaller than the number needed to trigger its creation. */
1372 n_subfacets = hmap_count(&ofproto->subfacets);
1373 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1374 && governor_is_idle(ofproto->governor)) {
1375 governor_destroy(ofproto->governor);
1376 ofproto->governor = NULL;
1384 wait(struct ofproto *ofproto_)
1386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1387 struct ofport_dpif *ofport;
1388 struct ofbundle *bundle;
1390 if (!clogged && !list_is_empty(&ofproto->completions)) {
1391 poll_immediate_wake();
1394 dpif_wait(ofproto->backer->dpif);
1395 dpif_recv_wait(ofproto->backer->dpif);
1396 if (ofproto->sflow) {
1397 dpif_sflow_wait(ofproto->sflow);
1399 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1400 poll_immediate_wake();
1402 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1405 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1406 bundle_wait(bundle);
1408 if (ofproto->netflow) {
1409 netflow_wait(ofproto->netflow);
1411 mac_learning_wait(ofproto->ml);
1413 if (ofproto->need_revalidate) {
1414 /* Shouldn't happen, but if it does just go around again. */
1415 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1416 poll_immediate_wake();
1418 if (ofproto->governor) {
1419 governor_wait(ofproto->governor);
1424 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1426 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1428 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1429 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1433 flush(struct ofproto *ofproto_)
1435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1436 struct subfacet *subfacet, *next_subfacet;
1437 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1441 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1442 &ofproto->subfacets) {
1443 if (subfacet->path != SF_NOT_INSTALLED) {
1444 batch[n_batch++] = subfacet;
1445 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1446 subfacet_destroy_batch(ofproto, batch, n_batch);
1450 subfacet_destroy(subfacet);
1455 subfacet_destroy_batch(ofproto, batch, n_batch);
1460 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1461 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1463 *arp_match_ip = true;
1464 *actions = (OFPUTIL_A_OUTPUT |
1465 OFPUTIL_A_SET_VLAN_VID |
1466 OFPUTIL_A_SET_VLAN_PCP |
1467 OFPUTIL_A_STRIP_VLAN |
1468 OFPUTIL_A_SET_DL_SRC |
1469 OFPUTIL_A_SET_DL_DST |
1470 OFPUTIL_A_SET_NW_SRC |
1471 OFPUTIL_A_SET_NW_DST |
1472 OFPUTIL_A_SET_NW_TOS |
1473 OFPUTIL_A_SET_TP_SRC |
1474 OFPUTIL_A_SET_TP_DST |
1479 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1482 struct dpif_dp_stats s;
1484 strcpy(ots->name, "classifier");
1486 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1488 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1489 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1492 static struct ofport *
1495 struct ofport_dpif *port = xmalloc(sizeof *port);
1500 port_dealloc(struct ofport *port_)
1502 struct ofport_dpif *port = ofport_dpif_cast(port_);
1507 port_construct(struct ofport *port_)
1509 struct ofport_dpif *port = ofport_dpif_cast(port_);
1510 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1511 struct dpif_port dpif_port;
1514 ofproto->need_revalidate = REV_RECONFIGURE;
1515 port->bundle = NULL;
1517 port->tag = tag_create_random();
1518 port->may_enable = true;
1519 port->stp_port = NULL;
1520 port->stp_state = STP_DISABLED;
1521 hmap_init(&port->priorities);
1522 port->realdev_ofp_port = 0;
1523 port->vlandev_vid = 0;
1524 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1526 error = dpif_port_query_by_name(ofproto->backer->dpif,
1527 netdev_get_name(port->up.netdev),
1533 port->odp_port = dpif_port.port_no;
1535 /* Sanity-check that a mapping doesn't already exist. This
1536 * shouldn't happen. */
1537 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1538 VLOG_ERR("port %s already has an OpenFlow port number\n",
1543 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1544 hash_int(port->odp_port, 0));
1546 if (ofproto->sflow) {
1547 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1554 port_destruct(struct ofport *port_)
1556 struct ofport_dpif *port = ofport_dpif_cast(port_);
1557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1558 const char *devname = netdev_get_name(port->up.netdev);
1560 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1561 /* The underlying device is still there, so delete it. This
1562 * happens when the ofproto is being destroyed, since the caller
1563 * assumes that removal of attached ports will happen as part of
1565 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1568 sset_find_and_delete(&ofproto->ports, devname);
1569 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1570 ofproto->need_revalidate = REV_RECONFIGURE;
1571 bundle_remove(port_);
1572 set_cfm(port_, NULL);
1573 if (ofproto->sflow) {
1574 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1577 ofport_clear_priorities(port);
1578 hmap_destroy(&port->priorities);
1582 port_modified(struct ofport *port_)
1584 struct ofport_dpif *port = ofport_dpif_cast(port_);
1586 if (port->bundle && port->bundle->bond) {
1587 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1592 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1594 struct ofport_dpif *port = ofport_dpif_cast(port_);
1595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1596 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1598 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1599 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1600 OFPUTIL_PC_NO_PACKET_IN)) {
1601 ofproto->need_revalidate = REV_RECONFIGURE;
1603 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1604 bundle_update(port->bundle);
1610 set_sflow(struct ofproto *ofproto_,
1611 const struct ofproto_sflow_options *sflow_options)
1613 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1614 struct dpif_sflow *ds = ofproto->sflow;
1616 if (sflow_options) {
1618 struct ofport_dpif *ofport;
1620 ds = ofproto->sflow = dpif_sflow_create();
1621 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1622 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1624 ofproto->need_revalidate = REV_RECONFIGURE;
1626 dpif_sflow_set_options(ds, sflow_options);
1629 dpif_sflow_destroy(ds);
1630 ofproto->need_revalidate = REV_RECONFIGURE;
1631 ofproto->sflow = NULL;
1638 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1640 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1647 struct ofproto_dpif *ofproto;
1649 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1650 ofproto->need_revalidate = REV_RECONFIGURE;
1651 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1654 if (cfm_configure(ofport->cfm, s)) {
1660 cfm_destroy(ofport->cfm);
1666 get_cfm_fault(const struct ofport *ofport_)
1668 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1670 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1674 get_cfm_opup(const struct ofport *ofport_)
1676 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1678 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1682 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1685 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1688 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1696 get_cfm_health(const struct ofport *ofport_)
1698 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1700 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1703 /* Spanning Tree. */
1706 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1708 struct ofproto_dpif *ofproto = ofproto_;
1709 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1710 struct ofport_dpif *ofport;
1712 ofport = stp_port_get_aux(sp);
1714 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1715 ofproto->up.name, port_num);
1717 struct eth_header *eth = pkt->l2;
1719 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1720 if (eth_addr_is_zero(eth->eth_src)) {
1721 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1722 "with unknown MAC", ofproto->up.name, port_num);
1724 send_packet(ofport, pkt);
1730 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1732 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1734 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1736 /* Only revalidate flows if the configuration changed. */
1737 if (!s != !ofproto->stp) {
1738 ofproto->need_revalidate = REV_RECONFIGURE;
1742 if (!ofproto->stp) {
1743 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1744 send_bpdu_cb, ofproto);
1745 ofproto->stp_last_tick = time_msec();
1748 stp_set_bridge_id(ofproto->stp, s->system_id);
1749 stp_set_bridge_priority(ofproto->stp, s->priority);
1750 stp_set_hello_time(ofproto->stp, s->hello_time);
1751 stp_set_max_age(ofproto->stp, s->max_age);
1752 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1754 struct ofport *ofport;
1756 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1757 set_stp_port(ofport, NULL);
1760 stp_destroy(ofproto->stp);
1761 ofproto->stp = NULL;
1768 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1774 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1775 s->designated_root = stp_get_designated_root(ofproto->stp);
1776 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1785 update_stp_port_state(struct ofport_dpif *ofport)
1787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1788 enum stp_state state;
1790 /* Figure out new state. */
1791 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1795 if (ofport->stp_state != state) {
1796 enum ofputil_port_state of_state;
1799 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1800 netdev_get_name(ofport->up.netdev),
1801 stp_state_name(ofport->stp_state),
1802 stp_state_name(state));
1803 if (stp_learn_in_state(ofport->stp_state)
1804 != stp_learn_in_state(state)) {
1805 /* xxx Learning action flows should also be flushed. */
1806 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1808 fwd_change = stp_forward_in_state(ofport->stp_state)
1809 != stp_forward_in_state(state);
1811 ofproto->need_revalidate = REV_STP;
1812 ofport->stp_state = state;
1813 ofport->stp_state_entered = time_msec();
1815 if (fwd_change && ofport->bundle) {
1816 bundle_update(ofport->bundle);
1819 /* Update the STP state bits in the OpenFlow port description. */
1820 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1821 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1822 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1823 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1824 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1826 ofproto_port_set_state(&ofport->up, of_state);
1830 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1831 * caller is responsible for assigning STP port numbers and ensuring
1832 * there are no duplicates. */
1834 set_stp_port(struct ofport *ofport_,
1835 const struct ofproto_port_stp_settings *s)
1837 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1839 struct stp_port *sp = ofport->stp_port;
1841 if (!s || !s->enable) {
1843 ofport->stp_port = NULL;
1844 stp_port_disable(sp);
1845 update_stp_port_state(ofport);
1848 } else if (sp && stp_port_no(sp) != s->port_num
1849 && ofport == stp_port_get_aux(sp)) {
1850 /* The port-id changed, so disable the old one if it's not
1851 * already in use by another port. */
1852 stp_port_disable(sp);
1855 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1856 stp_port_enable(sp);
1858 stp_port_set_aux(sp, ofport);
1859 stp_port_set_priority(sp, s->priority);
1860 stp_port_set_path_cost(sp, s->path_cost);
1862 update_stp_port_state(ofport);
1868 get_stp_port_status(struct ofport *ofport_,
1869 struct ofproto_port_stp_status *s)
1871 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1873 struct stp_port *sp = ofport->stp_port;
1875 if (!ofproto->stp || !sp) {
1881 s->port_id = stp_port_get_id(sp);
1882 s->state = stp_port_get_state(sp);
1883 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1884 s->role = stp_port_get_role(sp);
1885 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1891 stp_run(struct ofproto_dpif *ofproto)
1894 long long int now = time_msec();
1895 long long int elapsed = now - ofproto->stp_last_tick;
1896 struct stp_port *sp;
1899 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1900 ofproto->stp_last_tick = now;
1902 while (stp_get_changed_port(ofproto->stp, &sp)) {
1903 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1906 update_stp_port_state(ofport);
1910 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1911 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1917 stp_wait(struct ofproto_dpif *ofproto)
1920 poll_timer_wait(1000);
1924 /* Returns true if STP should process 'flow'. */
1926 stp_should_process_flow(const struct flow *flow)
1928 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1932 stp_process_packet(const struct ofport_dpif *ofport,
1933 const struct ofpbuf *packet)
1935 struct ofpbuf payload = *packet;
1936 struct eth_header *eth = payload.data;
1937 struct stp_port *sp = ofport->stp_port;
1939 /* Sink packets on ports that have STP disabled when the bridge has
1941 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1945 /* Trim off padding on payload. */
1946 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1947 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1950 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1951 stp_received_bpdu(sp, payload.data, payload.size);
1955 static struct priority_to_dscp *
1956 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1958 struct priority_to_dscp *pdscp;
1961 hash = hash_int(priority, 0);
1962 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1963 if (pdscp->priority == priority) {
1971 ofport_clear_priorities(struct ofport_dpif *ofport)
1973 struct priority_to_dscp *pdscp, *next;
1975 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1976 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1982 set_queues(struct ofport *ofport_,
1983 const struct ofproto_port_queue *qdscp_list,
1986 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1988 struct hmap new = HMAP_INITIALIZER(&new);
1991 for (i = 0; i < n_qdscp; i++) {
1992 struct priority_to_dscp *pdscp;
1996 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1997 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2002 pdscp = get_priority(ofport, priority);
2004 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2006 pdscp = xmalloc(sizeof *pdscp);
2007 pdscp->priority = priority;
2009 ofproto->need_revalidate = REV_RECONFIGURE;
2012 if (pdscp->dscp != dscp) {
2014 ofproto->need_revalidate = REV_RECONFIGURE;
2017 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2020 if (!hmap_is_empty(&ofport->priorities)) {
2021 ofport_clear_priorities(ofport);
2022 ofproto->need_revalidate = REV_RECONFIGURE;
2025 hmap_swap(&new, &ofport->priorities);
2033 /* Expires all MAC learning entries associated with 'bundle' and forces its
2034 * ofproto to revalidate every flow.
2036 * Normally MAC learning entries are removed only from the ofproto associated
2037 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2038 * are removed from every ofproto. When patch ports and SLB bonds are in use
2039 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2040 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2041 * with the host from which it migrated. */
2043 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2045 struct ofproto_dpif *ofproto = bundle->ofproto;
2046 struct mac_learning *ml = ofproto->ml;
2047 struct mac_entry *mac, *next_mac;
2049 ofproto->need_revalidate = REV_RECONFIGURE;
2050 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2051 if (mac->port.p == bundle) {
2053 struct ofproto_dpif *o;
2055 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2057 struct mac_entry *e;
2059 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2062 tag_set_add(&o->revalidate_set, e->tag);
2063 mac_learning_expire(o->ml, e);
2069 mac_learning_expire(ml, mac);
2074 static struct ofbundle *
2075 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2077 struct ofbundle *bundle;
2079 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2080 &ofproto->bundles) {
2081 if (bundle->aux == aux) {
2088 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2089 * ones that are found to 'bundles'. */
2091 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2092 void **auxes, size_t n_auxes,
2093 struct hmapx *bundles)
2097 hmapx_init(bundles);
2098 for (i = 0; i < n_auxes; i++) {
2099 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2101 hmapx_add(bundles, bundle);
2107 bundle_update(struct ofbundle *bundle)
2109 struct ofport_dpif *port;
2111 bundle->floodable = true;
2112 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2113 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2114 || !stp_forward_in_state(port->stp_state)) {
2115 bundle->floodable = false;
2122 bundle_del_port(struct ofport_dpif *port)
2124 struct ofbundle *bundle = port->bundle;
2126 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2128 list_remove(&port->bundle_node);
2129 port->bundle = NULL;
2132 lacp_slave_unregister(bundle->lacp, port);
2135 bond_slave_unregister(bundle->bond, port);
2138 bundle_update(bundle);
2142 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2143 struct lacp_slave_settings *lacp,
2144 uint32_t bond_stable_id)
2146 struct ofport_dpif *port;
2148 port = get_ofp_port(bundle->ofproto, ofp_port);
2153 if (port->bundle != bundle) {
2154 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2156 bundle_del_port(port);
2159 port->bundle = bundle;
2160 list_push_back(&bundle->ports, &port->bundle_node);
2161 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2162 || !stp_forward_in_state(port->stp_state)) {
2163 bundle->floodable = false;
2167 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2168 lacp_slave_register(bundle->lacp, port, lacp);
2171 port->bond_stable_id = bond_stable_id;
2177 bundle_destroy(struct ofbundle *bundle)
2179 struct ofproto_dpif *ofproto;
2180 struct ofport_dpif *port, *next_port;
2187 ofproto = bundle->ofproto;
2188 for (i = 0; i < MAX_MIRRORS; i++) {
2189 struct ofmirror *m = ofproto->mirrors[i];
2191 if (m->out == bundle) {
2193 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2194 || hmapx_find_and_delete(&m->dsts, bundle)) {
2195 ofproto->need_revalidate = REV_RECONFIGURE;
2200 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2201 bundle_del_port(port);
2204 bundle_flush_macs(bundle, true);
2205 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2207 free(bundle->trunks);
2208 lacp_destroy(bundle->lacp);
2209 bond_destroy(bundle->bond);
2214 bundle_set(struct ofproto *ofproto_, void *aux,
2215 const struct ofproto_bundle_settings *s)
2217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2218 bool need_flush = false;
2219 struct ofport_dpif *port;
2220 struct ofbundle *bundle;
2221 unsigned long *trunks;
2227 bundle_destroy(bundle_lookup(ofproto, aux));
2231 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2232 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2234 bundle = bundle_lookup(ofproto, aux);
2236 bundle = xmalloc(sizeof *bundle);
2238 bundle->ofproto = ofproto;
2239 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2240 hash_pointer(aux, 0));
2242 bundle->name = NULL;
2244 list_init(&bundle->ports);
2245 bundle->vlan_mode = PORT_VLAN_TRUNK;
2247 bundle->trunks = NULL;
2248 bundle->use_priority_tags = s->use_priority_tags;
2249 bundle->lacp = NULL;
2250 bundle->bond = NULL;
2252 bundle->floodable = true;
2254 bundle->src_mirrors = 0;
2255 bundle->dst_mirrors = 0;
2256 bundle->mirror_out = 0;
2259 if (!bundle->name || strcmp(s->name, bundle->name)) {
2261 bundle->name = xstrdup(s->name);
2266 if (!bundle->lacp) {
2267 ofproto->need_revalidate = REV_RECONFIGURE;
2268 bundle->lacp = lacp_create();
2270 lacp_configure(bundle->lacp, s->lacp);
2272 lacp_destroy(bundle->lacp);
2273 bundle->lacp = NULL;
2276 /* Update set of ports. */
2278 for (i = 0; i < s->n_slaves; i++) {
2279 if (!bundle_add_port(bundle, s->slaves[i],
2280 s->lacp ? &s->lacp_slaves[i] : NULL,
2281 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2285 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2286 struct ofport_dpif *next_port;
2288 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2289 for (i = 0; i < s->n_slaves; i++) {
2290 if (s->slaves[i] == port->up.ofp_port) {
2295 bundle_del_port(port);
2299 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2301 if (list_is_empty(&bundle->ports)) {
2302 bundle_destroy(bundle);
2306 /* Set VLAN tagging mode */
2307 if (s->vlan_mode != bundle->vlan_mode
2308 || s->use_priority_tags != bundle->use_priority_tags) {
2309 bundle->vlan_mode = s->vlan_mode;
2310 bundle->use_priority_tags = s->use_priority_tags;
2315 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2316 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2318 if (vlan != bundle->vlan) {
2319 bundle->vlan = vlan;
2323 /* Get trunked VLANs. */
2324 switch (s->vlan_mode) {
2325 case PORT_VLAN_ACCESS:
2329 case PORT_VLAN_TRUNK:
2330 trunks = CONST_CAST(unsigned long *, s->trunks);
2333 case PORT_VLAN_NATIVE_UNTAGGED:
2334 case PORT_VLAN_NATIVE_TAGGED:
2335 if (vlan != 0 && (!s->trunks
2336 || !bitmap_is_set(s->trunks, vlan)
2337 || bitmap_is_set(s->trunks, 0))) {
2338 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2340 trunks = bitmap_clone(s->trunks, 4096);
2342 trunks = bitmap_allocate1(4096);
2344 bitmap_set1(trunks, vlan);
2345 bitmap_set0(trunks, 0);
2347 trunks = CONST_CAST(unsigned long *, s->trunks);
2354 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2355 free(bundle->trunks);
2356 if (trunks == s->trunks) {
2357 bundle->trunks = vlan_bitmap_clone(trunks);
2359 bundle->trunks = trunks;
2364 if (trunks != s->trunks) {
2369 if (!list_is_short(&bundle->ports)) {
2370 bundle->ofproto->has_bonded_bundles = true;
2372 if (bond_reconfigure(bundle->bond, s->bond)) {
2373 ofproto->need_revalidate = REV_RECONFIGURE;
2376 bundle->bond = bond_create(s->bond);
2377 ofproto->need_revalidate = REV_RECONFIGURE;
2380 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2381 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2385 bond_destroy(bundle->bond);
2386 bundle->bond = NULL;
2389 /* If we changed something that would affect MAC learning, un-learn
2390 * everything on this port and force flow revalidation. */
2392 bundle_flush_macs(bundle, false);
2399 bundle_remove(struct ofport *port_)
2401 struct ofport_dpif *port = ofport_dpif_cast(port_);
2402 struct ofbundle *bundle = port->bundle;
2405 bundle_del_port(port);
2406 if (list_is_empty(&bundle->ports)) {
2407 bundle_destroy(bundle);
2408 } else if (list_is_short(&bundle->ports)) {
2409 bond_destroy(bundle->bond);
2410 bundle->bond = NULL;
2416 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2418 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2419 struct ofport_dpif *port = port_;
2420 uint8_t ea[ETH_ADDR_LEN];
2423 error = netdev_get_etheraddr(port->up.netdev, ea);
2425 struct ofpbuf packet;
2428 ofpbuf_init(&packet, 0);
2429 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2431 memcpy(packet_pdu, pdu, pdu_size);
2433 send_packet(port, &packet);
2434 ofpbuf_uninit(&packet);
2436 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2437 "%s (%s)", port->bundle->name,
2438 netdev_get_name(port->up.netdev), strerror(error));
2443 bundle_send_learning_packets(struct ofbundle *bundle)
2445 struct ofproto_dpif *ofproto = bundle->ofproto;
2446 int error, n_packets, n_errors;
2447 struct mac_entry *e;
2449 error = n_packets = n_errors = 0;
2450 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2451 if (e->port.p != bundle) {
2452 struct ofpbuf *learning_packet;
2453 struct ofport_dpif *port;
2457 /* The assignment to "port" is unnecessary but makes "grep"ing for
2458 * struct ofport_dpif more effective. */
2459 learning_packet = bond_compose_learning_packet(bundle->bond,
2463 ret = send_packet(port, learning_packet);
2464 ofpbuf_delete(learning_packet);
2474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2475 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2476 "packets, last error was: %s",
2477 bundle->name, n_errors, n_packets, strerror(error));
2479 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2480 bundle->name, n_packets);
2485 bundle_run(struct ofbundle *bundle)
2488 lacp_run(bundle->lacp, send_pdu_cb);
2491 struct ofport_dpif *port;
2493 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2494 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2497 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2498 lacp_status(bundle->lacp));
2499 if (bond_should_send_learning_packets(bundle->bond)) {
2500 bundle_send_learning_packets(bundle);
2506 bundle_wait(struct ofbundle *bundle)
2509 lacp_wait(bundle->lacp);
2512 bond_wait(bundle->bond);
2519 mirror_scan(struct ofproto_dpif *ofproto)
2523 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2524 if (!ofproto->mirrors[idx]) {
2531 static struct ofmirror *
2532 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2536 for (i = 0; i < MAX_MIRRORS; i++) {
2537 struct ofmirror *mirror = ofproto->mirrors[i];
2538 if (mirror && mirror->aux == aux) {
2546 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2548 mirror_update_dups(struct ofproto_dpif *ofproto)
2552 for (i = 0; i < MAX_MIRRORS; i++) {
2553 struct ofmirror *m = ofproto->mirrors[i];
2556 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2560 for (i = 0; i < MAX_MIRRORS; i++) {
2561 struct ofmirror *m1 = ofproto->mirrors[i];
2568 for (j = i + 1; j < MAX_MIRRORS; j++) {
2569 struct ofmirror *m2 = ofproto->mirrors[j];
2571 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2572 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2573 m2->dup_mirrors |= m1->dup_mirrors;
2580 mirror_set(struct ofproto *ofproto_, void *aux,
2581 const struct ofproto_mirror_settings *s)
2583 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2584 mirror_mask_t mirror_bit;
2585 struct ofbundle *bundle;
2586 struct ofmirror *mirror;
2587 struct ofbundle *out;
2588 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2589 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2592 mirror = mirror_lookup(ofproto, aux);
2594 mirror_destroy(mirror);
2600 idx = mirror_scan(ofproto);
2602 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2604 ofproto->up.name, MAX_MIRRORS, s->name);
2608 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2609 mirror->ofproto = ofproto;
2612 mirror->out_vlan = -1;
2613 mirror->name = NULL;
2616 if (!mirror->name || strcmp(s->name, mirror->name)) {
2618 mirror->name = xstrdup(s->name);
2621 /* Get the new configuration. */
2622 if (s->out_bundle) {
2623 out = bundle_lookup(ofproto, s->out_bundle);
2625 mirror_destroy(mirror);
2631 out_vlan = s->out_vlan;
2633 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2634 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2636 /* If the configuration has not changed, do nothing. */
2637 if (hmapx_equals(&srcs, &mirror->srcs)
2638 && hmapx_equals(&dsts, &mirror->dsts)
2639 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2640 && mirror->out == out
2641 && mirror->out_vlan == out_vlan)
2643 hmapx_destroy(&srcs);
2644 hmapx_destroy(&dsts);
2648 hmapx_swap(&srcs, &mirror->srcs);
2649 hmapx_destroy(&srcs);
2651 hmapx_swap(&dsts, &mirror->dsts);
2652 hmapx_destroy(&dsts);
2654 free(mirror->vlans);
2655 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2658 mirror->out_vlan = out_vlan;
2660 /* Update bundles. */
2661 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2662 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2663 if (hmapx_contains(&mirror->srcs, bundle)) {
2664 bundle->src_mirrors |= mirror_bit;
2666 bundle->src_mirrors &= ~mirror_bit;
2669 if (hmapx_contains(&mirror->dsts, bundle)) {
2670 bundle->dst_mirrors |= mirror_bit;
2672 bundle->dst_mirrors &= ~mirror_bit;
2675 if (mirror->out == bundle) {
2676 bundle->mirror_out |= mirror_bit;
2678 bundle->mirror_out &= ~mirror_bit;
2682 ofproto->need_revalidate = REV_RECONFIGURE;
2683 ofproto->has_mirrors = true;
2684 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2685 mirror_update_dups(ofproto);
2691 mirror_destroy(struct ofmirror *mirror)
2693 struct ofproto_dpif *ofproto;
2694 mirror_mask_t mirror_bit;
2695 struct ofbundle *bundle;
2702 ofproto = mirror->ofproto;
2703 ofproto->need_revalidate = REV_RECONFIGURE;
2704 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2706 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2707 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2708 bundle->src_mirrors &= ~mirror_bit;
2709 bundle->dst_mirrors &= ~mirror_bit;
2710 bundle->mirror_out &= ~mirror_bit;
2713 hmapx_destroy(&mirror->srcs);
2714 hmapx_destroy(&mirror->dsts);
2715 free(mirror->vlans);
2717 ofproto->mirrors[mirror->idx] = NULL;
2721 mirror_update_dups(ofproto);
2723 ofproto->has_mirrors = false;
2724 for (i = 0; i < MAX_MIRRORS; i++) {
2725 if (ofproto->mirrors[i]) {
2726 ofproto->has_mirrors = true;
2733 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2734 uint64_t *packets, uint64_t *bytes)
2736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2737 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2740 *packets = *bytes = UINT64_MAX;
2744 *packets = mirror->packet_count;
2745 *bytes = mirror->byte_count;
2751 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2755 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2761 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2763 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2764 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2765 return bundle && bundle->mirror_out != 0;
2769 forward_bpdu_changed(struct ofproto *ofproto_)
2771 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772 ofproto->need_revalidate = REV_RECONFIGURE;
2776 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780 mac_learning_set_idle_time(ofproto->ml, idle_time);
2781 mac_learning_set_max_entries(ofproto->ml, max_entries);
2786 static struct ofport_dpif *
2787 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2789 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2790 return ofport ? ofport_dpif_cast(ofport) : NULL;
2793 static struct ofport_dpif *
2794 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2796 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2797 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2801 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2802 struct ofproto_port *ofproto_port,
2803 struct dpif_port *dpif_port)
2805 ofproto_port->name = dpif_port->name;
2806 ofproto_port->type = dpif_port->type;
2807 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2811 port_run_fast(struct ofport_dpif *ofport)
2813 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2814 struct ofpbuf packet;
2816 ofpbuf_init(&packet, 0);
2817 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2818 send_packet(ofport, &packet);
2819 ofpbuf_uninit(&packet);
2824 port_run(struct ofport_dpif *ofport)
2826 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2827 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2828 bool enable = netdev_get_carrier(ofport->up.netdev);
2830 ofport->carrier_seq = carrier_seq;
2832 port_run_fast(ofport);
2834 int cfm_opup = cfm_get_opup(ofport->cfm);
2836 cfm_run(ofport->cfm);
2837 enable = enable && !cfm_get_fault(ofport->cfm);
2839 if (cfm_opup >= 0) {
2840 enable = enable && cfm_opup;
2844 if (ofport->bundle) {
2845 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2846 if (carrier_changed) {
2847 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2851 if (ofport->may_enable != enable) {
2852 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2854 if (ofproto->has_bundle_action) {
2855 ofproto->need_revalidate = REV_PORT_TOGGLED;
2859 ofport->may_enable = enable;
2863 port_wait(struct ofport_dpif *ofport)
2866 cfm_wait(ofport->cfm);
2871 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2872 struct ofproto_port *ofproto_port)
2874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2875 struct dpif_port dpif_port;
2878 if (!sset_contains(&ofproto->ports, devname)) {
2881 error = dpif_port_query_by_name(ofproto->backer->dpif,
2882 devname, &dpif_port);
2884 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2890 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2893 uint32_t odp_port = UINT32_MAX;
2896 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2898 sset_add(&ofproto->ports, netdev_get_name(netdev));
2904 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2907 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2910 if (odp_port != OFPP_NONE) {
2911 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2914 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2916 /* The caller is going to close ofport->up.netdev. If this is a
2917 * bonded port, then the bond is using that netdev, so remove it
2918 * from the bond. The client will need to reconfigure everything
2919 * after deleting ports, so then the slave will get re-added. */
2920 bundle_remove(&ofport->up);
2927 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2929 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2932 error = netdev_get_stats(ofport->up.netdev, stats);
2934 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2935 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2937 /* ofproto->stats.tx_packets represents packets that we created
2938 * internally and sent to some port (e.g. packets sent with
2939 * send_packet()). Account for them as if they had come from
2940 * OFPP_LOCAL and got forwarded. */
2942 if (stats->rx_packets != UINT64_MAX) {
2943 stats->rx_packets += ofproto->stats.tx_packets;
2946 if (stats->rx_bytes != UINT64_MAX) {
2947 stats->rx_bytes += ofproto->stats.tx_bytes;
2950 /* ofproto->stats.rx_packets represents packets that were received on
2951 * some port and we processed internally and dropped (e.g. STP).
2952 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2954 if (stats->tx_packets != UINT64_MAX) {
2955 stats->tx_packets += ofproto->stats.rx_packets;
2958 if (stats->tx_bytes != UINT64_MAX) {
2959 stats->tx_bytes += ofproto->stats.rx_bytes;
2966 /* Account packets for LOCAL port. */
2968 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2969 size_t tx_size, size_t rx_size)
2971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2974 ofproto->stats.rx_packets++;
2975 ofproto->stats.rx_bytes += rx_size;
2978 ofproto->stats.tx_packets++;
2979 ofproto->stats.tx_bytes += tx_size;
2983 struct port_dump_state {
2989 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2991 struct port_dump_state *state;
2993 *statep = state = xmalloc(sizeof *state);
3000 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
3001 struct ofproto_port *port)
3003 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3004 struct port_dump_state *state = state_;
3005 struct sset_node *node;
3007 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
3011 error = port_query_by_name(ofproto_, node->name, port);
3012 if (error != ENODEV) {
3021 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3023 struct port_dump_state *state = state_;
3030 port_poll(const struct ofproto *ofproto_, char **devnamep)
3032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3034 if (ofproto->port_poll_errno) {
3035 int error = ofproto->port_poll_errno;
3036 ofproto->port_poll_errno = 0;
3040 if (sset_is_empty(&ofproto->port_poll_set)) {
3044 *devnamep = sset_pop(&ofproto->port_poll_set);
3049 port_poll_wait(const struct ofproto *ofproto_)
3051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052 dpif_port_poll_wait(ofproto->backer->dpif);
3056 port_is_lacp_current(const struct ofport *ofport_)
3058 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3059 return (ofport->bundle && ofport->bundle->lacp
3060 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3064 /* Upcall handling. */
3066 /* Flow miss batching.
3068 * Some dpifs implement operations faster when you hand them off in a batch.
3069 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3070 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3071 * more packets, plus possibly installing the flow in the dpif.
3073 * So far we only batch the operations that affect flow setup time the most.
3074 * It's possible to batch more than that, but the benefit might be minimal. */
3076 struct hmap_node hmap_node;
3077 struct ofproto_dpif *ofproto;
3079 enum odp_key_fitness key_fitness;
3080 const struct nlattr *key;
3082 ovs_be16 initial_tci;
3083 struct list packets;
3084 enum dpif_upcall_type upcall_type;
3085 uint32_t odp_in_port;
3088 struct flow_miss_op {
3089 struct dpif_op dpif_op;
3090 struct subfacet *subfacet; /* Subfacet */
3091 void *garbage; /* Pointer to pass to free(), NULL if none. */
3092 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3095 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3096 * OpenFlow controller as necessary according to their individual
3097 * configurations. */
3099 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3100 const struct flow *flow)
3102 struct ofputil_packet_in pin;
3104 pin.packet = packet->data;
3105 pin.packet_len = packet->size;
3106 pin.reason = OFPR_NO_MATCH;
3107 pin.controller_id = 0;
3112 pin.send_len = 0; /* not used for flow table misses */
3114 flow_get_metadata(flow, &pin.fmd);
3116 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3119 static enum slow_path_reason
3120 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3121 const struct ofpbuf *packet)
3123 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3129 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3131 cfm_process_heartbeat(ofport->cfm, packet);
3134 } else if (ofport->bundle && ofport->bundle->lacp
3135 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3137 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3140 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3142 stp_process_packet(ofport, packet);
3149 static struct flow_miss *
3150 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3152 struct flow_miss *miss;
3154 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3155 if (flow_equal(&miss->flow, flow)) {
3163 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3164 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3165 * 'miss' is associated with a subfacet the caller must also initialize the
3166 * returned op->subfacet, and if anything needs to be freed after processing
3167 * the op, the caller must initialize op->garbage also. */
3169 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3170 struct flow_miss_op *op)
3172 if (miss->flow.vlan_tci != miss->initial_tci) {
3173 /* This packet was received on a VLAN splinter port. We
3174 * added a VLAN to the packet to make the packet resemble
3175 * the flow, but the actions were composed assuming that
3176 * the packet contained no VLAN. So, we must remove the
3177 * VLAN header from the packet before trying to execute the
3179 eth_pop_vlan(packet);
3182 op->subfacet = NULL;
3184 op->dpif_op.type = DPIF_OP_EXECUTE;
3185 op->dpif_op.u.execute.key = miss->key;
3186 op->dpif_op.u.execute.key_len = miss->key_len;
3187 op->dpif_op.u.execute.packet = packet;
3190 /* Helper for handle_flow_miss_without_facet() and
3191 * handle_flow_miss_with_facet(). */
3193 handle_flow_miss_common(struct rule_dpif *rule,
3194 struct ofpbuf *packet, const struct flow *flow)
3196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3198 ofproto->n_matches++;
3200 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3202 * Extra-special case for fail-open mode.
3204 * We are in fail-open mode and the packet matched the fail-open
3205 * rule, but we are connected to a controller too. We should send
3206 * the packet up to the controller in the hope that it will try to
3207 * set up a flow and thereby allow us to exit fail-open.
3209 * See the top-level comment in fail-open.c for more information.
3211 send_packet_in_miss(ofproto, packet, flow);
3215 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3216 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3217 * installing a datapath flow. The answer is usually "yes" (a return value of
3218 * true). However, for short flows the cost of bookkeeping is much higher than
3219 * the benefits, so when the datapath holds a large number of flows we impose
3220 * some heuristics to decide which flows are likely to be worth tracking. */
3222 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3223 struct flow_miss *miss, uint32_t hash)
3225 if (!ofproto->governor) {
3228 n_subfacets = hmap_count(&ofproto->subfacets);
3229 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3233 ofproto->governor = governor_create(ofproto->up.name);
3236 return governor_should_install_flow(ofproto->governor, hash,
3237 list_size(&miss->packets));
3240 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3241 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3242 * increment '*n_ops'. */
3244 handle_flow_miss_without_facet(struct flow_miss *miss,
3245 struct rule_dpif *rule,
3246 struct flow_miss_op *ops, size_t *n_ops)
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3249 long long int now = time_msec();
3250 struct action_xlate_ctx ctx;
3251 struct ofpbuf *packet;
3253 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3254 struct flow_miss_op *op = &ops[*n_ops];
3255 struct dpif_flow_stats stats;
3256 struct ofpbuf odp_actions;
3258 COVERAGE_INC(facet_suppress);
3260 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3262 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3263 rule_credit_stats(rule, &stats);
3265 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3267 ctx.resubmit_stats = &stats;
3268 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3271 if (odp_actions.size) {
3272 struct dpif_execute *execute = &op->dpif_op.u.execute;
3274 init_flow_miss_execute_op(miss, packet, op);
3275 execute->actions = odp_actions.data;
3276 execute->actions_len = odp_actions.size;
3277 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3281 ofpbuf_uninit(&odp_actions);
3286 /* Handles 'miss', which matches 'facet'. May add any required datapath
3287 * operations to 'ops', incrementing '*n_ops' for each new op.
3289 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3290 * This is really important only for new facets: if we just called time_msec()
3291 * here, then the new subfacet or its packets could look (occasionally) as
3292 * though it was used some time after the facet was used. That can make a
3293 * one-packet flow look like it has a nonzero duration, which looks odd in
3294 * e.g. NetFlow statistics. */
3296 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3298 struct flow_miss_op *ops, size_t *n_ops)
3300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3301 enum subfacet_path want_path;
3302 struct subfacet *subfacet;
3303 struct ofpbuf *packet;
3305 subfacet = subfacet_create(facet, miss, now);
3307 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3308 struct flow_miss_op *op = &ops[*n_ops];
3309 struct dpif_flow_stats stats;
3310 struct ofpbuf odp_actions;
3312 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3314 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3315 if (!subfacet->actions || subfacet->slow) {
3316 subfacet_make_actions(subfacet, packet, &odp_actions);
3319 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3320 subfacet_update_stats(subfacet, &stats);
3322 if (subfacet->actions_len) {
3323 struct dpif_execute *execute = &op->dpif_op.u.execute;
3325 init_flow_miss_execute_op(miss, packet, op);
3326 op->subfacet = subfacet;
3327 if (!subfacet->slow) {
3328 execute->actions = subfacet->actions;
3329 execute->actions_len = subfacet->actions_len;
3330 ofpbuf_uninit(&odp_actions);
3332 execute->actions = odp_actions.data;
3333 execute->actions_len = odp_actions.size;
3334 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3339 ofpbuf_uninit(&odp_actions);
3343 want_path = subfacet_want_path(subfacet->slow);
3344 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3345 struct flow_miss_op *op = &ops[(*n_ops)++];
3346 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3348 op->subfacet = subfacet;
3350 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3351 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3352 put->key = miss->key;
3353 put->key_len = miss->key_len;
3354 if (want_path == SF_FAST_PATH) {
3355 put->actions = subfacet->actions;
3356 put->actions_len = subfacet->actions_len;
3358 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3359 op->stub, sizeof op->stub,
3360 &put->actions, &put->actions_len);
3366 /* Handles flow miss 'miss'. May add any required datapath operations
3367 * to 'ops', incrementing '*n_ops' for each new op. */
3369 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3372 struct ofproto_dpif *ofproto = miss->ofproto;
3373 struct facet *facet;
3377 /* The caller must ensure that miss->hmap_node.hash contains
3378 * flow_hash(miss->flow, 0). */
3379 hash = miss->hmap_node.hash;
3381 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3383 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3385 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3386 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3390 facet = facet_create(rule, &miss->flow, hash);
3395 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3398 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3399 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3400 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3401 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3402 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3403 * 'packet' ingressed.
3405 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3406 * 'flow''s in_port to OFPP_NONE.
3408 * This function does post-processing on data returned from
3409 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3410 * of the upcall processing logic. In particular, if the extracted in_port is
3411 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3412 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3413 * a VLAN header onto 'packet' (if it is nonnull).
3415 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3416 * packet was really received, that is, the actual VLAN TCI extracted by
3417 * odp_flow_key_to_flow(). (This differs from the value returned in
3418 * flow->vlan_tci only for packets received on VLAN splinters.)
3420 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3421 * or some other positive errno if there are other problems. */
3423 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3424 const struct nlattr *key, size_t key_len,
3425 struct flow *flow, enum odp_key_fitness *fitnessp,
3426 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3427 ovs_be16 *initial_tci)
3429 const struct ofport_dpif *port;
3430 enum odp_key_fitness fitness;
3433 fitness = odp_flow_key_to_flow(key, key_len, flow);
3434 if (fitness == ODP_FIT_ERROR) {
3440 *initial_tci = flow->vlan_tci;
3444 *odp_in_port = flow->in_port;
3447 port = odp_port_to_ofport(backer, flow->in_port);
3449 flow->in_port = OFPP_NONE;
3450 error = ofproto ? ENODEV : 0;
3455 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3458 flow->in_port = port->up.ofp_port;
3459 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3461 /* Make the packet resemble the flow, so that it gets sent to an
3462 * OpenFlow controller properly, so that it looks correct for
3463 * sFlow, and so that flow_extract() will get the correct vlan_tci
3464 * if it is called on 'packet'.
3466 * The allocated space inside 'packet' probably also contains
3467 * 'key', that is, both 'packet' and 'key' are probably part of a
3468 * struct dpif_upcall (see the large comment on that structure
3469 * definition), so pushing data on 'packet' is in general not a
3470 * good idea since it could overwrite 'key' or free it as a side
3471 * effect. However, it's OK in this special case because we know
3472 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3473 * will just overwrite the 4-byte "struct nlattr", which is fine
3474 * since we don't need that header anymore. */
3475 eth_push_vlan(packet, flow->vlan_tci);
3478 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3479 if (fitness == ODP_FIT_PERFECT) {
3480 fitness = ODP_FIT_TOO_MUCH;
3487 *fitnessp = fitness;
3493 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3496 struct dpif_upcall *upcall;
3497 struct flow_miss *miss;
3498 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3499 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3500 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3510 /* Construct the to-do list.
3512 * This just amounts to extracting the flow from each packet and sticking
3513 * the packets that have the same flow in the same "flow_miss" structure so
3514 * that we can process them together. */
3517 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3518 struct flow_miss *miss = &misses[n_misses];
3519 struct flow_miss *existing_miss;
3520 struct ofproto_dpif *ofproto;
3521 uint32_t odp_in_port;
3526 error = ofproto_receive(backer, upcall->packet, upcall->key,
3527 upcall->key_len, &flow, &miss->key_fitness,
3528 &ofproto, &odp_in_port, &miss->initial_tci);
3529 if (error == ENODEV) {
3530 /* Received packet on port for which we couldn't associate
3531 * an ofproto. This can happen if a port is removed while
3532 * traffic is being received. Print a rate-limited message
3533 * in case it happens frequently. */
3534 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3540 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3541 &flow.tunnel, flow.in_port, &miss->flow);
3543 /* Add other packets to a to-do list. */
3544 hash = flow_hash(&miss->flow, 0);
3545 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3546 if (!existing_miss) {
3547 hmap_insert(&todo, &miss->hmap_node, hash);
3548 miss->ofproto = ofproto;
3549 miss->key = upcall->key;
3550 miss->key_len = upcall->key_len;
3551 miss->upcall_type = upcall->type;
3552 miss->odp_in_port = odp_in_port;
3553 list_init(&miss->packets);
3557 miss = existing_miss;
3559 list_push_back(&miss->packets, &upcall->packet->list_node);
3562 /* Process each element in the to-do list, constructing the set of
3563 * operations to batch. */
3565 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3566 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3568 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3570 /* Execute batch. */
3571 for (i = 0; i < n_ops; i++) {
3572 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3574 dpif_operate(backer->dpif, dpif_ops, n_ops);
3576 /* Free memory and update facets. */
3577 for (i = 0; i < n_ops; i++) {
3578 struct flow_miss_op *op = &flow_miss_ops[i];
3580 switch (op->dpif_op.type) {
3581 case DPIF_OP_EXECUTE:
3584 case DPIF_OP_FLOW_PUT:
3585 if (!op->dpif_op.error) {
3586 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3590 case DPIF_OP_FLOW_DEL:
3596 hmap_destroy(&todo);
3599 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3600 classify_upcall(const struct dpif_upcall *upcall)
3602 union user_action_cookie cookie;
3604 /* First look at the upcall type. */
3605 switch (upcall->type) {
3606 case DPIF_UC_ACTION:
3612 case DPIF_N_UC_TYPES:
3614 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3618 /* "action" upcalls need a closer look. */
3619 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3620 switch (cookie.type) {
3621 case USER_ACTION_COOKIE_SFLOW:
3622 return SFLOW_UPCALL;
3624 case USER_ACTION_COOKIE_SLOW_PATH:
3627 case USER_ACTION_COOKIE_UNSPEC:
3629 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3635 handle_sflow_upcall(struct dpif_backer *backer,
3636 const struct dpif_upcall *upcall)
3638 struct ofproto_dpif *ofproto;
3639 union user_action_cookie cookie;
3641 uint32_t odp_in_port;
3643 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3644 &flow, NULL, &ofproto, &odp_in_port, NULL)
3645 || !ofproto->sflow) {
3649 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3650 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3651 odp_in_port, &cookie);
3655 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3657 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3658 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3659 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3664 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3667 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3668 struct dpif_upcall *upcall = &misses[n_misses];
3669 struct ofpbuf *buf = &miss_bufs[n_misses];
3672 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3673 sizeof miss_buf_stubs[n_misses]);
3674 error = dpif_recv(backer->dpif, upcall, buf);
3680 switch (classify_upcall(upcall)) {
3682 /* Handle it later. */
3687 handle_sflow_upcall(backer, upcall);
3697 /* Handle deferred MISS_UPCALL processing. */
3698 handle_miss_upcalls(backer, misses, n_misses);
3699 for (i = 0; i < n_misses; i++) {
3700 ofpbuf_uninit(&miss_bufs[i]);
3706 /* Flow expiration. */
3708 static int subfacet_max_idle(const struct ofproto_dpif *);
3709 static void update_stats(struct dpif_backer *);
3710 static void rule_expire(struct rule_dpif *);
3711 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3713 /* This function is called periodically by run(). Its job is to collect
3714 * updates for the flows that have been installed into the datapath, most
3715 * importantly when they last were used, and then use that information to
3716 * expire flows that have not been used recently.
3718 * Returns the number of milliseconds after which it should be called again. */
3720 expire(struct dpif_backer *backer)
3722 struct ofproto_dpif *ofproto;
3723 int max_idle = INT32_MAX;
3725 /* Update stats for each flow in the backer. */
3726 update_stats(backer);
3728 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3729 struct rule_dpif *rule, *next_rule;
3730 struct oftable *table;
3733 if (ofproto->backer != backer) {
3737 /* Expire subfacets that have been idle too long. */
3738 dp_max_idle = subfacet_max_idle(ofproto);
3739 expire_subfacets(ofproto, dp_max_idle);
3741 max_idle = MIN(max_idle, dp_max_idle);
3743 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3745 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3746 struct cls_cursor cursor;
3748 cls_cursor_init(&cursor, &table->cls, NULL);
3749 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3754 /* All outstanding data in existing flows has been accounted, so it's a
3755 * good time to do bond rebalancing. */
3756 if (ofproto->has_bonded_bundles) {
3757 struct ofbundle *bundle;
3759 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3761 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3767 return MIN(max_idle, 1000);
3770 /* Updates flow table statistics given that the datapath just reported 'stats'
3771 * as 'subfacet''s statistics. */
3773 update_subfacet_stats(struct subfacet *subfacet,
3774 const struct dpif_flow_stats *stats)
3776 struct facet *facet = subfacet->facet;
3778 if (stats->n_packets >= subfacet->dp_packet_count) {
3779 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3780 facet->packet_count += extra;
3782 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3785 if (stats->n_bytes >= subfacet->dp_byte_count) {
3786 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3788 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3791 subfacet->dp_packet_count = stats->n_packets;
3792 subfacet->dp_byte_count = stats->n_bytes;
3794 facet->tcp_flags |= stats->tcp_flags;
3796 subfacet_update_time(subfacet, stats->used);
3797 if (facet->accounted_bytes < facet->byte_count) {
3799 facet_account(facet);
3800 facet->accounted_bytes = facet->byte_count;
3802 facet_push_stats(facet);
3805 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3806 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3808 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3809 const struct nlattr *key, size_t key_len)
3811 if (!VLOG_DROP_WARN(&rl)) {
3815 odp_flow_key_format(key, key_len, &s);
3816 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3820 COVERAGE_INC(facet_unexpected);
3821 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3824 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3826 * This function also pushes statistics updates to rules which each facet
3827 * resubmits into. Generally these statistics will be accurate. However, if a
3828 * facet changes the rule it resubmits into at some time in between
3829 * update_stats() runs, it is possible that statistics accrued to the
3830 * old rule will be incorrectly attributed to the new rule. This could be
3831 * avoided by calling update_stats() whenever rules are created or
3832 * deleted. However, the performance impact of making so many calls to the
3833 * datapath do not justify the benefit of having perfectly accurate statistics.
3836 update_stats(struct dpif_backer *backer)
3838 const struct dpif_flow_stats *stats;
3839 struct dpif_flow_dump dump;
3840 const struct nlattr *key;
3843 dpif_flow_dump_start(&dump, backer->dpif);
3844 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3846 struct subfacet *subfacet;
3847 enum odp_key_fitness fitness;
3848 struct ofproto_dpif *ofproto;
3849 struct ofport_dpif *port;
3852 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3853 if (fitness == ODP_FIT_ERROR) {
3857 port = odp_port_to_ofport(backer, flow.in_port);
3859 /* This flow is for a port for which we couldn't associate an
3860 * ofproto. This can happen if a port is removed while
3861 * traffic is being received. Ignore this flow, since it
3862 * will get timed out. */
3866 ofproto = ofproto_dpif_cast(port->up.ofproto);
3867 flow.in_port = port->up.ofp_port;
3868 key_hash = odp_flow_key_hash(key, key_len);
3870 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3871 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3873 update_subfacet_stats(subfacet, stats);
3877 /* Stats are updated per-packet. */
3880 case SF_NOT_INSTALLED:
3882 delete_unexpected_flow(ofproto, key, key_len);
3886 dpif_flow_dump_done(&dump);
3889 /* Calculates and returns the number of milliseconds of idle time after which
3890 * subfacets should expire from the datapath. When a subfacet expires, we fold
3891 * its statistics into its facet, and when a facet's last subfacet expires, we
3892 * fold its statistic into its rule. */
3894 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3897 * Idle time histogram.
3899 * Most of the time a switch has a relatively small number of subfacets.
3900 * When this is the case we might as well keep statistics for all of them
3901 * in userspace and to cache them in the kernel datapath for performance as
3904 * As the number of subfacets increases, the memory required to maintain
3905 * statistics about them in userspace and in the kernel becomes
3906 * significant. However, with a large number of subfacets it is likely
3907 * that only a few of them are "heavy hitters" that consume a large amount
3908 * of bandwidth. At this point, only heavy hitters are worth caching in
3909 * the kernel and maintaining in userspaces; other subfacets we can
3912 * The technique used to compute the idle time is to build a histogram with
3913 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3914 * that is installed in the kernel gets dropped in the appropriate bucket.
3915 * After the histogram has been built, we compute the cutoff so that only
3916 * the most-recently-used 1% of subfacets (but at least
3917 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3918 * the most-recently-used bucket of subfacets is kept, so actually an
3919 * arbitrary number of subfacets can be kept in any given expiration run
3920 * (though the next run will delete most of those unless they receive
3923 * This requires a second pass through the subfacets, in addition to the
3924 * pass made by update_stats(), because the former function never looks at
3925 * uninstallable subfacets.
3927 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3928 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3929 int buckets[N_BUCKETS] = { 0 };
3930 int total, subtotal, bucket;
3931 struct subfacet *subfacet;
3935 total = hmap_count(&ofproto->subfacets);
3936 if (total <= ofproto->up.flow_eviction_threshold) {
3937 return N_BUCKETS * BUCKET_WIDTH;
3940 /* Build histogram. */
3942 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3943 long long int idle = now - subfacet->used;
3944 int bucket = (idle <= 0 ? 0
3945 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3946 : (unsigned int) idle / BUCKET_WIDTH);
3950 /* Find the first bucket whose flows should be expired. */
3951 subtotal = bucket = 0;
3953 subtotal += buckets[bucket++];
3954 } while (bucket < N_BUCKETS &&
3955 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3957 if (VLOG_IS_DBG_ENABLED()) {
3961 ds_put_cstr(&s, "keep");
3962 for (i = 0; i < N_BUCKETS; i++) {
3964 ds_put_cstr(&s, ", drop");
3967 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3970 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3974 return bucket * BUCKET_WIDTH;
3978 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3980 /* Cutoff time for most flows. */
3981 long long int normal_cutoff = time_msec() - dp_max_idle;
3983 /* We really want to keep flows for special protocols around, so use a more
3984 * conservative cutoff. */
3985 long long int special_cutoff = time_msec() - 10000;
3987 struct subfacet *subfacet, *next_subfacet;
3988 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3992 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3993 &ofproto->subfacets) {
3994 long long int cutoff;
3996 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3999 if (subfacet->used < cutoff) {
4000 if (subfacet->path != SF_NOT_INSTALLED) {
4001 batch[n_batch++] = subfacet;
4002 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4003 subfacet_destroy_batch(ofproto, batch, n_batch);
4007 subfacet_destroy(subfacet);
4013 subfacet_destroy_batch(ofproto, batch, n_batch);
4017 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4018 * then delete it entirely. */
4020 rule_expire(struct rule_dpif *rule)
4022 struct facet *facet, *next_facet;
4026 if (rule->up.pending) {
4027 /* We'll have to expire it later. */
4031 /* Has 'rule' expired? */
4033 if (rule->up.hard_timeout
4034 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4035 reason = OFPRR_HARD_TIMEOUT;
4036 } else if (rule->up.idle_timeout
4037 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4038 reason = OFPRR_IDLE_TIMEOUT;
4043 COVERAGE_INC(ofproto_dpif_expired);
4045 /* Update stats. (This is a no-op if the rule expired due to an idle
4046 * timeout, because that only happens when the rule has no facets left.) */
4047 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4048 facet_remove(facet);
4051 /* Get rid of the rule. */
4052 ofproto_rule_expire(&rule->up, reason);
4057 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4059 * The caller must already have determined that no facet with an identical
4060 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4061 * the ofproto's classifier table.
4063 * 'hash' must be the return value of flow_hash(flow, 0).
4065 * The facet will initially have no subfacets. The caller should create (at
4066 * least) one subfacet with subfacet_create(). */
4067 static struct facet *
4068 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4071 struct facet *facet;
4073 facet = xzalloc(sizeof *facet);
4074 facet->used = time_msec();
4075 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4076 list_push_back(&rule->facets, &facet->list_node);
4078 facet->flow = *flow;
4079 list_init(&facet->subfacets);
4080 netflow_flow_init(&facet->nf_flow);
4081 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4087 facet_free(struct facet *facet)
4092 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4093 * 'packet', which arrived on 'in_port'.
4095 * Takes ownership of 'packet'. */
4097 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4098 const struct nlattr *odp_actions, size_t actions_len,
4099 struct ofpbuf *packet)
4101 struct odputil_keybuf keybuf;
4105 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4106 odp_flow_key_from_flow(&key, flow,
4107 ofp_port_to_odp_port(ofproto, flow->in_port));
4109 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4110 odp_actions, actions_len, packet);
4112 ofpbuf_delete(packet);
4116 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4118 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4119 * rule's statistics, via subfacet_uninstall().
4121 * - Removes 'facet' from its rule and from ofproto->facets.
4124 facet_remove(struct facet *facet)
4126 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4127 struct subfacet *subfacet, *next_subfacet;
4129 ovs_assert(!list_is_empty(&facet->subfacets));
4131 /* First uninstall all of the subfacets to get final statistics. */
4132 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4133 subfacet_uninstall(subfacet);
4136 /* Flush the final stats to the rule.
4138 * This might require us to have at least one subfacet around so that we
4139 * can use its actions for accounting in facet_account(), which is why we
4140 * have uninstalled but not yet destroyed the subfacets. */
4141 facet_flush_stats(facet);
4143 /* Now we're really all done so destroy everything. */
4144 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4145 &facet->subfacets) {
4146 subfacet_destroy__(subfacet);
4148 hmap_remove(&ofproto->facets, &facet->hmap_node);
4149 list_remove(&facet->list_node);
4153 /* Feed information from 'facet' back into the learning table to keep it in
4154 * sync with what is actually flowing through the datapath. */
4156 facet_learn(struct facet *facet)
4158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4159 struct action_xlate_ctx ctx;
4161 if (!facet->has_learn
4162 && !facet->has_normal
4163 && (!facet->has_fin_timeout
4164 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4168 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4169 facet->flow.vlan_tci,
4170 facet->rule, facet->tcp_flags, NULL);
4171 ctx.may_learn = true;
4172 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4173 facet->rule->up.ofpacts_len);
4177 facet_account(struct facet *facet)
4179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4180 struct subfacet *subfacet;
4181 const struct nlattr *a;
4186 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4189 n_bytes = facet->byte_count - facet->accounted_bytes;
4191 /* This loop feeds byte counters to bond_account() for rebalancing to use
4192 * as a basis. We also need to track the actual VLAN on which the packet
4193 * is going to be sent to ensure that it matches the one passed to
4194 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4197 * We use the actions from an arbitrary subfacet because they should all
4198 * be equally valid for our purpose. */
4199 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4200 struct subfacet, list_node);
4201 vlan_tci = facet->flow.vlan_tci;
4202 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4203 subfacet->actions, subfacet->actions_len) {
4204 const struct ovs_action_push_vlan *vlan;
4205 struct ofport_dpif *port;
4207 switch (nl_attr_type(a)) {
4208 case OVS_ACTION_ATTR_OUTPUT:
4209 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4210 if (port && port->bundle && port->bundle->bond) {
4211 bond_account(port->bundle->bond, &facet->flow,
4212 vlan_tci_to_vid(vlan_tci), n_bytes);
4216 case OVS_ACTION_ATTR_POP_VLAN:
4217 vlan_tci = htons(0);
4220 case OVS_ACTION_ATTR_PUSH_VLAN:
4221 vlan = nl_attr_get(a);
4222 vlan_tci = vlan->vlan_tci;
4228 /* Returns true if the only action for 'facet' is to send to the controller.
4229 * (We don't report NetFlow expiration messages for such facets because they
4230 * are just part of the control logic for the network, not real traffic). */
4232 facet_is_controller_flow(struct facet *facet)
4235 const struct rule *rule = &facet->rule->up;
4236 const struct ofpact *ofpacts = rule->ofpacts;
4237 size_t ofpacts_len = rule->ofpacts_len;
4239 if (ofpacts_len > 0 &&
4240 ofpacts->type == OFPACT_CONTROLLER &&
4241 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4248 /* Folds all of 'facet''s statistics into its rule. Also updates the
4249 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4250 * 'facet''s statistics in the datapath should have been zeroed and folded into
4251 * its packet and byte counts before this function is called. */
4253 facet_flush_stats(struct facet *facet)
4255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4256 struct subfacet *subfacet;
4258 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4259 ovs_assert(!subfacet->dp_byte_count);
4260 ovs_assert(!subfacet->dp_packet_count);
4263 facet_push_stats(facet);
4264 if (facet->accounted_bytes < facet->byte_count) {
4265 facet_account(facet);
4266 facet->accounted_bytes = facet->byte_count;
4269 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4270 struct ofexpired expired;
4271 expired.flow = facet->flow;
4272 expired.packet_count = facet->packet_count;
4273 expired.byte_count = facet->byte_count;
4274 expired.used = facet->used;
4275 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4278 facet->rule->packet_count += facet->packet_count;
4279 facet->rule->byte_count += facet->byte_count;
4281 /* Reset counters to prevent double counting if 'facet' ever gets
4283 facet_reset_counters(facet);
4285 netflow_flow_clear(&facet->nf_flow);
4286 facet->tcp_flags = 0;
4289 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4290 * Returns it if found, otherwise a null pointer.
4292 * 'hash' must be the return value of flow_hash(flow, 0).
4294 * The returned facet might need revalidation; use facet_lookup_valid()
4295 * instead if that is important. */
4296 static struct facet *
4297 facet_find(struct ofproto_dpif *ofproto,
4298 const struct flow *flow, uint32_t hash)
4300 struct facet *facet;
4302 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4303 if (flow_equal(flow, &facet->flow)) {
4311 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4312 * Returns it if found, otherwise a null pointer.
4314 * 'hash' must be the return value of flow_hash(flow, 0).
4316 * The returned facet is guaranteed to be valid. */
4317 static struct facet *
4318 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4321 struct facet *facet;
4323 facet = facet_find(ofproto, flow, hash);
4325 && (ofproto->need_revalidate
4326 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4327 facet_revalidate(facet);
4334 subfacet_path_to_string(enum subfacet_path path)
4337 case SF_NOT_INSTALLED:
4338 return "not installed";
4340 return "in fast path";
4342 return "in slow path";
4348 /* Returns the path in which a subfacet should be installed if its 'slow'
4349 * member has the specified value. */
4350 static enum subfacet_path
4351 subfacet_want_path(enum slow_path_reason slow)
4353 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4356 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4357 * supposing that its actions have been recalculated as 'want_actions' and that
4358 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4360 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4361 const struct ofpbuf *want_actions)
4363 enum subfacet_path want_path = subfacet_want_path(slow);
4364 return (want_path != subfacet->path
4365 || (want_path == SF_FAST_PATH
4366 && (subfacet->actions_len != want_actions->size
4367 || memcmp(subfacet->actions, want_actions->data,
4368 subfacet->actions_len))));
4372 facet_check_consistency(struct facet *facet)
4374 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4378 uint64_t odp_actions_stub[1024 / 8];
4379 struct ofpbuf odp_actions;
4381 struct rule_dpif *rule;
4382 struct subfacet *subfacet;
4383 bool may_log = false;
4386 /* Check the rule for consistency. */
4387 rule = rule_dpif_lookup(ofproto, &facet->flow);
4388 ok = rule == facet->rule;
4390 may_log = !VLOG_DROP_WARN(&rl);
4395 flow_format(&s, &facet->flow);
4396 ds_put_format(&s, ": facet associated with wrong rule (was "
4397 "table=%"PRIu8",", facet->rule->up.table_id);
4398 cls_rule_format(&facet->rule->up.cr, &s);
4399 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4401 cls_rule_format(&rule->up.cr, &s);
4402 ds_put_char(&s, ')');
4404 VLOG_WARN("%s", ds_cstr(&s));
4409 /* Check the datapath actions for consistency. */
4410 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4411 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4412 enum subfacet_path want_path;
4413 struct odputil_keybuf keybuf;
4414 struct action_xlate_ctx ctx;
4418 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4419 subfacet->initial_tci, rule, 0, NULL);
4420 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4423 if (subfacet->path == SF_NOT_INSTALLED) {
4424 /* This only happens if the datapath reported an error when we
4425 * tried to install the flow. Don't flag another error here. */
4429 want_path = subfacet_want_path(subfacet->slow);
4430 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4431 /* The actions for slow-path flows may legitimately vary from one
4432 * packet to the next. We're done. */
4436 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4440 /* Inconsistency! */
4442 may_log = !VLOG_DROP_WARN(&rl);
4446 /* Rate-limited, skip reporting. */
4451 subfacet_get_key(subfacet, &keybuf, &key);
4452 odp_flow_key_format(key.data, key.size, &s);
4454 ds_put_cstr(&s, ": inconsistency in subfacet");
4455 if (want_path != subfacet->path) {
4456 enum odp_key_fitness fitness = subfacet->key_fitness;
4458 ds_put_format(&s, " (%s, fitness=%s)",
4459 subfacet_path_to_string(subfacet->path),
4460 odp_key_fitness_to_string(fitness));
4461 ds_put_format(&s, " (should have been %s)",
4462 subfacet_path_to_string(want_path));
4463 } else if (want_path == SF_FAST_PATH) {
4464 ds_put_cstr(&s, " (actions were: ");
4465 format_odp_actions(&s, subfacet->actions,
4466 subfacet->actions_len);
4467 ds_put_cstr(&s, ") (correct actions: ");
4468 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4469 ds_put_char(&s, ')');
4471 ds_put_cstr(&s, " (actions: ");
4472 format_odp_actions(&s, subfacet->actions,
4473 subfacet->actions_len);
4474 ds_put_char(&s, ')');
4476 VLOG_WARN("%s", ds_cstr(&s));
4479 ofpbuf_uninit(&odp_actions);
4484 /* Re-searches the classifier for 'facet':
4486 * - If the rule found is different from 'facet''s current rule, moves
4487 * 'facet' to the new rule and recompiles its actions.
4489 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4490 * where it is and recompiles its actions anyway. */
4492 facet_revalidate(struct facet *facet)
4494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4496 struct nlattr *odp_actions;
4499 struct actions *new_actions;
4501 struct action_xlate_ctx ctx;
4502 uint64_t odp_actions_stub[1024 / 8];
4503 struct ofpbuf odp_actions;
4505 struct rule_dpif *new_rule;
4506 struct subfacet *subfacet;
4509 COVERAGE_INC(facet_revalidate);
4511 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4513 /* Calculate new datapath actions.
4515 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4516 * emit a NetFlow expiration and, if so, we need to have the old state
4517 * around to properly compose it. */
4519 /* If the datapath actions changed or the installability changed,
4520 * then we need to talk to the datapath. */
4523 memset(&ctx, 0, sizeof ctx);
4524 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4525 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4526 enum slow_path_reason slow;
4528 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4529 subfacet->initial_tci, new_rule, 0, NULL);
4530 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4533 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4534 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4535 struct dpif_flow_stats stats;
4537 subfacet_install(subfacet,
4538 odp_actions.data, odp_actions.size, &stats, slow);
4539 subfacet_update_stats(subfacet, &stats);
4542 new_actions = xcalloc(list_size(&facet->subfacets),
4543 sizeof *new_actions);
4545 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4547 new_actions[i].actions_len = odp_actions.size;
4552 ofpbuf_uninit(&odp_actions);
4555 facet_flush_stats(facet);
4558 /* Update 'facet' now that we've taken care of all the old state. */
4559 facet->tags = ctx.tags;
4560 facet->nf_flow.output_iface = ctx.nf_output_iface;
4561 facet->has_learn = ctx.has_learn;
4562 facet->has_normal = ctx.has_normal;
4563 facet->has_fin_timeout = ctx.has_fin_timeout;
4564 facet->mirrors = ctx.mirrors;
4567 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4568 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4570 if (new_actions && new_actions[i].odp_actions) {
4571 free(subfacet->actions);
4572 subfacet->actions = new_actions[i].odp_actions;
4573 subfacet->actions_len = new_actions[i].actions_len;
4579 if (facet->rule != new_rule) {
4580 COVERAGE_INC(facet_changed_rule);
4581 list_remove(&facet->list_node);
4582 list_push_back(&new_rule->facets, &facet->list_node);
4583 facet->rule = new_rule;
4584 facet->used = new_rule->up.created;
4585 facet->prev_used = facet->used;
4589 /* Updates 'facet''s used time. Caller is responsible for calling
4590 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4592 facet_update_time(struct facet *facet, long long int used)
4594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4595 if (used > facet->used) {
4597 ofproto_rule_update_used(&facet->rule->up, used);
4598 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4603 facet_reset_counters(struct facet *facet)
4605 facet->packet_count = 0;
4606 facet->byte_count = 0;
4607 facet->prev_packet_count = 0;
4608 facet->prev_byte_count = 0;
4609 facet->accounted_bytes = 0;
4613 facet_push_stats(struct facet *facet)
4615 struct dpif_flow_stats stats;
4617 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4618 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4619 ovs_assert(facet->used >= facet->prev_used);
4621 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4622 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4623 stats.used = facet->used;
4624 stats.tcp_flags = 0;
4626 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4627 facet->prev_packet_count = facet->packet_count;
4628 facet->prev_byte_count = facet->byte_count;
4629 facet->prev_used = facet->used;
4631 flow_push_stats(facet->rule, &facet->flow, &stats);
4633 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4634 facet->mirrors, stats.n_packets, stats.n_bytes);
4639 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4641 rule->packet_count += stats->n_packets;
4642 rule->byte_count += stats->n_bytes;
4643 ofproto_rule_update_used(&rule->up, stats->used);
4646 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4647 * 'rule''s actions and mirrors. */
4649 flow_push_stats(struct rule_dpif *rule,
4650 const struct flow *flow, const struct dpif_flow_stats *stats)
4652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4653 struct action_xlate_ctx ctx;
4655 ofproto_rule_update_used(&rule->up, stats->used);
4657 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4659 ctx.resubmit_stats = stats;
4660 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4661 rule->up.ofpacts_len);
4666 static struct subfacet *
4667 subfacet_find(struct ofproto_dpif *ofproto,
4668 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4669 const struct flow *flow)
4671 struct subfacet *subfacet;
4673 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4674 &ofproto->subfacets) {
4676 ? (subfacet->key_len == key_len
4677 && !memcmp(key, subfacet->key, key_len))
4678 : flow_equal(flow, &subfacet->facet->flow)) {
4686 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4687 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4688 * existing subfacet if there is one, otherwise creates and returns a
4691 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4692 * which case the caller must populate the actions with
4693 * subfacet_make_actions(). */
4694 static struct subfacet *
4695 subfacet_create(struct facet *facet, struct flow_miss *miss,
4698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4699 enum odp_key_fitness key_fitness = miss->key_fitness;
4700 const struct nlattr *key = miss->key;
4701 size_t key_len = miss->key_len;
4703 struct subfacet *subfacet;
4705 key_hash = odp_flow_key_hash(key, key_len);
4707 if (list_is_empty(&facet->subfacets)) {
4708 subfacet = &facet->one_subfacet;
4710 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4713 if (subfacet->facet == facet) {
4717 /* This shouldn't happen. */
4718 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4719 subfacet_destroy(subfacet);
4722 subfacet = xmalloc(sizeof *subfacet);
4725 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4726 list_push_back(&facet->subfacets, &subfacet->list_node);
4727 subfacet->facet = facet;
4728 subfacet->key_fitness = key_fitness;
4729 if (key_fitness != ODP_FIT_PERFECT) {
4730 subfacet->key = xmemdup(key, key_len);
4731 subfacet->key_len = key_len;
4733 subfacet->key = NULL;
4734 subfacet->key_len = 0;
4736 subfacet->used = now;
4737 subfacet->dp_packet_count = 0;
4738 subfacet->dp_byte_count = 0;
4739 subfacet->actions_len = 0;
4740 subfacet->actions = NULL;
4741 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4744 subfacet->path = SF_NOT_INSTALLED;
4745 subfacet->initial_tci = miss->initial_tci;
4746 subfacet->odp_in_port = miss->odp_in_port;
4751 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4752 * its facet within 'ofproto', and frees it. */
4754 subfacet_destroy__(struct subfacet *subfacet)
4756 struct facet *facet = subfacet->facet;
4757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4759 subfacet_uninstall(subfacet);
4760 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4761 list_remove(&subfacet->list_node);
4762 free(subfacet->key);
4763 free(subfacet->actions);
4764 if (subfacet != &facet->one_subfacet) {
4769 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4770 * last remaining subfacet in its facet destroys the facet too. */
4772 subfacet_destroy(struct subfacet *subfacet)
4774 struct facet *facet = subfacet->facet;
4776 if (list_is_singleton(&facet->subfacets)) {
4777 /* facet_remove() needs at least one subfacet (it will remove it). */
4778 facet_remove(facet);
4780 subfacet_destroy__(subfacet);
4785 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4786 struct subfacet **subfacets, int n)
4788 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4789 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4790 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4791 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4792 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4795 for (i = 0; i < n; i++) {
4796 ops[i].type = DPIF_OP_FLOW_DEL;
4797 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4798 ops[i].u.flow_del.key = keys[i].data;
4799 ops[i].u.flow_del.key_len = keys[i].size;
4800 ops[i].u.flow_del.stats = &stats[i];
4804 dpif_operate(ofproto->backer->dpif, opsp, n);
4805 for (i = 0; i < n; i++) {
4806 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4807 subfacets[i]->path = SF_NOT_INSTALLED;
4808 subfacet_destroy(subfacets[i]);
4812 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4813 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4814 * for use as temporary storage. */
4816 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4820 if (!subfacet->key) {
4821 struct flow *flow = &subfacet->facet->flow;
4823 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4824 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
4826 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4830 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4831 * Translates the actions into 'odp_actions', which the caller must have
4832 * initialized and is responsible for uninitializing. */
4834 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4835 struct ofpbuf *odp_actions)
4837 struct facet *facet = subfacet->facet;
4838 struct rule_dpif *rule = facet->rule;
4839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4841 struct action_xlate_ctx ctx;
4843 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4845 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4846 facet->tags = ctx.tags;
4847 facet->has_learn = ctx.has_learn;
4848 facet->has_normal = ctx.has_normal;
4849 facet->has_fin_timeout = ctx.has_fin_timeout;
4850 facet->nf_flow.output_iface = ctx.nf_output_iface;
4851 facet->mirrors = ctx.mirrors;
4853 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4854 if (subfacet->actions_len != odp_actions->size
4855 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4856 free(subfacet->actions);
4857 subfacet->actions_len = odp_actions->size;
4858 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4862 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4863 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4864 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4865 * since 'subfacet' was last updated.
4867 * Returns 0 if successful, otherwise a positive errno value. */
4869 subfacet_install(struct subfacet *subfacet,
4870 const struct nlattr *actions, size_t actions_len,
4871 struct dpif_flow_stats *stats,
4872 enum slow_path_reason slow)
4874 struct facet *facet = subfacet->facet;
4875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4876 enum subfacet_path path = subfacet_want_path(slow);
4877 uint64_t slow_path_stub[128 / 8];
4878 struct odputil_keybuf keybuf;
4879 enum dpif_flow_put_flags flags;
4883 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4885 flags |= DPIF_FP_ZERO_STATS;
4888 if (path == SF_SLOW_PATH) {
4889 compose_slow_path(ofproto, &facet->flow, slow,
4890 slow_path_stub, sizeof slow_path_stub,
4891 &actions, &actions_len);
4894 subfacet_get_key(subfacet, &keybuf, &key);
4895 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4896 actions, actions_len, stats);
4899 subfacet_reset_dp_stats(subfacet, stats);
4903 subfacet->path = path;
4909 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4911 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4912 stats, subfacet->slow);
4915 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4917 subfacet_uninstall(struct subfacet *subfacet)
4919 if (subfacet->path != SF_NOT_INSTALLED) {
4920 struct rule_dpif *rule = subfacet->facet->rule;
4921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4922 struct odputil_keybuf keybuf;
4923 struct dpif_flow_stats stats;
4927 subfacet_get_key(subfacet, &keybuf, &key);
4928 error = dpif_flow_del(ofproto->backer->dpif,
4929 key.data, key.size, &stats);
4930 subfacet_reset_dp_stats(subfacet, &stats);
4932 subfacet_update_stats(subfacet, &stats);
4934 subfacet->path = SF_NOT_INSTALLED;
4936 ovs_assert(subfacet->dp_packet_count == 0);
4937 ovs_assert(subfacet->dp_byte_count == 0);
4941 /* Resets 'subfacet''s datapath statistics counters. This should be called
4942 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4943 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4944 * was reset in the datapath. 'stats' will be modified to include only
4945 * statistics new since 'subfacet' was last updated. */
4947 subfacet_reset_dp_stats(struct subfacet *subfacet,
4948 struct dpif_flow_stats *stats)
4951 && subfacet->dp_packet_count <= stats->n_packets
4952 && subfacet->dp_byte_count <= stats->n_bytes) {
4953 stats->n_packets -= subfacet->dp_packet_count;
4954 stats->n_bytes -= subfacet->dp_byte_count;
4957 subfacet->dp_packet_count = 0;
4958 subfacet->dp_byte_count = 0;
4961 /* Updates 'subfacet''s used time. The caller is responsible for calling
4962 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4964 subfacet_update_time(struct subfacet *subfacet, long long int used)
4966 if (used > subfacet->used) {
4967 subfacet->used = used;
4968 facet_update_time(subfacet->facet, used);
4972 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4974 * Because of the meaning of a subfacet's counters, it only makes sense to do
4975 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4976 * represents a packet that was sent by hand or if it represents statistics
4977 * that have been cleared out of the datapath. */
4979 subfacet_update_stats(struct subfacet *subfacet,
4980 const struct dpif_flow_stats *stats)
4982 if (stats->n_packets || stats->used > subfacet->used) {
4983 struct facet *facet = subfacet->facet;
4985 subfacet_update_time(subfacet, stats->used);
4986 facet->packet_count += stats->n_packets;
4987 facet->byte_count += stats->n_bytes;
4988 facet->tcp_flags |= stats->tcp_flags;
4989 facet_push_stats(facet);
4990 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4996 static struct rule_dpif *
4997 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4999 struct rule_dpif *rule;
5001 rule = rule_dpif_lookup__(ofproto, flow, 0);
5006 return rule_dpif_miss_rule(ofproto, flow);
5009 static struct rule_dpif *
5010 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5013 struct cls_rule *cls_rule;
5014 struct classifier *cls;
5016 if (table_id >= N_TABLES) {
5020 cls = &ofproto->up.tables[table_id].cls;
5021 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5022 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5023 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5024 * are unavailable. */
5025 struct flow ofpc_normal_flow = *flow;
5026 ofpc_normal_flow.tp_src = htons(0);
5027 ofpc_normal_flow.tp_dst = htons(0);
5028 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5030 cls_rule = classifier_lookup(cls, flow);
5032 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5035 static struct rule_dpif *
5036 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5038 struct ofport_dpif *port;
5040 port = get_ofp_port(ofproto, flow->in_port);
5042 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5043 return ofproto->miss_rule;
5046 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5047 return ofproto->no_packet_in_rule;
5049 return ofproto->miss_rule;
5053 complete_operation(struct rule_dpif *rule)
5055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5057 rule_invalidate(rule);
5059 struct dpif_completion *c = xmalloc(sizeof *c);
5060 c->op = rule->up.pending;
5061 list_push_back(&ofproto->completions, &c->list_node);
5063 ofoperation_complete(rule->up.pending, 0);
5067 static struct rule *
5070 struct rule_dpif *rule = xmalloc(sizeof *rule);
5075 rule_dealloc(struct rule *rule_)
5077 struct rule_dpif *rule = rule_dpif_cast(rule_);
5082 rule_construct(struct rule *rule_)
5084 struct rule_dpif *rule = rule_dpif_cast(rule_);
5085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5086 struct rule_dpif *victim;
5089 rule->packet_count = 0;
5090 rule->byte_count = 0;
5092 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5093 if (victim && !list_is_empty(&victim->facets)) {
5094 struct facet *facet;
5096 rule->facets = victim->facets;
5097 list_moved(&rule->facets);
5098 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5099 /* XXX: We're only clearing our local counters here. It's possible
5100 * that quite a few packets are unaccounted for in the datapath
5101 * statistics. These will be accounted to the new rule instead of
5102 * cleared as required. This could be fixed by clearing out the
5103 * datapath statistics for this facet, but currently it doesn't
5105 facet_reset_counters(facet);
5109 /* Must avoid list_moved() in this case. */
5110 list_init(&rule->facets);
5113 table_id = rule->up.table_id;
5115 rule->tag = victim->tag;
5116 } else if (table_id == 0) {
5121 miniflow_expand(&rule->up.cr.match.flow, &flow);
5122 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5123 ofproto->tables[table_id].basis);
5126 complete_operation(rule);
5131 rule_destruct(struct rule *rule_)
5133 struct rule_dpif *rule = rule_dpif_cast(rule_);
5134 struct facet *facet, *next_facet;
5136 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5137 facet_revalidate(facet);
5140 complete_operation(rule);
5144 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5146 struct rule_dpif *rule = rule_dpif_cast(rule_);
5147 struct facet *facet;
5149 /* Start from historical data for 'rule' itself that are no longer tracked
5150 * in facets. This counts, for example, facets that have expired. */
5151 *packets = rule->packet_count;
5152 *bytes = rule->byte_count;
5154 /* Add any statistics that are tracked by facets. This includes
5155 * statistical data recently updated by ofproto_update_stats() as well as
5156 * stats for packets that were executed "by hand" via dpif_execute(). */
5157 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5158 *packets += facet->packet_count;
5159 *bytes += facet->byte_count;
5164 rule_execute(struct rule *rule_, const struct flow *flow,
5165 struct ofpbuf *packet)
5167 struct rule_dpif *rule = rule_dpif_cast(rule_);
5168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5170 struct dpif_flow_stats stats;
5172 struct action_xlate_ctx ctx;
5173 uint64_t odp_actions_stub[1024 / 8];
5174 struct ofpbuf odp_actions;
5176 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5177 rule_credit_stats(rule, &stats);
5179 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5180 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5181 rule, stats.tcp_flags, packet);
5182 ctx.resubmit_stats = &stats;
5183 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5185 execute_odp_actions(ofproto, flow, odp_actions.data,
5186 odp_actions.size, packet);
5188 ofpbuf_uninit(&odp_actions);
5194 rule_modify_actions(struct rule *rule_)
5196 struct rule_dpif *rule = rule_dpif_cast(rule_);
5198 complete_operation(rule);
5201 /* Sends 'packet' out 'ofport'.
5202 * May modify 'packet'.
5203 * Returns 0 if successful, otherwise a positive errno value. */
5205 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5207 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5208 struct ofpbuf key, odp_actions;
5209 struct odputil_keybuf keybuf;
5214 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5215 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5217 if (odp_port != ofport->odp_port) {
5218 eth_pop_vlan(packet);
5219 flow.vlan_tci = htons(0);
5222 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5223 odp_flow_key_from_flow(&key, &flow,
5224 ofp_port_to_odp_port(ofproto, flow.in_port));
5226 ofpbuf_init(&odp_actions, 32);
5227 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5229 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5230 error = dpif_execute(ofproto->backer->dpif,
5232 odp_actions.data, odp_actions.size,
5234 ofpbuf_uninit(&odp_actions);
5237 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5238 ofproto->up.name, odp_port, strerror(error));
5240 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5244 /* OpenFlow to datapath action translation. */
5246 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5247 struct action_xlate_ctx *);
5248 static void xlate_normal(struct action_xlate_ctx *);
5250 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5251 * The action will state 'slow' as the reason that the action is in the slow
5252 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5253 * dump-flows" output to see why a flow is in the slow path.)
5255 * The 'stub_size' bytes in 'stub' will be used to store the action.
5256 * 'stub_size' must be large enough for the action.
5258 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5261 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5262 enum slow_path_reason slow,
5263 uint64_t *stub, size_t stub_size,
5264 const struct nlattr **actionsp, size_t *actions_lenp)
5266 union user_action_cookie cookie;
5269 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5270 cookie.slow_path.unused = 0;
5271 cookie.slow_path.reason = slow;
5273 ofpbuf_use_stack(&buf, stub, stub_size);
5274 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5275 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5276 odp_put_userspace_action(pid, &cookie, &buf);
5278 put_userspace_action(ofproto, &buf, flow, &cookie);
5280 *actionsp = buf.data;
5281 *actions_lenp = buf.size;
5285 put_userspace_action(const struct ofproto_dpif *ofproto,
5286 struct ofpbuf *odp_actions,
5287 const struct flow *flow,
5288 const union user_action_cookie *cookie)
5292 pid = dpif_port_get_pid(ofproto->backer->dpif,
5293 ofp_port_to_odp_port(ofproto, flow->in_port));
5295 return odp_put_userspace_action(pid, cookie, odp_actions);
5299 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5300 ovs_be16 vlan_tci, uint32_t odp_port,
5301 unsigned int n_outputs, union user_action_cookie *cookie)
5305 cookie->type = USER_ACTION_COOKIE_SFLOW;
5306 cookie->sflow.vlan_tci = vlan_tci;
5308 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5309 * port information") for the interpretation of cookie->output. */
5310 switch (n_outputs) {
5312 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5313 cookie->sflow.output = 0x40000000 | 256;
5317 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5319 cookie->sflow.output = ifindex;
5324 /* 0x80000000 means "multiple output ports. */
5325 cookie->sflow.output = 0x80000000 | n_outputs;
5330 /* Compose SAMPLE action for sFlow. */
5332 compose_sflow_action(const struct ofproto_dpif *ofproto,
5333 struct ofpbuf *odp_actions,
5334 const struct flow *flow,
5337 uint32_t probability;
5338 union user_action_cookie cookie;
5339 size_t sample_offset, actions_offset;
5342 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5346 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5348 /* Number of packets out of UINT_MAX to sample. */
5349 probability = dpif_sflow_get_probability(ofproto->sflow);
5350 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5352 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5353 compose_sflow_cookie(ofproto, htons(0), odp_port,
5354 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5355 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5357 nl_msg_end_nested(odp_actions, actions_offset);
5358 nl_msg_end_nested(odp_actions, sample_offset);
5359 return cookie_offset;
5362 /* SAMPLE action must be first action in any given list of actions.
5363 * At this point we do not have all information required to build it. So try to
5364 * build sample action as complete as possible. */
5366 add_sflow_action(struct action_xlate_ctx *ctx)
5368 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5370 &ctx->flow, OVSP_NONE);
5371 ctx->sflow_odp_port = 0;
5372 ctx->sflow_n_outputs = 0;
5375 /* Fix SAMPLE action according to data collected while composing ODP actions.
5376 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5377 * USERSPACE action's user-cookie which is required for sflow. */
5379 fix_sflow_action(struct action_xlate_ctx *ctx)
5381 const struct flow *base = &ctx->base_flow;
5382 union user_action_cookie *cookie;
5384 if (!ctx->user_cookie_offset) {
5388 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5390 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5392 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5393 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5397 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5400 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5401 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5402 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5403 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5404 struct priority_to_dscp *pdscp;
5408 xlate_report(ctx, "Nonexistent output port");
5410 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5411 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5413 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5414 xlate_report(ctx, "STP not in forwarding state, skipping output");
5418 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5420 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5421 ctx->flow.nw_tos |= pdscp->dscp;
5424 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5425 ctx->flow.vlan_tci);
5426 if (out_port != odp_port) {
5427 ctx->flow.vlan_tci = htons(0);
5429 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5430 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5432 ctx->sflow_odp_port = odp_port;
5433 ctx->sflow_n_outputs++;
5434 ctx->nf_output_iface = ofp_port;
5435 ctx->flow.vlan_tci = flow_vlan_tci;
5436 ctx->flow.nw_tos = flow_nw_tos;
5440 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5442 compose_output_action__(ctx, ofp_port, true);
5446 xlate_table_action(struct action_xlate_ctx *ctx,
5447 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5449 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5450 struct ofproto_dpif *ofproto = ctx->ofproto;
5451 struct rule_dpif *rule;
5452 uint16_t old_in_port;
5453 uint8_t old_table_id;
5455 old_table_id = ctx->table_id;
5456 ctx->table_id = table_id;
5458 /* Look up a flow with 'in_port' as the input port. */
5459 old_in_port = ctx->flow.in_port;
5460 ctx->flow.in_port = in_port;
5461 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5464 if (table_id > 0 && table_id < N_TABLES) {
5465 struct table_dpif *table = &ofproto->tables[table_id];
5466 if (table->other_table) {
5467 ctx->tags |= (rule && rule->tag
5469 : rule_calculate_tag(&ctx->flow,
5470 &table->other_table->mask,
5475 /* Restore the original input port. Otherwise OFPP_NORMAL and
5476 * OFPP_IN_PORT will have surprising behavior. */
5477 ctx->flow.in_port = old_in_port;
5479 if (ctx->resubmit_hook) {
5480 ctx->resubmit_hook(ctx, rule);
5483 if (rule == NULL && may_packet_in) {
5485 * check if table configuration flags
5486 * OFPTC_TABLE_MISS_CONTROLLER, default.
5487 * OFPTC_TABLE_MISS_CONTINUE,
5488 * OFPTC_TABLE_MISS_DROP
5489 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5491 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5495 struct rule_dpif *old_rule = ctx->rule;
5497 if (ctx->resubmit_stats) {
5498 rule_credit_stats(rule, ctx->resubmit_stats);
5503 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5504 ctx->rule = old_rule;
5508 ctx->table_id = old_table_id;
5510 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5512 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5513 MAX_RESUBMIT_RECURSION);
5514 ctx->max_resubmit_trigger = true;
5519 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5520 const struct ofpact_resubmit *resubmit)
5525 in_port = resubmit->in_port;
5526 if (in_port == OFPP_IN_PORT) {
5527 in_port = ctx->flow.in_port;
5530 table_id = resubmit->table_id;
5531 if (table_id == 255) {
5532 table_id = ctx->table_id;
5535 xlate_table_action(ctx, in_port, table_id, false);
5539 flood_packets(struct action_xlate_ctx *ctx, bool all)
5541 struct ofport_dpif *ofport;
5543 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5544 uint16_t ofp_port = ofport->up.ofp_port;
5546 if (ofp_port == ctx->flow.in_port) {
5551 compose_output_action__(ctx, ofp_port, false);
5552 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5553 compose_output_action(ctx, ofp_port);
5557 ctx->nf_output_iface = NF_OUT_FLOOD;
5561 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5562 enum ofp_packet_in_reason reason,
5563 uint16_t controller_id)
5565 struct ofputil_packet_in pin;
5566 struct ofpbuf *packet;
5568 ctx->slow |= SLOW_CONTROLLER;
5573 packet = ofpbuf_clone(ctx->packet);
5575 if (packet->l2 && packet->l3) {
5576 struct eth_header *eh;
5578 eth_pop_vlan(packet);
5581 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5582 * LLC frame. Calculating the Ethernet type of these frames is more
5583 * trouble than seems appropriate for a simple assertion. */
5584 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5585 || eh->eth_type == ctx->flow.dl_type);
5587 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5588 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5590 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5591 eth_push_vlan(packet, ctx->flow.vlan_tci);
5595 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5596 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5597 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5601 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5602 packet_set_tcp_port(packet, ctx->flow.tp_src,
5604 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5605 packet_set_udp_port(packet, ctx->flow.tp_src,
5612 pin.packet = packet->data;
5613 pin.packet_len = packet->size;
5614 pin.reason = reason;
5615 pin.controller_id = controller_id;
5616 pin.table_id = ctx->table_id;
5617 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5620 flow_get_metadata(&ctx->flow, &pin.fmd);
5622 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5623 ofpbuf_delete(packet);
5627 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5629 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5630 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5634 if (ctx->flow.nw_ttl > 1) {
5640 for (i = 0; i < ids->n_controllers; i++) {
5641 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5645 /* Stop processing for current table. */
5651 xlate_output_action(struct action_xlate_ctx *ctx,
5652 uint16_t port, uint16_t max_len, bool may_packet_in)
5654 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5656 ctx->nf_output_iface = NF_OUT_DROP;
5660 compose_output_action(ctx, ctx->flow.in_port);
5663 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5669 flood_packets(ctx, false);
5672 flood_packets(ctx, true);
5674 case OFPP_CONTROLLER:
5675 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5681 if (port != ctx->flow.in_port) {
5682 compose_output_action(ctx, port);
5684 xlate_report(ctx, "skipping output to input port");
5689 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5690 ctx->nf_output_iface = NF_OUT_FLOOD;
5691 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5692 ctx->nf_output_iface = prev_nf_output_iface;
5693 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5694 ctx->nf_output_iface != NF_OUT_FLOOD) {
5695 ctx->nf_output_iface = NF_OUT_MULTI;
5700 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5701 const struct ofpact_output_reg *or)
5703 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5704 if (port <= UINT16_MAX) {
5705 xlate_output_action(ctx, port, or->max_len, false);
5710 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5711 const struct ofpact_enqueue *enqueue)
5713 uint16_t ofp_port = enqueue->port;
5714 uint32_t queue_id = enqueue->queue;
5715 uint32_t flow_priority, priority;
5718 /* Translate queue to priority. */
5719 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5720 queue_id, &priority);
5722 /* Fall back to ordinary output action. */
5723 xlate_output_action(ctx, enqueue->port, 0, false);
5727 /* Check output port. */
5728 if (ofp_port == OFPP_IN_PORT) {
5729 ofp_port = ctx->flow.in_port;
5730 } else if (ofp_port == ctx->flow.in_port) {
5734 /* Add datapath actions. */
5735 flow_priority = ctx->flow.skb_priority;
5736 ctx->flow.skb_priority = priority;
5737 compose_output_action(ctx, ofp_port);
5738 ctx->flow.skb_priority = flow_priority;
5740 /* Update NetFlow output port. */
5741 if (ctx->nf_output_iface == NF_OUT_DROP) {
5742 ctx->nf_output_iface = ofp_port;
5743 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5744 ctx->nf_output_iface = NF_OUT_MULTI;
5749 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5751 uint32_t skb_priority;
5753 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5754 queue_id, &skb_priority)) {
5755 ctx->flow.skb_priority = skb_priority;
5757 /* Couldn't translate queue to a priority. Nothing to do. A warning
5758 * has already been logged. */
5762 struct xlate_reg_state {
5768 xlate_autopath(struct action_xlate_ctx *ctx,
5769 const struct ofpact_autopath *ap)
5771 uint16_t ofp_port = ap->port;
5772 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5774 if (!port || !port->bundle) {
5775 ofp_port = OFPP_NONE;
5776 } else if (port->bundle->bond) {
5777 /* Autopath does not support VLAN hashing. */
5778 struct ofport_dpif *slave = bond_choose_output_slave(
5779 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5781 ofp_port = slave->up.ofp_port;
5784 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5788 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5790 struct ofproto_dpif *ofproto = ofproto_;
5791 struct ofport_dpif *port;
5801 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5804 port = get_ofp_port(ofproto, ofp_port);
5805 return port ? port->may_enable : false;
5810 xlate_bundle_action(struct action_xlate_ctx *ctx,
5811 const struct ofpact_bundle *bundle)
5815 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5816 if (bundle->dst.field) {
5817 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5819 xlate_output_action(ctx, port, 0, false);
5824 xlate_learn_action(struct action_xlate_ctx *ctx,
5825 const struct ofpact_learn *learn)
5827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5828 struct ofputil_flow_mod fm;
5829 uint64_t ofpacts_stub[1024 / 8];
5830 struct ofpbuf ofpacts;
5833 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5834 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5836 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5837 if (error && !VLOG_DROP_WARN(&rl)) {
5838 VLOG_WARN("learning action failed to modify flow table (%s)",
5839 ofperr_get_name(error));
5842 ofpbuf_uninit(&ofpacts);
5845 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5846 * means "infinite". */
5848 reduce_timeout(uint16_t max, uint16_t *timeout)
5850 if (max && (!*timeout || *timeout > max)) {
5856 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5857 const struct ofpact_fin_timeout *oft)
5859 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5860 struct rule_dpif *rule = ctx->rule;
5862 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5863 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5868 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5870 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5871 ? OFPUTIL_PC_NO_RECV_STP
5872 : OFPUTIL_PC_NO_RECV)) {
5876 /* Only drop packets here if both forwarding and learning are
5877 * disabled. If just learning is enabled, we need to have
5878 * OFPP_NORMAL and the learning action have a look at the packet
5879 * before we can drop it. */
5880 if (!stp_forward_in_state(port->stp_state)
5881 && !stp_learn_in_state(port->stp_state)) {
5889 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5890 struct action_xlate_ctx *ctx)
5892 const struct ofport_dpif *port;
5893 bool was_evictable = true;
5894 const struct ofpact *a;
5896 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5897 if (port && !may_receive(port, ctx)) {
5898 /* Drop this flow. */
5903 /* Don't let the rule we're working on get evicted underneath us. */
5904 was_evictable = ctx->rule->up.evictable;
5905 ctx->rule->up.evictable = false;
5907 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5908 struct ofpact_controller *controller;
5909 const struct ofpact_metadata *metadata;
5917 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5918 ofpact_get_OUTPUT(a)->max_len, true);
5921 case OFPACT_CONTROLLER:
5922 controller = ofpact_get_CONTROLLER(a);
5923 execute_controller_action(ctx, controller->max_len,
5925 controller->controller_id);
5928 case OFPACT_ENQUEUE:
5929 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5932 case OFPACT_SET_VLAN_VID:
5933 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5934 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5938 case OFPACT_SET_VLAN_PCP:
5939 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5940 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5945 case OFPACT_STRIP_VLAN:
5946 ctx->flow.vlan_tci = htons(0);
5949 case OFPACT_PUSH_VLAN:
5950 /* XXX 802.1AD(QinQ) */
5951 ctx->flow.vlan_tci = htons(VLAN_CFI);
5954 case OFPACT_SET_ETH_SRC:
5955 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5959 case OFPACT_SET_ETH_DST:
5960 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5964 case OFPACT_SET_IPV4_SRC:
5965 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5968 case OFPACT_SET_IPV4_DST:
5969 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5972 case OFPACT_SET_IPV4_DSCP:
5973 /* OpenFlow 1.0 only supports IPv4. */
5974 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5975 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5976 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5980 case OFPACT_SET_L4_SRC_PORT:
5981 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5984 case OFPACT_SET_L4_DST_PORT:
5985 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5988 case OFPACT_RESUBMIT:
5989 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5992 case OFPACT_SET_TUNNEL:
5993 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5996 case OFPACT_SET_QUEUE:
5997 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6000 case OFPACT_POP_QUEUE:
6001 ctx->flow.skb_priority = ctx->orig_skb_priority;
6004 case OFPACT_REG_MOVE:
6005 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6008 case OFPACT_REG_LOAD:
6009 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6012 case OFPACT_DEC_TTL:
6013 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6019 /* Nothing to do. */
6022 case OFPACT_MULTIPATH:
6023 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6026 case OFPACT_AUTOPATH:
6027 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6031 ctx->ofproto->has_bundle_action = true;
6032 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6035 case OFPACT_OUTPUT_REG:
6036 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6040 ctx->has_learn = true;
6041 if (ctx->may_learn) {
6042 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6050 case OFPACT_FIN_TIMEOUT:
6051 ctx->has_fin_timeout = true;
6052 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6055 case OFPACT_CLEAR_ACTIONS:
6057 * Nothing to do because writa-actions is not supported for now.
6058 * When writa-actions is supported, clear-actions also must
6059 * be supported at the same time.
6063 case OFPACT_WRITE_METADATA:
6064 metadata = ofpact_get_WRITE_METADATA(a);
6065 ctx->flow.metadata &= ~metadata->mask;
6066 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6069 case OFPACT_GOTO_TABLE: {
6070 /* XXX remove recursion */
6071 /* It is assumed that goto-table is last action */
6072 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6073 ovs_assert(ctx->table_id < ogt->table_id);
6074 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6081 /* We've let OFPP_NORMAL and the learning action look at the packet,
6082 * so drop it now if forwarding is disabled. */
6083 if (port && !stp_forward_in_state(port->stp_state)) {
6084 ofpbuf_clear(ctx->odp_actions);
6085 add_sflow_action(ctx);
6088 ctx->rule->up.evictable = was_evictable;
6093 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6094 struct ofproto_dpif *ofproto, const struct flow *flow,
6095 ovs_be16 initial_tci, struct rule_dpif *rule,
6096 uint8_t tcp_flags, const struct ofpbuf *packet)
6098 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6100 /* Flow initialization rules:
6101 * - 'base_flow' must match the kernel's view of the packet at the
6102 * time that action processing starts. 'flow' represents any
6103 * transformations we wish to make through actions.
6104 * - By default 'base_flow' and 'flow' are the same since the input
6105 * packet matches the output before any actions are applied.
6106 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6107 * of the received packet as seen by the kernel. If we later output
6108 * to another device without any modifications this will cause us to
6109 * insert a new tag since the original one was stripped off by the
6111 * - Tunnel 'flow' is largely cleared when transitioning between
6112 * the input and output stages since it does not make sense to output
6113 * a packet with the exact headers that it was received with (i.e.
6114 * the destination IP is us). The one exception is the tun_id, which
6115 * is preserved to allow use in later resubmit lookups and loads into
6117 * - Tunnel 'base_flow' is completely cleared since that is what the
6118 * kernel does. If we wish to maintain the original values an action
6119 * needs to be generated. */
6121 ctx->ofproto = ofproto;
6123 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6124 ctx->base_flow = ctx->flow;
6125 ctx->base_flow.vlan_tci = initial_tci;
6126 ctx->flow.tunnel.tun_id = initial_tun_id;
6128 ctx->packet = packet;
6129 ctx->may_learn = packet != NULL;
6130 ctx->tcp_flags = tcp_flags;
6131 ctx->resubmit_hook = NULL;
6132 ctx->report_hook = NULL;
6133 ctx->resubmit_stats = NULL;
6136 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6137 * into datapath actions in 'odp_actions', using 'ctx'. */
6139 xlate_actions(struct action_xlate_ctx *ctx,
6140 const struct ofpact *ofpacts, size_t ofpacts_len,
6141 struct ofpbuf *odp_actions)
6143 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6144 * that in the future we always keep a copy of the original flow for
6145 * tracing purposes. */
6146 static bool hit_resubmit_limit;
6148 enum slow_path_reason special;
6150 COVERAGE_INC(ofproto_dpif_xlate);
6152 ofpbuf_clear(odp_actions);
6153 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6155 ctx->odp_actions = odp_actions;
6158 ctx->has_learn = false;
6159 ctx->has_normal = false;
6160 ctx->has_fin_timeout = false;
6161 ctx->nf_output_iface = NF_OUT_DROP;
6164 ctx->max_resubmit_trigger = false;
6165 ctx->orig_skb_priority = ctx->flow.skb_priority;
6169 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6170 /* Do this conditionally because the copy is expensive enough that it
6171 * shows up in profiles.
6173 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6174 * believe that I wasn't using it without initializing it if I kept it
6175 * in a local variable. */
6176 ctx->orig_flow = ctx->flow;
6179 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6180 switch (ctx->ofproto->up.frag_handling) {
6181 case OFPC_FRAG_NORMAL:
6182 /* We must pretend that transport ports are unavailable. */
6183 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6184 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6187 case OFPC_FRAG_DROP:
6190 case OFPC_FRAG_REASM:
6193 case OFPC_FRAG_NX_MATCH:
6194 /* Nothing to do. */
6197 case OFPC_INVALID_TTL_TO_CONTROLLER:
6202 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6204 ctx->slow |= special;
6206 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6207 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6208 uint32_t local_odp_port;
6210 add_sflow_action(ctx);
6211 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6213 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6214 if (!hit_resubmit_limit) {
6215 /* We didn't record the original flow. Make sure we do from
6217 hit_resubmit_limit = true;
6218 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6219 struct ds ds = DS_EMPTY_INITIALIZER;
6221 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6223 VLOG_ERR("Trace triggered by excessive resubmit "
6224 "recursion:\n%s", ds_cstr(&ds));
6229 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6230 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6232 ctx->odp_actions->data,
6233 ctx->odp_actions->size)) {
6234 ctx->slow |= SLOW_IN_BAND;
6236 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6238 compose_output_action(ctx, OFPP_LOCAL);
6241 if (ctx->ofproto->has_mirrors) {
6242 add_mirror_actions(ctx, &ctx->orig_flow);
6244 fix_sflow_action(ctx);
6248 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6249 * into datapath actions, using 'ctx', and discards the datapath actions. */
6251 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6252 const struct ofpact *ofpacts,
6255 uint64_t odp_actions_stub[1024 / 8];
6256 struct ofpbuf odp_actions;
6258 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6259 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6260 ofpbuf_uninit(&odp_actions);
6264 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6266 if (ctx->report_hook) {
6267 ctx->report_hook(ctx, s);
6271 /* OFPP_NORMAL implementation. */
6273 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6275 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6276 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6277 * the bundle on which the packet was received, returns the VLAN to which the
6280 * Both 'vid' and the return value are in the range 0...4095. */
6282 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6284 switch (in_bundle->vlan_mode) {
6285 case PORT_VLAN_ACCESS:
6286 return in_bundle->vlan;
6289 case PORT_VLAN_TRUNK:
6292 case PORT_VLAN_NATIVE_UNTAGGED:
6293 case PORT_VLAN_NATIVE_TAGGED:
6294 return vid ? vid : in_bundle->vlan;
6301 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6302 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6305 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6306 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6309 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6311 /* Allow any VID on the OFPP_NONE port. */
6312 if (in_bundle == &ofpp_none_bundle) {
6316 switch (in_bundle->vlan_mode) {
6317 case PORT_VLAN_ACCESS:
6320 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6321 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6322 "packet received on port %s configured as VLAN "
6323 "%"PRIu16" access port",
6324 in_bundle->ofproto->up.name, vid,
6325 in_bundle->name, in_bundle->vlan);
6331 case PORT_VLAN_NATIVE_UNTAGGED:
6332 case PORT_VLAN_NATIVE_TAGGED:
6334 /* Port must always carry its native VLAN. */
6338 case PORT_VLAN_TRUNK:
6339 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6341 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6342 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6343 "received on port %s not configured for trunking "
6345 in_bundle->ofproto->up.name, vid,
6346 in_bundle->name, vid);
6358 /* Given 'vlan', the VLAN that a packet belongs to, and
6359 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6360 * that should be included in the 802.1Q header. (If the return value is 0,
6361 * then the 802.1Q header should only be included in the packet if there is a
6364 * Both 'vlan' and the return value are in the range 0...4095. */
6366 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6368 switch (out_bundle->vlan_mode) {
6369 case PORT_VLAN_ACCESS:
6372 case PORT_VLAN_TRUNK:
6373 case PORT_VLAN_NATIVE_TAGGED:
6376 case PORT_VLAN_NATIVE_UNTAGGED:
6377 return vlan == out_bundle->vlan ? 0 : vlan;
6385 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6388 struct ofport_dpif *port;
6390 ovs_be16 tci, old_tci;
6392 vid = output_vlan_to_vid(out_bundle, vlan);
6393 if (!out_bundle->bond) {
6394 port = ofbundle_get_a_port(out_bundle);
6396 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6399 /* No slaves enabled, so drop packet. */
6404 old_tci = ctx->flow.vlan_tci;
6406 if (tci || out_bundle->use_priority_tags) {
6407 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6409 tci |= htons(VLAN_CFI);
6412 ctx->flow.vlan_tci = tci;
6414 compose_output_action(ctx, port->up.ofp_port);
6415 ctx->flow.vlan_tci = old_tci;
6419 mirror_mask_ffs(mirror_mask_t mask)
6421 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6426 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6428 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6429 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6433 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6435 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6438 /* Returns an arbitrary interface within 'bundle'. */
6439 static struct ofport_dpif *
6440 ofbundle_get_a_port(const struct ofbundle *bundle)
6442 return CONTAINER_OF(list_front(&bundle->ports),
6443 struct ofport_dpif, bundle_node);
6447 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6449 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6453 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6455 struct ofproto_dpif *ofproto = ctx->ofproto;
6456 mirror_mask_t mirrors;
6457 struct ofbundle *in_bundle;
6460 const struct nlattr *a;
6463 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6464 ctx->packet != NULL, NULL);
6468 mirrors = in_bundle->src_mirrors;
6470 /* Drop frames on bundles reserved for mirroring. */
6471 if (in_bundle->mirror_out) {
6472 if (ctx->packet != NULL) {
6473 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6474 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6475 "%s, which is reserved exclusively for mirroring",
6476 ctx->ofproto->up.name, in_bundle->name);
6482 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6483 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6486 vlan = input_vid_to_vlan(in_bundle, vid);
6488 /* Look at the output ports to check for destination selections. */
6490 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6491 ctx->odp_actions->size) {
6492 enum ovs_action_attr type = nl_attr_type(a);
6493 struct ofport_dpif *ofport;
6495 if (type != OVS_ACTION_ATTR_OUTPUT) {
6499 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6500 if (ofport && ofport->bundle) {
6501 mirrors |= ofport->bundle->dst_mirrors;
6509 /* Restore the original packet before adding the mirror actions. */
6510 ctx->flow = *orig_flow;
6515 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6517 if (!vlan_is_mirrored(m, vlan)) {
6518 mirrors = zero_rightmost_1bit(mirrors);
6522 mirrors &= ~m->dup_mirrors;
6523 ctx->mirrors |= m->dup_mirrors;
6525 output_normal(ctx, m->out, vlan);
6526 } else if (vlan != m->out_vlan
6527 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6528 struct ofbundle *bundle;
6530 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6531 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6532 && !bundle->mirror_out) {
6533 output_normal(ctx, bundle, m->out_vlan);
6541 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6542 uint64_t packets, uint64_t bytes)
6548 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6551 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6554 /* In normal circumstances 'm' will not be NULL. However,
6555 * if mirrors are reconfigured, we can temporarily get out
6556 * of sync in facet_revalidate(). We could "correct" the
6557 * mirror list before reaching here, but doing that would
6558 * not properly account the traffic stats we've currently
6559 * accumulated for previous mirror configuration. */
6563 m->packet_count += packets;
6564 m->byte_count += bytes;
6568 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6569 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6570 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6572 is_gratuitous_arp(const struct flow *flow)
6574 return (flow->dl_type == htons(ETH_TYPE_ARP)
6575 && eth_addr_is_broadcast(flow->dl_dst)
6576 && (flow->nw_proto == ARP_OP_REPLY
6577 || (flow->nw_proto == ARP_OP_REQUEST
6578 && flow->nw_src == flow->nw_dst)));
6582 update_learning_table(struct ofproto_dpif *ofproto,
6583 const struct flow *flow, int vlan,
6584 struct ofbundle *in_bundle)
6586 struct mac_entry *mac;
6588 /* Don't learn the OFPP_NONE port. */
6589 if (in_bundle == &ofpp_none_bundle) {
6593 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6597 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6598 if (is_gratuitous_arp(flow)) {
6599 /* We don't want to learn from gratuitous ARP packets that are
6600 * reflected back over bond slaves so we lock the learning table. */
6601 if (!in_bundle->bond) {
6602 mac_entry_set_grat_arp_lock(mac);
6603 } else if (mac_entry_is_grat_arp_locked(mac)) {
6608 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6609 /* The log messages here could actually be useful in debugging,
6610 * so keep the rate limit relatively high. */
6611 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6612 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6613 "on port %s in VLAN %d",
6614 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6615 in_bundle->name, vlan);
6617 mac->port.p = in_bundle;
6618 tag_set_add(&ofproto->revalidate_set,
6619 mac_learning_changed(ofproto->ml, mac));
6623 static struct ofbundle *
6624 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6625 bool warn, struct ofport_dpif **in_ofportp)
6627 struct ofport_dpif *ofport;
6629 /* Find the port and bundle for the received packet. */
6630 ofport = get_ofp_port(ofproto, in_port);
6632 *in_ofportp = ofport;
6634 if (ofport && ofport->bundle) {
6635 return ofport->bundle;
6638 /* Special-case OFPP_NONE, which a controller may use as the ingress
6639 * port for traffic that it is sourcing. */
6640 if (in_port == OFPP_NONE) {
6641 return &ofpp_none_bundle;
6644 /* Odd. A few possible reasons here:
6646 * - We deleted a port but there are still a few packets queued up
6649 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6650 * we don't know about.
6652 * - The ofproto client didn't configure the port as part of a bundle.
6653 * This is particularly likely to happen if a packet was received on the
6654 * port after it was created, but before the client had a chance to
6655 * configure its bundle.
6658 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6660 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6661 "port %"PRIu16, ofproto->up.name, in_port);
6666 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6667 * dropped. Returns true if they may be forwarded, false if they should be
6670 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6671 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6673 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6674 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6675 * checked by input_vid_is_valid().
6677 * May also add tags to '*tags', although the current implementation only does
6678 * so in one special case.
6681 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6684 struct ofproto_dpif *ofproto = ctx->ofproto;
6685 struct flow *flow = &ctx->flow;
6686 struct ofbundle *in_bundle = in_port->bundle;
6688 /* Drop frames for reserved multicast addresses
6689 * only if forward_bpdu option is absent. */
6690 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6691 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6695 if (in_bundle->bond) {
6696 struct mac_entry *mac;
6698 switch (bond_check_admissibility(in_bundle->bond, in_port,
6699 flow->dl_dst, &ctx->tags)) {
6704 xlate_report(ctx, "bonding refused admissibility, dropping");
6707 case BV_DROP_IF_MOVED:
6708 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6709 if (mac && mac->port.p != in_bundle &&
6710 (!is_gratuitous_arp(flow)
6711 || mac_entry_is_grat_arp_locked(mac))) {
6712 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6724 xlate_normal(struct action_xlate_ctx *ctx)
6726 struct ofport_dpif *in_port;
6727 struct ofbundle *in_bundle;
6728 struct mac_entry *mac;
6732 ctx->has_normal = true;
6734 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6735 ctx->packet != NULL, &in_port);
6737 xlate_report(ctx, "no input bundle, dropping");
6741 /* Drop malformed frames. */
6742 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6743 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6744 if (ctx->packet != NULL) {
6745 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6746 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6747 "VLAN tag received on port %s",
6748 ctx->ofproto->up.name, in_bundle->name);
6750 xlate_report(ctx, "partial VLAN tag, dropping");
6754 /* Drop frames on bundles reserved for mirroring. */
6755 if (in_bundle->mirror_out) {
6756 if (ctx->packet != NULL) {
6757 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6758 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6759 "%s, which is reserved exclusively for mirroring",
6760 ctx->ofproto->up.name, in_bundle->name);
6762 xlate_report(ctx, "input port is mirror output port, dropping");
6767 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6768 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6769 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6772 vlan = input_vid_to_vlan(in_bundle, vid);
6774 /* Check other admissibility requirements. */
6775 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6779 /* Learn source MAC. */
6780 if (ctx->may_learn) {
6781 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6784 /* Determine output bundle. */
6785 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6788 if (mac->port.p != in_bundle) {
6789 xlate_report(ctx, "forwarding to learned port");
6790 output_normal(ctx, mac->port.p, vlan);
6792 xlate_report(ctx, "learned port is input port, dropping");
6795 struct ofbundle *bundle;
6797 xlate_report(ctx, "no learned MAC for destination, flooding");
6798 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6799 if (bundle != in_bundle
6800 && ofbundle_includes_vlan(bundle, vlan)
6801 && bundle->floodable
6802 && !bundle->mirror_out) {
6803 output_normal(ctx, bundle, vlan);
6806 ctx->nf_output_iface = NF_OUT_FLOOD;
6810 /* Optimized flow revalidation.
6812 * It's a difficult problem, in general, to tell which facets need to have
6813 * their actions recalculated whenever the OpenFlow flow table changes. We
6814 * don't try to solve that general problem: for most kinds of OpenFlow flow
6815 * table changes, we recalculate the actions for every facet. This is
6816 * relatively expensive, but it's good enough if the OpenFlow flow table
6817 * doesn't change very often.
6819 * However, we can expect one particular kind of OpenFlow flow table change to
6820 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6821 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6822 * table, we add a special case that applies to flow tables in which every rule
6823 * has the same form (that is, the same wildcards), except that the table is
6824 * also allowed to have a single "catch-all" flow that matches all packets. We
6825 * optimize this case by tagging all of the facets that resubmit into the table
6826 * and invalidating the same tag whenever a flow changes in that table. The
6827 * end result is that we revalidate just the facets that need it (and sometimes
6828 * a few more, but not all of the facets or even all of the facets that
6829 * resubmit to the table modified by MAC learning). */
6831 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6832 * into an OpenFlow table with the given 'basis'. */
6834 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6837 if (minimask_is_catchall(mask)) {
6840 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6841 return tag_create_deterministic(hash);
6845 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6846 * taggability of that table.
6848 * This function must be called after *each* change to a flow table. If you
6849 * skip calling it on some changes then the pointer comparisons at the end can
6850 * be invalid if you get unlucky. For example, if a flow removal causes a
6851 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6852 * different wildcards to be created with the same address, then this function
6853 * will incorrectly skip revalidation. */
6855 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6857 struct table_dpif *table = &ofproto->tables[table_id];
6858 const struct oftable *oftable = &ofproto->up.tables[table_id];
6859 struct cls_table *catchall, *other;
6860 struct cls_table *t;
6862 catchall = other = NULL;
6864 switch (hmap_count(&oftable->cls.tables)) {
6866 /* We could tag this OpenFlow table but it would make the logic a
6867 * little harder and it's a corner case that doesn't seem worth it
6873 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6874 if (cls_table_is_catchall(t)) {
6876 } else if (!other) {
6879 /* Indicate that we can't tag this by setting both tables to
6880 * NULL. (We know that 'catchall' is already NULL.) */
6887 /* Can't tag this table. */
6891 if (table->catchall_table != catchall || table->other_table != other) {
6892 table->catchall_table = catchall;
6893 table->other_table = other;
6894 ofproto->need_revalidate = REV_FLOW_TABLE;
6898 /* Given 'rule' that has changed in some way (either it is a rule being
6899 * inserted, a rule being deleted, or a rule whose actions are being
6900 * modified), marks facets for revalidation to ensure that packets will be
6901 * forwarded correctly according to the new state of the flow table.
6903 * This function must be called after *each* change to a flow table. See
6904 * the comment on table_update_taggable() for more information. */
6906 rule_invalidate(const struct rule_dpif *rule)
6908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6910 table_update_taggable(ofproto, rule->up.table_id);
6912 if (!ofproto->need_revalidate) {
6913 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6915 if (table->other_table && rule->tag) {
6916 tag_set_add(&ofproto->revalidate_set, rule->tag);
6918 ofproto->need_revalidate = REV_FLOW_TABLE;
6924 set_frag_handling(struct ofproto *ofproto_,
6925 enum ofp_config_flags frag_handling)
6927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6929 if (frag_handling != OFPC_FRAG_REASM) {
6930 ofproto->need_revalidate = REV_RECONFIGURE;
6938 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6939 const struct flow *flow,
6940 const struct ofpact *ofpacts, size_t ofpacts_len)
6942 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6943 struct odputil_keybuf keybuf;
6944 struct dpif_flow_stats stats;
6948 struct action_xlate_ctx ctx;
6949 uint64_t odp_actions_stub[1024 / 8];
6950 struct ofpbuf odp_actions;
6952 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6953 odp_flow_key_from_flow(&key, flow,
6954 ofp_port_to_odp_port(ofproto, flow->in_port));
6956 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6958 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6959 packet_get_tcp_flags(packet, flow), packet);
6960 ctx.resubmit_stats = &stats;
6962 ofpbuf_use_stub(&odp_actions,
6963 odp_actions_stub, sizeof odp_actions_stub);
6964 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6965 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6966 odp_actions.data, odp_actions.size, packet);
6967 ofpbuf_uninit(&odp_actions);
6975 set_netflow(struct ofproto *ofproto_,
6976 const struct netflow_options *netflow_options)
6978 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6980 if (netflow_options) {
6981 if (!ofproto->netflow) {
6982 ofproto->netflow = netflow_create();
6984 return netflow_set_options(ofproto->netflow, netflow_options);
6986 netflow_destroy(ofproto->netflow);
6987 ofproto->netflow = NULL;
6993 get_netflow_ids(const struct ofproto *ofproto_,
6994 uint8_t *engine_type, uint8_t *engine_id)
6996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6998 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7002 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7004 if (!facet_is_controller_flow(facet) &&
7005 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7006 struct subfacet *subfacet;
7007 struct ofexpired expired;
7009 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7010 if (subfacet->path == SF_FAST_PATH) {
7011 struct dpif_flow_stats stats;
7013 subfacet_reinstall(subfacet, &stats);
7014 subfacet_update_stats(subfacet, &stats);
7018 expired.flow = facet->flow;
7019 expired.packet_count = facet->packet_count;
7020 expired.byte_count = facet->byte_count;
7021 expired.used = facet->used;
7022 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7027 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7029 struct facet *facet;
7031 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7032 send_active_timeout(ofproto, facet);
7036 static struct ofproto_dpif *
7037 ofproto_dpif_lookup(const char *name)
7039 struct ofproto_dpif *ofproto;
7041 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7042 hash_string(name, 0), &all_ofproto_dpifs) {
7043 if (!strcmp(ofproto->up.name, name)) {
7051 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7052 const char *argv[], void *aux OVS_UNUSED)
7054 struct ofproto_dpif *ofproto;
7057 ofproto = ofproto_dpif_lookup(argv[1]);
7059 unixctl_command_reply_error(conn, "no such bridge");
7062 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7064 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7065 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7069 unixctl_command_reply(conn, "table successfully flushed");
7073 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7074 const char *argv[], void *aux OVS_UNUSED)
7076 struct ds ds = DS_EMPTY_INITIALIZER;
7077 const struct ofproto_dpif *ofproto;
7078 const struct mac_entry *e;
7080 ofproto = ofproto_dpif_lookup(argv[1]);
7082 unixctl_command_reply_error(conn, "no such bridge");
7086 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7087 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7088 struct ofbundle *bundle = e->port.p;
7089 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7090 ofbundle_get_a_port(bundle)->odp_port,
7091 e->vlan, ETH_ADDR_ARGS(e->mac),
7092 mac_entry_age(ofproto->ml, e));
7094 unixctl_command_reply(conn, ds_cstr(&ds));
7099 struct action_xlate_ctx ctx;
7105 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7106 const struct rule_dpif *rule)
7108 ds_put_char_multiple(result, '\t', level);
7110 ds_put_cstr(result, "No match\n");
7114 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7115 table_id, ntohll(rule->up.flow_cookie));
7116 cls_rule_format(&rule->up.cr, result);
7117 ds_put_char(result, '\n');
7119 ds_put_char_multiple(result, '\t', level);
7120 ds_put_cstr(result, "OpenFlow ");
7121 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7122 ds_put_char(result, '\n');
7126 trace_format_flow(struct ds *result, int level, const char *title,
7127 struct trace_ctx *trace)
7129 ds_put_char_multiple(result, '\t', level);
7130 ds_put_format(result, "%s: ", title);
7131 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7132 ds_put_cstr(result, "unchanged");
7134 flow_format(result, &trace->ctx.flow);
7135 trace->flow = trace->ctx.flow;
7137 ds_put_char(result, '\n');
7141 trace_format_regs(struct ds *result, int level, const char *title,
7142 struct trace_ctx *trace)
7146 ds_put_char_multiple(result, '\t', level);
7147 ds_put_format(result, "%s:", title);
7148 for (i = 0; i < FLOW_N_REGS; i++) {
7149 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7151 ds_put_char(result, '\n');
7155 trace_format_odp(struct ds *result, int level, const char *title,
7156 struct trace_ctx *trace)
7158 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7160 ds_put_char_multiple(result, '\t', level);
7161 ds_put_format(result, "%s: ", title);
7162 format_odp_actions(result, odp_actions->data, odp_actions->size);
7163 ds_put_char(result, '\n');
7167 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7169 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7170 struct ds *result = trace->result;
7172 ds_put_char(result, '\n');
7173 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7174 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7175 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7176 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7180 trace_report(struct action_xlate_ctx *ctx, const char *s)
7182 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7183 struct ds *result = trace->result;
7185 ds_put_char_multiple(result, '\t', ctx->recurse);
7186 ds_put_cstr(result, s);
7187 ds_put_char(result, '\n');
7191 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7192 void *aux OVS_UNUSED)
7194 const char *dpname = argv[1];
7195 struct ofproto_dpif *ofproto;
7196 struct ofpbuf odp_key;
7197 struct ofpbuf *packet;
7198 ovs_be16 initial_tci;
7204 ofpbuf_init(&odp_key, 0);
7207 ofproto = ofproto_dpif_lookup(dpname);
7209 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7213 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7214 /* ofproto/trace dpname flow [-generate] */
7215 const char *flow_s = argv[2];
7216 const char *generate_s = argv[3];
7218 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7219 * flow. We guess which type it is based on whether 'flow_s' contains
7220 * an '(', since a datapath flow always contains '(') but an
7221 * OpenFlow-like flow should not (in fact it's allowed but I believe
7222 * that's not documented anywhere).
7224 * An alternative would be to try to parse 'flow_s' both ways, but then
7225 * it would be tricky giving a sensible error message. After all, do
7226 * you just say "syntax error" or do you present both error messages?
7227 * Both choices seem lousy. */
7228 if (strchr(flow_s, '(')) {
7231 /* Convert string to datapath key. */
7232 ofpbuf_init(&odp_key, 0);
7233 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7235 unixctl_command_reply_error(conn, "Bad flow syntax");
7239 /* XXX: Since we allow the user to specify an ofproto, it's
7240 * possible they will specify a different ofproto than the one the
7241 * port actually belongs too. Ideally we should simply remove the
7242 * ability to specify the ofproto. */
7243 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7244 odp_key.size, &flow, NULL, NULL, NULL,
7246 unixctl_command_reply_error(conn, "Invalid flow");
7252 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7254 unixctl_command_reply_error(conn, error_s);
7259 initial_tci = flow.vlan_tci;
7262 /* Generate a packet, if requested. */
7264 packet = ofpbuf_new(0);
7265 flow_compose(packet, &flow);
7267 } else if (argc == 7) {
7268 /* ofproto/trace dpname priority tun_id in_port mark packet */
7269 const char *priority_s = argv[2];
7270 const char *tun_id_s = argv[3];
7271 const char *in_port_s = argv[4];
7272 const char *mark_s = argv[5];
7273 const char *packet_s = argv[6];
7274 uint32_t in_port = atoi(in_port_s);
7275 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7276 uint32_t priority = atoi(priority_s);
7277 uint32_t mark = atoi(mark_s);
7280 msg = eth_from_hex(packet_s, &packet);
7282 unixctl_command_reply_error(conn, msg);
7286 ds_put_cstr(&result, "Packet: ");
7287 s = ofp_packet_to_string(packet->data, packet->size);
7288 ds_put_cstr(&result, s);
7291 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7292 flow.tunnel.tun_id = tun_id;
7293 initial_tci = flow.vlan_tci;
7295 unixctl_command_reply_error(conn, "Bad command syntax");
7299 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7300 unixctl_command_reply(conn, ds_cstr(&result));
7303 ds_destroy(&result);
7304 ofpbuf_delete(packet);
7305 ofpbuf_uninit(&odp_key);
7309 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7310 const struct ofpbuf *packet, ovs_be16 initial_tci,
7313 struct rule_dpif *rule;
7315 ds_put_cstr(ds, "Flow: ");
7316 flow_format(ds, flow);
7317 ds_put_char(ds, '\n');
7319 rule = rule_dpif_lookup(ofproto, flow);
7321 trace_format_rule(ds, 0, 0, rule);
7322 if (rule == ofproto->miss_rule) {
7323 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7324 } else if (rule == ofproto->no_packet_in_rule) {
7325 ds_put_cstr(ds, "\nNo match, packets dropped because "
7326 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7330 uint64_t odp_actions_stub[1024 / 8];
7331 struct ofpbuf odp_actions;
7333 struct trace_ctx trace;
7336 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7339 ofpbuf_use_stub(&odp_actions,
7340 odp_actions_stub, sizeof odp_actions_stub);
7341 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7342 rule, tcp_flags, packet);
7343 trace.ctx.resubmit_hook = trace_resubmit;
7344 trace.ctx.report_hook = trace_report;
7345 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7348 ds_put_char(ds, '\n');
7349 trace_format_flow(ds, 0, "Final flow", &trace);
7350 ds_put_cstr(ds, "Datapath actions: ");
7351 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7352 ofpbuf_uninit(&odp_actions);
7354 if (trace.ctx.slow) {
7355 enum slow_path_reason slow;
7357 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7358 "slow path because it:");
7359 for (slow = trace.ctx.slow; slow; ) {
7360 enum slow_path_reason bit = rightmost_1bit(slow);
7364 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7367 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7370 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7373 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7376 ds_put_cstr(ds, "\n\t (The datapath actions are "
7377 "incomplete--for complete actions, "
7378 "please supply a packet.)");
7381 case SLOW_CONTROLLER:
7382 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7383 "to the OpenFlow controller.");
7386 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7387 "than the datapath supports.");
7394 if (slow & ~SLOW_MATCH) {
7395 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7396 "the special slow-path processing.");
7403 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7404 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7407 unixctl_command_reply(conn, NULL);
7411 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7412 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7415 unixctl_command_reply(conn, NULL);
7418 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7419 * 'reply' describing the results. */
7421 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7423 struct facet *facet;
7427 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7428 if (!facet_check_consistency(facet)) {
7433 ofproto->need_revalidate = REV_INCONSISTENCY;
7437 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7438 ofproto->up.name, errors);
7440 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7445 ofproto_dpif_self_check(struct unixctl_conn *conn,
7446 int argc, const char *argv[], void *aux OVS_UNUSED)
7448 struct ds reply = DS_EMPTY_INITIALIZER;
7449 struct ofproto_dpif *ofproto;
7452 ofproto = ofproto_dpif_lookup(argv[1]);
7454 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7455 "ofproto/list for help)");
7458 ofproto_dpif_self_check__(ofproto, &reply);
7460 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7461 ofproto_dpif_self_check__(ofproto, &reply);
7465 unixctl_command_reply(conn, ds_cstr(&reply));
7469 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7470 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7471 * to destroy 'ofproto_shash' and free the returned value. */
7472 static const struct shash_node **
7473 get_ofprotos(struct shash *ofproto_shash)
7475 const struct ofproto_dpif *ofproto;
7477 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7478 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7479 shash_add_nocopy(ofproto_shash, name, ofproto);
7482 return shash_sort(ofproto_shash);
7486 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7487 const char *argv[] OVS_UNUSED,
7488 void *aux OVS_UNUSED)
7490 struct ds ds = DS_EMPTY_INITIALIZER;
7491 struct shash ofproto_shash;
7492 const struct shash_node **sorted_ofprotos;
7495 shash_init(&ofproto_shash);
7496 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7497 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7498 const struct shash_node *node = sorted_ofprotos[i];
7499 ds_put_format(&ds, "%s\n", node->name);
7502 shash_destroy(&ofproto_shash);
7503 free(sorted_ofprotos);
7505 unixctl_command_reply(conn, ds_cstr(&ds));
7510 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7512 struct dpif_dp_stats s;
7513 const struct shash_node **ports;
7516 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7518 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7519 dpif_name(ofproto->backer->dpif));
7520 /* xxx It would be better to show bridge-specific stats instead
7521 * xxx of dp ones. */
7523 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7524 s.n_hit, s.n_missed, s.n_lost);
7525 ds_put_format(ds, "\tflows: %zu\n",
7526 hmap_count(&ofproto->subfacets));
7528 ports = shash_sort(&ofproto->up.port_by_name);
7529 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7530 const struct shash_node *node = ports[i];
7531 struct ofport *ofport = node->data;
7532 const char *name = netdev_get_name(ofport->netdev);
7533 const char *type = netdev_get_type(ofport->netdev);
7535 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7536 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7537 if (strcmp(type, "system")) {
7538 struct netdev *netdev;
7541 ds_put_format(ds, " (%s", type);
7543 error = netdev_open(name, type, &netdev);
7548 error = netdev_get_config(netdev, &config);
7550 const struct smap_node **nodes;
7553 nodes = smap_sort(&config);
7554 for (i = 0; i < smap_count(&config); i++) {
7555 const struct smap_node *node = nodes[i];
7556 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7557 node->key, node->value);
7561 smap_destroy(&config);
7563 netdev_close(netdev);
7565 ds_put_char(ds, ')');
7567 ds_put_char(ds, '\n');
7573 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7574 const char *argv[], void *aux OVS_UNUSED)
7576 struct ds ds = DS_EMPTY_INITIALIZER;
7577 const struct ofproto_dpif *ofproto;
7581 for (i = 1; i < argc; i++) {
7582 ofproto = ofproto_dpif_lookup(argv[i]);
7584 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7585 "for help)", argv[i]);
7586 unixctl_command_reply_error(conn, ds_cstr(&ds));
7589 show_dp_format(ofproto, &ds);
7592 struct shash ofproto_shash;
7593 const struct shash_node **sorted_ofprotos;
7596 shash_init(&ofproto_shash);
7597 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7598 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7599 const struct shash_node *node = sorted_ofprotos[i];
7600 show_dp_format(node->data, &ds);
7603 shash_destroy(&ofproto_shash);
7604 free(sorted_ofprotos);
7607 unixctl_command_reply(conn, ds_cstr(&ds));
7612 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7613 int argc OVS_UNUSED, const char *argv[],
7614 void *aux OVS_UNUSED)
7616 struct ds ds = DS_EMPTY_INITIALIZER;
7617 const struct ofproto_dpif *ofproto;
7618 struct subfacet *subfacet;
7620 ofproto = ofproto_dpif_lookup(argv[1]);
7622 unixctl_command_reply_error(conn, "no such bridge");
7626 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7627 struct odputil_keybuf keybuf;
7630 subfacet_get_key(subfacet, &keybuf, &key);
7631 odp_flow_key_format(key.data, key.size, &ds);
7633 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7634 subfacet->dp_packet_count, subfacet->dp_byte_count);
7635 if (subfacet->used) {
7636 ds_put_format(&ds, "%.3fs",
7637 (time_msec() - subfacet->used) / 1000.0);
7639 ds_put_format(&ds, "never");
7641 if (subfacet->facet->tcp_flags) {
7642 ds_put_cstr(&ds, ", flags:");
7643 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7646 ds_put_cstr(&ds, ", actions:");
7647 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7648 ds_put_char(&ds, '\n');
7651 unixctl_command_reply(conn, ds_cstr(&ds));
7656 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7657 int argc OVS_UNUSED, const char *argv[],
7658 void *aux OVS_UNUSED)
7660 struct ds ds = DS_EMPTY_INITIALIZER;
7661 struct ofproto_dpif *ofproto;
7663 ofproto = ofproto_dpif_lookup(argv[1]);
7665 unixctl_command_reply_error(conn, "no such bridge");
7669 flush(&ofproto->up);
7671 unixctl_command_reply(conn, ds_cstr(&ds));
7676 ofproto_dpif_unixctl_init(void)
7678 static bool registered;
7684 unixctl_command_register(
7686 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7687 2, 6, ofproto_unixctl_trace, NULL);
7688 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7689 ofproto_unixctl_fdb_flush, NULL);
7690 unixctl_command_register("fdb/show", "bridge", 1, 1,
7691 ofproto_unixctl_fdb_show, NULL);
7692 unixctl_command_register("ofproto/clog", "", 0, 0,
7693 ofproto_dpif_clog, NULL);
7694 unixctl_command_register("ofproto/unclog", "", 0, 0,
7695 ofproto_dpif_unclog, NULL);
7696 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7697 ofproto_dpif_self_check, NULL);
7698 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7699 ofproto_unixctl_dpif_dump_dps, NULL);
7700 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7701 ofproto_unixctl_dpif_show, NULL);
7702 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7703 ofproto_unixctl_dpif_dump_flows, NULL);
7704 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7705 ofproto_unixctl_dpif_del_flows, NULL);
7708 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7710 * This is deprecated. It is only for compatibility with broken device drivers
7711 * in old versions of Linux that do not properly support VLANs when VLAN
7712 * devices are not used. When broken device drivers are no longer in
7713 * widespread use, we will delete these interfaces. */
7716 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7718 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7719 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7721 if (realdev_ofp_port == ofport->realdev_ofp_port
7722 && vid == ofport->vlandev_vid) {
7726 ofproto->need_revalidate = REV_RECONFIGURE;
7728 if (ofport->realdev_ofp_port) {
7731 if (realdev_ofp_port && ofport->bundle) {
7732 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7733 * themselves be part of a bundle. */
7734 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7737 ofport->realdev_ofp_port = realdev_ofp_port;
7738 ofport->vlandev_vid = vid;
7740 if (realdev_ofp_port) {
7741 vsp_add(ofport, realdev_ofp_port, vid);
7748 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7750 return hash_2words(realdev_ofp_port, vid);
7753 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7754 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7755 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7756 * it would return the port number of eth0.9.
7758 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7759 * function just returns its 'realdev_odp_port' argument. */
7761 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7762 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7764 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7765 uint16_t realdev_ofp_port;
7766 int vid = vlan_tci_to_vid(vlan_tci);
7767 const struct vlan_splinter *vsp;
7769 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7770 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7771 hash_realdev_vid(realdev_ofp_port, vid),
7772 &ofproto->realdev_vid_map) {
7773 if (vsp->realdev_ofp_port == realdev_ofp_port
7774 && vsp->vid == vid) {
7775 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7779 return realdev_odp_port;
7782 static struct vlan_splinter *
7783 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7785 struct vlan_splinter *vsp;
7787 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7788 &ofproto->vlandev_map) {
7789 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7797 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7798 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7799 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7800 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7801 * eth0 and store 9 in '*vid'.
7803 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7804 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7807 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7808 uint16_t vlandev_ofp_port, int *vid)
7810 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7811 const struct vlan_splinter *vsp;
7813 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7818 return vsp->realdev_ofp_port;
7824 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7825 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7826 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7827 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7828 * always the case unless VLAN splinters are enabled), returns false without
7829 * making any changes. */
7831 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7836 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7841 /* Cause the flow to be processed as if it came in on the real device with
7842 * the VLAN device's VLAN ID. */
7843 flow->in_port = realdev;
7844 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7849 vsp_remove(struct ofport_dpif *port)
7851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7852 struct vlan_splinter *vsp;
7854 vsp = vlandev_find(ofproto, port->up.ofp_port);
7856 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7857 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7860 port->realdev_ofp_port = 0;
7862 VLOG_ERR("missing vlan device record");
7867 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7869 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7871 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7872 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7873 == realdev_ofp_port)) {
7874 struct vlan_splinter *vsp;
7876 vsp = xmalloc(sizeof *vsp);
7877 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7878 hash_int(port->up.ofp_port, 0));
7879 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7880 hash_realdev_vid(realdev_ofp_port, vid));
7881 vsp->realdev_ofp_port = realdev_ofp_port;
7882 vsp->vlandev_ofp_port = port->up.ofp_port;
7885 port->realdev_ofp_port = realdev_ofp_port;
7887 VLOG_ERR("duplicate vlan device record");
7892 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7894 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7895 return ofport ? ofport->odp_port : OVSP_NONE;
7898 static struct ofport_dpif *
7899 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7901 struct ofport_dpif *port;
7903 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7904 hash_int(odp_port, 0),
7905 &backer->odp_to_ofport_map) {
7906 if (port->odp_port == odp_port) {
7915 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7917 struct ofport_dpif *port;
7919 port = odp_port_to_ofport(ofproto->backer, odp_port);
7920 if (port && &ofproto->up == port->up.ofproto) {
7921 return port->up.ofp_port;
7927 const struct ofproto_class ofproto_dpif_class = {
7962 port_is_lacp_current,
7963 NULL, /* rule_choose_table */
7970 rule_modify_actions,
7979 get_cfm_remote_mpids,
7984 get_stp_port_status,
7991 is_mirror_output_bundle,
7992 forward_bpdu_changed,
7993 set_mac_table_config,