2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
53 #include "unaligned.h"
55 #include "vlan-bitmap.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
60 COVERAGE_DEFINE(ofproto_dpif_expired);
61 COVERAGE_DEFINE(ofproto_dpif_xlate);
62 COVERAGE_DEFINE(facet_changed_rule);
63 COVERAGE_DEFINE(facet_revalidate);
64 COVERAGE_DEFINE(facet_unexpected);
65 COVERAGE_DEFINE(facet_suppress);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 64
71 /* Number of implemented OpenFlow tables. */
72 enum { N_TABLES = 255 };
73 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
74 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
84 * - Do include packets and bytes from facets that have been deleted or
85 * whose own statistics have been folded into the rule.
87 * - Do include packets and bytes sent "by hand" that were accounted to
88 * the rule without any facet being involved (this is a rare corner
89 * case in rule_execute()).
91 * - Do not include packet or bytes that can be obtained from any facet's
92 * packet_count or byte_count member or that can be obtained from the
93 * datapath by, e.g., dpif_flow_get() for any subfacet.
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
98 tag_type tag; /* Caches rule_calculate_tag() result. */
100 struct list facets; /* List of "struct facet"s. */
103 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
105 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
108 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
109 const struct flow *);
110 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
113 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
114 const struct flow *flow);
116 static void rule_credit_stats(struct rule_dpif *,
117 const struct dpif_flow_stats *);
118 static void flow_push_stats(struct rule_dpif *, const struct flow *,
119 const struct dpif_flow_stats *);
120 static tag_type rule_calculate_tag(const struct flow *,
121 const struct flow_wildcards *,
123 static void rule_invalidate(const struct rule_dpif *);
125 #define MAX_MIRRORS 32
126 typedef uint32_t mirror_mask_t;
127 #define MIRROR_MASK_C(X) UINT32_C(X)
128 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
130 struct ofproto_dpif *ofproto; /* Owning ofproto. */
131 size_t idx; /* In ofproto's "mirrors" array. */
132 void *aux; /* Key supplied by ofproto's client. */
133 char *name; /* Identifier for log messages. */
135 /* Selection criteria. */
136 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
137 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
138 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
140 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
141 struct ofbundle *out; /* Output port or NULL. */
142 int out_vlan; /* Output VLAN or -1. */
143 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
146 int64_t packet_count; /* Number of packets sent. */
147 int64_t byte_count; /* Number of bytes sent. */
150 static void mirror_destroy(struct ofmirror *);
151 static void update_mirror_stats(struct ofproto_dpif *ofproto,
152 mirror_mask_t mirrors,
153 uint64_t packets, uint64_t bytes);
156 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
157 struct ofproto_dpif *ofproto; /* Owning ofproto. */
158 void *aux; /* Key supplied by ofproto's client. */
159 char *name; /* Identifier for log messages. */
162 struct list ports; /* Contains "struct ofport"s. */
163 enum port_vlan_mode vlan_mode; /* VLAN mode */
164 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
165 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
166 * NULL if all VLANs are trunked. */
167 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
168 struct bond *bond; /* Nonnull iff more than one port. */
169 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
172 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
174 /* Port mirroring info. */
175 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
176 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
177 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
180 static void bundle_remove(struct ofport *);
181 static void bundle_update(struct ofbundle *);
182 static void bundle_destroy(struct ofbundle *);
183 static void bundle_del_port(struct ofport_dpif *);
184 static void bundle_run(struct ofbundle *);
185 static void bundle_wait(struct ofbundle *);
186 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
187 uint16_t in_port, bool warn,
188 struct ofport_dpif **in_ofportp);
190 /* A controller may use OFPP_NONE as the ingress port to indicate that
191 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
192 * when an input bundle is needed for validation (e.g., mirroring or
193 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
194 * any 'port' structs, so care must be taken when dealing with it. */
195 static struct ofbundle ofpp_none_bundle = {
197 .vlan_mode = PORT_VLAN_TRUNK
200 static void stp_run(struct ofproto_dpif *ofproto);
201 static void stp_wait(struct ofproto_dpif *ofproto);
202 static int set_stp_port(struct ofport *,
203 const struct ofproto_port_stp_settings *);
205 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
207 struct action_xlate_ctx {
208 /* action_xlate_ctx_init() initializes these members. */
211 struct ofproto_dpif *ofproto;
213 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
214 * this flow when actions change header fields. */
217 /* The packet corresponding to 'flow', or a null pointer if we are
218 * revalidating without a packet to refer to. */
219 const struct ofpbuf *packet;
221 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
222 * actions update the flow table?
224 * We want to update these tables if we are actually processing a packet,
225 * or if we are accounting for packets that the datapath has processed, but
226 * not if we are just revalidating. */
229 /* The rule that we are currently translating, or NULL. */
230 struct rule_dpif *rule;
232 /* Union of the set of TCP flags seen so far in this flow. (Used only by
233 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
237 /* If nonnull, flow translation calls this function just before executing a
238 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
239 * when the recursion depth is exceeded.
241 * 'rule' is the rule being submitted into. It will be null if the
242 * resubmit or OFPP_TABLE action didn't find a matching rule.
244 * This is normally null so the client has to set it manually after
245 * calling action_xlate_ctx_init(). */
246 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
248 /* If nonnull, flow translation calls this function to report some
249 * significant decision, e.g. to explain why OFPP_NORMAL translation
250 * dropped a packet. */
251 void (*report_hook)(struct action_xlate_ctx *, const char *s);
253 /* If nonnull, flow translation credits the specified statistics to each
254 * rule reached through a resubmit or OFPP_TABLE action.
256 * This is normally null so the client has to set it manually after
257 * calling action_xlate_ctx_init(). */
258 const struct dpif_flow_stats *resubmit_stats;
260 /* xlate_actions() initializes and uses these members. The client might want
261 * to look at them after it returns. */
263 struct ofpbuf *odp_actions; /* Datapath actions. */
264 tag_type tags; /* Tags associated with actions. */
265 enum slow_path_reason slow; /* 0 if fast path may be used. */
266 bool has_learn; /* Actions include NXAST_LEARN? */
267 bool has_normal; /* Actions output to OFPP_NORMAL? */
268 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
269 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
270 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
272 /* xlate_actions() initializes and uses these members, but the client has no
273 * reason to look at them. */
275 int recurse; /* Recursion level, via xlate_table_action. */
276 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
277 struct flow base_flow; /* Flow at the last commit. */
278 uint32_t orig_skb_priority; /* Priority when packet arrived. */
279 uint8_t table_id; /* OpenFlow table ID where flow was found. */
280 uint32_t sflow_n_outputs; /* Number of output ports. */
281 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
282 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
283 bool exit; /* No further actions should be processed. */
284 struct flow orig_flow; /* Copy of original flow. */
287 static void action_xlate_ctx_init(struct action_xlate_ctx *,
288 struct ofproto_dpif *, const struct flow *,
289 ovs_be16 initial_tci, struct rule_dpif *,
290 uint8_t tcp_flags, const struct ofpbuf *);
291 static void xlate_actions(struct action_xlate_ctx *,
292 const struct ofpact *ofpacts, size_t ofpacts_len,
293 struct ofpbuf *odp_actions);
294 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
295 const struct ofpact *ofpacts,
298 static size_t put_userspace_action(const struct ofproto_dpif *,
299 struct ofpbuf *odp_actions,
301 const union user_action_cookie *);
303 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
304 enum slow_path_reason,
305 uint64_t *stub, size_t stub_size,
306 const struct nlattr **actionsp,
307 size_t *actions_lenp);
309 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
311 /* A subfacet (see "struct subfacet" below) has three possible installation
314 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
315 * case just after the subfacet is created, just before the subfacet is
316 * destroyed, or if the datapath returns an error when we try to install a
319 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
321 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
322 * ofproto_dpif is installed in the datapath.
325 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
326 SF_FAST_PATH, /* Full actions are installed. */
327 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
330 static const char *subfacet_path_to_string(enum subfacet_path);
332 /* A dpif flow and actions associated with a facet.
334 * See also the large comment on struct facet. */
337 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
338 struct list list_node; /* In struct facet's 'facets' list. */
339 struct facet *facet; /* Owning facet. */
343 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
344 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
345 * regenerate the ODP flow key from ->facet->flow. */
346 enum odp_key_fitness key_fitness;
350 long long int used; /* Time last used; time created if not used. */
352 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
353 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
357 * These should be essentially identical for every subfacet in a facet, but
358 * may differ in trivial ways due to VLAN splinters. */
359 size_t actions_len; /* Number of bytes in actions[]. */
360 struct nlattr *actions; /* Datapath actions. */
362 enum slow_path_reason slow; /* 0 if fast path may be used. */
363 enum subfacet_path path; /* Installed in datapath? */
365 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
366 * splinters can cause it to differ. This value should be removed when
367 * the VLAN splinters feature is no longer needed. */
368 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
371 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
372 const struct nlattr *key,
373 size_t key_len, ovs_be16 initial_tci,
375 static struct subfacet *subfacet_find(struct ofproto_dpif *,
376 const struct nlattr *key, size_t key_len);
377 static void subfacet_destroy(struct subfacet *);
378 static void subfacet_destroy__(struct subfacet *);
379 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
381 static void subfacet_reset_dp_stats(struct subfacet *,
382 struct dpif_flow_stats *);
383 static void subfacet_update_time(struct subfacet *, long long int used);
384 static void subfacet_update_stats(struct subfacet *,
385 const struct dpif_flow_stats *);
386 static void subfacet_make_actions(struct subfacet *,
387 const struct ofpbuf *packet,
388 struct ofpbuf *odp_actions);
389 static int subfacet_install(struct subfacet *,
390 const struct nlattr *actions, size_t actions_len,
391 struct dpif_flow_stats *, enum slow_path_reason);
392 static void subfacet_uninstall(struct subfacet *);
394 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
396 /* An exact-match instantiation of an OpenFlow flow.
398 * A facet associates a "struct flow", which represents the Open vSwitch
399 * userspace idea of an exact-match flow, with one or more subfacets. Each
400 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
401 * the facet. When the kernel module (or other dpif implementation) and Open
402 * vSwitch userspace agree on the definition of a flow key, there is exactly
403 * one subfacet per facet. If the dpif implementation supports more-specific
404 * flow matching than userspace, however, a facet can have more than one
405 * subfacet, each of which corresponds to some distinction in flow that
406 * userspace simply doesn't understand.
408 * Flow expiration works in terms of subfacets, so a facet must have at least
409 * one subfacet or it will never expire, leaking memory. */
412 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
413 struct list list_node; /* In owning rule's 'facets' list. */
414 struct rule_dpif *rule; /* Owning rule. */
417 struct list subfacets;
418 long long int used; /* Time last used; time created if not used. */
425 * - Do include packets and bytes sent "by hand", e.g. with
428 * - Do include packets and bytes that were obtained from the datapath
429 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
430 * DPIF_FP_ZERO_STATS).
432 * - Do not include packets or bytes that can be obtained from the
433 * datapath for any existing subfacet.
435 uint64_t packet_count; /* Number of packets received. */
436 uint64_t byte_count; /* Number of bytes received. */
438 /* Resubmit statistics. */
439 uint64_t prev_packet_count; /* Number of packets from last stats push. */
440 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
441 long long int prev_used; /* Used time from last stats push. */
444 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
445 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
446 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
448 /* Properties of datapath actions.
450 * Every subfacet has its own actions because actions can differ slightly
451 * between splintered and non-splintered subfacets due to the VLAN tag
452 * being initially different (present vs. absent). All of them have these
453 * properties in common so we just store one copy of them here. */
454 bool has_learn; /* Actions include NXAST_LEARN? */
455 bool has_normal; /* Actions output to OFPP_NORMAL? */
456 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
457 tag_type tags; /* Tags that would require revalidation. */
458 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
460 /* Storage for a single subfacet, to reduce malloc() time and space
461 * overhead. (A facet always has at least one subfacet and in the common
462 * case has exactly one subfacet.) */
463 struct subfacet one_subfacet;
466 static struct facet *facet_create(struct rule_dpif *,
467 const struct flow *, uint32_t hash);
468 static void facet_remove(struct facet *);
469 static void facet_free(struct facet *);
471 static struct facet *facet_find(struct ofproto_dpif *,
472 const struct flow *, uint32_t hash);
473 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
474 const struct flow *, uint32_t hash);
475 static void facet_revalidate(struct facet *);
476 static bool facet_check_consistency(struct facet *);
478 static void facet_flush_stats(struct facet *);
480 static void facet_update_time(struct facet *, long long int used);
481 static void facet_reset_counters(struct facet *);
482 static void facet_push_stats(struct facet *);
483 static void facet_learn(struct facet *);
484 static void facet_account(struct facet *);
486 static bool facet_is_controller_flow(struct facet *);
492 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
493 struct list bundle_node; /* In struct ofbundle's "ports" list. */
494 struct cfm *cfm; /* Connectivity Fault Management, if any. */
495 tag_type tag; /* Tag associated with this port. */
496 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
497 bool may_enable; /* May be enabled in bonds. */
498 long long int carrier_seq; /* Carrier status changes. */
501 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
502 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
503 long long int stp_state_entered;
505 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
507 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
509 * This is deprecated. It is only for compatibility with broken device
510 * drivers in old versions of Linux that do not properly support VLANs when
511 * VLAN devices are not used. When broken device drivers are no longer in
512 * widespread use, we will delete these interfaces. */
513 uint16_t realdev_ofp_port;
517 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
518 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
519 * traffic egressing the 'ofport' with that priority should be marked with. */
520 struct priority_to_dscp {
521 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
522 uint32_t priority; /* Priority of this queue (see struct flow). */
524 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
527 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
529 * This is deprecated. It is only for compatibility with broken device drivers
530 * in old versions of Linux that do not properly support VLANs when VLAN
531 * devices are not used. When broken device drivers are no longer in
532 * widespread use, we will delete these interfaces. */
533 struct vlan_splinter {
534 struct hmap_node realdev_vid_node;
535 struct hmap_node vlandev_node;
536 uint16_t realdev_ofp_port;
537 uint16_t vlandev_ofp_port;
541 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
542 uint32_t realdev, ovs_be16 vlan_tci);
543 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
544 static void vsp_remove(struct ofport_dpif *);
545 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
547 static struct ofport_dpif *
548 ofport_dpif_cast(const struct ofport *ofport)
550 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
551 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
554 static void port_run(struct ofport_dpif *);
555 static void port_run_fast(struct ofport_dpif *);
556 static void port_wait(struct ofport_dpif *);
557 static int set_cfm(struct ofport *, const struct cfm_settings *);
558 static void ofport_clear_priorities(struct ofport_dpif *);
560 struct dpif_completion {
561 struct list list_node;
562 struct ofoperation *op;
565 /* Extra information about a classifier table.
566 * Currently used just for optimized flow revalidation. */
568 /* If either of these is nonnull, then this table has a form that allows
569 * flows to be tagged to avoid revalidating most flows for the most common
570 * kinds of flow table changes. */
571 struct cls_table *catchall_table; /* Table that wildcards all fields. */
572 struct cls_table *other_table; /* Table with any other wildcard set. */
573 uint32_t basis; /* Keeps each table's tags separate. */
576 /* Reasons that we might need to revalidate every facet, and corresponding
579 * A value of 0 means that there is no need to revalidate.
581 * It would be nice to have some cleaner way to integrate with coverage
582 * counters, but with only a few reasons I guess this is good enough for
584 enum revalidate_reason {
585 REV_RECONFIGURE = 1, /* Switch configuration changed. */
586 REV_STP, /* Spanning tree protocol port status change. */
587 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
588 REV_FLOW_TABLE, /* Flow table changed. */
589 REV_INCONSISTENCY /* Facet self-check failed. */
591 COVERAGE_DEFINE(rev_reconfigure);
592 COVERAGE_DEFINE(rev_stp);
593 COVERAGE_DEFINE(rev_port_toggled);
594 COVERAGE_DEFINE(rev_flow_table);
595 COVERAGE_DEFINE(rev_inconsistency);
597 struct ofproto_dpif {
598 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
603 /* Special OpenFlow rules. */
604 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
605 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
611 struct netflow *netflow;
612 struct dpif_sflow *sflow;
613 struct hmap bundles; /* Contains "struct ofbundle"s. */
614 struct mac_learning *ml;
615 struct ofmirror *mirrors[MAX_MIRRORS];
617 bool has_bonded_bundles;
620 struct timer next_expiration;
624 struct hmap subfacets;
625 struct governor *governor;
628 struct table_dpif tables[N_TABLES];
629 enum revalidate_reason need_revalidate;
630 struct tag_set revalidate_set;
632 /* Support for debugging async flow mods. */
633 struct list completions;
635 bool has_bundle_action; /* True when the first bundle action appears. */
636 struct netdev_stats stats; /* To account packets generated and consumed in
641 long long int stp_last_tick;
643 /* VLAN splinters. */
644 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
645 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
648 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
649 * for debugging the asynchronous flow_mod implementation.) */
652 /* All existing ofproto_dpif instances, indexed by ->up.name. */
653 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
655 static void ofproto_dpif_unixctl_init(void);
657 static struct ofproto_dpif *
658 ofproto_dpif_cast(const struct ofproto *ofproto)
660 assert(ofproto->ofproto_class == &ofproto_dpif_class);
661 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
664 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
666 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
668 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
669 const struct ofpbuf *, ovs_be16 initial_tci,
672 /* Packet processing. */
673 static void update_learning_table(struct ofproto_dpif *,
674 const struct flow *, int vlan,
677 #define FLOW_MISS_MAX_BATCH 50
678 static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
680 /* Flow expiration. */
681 static int expire(struct ofproto_dpif *);
684 static void send_netflow_active_timeouts(struct ofproto_dpif *);
687 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
688 static size_t compose_sflow_action(const struct ofproto_dpif *,
689 struct ofpbuf *odp_actions,
690 const struct flow *, uint32_t odp_port);
691 static void add_mirror_actions(struct action_xlate_ctx *ctx,
692 const struct flow *flow);
693 /* Global variables. */
694 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
696 /* Factory functions. */
699 enumerate_types(struct sset *types)
701 dp_enumerate_types(types);
705 enumerate_names(const char *type, struct sset *names)
707 return dp_enumerate_names(type, names);
711 del(const char *type, const char *name)
716 error = dpif_open(name, type, &dpif);
718 error = dpif_delete(dpif);
724 /* Basic life-cycle. */
726 static int add_internal_flows(struct ofproto_dpif *);
728 static struct ofproto *
731 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
736 dealloc(struct ofproto *ofproto_)
738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
743 construct(struct ofproto *ofproto_)
745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
746 const char *name = ofproto->up.name;
750 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
752 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
756 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
757 ofproto->n_matches = 0;
759 dpif_flow_flush(ofproto->dpif);
760 dpif_recv_purge(ofproto->dpif);
762 error = dpif_recv_set(ofproto->dpif, true);
764 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
765 dpif_close(ofproto->dpif);
769 ofproto->netflow = NULL;
770 ofproto->sflow = NULL;
772 hmap_init(&ofproto->bundles);
773 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
774 for (i = 0; i < MAX_MIRRORS; i++) {
775 ofproto->mirrors[i] = NULL;
777 ofproto->has_bonded_bundles = false;
779 timer_set_duration(&ofproto->next_expiration, 1000);
781 hmap_init(&ofproto->facets);
782 hmap_init(&ofproto->subfacets);
783 ofproto->governor = NULL;
785 for (i = 0; i < N_TABLES; i++) {
786 struct table_dpif *table = &ofproto->tables[i];
788 table->catchall_table = NULL;
789 table->other_table = NULL;
790 table->basis = random_uint32();
792 ofproto->need_revalidate = 0;
793 tag_set_init(&ofproto->revalidate_set);
795 list_init(&ofproto->completions);
797 ofproto_dpif_unixctl_init();
799 ofproto->has_mirrors = false;
800 ofproto->has_bundle_action = false;
802 hmap_init(&ofproto->vlandev_map);
803 hmap_init(&ofproto->realdev_vid_map);
805 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
806 hash_string(ofproto->up.name, 0));
807 memset(&ofproto->stats, 0, sizeof ofproto->stats);
809 ofproto_init_tables(ofproto_, N_TABLES);
810 error = add_internal_flows(ofproto);
811 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
817 add_internal_flow(struct ofproto_dpif *ofproto, int id,
818 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
820 struct ofputil_flow_mod fm;
823 cls_rule_init_catchall(&fm.cr, 0);
824 cls_rule_set_reg(&fm.cr, 0, id);
825 fm.new_cookie = htonll(0);
826 fm.cookie = htonll(0);
827 fm.cookie_mask = htonll(0);
828 fm.table_id = TBL_INTERNAL;
829 fm.command = OFPFC_ADD;
835 fm.ofpacts = ofpacts->data;
836 fm.ofpacts_len = ofpacts->size;
838 error = ofproto_flow_mod(&ofproto->up, &fm);
840 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
841 id, ofperr_to_string(error));
845 *rulep = rule_dpif_lookup__(ofproto, &fm.cr.flow, TBL_INTERNAL);
846 assert(*rulep != NULL);
852 add_internal_flows(struct ofproto_dpif *ofproto)
854 struct ofpact_controller *controller;
855 uint64_t ofpacts_stub[128 / 8];
856 struct ofpbuf ofpacts;
860 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
863 controller = ofpact_put_CONTROLLER(&ofpacts);
864 controller->max_len = UINT16_MAX;
865 controller->controller_id = 0;
866 controller->reason = OFPR_NO_MATCH;
867 ofpact_pad(&ofpacts);
869 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
874 ofpbuf_clear(&ofpacts);
875 error = add_internal_flow(ofproto, id++, &ofpacts,
876 &ofproto->no_packet_in_rule);
881 complete_operations(struct ofproto_dpif *ofproto)
883 struct dpif_completion *c, *next;
885 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
886 ofoperation_complete(c->op, 0);
887 list_remove(&c->list_node);
893 destruct(struct ofproto *ofproto_)
895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
896 struct rule_dpif *rule, *next_rule;
897 struct oftable *table;
900 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
901 complete_operations(ofproto);
903 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
904 struct cls_cursor cursor;
906 cls_cursor_init(&cursor, &table->cls, NULL);
907 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
908 ofproto_rule_destroy(&rule->up);
912 for (i = 0; i < MAX_MIRRORS; i++) {
913 mirror_destroy(ofproto->mirrors[i]);
916 netflow_destroy(ofproto->netflow);
917 dpif_sflow_destroy(ofproto->sflow);
918 hmap_destroy(&ofproto->bundles);
919 mac_learning_destroy(ofproto->ml);
921 hmap_destroy(&ofproto->facets);
922 hmap_destroy(&ofproto->subfacets);
923 governor_destroy(ofproto->governor);
925 hmap_destroy(&ofproto->vlandev_map);
926 hmap_destroy(&ofproto->realdev_vid_map);
928 dpif_close(ofproto->dpif);
932 run_fast(struct ofproto *ofproto_)
934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
935 struct ofport_dpif *ofport;
938 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
939 port_run_fast(ofport);
942 /* Handle one or more batches of upcalls, until there's nothing left to do
943 * or until we do a fixed total amount of work.
945 * We do work in batches because it can be much cheaper to set up a number
946 * of flows and fire off their patches all at once. We do multiple batches
947 * because in some cases handling a packet can cause another packet to be
948 * queued almost immediately as part of the return flow. Both
949 * optimizations can make major improvements on some benchmarks and
950 * presumably for real traffic as well. */
952 while (work < FLOW_MISS_MAX_BATCH) {
953 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
963 run(struct ofproto *ofproto_)
965 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
966 struct ofport_dpif *ofport;
967 struct ofbundle *bundle;
971 complete_operations(ofproto);
973 dpif_run(ofproto->dpif);
975 error = run_fast(ofproto_);
980 if (timer_expired(&ofproto->next_expiration)) {
981 int delay = expire(ofproto);
982 timer_set_duration(&ofproto->next_expiration, delay);
985 if (ofproto->netflow) {
986 if (netflow_run(ofproto->netflow)) {
987 send_netflow_active_timeouts(ofproto);
990 if (ofproto->sflow) {
991 dpif_sflow_run(ofproto->sflow);
994 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
997 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1002 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1004 /* Now revalidate if there's anything to do. */
1005 if (ofproto->need_revalidate
1006 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1007 struct tag_set revalidate_set = ofproto->revalidate_set;
1008 bool revalidate_all = ofproto->need_revalidate;
1009 struct facet *facet;
1011 switch (ofproto->need_revalidate) {
1012 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1013 case REV_STP: COVERAGE_INC(rev_stp); break;
1014 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1015 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1016 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1019 /* Clear the revalidation flags. */
1020 tag_set_init(&ofproto->revalidate_set);
1021 ofproto->need_revalidate = 0;
1023 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1025 || tag_set_intersects(&revalidate_set, facet->tags)) {
1026 facet_revalidate(facet);
1031 /* Check the consistency of a random facet, to aid debugging. */
1032 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1033 struct facet *facet;
1035 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1036 struct facet, hmap_node);
1037 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1038 if (!facet_check_consistency(facet)) {
1039 ofproto->need_revalidate = REV_INCONSISTENCY;
1044 if (ofproto->governor) {
1047 governor_run(ofproto->governor);
1049 /* If the governor has shrunk to its minimum size and the number of
1050 * subfacets has dwindled, then drop the governor entirely.
1052 * For hysteresis, the number of subfacets to drop the governor is
1053 * smaller than the number needed to trigger its creation. */
1054 n_subfacets = hmap_count(&ofproto->subfacets);
1055 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1056 && governor_is_idle(ofproto->governor)) {
1057 governor_destroy(ofproto->governor);
1058 ofproto->governor = NULL;
1066 wait(struct ofproto *ofproto_)
1068 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1069 struct ofport_dpif *ofport;
1070 struct ofbundle *bundle;
1072 if (!clogged && !list_is_empty(&ofproto->completions)) {
1073 poll_immediate_wake();
1076 dpif_wait(ofproto->dpif);
1077 dpif_recv_wait(ofproto->dpif);
1078 if (ofproto->sflow) {
1079 dpif_sflow_wait(ofproto->sflow);
1081 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1082 poll_immediate_wake();
1084 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1087 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1088 bundle_wait(bundle);
1090 if (ofproto->netflow) {
1091 netflow_wait(ofproto->netflow);
1093 mac_learning_wait(ofproto->ml);
1095 if (ofproto->need_revalidate) {
1096 /* Shouldn't happen, but if it does just go around again. */
1097 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1098 poll_immediate_wake();
1100 timer_wait(&ofproto->next_expiration);
1102 if (ofproto->governor) {
1103 governor_wait(ofproto->governor);
1108 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1110 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1112 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1113 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1117 flush(struct ofproto *ofproto_)
1119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1120 struct facet *facet, *next_facet;
1122 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1123 /* Mark the facet as not installed so that facet_remove() doesn't
1124 * bother trying to uninstall it. There is no point in uninstalling it
1125 * individually since we are about to blow away all the facets with
1126 * dpif_flow_flush(). */
1127 struct subfacet *subfacet;
1129 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
1130 subfacet->path = SF_NOT_INSTALLED;
1131 subfacet->dp_packet_count = 0;
1132 subfacet->dp_byte_count = 0;
1134 facet_remove(facet);
1136 dpif_flow_flush(ofproto->dpif);
1140 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1141 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1143 *arp_match_ip = true;
1144 *actions = (OFPUTIL_A_OUTPUT |
1145 OFPUTIL_A_SET_VLAN_VID |
1146 OFPUTIL_A_SET_VLAN_PCP |
1147 OFPUTIL_A_STRIP_VLAN |
1148 OFPUTIL_A_SET_DL_SRC |
1149 OFPUTIL_A_SET_DL_DST |
1150 OFPUTIL_A_SET_NW_SRC |
1151 OFPUTIL_A_SET_NW_DST |
1152 OFPUTIL_A_SET_NW_TOS |
1153 OFPUTIL_A_SET_TP_SRC |
1154 OFPUTIL_A_SET_TP_DST |
1159 get_tables(struct ofproto *ofproto_, struct ofp10_table_stats *ots)
1161 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1162 struct dpif_dp_stats s;
1164 strcpy(ots->name, "classifier");
1166 dpif_get_dp_stats(ofproto->dpif, &s);
1167 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
1168 put_32aligned_be64(&ots->matched_count,
1169 htonll(s.n_hit + ofproto->n_matches));
1172 static struct ofport *
1175 struct ofport_dpif *port = xmalloc(sizeof *port);
1180 port_dealloc(struct ofport *port_)
1182 struct ofport_dpif *port = ofport_dpif_cast(port_);
1187 port_construct(struct ofport *port_)
1189 struct ofport_dpif *port = ofport_dpif_cast(port_);
1190 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1192 ofproto->need_revalidate = REV_RECONFIGURE;
1193 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
1194 port->bundle = NULL;
1196 port->tag = tag_create_random();
1197 port->may_enable = true;
1198 port->stp_port = NULL;
1199 port->stp_state = STP_DISABLED;
1200 hmap_init(&port->priorities);
1201 port->realdev_ofp_port = 0;
1202 port->vlandev_vid = 0;
1203 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1205 if (ofproto->sflow) {
1206 dpif_sflow_add_port(ofproto->sflow, port_);
1213 port_destruct(struct ofport *port_)
1215 struct ofport_dpif *port = ofport_dpif_cast(port_);
1216 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1218 ofproto->need_revalidate = REV_RECONFIGURE;
1219 bundle_remove(port_);
1220 set_cfm(port_, NULL);
1221 if (ofproto->sflow) {
1222 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1225 ofport_clear_priorities(port);
1226 hmap_destroy(&port->priorities);
1230 port_modified(struct ofport *port_)
1232 struct ofport_dpif *port = ofport_dpif_cast(port_);
1234 if (port->bundle && port->bundle->bond) {
1235 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1240 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1242 struct ofport_dpif *port = ofport_dpif_cast(port_);
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1244 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1246 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1247 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1248 OFPUTIL_PC_NO_PACKET_IN)) {
1249 ofproto->need_revalidate = REV_RECONFIGURE;
1251 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1252 bundle_update(port->bundle);
1258 set_sflow(struct ofproto *ofproto_,
1259 const struct ofproto_sflow_options *sflow_options)
1261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1262 struct dpif_sflow *ds = ofproto->sflow;
1264 if (sflow_options) {
1266 struct ofport_dpif *ofport;
1268 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
1269 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1270 dpif_sflow_add_port(ds, &ofport->up);
1272 ofproto->need_revalidate = REV_RECONFIGURE;
1274 dpif_sflow_set_options(ds, sflow_options);
1277 dpif_sflow_destroy(ds);
1278 ofproto->need_revalidate = REV_RECONFIGURE;
1279 ofproto->sflow = NULL;
1286 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1288 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1295 struct ofproto_dpif *ofproto;
1297 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1298 ofproto->need_revalidate = REV_RECONFIGURE;
1299 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1302 if (cfm_configure(ofport->cfm, s)) {
1308 cfm_destroy(ofport->cfm);
1314 get_cfm_fault(const struct ofport *ofport_)
1316 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1318 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1322 get_cfm_opup(const struct ofport *ofport_)
1324 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1326 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1330 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1333 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1336 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1344 get_cfm_health(const struct ofport *ofport_)
1346 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1348 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1351 /* Spanning Tree. */
1354 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1356 struct ofproto_dpif *ofproto = ofproto_;
1357 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1358 struct ofport_dpif *ofport;
1360 ofport = stp_port_get_aux(sp);
1362 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1363 ofproto->up.name, port_num);
1365 struct eth_header *eth = pkt->l2;
1367 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1368 if (eth_addr_is_zero(eth->eth_src)) {
1369 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1370 "with unknown MAC", ofproto->up.name, port_num);
1372 send_packet(ofport, pkt);
1378 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1380 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1384 /* Only revalidate flows if the configuration changed. */
1385 if (!s != !ofproto->stp) {
1386 ofproto->need_revalidate = REV_RECONFIGURE;
1390 if (!ofproto->stp) {
1391 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1392 send_bpdu_cb, ofproto);
1393 ofproto->stp_last_tick = time_msec();
1396 stp_set_bridge_id(ofproto->stp, s->system_id);
1397 stp_set_bridge_priority(ofproto->stp, s->priority);
1398 stp_set_hello_time(ofproto->stp, s->hello_time);
1399 stp_set_max_age(ofproto->stp, s->max_age);
1400 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1402 struct ofport *ofport;
1404 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1405 set_stp_port(ofport, NULL);
1408 stp_destroy(ofproto->stp);
1409 ofproto->stp = NULL;
1416 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1422 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1423 s->designated_root = stp_get_designated_root(ofproto->stp);
1424 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1433 update_stp_port_state(struct ofport_dpif *ofport)
1435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1436 enum stp_state state;
1438 /* Figure out new state. */
1439 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1443 if (ofport->stp_state != state) {
1444 enum ofputil_port_state of_state;
1447 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1448 netdev_get_name(ofport->up.netdev),
1449 stp_state_name(ofport->stp_state),
1450 stp_state_name(state));
1451 if (stp_learn_in_state(ofport->stp_state)
1452 != stp_learn_in_state(state)) {
1453 /* xxx Learning action flows should also be flushed. */
1454 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1456 fwd_change = stp_forward_in_state(ofport->stp_state)
1457 != stp_forward_in_state(state);
1459 ofproto->need_revalidate = REV_STP;
1460 ofport->stp_state = state;
1461 ofport->stp_state_entered = time_msec();
1463 if (fwd_change && ofport->bundle) {
1464 bundle_update(ofport->bundle);
1467 /* Update the STP state bits in the OpenFlow port description. */
1468 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1469 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1470 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1471 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1472 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1474 ofproto_port_set_state(&ofport->up, of_state);
1478 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1479 * caller is responsible for assigning STP port numbers and ensuring
1480 * there are no duplicates. */
1482 set_stp_port(struct ofport *ofport_,
1483 const struct ofproto_port_stp_settings *s)
1485 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1487 struct stp_port *sp = ofport->stp_port;
1489 if (!s || !s->enable) {
1491 ofport->stp_port = NULL;
1492 stp_port_disable(sp);
1493 update_stp_port_state(ofport);
1496 } else if (sp && stp_port_no(sp) != s->port_num
1497 && ofport == stp_port_get_aux(sp)) {
1498 /* The port-id changed, so disable the old one if it's not
1499 * already in use by another port. */
1500 stp_port_disable(sp);
1503 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1504 stp_port_enable(sp);
1506 stp_port_set_aux(sp, ofport);
1507 stp_port_set_priority(sp, s->priority);
1508 stp_port_set_path_cost(sp, s->path_cost);
1510 update_stp_port_state(ofport);
1516 get_stp_port_status(struct ofport *ofport_,
1517 struct ofproto_port_stp_status *s)
1519 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1521 struct stp_port *sp = ofport->stp_port;
1523 if (!ofproto->stp || !sp) {
1529 s->port_id = stp_port_get_id(sp);
1530 s->state = stp_port_get_state(sp);
1531 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1532 s->role = stp_port_get_role(sp);
1533 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1539 stp_run(struct ofproto_dpif *ofproto)
1542 long long int now = time_msec();
1543 long long int elapsed = now - ofproto->stp_last_tick;
1544 struct stp_port *sp;
1547 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1548 ofproto->stp_last_tick = now;
1550 while (stp_get_changed_port(ofproto->stp, &sp)) {
1551 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1554 update_stp_port_state(ofport);
1558 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1559 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1565 stp_wait(struct ofproto_dpif *ofproto)
1568 poll_timer_wait(1000);
1572 /* Returns true if STP should process 'flow'. */
1574 stp_should_process_flow(const struct flow *flow)
1576 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1580 stp_process_packet(const struct ofport_dpif *ofport,
1581 const struct ofpbuf *packet)
1583 struct ofpbuf payload = *packet;
1584 struct eth_header *eth = payload.data;
1585 struct stp_port *sp = ofport->stp_port;
1587 /* Sink packets on ports that have STP disabled when the bridge has
1589 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1593 /* Trim off padding on payload. */
1594 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1595 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1598 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1599 stp_received_bpdu(sp, payload.data, payload.size);
1603 static struct priority_to_dscp *
1604 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1606 struct priority_to_dscp *pdscp;
1609 hash = hash_int(priority, 0);
1610 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1611 if (pdscp->priority == priority) {
1619 ofport_clear_priorities(struct ofport_dpif *ofport)
1621 struct priority_to_dscp *pdscp, *next;
1623 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1624 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1630 set_queues(struct ofport *ofport_,
1631 const struct ofproto_port_queue *qdscp_list,
1634 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1635 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1636 struct hmap new = HMAP_INITIALIZER(&new);
1639 for (i = 0; i < n_qdscp; i++) {
1640 struct priority_to_dscp *pdscp;
1644 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1645 if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue,
1650 pdscp = get_priority(ofport, priority);
1652 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1654 pdscp = xmalloc(sizeof *pdscp);
1655 pdscp->priority = priority;
1657 ofproto->need_revalidate = REV_RECONFIGURE;
1660 if (pdscp->dscp != dscp) {
1662 ofproto->need_revalidate = REV_RECONFIGURE;
1665 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1668 if (!hmap_is_empty(&ofport->priorities)) {
1669 ofport_clear_priorities(ofport);
1670 ofproto->need_revalidate = REV_RECONFIGURE;
1673 hmap_swap(&new, &ofport->priorities);
1681 /* Expires all MAC learning entries associated with 'bundle' and forces its
1682 * ofproto to revalidate every flow.
1684 * Normally MAC learning entries are removed only from the ofproto associated
1685 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1686 * are removed from every ofproto. When patch ports and SLB bonds are in use
1687 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1688 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1689 * with the host from which it migrated. */
1691 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1693 struct ofproto_dpif *ofproto = bundle->ofproto;
1694 struct mac_learning *ml = ofproto->ml;
1695 struct mac_entry *mac, *next_mac;
1697 ofproto->need_revalidate = REV_RECONFIGURE;
1698 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1699 if (mac->port.p == bundle) {
1701 struct ofproto_dpif *o;
1703 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1705 struct mac_entry *e;
1707 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
1710 tag_set_add(&o->revalidate_set, e->tag);
1711 mac_learning_expire(o->ml, e);
1717 mac_learning_expire(ml, mac);
1722 static struct ofbundle *
1723 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1725 struct ofbundle *bundle;
1727 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1728 &ofproto->bundles) {
1729 if (bundle->aux == aux) {
1736 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1737 * ones that are found to 'bundles'. */
1739 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1740 void **auxes, size_t n_auxes,
1741 struct hmapx *bundles)
1745 hmapx_init(bundles);
1746 for (i = 0; i < n_auxes; i++) {
1747 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1749 hmapx_add(bundles, bundle);
1755 bundle_update(struct ofbundle *bundle)
1757 struct ofport_dpif *port;
1759 bundle->floodable = true;
1760 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1761 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1762 || !stp_forward_in_state(port->stp_state)) {
1763 bundle->floodable = false;
1770 bundle_del_port(struct ofport_dpif *port)
1772 struct ofbundle *bundle = port->bundle;
1774 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1776 list_remove(&port->bundle_node);
1777 port->bundle = NULL;
1780 lacp_slave_unregister(bundle->lacp, port);
1783 bond_slave_unregister(bundle->bond, port);
1786 bundle_update(bundle);
1790 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1791 struct lacp_slave_settings *lacp,
1792 uint32_t bond_stable_id)
1794 struct ofport_dpif *port;
1796 port = get_ofp_port(bundle->ofproto, ofp_port);
1801 if (port->bundle != bundle) {
1802 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1804 bundle_del_port(port);
1807 port->bundle = bundle;
1808 list_push_back(&bundle->ports, &port->bundle_node);
1809 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1810 || !stp_forward_in_state(port->stp_state)) {
1811 bundle->floodable = false;
1815 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
1816 lacp_slave_register(bundle->lacp, port, lacp);
1819 port->bond_stable_id = bond_stable_id;
1825 bundle_destroy(struct ofbundle *bundle)
1827 struct ofproto_dpif *ofproto;
1828 struct ofport_dpif *port, *next_port;
1835 ofproto = bundle->ofproto;
1836 for (i = 0; i < MAX_MIRRORS; i++) {
1837 struct ofmirror *m = ofproto->mirrors[i];
1839 if (m->out == bundle) {
1841 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1842 || hmapx_find_and_delete(&m->dsts, bundle)) {
1843 ofproto->need_revalidate = REV_RECONFIGURE;
1848 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1849 bundle_del_port(port);
1852 bundle_flush_macs(bundle, true);
1853 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1855 free(bundle->trunks);
1856 lacp_destroy(bundle->lacp);
1857 bond_destroy(bundle->bond);
1862 bundle_set(struct ofproto *ofproto_, void *aux,
1863 const struct ofproto_bundle_settings *s)
1865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1866 bool need_flush = false;
1867 struct ofport_dpif *port;
1868 struct ofbundle *bundle;
1869 unsigned long *trunks;
1875 bundle_destroy(bundle_lookup(ofproto, aux));
1879 assert(s->n_slaves == 1 || s->bond != NULL);
1880 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1882 bundle = bundle_lookup(ofproto, aux);
1884 bundle = xmalloc(sizeof *bundle);
1886 bundle->ofproto = ofproto;
1887 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1888 hash_pointer(aux, 0));
1890 bundle->name = NULL;
1892 list_init(&bundle->ports);
1893 bundle->vlan_mode = PORT_VLAN_TRUNK;
1895 bundle->trunks = NULL;
1896 bundle->use_priority_tags = s->use_priority_tags;
1897 bundle->lacp = NULL;
1898 bundle->bond = NULL;
1900 bundle->floodable = true;
1902 bundle->src_mirrors = 0;
1903 bundle->dst_mirrors = 0;
1904 bundle->mirror_out = 0;
1907 if (!bundle->name || strcmp(s->name, bundle->name)) {
1909 bundle->name = xstrdup(s->name);
1914 if (!bundle->lacp) {
1915 ofproto->need_revalidate = REV_RECONFIGURE;
1916 bundle->lacp = lacp_create();
1918 lacp_configure(bundle->lacp, s->lacp);
1920 lacp_destroy(bundle->lacp);
1921 bundle->lacp = NULL;
1924 /* Update set of ports. */
1926 for (i = 0; i < s->n_slaves; i++) {
1927 if (!bundle_add_port(bundle, s->slaves[i],
1928 s->lacp ? &s->lacp_slaves[i] : NULL,
1929 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1933 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1934 struct ofport_dpif *next_port;
1936 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1937 for (i = 0; i < s->n_slaves; i++) {
1938 if (s->slaves[i] == port->up.ofp_port) {
1943 bundle_del_port(port);
1947 assert(list_size(&bundle->ports) <= s->n_slaves);
1949 if (list_is_empty(&bundle->ports)) {
1950 bundle_destroy(bundle);
1954 /* Set VLAN tagging mode */
1955 if (s->vlan_mode != bundle->vlan_mode
1956 || s->use_priority_tags != bundle->use_priority_tags) {
1957 bundle->vlan_mode = s->vlan_mode;
1958 bundle->use_priority_tags = s->use_priority_tags;
1963 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1964 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1966 if (vlan != bundle->vlan) {
1967 bundle->vlan = vlan;
1971 /* Get trunked VLANs. */
1972 switch (s->vlan_mode) {
1973 case PORT_VLAN_ACCESS:
1977 case PORT_VLAN_TRUNK:
1978 trunks = CONST_CAST(unsigned long *, s->trunks);
1981 case PORT_VLAN_NATIVE_UNTAGGED:
1982 case PORT_VLAN_NATIVE_TAGGED:
1983 if (vlan != 0 && (!s->trunks
1984 || !bitmap_is_set(s->trunks, vlan)
1985 || bitmap_is_set(s->trunks, 0))) {
1986 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1988 trunks = bitmap_clone(s->trunks, 4096);
1990 trunks = bitmap_allocate1(4096);
1992 bitmap_set1(trunks, vlan);
1993 bitmap_set0(trunks, 0);
1995 trunks = CONST_CAST(unsigned long *, s->trunks);
2002 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2003 free(bundle->trunks);
2004 if (trunks == s->trunks) {
2005 bundle->trunks = vlan_bitmap_clone(trunks);
2007 bundle->trunks = trunks;
2012 if (trunks != s->trunks) {
2017 if (!list_is_short(&bundle->ports)) {
2018 bundle->ofproto->has_bonded_bundles = true;
2020 if (bond_reconfigure(bundle->bond, s->bond)) {
2021 ofproto->need_revalidate = REV_RECONFIGURE;
2024 bundle->bond = bond_create(s->bond);
2025 ofproto->need_revalidate = REV_RECONFIGURE;
2028 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2029 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2033 bond_destroy(bundle->bond);
2034 bundle->bond = NULL;
2037 /* If we changed something that would affect MAC learning, un-learn
2038 * everything on this port and force flow revalidation. */
2040 bundle_flush_macs(bundle, false);
2047 bundle_remove(struct ofport *port_)
2049 struct ofport_dpif *port = ofport_dpif_cast(port_);
2050 struct ofbundle *bundle = port->bundle;
2053 bundle_del_port(port);
2054 if (list_is_empty(&bundle->ports)) {
2055 bundle_destroy(bundle);
2056 } else if (list_is_short(&bundle->ports)) {
2057 bond_destroy(bundle->bond);
2058 bundle->bond = NULL;
2064 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2067 struct ofport_dpif *port = port_;
2068 uint8_t ea[ETH_ADDR_LEN];
2071 error = netdev_get_etheraddr(port->up.netdev, ea);
2073 struct ofpbuf packet;
2076 ofpbuf_init(&packet, 0);
2077 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2079 memcpy(packet_pdu, pdu, pdu_size);
2081 send_packet(port, &packet);
2082 ofpbuf_uninit(&packet);
2084 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2085 "%s (%s)", port->bundle->name,
2086 netdev_get_name(port->up.netdev), strerror(error));
2091 bundle_send_learning_packets(struct ofbundle *bundle)
2093 struct ofproto_dpif *ofproto = bundle->ofproto;
2094 int error, n_packets, n_errors;
2095 struct mac_entry *e;
2097 error = n_packets = n_errors = 0;
2098 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2099 if (e->port.p != bundle) {
2100 struct ofpbuf *learning_packet;
2101 struct ofport_dpif *port;
2105 /* The assignment to "port" is unnecessary but makes "grep"ing for
2106 * struct ofport_dpif more effective. */
2107 learning_packet = bond_compose_learning_packet(bundle->bond,
2111 ret = send_packet(port, learning_packet);
2112 ofpbuf_delete(learning_packet);
2122 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2123 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2124 "packets, last error was: %s",
2125 bundle->name, n_errors, n_packets, strerror(error));
2127 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2128 bundle->name, n_packets);
2133 bundle_run(struct ofbundle *bundle)
2136 lacp_run(bundle->lacp, send_pdu_cb);
2139 struct ofport_dpif *port;
2141 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2142 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2145 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2146 lacp_status(bundle->lacp));
2147 if (bond_should_send_learning_packets(bundle->bond)) {
2148 bundle_send_learning_packets(bundle);
2154 bundle_wait(struct ofbundle *bundle)
2157 lacp_wait(bundle->lacp);
2160 bond_wait(bundle->bond);
2167 mirror_scan(struct ofproto_dpif *ofproto)
2171 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2172 if (!ofproto->mirrors[idx]) {
2179 static struct ofmirror *
2180 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2184 for (i = 0; i < MAX_MIRRORS; i++) {
2185 struct ofmirror *mirror = ofproto->mirrors[i];
2186 if (mirror && mirror->aux == aux) {
2194 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2196 mirror_update_dups(struct ofproto_dpif *ofproto)
2200 for (i = 0; i < MAX_MIRRORS; i++) {
2201 struct ofmirror *m = ofproto->mirrors[i];
2204 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2208 for (i = 0; i < MAX_MIRRORS; i++) {
2209 struct ofmirror *m1 = ofproto->mirrors[i];
2216 for (j = i + 1; j < MAX_MIRRORS; j++) {
2217 struct ofmirror *m2 = ofproto->mirrors[j];
2219 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2220 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2221 m2->dup_mirrors |= m1->dup_mirrors;
2228 mirror_set(struct ofproto *ofproto_, void *aux,
2229 const struct ofproto_mirror_settings *s)
2231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2232 mirror_mask_t mirror_bit;
2233 struct ofbundle *bundle;
2234 struct ofmirror *mirror;
2235 struct ofbundle *out;
2236 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2237 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2240 mirror = mirror_lookup(ofproto, aux);
2242 mirror_destroy(mirror);
2248 idx = mirror_scan(ofproto);
2250 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2252 ofproto->up.name, MAX_MIRRORS, s->name);
2256 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2257 mirror->ofproto = ofproto;
2260 mirror->out_vlan = -1;
2261 mirror->name = NULL;
2264 if (!mirror->name || strcmp(s->name, mirror->name)) {
2266 mirror->name = xstrdup(s->name);
2269 /* Get the new configuration. */
2270 if (s->out_bundle) {
2271 out = bundle_lookup(ofproto, s->out_bundle);
2273 mirror_destroy(mirror);
2279 out_vlan = s->out_vlan;
2281 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2282 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2284 /* If the configuration has not changed, do nothing. */
2285 if (hmapx_equals(&srcs, &mirror->srcs)
2286 && hmapx_equals(&dsts, &mirror->dsts)
2287 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2288 && mirror->out == out
2289 && mirror->out_vlan == out_vlan)
2291 hmapx_destroy(&srcs);
2292 hmapx_destroy(&dsts);
2296 hmapx_swap(&srcs, &mirror->srcs);
2297 hmapx_destroy(&srcs);
2299 hmapx_swap(&dsts, &mirror->dsts);
2300 hmapx_destroy(&dsts);
2302 free(mirror->vlans);
2303 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2306 mirror->out_vlan = out_vlan;
2308 /* Update bundles. */
2309 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2310 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2311 if (hmapx_contains(&mirror->srcs, bundle)) {
2312 bundle->src_mirrors |= mirror_bit;
2314 bundle->src_mirrors &= ~mirror_bit;
2317 if (hmapx_contains(&mirror->dsts, bundle)) {
2318 bundle->dst_mirrors |= mirror_bit;
2320 bundle->dst_mirrors &= ~mirror_bit;
2323 if (mirror->out == bundle) {
2324 bundle->mirror_out |= mirror_bit;
2326 bundle->mirror_out &= ~mirror_bit;
2330 ofproto->need_revalidate = REV_RECONFIGURE;
2331 ofproto->has_mirrors = true;
2332 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2333 mirror_update_dups(ofproto);
2339 mirror_destroy(struct ofmirror *mirror)
2341 struct ofproto_dpif *ofproto;
2342 mirror_mask_t mirror_bit;
2343 struct ofbundle *bundle;
2350 ofproto = mirror->ofproto;
2351 ofproto->need_revalidate = REV_RECONFIGURE;
2352 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2354 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2355 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2356 bundle->src_mirrors &= ~mirror_bit;
2357 bundle->dst_mirrors &= ~mirror_bit;
2358 bundle->mirror_out &= ~mirror_bit;
2361 hmapx_destroy(&mirror->srcs);
2362 hmapx_destroy(&mirror->dsts);
2363 free(mirror->vlans);
2365 ofproto->mirrors[mirror->idx] = NULL;
2369 mirror_update_dups(ofproto);
2371 ofproto->has_mirrors = false;
2372 for (i = 0; i < MAX_MIRRORS; i++) {
2373 if (ofproto->mirrors[i]) {
2374 ofproto->has_mirrors = true;
2381 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2382 uint64_t *packets, uint64_t *bytes)
2384 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2385 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2388 *packets = *bytes = UINT64_MAX;
2392 *packets = mirror->packet_count;
2393 *bytes = mirror->byte_count;
2399 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2401 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2402 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2403 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2409 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2412 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2413 return bundle && bundle->mirror_out != 0;
2417 forward_bpdu_changed(struct ofproto *ofproto_)
2419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2420 ofproto->need_revalidate = REV_RECONFIGURE;
2424 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2426 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2427 mac_learning_set_idle_time(ofproto->ml, idle_time);
2432 static struct ofport_dpif *
2433 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2435 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2436 return ofport ? ofport_dpif_cast(ofport) : NULL;
2439 static struct ofport_dpif *
2440 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2442 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
2446 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
2447 struct dpif_port *dpif_port)
2449 ofproto_port->name = dpif_port->name;
2450 ofproto_port->type = dpif_port->type;
2451 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
2455 port_run_fast(struct ofport_dpif *ofport)
2457 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2458 struct ofpbuf packet;
2460 ofpbuf_init(&packet, 0);
2461 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2462 send_packet(ofport, &packet);
2463 ofpbuf_uninit(&packet);
2468 port_run(struct ofport_dpif *ofport)
2470 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2471 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2472 bool enable = netdev_get_carrier(ofport->up.netdev);
2474 ofport->carrier_seq = carrier_seq;
2476 port_run_fast(ofport);
2478 int cfm_opup = cfm_get_opup(ofport->cfm);
2480 cfm_run(ofport->cfm);
2481 enable = enable && !cfm_get_fault(ofport->cfm);
2483 if (cfm_opup >= 0) {
2484 enable = enable && cfm_opup;
2488 if (ofport->bundle) {
2489 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2490 if (carrier_changed) {
2491 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2495 if (ofport->may_enable != enable) {
2496 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2498 if (ofproto->has_bundle_action) {
2499 ofproto->need_revalidate = REV_PORT_TOGGLED;
2503 ofport->may_enable = enable;
2507 port_wait(struct ofport_dpif *ofport)
2510 cfm_wait(ofport->cfm);
2515 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2516 struct ofproto_port *ofproto_port)
2518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2519 struct dpif_port dpif_port;
2522 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
2524 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
2530 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
2532 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2533 uint16_t odp_port = UINT16_MAX;
2536 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
2538 *ofp_portp = odp_port_to_ofp_port(odp_port);
2544 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2546 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2549 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
2551 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2553 /* The caller is going to close ofport->up.netdev. If this is a
2554 * bonded port, then the bond is using that netdev, so remove it
2555 * from the bond. The client will need to reconfigure everything
2556 * after deleting ports, so then the slave will get re-added. */
2557 bundle_remove(&ofport->up);
2564 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2566 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2569 error = netdev_get_stats(ofport->up.netdev, stats);
2571 if (!error && ofport->odp_port == OVSP_LOCAL) {
2572 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2574 /* ofproto->stats.tx_packets represents packets that we created
2575 * internally and sent to some port (e.g. packets sent with
2576 * send_packet()). Account for them as if they had come from
2577 * OFPP_LOCAL and got forwarded. */
2579 if (stats->rx_packets != UINT64_MAX) {
2580 stats->rx_packets += ofproto->stats.tx_packets;
2583 if (stats->rx_bytes != UINT64_MAX) {
2584 stats->rx_bytes += ofproto->stats.tx_bytes;
2587 /* ofproto->stats.rx_packets represents packets that were received on
2588 * some port and we processed internally and dropped (e.g. STP).
2589 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2591 if (stats->tx_packets != UINT64_MAX) {
2592 stats->tx_packets += ofproto->stats.rx_packets;
2595 if (stats->tx_bytes != UINT64_MAX) {
2596 stats->tx_bytes += ofproto->stats.rx_bytes;
2603 /* Account packets for LOCAL port. */
2605 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2606 size_t tx_size, size_t rx_size)
2608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2611 ofproto->stats.rx_packets++;
2612 ofproto->stats.rx_bytes += rx_size;
2615 ofproto->stats.tx_packets++;
2616 ofproto->stats.tx_bytes += tx_size;
2620 struct port_dump_state {
2621 struct dpif_port_dump dump;
2626 port_dump_start(const struct ofproto *ofproto_, void **statep)
2628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2629 struct port_dump_state *state;
2631 *statep = state = xmalloc(sizeof *state);
2632 dpif_port_dump_start(&state->dump, ofproto->dpif);
2633 state->done = false;
2638 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2639 struct ofproto_port *port)
2641 struct port_dump_state *state = state_;
2642 struct dpif_port dpif_port;
2644 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2645 ofproto_port_from_dpif_port(port, &dpif_port);
2648 int error = dpif_port_dump_done(&state->dump);
2650 return error ? error : EOF;
2655 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2657 struct port_dump_state *state = state_;
2660 dpif_port_dump_done(&state->dump);
2667 port_poll(const struct ofproto *ofproto_, char **devnamep)
2669 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2670 return dpif_port_poll(ofproto->dpif, devnamep);
2674 port_poll_wait(const struct ofproto *ofproto_)
2676 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2677 dpif_port_poll_wait(ofproto->dpif);
2681 port_is_lacp_current(const struct ofport *ofport_)
2683 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2684 return (ofport->bundle && ofport->bundle->lacp
2685 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2689 /* Upcall handling. */
2691 /* Flow miss batching.
2693 * Some dpifs implement operations faster when you hand them off in a batch.
2694 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2695 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2696 * more packets, plus possibly installing the flow in the dpif.
2698 * So far we only batch the operations that affect flow setup time the most.
2699 * It's possible to batch more than that, but the benefit might be minimal. */
2701 struct hmap_node hmap_node;
2703 enum odp_key_fitness key_fitness;
2704 const struct nlattr *key;
2706 ovs_be16 initial_tci;
2707 struct list packets;
2708 enum dpif_upcall_type upcall_type;
2711 struct flow_miss_op {
2712 struct dpif_op dpif_op;
2713 struct subfacet *subfacet; /* Subfacet */
2714 void *garbage; /* Pointer to pass to free(), NULL if none. */
2715 uint64_t stub[1024 / 8]; /* Temporary buffer. */
2718 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2719 * OpenFlow controller as necessary according to their individual
2720 * configurations. */
2722 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
2723 const struct flow *flow)
2725 struct ofputil_packet_in pin;
2727 pin.packet = packet->data;
2728 pin.packet_len = packet->size;
2729 pin.reason = OFPR_NO_MATCH;
2730 pin.controller_id = 0;
2735 pin.send_len = 0; /* not used for flow table misses */
2737 flow_get_metadata(flow, &pin.fmd);
2739 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
2742 static enum slow_path_reason
2743 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2744 const struct ofpbuf *packet)
2746 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2752 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2754 cfm_process_heartbeat(ofport->cfm, packet);
2757 } else if (ofport->bundle && ofport->bundle->lacp
2758 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2760 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2763 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2765 stp_process_packet(ofport, packet);
2772 static struct flow_miss *
2773 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
2775 struct flow_miss *miss;
2777 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2778 if (flow_equal(&miss->flow, flow)) {
2786 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
2787 * 'packet'. The caller must initialize op->actions and op->actions_len. If
2788 * 'miss' is associated with a subfacet the caller must also initialize the
2789 * returned op->subfacet, and if anything needs to be freed after processing
2790 * the op, the caller must initialize op->garbage also. */
2792 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
2793 struct flow_miss_op *op)
2795 if (miss->flow.vlan_tci != miss->initial_tci) {
2796 /* This packet was received on a VLAN splinter port. We
2797 * added a VLAN to the packet to make the packet resemble
2798 * the flow, but the actions were composed assuming that
2799 * the packet contained no VLAN. So, we must remove the
2800 * VLAN header from the packet before trying to execute the
2802 eth_pop_vlan(packet);
2805 op->subfacet = NULL;
2807 op->dpif_op.type = DPIF_OP_EXECUTE;
2808 op->dpif_op.u.execute.key = miss->key;
2809 op->dpif_op.u.execute.key_len = miss->key_len;
2810 op->dpif_op.u.execute.packet = packet;
2813 /* Helper for handle_flow_miss_without_facet() and
2814 * handle_flow_miss_with_facet(). */
2816 handle_flow_miss_common(struct rule_dpif *rule,
2817 struct ofpbuf *packet, const struct flow *flow)
2819 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2821 ofproto->n_matches++;
2823 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2825 * Extra-special case for fail-open mode.
2827 * We are in fail-open mode and the packet matched the fail-open
2828 * rule, but we are connected to a controller too. We should send
2829 * the packet up to the controller in the hope that it will try to
2830 * set up a flow and thereby allow us to exit fail-open.
2832 * See the top-level comment in fail-open.c for more information.
2834 send_packet_in_miss(ofproto, packet, flow);
2838 /* Figures out whether a flow that missed in 'ofproto', whose details are in
2839 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
2840 * installing a datapath flow. The answer is usually "yes" (a return value of
2841 * true). However, for short flows the cost of bookkeeping is much higher than
2842 * the benefits, so when the datapath holds a large number of flows we impose
2843 * some heuristics to decide which flows are likely to be worth tracking. */
2845 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
2846 struct flow_miss *miss, uint32_t hash)
2848 if (!ofproto->governor) {
2851 n_subfacets = hmap_count(&ofproto->subfacets);
2852 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
2856 ofproto->governor = governor_create(ofproto->up.name);
2859 return governor_should_install_flow(ofproto->governor, hash,
2860 list_size(&miss->packets));
2863 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
2864 * or creating any datapath flow. May add an "execute" operation to 'ops' and
2865 * increment '*n_ops'. */
2867 handle_flow_miss_without_facet(struct flow_miss *miss,
2868 struct rule_dpif *rule,
2869 struct flow_miss_op *ops, size_t *n_ops)
2871 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2872 long long int now = time_msec();
2873 struct action_xlate_ctx ctx;
2874 struct ofpbuf *packet;
2876 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2877 struct flow_miss_op *op = &ops[*n_ops];
2878 struct dpif_flow_stats stats;
2879 struct ofpbuf odp_actions;
2881 COVERAGE_INC(facet_suppress);
2883 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2885 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
2886 rule_credit_stats(rule, &stats);
2888 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
2890 ctx.resubmit_stats = &stats;
2891 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
2894 if (odp_actions.size) {
2895 struct dpif_execute *execute = &op->dpif_op.u.execute;
2897 init_flow_miss_execute_op(miss, packet, op);
2898 execute->actions = odp_actions.data;
2899 execute->actions_len = odp_actions.size;
2900 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2904 ofpbuf_uninit(&odp_actions);
2909 /* Handles 'miss', which matches 'facet'. May add any required datapath
2910 * operations to 'ops', incrementing '*n_ops' for each new op.
2912 * All of the packets in 'miss' are considered to have arrived at time 'now'.
2913 * This is really important only for new facets: if we just called time_msec()
2914 * here, then the new subfacet or its packets could look (occasionally) as
2915 * though it was used some time after the facet was used. That can make a
2916 * one-packet flow look like it has a nonzero duration, which looks odd in
2917 * e.g. NetFlow statistics. */
2919 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
2921 struct flow_miss_op *ops, size_t *n_ops)
2923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
2924 enum subfacet_path want_path;
2925 struct subfacet *subfacet;
2926 struct ofpbuf *packet;
2928 subfacet = subfacet_create(facet,
2929 miss->key_fitness, miss->key, miss->key_len,
2930 miss->initial_tci, now);
2932 LIST_FOR_EACH (packet, list_node, &miss->packets) {
2933 struct flow_miss_op *op = &ops[*n_ops];
2934 struct dpif_flow_stats stats;
2935 struct ofpbuf odp_actions;
2937 handle_flow_miss_common(facet->rule, packet, &miss->flow);
2939 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
2940 if (!subfacet->actions || subfacet->slow) {
2941 subfacet_make_actions(subfacet, packet, &odp_actions);
2944 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
2945 subfacet_update_stats(subfacet, &stats);
2947 if (subfacet->actions_len) {
2948 struct dpif_execute *execute = &op->dpif_op.u.execute;
2950 init_flow_miss_execute_op(miss, packet, op);
2951 op->subfacet = subfacet;
2952 if (!subfacet->slow) {
2953 execute->actions = subfacet->actions;
2954 execute->actions_len = subfacet->actions_len;
2955 ofpbuf_uninit(&odp_actions);
2957 execute->actions = odp_actions.data;
2958 execute->actions_len = odp_actions.size;
2959 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
2964 ofpbuf_uninit(&odp_actions);
2968 want_path = subfacet_want_path(subfacet->slow);
2969 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
2970 struct flow_miss_op *op = &ops[(*n_ops)++];
2971 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
2973 op->subfacet = subfacet;
2975 op->dpif_op.type = DPIF_OP_FLOW_PUT;
2976 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2977 put->key = miss->key;
2978 put->key_len = miss->key_len;
2979 if (want_path == SF_FAST_PATH) {
2980 put->actions = subfacet->actions;
2981 put->actions_len = subfacet->actions_len;
2983 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
2984 op->stub, sizeof op->stub,
2985 &put->actions, &put->actions_len);
2991 /* Handles flow miss 'miss' on 'ofproto'. May add any required datapath
2992 * operations to 'ops', incrementing '*n_ops' for each new op. */
2994 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2995 struct flow_miss_op *ops, size_t *n_ops)
2997 struct facet *facet;
3001 /* The caller must ensure that miss->hmap_node.hash contains
3002 * flow_hash(miss->flow, 0). */
3003 hash = miss->hmap_node.hash;
3005 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3007 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3009 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3010 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3014 facet = facet_create(rule, &miss->flow, hash);
3019 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3022 /* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of
3023 * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns
3024 * an ODP_FIT_* value that indicates how well 'key' fits our expectations for
3025 * what a flow key should contain.
3027 * This function also includes some logic to help make VLAN splinters
3028 * transparent to the rest of the upcall processing logic. In particular, if
3029 * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by
3030 * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN
3031 * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull).
3033 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3034 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3035 * (This differs from the value returned in flow->vlan_tci only for packets
3036 * received on VLAN splinters.)
3038 static enum odp_key_fitness
3039 ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto,
3040 const struct nlattr *key, size_t key_len,
3041 struct flow *flow, ovs_be16 *initial_tci,
3042 struct ofpbuf *packet)
3044 enum odp_key_fitness fitness;
3046 fitness = odp_flow_key_to_flow(key, key_len, flow);
3047 if (fitness == ODP_FIT_ERROR) {
3050 *initial_tci = flow->vlan_tci;
3052 if (vsp_adjust_flow(ofproto, flow)) {
3054 /* Make the packet resemble the flow, so that it gets sent to an
3055 * OpenFlow controller properly, so that it looks correct for
3056 * sFlow, and so that flow_extract() will get the correct vlan_tci
3057 * if it is called on 'packet'.
3059 * The allocated space inside 'packet' probably also contains
3060 * 'key', that is, both 'packet' and 'key' are probably part of a
3061 * struct dpif_upcall (see the large comment on that structure
3062 * definition), so pushing data on 'packet' is in general not a
3063 * good idea since it could overwrite 'key' or free it as a side
3064 * effect. However, it's OK in this special case because we know
3065 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3066 * will just overwrite the 4-byte "struct nlattr", which is fine
3067 * since we don't need that header anymore. */
3068 eth_push_vlan(packet, flow->vlan_tci);
3071 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3072 if (fitness == ODP_FIT_PERFECT) {
3073 fitness = ODP_FIT_TOO_MUCH;
3081 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
3084 struct dpif_upcall *upcall;
3085 struct flow_miss *miss;
3086 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3087 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3088 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3098 /* Construct the to-do list.
3100 * This just amounts to extracting the flow from each packet and sticking
3101 * the packets that have the same flow in the same "flow_miss" structure so
3102 * that we can process them together. */
3105 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3106 struct flow_miss *miss = &misses[n_misses];
3107 struct flow_miss *existing_miss;
3110 /* Obtain metadata and check userspace/kernel agreement on flow match,
3111 * then set 'flow''s header pointers. */
3112 miss->key_fitness = ofproto_dpif_extract_flow_key(
3113 ofproto, upcall->key, upcall->key_len,
3114 &miss->flow, &miss->initial_tci, upcall->packet);
3115 if (miss->key_fitness == ODP_FIT_ERROR) {
3118 flow_extract(upcall->packet, miss->flow.skb_priority,
3119 miss->flow.tun_id, miss->flow.in_port, &miss->flow);
3121 /* Add other packets to a to-do list. */
3122 hash = flow_hash(&miss->flow, 0);
3123 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3124 if (!existing_miss) {
3125 hmap_insert(&todo, &miss->hmap_node, hash);
3126 miss->key = upcall->key;
3127 miss->key_len = upcall->key_len;
3128 miss->upcall_type = upcall->type;
3129 list_init(&miss->packets);
3133 miss = existing_miss;
3135 list_push_back(&miss->packets, &upcall->packet->list_node);
3138 /* Process each element in the to-do list, constructing the set of
3139 * operations to batch. */
3141 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3142 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
3144 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3146 /* Execute batch. */
3147 for (i = 0; i < n_ops; i++) {
3148 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3150 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
3152 /* Free memory and update facets. */
3153 for (i = 0; i < n_ops; i++) {
3154 struct flow_miss_op *op = &flow_miss_ops[i];
3156 switch (op->dpif_op.type) {
3157 case DPIF_OP_EXECUTE:
3160 case DPIF_OP_FLOW_PUT:
3161 if (!op->dpif_op.error) {
3162 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3166 case DPIF_OP_FLOW_DEL:
3172 hmap_destroy(&todo);
3175 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3176 classify_upcall(const struct dpif_upcall *upcall)
3178 union user_action_cookie cookie;
3180 /* First look at the upcall type. */
3181 switch (upcall->type) {
3182 case DPIF_UC_ACTION:
3188 case DPIF_N_UC_TYPES:
3190 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3194 /* "action" upcalls need a closer look. */
3195 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3196 switch (cookie.type) {
3197 case USER_ACTION_COOKIE_SFLOW:
3198 return SFLOW_UPCALL;
3200 case USER_ACTION_COOKIE_SLOW_PATH:
3203 case USER_ACTION_COOKIE_UNSPEC:
3205 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3211 handle_sflow_upcall(struct ofproto_dpif *ofproto,
3212 const struct dpif_upcall *upcall)
3214 union user_action_cookie cookie;
3215 enum odp_key_fitness fitness;
3216 ovs_be16 initial_tci;
3219 fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key,
3220 upcall->key_len, &flow,
3221 &initial_tci, upcall->packet);
3222 if (fitness == ODP_FIT_ERROR) {
3226 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3227 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
3231 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
3233 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3234 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3235 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3240 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3243 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3244 struct dpif_upcall *upcall = &misses[n_misses];
3245 struct ofpbuf *buf = &miss_bufs[n_misses];
3248 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3249 sizeof miss_buf_stubs[n_misses]);
3250 error = dpif_recv(ofproto->dpif, upcall, buf);
3256 switch (classify_upcall(upcall)) {
3258 /* Handle it later. */
3263 if (ofproto->sflow) {
3264 handle_sflow_upcall(ofproto, upcall);
3275 /* Handle deferred MISS_UPCALL processing. */
3276 handle_miss_upcalls(ofproto, misses, n_misses);
3277 for (i = 0; i < n_misses; i++) {
3278 ofpbuf_uninit(&miss_bufs[i]);
3284 /* Flow expiration. */
3286 static int subfacet_max_idle(const struct ofproto_dpif *);
3287 static void update_stats(struct ofproto_dpif *);
3288 static void rule_expire(struct rule_dpif *);
3289 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3291 /* This function is called periodically by run(). Its job is to collect
3292 * updates for the flows that have been installed into the datapath, most
3293 * importantly when they last were used, and then use that information to
3294 * expire flows that have not been used recently.
3296 * Returns the number of milliseconds after which it should be called again. */
3298 expire(struct ofproto_dpif *ofproto)
3300 struct rule_dpif *rule, *next_rule;
3301 struct oftable *table;
3304 /* Update stats for each flow in the datapath. */
3305 update_stats(ofproto);
3307 /* Expire subfacets that have been idle too long. */
3308 dp_max_idle = subfacet_max_idle(ofproto);
3309 expire_subfacets(ofproto, dp_max_idle);
3311 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
3312 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3313 struct cls_cursor cursor;
3315 cls_cursor_init(&cursor, &table->cls, NULL);
3316 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3321 /* All outstanding data in existing flows has been accounted, so it's a
3322 * good time to do bond rebalancing. */
3323 if (ofproto->has_bonded_bundles) {
3324 struct ofbundle *bundle;
3326 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3328 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3333 return MIN(dp_max_idle, 1000);
3336 /* Updates flow table statistics given that the datapath just reported 'stats'
3337 * as 'subfacet''s statistics. */
3339 update_subfacet_stats(struct subfacet *subfacet,
3340 const struct dpif_flow_stats *stats)
3342 struct facet *facet = subfacet->facet;
3344 if (stats->n_packets >= subfacet->dp_packet_count) {
3345 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3346 facet->packet_count += extra;
3348 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3351 if (stats->n_bytes >= subfacet->dp_byte_count) {
3352 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3354 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3357 subfacet->dp_packet_count = stats->n_packets;
3358 subfacet->dp_byte_count = stats->n_bytes;
3360 facet->tcp_flags |= stats->tcp_flags;
3362 subfacet_update_time(subfacet, stats->used);
3363 if (facet->accounted_bytes < facet->byte_count) {
3365 facet_account(facet);
3366 facet->accounted_bytes = facet->byte_count;
3368 facet_push_stats(facet);
3371 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3372 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3374 delete_unexpected_flow(struct dpif *dpif,
3375 const struct nlattr *key, size_t key_len)
3377 if (!VLOG_DROP_WARN(&rl)) {
3381 odp_flow_key_format(key, key_len, &s);
3382 VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s));
3386 COVERAGE_INC(facet_unexpected);
3387 dpif_flow_del(dpif, key, key_len, NULL);
3390 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3392 * This function also pushes statistics updates to rules which each facet
3393 * resubmits into. Generally these statistics will be accurate. However, if a
3394 * facet changes the rule it resubmits into at some time in between
3395 * update_stats() runs, it is possible that statistics accrued to the
3396 * old rule will be incorrectly attributed to the new rule. This could be
3397 * avoided by calling update_stats() whenever rules are created or
3398 * deleted. However, the performance impact of making so many calls to the
3399 * datapath do not justify the benefit of having perfectly accurate statistics.
3402 update_stats(struct ofproto_dpif *p)
3404 const struct dpif_flow_stats *stats;
3405 struct dpif_flow_dump dump;
3406 const struct nlattr *key;
3409 dpif_flow_dump_start(&dump, p->dpif);
3410 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3411 struct subfacet *subfacet;
3413 subfacet = subfacet_find(p, key, key_len);
3414 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3416 update_subfacet_stats(subfacet, stats);
3420 /* Stats are updated per-packet. */
3423 case SF_NOT_INSTALLED:
3425 delete_unexpected_flow(p->dpif, key, key_len);
3429 dpif_flow_dump_done(&dump);
3432 /* Calculates and returns the number of milliseconds of idle time after which
3433 * subfacets should expire from the datapath. When a subfacet expires, we fold
3434 * its statistics into its facet, and when a facet's last subfacet expires, we
3435 * fold its statistic into its rule. */
3437 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3440 * Idle time histogram.
3442 * Most of the time a switch has a relatively small number of subfacets.
3443 * When this is the case we might as well keep statistics for all of them
3444 * in userspace and to cache them in the kernel datapath for performance as
3447 * As the number of subfacets increases, the memory required to maintain
3448 * statistics about them in userspace and in the kernel becomes
3449 * significant. However, with a large number of subfacets it is likely
3450 * that only a few of them are "heavy hitters" that consume a large amount
3451 * of bandwidth. At this point, only heavy hitters are worth caching in
3452 * the kernel and maintaining in userspaces; other subfacets we can
3455 * The technique used to compute the idle time is to build a histogram with
3456 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3457 * that is installed in the kernel gets dropped in the appropriate bucket.
3458 * After the histogram has been built, we compute the cutoff so that only
3459 * the most-recently-used 1% of subfacets (but at least
3460 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3461 * the most-recently-used bucket of subfacets is kept, so actually an
3462 * arbitrary number of subfacets can be kept in any given expiration run
3463 * (though the next run will delete most of those unless they receive
3466 * This requires a second pass through the subfacets, in addition to the
3467 * pass made by update_stats(), because the former function never looks at
3468 * uninstallable subfacets.
3470 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3471 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3472 int buckets[N_BUCKETS] = { 0 };
3473 int total, subtotal, bucket;
3474 struct subfacet *subfacet;
3478 total = hmap_count(&ofproto->subfacets);
3479 if (total <= ofproto->up.flow_eviction_threshold) {
3480 return N_BUCKETS * BUCKET_WIDTH;
3483 /* Build histogram. */
3485 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3486 long long int idle = now - subfacet->used;
3487 int bucket = (idle <= 0 ? 0
3488 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3489 : (unsigned int) idle / BUCKET_WIDTH);
3493 /* Find the first bucket whose flows should be expired. */
3494 subtotal = bucket = 0;
3496 subtotal += buckets[bucket++];
3497 } while (bucket < N_BUCKETS &&
3498 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3500 if (VLOG_IS_DBG_ENABLED()) {
3504 ds_put_cstr(&s, "keep");
3505 for (i = 0; i < N_BUCKETS; i++) {
3507 ds_put_cstr(&s, ", drop");
3510 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3513 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3517 return bucket * BUCKET_WIDTH;
3520 enum { EXPIRE_MAX_BATCH = 50 };
3523 expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n)
3525 struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH];
3526 struct dpif_op ops[EXPIRE_MAX_BATCH];
3527 struct dpif_op *opsp[EXPIRE_MAX_BATCH];
3528 struct ofpbuf keys[EXPIRE_MAX_BATCH];
3529 struct dpif_flow_stats stats[EXPIRE_MAX_BATCH];
3532 for (i = 0; i < n; i++) {
3533 ops[i].type = DPIF_OP_FLOW_DEL;
3534 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
3535 ops[i].u.flow_del.key = keys[i].data;
3536 ops[i].u.flow_del.key_len = keys[i].size;
3537 ops[i].u.flow_del.stats = &stats[i];
3541 dpif_operate(ofproto->dpif, opsp, n);
3542 for (i = 0; i < n; i++) {
3543 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
3544 subfacets[i]->path = SF_NOT_INSTALLED;
3545 subfacet_destroy(subfacets[i]);
3550 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3552 /* Cutoff time for most flows. */
3553 long long int normal_cutoff = time_msec() - dp_max_idle;
3555 /* We really want to keep flows for special protocols around, so use a more
3556 * conservative cutoff. */
3557 long long int special_cutoff = time_msec() - 10000;
3559 struct subfacet *subfacet, *next_subfacet;
3560 struct subfacet *batch[EXPIRE_MAX_BATCH];
3564 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3565 &ofproto->subfacets) {
3566 long long int cutoff;
3568 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3571 if (subfacet->used < cutoff) {
3572 if (subfacet->path != SF_NOT_INSTALLED) {
3573 batch[n_batch++] = subfacet;
3574 if (n_batch >= EXPIRE_MAX_BATCH) {
3575 expire_batch(ofproto, batch, n_batch);
3579 subfacet_destroy(subfacet);
3585 expire_batch(ofproto, batch, n_batch);
3589 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3590 * then delete it entirely. */
3592 rule_expire(struct rule_dpif *rule)
3594 struct facet *facet, *next_facet;
3598 if (rule->up.pending) {
3599 /* We'll have to expire it later. */
3603 /* Has 'rule' expired? */
3605 if (rule->up.hard_timeout
3606 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3607 reason = OFPRR_HARD_TIMEOUT;
3608 } else if (rule->up.idle_timeout
3609 && now > rule->up.used + rule->up.idle_timeout * 1000) {
3610 reason = OFPRR_IDLE_TIMEOUT;
3615 COVERAGE_INC(ofproto_dpif_expired);
3617 /* Update stats. (This is a no-op if the rule expired due to an idle
3618 * timeout, because that only happens when the rule has no facets left.) */
3619 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3620 facet_remove(facet);
3623 /* Get rid of the rule. */
3624 ofproto_rule_expire(&rule->up, reason);
3629 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
3631 * The caller must already have determined that no facet with an identical
3632 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
3633 * the ofproto's classifier table.
3635 * 'hash' must be the return value of flow_hash(flow, 0).
3637 * The facet will initially have no subfacets. The caller should create (at
3638 * least) one subfacet with subfacet_create(). */
3639 static struct facet *
3640 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
3642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3643 struct facet *facet;
3645 facet = xzalloc(sizeof *facet);
3646 facet->used = time_msec();
3647 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
3648 list_push_back(&rule->facets, &facet->list_node);
3650 facet->flow = *flow;
3651 list_init(&facet->subfacets);
3652 netflow_flow_init(&facet->nf_flow);
3653 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3659 facet_free(struct facet *facet)
3664 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3665 * 'packet', which arrived on 'in_port'.
3667 * Takes ownership of 'packet'. */
3669 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3670 const struct nlattr *odp_actions, size_t actions_len,
3671 struct ofpbuf *packet)
3673 struct odputil_keybuf keybuf;
3677 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3678 odp_flow_key_from_flow(&key, flow);
3680 error = dpif_execute(ofproto->dpif, key.data, key.size,
3681 odp_actions, actions_len, packet);
3683 ofpbuf_delete(packet);
3687 /* Remove 'facet' from 'ofproto' and free up the associated memory:
3689 * - If 'facet' was installed in the datapath, uninstalls it and updates its
3690 * rule's statistics, via subfacet_uninstall().
3692 * - Removes 'facet' from its rule and from ofproto->facets.
3695 facet_remove(struct facet *facet)
3697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3698 struct subfacet *subfacet, *next_subfacet;
3700 assert(!list_is_empty(&facet->subfacets));
3702 /* First uninstall all of the subfacets to get final statistics. */
3703 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3704 subfacet_uninstall(subfacet);
3707 /* Flush the final stats to the rule.
3709 * This might require us to have at least one subfacet around so that we
3710 * can use its actions for accounting in facet_account(), which is why we
3711 * have uninstalled but not yet destroyed the subfacets. */
3712 facet_flush_stats(facet);
3714 /* Now we're really all done so destroy everything. */
3715 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3716 &facet->subfacets) {
3717 subfacet_destroy__(subfacet);
3719 hmap_remove(&ofproto->facets, &facet->hmap_node);
3720 list_remove(&facet->list_node);
3724 /* Feed information from 'facet' back into the learning table to keep it in
3725 * sync with what is actually flowing through the datapath. */
3727 facet_learn(struct facet *facet)
3729 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3730 struct action_xlate_ctx ctx;
3732 if (!facet->has_learn
3733 && !facet->has_normal
3734 && (!facet->has_fin_timeout
3735 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3739 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3740 facet->flow.vlan_tci,
3741 facet->rule, facet->tcp_flags, NULL);
3742 ctx.may_learn = true;
3743 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3744 facet->rule->up.ofpacts_len);
3748 facet_account(struct facet *facet)
3750 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3751 struct subfacet *subfacet;
3752 const struct nlattr *a;
3757 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
3760 n_bytes = facet->byte_count - facet->accounted_bytes;
3762 /* This loop feeds byte counters to bond_account() for rebalancing to use
3763 * as a basis. We also need to track the actual VLAN on which the packet
3764 * is going to be sent to ensure that it matches the one passed to
3765 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
3768 * We use the actions from an arbitrary subfacet because they should all
3769 * be equally valid for our purpose. */
3770 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
3771 struct subfacet, list_node);
3772 vlan_tci = facet->flow.vlan_tci;
3773 NL_ATTR_FOR_EACH_UNSAFE (a, left,
3774 subfacet->actions, subfacet->actions_len) {
3775 const struct ovs_action_push_vlan *vlan;
3776 struct ofport_dpif *port;
3778 switch (nl_attr_type(a)) {
3779 case OVS_ACTION_ATTR_OUTPUT:
3780 port = get_odp_port(ofproto, nl_attr_get_u32(a));
3781 if (port && port->bundle && port->bundle->bond) {
3782 bond_account(port->bundle->bond, &facet->flow,
3783 vlan_tci_to_vid(vlan_tci), n_bytes);
3787 case OVS_ACTION_ATTR_POP_VLAN:
3788 vlan_tci = htons(0);
3791 case OVS_ACTION_ATTR_PUSH_VLAN:
3792 vlan = nl_attr_get(a);
3793 vlan_tci = vlan->vlan_tci;
3799 /* Returns true if the only action for 'facet' is to send to the controller.
3800 * (We don't report NetFlow expiration messages for such facets because they
3801 * are just part of the control logic for the network, not real traffic). */
3803 facet_is_controller_flow(struct facet *facet)
3806 const struct rule *rule = &facet->rule->up;
3807 const struct ofpact *ofpacts = rule->ofpacts;
3808 size_t ofpacts_len = rule->ofpacts_len;
3810 if (ofpacts_len > 0 &&
3811 ofpacts->type == OFPACT_CONTROLLER &&
3812 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
3819 /* Folds all of 'facet''s statistics into its rule. Also updates the
3820 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3821 * 'facet''s statistics in the datapath should have been zeroed and folded into
3822 * its packet and byte counts before this function is called. */
3824 facet_flush_stats(struct facet *facet)
3826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3827 struct subfacet *subfacet;
3829 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3830 assert(!subfacet->dp_byte_count);
3831 assert(!subfacet->dp_packet_count);
3834 facet_push_stats(facet);
3835 if (facet->accounted_bytes < facet->byte_count) {
3836 facet_account(facet);
3837 facet->accounted_bytes = facet->byte_count;
3840 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3841 struct ofexpired expired;
3842 expired.flow = facet->flow;
3843 expired.packet_count = facet->packet_count;
3844 expired.byte_count = facet->byte_count;
3845 expired.used = facet->used;
3846 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3849 facet->rule->packet_count += facet->packet_count;
3850 facet->rule->byte_count += facet->byte_count;
3852 /* Reset counters to prevent double counting if 'facet' ever gets
3854 facet_reset_counters(facet);
3856 netflow_flow_clear(&facet->nf_flow);
3857 facet->tcp_flags = 0;
3860 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3861 * Returns it if found, otherwise a null pointer.
3863 * 'hash' must be the return value of flow_hash(flow, 0).
3865 * The returned facet might need revalidation; use facet_lookup_valid()
3866 * instead if that is important. */
3867 static struct facet *
3868 facet_find(struct ofproto_dpif *ofproto,
3869 const struct flow *flow, uint32_t hash)
3871 struct facet *facet;
3873 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
3874 if (flow_equal(flow, &facet->flow)) {
3882 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3883 * Returns it if found, otherwise a null pointer.
3885 * 'hash' must be the return value of flow_hash(flow, 0).
3887 * The returned facet is guaranteed to be valid. */
3888 static struct facet *
3889 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
3892 struct facet *facet;
3894 facet = facet_find(ofproto, flow, hash);
3896 && (ofproto->need_revalidate
3897 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
3898 facet_revalidate(facet);
3905 subfacet_path_to_string(enum subfacet_path path)
3908 case SF_NOT_INSTALLED:
3909 return "not installed";
3911 return "in fast path";
3913 return "in slow path";
3919 /* Returns the path in which a subfacet should be installed if its 'slow'
3920 * member has the specified value. */
3921 static enum subfacet_path
3922 subfacet_want_path(enum slow_path_reason slow)
3924 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
3927 /* Returns true if 'subfacet' needs to have its datapath flow updated,
3928 * supposing that its actions have been recalculated as 'want_actions' and that
3929 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
3931 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
3932 const struct ofpbuf *want_actions)
3934 enum subfacet_path want_path = subfacet_want_path(slow);
3935 return (want_path != subfacet->path
3936 || (want_path == SF_FAST_PATH
3937 && (subfacet->actions_len != want_actions->size
3938 || memcmp(subfacet->actions, want_actions->data,
3939 subfacet->actions_len))));
3943 facet_check_consistency(struct facet *facet)
3945 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3949 uint64_t odp_actions_stub[1024 / 8];
3950 struct ofpbuf odp_actions;
3952 struct rule_dpif *rule;
3953 struct subfacet *subfacet;
3954 bool may_log = false;
3957 /* Check the rule for consistency. */
3958 rule = rule_dpif_lookup(ofproto, &facet->flow);
3959 ok = rule == facet->rule;
3961 may_log = !VLOG_DROP_WARN(&rl);
3966 flow_format(&s, &facet->flow);
3967 ds_put_format(&s, ": facet associated with wrong rule (was "
3968 "table=%"PRIu8",", facet->rule->up.table_id);
3969 cls_rule_format(&facet->rule->up.cr, &s);
3970 ds_put_format(&s, ") (should have been table=%"PRIu8",",
3972 cls_rule_format(&rule->up.cr, &s);
3973 ds_put_char(&s, ')');
3975 VLOG_WARN("%s", ds_cstr(&s));
3980 /* Check the datapath actions for consistency. */
3981 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
3982 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3983 enum subfacet_path want_path;
3984 struct odputil_keybuf keybuf;
3985 struct action_xlate_ctx ctx;
3989 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3990 subfacet->initial_tci, rule, 0, NULL);
3991 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3994 if (subfacet->path == SF_NOT_INSTALLED) {
3995 /* This only happens if the datapath reported an error when we
3996 * tried to install the flow. Don't flag another error here. */
4000 want_path = subfacet_want_path(subfacet->slow);
4001 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4002 /* The actions for slow-path flows may legitimately vary from one
4003 * packet to the next. We're done. */
4007 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4011 /* Inconsistency! */
4013 may_log = !VLOG_DROP_WARN(&rl);
4017 /* Rate-limited, skip reporting. */
4022 subfacet_get_key(subfacet, &keybuf, &key);
4023 odp_flow_key_format(key.data, key.size, &s);
4025 ds_put_cstr(&s, ": inconsistency in subfacet");
4026 if (want_path != subfacet->path) {
4027 enum odp_key_fitness fitness = subfacet->key_fitness;
4029 ds_put_format(&s, " (%s, fitness=%s)",
4030 subfacet_path_to_string(subfacet->path),
4031 odp_key_fitness_to_string(fitness));
4032 ds_put_format(&s, " (should have been %s)",
4033 subfacet_path_to_string(want_path));
4034 } else if (want_path == SF_FAST_PATH) {
4035 ds_put_cstr(&s, " (actions were: ");
4036 format_odp_actions(&s, subfacet->actions,
4037 subfacet->actions_len);
4038 ds_put_cstr(&s, ") (correct actions: ");
4039 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4040 ds_put_char(&s, ')');
4042 ds_put_cstr(&s, " (actions: ");
4043 format_odp_actions(&s, subfacet->actions,
4044 subfacet->actions_len);
4045 ds_put_char(&s, ')');
4047 VLOG_WARN("%s", ds_cstr(&s));
4050 ofpbuf_uninit(&odp_actions);
4055 /* Re-searches the classifier for 'facet':
4057 * - If the rule found is different from 'facet''s current rule, moves
4058 * 'facet' to the new rule and recompiles its actions.
4060 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4061 * where it is and recompiles its actions anyway. */
4063 facet_revalidate(struct facet *facet)
4065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4067 struct nlattr *odp_actions;
4070 struct actions *new_actions;
4072 struct action_xlate_ctx ctx;
4073 uint64_t odp_actions_stub[1024 / 8];
4074 struct ofpbuf odp_actions;
4076 struct rule_dpif *new_rule;
4077 struct subfacet *subfacet;
4080 COVERAGE_INC(facet_revalidate);
4082 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4084 /* Calculate new datapath actions.
4086 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4087 * emit a NetFlow expiration and, if so, we need to have the old state
4088 * around to properly compose it. */
4090 /* If the datapath actions changed or the installability changed,
4091 * then we need to talk to the datapath. */
4094 memset(&ctx, 0, sizeof ctx);
4095 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4096 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4097 enum slow_path_reason slow;
4099 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4100 subfacet->initial_tci, new_rule, 0, NULL);
4101 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4104 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4105 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4106 struct dpif_flow_stats stats;
4108 subfacet_install(subfacet,
4109 odp_actions.data, odp_actions.size, &stats, slow);
4110 subfacet_update_stats(subfacet, &stats);
4113 new_actions = xcalloc(list_size(&facet->subfacets),
4114 sizeof *new_actions);
4116 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4118 new_actions[i].actions_len = odp_actions.size;
4123 ofpbuf_uninit(&odp_actions);
4126 facet_flush_stats(facet);
4129 /* Update 'facet' now that we've taken care of all the old state. */
4130 facet->tags = ctx.tags;
4131 facet->nf_flow.output_iface = ctx.nf_output_iface;
4132 facet->has_learn = ctx.has_learn;
4133 facet->has_normal = ctx.has_normal;
4134 facet->has_fin_timeout = ctx.has_fin_timeout;
4135 facet->mirrors = ctx.mirrors;
4138 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4139 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4141 if (new_actions && new_actions[i].odp_actions) {
4142 free(subfacet->actions);
4143 subfacet->actions = new_actions[i].odp_actions;
4144 subfacet->actions_len = new_actions[i].actions_len;
4150 if (facet->rule != new_rule) {
4151 COVERAGE_INC(facet_changed_rule);
4152 list_remove(&facet->list_node);
4153 list_push_back(&new_rule->facets, &facet->list_node);
4154 facet->rule = new_rule;
4155 facet->used = new_rule->up.created;
4156 facet->prev_used = facet->used;
4160 /* Updates 'facet''s used time. Caller is responsible for calling
4161 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4163 facet_update_time(struct facet *facet, long long int used)
4165 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4166 if (used > facet->used) {
4168 ofproto_rule_update_used(&facet->rule->up, used);
4169 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4174 facet_reset_counters(struct facet *facet)
4176 facet->packet_count = 0;
4177 facet->byte_count = 0;
4178 facet->prev_packet_count = 0;
4179 facet->prev_byte_count = 0;
4180 facet->accounted_bytes = 0;
4184 facet_push_stats(struct facet *facet)
4186 struct dpif_flow_stats stats;
4188 assert(facet->packet_count >= facet->prev_packet_count);
4189 assert(facet->byte_count >= facet->prev_byte_count);
4190 assert(facet->used >= facet->prev_used);
4192 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4193 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4194 stats.used = facet->used;
4195 stats.tcp_flags = 0;
4197 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4198 facet->prev_packet_count = facet->packet_count;
4199 facet->prev_byte_count = facet->byte_count;
4200 facet->prev_used = facet->used;
4202 flow_push_stats(facet->rule, &facet->flow, &stats);
4204 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4205 facet->mirrors, stats.n_packets, stats.n_bytes);
4210 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4212 rule->packet_count += stats->n_packets;
4213 rule->byte_count += stats->n_bytes;
4214 ofproto_rule_update_used(&rule->up, stats->used);
4217 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4218 * 'rule''s actions and mirrors. */
4220 flow_push_stats(struct rule_dpif *rule,
4221 const struct flow *flow, const struct dpif_flow_stats *stats)
4223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4224 struct action_xlate_ctx ctx;
4226 ofproto_rule_update_used(&rule->up, stats->used);
4228 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4230 ctx.resubmit_stats = stats;
4231 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4232 rule->up.ofpacts_len);
4237 static struct subfacet *
4238 subfacet_find__(struct ofproto_dpif *ofproto,
4239 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4240 const struct flow *flow)
4242 struct subfacet *subfacet;
4244 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4245 &ofproto->subfacets) {
4247 ? (subfacet->key_len == key_len
4248 && !memcmp(key, subfacet->key, key_len))
4249 : flow_equal(flow, &subfacet->facet->flow)) {
4257 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4258 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4259 * there is one, otherwise creates and returns a new subfacet.
4261 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4262 * which case the caller must populate the actions with
4263 * subfacet_make_actions(). */
4264 static struct subfacet *
4265 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4266 const struct nlattr *key, size_t key_len,
4267 ovs_be16 initial_tci, long long int now)
4269 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4270 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4271 struct subfacet *subfacet;
4273 if (list_is_empty(&facet->subfacets)) {
4274 subfacet = &facet->one_subfacet;
4276 subfacet = subfacet_find__(ofproto, key, key_len, key_hash,
4279 if (subfacet->facet == facet) {
4283 /* This shouldn't happen. */
4284 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4285 subfacet_destroy(subfacet);
4288 subfacet = xmalloc(sizeof *subfacet);
4291 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4292 list_push_back(&facet->subfacets, &subfacet->list_node);
4293 subfacet->facet = facet;
4294 subfacet->key_fitness = key_fitness;
4295 if (key_fitness != ODP_FIT_PERFECT) {
4296 subfacet->key = xmemdup(key, key_len);
4297 subfacet->key_len = key_len;
4299 subfacet->key = NULL;
4300 subfacet->key_len = 0;
4302 subfacet->used = now;
4303 subfacet->dp_packet_count = 0;
4304 subfacet->dp_byte_count = 0;
4305 subfacet->actions_len = 0;
4306 subfacet->actions = NULL;
4307 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4310 subfacet->path = SF_NOT_INSTALLED;
4311 subfacet->initial_tci = initial_tci;
4316 /* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and
4317 * 'flow'. Returns the subfacet if one exists, otherwise NULL. */
4318 static struct subfacet *
4319 subfacet_find(struct ofproto_dpif *ofproto,
4320 const struct nlattr *key, size_t key_len)
4322 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4323 enum odp_key_fitness fitness;
4326 fitness = odp_flow_key_to_flow(key, key_len, &flow);
4327 if (fitness == ODP_FIT_ERROR) {
4331 return subfacet_find__(ofproto, key, key_len, key_hash, &flow);
4334 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4335 * its facet within 'ofproto', and frees it. */
4337 subfacet_destroy__(struct subfacet *subfacet)
4339 struct facet *facet = subfacet->facet;
4340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4342 subfacet_uninstall(subfacet);
4343 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4344 list_remove(&subfacet->list_node);
4345 free(subfacet->key);
4346 free(subfacet->actions);
4347 if (subfacet != &facet->one_subfacet) {
4352 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4353 * last remaining subfacet in its facet destroys the facet too. */
4355 subfacet_destroy(struct subfacet *subfacet)
4357 struct facet *facet = subfacet->facet;
4359 if (list_is_singleton(&facet->subfacets)) {
4360 /* facet_remove() needs at least one subfacet (it will remove it). */
4361 facet_remove(facet);
4363 subfacet_destroy__(subfacet);
4367 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4368 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4369 * for use as temporary storage. */
4371 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4374 if (!subfacet->key) {
4375 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4376 odp_flow_key_from_flow(key, &subfacet->facet->flow);
4378 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4382 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4383 * Translates the actions into 'odp_actions', which the caller must have
4384 * initialized and is responsible for uninitializing. */
4386 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4387 struct ofpbuf *odp_actions)
4389 struct facet *facet = subfacet->facet;
4390 struct rule_dpif *rule = facet->rule;
4391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4393 struct action_xlate_ctx ctx;
4395 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4397 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4398 facet->tags = ctx.tags;
4399 facet->has_learn = ctx.has_learn;
4400 facet->has_normal = ctx.has_normal;
4401 facet->has_fin_timeout = ctx.has_fin_timeout;
4402 facet->nf_flow.output_iface = ctx.nf_output_iface;
4403 facet->mirrors = ctx.mirrors;
4405 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4406 if (subfacet->actions_len != odp_actions->size
4407 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4408 free(subfacet->actions);
4409 subfacet->actions_len = odp_actions->size;
4410 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4414 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4415 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4416 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4417 * since 'subfacet' was last updated.
4419 * Returns 0 if successful, otherwise a positive errno value. */
4421 subfacet_install(struct subfacet *subfacet,
4422 const struct nlattr *actions, size_t actions_len,
4423 struct dpif_flow_stats *stats,
4424 enum slow_path_reason slow)
4426 struct facet *facet = subfacet->facet;
4427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4428 enum subfacet_path path = subfacet_want_path(slow);
4429 uint64_t slow_path_stub[128 / 8];
4430 struct odputil_keybuf keybuf;
4431 enum dpif_flow_put_flags flags;
4435 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4437 flags |= DPIF_FP_ZERO_STATS;
4440 if (path == SF_SLOW_PATH) {
4441 compose_slow_path(ofproto, &facet->flow, slow,
4442 slow_path_stub, sizeof slow_path_stub,
4443 &actions, &actions_len);
4446 subfacet_get_key(subfacet, &keybuf, &key);
4447 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
4448 actions, actions_len, stats);
4451 subfacet_reset_dp_stats(subfacet, stats);
4455 subfacet->path = path;
4461 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4463 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4464 stats, subfacet->slow);
4467 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4469 subfacet_uninstall(struct subfacet *subfacet)
4471 if (subfacet->path != SF_NOT_INSTALLED) {
4472 struct rule_dpif *rule = subfacet->facet->rule;
4473 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4474 struct odputil_keybuf keybuf;
4475 struct dpif_flow_stats stats;
4479 subfacet_get_key(subfacet, &keybuf, &key);
4480 error = dpif_flow_del(ofproto->dpif, key.data, key.size, &stats);
4481 subfacet_reset_dp_stats(subfacet, &stats);
4483 subfacet_update_stats(subfacet, &stats);
4485 subfacet->path = SF_NOT_INSTALLED;
4487 assert(subfacet->dp_packet_count == 0);
4488 assert(subfacet->dp_byte_count == 0);
4492 /* Resets 'subfacet''s datapath statistics counters. This should be called
4493 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4494 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4495 * was reset in the datapath. 'stats' will be modified to include only
4496 * statistics new since 'subfacet' was last updated. */
4498 subfacet_reset_dp_stats(struct subfacet *subfacet,
4499 struct dpif_flow_stats *stats)
4502 && subfacet->dp_packet_count <= stats->n_packets
4503 && subfacet->dp_byte_count <= stats->n_bytes) {
4504 stats->n_packets -= subfacet->dp_packet_count;
4505 stats->n_bytes -= subfacet->dp_byte_count;
4508 subfacet->dp_packet_count = 0;
4509 subfacet->dp_byte_count = 0;
4512 /* Updates 'subfacet''s used time. The caller is responsible for calling
4513 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4515 subfacet_update_time(struct subfacet *subfacet, long long int used)
4517 if (used > subfacet->used) {
4518 subfacet->used = used;
4519 facet_update_time(subfacet->facet, used);
4523 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4525 * Because of the meaning of a subfacet's counters, it only makes sense to do
4526 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4527 * represents a packet that was sent by hand or if it represents statistics
4528 * that have been cleared out of the datapath. */
4530 subfacet_update_stats(struct subfacet *subfacet,
4531 const struct dpif_flow_stats *stats)
4533 if (stats->n_packets || stats->used > subfacet->used) {
4534 struct facet *facet = subfacet->facet;
4536 subfacet_update_time(subfacet, stats->used);
4537 facet->packet_count += stats->n_packets;
4538 facet->byte_count += stats->n_bytes;
4539 facet->tcp_flags |= stats->tcp_flags;
4540 facet_push_stats(facet);
4541 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4547 static struct rule_dpif *
4548 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4550 struct rule_dpif *rule;
4552 rule = rule_dpif_lookup__(ofproto, flow, 0);
4557 return rule_dpif_miss_rule(ofproto, flow);
4560 static struct rule_dpif *
4561 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4564 struct cls_rule *cls_rule;
4565 struct classifier *cls;
4567 if (table_id >= N_TABLES) {
4571 cls = &ofproto->up.tables[table_id].cls;
4572 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4573 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4574 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4575 * are unavailable. */
4576 struct flow ofpc_normal_flow = *flow;
4577 ofpc_normal_flow.tp_src = htons(0);
4578 ofpc_normal_flow.tp_dst = htons(0);
4579 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4581 cls_rule = classifier_lookup(cls, flow);
4583 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4586 static struct rule_dpif *
4587 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4589 struct ofport_dpif *port;
4591 port = get_ofp_port(ofproto, flow->in_port);
4593 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
4594 return ofproto->miss_rule;
4597 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
4598 return ofproto->no_packet_in_rule;
4600 return ofproto->miss_rule;
4604 complete_operation(struct rule_dpif *rule)
4606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4608 rule_invalidate(rule);
4610 struct dpif_completion *c = xmalloc(sizeof *c);
4611 c->op = rule->up.pending;
4612 list_push_back(&ofproto->completions, &c->list_node);
4614 ofoperation_complete(rule->up.pending, 0);
4618 static struct rule *
4621 struct rule_dpif *rule = xmalloc(sizeof *rule);
4626 rule_dealloc(struct rule *rule_)
4628 struct rule_dpif *rule = rule_dpif_cast(rule_);
4633 rule_construct(struct rule *rule_)
4635 struct rule_dpif *rule = rule_dpif_cast(rule_);
4636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4637 struct rule_dpif *victim;
4641 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4642 &rule->up.cr.flow, ofproto->max_ports);
4647 rule->packet_count = 0;
4648 rule->byte_count = 0;
4650 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
4651 if (victim && !list_is_empty(&victim->facets)) {
4652 struct facet *facet;
4654 rule->facets = victim->facets;
4655 list_moved(&rule->facets);
4656 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4657 /* XXX: We're only clearing our local counters here. It's possible
4658 * that quite a few packets are unaccounted for in the datapath
4659 * statistics. These will be accounted to the new rule instead of
4660 * cleared as required. This could be fixed by clearing out the
4661 * datapath statistics for this facet, but currently it doesn't
4663 facet_reset_counters(facet);
4667 /* Must avoid list_moved() in this case. */
4668 list_init(&rule->facets);
4671 table_id = rule->up.table_id;
4672 rule->tag = (victim ? victim->tag
4674 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
4675 ofproto->tables[table_id].basis));
4677 complete_operation(rule);
4682 rule_destruct(struct rule *rule_)
4684 struct rule_dpif *rule = rule_dpif_cast(rule_);
4685 struct facet *facet, *next_facet;
4687 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4688 facet_revalidate(facet);
4691 complete_operation(rule);
4695 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4697 struct rule_dpif *rule = rule_dpif_cast(rule_);
4698 struct facet *facet;
4700 /* Start from historical data for 'rule' itself that are no longer tracked
4701 * in facets. This counts, for example, facets that have expired. */
4702 *packets = rule->packet_count;
4703 *bytes = rule->byte_count;
4705 /* Add any statistics that are tracked by facets. This includes
4706 * statistical data recently updated by ofproto_update_stats() as well as
4707 * stats for packets that were executed "by hand" via dpif_execute(). */
4708 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4709 *packets += facet->packet_count;
4710 *bytes += facet->byte_count;
4715 rule_execute(struct rule *rule_, const struct flow *flow,
4716 struct ofpbuf *packet)
4718 struct rule_dpif *rule = rule_dpif_cast(rule_);
4719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4721 struct dpif_flow_stats stats;
4723 struct action_xlate_ctx ctx;
4724 uint64_t odp_actions_stub[1024 / 8];
4725 struct ofpbuf odp_actions;
4727 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4728 rule_credit_stats(rule, &stats);
4730 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4731 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
4732 rule, stats.tcp_flags, packet);
4733 ctx.resubmit_stats = &stats;
4734 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
4736 execute_odp_actions(ofproto, flow, odp_actions.data,
4737 odp_actions.size, packet);
4739 ofpbuf_uninit(&odp_actions);
4745 rule_modify_actions(struct rule *rule_)
4747 struct rule_dpif *rule = rule_dpif_cast(rule_);
4748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4751 error = ofpacts_check(rule->up.ofpacts, rule->up.ofpacts_len,
4752 &rule->up.cr.flow, ofproto->max_ports);
4754 ofoperation_complete(rule->up.pending, error);
4758 complete_operation(rule);
4761 /* Sends 'packet' out 'ofport'.
4762 * May modify 'packet'.
4763 * Returns 0 if successful, otherwise a positive errno value. */
4765 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4767 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4768 struct ofpbuf key, odp_actions;
4769 struct odputil_keybuf keybuf;
4774 flow_extract(packet, 0, 0, 0, &flow);
4775 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
4777 if (odp_port != ofport->odp_port) {
4778 eth_pop_vlan(packet);
4779 flow.vlan_tci = htons(0);
4782 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4783 odp_flow_key_from_flow(&key, &flow);
4785 ofpbuf_init(&odp_actions, 32);
4786 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
4788 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
4789 error = dpif_execute(ofproto->dpif,
4791 odp_actions.data, odp_actions.size,
4793 ofpbuf_uninit(&odp_actions);
4796 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
4797 ofproto->up.name, odp_port, strerror(error));
4799 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
4803 /* OpenFlow to datapath action translation. */
4805 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
4806 struct action_xlate_ctx *);
4807 static void xlate_normal(struct action_xlate_ctx *);
4809 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4810 * The action will state 'slow' as the reason that the action is in the slow
4811 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
4812 * dump-flows" output to see why a flow is in the slow path.)
4814 * The 'stub_size' bytes in 'stub' will be used to store the action.
4815 * 'stub_size' must be large enough for the action.
4817 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4820 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4821 enum slow_path_reason slow,
4822 uint64_t *stub, size_t stub_size,
4823 const struct nlattr **actionsp, size_t *actions_lenp)
4825 union user_action_cookie cookie;
4828 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4829 cookie.slow_path.unused = 0;
4830 cookie.slow_path.reason = slow;
4832 ofpbuf_use_stack(&buf, stub, stub_size);
4833 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
4834 uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX);
4835 odp_put_userspace_action(pid, &cookie, &buf);
4837 put_userspace_action(ofproto, &buf, flow, &cookie);
4839 *actionsp = buf.data;
4840 *actions_lenp = buf.size;
4844 put_userspace_action(const struct ofproto_dpif *ofproto,
4845 struct ofpbuf *odp_actions,
4846 const struct flow *flow,
4847 const union user_action_cookie *cookie)
4851 pid = dpif_port_get_pid(ofproto->dpif,
4852 ofp_port_to_odp_port(flow->in_port));
4854 return odp_put_userspace_action(pid, cookie, odp_actions);
4858 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
4859 ovs_be16 vlan_tci, uint32_t odp_port,
4860 unsigned int n_outputs, union user_action_cookie *cookie)
4864 cookie->type = USER_ACTION_COOKIE_SFLOW;
4865 cookie->sflow.vlan_tci = vlan_tci;
4867 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
4868 * port information") for the interpretation of cookie->output. */
4869 switch (n_outputs) {
4871 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
4872 cookie->sflow.output = 0x40000000 | 256;
4876 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
4878 cookie->sflow.output = ifindex;
4883 /* 0x80000000 means "multiple output ports. */
4884 cookie->sflow.output = 0x80000000 | n_outputs;
4889 /* Compose SAMPLE action for sFlow. */
4891 compose_sflow_action(const struct ofproto_dpif *ofproto,
4892 struct ofpbuf *odp_actions,
4893 const struct flow *flow,
4896 uint32_t probability;
4897 union user_action_cookie cookie;
4898 size_t sample_offset, actions_offset;
4901 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
4905 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
4907 /* Number of packets out of UINT_MAX to sample. */
4908 probability = dpif_sflow_get_probability(ofproto->sflow);
4909 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
4911 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
4912 compose_sflow_cookie(ofproto, htons(0), odp_port,
4913 odp_port == OVSP_NONE ? 0 : 1, &cookie);
4914 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
4916 nl_msg_end_nested(odp_actions, actions_offset);
4917 nl_msg_end_nested(odp_actions, sample_offset);
4918 return cookie_offset;
4921 /* SAMPLE action must be first action in any given list of actions.
4922 * At this point we do not have all information required to build it. So try to
4923 * build sample action as complete as possible. */
4925 add_sflow_action(struct action_xlate_ctx *ctx)
4927 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
4929 &ctx->flow, OVSP_NONE);
4930 ctx->sflow_odp_port = 0;
4931 ctx->sflow_n_outputs = 0;
4934 /* Fix SAMPLE action according to data collected while composing ODP actions.
4935 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
4936 * USERSPACE action's user-cookie which is required for sflow. */
4938 fix_sflow_action(struct action_xlate_ctx *ctx)
4940 const struct flow *base = &ctx->base_flow;
4941 union user_action_cookie *cookie;
4943 if (!ctx->user_cookie_offset) {
4947 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
4949 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
4951 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
4952 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
4956 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
4959 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
4960 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
4961 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
4962 uint8_t flow_nw_tos = ctx->flow.nw_tos;
4966 struct priority_to_dscp *pdscp;
4968 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD
4969 || (check_stp && !stp_forward_in_state(ofport->stp_state))) {
4973 pdscp = get_priority(ofport, ctx->flow.skb_priority);
4975 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4976 ctx->flow.nw_tos |= pdscp->dscp;
4979 /* We may not have an ofport record for this port, but it doesn't hurt
4980 * to allow forwarding to it anyhow. Maybe such a port will appear
4981 * later and we're pre-populating the flow table. */
4984 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
4985 ctx->flow.vlan_tci);
4986 if (out_port != odp_port) {
4987 ctx->flow.vlan_tci = htons(0);
4989 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
4990 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
4992 ctx->sflow_odp_port = odp_port;
4993 ctx->sflow_n_outputs++;
4994 ctx->nf_output_iface = ofp_port;
4995 ctx->flow.vlan_tci = flow_vlan_tci;
4996 ctx->flow.nw_tos = flow_nw_tos;
5000 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5002 compose_output_action__(ctx, ofp_port, true);
5006 xlate_table_action(struct action_xlate_ctx *ctx,
5007 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5009 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5010 struct ofproto_dpif *ofproto = ctx->ofproto;
5011 struct rule_dpif *rule;
5012 uint16_t old_in_port;
5013 uint8_t old_table_id;
5015 old_table_id = ctx->table_id;
5016 ctx->table_id = table_id;
5018 /* Look up a flow with 'in_port' as the input port. */
5019 old_in_port = ctx->flow.in_port;
5020 ctx->flow.in_port = in_port;
5021 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5024 if (table_id > 0 && table_id < N_TABLES) {
5025 struct table_dpif *table = &ofproto->tables[table_id];
5026 if (table->other_table) {
5027 ctx->tags |= (rule && rule->tag
5029 : rule_calculate_tag(&ctx->flow,
5030 &table->other_table->wc,
5035 /* Restore the original input port. Otherwise OFPP_NORMAL and
5036 * OFPP_IN_PORT will have surprising behavior. */
5037 ctx->flow.in_port = old_in_port;
5039 if (ctx->resubmit_hook) {
5040 ctx->resubmit_hook(ctx, rule);
5043 if (rule == NULL && may_packet_in) {
5045 * check if table configuration flags
5046 * OFPTC_TABLE_MISS_CONTROLLER, default.
5047 * OFPTC_TABLE_MISS_CONTINUE,
5048 * OFPTC_TABLE_MISS_DROP
5049 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5051 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5055 struct rule_dpif *old_rule = ctx->rule;
5057 if (ctx->resubmit_stats) {
5058 rule_credit_stats(rule, ctx->resubmit_stats);
5063 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5064 ctx->rule = old_rule;
5068 ctx->table_id = old_table_id;
5070 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5072 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5073 MAX_RESUBMIT_RECURSION);
5074 ctx->max_resubmit_trigger = true;
5079 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5080 const struct ofpact_resubmit *resubmit)
5085 in_port = resubmit->in_port;
5086 if (in_port == OFPP_IN_PORT) {
5087 in_port = ctx->flow.in_port;
5090 table_id = resubmit->table_id;
5091 if (table_id == 255) {
5092 table_id = ctx->table_id;
5095 xlate_table_action(ctx, in_port, table_id, false);
5099 flood_packets(struct action_xlate_ctx *ctx, bool all)
5101 struct ofport_dpif *ofport;
5103 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5104 uint16_t ofp_port = ofport->up.ofp_port;
5106 if (ofp_port == ctx->flow.in_port) {
5111 compose_output_action__(ctx, ofp_port, false);
5112 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5113 compose_output_action(ctx, ofp_port);
5117 ctx->nf_output_iface = NF_OUT_FLOOD;
5121 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5122 enum ofp_packet_in_reason reason,
5123 uint16_t controller_id)
5125 struct ofputil_packet_in pin;
5126 struct ofpbuf *packet;
5128 ctx->slow |= SLOW_CONTROLLER;
5133 packet = ofpbuf_clone(ctx->packet);
5135 if (packet->l2 && packet->l3) {
5136 struct eth_header *eh;
5138 eth_pop_vlan(packet);
5141 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5142 * LLC frame. Calculating the Ethernet type of these frames is more
5143 * trouble than seems appropriate for a simple assertion. */
5144 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5145 || eh->eth_type == ctx->flow.dl_type);
5147 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5148 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5150 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5151 eth_push_vlan(packet, ctx->flow.vlan_tci);
5155 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5156 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5157 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5161 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5162 packet_set_tcp_port(packet, ctx->flow.tp_src,
5164 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5165 packet_set_udp_port(packet, ctx->flow.tp_src,
5172 pin.packet = packet->data;
5173 pin.packet_len = packet->size;
5174 pin.reason = reason;
5175 pin.controller_id = controller_id;
5176 pin.table_id = ctx->table_id;
5177 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5180 flow_get_metadata(&ctx->flow, &pin.fmd);
5182 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5183 ofpbuf_delete(packet);
5187 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5189 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5190 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5194 if (ctx->flow.nw_ttl > 1) {
5200 for (i = 0; i < ids->n_controllers; i++) {
5201 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5205 /* Stop processing for current table. */
5211 xlate_output_action(struct action_xlate_ctx *ctx,
5212 uint16_t port, uint16_t max_len, bool may_packet_in)
5214 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5216 ctx->nf_output_iface = NF_OUT_DROP;
5220 compose_output_action(ctx, ctx->flow.in_port);
5223 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5229 flood_packets(ctx, false);
5232 flood_packets(ctx, true);
5234 case OFPP_CONTROLLER:
5235 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5241 if (port != ctx->flow.in_port) {
5242 compose_output_action(ctx, port);
5247 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5248 ctx->nf_output_iface = NF_OUT_FLOOD;
5249 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5250 ctx->nf_output_iface = prev_nf_output_iface;
5251 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5252 ctx->nf_output_iface != NF_OUT_FLOOD) {
5253 ctx->nf_output_iface = NF_OUT_MULTI;
5258 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5259 const struct ofpact_output_reg *or)
5261 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5262 if (port <= UINT16_MAX) {
5263 xlate_output_action(ctx, port, or->max_len, false);
5268 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5269 const struct ofpact_enqueue *enqueue)
5271 uint16_t ofp_port = enqueue->port;
5272 uint32_t queue_id = enqueue->queue;
5273 uint32_t flow_priority, priority;
5276 /* Translate queue to priority. */
5277 error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority);
5279 /* Fall back to ordinary output action. */
5280 xlate_output_action(ctx, enqueue->port, 0, false);
5284 /* Check output port. */
5285 if (ofp_port == OFPP_IN_PORT) {
5286 ofp_port = ctx->flow.in_port;
5287 } else if (ofp_port == ctx->flow.in_port) {
5291 /* Add datapath actions. */
5292 flow_priority = ctx->flow.skb_priority;
5293 ctx->flow.skb_priority = priority;
5294 compose_output_action(ctx, ofp_port);
5295 ctx->flow.skb_priority = flow_priority;
5297 /* Update NetFlow output port. */
5298 if (ctx->nf_output_iface == NF_OUT_DROP) {
5299 ctx->nf_output_iface = ofp_port;
5300 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5301 ctx->nf_output_iface = NF_OUT_MULTI;
5306 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5308 uint32_t skb_priority;
5310 if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) {
5311 ctx->flow.skb_priority = skb_priority;
5313 /* Couldn't translate queue to a priority. Nothing to do. A warning
5314 * has already been logged. */
5318 struct xlate_reg_state {
5324 xlate_autopath(struct action_xlate_ctx *ctx,
5325 const struct ofpact_autopath *ap)
5327 uint16_t ofp_port = ap->port;
5328 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5330 if (!port || !port->bundle) {
5331 ofp_port = OFPP_NONE;
5332 } else if (port->bundle->bond) {
5333 /* Autopath does not support VLAN hashing. */
5334 struct ofport_dpif *slave = bond_choose_output_slave(
5335 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5337 ofp_port = slave->up.ofp_port;
5340 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5344 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5346 struct ofproto_dpif *ofproto = ofproto_;
5347 struct ofport_dpif *port;
5357 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5360 port = get_ofp_port(ofproto, ofp_port);
5361 return port ? port->may_enable : false;
5366 xlate_bundle_action(struct action_xlate_ctx *ctx,
5367 const struct ofpact_bundle *bundle)
5371 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5372 if (bundle->dst.field) {
5373 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5375 xlate_output_action(ctx, port, 0, false);
5380 xlate_learn_action(struct action_xlate_ctx *ctx,
5381 const struct ofpact_learn *learn)
5383 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5384 struct ofputil_flow_mod fm;
5385 uint64_t ofpacts_stub[1024 / 8];
5386 struct ofpbuf ofpacts;
5389 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5390 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5392 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5393 if (error && !VLOG_DROP_WARN(&rl)) {
5394 VLOG_WARN("learning action failed to modify flow table (%s)",
5395 ofperr_get_name(error));
5398 ofpbuf_uninit(&ofpacts);
5401 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5402 * means "infinite". */
5404 reduce_timeout(uint16_t max, uint16_t *timeout)
5406 if (max && (!*timeout || *timeout > max)) {
5412 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5413 const struct ofpact_fin_timeout *oft)
5415 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5416 struct rule_dpif *rule = ctx->rule;
5418 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5419 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5424 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5426 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5427 ? OFPUTIL_PC_NO_RECV_STP
5428 : OFPUTIL_PC_NO_RECV)) {
5432 /* Only drop packets here if both forwarding and learning are
5433 * disabled. If just learning is enabled, we need to have
5434 * OFPP_NORMAL and the learning action have a look at the packet
5435 * before we can drop it. */
5436 if (!stp_forward_in_state(port->stp_state)
5437 && !stp_learn_in_state(port->stp_state)) {
5445 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5446 struct action_xlate_ctx *ctx)
5448 const struct ofport_dpif *port;
5449 bool was_evictable = true;
5450 const struct ofpact *a;
5452 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5453 if (port && !may_receive(port, ctx)) {
5454 /* Drop this flow. */
5459 /* Don't let the rule we're working on get evicted underneath us. */
5460 was_evictable = ctx->rule->up.evictable;
5461 ctx->rule->up.evictable = false;
5463 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5464 struct ofpact_controller *controller;
5472 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5473 ofpact_get_OUTPUT(a)->max_len, true);
5476 case OFPACT_CONTROLLER:
5477 controller = ofpact_get_CONTROLLER(a);
5478 execute_controller_action(ctx, controller->max_len,
5480 controller->controller_id);
5483 case OFPACT_ENQUEUE:
5484 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5487 case OFPACT_SET_VLAN_VID:
5488 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5489 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5493 case OFPACT_SET_VLAN_PCP:
5494 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5495 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5500 case OFPACT_STRIP_VLAN:
5501 ctx->flow.vlan_tci = htons(0);
5504 case OFPACT_SET_ETH_SRC:
5505 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5509 case OFPACT_SET_ETH_DST:
5510 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5514 case OFPACT_SET_IPV4_SRC:
5515 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5518 case OFPACT_SET_IPV4_DST:
5519 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5522 case OFPACT_SET_IPV4_DSCP:
5523 /* OpenFlow 1.0 only supports IPv4. */
5524 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5525 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5526 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5530 case OFPACT_SET_L4_SRC_PORT:
5531 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5534 case OFPACT_SET_L4_DST_PORT:
5535 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5538 case OFPACT_RESUBMIT:
5539 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5542 case OFPACT_SET_TUNNEL:
5543 ctx->flow.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5546 case OFPACT_SET_QUEUE:
5547 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5550 case OFPACT_POP_QUEUE:
5551 ctx->flow.skb_priority = ctx->orig_skb_priority;
5554 case OFPACT_REG_MOVE:
5555 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5558 case OFPACT_REG_LOAD:
5559 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5562 case OFPACT_DEC_TTL:
5563 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5569 /* Nothing to do. */
5572 case OFPACT_MULTIPATH:
5573 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5576 case OFPACT_AUTOPATH:
5577 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5581 ctx->ofproto->has_bundle_action = true;
5582 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5585 case OFPACT_OUTPUT_REG:
5586 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
5590 ctx->has_learn = true;
5591 if (ctx->may_learn) {
5592 xlate_learn_action(ctx, ofpact_get_LEARN(a));
5600 case OFPACT_FIN_TIMEOUT:
5601 ctx->has_fin_timeout = true;
5602 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
5608 /* We've let OFPP_NORMAL and the learning action look at the packet,
5609 * so drop it now if forwarding is disabled. */
5610 if (port && !stp_forward_in_state(port->stp_state)) {
5611 ofpbuf_clear(ctx->odp_actions);
5612 add_sflow_action(ctx);
5615 ctx->rule->up.evictable = was_evictable;
5620 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
5621 struct ofproto_dpif *ofproto, const struct flow *flow,
5622 ovs_be16 initial_tci, struct rule_dpif *rule,
5623 uint8_t tcp_flags, const struct ofpbuf *packet)
5625 ctx->ofproto = ofproto;
5627 ctx->base_flow = ctx->flow;
5628 ctx->base_flow.tun_id = 0;
5629 ctx->base_flow.vlan_tci = initial_tci;
5631 ctx->packet = packet;
5632 ctx->may_learn = packet != NULL;
5633 ctx->tcp_flags = tcp_flags;
5634 ctx->resubmit_hook = NULL;
5635 ctx->report_hook = NULL;
5636 ctx->resubmit_stats = NULL;
5639 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
5640 * into datapath actions in 'odp_actions', using 'ctx'. */
5642 xlate_actions(struct action_xlate_ctx *ctx,
5643 const struct ofpact *ofpacts, size_t ofpacts_len,
5644 struct ofpbuf *odp_actions)
5646 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
5647 * that in the future we always keep a copy of the original flow for
5648 * tracing purposes. */
5649 static bool hit_resubmit_limit;
5651 enum slow_path_reason special;
5653 COVERAGE_INC(ofproto_dpif_xlate);
5655 ofpbuf_clear(odp_actions);
5656 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
5658 ctx->odp_actions = odp_actions;
5661 ctx->has_learn = false;
5662 ctx->has_normal = false;
5663 ctx->has_fin_timeout = false;
5664 ctx->nf_output_iface = NF_OUT_DROP;
5667 ctx->max_resubmit_trigger = false;
5668 ctx->orig_skb_priority = ctx->flow.skb_priority;
5672 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
5673 /* Do this conditionally because the copy is expensive enough that it
5674 * shows up in profiles.
5676 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
5677 * believe that I wasn't using it without initializing it if I kept it
5678 * in a local variable. */
5679 ctx->orig_flow = ctx->flow;
5682 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
5683 switch (ctx->ofproto->up.frag_handling) {
5684 case OFPC_FRAG_NORMAL:
5685 /* We must pretend that transport ports are unavailable. */
5686 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
5687 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
5690 case OFPC_FRAG_DROP:
5693 case OFPC_FRAG_REASM:
5696 case OFPC_FRAG_NX_MATCH:
5697 /* Nothing to do. */
5700 case OFPC_INVALID_TTL_TO_CONTROLLER:
5705 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
5707 ctx->slow |= special;
5709 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5710 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
5712 add_sflow_action(ctx);
5713 do_xlate_actions(ofpacts, ofpacts_len, ctx);
5715 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
5716 if (!hit_resubmit_limit) {
5717 /* We didn't record the original flow. Make sure we do from
5719 hit_resubmit_limit = true;
5720 } else if (!VLOG_DROP_ERR(&trace_rl)) {
5721 struct ds ds = DS_EMPTY_INITIALIZER;
5723 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
5725 VLOG_ERR("Trace triggered by excessive resubmit "
5726 "recursion:\n%s", ds_cstr(&ds));
5731 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
5732 ctx->odp_actions->data,
5733 ctx->odp_actions->size)) {
5734 ctx->slow |= SLOW_IN_BAND;
5736 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
5738 compose_output_action(ctx, OFPP_LOCAL);
5741 if (ctx->ofproto->has_mirrors) {
5742 add_mirror_actions(ctx, &ctx->orig_flow);
5744 fix_sflow_action(ctx);
5748 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5749 * into datapath actions, using 'ctx', and discards the datapath actions. */
5751 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
5752 const struct ofpact *ofpacts,
5755 uint64_t odp_actions_stub[1024 / 8];
5756 struct ofpbuf odp_actions;
5758 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5759 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
5760 ofpbuf_uninit(&odp_actions);
5764 xlate_report(struct action_xlate_ctx *ctx, const char *s)
5766 if (ctx->report_hook) {
5767 ctx->report_hook(ctx, s);
5771 /* OFPP_NORMAL implementation. */
5773 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
5775 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
5776 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
5777 * the bundle on which the packet was received, returns the VLAN to which the
5780 * Both 'vid' and the return value are in the range 0...4095. */
5782 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
5784 switch (in_bundle->vlan_mode) {
5785 case PORT_VLAN_ACCESS:
5786 return in_bundle->vlan;
5789 case PORT_VLAN_TRUNK:
5792 case PORT_VLAN_NATIVE_UNTAGGED:
5793 case PORT_VLAN_NATIVE_TAGGED:
5794 return vid ? vid : in_bundle->vlan;
5801 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
5802 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
5805 * 'vid' should be the VID obtained from the 802.1Q header that was received as
5806 * part of a packet (specify 0 if there was no 802.1Q header), in the range
5809 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
5811 /* Allow any VID on the OFPP_NONE port. */
5812 if (in_bundle == &ofpp_none_bundle) {
5816 switch (in_bundle->vlan_mode) {
5817 case PORT_VLAN_ACCESS:
5820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5821 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
5822 "packet received on port %s configured as VLAN "
5823 "%"PRIu16" access port",
5824 in_bundle->ofproto->up.name, vid,
5825 in_bundle->name, in_bundle->vlan);
5831 case PORT_VLAN_NATIVE_UNTAGGED:
5832 case PORT_VLAN_NATIVE_TAGGED:
5834 /* Port must always carry its native VLAN. */
5838 case PORT_VLAN_TRUNK:
5839 if (!ofbundle_includes_vlan(in_bundle, vid)) {
5841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5842 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
5843 "received on port %s not configured for trunking "
5845 in_bundle->ofproto->up.name, vid,
5846 in_bundle->name, vid);
5858 /* Given 'vlan', the VLAN that a packet belongs to, and
5859 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
5860 * that should be included in the 802.1Q header. (If the return value is 0,
5861 * then the 802.1Q header should only be included in the packet if there is a
5864 * Both 'vlan' and the return value are in the range 0...4095. */
5866 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
5868 switch (out_bundle->vlan_mode) {
5869 case PORT_VLAN_ACCESS:
5872 case PORT_VLAN_TRUNK:
5873 case PORT_VLAN_NATIVE_TAGGED:
5876 case PORT_VLAN_NATIVE_UNTAGGED:
5877 return vlan == out_bundle->vlan ? 0 : vlan;
5885 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
5888 struct ofport_dpif *port;
5890 ovs_be16 tci, old_tci;
5892 vid = output_vlan_to_vid(out_bundle, vlan);
5893 if (!out_bundle->bond) {
5894 port = ofbundle_get_a_port(out_bundle);
5896 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
5899 /* No slaves enabled, so drop packet. */
5904 old_tci = ctx->flow.vlan_tci;
5906 if (tci || out_bundle->use_priority_tags) {
5907 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
5909 tci |= htons(VLAN_CFI);
5912 ctx->flow.vlan_tci = tci;
5914 compose_output_action(ctx, port->up.ofp_port);
5915 ctx->flow.vlan_tci = old_tci;
5919 mirror_mask_ffs(mirror_mask_t mask)
5921 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
5926 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
5928 return (bundle->vlan_mode != PORT_VLAN_ACCESS
5929 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
5933 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
5935 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
5938 /* Returns an arbitrary interface within 'bundle'. */
5939 static struct ofport_dpif *
5940 ofbundle_get_a_port(const struct ofbundle *bundle)
5942 return CONTAINER_OF(list_front(&bundle->ports),
5943 struct ofport_dpif, bundle_node);
5947 vlan_is_mirrored(const struct ofmirror *m, int vlan)
5949 return !m->vlans || bitmap_is_set(m->vlans, vlan);
5953 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
5955 struct ofproto_dpif *ofproto = ctx->ofproto;
5956 mirror_mask_t mirrors;
5957 struct ofbundle *in_bundle;
5960 const struct nlattr *a;
5963 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
5964 ctx->packet != NULL, NULL);
5968 mirrors = in_bundle->src_mirrors;
5970 /* Drop frames on bundles reserved for mirroring. */
5971 if (in_bundle->mirror_out) {
5972 if (ctx->packet != NULL) {
5973 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5974 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5975 "%s, which is reserved exclusively for mirroring",
5976 ctx->ofproto->up.name, in_bundle->name);
5982 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
5983 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
5986 vlan = input_vid_to_vlan(in_bundle, vid);
5988 /* Look at the output ports to check for destination selections. */
5990 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
5991 ctx->odp_actions->size) {
5992 enum ovs_action_attr type = nl_attr_type(a);
5993 struct ofport_dpif *ofport;
5995 if (type != OVS_ACTION_ATTR_OUTPUT) {
5999 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6000 if (ofport && ofport->bundle) {
6001 mirrors |= ofport->bundle->dst_mirrors;
6009 /* Restore the original packet before adding the mirror actions. */
6010 ctx->flow = *orig_flow;
6015 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6017 if (!vlan_is_mirrored(m, vlan)) {
6018 mirrors &= mirrors - 1;
6022 mirrors &= ~m->dup_mirrors;
6023 ctx->mirrors |= m->dup_mirrors;
6025 output_normal(ctx, m->out, vlan);
6026 } else if (vlan != m->out_vlan
6027 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6028 struct ofbundle *bundle;
6030 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6031 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6032 && !bundle->mirror_out) {
6033 output_normal(ctx, bundle, m->out_vlan);
6041 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6042 uint64_t packets, uint64_t bytes)
6048 for (; mirrors; mirrors &= mirrors - 1) {
6051 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6054 /* In normal circumstances 'm' will not be NULL. However,
6055 * if mirrors are reconfigured, we can temporarily get out
6056 * of sync in facet_revalidate(). We could "correct" the
6057 * mirror list before reaching here, but doing that would
6058 * not properly account the traffic stats we've currently
6059 * accumulated for previous mirror configuration. */
6063 m->packet_count += packets;
6064 m->byte_count += bytes;
6068 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6069 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6070 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6072 is_gratuitous_arp(const struct flow *flow)
6074 return (flow->dl_type == htons(ETH_TYPE_ARP)
6075 && eth_addr_is_broadcast(flow->dl_dst)
6076 && (flow->nw_proto == ARP_OP_REPLY
6077 || (flow->nw_proto == ARP_OP_REQUEST
6078 && flow->nw_src == flow->nw_dst)));
6082 update_learning_table(struct ofproto_dpif *ofproto,
6083 const struct flow *flow, int vlan,
6084 struct ofbundle *in_bundle)
6086 struct mac_entry *mac;
6088 /* Don't learn the OFPP_NONE port. */
6089 if (in_bundle == &ofpp_none_bundle) {
6093 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6097 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6098 if (is_gratuitous_arp(flow)) {
6099 /* We don't want to learn from gratuitous ARP packets that are
6100 * reflected back over bond slaves so we lock the learning table. */
6101 if (!in_bundle->bond) {
6102 mac_entry_set_grat_arp_lock(mac);
6103 } else if (mac_entry_is_grat_arp_locked(mac)) {
6108 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6109 /* The log messages here could actually be useful in debugging,
6110 * so keep the rate limit relatively high. */
6111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6112 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6113 "on port %s in VLAN %d",
6114 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6115 in_bundle->name, vlan);
6117 mac->port.p = in_bundle;
6118 tag_set_add(&ofproto->revalidate_set,
6119 mac_learning_changed(ofproto->ml, mac));
6123 static struct ofbundle *
6124 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6125 bool warn, struct ofport_dpif **in_ofportp)
6127 struct ofport_dpif *ofport;
6129 /* Find the port and bundle for the received packet. */
6130 ofport = get_ofp_port(ofproto, in_port);
6132 *in_ofportp = ofport;
6134 if (ofport && ofport->bundle) {
6135 return ofport->bundle;
6138 /* Special-case OFPP_NONE, which a controller may use as the ingress
6139 * port for traffic that it is sourcing. */
6140 if (in_port == OFPP_NONE) {
6141 return &ofpp_none_bundle;
6144 /* Odd. A few possible reasons here:
6146 * - We deleted a port but there are still a few packets queued up
6149 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6150 * we don't know about.
6152 * - The ofproto client didn't configure the port as part of a bundle.
6153 * This is particularly likely to happen if a packet was received on the
6154 * port after it was created, but before the client had a chance to
6155 * configure its bundle.
6158 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6160 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6161 "port %"PRIu16, ofproto->up.name, in_port);
6166 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6167 * dropped. Returns true if they may be forwarded, false if they should be
6170 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6171 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6173 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6174 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6175 * checked by input_vid_is_valid().
6177 * May also add tags to '*tags', although the current implementation only does
6178 * so in one special case.
6181 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6184 struct ofproto_dpif *ofproto = ctx->ofproto;
6185 struct flow *flow = &ctx->flow;
6186 struct ofbundle *in_bundle = in_port->bundle;
6188 /* Drop frames for reserved multicast addresses
6189 * only if forward_bpdu option is absent. */
6190 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6191 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6195 if (in_bundle->bond) {
6196 struct mac_entry *mac;
6198 switch (bond_check_admissibility(in_bundle->bond, in_port,
6199 flow->dl_dst, &ctx->tags)) {
6204 xlate_report(ctx, "bonding refused admissibility, dropping");
6207 case BV_DROP_IF_MOVED:
6208 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6209 if (mac && mac->port.p != in_bundle &&
6210 (!is_gratuitous_arp(flow)
6211 || mac_entry_is_grat_arp_locked(mac))) {
6212 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6224 xlate_normal(struct action_xlate_ctx *ctx)
6226 struct ofport_dpif *in_port;
6227 struct ofbundle *in_bundle;
6228 struct mac_entry *mac;
6232 ctx->has_normal = true;
6234 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6235 ctx->packet != NULL, &in_port);
6237 xlate_report(ctx, "no input bundle, dropping");
6241 /* Drop malformed frames. */
6242 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6243 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6244 if (ctx->packet != NULL) {
6245 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6246 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6247 "VLAN tag received on port %s",
6248 ctx->ofproto->up.name, in_bundle->name);
6250 xlate_report(ctx, "partial VLAN tag, dropping");
6254 /* Drop frames on bundles reserved for mirroring. */
6255 if (in_bundle->mirror_out) {
6256 if (ctx->packet != NULL) {
6257 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6258 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6259 "%s, which is reserved exclusively for mirroring",
6260 ctx->ofproto->up.name, in_bundle->name);
6262 xlate_report(ctx, "input port is mirror output port, dropping");
6267 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6268 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6269 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6272 vlan = input_vid_to_vlan(in_bundle, vid);
6274 /* Check other admissibility requirements. */
6275 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6279 /* Learn source MAC. */
6280 if (ctx->may_learn) {
6281 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6284 /* Determine output bundle. */
6285 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6288 if (mac->port.p != in_bundle) {
6289 xlate_report(ctx, "forwarding to learned port");
6290 output_normal(ctx, mac->port.p, vlan);
6292 xlate_report(ctx, "learned port is input port, dropping");
6295 struct ofbundle *bundle;
6297 xlate_report(ctx, "no learned MAC for destination, flooding");
6298 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6299 if (bundle != in_bundle
6300 && ofbundle_includes_vlan(bundle, vlan)
6301 && bundle->floodable
6302 && !bundle->mirror_out) {
6303 output_normal(ctx, bundle, vlan);
6306 ctx->nf_output_iface = NF_OUT_FLOOD;
6310 /* Optimized flow revalidation.
6312 * It's a difficult problem, in general, to tell which facets need to have
6313 * their actions recalculated whenever the OpenFlow flow table changes. We
6314 * don't try to solve that general problem: for most kinds of OpenFlow flow
6315 * table changes, we recalculate the actions for every facet. This is
6316 * relatively expensive, but it's good enough if the OpenFlow flow table
6317 * doesn't change very often.
6319 * However, we can expect one particular kind of OpenFlow flow table change to
6320 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6321 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6322 * table, we add a special case that applies to flow tables in which every rule
6323 * has the same form (that is, the same wildcards), except that the table is
6324 * also allowed to have a single "catch-all" flow that matches all packets. We
6325 * optimize this case by tagging all of the facets that resubmit into the table
6326 * and invalidating the same tag whenever a flow changes in that table. The
6327 * end result is that we revalidate just the facets that need it (and sometimes
6328 * a few more, but not all of the facets or even all of the facets that
6329 * resubmit to the table modified by MAC learning). */
6331 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
6332 * into an OpenFlow table with the given 'basis'. */
6334 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
6337 if (flow_wildcards_is_catchall(wc)) {
6340 struct flow tag_flow = *flow;
6341 flow_zero_wildcards(&tag_flow, wc);
6342 return tag_create_deterministic(flow_hash(&tag_flow, secret));
6346 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6347 * taggability of that table.
6349 * This function must be called after *each* change to a flow table. If you
6350 * skip calling it on some changes then the pointer comparisons at the end can
6351 * be invalid if you get unlucky. For example, if a flow removal causes a
6352 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6353 * different wildcards to be created with the same address, then this function
6354 * will incorrectly skip revalidation. */
6356 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6358 struct table_dpif *table = &ofproto->tables[table_id];
6359 const struct oftable *oftable = &ofproto->up.tables[table_id];
6360 struct cls_table *catchall, *other;
6361 struct cls_table *t;
6363 catchall = other = NULL;
6365 switch (hmap_count(&oftable->cls.tables)) {
6367 /* We could tag this OpenFlow table but it would make the logic a
6368 * little harder and it's a corner case that doesn't seem worth it
6374 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6375 if (cls_table_is_catchall(t)) {
6377 } else if (!other) {
6380 /* Indicate that we can't tag this by setting both tables to
6381 * NULL. (We know that 'catchall' is already NULL.) */
6388 /* Can't tag this table. */
6392 if (table->catchall_table != catchall || table->other_table != other) {
6393 table->catchall_table = catchall;
6394 table->other_table = other;
6395 ofproto->need_revalidate = REV_FLOW_TABLE;
6399 /* Given 'rule' that has changed in some way (either it is a rule being
6400 * inserted, a rule being deleted, or a rule whose actions are being
6401 * modified), marks facets for revalidation to ensure that packets will be
6402 * forwarded correctly according to the new state of the flow table.
6404 * This function must be called after *each* change to a flow table. See
6405 * the comment on table_update_taggable() for more information. */
6407 rule_invalidate(const struct rule_dpif *rule)
6409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6411 table_update_taggable(ofproto, rule->up.table_id);
6413 if (!ofproto->need_revalidate) {
6414 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6416 if (table->other_table && rule->tag) {
6417 tag_set_add(&ofproto->revalidate_set, rule->tag);
6419 ofproto->need_revalidate = REV_FLOW_TABLE;
6425 set_frag_handling(struct ofproto *ofproto_,
6426 enum ofp_config_flags frag_handling)
6428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6430 if (frag_handling != OFPC_FRAG_REASM) {
6431 ofproto->need_revalidate = REV_RECONFIGURE;
6439 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6440 const struct flow *flow,
6441 const struct ofpact *ofpacts, size_t ofpacts_len)
6443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6446 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
6447 return OFPERR_NXBRC_BAD_IN_PORT;
6450 error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports);
6452 struct odputil_keybuf keybuf;
6453 struct dpif_flow_stats stats;
6457 struct action_xlate_ctx ctx;
6458 uint64_t odp_actions_stub[1024 / 8];
6459 struct ofpbuf odp_actions;
6461 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6462 odp_flow_key_from_flow(&key, flow);
6464 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6466 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6467 packet_get_tcp_flags(packet, flow), packet);
6468 ctx.resubmit_stats = &stats;
6470 ofpbuf_use_stub(&odp_actions,
6471 odp_actions_stub, sizeof odp_actions_stub);
6472 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6473 dpif_execute(ofproto->dpif, key.data, key.size,
6474 odp_actions.data, odp_actions.size, packet);
6475 ofpbuf_uninit(&odp_actions);
6483 set_netflow(struct ofproto *ofproto_,
6484 const struct netflow_options *netflow_options)
6486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6488 if (netflow_options) {
6489 if (!ofproto->netflow) {
6490 ofproto->netflow = netflow_create();
6492 return netflow_set_options(ofproto->netflow, netflow_options);
6494 netflow_destroy(ofproto->netflow);
6495 ofproto->netflow = NULL;
6501 get_netflow_ids(const struct ofproto *ofproto_,
6502 uint8_t *engine_type, uint8_t *engine_id)
6504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6506 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
6510 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6512 if (!facet_is_controller_flow(facet) &&
6513 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6514 struct subfacet *subfacet;
6515 struct ofexpired expired;
6517 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6518 if (subfacet->path == SF_FAST_PATH) {
6519 struct dpif_flow_stats stats;
6521 subfacet_reinstall(subfacet, &stats);
6522 subfacet_update_stats(subfacet, &stats);
6526 expired.flow = facet->flow;
6527 expired.packet_count = facet->packet_count;
6528 expired.byte_count = facet->byte_count;
6529 expired.used = facet->used;
6530 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6535 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6537 struct facet *facet;
6539 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6540 send_active_timeout(ofproto, facet);
6544 static struct ofproto_dpif *
6545 ofproto_dpif_lookup(const char *name)
6547 struct ofproto_dpif *ofproto;
6549 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6550 hash_string(name, 0), &all_ofproto_dpifs) {
6551 if (!strcmp(ofproto->up.name, name)) {
6559 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6560 const char *argv[], void *aux OVS_UNUSED)
6562 struct ofproto_dpif *ofproto;
6565 ofproto = ofproto_dpif_lookup(argv[1]);
6567 unixctl_command_reply_error(conn, "no such bridge");
6570 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6572 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6573 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
6577 unixctl_command_reply(conn, "table successfully flushed");
6581 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
6582 const char *argv[], void *aux OVS_UNUSED)
6584 struct ds ds = DS_EMPTY_INITIALIZER;
6585 const struct ofproto_dpif *ofproto;
6586 const struct mac_entry *e;
6588 ofproto = ofproto_dpif_lookup(argv[1]);
6590 unixctl_command_reply_error(conn, "no such bridge");
6594 ds_put_cstr(&ds, " port VLAN MAC Age\n");
6595 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
6596 struct ofbundle *bundle = e->port.p;
6597 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
6598 ofbundle_get_a_port(bundle)->odp_port,
6599 e->vlan, ETH_ADDR_ARGS(e->mac),
6600 mac_entry_age(ofproto->ml, e));
6602 unixctl_command_reply(conn, ds_cstr(&ds));
6607 struct action_xlate_ctx ctx;
6613 trace_format_rule(struct ds *result, uint8_t table_id, int level,
6614 const struct rule_dpif *rule)
6616 ds_put_char_multiple(result, '\t', level);
6618 ds_put_cstr(result, "No match\n");
6622 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
6623 table_id, ntohll(rule->up.flow_cookie));
6624 cls_rule_format(&rule->up.cr, result);
6625 ds_put_char(result, '\n');
6627 ds_put_char_multiple(result, '\t', level);
6628 ds_put_cstr(result, "OpenFlow ");
6629 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
6630 ds_put_char(result, '\n');
6634 trace_format_flow(struct ds *result, int level, const char *title,
6635 struct trace_ctx *trace)
6637 ds_put_char_multiple(result, '\t', level);
6638 ds_put_format(result, "%s: ", title);
6639 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
6640 ds_put_cstr(result, "unchanged");
6642 flow_format(result, &trace->ctx.flow);
6643 trace->flow = trace->ctx.flow;
6645 ds_put_char(result, '\n');
6649 trace_format_regs(struct ds *result, int level, const char *title,
6650 struct trace_ctx *trace)
6654 ds_put_char_multiple(result, '\t', level);
6655 ds_put_format(result, "%s:", title);
6656 for (i = 0; i < FLOW_N_REGS; i++) {
6657 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
6659 ds_put_char(result, '\n');
6663 trace_format_odp(struct ds *result, int level, const char *title,
6664 struct trace_ctx *trace)
6666 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
6668 ds_put_char_multiple(result, '\t', level);
6669 ds_put_format(result, "%s: ", title);
6670 format_odp_actions(result, odp_actions->data, odp_actions->size);
6671 ds_put_char(result, '\n');
6675 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6677 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6678 struct ds *result = trace->result;
6680 ds_put_char(result, '\n');
6681 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
6682 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
6683 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
6684 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
6688 trace_report(struct action_xlate_ctx *ctx, const char *s)
6690 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
6691 struct ds *result = trace->result;
6693 ds_put_char_multiple(result, '\t', ctx->recurse);
6694 ds_put_cstr(result, s);
6695 ds_put_char(result, '\n');
6699 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
6700 void *aux OVS_UNUSED)
6702 const char *dpname = argv[1];
6703 struct ofproto_dpif *ofproto;
6704 struct ofpbuf odp_key;
6705 struct ofpbuf *packet;
6706 ovs_be16 initial_tci;
6712 ofpbuf_init(&odp_key, 0);
6715 ofproto = ofproto_dpif_lookup(dpname);
6717 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
6721 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
6722 /* ofproto/trace dpname flow [-generate] */
6723 const char *flow_s = argv[2];
6724 const char *generate_s = argv[3];
6726 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
6727 * flow. We guess which type it is based on whether 'flow_s' contains
6728 * an '(', since a datapath flow always contains '(') but an
6729 * OpenFlow-like flow should not (in fact it's allowed but I believe
6730 * that's not documented anywhere).
6732 * An alternative would be to try to parse 'flow_s' both ways, but then
6733 * it would be tricky giving a sensible error message. After all, do
6734 * you just say "syntax error" or do you present both error messages?
6735 * Both choices seem lousy. */
6736 if (strchr(flow_s, '(')) {
6739 /* Convert string to datapath key. */
6740 ofpbuf_init(&odp_key, 0);
6741 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
6743 unixctl_command_reply_error(conn, "Bad flow syntax");
6747 /* Convert odp_key to flow. */
6748 error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data,
6749 odp_key.size, &flow,
6750 &initial_tci, NULL);
6751 if (error == ODP_FIT_ERROR) {
6752 unixctl_command_reply_error(conn, "Invalid flow");
6758 error_s = parse_ofp_exact_flow(&flow, argv[2]);
6760 unixctl_command_reply_error(conn, error_s);
6765 initial_tci = flow.vlan_tci;
6766 vsp_adjust_flow(ofproto, &flow);
6769 /* Generate a packet, if requested. */
6771 packet = ofpbuf_new(0);
6772 flow_compose(packet, &flow);
6774 } else if (argc == 6) {
6775 /* ofproto/trace dpname priority tun_id in_port packet */
6776 const char *priority_s = argv[2];
6777 const char *tun_id_s = argv[3];
6778 const char *in_port_s = argv[4];
6779 const char *packet_s = argv[5];
6780 uint16_t in_port = ofp_port_to_odp_port(atoi(in_port_s));
6781 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
6782 uint32_t priority = atoi(priority_s);
6785 msg = eth_from_hex(packet_s, &packet);
6787 unixctl_command_reply_error(conn, msg);
6791 ds_put_cstr(&result, "Packet: ");
6792 s = ofp_packet_to_string(packet->data, packet->size);
6793 ds_put_cstr(&result, s);
6796 flow_extract(packet, priority, tun_id, in_port, &flow);
6797 initial_tci = flow.vlan_tci;
6799 unixctl_command_reply_error(conn, "Bad command syntax");
6803 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
6804 unixctl_command_reply(conn, ds_cstr(&result));
6807 ds_destroy(&result);
6808 ofpbuf_delete(packet);
6809 ofpbuf_uninit(&odp_key);
6813 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
6814 const struct ofpbuf *packet, ovs_be16 initial_tci,
6817 struct rule_dpif *rule;
6819 ds_put_cstr(ds, "Flow: ");
6820 flow_format(ds, flow);
6821 ds_put_char(ds, '\n');
6823 rule = rule_dpif_lookup(ofproto, flow);
6825 trace_format_rule(ds, 0, 0, rule);
6826 if (rule == ofproto->miss_rule) {
6827 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
6828 } else if (rule == ofproto->no_packet_in_rule) {
6829 ds_put_cstr(ds, "\nNo match, packets dropped because "
6830 "OFPPC_NO_PACKET_IN is set on in_port.\n");
6834 uint64_t odp_actions_stub[1024 / 8];
6835 struct ofpbuf odp_actions;
6837 struct trace_ctx trace;
6840 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
6843 ofpbuf_use_stub(&odp_actions,
6844 odp_actions_stub, sizeof odp_actions_stub);
6845 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
6846 rule, tcp_flags, packet);
6847 trace.ctx.resubmit_hook = trace_resubmit;
6848 trace.ctx.report_hook = trace_report;
6849 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
6852 ds_put_char(ds, '\n');
6853 trace_format_flow(ds, 0, "Final flow", &trace);
6854 ds_put_cstr(ds, "Datapath actions: ");
6855 format_odp_actions(ds, odp_actions.data, odp_actions.size);
6856 ofpbuf_uninit(&odp_actions);
6858 if (trace.ctx.slow) {
6859 enum slow_path_reason slow;
6861 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
6862 "slow path because it:");
6863 for (slow = trace.ctx.slow; slow; ) {
6864 enum slow_path_reason bit = rightmost_1bit(slow);
6868 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
6871 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
6874 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
6877 ds_put_cstr(ds, "\n\t- Needs in-band special case "
6880 ds_put_cstr(ds, "\n\t (The datapath actions are "
6881 "incomplete--for complete actions, "
6882 "please supply a packet.)");
6885 case SLOW_CONTROLLER:
6886 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
6887 "to the OpenFlow controller.");
6890 ds_put_cstr(ds, "\n\t- Needs more specific matching "
6891 "than the datapath supports.");
6898 if (slow & ~SLOW_MATCH) {
6899 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
6900 "the special slow-path processing.");
6907 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6908 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6911 unixctl_command_reply(conn, NULL);
6915 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
6916 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6919 unixctl_command_reply(conn, NULL);
6922 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
6923 * 'reply' describing the results. */
6925 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
6927 struct facet *facet;
6931 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6932 if (!facet_check_consistency(facet)) {
6937 ofproto->need_revalidate = REV_INCONSISTENCY;
6941 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
6942 ofproto->up.name, errors);
6944 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
6949 ofproto_dpif_self_check(struct unixctl_conn *conn,
6950 int argc, const char *argv[], void *aux OVS_UNUSED)
6952 struct ds reply = DS_EMPTY_INITIALIZER;
6953 struct ofproto_dpif *ofproto;
6956 ofproto = ofproto_dpif_lookup(argv[1]);
6958 unixctl_command_reply_error(conn, "Unknown ofproto (use "
6959 "ofproto/list for help)");
6962 ofproto_dpif_self_check__(ofproto, &reply);
6964 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
6965 ofproto_dpif_self_check__(ofproto, &reply);
6969 unixctl_command_reply(conn, ds_cstr(&reply));
6974 ofproto_dpif_unixctl_init(void)
6976 static bool registered;
6982 unixctl_command_register(
6984 "bridge {tun_id in_port packet | odp_flow [-generate]}",
6985 2, 5, ofproto_unixctl_trace, NULL);
6986 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6987 ofproto_unixctl_fdb_flush, NULL);
6988 unixctl_command_register("fdb/show", "bridge", 1, 1,
6989 ofproto_unixctl_fdb_show, NULL);
6990 unixctl_command_register("ofproto/clog", "", 0, 0,
6991 ofproto_dpif_clog, NULL);
6992 unixctl_command_register("ofproto/unclog", "", 0, 0,
6993 ofproto_dpif_unclog, NULL);
6994 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6995 ofproto_dpif_self_check, NULL);
6998 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7000 * This is deprecated. It is only for compatibility with broken device drivers
7001 * in old versions of Linux that do not properly support VLANs when VLAN
7002 * devices are not used. When broken device drivers are no longer in
7003 * widespread use, we will delete these interfaces. */
7006 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7009 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7011 if (realdev_ofp_port == ofport->realdev_ofp_port
7012 && vid == ofport->vlandev_vid) {
7016 ofproto->need_revalidate = REV_RECONFIGURE;
7018 if (ofport->realdev_ofp_port) {
7021 if (realdev_ofp_port && ofport->bundle) {
7022 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7023 * themselves be part of a bundle. */
7024 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7027 ofport->realdev_ofp_port = realdev_ofp_port;
7028 ofport->vlandev_vid = vid;
7030 if (realdev_ofp_port) {
7031 vsp_add(ofport, realdev_ofp_port, vid);
7038 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7040 return hash_2words(realdev_ofp_port, vid);
7043 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7044 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7045 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7046 * it would return the port number of eth0.9.
7048 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7049 * function just returns its 'realdev_odp_port' argument. */
7051 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7052 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7054 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7055 uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port);
7056 int vid = vlan_tci_to_vid(vlan_tci);
7057 const struct vlan_splinter *vsp;
7059 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7060 hash_realdev_vid(realdev_ofp_port, vid),
7061 &ofproto->realdev_vid_map) {
7062 if (vsp->realdev_ofp_port == realdev_ofp_port
7063 && vsp->vid == vid) {
7064 return ofp_port_to_odp_port(vsp->vlandev_ofp_port);
7068 return realdev_odp_port;
7071 static struct vlan_splinter *
7072 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7074 struct vlan_splinter *vsp;
7076 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7077 &ofproto->vlandev_map) {
7078 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7086 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7087 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7088 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7089 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7090 * eth0 and store 9 in '*vid'.
7092 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7093 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7096 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7097 uint16_t vlandev_ofp_port, int *vid)
7099 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7100 const struct vlan_splinter *vsp;
7102 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7107 return vsp->realdev_ofp_port;
7113 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7114 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7115 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7116 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7117 * always the case unless VLAN splinters are enabled), returns false without
7118 * making any changes. */
7120 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7125 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7130 /* Cause the flow to be processed as if it came in on the real device with
7131 * the VLAN device's VLAN ID. */
7132 flow->in_port = realdev;
7133 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7138 vsp_remove(struct ofport_dpif *port)
7140 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7141 struct vlan_splinter *vsp;
7143 vsp = vlandev_find(ofproto, port->up.ofp_port);
7145 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7146 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7149 port->realdev_ofp_port = 0;
7151 VLOG_ERR("missing vlan device record");
7156 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7160 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7161 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7162 == realdev_ofp_port)) {
7163 struct vlan_splinter *vsp;
7165 vsp = xmalloc(sizeof *vsp);
7166 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7167 hash_int(port->up.ofp_port, 0));
7168 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7169 hash_realdev_vid(realdev_ofp_port, vid));
7170 vsp->realdev_ofp_port = realdev_ofp_port;
7171 vsp->vlandev_ofp_port = port->up.ofp_port;
7174 port->realdev_ofp_port = realdev_ofp_port;
7176 VLOG_ERR("duplicate vlan device record");
7180 const struct ofproto_class ofproto_dpif_class = {
7210 port_is_lacp_current,
7211 NULL, /* rule_choose_table */
7218 rule_modify_actions,
7227 get_cfm_remote_mpids,
7232 get_stp_port_status,
7239 is_mirror_output_bundle,
7240 forward_bpdu_changed,