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"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
61 COVERAGE_DEFINE(ofproto_dpif_expired);
62 COVERAGE_DEFINE(ofproto_dpif_xlate);
63 COVERAGE_DEFINE(facet_changed_rule);
64 COVERAGE_DEFINE(facet_revalidate);
65 COVERAGE_DEFINE(facet_unexpected);
66 COVERAGE_DEFINE(facet_suppress);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES = 255 };
74 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
85 * - Do include packets and bytes from facets that have been deleted or
86 * whose own statistics have been folded into the rule.
88 * - Do include packets and bytes sent "by hand" that were accounted to
89 * the rule without any facet being involved (this is a rare corner
90 * case in rule_execute()).
92 * - Do not include packet or bytes that can be obtained from any facet's
93 * packet_count or byte_count member or that can be obtained from the
94 * datapath by, e.g., dpif_flow_get() for any subfacet.
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
99 tag_type tag; /* Caches rule_calculate_tag() result. */
101 struct list facets; /* List of "struct facet"s. */
104 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
106 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
109 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
110 const struct flow *);
111 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
114 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
115 const struct flow *flow);
117 static void rule_credit_stats(struct rule_dpif *,
118 const struct dpif_flow_stats *);
119 static void flow_push_stats(struct rule_dpif *, const struct flow *,
120 const struct dpif_flow_stats *);
121 static tag_type rule_calculate_tag(const struct flow *,
122 const struct minimask *, uint32_t basis);
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 uint32_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 #define SUBFACET_DESTROY_MAX_BATCH 50
373 static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness,
374 const struct nlattr *key,
375 size_t key_len, ovs_be16 initial_tci,
377 static struct subfacet *subfacet_find(struct ofproto_dpif *,
378 const struct nlattr *key, size_t key_len,
380 const struct flow *flow);
381 static void subfacet_destroy(struct subfacet *);
382 static void subfacet_destroy__(struct subfacet *);
383 static void subfacet_destroy_batch(struct ofproto_dpif *,
384 struct subfacet **, int n);
385 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
387 static void subfacet_reset_dp_stats(struct subfacet *,
388 struct dpif_flow_stats *);
389 static void subfacet_update_time(struct subfacet *, long long int used);
390 static void subfacet_update_stats(struct subfacet *,
391 const struct dpif_flow_stats *);
392 static void subfacet_make_actions(struct subfacet *,
393 const struct ofpbuf *packet,
394 struct ofpbuf *odp_actions);
395 static int subfacet_install(struct subfacet *,
396 const struct nlattr *actions, size_t actions_len,
397 struct dpif_flow_stats *, enum slow_path_reason);
398 static void subfacet_uninstall(struct subfacet *);
400 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
402 /* An exact-match instantiation of an OpenFlow flow.
404 * A facet associates a "struct flow", which represents the Open vSwitch
405 * userspace idea of an exact-match flow, with one or more subfacets. Each
406 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
407 * the facet. When the kernel module (or other dpif implementation) and Open
408 * vSwitch userspace agree on the definition of a flow key, there is exactly
409 * one subfacet per facet. If the dpif implementation supports more-specific
410 * flow matching than userspace, however, a facet can have more than one
411 * subfacet, each of which corresponds to some distinction in flow that
412 * userspace simply doesn't understand.
414 * Flow expiration works in terms of subfacets, so a facet must have at least
415 * one subfacet or it will never expire, leaking memory. */
418 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
419 struct list list_node; /* In owning rule's 'facets' list. */
420 struct rule_dpif *rule; /* Owning rule. */
423 struct list subfacets;
424 long long int used; /* Time last used; time created if not used. */
431 * - Do include packets and bytes sent "by hand", e.g. with
434 * - Do include packets and bytes that were obtained from the datapath
435 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
436 * DPIF_FP_ZERO_STATS).
438 * - Do not include packets or bytes that can be obtained from the
439 * datapath for any existing subfacet.
441 uint64_t packet_count; /* Number of packets received. */
442 uint64_t byte_count; /* Number of bytes received. */
444 /* Resubmit statistics. */
445 uint64_t prev_packet_count; /* Number of packets from last stats push. */
446 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
447 long long int prev_used; /* Used time from last stats push. */
450 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
451 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
452 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
454 /* Properties of datapath actions.
456 * Every subfacet has its own actions because actions can differ slightly
457 * between splintered and non-splintered subfacets due to the VLAN tag
458 * being initially different (present vs. absent). All of them have these
459 * properties in common so we just store one copy of them here. */
460 bool has_learn; /* Actions include NXAST_LEARN? */
461 bool has_normal; /* Actions output to OFPP_NORMAL? */
462 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
463 tag_type tags; /* Tags that would require revalidation. */
464 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
466 /* Storage for a single subfacet, to reduce malloc() time and space
467 * overhead. (A facet always has at least one subfacet and in the common
468 * case has exactly one subfacet.) */
469 struct subfacet one_subfacet;
472 static struct facet *facet_create(struct rule_dpif *,
473 const struct flow *, uint32_t hash);
474 static void facet_remove(struct facet *);
475 static void facet_free(struct facet *);
477 static struct facet *facet_find(struct ofproto_dpif *,
478 const struct flow *, uint32_t hash);
479 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
480 const struct flow *, uint32_t hash);
481 static void facet_revalidate(struct facet *);
482 static bool facet_check_consistency(struct facet *);
484 static void facet_flush_stats(struct facet *);
486 static void facet_update_time(struct facet *, long long int used);
487 static void facet_reset_counters(struct facet *);
488 static void facet_push_stats(struct facet *);
489 static void facet_learn(struct facet *);
490 static void facet_account(struct facet *);
492 static bool facet_is_controller_flow(struct facet *);
495 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
499 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
500 struct list bundle_node; /* In struct ofbundle's "ports" list. */
501 struct cfm *cfm; /* Connectivity Fault Management, if any. */
502 tag_type tag; /* Tag associated with this port. */
503 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
504 bool may_enable; /* May be enabled in bonds. */
505 long long int carrier_seq; /* Carrier status changes. */
508 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
509 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
510 long long int stp_state_entered;
512 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
514 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
516 * This is deprecated. It is only for compatibility with broken device
517 * drivers in old versions of Linux that do not properly support VLANs when
518 * VLAN devices are not used. When broken device drivers are no longer in
519 * widespread use, we will delete these interfaces. */
520 uint16_t realdev_ofp_port;
524 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
525 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
526 * traffic egressing the 'ofport' with that priority should be marked with. */
527 struct priority_to_dscp {
528 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
529 uint32_t priority; /* Priority of this queue (see struct flow). */
531 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
534 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
536 * This is deprecated. It is only for compatibility with broken device drivers
537 * in old versions of Linux that do not properly support VLANs when VLAN
538 * devices are not used. When broken device drivers are no longer in
539 * widespread use, we will delete these interfaces. */
540 struct vlan_splinter {
541 struct hmap_node realdev_vid_node;
542 struct hmap_node vlandev_node;
543 uint16_t realdev_ofp_port;
544 uint16_t vlandev_ofp_port;
548 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
549 uint32_t realdev, ovs_be16 vlan_tci);
550 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
551 static void vsp_remove(struct ofport_dpif *);
552 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
554 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
556 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
559 static struct ofport_dpif *
560 ofport_dpif_cast(const struct ofport *ofport)
562 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
563 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
566 static void port_run(struct ofport_dpif *);
567 static void port_run_fast(struct ofport_dpif *);
568 static void port_wait(struct ofport_dpif *);
569 static int set_cfm(struct ofport *, const struct cfm_settings *);
570 static void ofport_clear_priorities(struct ofport_dpif *);
572 struct dpif_completion {
573 struct list list_node;
574 struct ofoperation *op;
577 /* Extra information about a classifier table.
578 * Currently used just for optimized flow revalidation. */
580 /* If either of these is nonnull, then this table has a form that allows
581 * flows to be tagged to avoid revalidating most flows for the most common
582 * kinds of flow table changes. */
583 struct cls_table *catchall_table; /* Table that wildcards all fields. */
584 struct cls_table *other_table; /* Table with any other wildcard set. */
585 uint32_t basis; /* Keeps each table's tags separate. */
588 /* Reasons that we might need to revalidate every facet, and corresponding
591 * A value of 0 means that there is no need to revalidate.
593 * It would be nice to have some cleaner way to integrate with coverage
594 * counters, but with only a few reasons I guess this is good enough for
596 enum revalidate_reason {
597 REV_RECONFIGURE = 1, /* Switch configuration changed. */
598 REV_STP, /* Spanning tree protocol port status change. */
599 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
600 REV_FLOW_TABLE, /* Flow table changed. */
601 REV_INCONSISTENCY /* Facet self-check failed. */
603 COVERAGE_DEFINE(rev_reconfigure);
604 COVERAGE_DEFINE(rev_stp);
605 COVERAGE_DEFINE(rev_port_toggled);
606 COVERAGE_DEFINE(rev_flow_table);
607 COVERAGE_DEFINE(rev_inconsistency);
609 /* All datapaths of a given type share a single dpif backer instance. */
614 struct timer next_expiration;
615 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
618 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
619 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
621 static struct ofport_dpif *
622 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
624 struct ofproto_dpif {
625 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
627 struct dpif_backer *backer;
629 /* Special OpenFlow rules. */
630 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
631 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
637 struct netflow *netflow;
638 struct dpif_sflow *sflow;
639 struct hmap bundles; /* Contains "struct ofbundle"s. */
640 struct mac_learning *ml;
641 struct ofmirror *mirrors[MAX_MIRRORS];
643 bool has_bonded_bundles;
647 struct hmap subfacets;
648 struct governor *governor;
651 struct table_dpif tables[N_TABLES];
652 enum revalidate_reason need_revalidate;
653 struct tag_set revalidate_set;
655 /* Support for debugging async flow mods. */
656 struct list completions;
658 bool has_bundle_action; /* True when the first bundle action appears. */
659 struct netdev_stats stats; /* To account packets generated and consumed in
664 long long int stp_last_tick;
666 /* VLAN splinters. */
667 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
668 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
671 struct sset ports; /* Set of port names. */
672 struct sset port_poll_set; /* Queued names for port_poll() reply. */
673 int port_poll_errno; /* Last errno for port_poll() reply. */
676 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
677 * for debugging the asynchronous flow_mod implementation.) */
680 /* All existing ofproto_dpif instances, indexed by ->up.name. */
681 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
683 static void ofproto_dpif_unixctl_init(void);
685 static struct ofproto_dpif *
686 ofproto_dpif_cast(const struct ofproto *ofproto)
688 assert(ofproto->ofproto_class == &ofproto_dpif_class);
689 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
692 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
694 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
696 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
697 const struct ofpbuf *, ovs_be16 initial_tci,
700 /* Packet processing. */
701 static void update_learning_table(struct ofproto_dpif *,
702 const struct flow *, int vlan,
705 #define FLOW_MISS_MAX_BATCH 50
706 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
708 /* Flow expiration. */
709 static int expire(struct dpif_backer *);
712 static void send_netflow_active_timeouts(struct ofproto_dpif *);
715 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
716 static size_t compose_sflow_action(const struct ofproto_dpif *,
717 struct ofpbuf *odp_actions,
718 const struct flow *, uint32_t odp_port);
719 static void add_mirror_actions(struct action_xlate_ctx *ctx,
720 const struct flow *flow);
721 /* Global variables. */
722 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
724 /* Initial mappings of port to bridge mappings. */
725 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
727 /* Factory functions. */
730 init(const struct shash *iface_hints)
732 struct shash_node *node;
734 /* Make a local copy, since we don't own 'iface_hints' elements. */
735 SHASH_FOR_EACH(node, iface_hints) {
736 const struct iface_hint *orig_hint = node->data;
737 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
739 new_hint->br_name = xstrdup(orig_hint->br_name);
740 new_hint->br_type = xstrdup(orig_hint->br_type);
741 new_hint->ofp_port = orig_hint->ofp_port;
743 shash_add(&init_ofp_ports, node->name, new_hint);
748 enumerate_types(struct sset *types)
750 dp_enumerate_types(types);
754 enumerate_names(const char *type, struct sset *names)
756 struct ofproto_dpif *ofproto;
759 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
760 if (strcmp(type, ofproto->up.type)) {
763 sset_add(names, ofproto->up.name);
770 del(const char *type, const char *name)
775 error = dpif_open(name, type, &dpif);
777 error = dpif_delete(dpif);
784 port_open_type(const char *datapath_type, const char *port_type)
786 return dpif_port_open_type(datapath_type, port_type);
789 /* Type functions. */
792 type_run(const char *type)
794 struct dpif_backer *backer;
798 backer = shash_find_data(&all_dpif_backers, type);
800 /* This is not necessarily a problem, since backers are only
801 * created on demand. */
805 dpif_run(backer->dpif);
807 if (timer_expired(&backer->next_expiration)) {
808 int delay = expire(backer);
809 timer_set_duration(&backer->next_expiration, delay);
812 /* Check for port changes in the dpif. */
813 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
814 struct ofproto_dpif *ofproto = NULL;
815 struct dpif_port port;
817 /* Don't report on the datapath's device. */
818 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
822 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
823 &all_ofproto_dpifs) {
824 if (sset_contains(&ofproto->ports, devname)) {
829 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
830 /* The port was removed. If we know the datapath,
831 * report it through poll_set(). If we don't, it may be
832 * notifying us of a removal we initiated, so ignore it.
833 * If there's a pending ENOBUFS, let it stand, since
834 * everything will be reevaluated. */
835 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
836 sset_add(&ofproto->port_poll_set, devname);
837 ofproto->port_poll_errno = 0;
839 dpif_port_destroy(&port);
840 } else if (!ofproto) {
841 /* The port was added, but we don't know with which
842 * ofproto we should associate it. Delete it. */
843 dpif_port_del(backer->dpif, port.port_no);
849 if (error != EAGAIN) {
850 struct ofproto_dpif *ofproto;
852 /* There was some sort of error, so propagate it to all
853 * ofprotos that use this backer. */
854 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
855 &all_ofproto_dpifs) {
856 if (ofproto->backer == backer) {
857 sset_clear(&ofproto->port_poll_set);
858 ofproto->port_poll_errno = error;
867 type_run_fast(const char *type)
869 struct dpif_backer *backer;
872 backer = shash_find_data(&all_dpif_backers, type);
874 /* This is not necessarily a problem, since backers are only
875 * created on demand. */
879 /* Handle one or more batches of upcalls, until there's nothing left to do
880 * or until we do a fixed total amount of work.
882 * We do work in batches because it can be much cheaper to set up a number
883 * of flows and fire off their patches all at once. We do multiple batches
884 * because in some cases handling a packet can cause another packet to be
885 * queued almost immediately as part of the return flow. Both
886 * optimizations can make major improvements on some benchmarks and
887 * presumably for real traffic as well. */
889 while (work < FLOW_MISS_MAX_BATCH) {
890 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
901 type_wait(const char *type)
903 struct dpif_backer *backer;
905 backer = shash_find_data(&all_dpif_backers, type);
907 /* This is not necessarily a problem, since backers are only
908 * created on demand. */
912 timer_wait(&backer->next_expiration);
915 /* Basic life-cycle. */
917 static int add_internal_flows(struct ofproto_dpif *);
919 static struct ofproto *
922 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
927 dealloc(struct ofproto *ofproto_)
929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
934 close_dpif_backer(struct dpif_backer *backer)
936 struct shash_node *node;
938 assert(backer->refcount > 0);
940 if (--backer->refcount) {
944 hmap_destroy(&backer->odp_to_ofport_map);
945 node = shash_find(&all_dpif_backers, backer->type);
947 shash_delete(&all_dpif_backers, node);
948 dpif_close(backer->dpif);
953 /* Datapath port slated for removal from datapath. */
955 struct list list_node;
960 open_dpif_backer(const char *type, struct dpif_backer **backerp)
962 struct dpif_backer *backer;
963 struct dpif_port_dump port_dump;
964 struct dpif_port port;
965 struct shash_node *node;
966 struct list garbage_list;
967 struct odp_garbage *garbage, *next;
973 backer = shash_find_data(&all_dpif_backers, type);
980 backer_name = xasprintf("ovs-%s", type);
982 /* Remove any existing datapaths, since we assume we're the only
983 * userspace controlling the datapath. */
985 dp_enumerate_names(type, &names);
986 SSET_FOR_EACH(name, &names) {
987 struct dpif *old_dpif;
989 /* Don't remove our backer if it exists. */
990 if (!strcmp(name, backer_name)) {
994 if (dpif_open(name, type, &old_dpif)) {
995 VLOG_WARN("couldn't open old datapath %s to remove it", name);
997 dpif_delete(old_dpif);
998 dpif_close(old_dpif);
1001 sset_destroy(&names);
1003 backer = xmalloc(sizeof *backer);
1005 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1008 VLOG_ERR("failed to open datapath of type %s: %s", type,
1013 backer->type = xstrdup(type);
1014 backer->refcount = 1;
1015 hmap_init(&backer->odp_to_ofport_map);
1016 timer_set_duration(&backer->next_expiration, 1000);
1019 dpif_flow_flush(backer->dpif);
1021 /* Loop through the ports already on the datapath and remove any
1022 * that we don't need anymore. */
1023 list_init(&garbage_list);
1024 dpif_port_dump_start(&port_dump, backer->dpif);
1025 while (dpif_port_dump_next(&port_dump, &port)) {
1026 node = shash_find(&init_ofp_ports, port.name);
1027 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1028 garbage = xmalloc(sizeof *garbage);
1029 garbage->odp_port = port.port_no;
1030 list_push_front(&garbage_list, &garbage->list_node);
1033 dpif_port_dump_done(&port_dump);
1035 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1036 dpif_port_del(backer->dpif, garbage->odp_port);
1037 list_remove(&garbage->list_node);
1041 shash_add(&all_dpif_backers, type, backer);
1043 error = dpif_recv_set(backer->dpif, true);
1045 VLOG_ERR("failed to listen on datapath of type %s: %s",
1046 type, strerror(error));
1047 close_dpif_backer(backer);
1055 construct(struct ofproto *ofproto_)
1057 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1058 struct shash_node *node, *next;
1063 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1068 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1069 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1071 ofproto->n_matches = 0;
1073 ofproto->netflow = NULL;
1074 ofproto->sflow = NULL;
1075 ofproto->stp = NULL;
1076 hmap_init(&ofproto->bundles);
1077 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1078 for (i = 0; i < MAX_MIRRORS; i++) {
1079 ofproto->mirrors[i] = NULL;
1081 ofproto->has_bonded_bundles = false;
1083 hmap_init(&ofproto->facets);
1084 hmap_init(&ofproto->subfacets);
1085 ofproto->governor = NULL;
1087 for (i = 0; i < N_TABLES; i++) {
1088 struct table_dpif *table = &ofproto->tables[i];
1090 table->catchall_table = NULL;
1091 table->other_table = NULL;
1092 table->basis = random_uint32();
1094 ofproto->need_revalidate = 0;
1095 tag_set_init(&ofproto->revalidate_set);
1097 list_init(&ofproto->completions);
1099 ofproto_dpif_unixctl_init();
1101 ofproto->has_mirrors = false;
1102 ofproto->has_bundle_action = false;
1104 hmap_init(&ofproto->vlandev_map);
1105 hmap_init(&ofproto->realdev_vid_map);
1107 sset_init(&ofproto->ports);
1108 sset_init(&ofproto->port_poll_set);
1109 ofproto->port_poll_errno = 0;
1111 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1112 const struct iface_hint *iface_hint = node->data;
1114 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1115 /* Check if the datapath already has this port. */
1116 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1117 sset_add(&ofproto->ports, node->name);
1120 free(iface_hint->br_name);
1121 free(iface_hint->br_type);
1122 shash_delete(&init_ofp_ports, node);
1126 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1127 hash_string(ofproto->up.name, 0));
1128 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1130 ofproto_init_tables(ofproto_, N_TABLES);
1131 error = add_internal_flows(ofproto);
1132 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1138 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1139 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1141 struct ofputil_flow_mod fm;
1144 match_init_catchall(&fm.match);
1146 match_set_reg(&fm.match, 0, id);
1147 fm.new_cookie = htonll(0);
1148 fm.cookie = htonll(0);
1149 fm.cookie_mask = htonll(0);
1150 fm.table_id = TBL_INTERNAL;
1151 fm.command = OFPFC_ADD;
1152 fm.idle_timeout = 0;
1153 fm.hard_timeout = 0;
1157 fm.ofpacts = ofpacts->data;
1158 fm.ofpacts_len = ofpacts->size;
1160 error = ofproto_flow_mod(&ofproto->up, &fm);
1162 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1163 id, ofperr_to_string(error));
1167 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1168 assert(*rulep != NULL);
1174 add_internal_flows(struct ofproto_dpif *ofproto)
1176 struct ofpact_controller *controller;
1177 uint64_t ofpacts_stub[128 / 8];
1178 struct ofpbuf ofpacts;
1182 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1185 controller = ofpact_put_CONTROLLER(&ofpacts);
1186 controller->max_len = UINT16_MAX;
1187 controller->controller_id = 0;
1188 controller->reason = OFPR_NO_MATCH;
1189 ofpact_pad(&ofpacts);
1191 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1196 ofpbuf_clear(&ofpacts);
1197 error = add_internal_flow(ofproto, id++, &ofpacts,
1198 &ofproto->no_packet_in_rule);
1203 complete_operations(struct ofproto_dpif *ofproto)
1205 struct dpif_completion *c, *next;
1207 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1208 ofoperation_complete(c->op, 0);
1209 list_remove(&c->list_node);
1215 destruct(struct ofproto *ofproto_)
1217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1218 struct rule_dpif *rule, *next_rule;
1219 struct oftable *table;
1222 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1223 complete_operations(ofproto);
1225 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1226 struct cls_cursor cursor;
1228 cls_cursor_init(&cursor, &table->cls, NULL);
1229 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1230 ofproto_rule_destroy(&rule->up);
1234 for (i = 0; i < MAX_MIRRORS; i++) {
1235 mirror_destroy(ofproto->mirrors[i]);
1238 netflow_destroy(ofproto->netflow);
1239 dpif_sflow_destroy(ofproto->sflow);
1240 hmap_destroy(&ofproto->bundles);
1241 mac_learning_destroy(ofproto->ml);
1243 hmap_destroy(&ofproto->facets);
1244 hmap_destroy(&ofproto->subfacets);
1245 governor_destroy(ofproto->governor);
1247 hmap_destroy(&ofproto->vlandev_map);
1248 hmap_destroy(&ofproto->realdev_vid_map);
1250 sset_destroy(&ofproto->ports);
1251 sset_destroy(&ofproto->port_poll_set);
1253 close_dpif_backer(ofproto->backer);
1257 run_fast(struct ofproto *ofproto_)
1259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1260 struct ofport_dpif *ofport;
1262 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1263 port_run_fast(ofport);
1270 run(struct ofproto *ofproto_)
1272 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1273 struct ofport_dpif *ofport;
1274 struct ofbundle *bundle;
1278 complete_operations(ofproto);
1281 error = run_fast(ofproto_);
1286 if (ofproto->netflow) {
1287 if (netflow_run(ofproto->netflow)) {
1288 send_netflow_active_timeouts(ofproto);
1291 if (ofproto->sflow) {
1292 dpif_sflow_run(ofproto->sflow);
1295 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1298 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1303 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1305 /* Now revalidate if there's anything to do. */
1306 if (ofproto->need_revalidate
1307 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1308 struct tag_set revalidate_set = ofproto->revalidate_set;
1309 bool revalidate_all = ofproto->need_revalidate;
1310 struct facet *facet;
1312 switch (ofproto->need_revalidate) {
1313 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1314 case REV_STP: COVERAGE_INC(rev_stp); break;
1315 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1316 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1317 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1320 /* Clear the revalidation flags. */
1321 tag_set_init(&ofproto->revalidate_set);
1322 ofproto->need_revalidate = 0;
1324 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1326 || tag_set_intersects(&revalidate_set, facet->tags)) {
1327 facet_revalidate(facet);
1332 /* Check the consistency of a random facet, to aid debugging. */
1333 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1334 struct facet *facet;
1336 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1337 struct facet, hmap_node);
1338 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1339 if (!facet_check_consistency(facet)) {
1340 ofproto->need_revalidate = REV_INCONSISTENCY;
1345 if (ofproto->governor) {
1348 governor_run(ofproto->governor);
1350 /* If the governor has shrunk to its minimum size and the number of
1351 * subfacets has dwindled, then drop the governor entirely.
1353 * For hysteresis, the number of subfacets to drop the governor is
1354 * smaller than the number needed to trigger its creation. */
1355 n_subfacets = hmap_count(&ofproto->subfacets);
1356 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1357 && governor_is_idle(ofproto->governor)) {
1358 governor_destroy(ofproto->governor);
1359 ofproto->governor = NULL;
1367 wait(struct ofproto *ofproto_)
1369 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1370 struct ofport_dpif *ofport;
1371 struct ofbundle *bundle;
1373 if (!clogged && !list_is_empty(&ofproto->completions)) {
1374 poll_immediate_wake();
1377 dpif_wait(ofproto->backer->dpif);
1378 dpif_recv_wait(ofproto->backer->dpif);
1379 if (ofproto->sflow) {
1380 dpif_sflow_wait(ofproto->sflow);
1382 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1383 poll_immediate_wake();
1385 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1388 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1389 bundle_wait(bundle);
1391 if (ofproto->netflow) {
1392 netflow_wait(ofproto->netflow);
1394 mac_learning_wait(ofproto->ml);
1396 if (ofproto->need_revalidate) {
1397 /* Shouldn't happen, but if it does just go around again. */
1398 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1399 poll_immediate_wake();
1401 if (ofproto->governor) {
1402 governor_wait(ofproto->governor);
1407 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1409 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1411 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1412 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1416 flush(struct ofproto *ofproto_)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1419 struct subfacet *subfacet, *next_subfacet;
1420 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1424 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1425 &ofproto->subfacets) {
1426 if (subfacet->path != SF_NOT_INSTALLED) {
1427 batch[n_batch++] = subfacet;
1428 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1429 subfacet_destroy_batch(ofproto, batch, n_batch);
1433 subfacet_destroy(subfacet);
1438 subfacet_destroy_batch(ofproto, batch, n_batch);
1443 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1444 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1446 *arp_match_ip = true;
1447 *actions = (OFPUTIL_A_OUTPUT |
1448 OFPUTIL_A_SET_VLAN_VID |
1449 OFPUTIL_A_SET_VLAN_PCP |
1450 OFPUTIL_A_STRIP_VLAN |
1451 OFPUTIL_A_SET_DL_SRC |
1452 OFPUTIL_A_SET_DL_DST |
1453 OFPUTIL_A_SET_NW_SRC |
1454 OFPUTIL_A_SET_NW_DST |
1455 OFPUTIL_A_SET_NW_TOS |
1456 OFPUTIL_A_SET_TP_SRC |
1457 OFPUTIL_A_SET_TP_DST |
1462 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1464 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1465 struct dpif_dp_stats s;
1467 strcpy(ots->name, "classifier");
1469 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1471 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1472 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1475 static struct ofport *
1478 struct ofport_dpif *port = xmalloc(sizeof *port);
1483 port_dealloc(struct ofport *port_)
1485 struct ofport_dpif *port = ofport_dpif_cast(port_);
1490 port_construct(struct ofport *port_)
1492 struct ofport_dpif *port = ofport_dpif_cast(port_);
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1494 struct dpif_port dpif_port;
1497 ofproto->need_revalidate = REV_RECONFIGURE;
1498 port->bundle = NULL;
1500 port->tag = tag_create_random();
1501 port->may_enable = true;
1502 port->stp_port = NULL;
1503 port->stp_state = STP_DISABLED;
1504 hmap_init(&port->priorities);
1505 port->realdev_ofp_port = 0;
1506 port->vlandev_vid = 0;
1507 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1509 error = dpif_port_query_by_name(ofproto->backer->dpif,
1510 netdev_get_name(port->up.netdev),
1516 port->odp_port = dpif_port.port_no;
1518 /* Sanity-check that a mapping doesn't already exist. This
1519 * shouldn't happen. */
1520 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1521 VLOG_ERR("port %s already has an OpenFlow port number\n",
1526 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1527 hash_int(port->odp_port, 0));
1529 if (ofproto->sflow) {
1530 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1537 port_destruct(struct ofport *port_)
1539 struct ofport_dpif *port = ofport_dpif_cast(port_);
1540 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1541 const char *devname = netdev_get_name(port->up.netdev);
1543 if (dpif_port_exists(ofproto->backer->dpif, devname)) {
1544 /* The underlying device is still there, so delete it. This
1545 * happens when the ofproto is being destroyed, since the caller
1546 * assumes that removal of attached ports will happen as part of
1548 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1551 sset_find_and_delete(&ofproto->ports, devname);
1552 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1553 ofproto->need_revalidate = REV_RECONFIGURE;
1554 bundle_remove(port_);
1555 set_cfm(port_, NULL);
1556 if (ofproto->sflow) {
1557 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1560 ofport_clear_priorities(port);
1561 hmap_destroy(&port->priorities);
1565 port_modified(struct ofport *port_)
1567 struct ofport_dpif *port = ofport_dpif_cast(port_);
1569 if (port->bundle && port->bundle->bond) {
1570 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1575 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1577 struct ofport_dpif *port = ofport_dpif_cast(port_);
1578 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1579 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1581 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1582 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1583 OFPUTIL_PC_NO_PACKET_IN)) {
1584 ofproto->need_revalidate = REV_RECONFIGURE;
1586 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1587 bundle_update(port->bundle);
1593 set_sflow(struct ofproto *ofproto_,
1594 const struct ofproto_sflow_options *sflow_options)
1596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597 struct dpif_sflow *ds = ofproto->sflow;
1599 if (sflow_options) {
1601 struct ofport_dpif *ofport;
1603 ds = ofproto->sflow = dpif_sflow_create();
1604 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1605 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1607 ofproto->need_revalidate = REV_RECONFIGURE;
1609 dpif_sflow_set_options(ds, sflow_options);
1612 dpif_sflow_destroy(ds);
1613 ofproto->need_revalidate = REV_RECONFIGURE;
1614 ofproto->sflow = NULL;
1621 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1623 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1630 struct ofproto_dpif *ofproto;
1632 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1633 ofproto->need_revalidate = REV_RECONFIGURE;
1634 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1637 if (cfm_configure(ofport->cfm, s)) {
1643 cfm_destroy(ofport->cfm);
1649 get_cfm_fault(const struct ofport *ofport_)
1651 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1653 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1657 get_cfm_opup(const struct ofport *ofport_)
1659 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1661 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1665 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1668 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1671 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1679 get_cfm_health(const struct ofport *ofport_)
1681 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1683 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1686 /* Spanning Tree. */
1689 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1691 struct ofproto_dpif *ofproto = ofproto_;
1692 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1693 struct ofport_dpif *ofport;
1695 ofport = stp_port_get_aux(sp);
1697 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1698 ofproto->up.name, port_num);
1700 struct eth_header *eth = pkt->l2;
1702 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1703 if (eth_addr_is_zero(eth->eth_src)) {
1704 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1705 "with unknown MAC", ofproto->up.name, port_num);
1707 send_packet(ofport, pkt);
1713 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1715 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1719 /* Only revalidate flows if the configuration changed. */
1720 if (!s != !ofproto->stp) {
1721 ofproto->need_revalidate = REV_RECONFIGURE;
1725 if (!ofproto->stp) {
1726 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1727 send_bpdu_cb, ofproto);
1728 ofproto->stp_last_tick = time_msec();
1731 stp_set_bridge_id(ofproto->stp, s->system_id);
1732 stp_set_bridge_priority(ofproto->stp, s->priority);
1733 stp_set_hello_time(ofproto->stp, s->hello_time);
1734 stp_set_max_age(ofproto->stp, s->max_age);
1735 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1737 struct ofport *ofport;
1739 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1740 set_stp_port(ofport, NULL);
1743 stp_destroy(ofproto->stp);
1744 ofproto->stp = NULL;
1751 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1757 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1758 s->designated_root = stp_get_designated_root(ofproto->stp);
1759 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1768 update_stp_port_state(struct ofport_dpif *ofport)
1770 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1771 enum stp_state state;
1773 /* Figure out new state. */
1774 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1778 if (ofport->stp_state != state) {
1779 enum ofputil_port_state of_state;
1782 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1783 netdev_get_name(ofport->up.netdev),
1784 stp_state_name(ofport->stp_state),
1785 stp_state_name(state));
1786 if (stp_learn_in_state(ofport->stp_state)
1787 != stp_learn_in_state(state)) {
1788 /* xxx Learning action flows should also be flushed. */
1789 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1791 fwd_change = stp_forward_in_state(ofport->stp_state)
1792 != stp_forward_in_state(state);
1794 ofproto->need_revalidate = REV_STP;
1795 ofport->stp_state = state;
1796 ofport->stp_state_entered = time_msec();
1798 if (fwd_change && ofport->bundle) {
1799 bundle_update(ofport->bundle);
1802 /* Update the STP state bits in the OpenFlow port description. */
1803 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1804 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1805 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1806 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1807 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1809 ofproto_port_set_state(&ofport->up, of_state);
1813 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1814 * caller is responsible for assigning STP port numbers and ensuring
1815 * there are no duplicates. */
1817 set_stp_port(struct ofport *ofport_,
1818 const struct ofproto_port_stp_settings *s)
1820 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1822 struct stp_port *sp = ofport->stp_port;
1824 if (!s || !s->enable) {
1826 ofport->stp_port = NULL;
1827 stp_port_disable(sp);
1828 update_stp_port_state(ofport);
1831 } else if (sp && stp_port_no(sp) != s->port_num
1832 && ofport == stp_port_get_aux(sp)) {
1833 /* The port-id changed, so disable the old one if it's not
1834 * already in use by another port. */
1835 stp_port_disable(sp);
1838 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1839 stp_port_enable(sp);
1841 stp_port_set_aux(sp, ofport);
1842 stp_port_set_priority(sp, s->priority);
1843 stp_port_set_path_cost(sp, s->path_cost);
1845 update_stp_port_state(ofport);
1851 get_stp_port_status(struct ofport *ofport_,
1852 struct ofproto_port_stp_status *s)
1854 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1856 struct stp_port *sp = ofport->stp_port;
1858 if (!ofproto->stp || !sp) {
1864 s->port_id = stp_port_get_id(sp);
1865 s->state = stp_port_get_state(sp);
1866 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1867 s->role = stp_port_get_role(sp);
1868 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1874 stp_run(struct ofproto_dpif *ofproto)
1877 long long int now = time_msec();
1878 long long int elapsed = now - ofproto->stp_last_tick;
1879 struct stp_port *sp;
1882 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1883 ofproto->stp_last_tick = now;
1885 while (stp_get_changed_port(ofproto->stp, &sp)) {
1886 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1889 update_stp_port_state(ofport);
1893 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1894 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1900 stp_wait(struct ofproto_dpif *ofproto)
1903 poll_timer_wait(1000);
1907 /* Returns true if STP should process 'flow'. */
1909 stp_should_process_flow(const struct flow *flow)
1911 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1915 stp_process_packet(const struct ofport_dpif *ofport,
1916 const struct ofpbuf *packet)
1918 struct ofpbuf payload = *packet;
1919 struct eth_header *eth = payload.data;
1920 struct stp_port *sp = ofport->stp_port;
1922 /* Sink packets on ports that have STP disabled when the bridge has
1924 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1928 /* Trim off padding on payload. */
1929 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1930 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1933 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1934 stp_received_bpdu(sp, payload.data, payload.size);
1938 static struct priority_to_dscp *
1939 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1941 struct priority_to_dscp *pdscp;
1944 hash = hash_int(priority, 0);
1945 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1946 if (pdscp->priority == priority) {
1954 ofport_clear_priorities(struct ofport_dpif *ofport)
1956 struct priority_to_dscp *pdscp, *next;
1958 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1959 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1965 set_queues(struct ofport *ofport_,
1966 const struct ofproto_port_queue *qdscp_list,
1969 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1971 struct hmap new = HMAP_INITIALIZER(&new);
1974 for (i = 0; i < n_qdscp; i++) {
1975 struct priority_to_dscp *pdscp;
1979 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1980 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1985 pdscp = get_priority(ofport, priority);
1987 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1989 pdscp = xmalloc(sizeof *pdscp);
1990 pdscp->priority = priority;
1992 ofproto->need_revalidate = REV_RECONFIGURE;
1995 if (pdscp->dscp != dscp) {
1997 ofproto->need_revalidate = REV_RECONFIGURE;
2000 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2003 if (!hmap_is_empty(&ofport->priorities)) {
2004 ofport_clear_priorities(ofport);
2005 ofproto->need_revalidate = REV_RECONFIGURE;
2008 hmap_swap(&new, &ofport->priorities);
2016 /* Expires all MAC learning entries associated with 'bundle' and forces its
2017 * ofproto to revalidate every flow.
2019 * Normally MAC learning entries are removed only from the ofproto associated
2020 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2021 * are removed from every ofproto. When patch ports and SLB bonds are in use
2022 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2023 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2024 * with the host from which it migrated. */
2026 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2028 struct ofproto_dpif *ofproto = bundle->ofproto;
2029 struct mac_learning *ml = ofproto->ml;
2030 struct mac_entry *mac, *next_mac;
2032 ofproto->need_revalidate = REV_RECONFIGURE;
2033 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2034 if (mac->port.p == bundle) {
2036 struct ofproto_dpif *o;
2038 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2040 struct mac_entry *e;
2042 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2045 tag_set_add(&o->revalidate_set, e->tag);
2046 mac_learning_expire(o->ml, e);
2052 mac_learning_expire(ml, mac);
2057 static struct ofbundle *
2058 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2060 struct ofbundle *bundle;
2062 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2063 &ofproto->bundles) {
2064 if (bundle->aux == aux) {
2071 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2072 * ones that are found to 'bundles'. */
2074 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2075 void **auxes, size_t n_auxes,
2076 struct hmapx *bundles)
2080 hmapx_init(bundles);
2081 for (i = 0; i < n_auxes; i++) {
2082 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2084 hmapx_add(bundles, bundle);
2090 bundle_update(struct ofbundle *bundle)
2092 struct ofport_dpif *port;
2094 bundle->floodable = true;
2095 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2096 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2097 || !stp_forward_in_state(port->stp_state)) {
2098 bundle->floodable = false;
2105 bundle_del_port(struct ofport_dpif *port)
2107 struct ofbundle *bundle = port->bundle;
2109 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2111 list_remove(&port->bundle_node);
2112 port->bundle = NULL;
2115 lacp_slave_unregister(bundle->lacp, port);
2118 bond_slave_unregister(bundle->bond, port);
2121 bundle_update(bundle);
2125 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2126 struct lacp_slave_settings *lacp,
2127 uint32_t bond_stable_id)
2129 struct ofport_dpif *port;
2131 port = get_ofp_port(bundle->ofproto, ofp_port);
2136 if (port->bundle != bundle) {
2137 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2139 bundle_del_port(port);
2142 port->bundle = bundle;
2143 list_push_back(&bundle->ports, &port->bundle_node);
2144 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2145 || !stp_forward_in_state(port->stp_state)) {
2146 bundle->floodable = false;
2150 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2151 lacp_slave_register(bundle->lacp, port, lacp);
2154 port->bond_stable_id = bond_stable_id;
2160 bundle_destroy(struct ofbundle *bundle)
2162 struct ofproto_dpif *ofproto;
2163 struct ofport_dpif *port, *next_port;
2170 ofproto = bundle->ofproto;
2171 for (i = 0; i < MAX_MIRRORS; i++) {
2172 struct ofmirror *m = ofproto->mirrors[i];
2174 if (m->out == bundle) {
2176 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2177 || hmapx_find_and_delete(&m->dsts, bundle)) {
2178 ofproto->need_revalidate = REV_RECONFIGURE;
2183 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2184 bundle_del_port(port);
2187 bundle_flush_macs(bundle, true);
2188 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2190 free(bundle->trunks);
2191 lacp_destroy(bundle->lacp);
2192 bond_destroy(bundle->bond);
2197 bundle_set(struct ofproto *ofproto_, void *aux,
2198 const struct ofproto_bundle_settings *s)
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2201 bool need_flush = false;
2202 struct ofport_dpif *port;
2203 struct ofbundle *bundle;
2204 unsigned long *trunks;
2210 bundle_destroy(bundle_lookup(ofproto, aux));
2214 assert(s->n_slaves == 1 || s->bond != NULL);
2215 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2217 bundle = bundle_lookup(ofproto, aux);
2219 bundle = xmalloc(sizeof *bundle);
2221 bundle->ofproto = ofproto;
2222 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2223 hash_pointer(aux, 0));
2225 bundle->name = NULL;
2227 list_init(&bundle->ports);
2228 bundle->vlan_mode = PORT_VLAN_TRUNK;
2230 bundle->trunks = NULL;
2231 bundle->use_priority_tags = s->use_priority_tags;
2232 bundle->lacp = NULL;
2233 bundle->bond = NULL;
2235 bundle->floodable = true;
2237 bundle->src_mirrors = 0;
2238 bundle->dst_mirrors = 0;
2239 bundle->mirror_out = 0;
2242 if (!bundle->name || strcmp(s->name, bundle->name)) {
2244 bundle->name = xstrdup(s->name);
2249 if (!bundle->lacp) {
2250 ofproto->need_revalidate = REV_RECONFIGURE;
2251 bundle->lacp = lacp_create();
2253 lacp_configure(bundle->lacp, s->lacp);
2255 lacp_destroy(bundle->lacp);
2256 bundle->lacp = NULL;
2259 /* Update set of ports. */
2261 for (i = 0; i < s->n_slaves; i++) {
2262 if (!bundle_add_port(bundle, s->slaves[i],
2263 s->lacp ? &s->lacp_slaves[i] : NULL,
2264 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2268 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2269 struct ofport_dpif *next_port;
2271 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2272 for (i = 0; i < s->n_slaves; i++) {
2273 if (s->slaves[i] == port->up.ofp_port) {
2278 bundle_del_port(port);
2282 assert(list_size(&bundle->ports) <= s->n_slaves);
2284 if (list_is_empty(&bundle->ports)) {
2285 bundle_destroy(bundle);
2289 /* Set VLAN tagging mode */
2290 if (s->vlan_mode != bundle->vlan_mode
2291 || s->use_priority_tags != bundle->use_priority_tags) {
2292 bundle->vlan_mode = s->vlan_mode;
2293 bundle->use_priority_tags = s->use_priority_tags;
2298 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2299 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2301 if (vlan != bundle->vlan) {
2302 bundle->vlan = vlan;
2306 /* Get trunked VLANs. */
2307 switch (s->vlan_mode) {
2308 case PORT_VLAN_ACCESS:
2312 case PORT_VLAN_TRUNK:
2313 trunks = CONST_CAST(unsigned long *, s->trunks);
2316 case PORT_VLAN_NATIVE_UNTAGGED:
2317 case PORT_VLAN_NATIVE_TAGGED:
2318 if (vlan != 0 && (!s->trunks
2319 || !bitmap_is_set(s->trunks, vlan)
2320 || bitmap_is_set(s->trunks, 0))) {
2321 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2323 trunks = bitmap_clone(s->trunks, 4096);
2325 trunks = bitmap_allocate1(4096);
2327 bitmap_set1(trunks, vlan);
2328 bitmap_set0(trunks, 0);
2330 trunks = CONST_CAST(unsigned long *, s->trunks);
2337 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2338 free(bundle->trunks);
2339 if (trunks == s->trunks) {
2340 bundle->trunks = vlan_bitmap_clone(trunks);
2342 bundle->trunks = trunks;
2347 if (trunks != s->trunks) {
2352 if (!list_is_short(&bundle->ports)) {
2353 bundle->ofproto->has_bonded_bundles = true;
2355 if (bond_reconfigure(bundle->bond, s->bond)) {
2356 ofproto->need_revalidate = REV_RECONFIGURE;
2359 bundle->bond = bond_create(s->bond);
2360 ofproto->need_revalidate = REV_RECONFIGURE;
2363 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2364 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2368 bond_destroy(bundle->bond);
2369 bundle->bond = NULL;
2372 /* If we changed something that would affect MAC learning, un-learn
2373 * everything on this port and force flow revalidation. */
2375 bundle_flush_macs(bundle, false);
2382 bundle_remove(struct ofport *port_)
2384 struct ofport_dpif *port = ofport_dpif_cast(port_);
2385 struct ofbundle *bundle = port->bundle;
2388 bundle_del_port(port);
2389 if (list_is_empty(&bundle->ports)) {
2390 bundle_destroy(bundle);
2391 } else if (list_is_short(&bundle->ports)) {
2392 bond_destroy(bundle->bond);
2393 bundle->bond = NULL;
2399 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2402 struct ofport_dpif *port = port_;
2403 uint8_t ea[ETH_ADDR_LEN];
2406 error = netdev_get_etheraddr(port->up.netdev, ea);
2408 struct ofpbuf packet;
2411 ofpbuf_init(&packet, 0);
2412 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2414 memcpy(packet_pdu, pdu, pdu_size);
2416 send_packet(port, &packet);
2417 ofpbuf_uninit(&packet);
2419 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2420 "%s (%s)", port->bundle->name,
2421 netdev_get_name(port->up.netdev), strerror(error));
2426 bundle_send_learning_packets(struct ofbundle *bundle)
2428 struct ofproto_dpif *ofproto = bundle->ofproto;
2429 int error, n_packets, n_errors;
2430 struct mac_entry *e;
2432 error = n_packets = n_errors = 0;
2433 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2434 if (e->port.p != bundle) {
2435 struct ofpbuf *learning_packet;
2436 struct ofport_dpif *port;
2440 /* The assignment to "port" is unnecessary but makes "grep"ing for
2441 * struct ofport_dpif more effective. */
2442 learning_packet = bond_compose_learning_packet(bundle->bond,
2446 ret = send_packet(port, learning_packet);
2447 ofpbuf_delete(learning_packet);
2457 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2458 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2459 "packets, last error was: %s",
2460 bundle->name, n_errors, n_packets, strerror(error));
2462 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2463 bundle->name, n_packets);
2468 bundle_run(struct ofbundle *bundle)
2471 lacp_run(bundle->lacp, send_pdu_cb);
2474 struct ofport_dpif *port;
2476 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2477 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2480 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2481 lacp_status(bundle->lacp));
2482 if (bond_should_send_learning_packets(bundle->bond)) {
2483 bundle_send_learning_packets(bundle);
2489 bundle_wait(struct ofbundle *bundle)
2492 lacp_wait(bundle->lacp);
2495 bond_wait(bundle->bond);
2502 mirror_scan(struct ofproto_dpif *ofproto)
2506 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2507 if (!ofproto->mirrors[idx]) {
2514 static struct ofmirror *
2515 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2519 for (i = 0; i < MAX_MIRRORS; i++) {
2520 struct ofmirror *mirror = ofproto->mirrors[i];
2521 if (mirror && mirror->aux == aux) {
2529 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2531 mirror_update_dups(struct ofproto_dpif *ofproto)
2535 for (i = 0; i < MAX_MIRRORS; i++) {
2536 struct ofmirror *m = ofproto->mirrors[i];
2539 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2543 for (i = 0; i < MAX_MIRRORS; i++) {
2544 struct ofmirror *m1 = ofproto->mirrors[i];
2551 for (j = i + 1; j < MAX_MIRRORS; j++) {
2552 struct ofmirror *m2 = ofproto->mirrors[j];
2554 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2555 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2556 m2->dup_mirrors |= m1->dup_mirrors;
2563 mirror_set(struct ofproto *ofproto_, void *aux,
2564 const struct ofproto_mirror_settings *s)
2566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2567 mirror_mask_t mirror_bit;
2568 struct ofbundle *bundle;
2569 struct ofmirror *mirror;
2570 struct ofbundle *out;
2571 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2572 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2575 mirror = mirror_lookup(ofproto, aux);
2577 mirror_destroy(mirror);
2583 idx = mirror_scan(ofproto);
2585 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2587 ofproto->up.name, MAX_MIRRORS, s->name);
2591 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2592 mirror->ofproto = ofproto;
2595 mirror->out_vlan = -1;
2596 mirror->name = NULL;
2599 if (!mirror->name || strcmp(s->name, mirror->name)) {
2601 mirror->name = xstrdup(s->name);
2604 /* Get the new configuration. */
2605 if (s->out_bundle) {
2606 out = bundle_lookup(ofproto, s->out_bundle);
2608 mirror_destroy(mirror);
2614 out_vlan = s->out_vlan;
2616 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2617 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2619 /* If the configuration has not changed, do nothing. */
2620 if (hmapx_equals(&srcs, &mirror->srcs)
2621 && hmapx_equals(&dsts, &mirror->dsts)
2622 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2623 && mirror->out == out
2624 && mirror->out_vlan == out_vlan)
2626 hmapx_destroy(&srcs);
2627 hmapx_destroy(&dsts);
2631 hmapx_swap(&srcs, &mirror->srcs);
2632 hmapx_destroy(&srcs);
2634 hmapx_swap(&dsts, &mirror->dsts);
2635 hmapx_destroy(&dsts);
2637 free(mirror->vlans);
2638 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2641 mirror->out_vlan = out_vlan;
2643 /* Update bundles. */
2644 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2645 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2646 if (hmapx_contains(&mirror->srcs, bundle)) {
2647 bundle->src_mirrors |= mirror_bit;
2649 bundle->src_mirrors &= ~mirror_bit;
2652 if (hmapx_contains(&mirror->dsts, bundle)) {
2653 bundle->dst_mirrors |= mirror_bit;
2655 bundle->dst_mirrors &= ~mirror_bit;
2658 if (mirror->out == bundle) {
2659 bundle->mirror_out |= mirror_bit;
2661 bundle->mirror_out &= ~mirror_bit;
2665 ofproto->need_revalidate = REV_RECONFIGURE;
2666 ofproto->has_mirrors = true;
2667 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2668 mirror_update_dups(ofproto);
2674 mirror_destroy(struct ofmirror *mirror)
2676 struct ofproto_dpif *ofproto;
2677 mirror_mask_t mirror_bit;
2678 struct ofbundle *bundle;
2685 ofproto = mirror->ofproto;
2686 ofproto->need_revalidate = REV_RECONFIGURE;
2687 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2689 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2690 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2691 bundle->src_mirrors &= ~mirror_bit;
2692 bundle->dst_mirrors &= ~mirror_bit;
2693 bundle->mirror_out &= ~mirror_bit;
2696 hmapx_destroy(&mirror->srcs);
2697 hmapx_destroy(&mirror->dsts);
2698 free(mirror->vlans);
2700 ofproto->mirrors[mirror->idx] = NULL;
2704 mirror_update_dups(ofproto);
2706 ofproto->has_mirrors = false;
2707 for (i = 0; i < MAX_MIRRORS; i++) {
2708 if (ofproto->mirrors[i]) {
2709 ofproto->has_mirrors = true;
2716 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2717 uint64_t *packets, uint64_t *bytes)
2719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2720 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2723 *packets = *bytes = UINT64_MAX;
2727 *packets = mirror->packet_count;
2728 *bytes = mirror->byte_count;
2734 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2737 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2738 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2744 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2747 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2748 return bundle && bundle->mirror_out != 0;
2752 forward_bpdu_changed(struct ofproto *ofproto_)
2754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2755 ofproto->need_revalidate = REV_RECONFIGURE;
2759 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2762 mac_learning_set_idle_time(ofproto->ml, idle_time);
2767 static struct ofport_dpif *
2768 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2770 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2771 return ofport ? ofport_dpif_cast(ofport) : NULL;
2774 static struct ofport_dpif *
2775 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2777 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2781 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2782 struct ofproto_port *ofproto_port,
2783 struct dpif_port *dpif_port)
2785 ofproto_port->name = dpif_port->name;
2786 ofproto_port->type = dpif_port->type;
2787 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2791 port_run_fast(struct ofport_dpif *ofport)
2793 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2794 struct ofpbuf packet;
2796 ofpbuf_init(&packet, 0);
2797 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2798 send_packet(ofport, &packet);
2799 ofpbuf_uninit(&packet);
2804 port_run(struct ofport_dpif *ofport)
2806 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2807 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2808 bool enable = netdev_get_carrier(ofport->up.netdev);
2810 ofport->carrier_seq = carrier_seq;
2812 port_run_fast(ofport);
2814 int cfm_opup = cfm_get_opup(ofport->cfm);
2816 cfm_run(ofport->cfm);
2817 enable = enable && !cfm_get_fault(ofport->cfm);
2819 if (cfm_opup >= 0) {
2820 enable = enable && cfm_opup;
2824 if (ofport->bundle) {
2825 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2826 if (carrier_changed) {
2827 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2831 if (ofport->may_enable != enable) {
2832 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2834 if (ofproto->has_bundle_action) {
2835 ofproto->need_revalidate = REV_PORT_TOGGLED;
2839 ofport->may_enable = enable;
2843 port_wait(struct ofport_dpif *ofport)
2846 cfm_wait(ofport->cfm);
2851 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2852 struct ofproto_port *ofproto_port)
2854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2855 struct dpif_port dpif_port;
2858 if (!sset_contains(&ofproto->ports, devname)) {
2861 error = dpif_port_query_by_name(ofproto->backer->dpif,
2862 devname, &dpif_port);
2864 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2870 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873 uint32_t odp_port = UINT32_MAX;
2876 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2878 sset_add(&ofproto->ports, netdev_get_name(netdev));
2884 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2887 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2890 if (odp_port != OFPP_NONE) {
2891 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2894 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2896 /* The caller is going to close ofport->up.netdev. If this is a
2897 * bonded port, then the bond is using that netdev, so remove it
2898 * from the bond. The client will need to reconfigure everything
2899 * after deleting ports, so then the slave will get re-added. */
2900 bundle_remove(&ofport->up);
2907 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2909 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2912 error = netdev_get_stats(ofport->up.netdev, stats);
2914 if (!error && ofport->odp_port == OVSP_LOCAL) {
2915 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2917 /* ofproto->stats.tx_packets represents packets that we created
2918 * internally and sent to some port (e.g. packets sent with
2919 * send_packet()). Account for them as if they had come from
2920 * OFPP_LOCAL and got forwarded. */
2922 if (stats->rx_packets != UINT64_MAX) {
2923 stats->rx_packets += ofproto->stats.tx_packets;
2926 if (stats->rx_bytes != UINT64_MAX) {
2927 stats->rx_bytes += ofproto->stats.tx_bytes;
2930 /* ofproto->stats.rx_packets represents packets that were received on
2931 * some port and we processed internally and dropped (e.g. STP).
2932 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2934 if (stats->tx_packets != UINT64_MAX) {
2935 stats->tx_packets += ofproto->stats.rx_packets;
2938 if (stats->tx_bytes != UINT64_MAX) {
2939 stats->tx_bytes += ofproto->stats.rx_bytes;
2946 /* Account packets for LOCAL port. */
2948 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2949 size_t tx_size, size_t rx_size)
2951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2954 ofproto->stats.rx_packets++;
2955 ofproto->stats.rx_bytes += rx_size;
2958 ofproto->stats.tx_packets++;
2959 ofproto->stats.tx_bytes += tx_size;
2963 struct port_dump_state {
2969 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2971 struct port_dump_state *state;
2973 *statep = state = xmalloc(sizeof *state);
2980 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2981 struct ofproto_port *port)
2983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2984 struct port_dump_state *state = state_;
2985 struct sset_node *node;
2987 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
2991 error = port_query_by_name(ofproto_, node->name, port);
2992 if (error != ENODEV) {
3001 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3003 struct port_dump_state *state = state_;
3010 port_poll(const struct ofproto *ofproto_, char **devnamep)
3012 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3014 if (ofproto->port_poll_errno) {
3015 int error = ofproto->port_poll_errno;
3016 ofproto->port_poll_errno = 0;
3020 if (sset_is_empty(&ofproto->port_poll_set)) {
3024 *devnamep = sset_pop(&ofproto->port_poll_set);
3029 port_poll_wait(const struct ofproto *ofproto_)
3031 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3032 dpif_port_poll_wait(ofproto->backer->dpif);
3036 port_is_lacp_current(const struct ofport *ofport_)
3038 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3039 return (ofport->bundle && ofport->bundle->lacp
3040 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3044 /* Upcall handling. */
3046 /* Flow miss batching.
3048 * Some dpifs implement operations faster when you hand them off in a batch.
3049 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3050 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3051 * more packets, plus possibly installing the flow in the dpif.
3053 * So far we only batch the operations that affect flow setup time the most.
3054 * It's possible to batch more than that, but the benefit might be minimal. */
3056 struct hmap_node hmap_node;
3057 struct ofproto_dpif *ofproto;
3059 enum odp_key_fitness key_fitness;
3060 const struct nlattr *key;
3062 ovs_be16 initial_tci;
3063 struct list packets;
3064 enum dpif_upcall_type upcall_type;
3067 struct flow_miss_op {
3068 struct dpif_op dpif_op;
3069 struct subfacet *subfacet; /* Subfacet */
3070 void *garbage; /* Pointer to pass to free(), NULL if none. */
3071 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3074 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3075 * OpenFlow controller as necessary according to their individual
3076 * configurations. */
3078 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3079 const struct flow *flow)
3081 struct ofputil_packet_in pin;
3083 pin.packet = packet->data;
3084 pin.packet_len = packet->size;
3085 pin.reason = OFPR_NO_MATCH;
3086 pin.controller_id = 0;
3091 pin.send_len = 0; /* not used for flow table misses */
3093 flow_get_metadata(flow, &pin.fmd);
3095 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3098 static enum slow_path_reason
3099 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3100 const struct ofpbuf *packet)
3102 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3108 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3110 cfm_process_heartbeat(ofport->cfm, packet);
3113 } else if (ofport->bundle && ofport->bundle->lacp
3114 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3116 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3119 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3121 stp_process_packet(ofport, packet);
3128 static struct flow_miss *
3129 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3131 struct flow_miss *miss;
3133 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3134 if (flow_equal(&miss->flow, flow)) {
3142 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3143 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3144 * 'miss' is associated with a subfacet the caller must also initialize the
3145 * returned op->subfacet, and if anything needs to be freed after processing
3146 * the op, the caller must initialize op->garbage also. */
3148 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3149 struct flow_miss_op *op)
3151 if (miss->flow.vlan_tci != miss->initial_tci) {
3152 /* This packet was received on a VLAN splinter port. We
3153 * added a VLAN to the packet to make the packet resemble
3154 * the flow, but the actions were composed assuming that
3155 * the packet contained no VLAN. So, we must remove the
3156 * VLAN header from the packet before trying to execute the
3158 eth_pop_vlan(packet);
3161 op->subfacet = NULL;
3163 op->dpif_op.type = DPIF_OP_EXECUTE;
3164 op->dpif_op.u.execute.key = miss->key;
3165 op->dpif_op.u.execute.key_len = miss->key_len;
3166 op->dpif_op.u.execute.packet = packet;
3169 /* Helper for handle_flow_miss_without_facet() and
3170 * handle_flow_miss_with_facet(). */
3172 handle_flow_miss_common(struct rule_dpif *rule,
3173 struct ofpbuf *packet, const struct flow *flow)
3175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3177 ofproto->n_matches++;
3179 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3181 * Extra-special case for fail-open mode.
3183 * We are in fail-open mode and the packet matched the fail-open
3184 * rule, but we are connected to a controller too. We should send
3185 * the packet up to the controller in the hope that it will try to
3186 * set up a flow and thereby allow us to exit fail-open.
3188 * See the top-level comment in fail-open.c for more information.
3190 send_packet_in_miss(ofproto, packet, flow);
3194 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3195 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3196 * installing a datapath flow. The answer is usually "yes" (a return value of
3197 * true). However, for short flows the cost of bookkeeping is much higher than
3198 * the benefits, so when the datapath holds a large number of flows we impose
3199 * some heuristics to decide which flows are likely to be worth tracking. */
3201 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3202 struct flow_miss *miss, uint32_t hash)
3204 if (!ofproto->governor) {
3207 n_subfacets = hmap_count(&ofproto->subfacets);
3208 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3212 ofproto->governor = governor_create(ofproto->up.name);
3215 return governor_should_install_flow(ofproto->governor, hash,
3216 list_size(&miss->packets));
3219 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3220 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3221 * increment '*n_ops'. */
3223 handle_flow_miss_without_facet(struct flow_miss *miss,
3224 struct rule_dpif *rule,
3225 struct flow_miss_op *ops, size_t *n_ops)
3227 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3228 long long int now = time_msec();
3229 struct action_xlate_ctx ctx;
3230 struct ofpbuf *packet;
3232 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3233 struct flow_miss_op *op = &ops[*n_ops];
3234 struct dpif_flow_stats stats;
3235 struct ofpbuf odp_actions;
3237 COVERAGE_INC(facet_suppress);
3239 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3241 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3242 rule_credit_stats(rule, &stats);
3244 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3246 ctx.resubmit_stats = &stats;
3247 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3250 if (odp_actions.size) {
3251 struct dpif_execute *execute = &op->dpif_op.u.execute;
3253 init_flow_miss_execute_op(miss, packet, op);
3254 execute->actions = odp_actions.data;
3255 execute->actions_len = odp_actions.size;
3256 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3260 ofpbuf_uninit(&odp_actions);
3265 /* Handles 'miss', which matches 'facet'. May add any required datapath
3266 * operations to 'ops', incrementing '*n_ops' for each new op.
3268 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3269 * This is really important only for new facets: if we just called time_msec()
3270 * here, then the new subfacet or its packets could look (occasionally) as
3271 * though it was used some time after the facet was used. That can make a
3272 * one-packet flow look like it has a nonzero duration, which looks odd in
3273 * e.g. NetFlow statistics. */
3275 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3277 struct flow_miss_op *ops, size_t *n_ops)
3279 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3280 enum subfacet_path want_path;
3281 struct subfacet *subfacet;
3282 struct ofpbuf *packet;
3284 subfacet = subfacet_create(facet,
3285 miss->key_fitness, miss->key, miss->key_len,
3286 miss->initial_tci, now);
3288 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3289 struct flow_miss_op *op = &ops[*n_ops];
3290 struct dpif_flow_stats stats;
3291 struct ofpbuf odp_actions;
3293 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3295 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3296 if (!subfacet->actions || subfacet->slow) {
3297 subfacet_make_actions(subfacet, packet, &odp_actions);
3300 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3301 subfacet_update_stats(subfacet, &stats);
3303 if (subfacet->actions_len) {
3304 struct dpif_execute *execute = &op->dpif_op.u.execute;
3306 init_flow_miss_execute_op(miss, packet, op);
3307 op->subfacet = subfacet;
3308 if (!subfacet->slow) {
3309 execute->actions = subfacet->actions;
3310 execute->actions_len = subfacet->actions_len;
3311 ofpbuf_uninit(&odp_actions);
3313 execute->actions = odp_actions.data;
3314 execute->actions_len = odp_actions.size;
3315 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3320 ofpbuf_uninit(&odp_actions);
3324 want_path = subfacet_want_path(subfacet->slow);
3325 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3326 struct flow_miss_op *op = &ops[(*n_ops)++];
3327 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3329 op->subfacet = subfacet;
3331 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3332 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3333 put->key = miss->key;
3334 put->key_len = miss->key_len;
3335 if (want_path == SF_FAST_PATH) {
3336 put->actions = subfacet->actions;
3337 put->actions_len = subfacet->actions_len;
3339 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3340 op->stub, sizeof op->stub,
3341 &put->actions, &put->actions_len);
3347 /* Handles flow miss 'miss'. May add any required datapath operations
3348 * to 'ops', incrementing '*n_ops' for each new op. */
3350 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3353 struct ofproto_dpif *ofproto = miss->ofproto;
3354 struct facet *facet;
3358 /* The caller must ensure that miss->hmap_node.hash contains
3359 * flow_hash(miss->flow, 0). */
3360 hash = miss->hmap_node.hash;
3362 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3364 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3366 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3367 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3371 facet = facet_create(rule, &miss->flow, hash);
3376 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3379 /* This function does post-processing on data returned from
3380 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3381 * rest of the upcall processing logic. In particular, if the extracted
3382 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3383 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3384 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3385 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3386 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3387 * member converted to the OpenFlow number.
3389 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3390 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3391 * (This differs from the value returned in flow->vlan_tci only for packets
3392 * received on VLAN splinters.) */
3393 static enum odp_key_fitness
3394 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3395 enum odp_key_fitness fitness,
3396 struct flow *flow, ovs_be16 *initial_tci,
3397 struct ofpbuf *packet)
3399 if (fitness == ODP_FIT_ERROR) {
3402 *initial_tci = flow->vlan_tci;
3404 if (vsp_adjust_flow(ofproto, flow)) {
3406 /* Make the packet resemble the flow, so that it gets sent to an
3407 * OpenFlow controller properly, so that it looks correct for
3408 * sFlow, and so that flow_extract() will get the correct vlan_tci
3409 * if it is called on 'packet'.
3411 * The allocated space inside 'packet' probably also contains
3412 * 'key', that is, both 'packet' and 'key' are probably part of a
3413 * struct dpif_upcall (see the large comment on that structure
3414 * definition), so pushing data on 'packet' is in general not a
3415 * good idea since it could overwrite 'key' or free it as a side
3416 * effect. However, it's OK in this special case because we know
3417 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3418 * will just overwrite the 4-byte "struct nlattr", which is fine
3419 * since we don't need that header anymore. */
3420 eth_push_vlan(packet, flow->vlan_tci);
3423 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3424 if (fitness == ODP_FIT_PERFECT) {
3425 fitness = ODP_FIT_TOO_MUCH;
3433 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3436 struct dpif_upcall *upcall;
3437 struct flow_miss *miss;
3438 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3439 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3440 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3450 /* Construct the to-do list.
3452 * This just amounts to extracting the flow from each packet and sticking
3453 * the packets that have the same flow in the same "flow_miss" structure so
3454 * that we can process them together. */
3457 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3458 struct flow_miss *miss = &misses[n_misses];
3459 struct flow_miss *existing_miss;
3460 enum odp_key_fitness fitness;
3461 struct ofproto_dpif *ofproto;
3462 struct ofport_dpif *port;
3466 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3467 port = odp_port_to_ofport(backer, flow.in_port);
3469 /* Received packet on port for which we couldn't associate
3470 * an ofproto. This can happen if a port is removed while
3471 * traffic is being received. Print a rate-limited message
3472 * in case it happens frequently. */
3473 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3477 ofproto = ofproto_dpif_cast(port->up.ofproto);
3478 flow.in_port = port->up.ofp_port;
3480 /* Obtain metadata and check userspace/kernel agreement on flow match,
3481 * then set 'flow''s header pointers. */
3482 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3483 &flow, &miss->initial_tci, upcall->packet);
3484 if (miss->key_fitness == ODP_FIT_ERROR) {
3487 flow_extract(upcall->packet, flow.skb_priority,
3488 &flow.tunnel, flow.in_port, &miss->flow);
3490 /* Add other packets to a to-do list. */
3491 hash = flow_hash(&miss->flow, 0);
3492 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3493 if (!existing_miss) {
3494 hmap_insert(&todo, &miss->hmap_node, hash);
3495 miss->ofproto = ofproto;
3496 miss->key = upcall->key;
3497 miss->key_len = upcall->key_len;
3498 miss->upcall_type = upcall->type;
3499 list_init(&miss->packets);
3503 miss = existing_miss;
3505 list_push_back(&miss->packets, &upcall->packet->list_node);
3508 /* Process each element in the to-do list, constructing the set of
3509 * operations to batch. */
3511 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3512 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3514 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3516 /* Execute batch. */
3517 for (i = 0; i < n_ops; i++) {
3518 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3520 dpif_operate(backer->dpif, dpif_ops, n_ops);
3522 /* Free memory and update facets. */
3523 for (i = 0; i < n_ops; i++) {
3524 struct flow_miss_op *op = &flow_miss_ops[i];
3526 switch (op->dpif_op.type) {
3527 case DPIF_OP_EXECUTE:
3530 case DPIF_OP_FLOW_PUT:
3531 if (!op->dpif_op.error) {
3532 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3536 case DPIF_OP_FLOW_DEL:
3542 hmap_destroy(&todo);
3545 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3546 classify_upcall(const struct dpif_upcall *upcall)
3548 union user_action_cookie cookie;
3550 /* First look at the upcall type. */
3551 switch (upcall->type) {
3552 case DPIF_UC_ACTION:
3558 case DPIF_N_UC_TYPES:
3560 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3564 /* "action" upcalls need a closer look. */
3565 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3566 switch (cookie.type) {
3567 case USER_ACTION_COOKIE_SFLOW:
3568 return SFLOW_UPCALL;
3570 case USER_ACTION_COOKIE_SLOW_PATH:
3573 case USER_ACTION_COOKIE_UNSPEC:
3575 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3581 handle_sflow_upcall(struct dpif_backer *backer,
3582 const struct dpif_upcall *upcall)
3584 struct ofproto_dpif *ofproto;
3585 union user_action_cookie cookie;
3586 enum odp_key_fitness fitness;
3587 struct ofport_dpif *port;
3588 ovs_be16 initial_tci;
3590 uint32_t odp_in_port;
3592 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3594 port = odp_port_to_ofport(backer, flow.in_port);
3599 ofproto = ofproto_dpif_cast(port->up.ofproto);
3600 if (!ofproto->sflow) {
3604 odp_in_port = flow.in_port;
3605 flow.in_port = port->up.ofp_port;
3606 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3607 &initial_tci, upcall->packet);
3608 if (fitness == ODP_FIT_ERROR) {
3612 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3613 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3614 odp_in_port, &cookie);
3618 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3620 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3621 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3622 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3627 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3630 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3631 struct dpif_upcall *upcall = &misses[n_misses];
3632 struct ofpbuf *buf = &miss_bufs[n_misses];
3635 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3636 sizeof miss_buf_stubs[n_misses]);
3637 error = dpif_recv(backer->dpif, upcall, buf);
3643 switch (classify_upcall(upcall)) {
3645 /* Handle it later. */
3650 handle_sflow_upcall(backer, upcall);
3660 /* Handle deferred MISS_UPCALL processing. */
3661 handle_miss_upcalls(backer, misses, n_misses);
3662 for (i = 0; i < n_misses; i++) {
3663 ofpbuf_uninit(&miss_bufs[i]);
3669 /* Flow expiration. */
3671 static int subfacet_max_idle(const struct ofproto_dpif *);
3672 static void update_stats(struct dpif_backer *);
3673 static void rule_expire(struct rule_dpif *);
3674 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3676 /* This function is called periodically by run(). Its job is to collect
3677 * updates for the flows that have been installed into the datapath, most
3678 * importantly when they last were used, and then use that information to
3679 * expire flows that have not been used recently.
3681 * Returns the number of milliseconds after which it should be called again. */
3683 expire(struct dpif_backer *backer)
3685 struct ofproto_dpif *ofproto;
3686 int max_idle = INT32_MAX;
3688 /* Update stats for each flow in the backer. */
3689 update_stats(backer);
3691 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3692 struct rule_dpif *rule, *next_rule;
3693 struct oftable *table;
3696 if (ofproto->backer != backer) {
3700 /* Expire subfacets that have been idle too long. */
3701 dp_max_idle = subfacet_max_idle(ofproto);
3702 expire_subfacets(ofproto, dp_max_idle);
3704 max_idle = MIN(max_idle, dp_max_idle);
3706 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3708 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3709 struct cls_cursor cursor;
3711 cls_cursor_init(&cursor, &table->cls, NULL);
3712 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3717 /* All outstanding data in existing flows has been accounted, so it's a
3718 * good time to do bond rebalancing. */
3719 if (ofproto->has_bonded_bundles) {
3720 struct ofbundle *bundle;
3722 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3724 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3730 return MIN(max_idle, 1000);
3733 /* Updates flow table statistics given that the datapath just reported 'stats'
3734 * as 'subfacet''s statistics. */
3736 update_subfacet_stats(struct subfacet *subfacet,
3737 const struct dpif_flow_stats *stats)
3739 struct facet *facet = subfacet->facet;
3741 if (stats->n_packets >= subfacet->dp_packet_count) {
3742 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3743 facet->packet_count += extra;
3745 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3748 if (stats->n_bytes >= subfacet->dp_byte_count) {
3749 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3751 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3754 subfacet->dp_packet_count = stats->n_packets;
3755 subfacet->dp_byte_count = stats->n_bytes;
3757 facet->tcp_flags |= stats->tcp_flags;
3759 subfacet_update_time(subfacet, stats->used);
3760 if (facet->accounted_bytes < facet->byte_count) {
3762 facet_account(facet);
3763 facet->accounted_bytes = facet->byte_count;
3765 facet_push_stats(facet);
3768 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3769 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3771 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3772 const struct nlattr *key, size_t key_len)
3774 if (!VLOG_DROP_WARN(&rl)) {
3778 odp_flow_key_format(key, key_len, &s);
3779 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3783 COVERAGE_INC(facet_unexpected);
3784 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3787 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3789 * This function also pushes statistics updates to rules which each facet
3790 * resubmits into. Generally these statistics will be accurate. However, if a
3791 * facet changes the rule it resubmits into at some time in between
3792 * update_stats() runs, it is possible that statistics accrued to the
3793 * old rule will be incorrectly attributed to the new rule. This could be
3794 * avoided by calling update_stats() whenever rules are created or
3795 * deleted. However, the performance impact of making so many calls to the
3796 * datapath do not justify the benefit of having perfectly accurate statistics.
3799 update_stats(struct dpif_backer *backer)
3801 const struct dpif_flow_stats *stats;
3802 struct dpif_flow_dump dump;
3803 const struct nlattr *key;
3806 dpif_flow_dump_start(&dump, backer->dpif);
3807 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3809 struct subfacet *subfacet;
3810 enum odp_key_fitness fitness;
3811 struct ofproto_dpif *ofproto;
3812 struct ofport_dpif *port;
3815 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3816 if (fitness == ODP_FIT_ERROR) {
3820 port = odp_port_to_ofport(backer, flow.in_port);
3822 /* This flow is for a port for which we couldn't associate an
3823 * ofproto. This can happen if a port is removed while
3824 * traffic is being received. Print a rate-limited message
3825 * in case it happens frequently. */
3827 "stats update for flow with unassociated port %"PRIu32,
3832 ofproto = ofproto_dpif_cast(port->up.ofproto);
3833 flow.in_port = port->up.ofp_port;
3834 key_hash = odp_flow_key_hash(key, key_len);
3836 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3837 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3839 update_subfacet_stats(subfacet, stats);
3843 /* Stats are updated per-packet. */
3846 case SF_NOT_INSTALLED:
3848 delete_unexpected_flow(ofproto, key, key_len);
3852 dpif_flow_dump_done(&dump);
3855 /* Calculates and returns the number of milliseconds of idle time after which
3856 * subfacets should expire from the datapath. When a subfacet expires, we fold
3857 * its statistics into its facet, and when a facet's last subfacet expires, we
3858 * fold its statistic into its rule. */
3860 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3863 * Idle time histogram.
3865 * Most of the time a switch has a relatively small number of subfacets.
3866 * When this is the case we might as well keep statistics for all of them
3867 * in userspace and to cache them in the kernel datapath for performance as
3870 * As the number of subfacets increases, the memory required to maintain
3871 * statistics about them in userspace and in the kernel becomes
3872 * significant. However, with a large number of subfacets it is likely
3873 * that only a few of them are "heavy hitters" that consume a large amount
3874 * of bandwidth. At this point, only heavy hitters are worth caching in
3875 * the kernel and maintaining in userspaces; other subfacets we can
3878 * The technique used to compute the idle time is to build a histogram with
3879 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3880 * that is installed in the kernel gets dropped in the appropriate bucket.
3881 * After the histogram has been built, we compute the cutoff so that only
3882 * the most-recently-used 1% of subfacets (but at least
3883 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3884 * the most-recently-used bucket of subfacets is kept, so actually an
3885 * arbitrary number of subfacets can be kept in any given expiration run
3886 * (though the next run will delete most of those unless they receive
3889 * This requires a second pass through the subfacets, in addition to the
3890 * pass made by update_stats(), because the former function never looks at
3891 * uninstallable subfacets.
3893 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3894 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3895 int buckets[N_BUCKETS] = { 0 };
3896 int total, subtotal, bucket;
3897 struct subfacet *subfacet;
3901 total = hmap_count(&ofproto->subfacets);
3902 if (total <= ofproto->up.flow_eviction_threshold) {
3903 return N_BUCKETS * BUCKET_WIDTH;
3906 /* Build histogram. */
3908 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3909 long long int idle = now - subfacet->used;
3910 int bucket = (idle <= 0 ? 0
3911 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3912 : (unsigned int) idle / BUCKET_WIDTH);
3916 /* Find the first bucket whose flows should be expired. */
3917 subtotal = bucket = 0;
3919 subtotal += buckets[bucket++];
3920 } while (bucket < N_BUCKETS &&
3921 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3923 if (VLOG_IS_DBG_ENABLED()) {
3927 ds_put_cstr(&s, "keep");
3928 for (i = 0; i < N_BUCKETS; i++) {
3930 ds_put_cstr(&s, ", drop");
3933 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3936 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3940 return bucket * BUCKET_WIDTH;
3944 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3946 /* Cutoff time for most flows. */
3947 long long int normal_cutoff = time_msec() - dp_max_idle;
3949 /* We really want to keep flows for special protocols around, so use a more
3950 * conservative cutoff. */
3951 long long int special_cutoff = time_msec() - 10000;
3953 struct subfacet *subfacet, *next_subfacet;
3954 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3958 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3959 &ofproto->subfacets) {
3960 long long int cutoff;
3962 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3965 if (subfacet->used < cutoff) {
3966 if (subfacet->path != SF_NOT_INSTALLED) {
3967 batch[n_batch++] = subfacet;
3968 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3969 subfacet_destroy_batch(ofproto, batch, n_batch);
3973 subfacet_destroy(subfacet);
3979 subfacet_destroy_batch(ofproto, batch, n_batch);
3983 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3984 * then delete it entirely. */
3986 rule_expire(struct rule_dpif *rule)
3988 struct facet *facet, *next_facet;
3992 if (rule->up.pending) {
3993 /* We'll have to expire it later. */
3997 /* Has 'rule' expired? */
3999 if (rule->up.hard_timeout
4000 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4001 reason = OFPRR_HARD_TIMEOUT;
4002 } else if (rule->up.idle_timeout
4003 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4004 reason = OFPRR_IDLE_TIMEOUT;
4009 COVERAGE_INC(ofproto_dpif_expired);
4011 /* Update stats. (This is a no-op if the rule expired due to an idle
4012 * timeout, because that only happens when the rule has no facets left.) */
4013 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4014 facet_remove(facet);
4017 /* Get rid of the rule. */
4018 ofproto_rule_expire(&rule->up, reason);
4023 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4025 * The caller must already have determined that no facet with an identical
4026 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4027 * the ofproto's classifier table.
4029 * 'hash' must be the return value of flow_hash(flow, 0).
4031 * The facet will initially have no subfacets. The caller should create (at
4032 * least) one subfacet with subfacet_create(). */
4033 static struct facet *
4034 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4037 struct facet *facet;
4039 facet = xzalloc(sizeof *facet);
4040 facet->used = time_msec();
4041 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4042 list_push_back(&rule->facets, &facet->list_node);
4044 facet->flow = *flow;
4045 list_init(&facet->subfacets);
4046 netflow_flow_init(&facet->nf_flow);
4047 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4053 facet_free(struct facet *facet)
4058 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4059 * 'packet', which arrived on 'in_port'.
4061 * Takes ownership of 'packet'. */
4063 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4064 const struct nlattr *odp_actions, size_t actions_len,
4065 struct ofpbuf *packet)
4067 struct odputil_keybuf keybuf;
4071 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4072 odp_flow_key_from_flow(&key, flow,
4073 ofp_port_to_odp_port(ofproto, flow->in_port));
4075 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4076 odp_actions, actions_len, packet);
4078 ofpbuf_delete(packet);
4082 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4084 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4085 * rule's statistics, via subfacet_uninstall().
4087 * - Removes 'facet' from its rule and from ofproto->facets.
4090 facet_remove(struct facet *facet)
4092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4093 struct subfacet *subfacet, *next_subfacet;
4095 assert(!list_is_empty(&facet->subfacets));
4097 /* First uninstall all of the subfacets to get final statistics. */
4098 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4099 subfacet_uninstall(subfacet);
4102 /* Flush the final stats to the rule.
4104 * This might require us to have at least one subfacet around so that we
4105 * can use its actions for accounting in facet_account(), which is why we
4106 * have uninstalled but not yet destroyed the subfacets. */
4107 facet_flush_stats(facet);
4109 /* Now we're really all done so destroy everything. */
4110 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4111 &facet->subfacets) {
4112 subfacet_destroy__(subfacet);
4114 hmap_remove(&ofproto->facets, &facet->hmap_node);
4115 list_remove(&facet->list_node);
4119 /* Feed information from 'facet' back into the learning table to keep it in
4120 * sync with what is actually flowing through the datapath. */
4122 facet_learn(struct facet *facet)
4124 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4125 struct action_xlate_ctx ctx;
4127 if (!facet->has_learn
4128 && !facet->has_normal
4129 && (!facet->has_fin_timeout
4130 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4134 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4135 facet->flow.vlan_tci,
4136 facet->rule, facet->tcp_flags, NULL);
4137 ctx.may_learn = true;
4138 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4139 facet->rule->up.ofpacts_len);
4143 facet_account(struct facet *facet)
4145 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4146 struct subfacet *subfacet;
4147 const struct nlattr *a;
4152 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4155 n_bytes = facet->byte_count - facet->accounted_bytes;
4157 /* This loop feeds byte counters to bond_account() for rebalancing to use
4158 * as a basis. We also need to track the actual VLAN on which the packet
4159 * is going to be sent to ensure that it matches the one passed to
4160 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4163 * We use the actions from an arbitrary subfacet because they should all
4164 * be equally valid for our purpose. */
4165 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4166 struct subfacet, list_node);
4167 vlan_tci = facet->flow.vlan_tci;
4168 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4169 subfacet->actions, subfacet->actions_len) {
4170 const struct ovs_action_push_vlan *vlan;
4171 struct ofport_dpif *port;
4173 switch (nl_attr_type(a)) {
4174 case OVS_ACTION_ATTR_OUTPUT:
4175 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4176 if (port && port->bundle && port->bundle->bond) {
4177 bond_account(port->bundle->bond, &facet->flow,
4178 vlan_tci_to_vid(vlan_tci), n_bytes);
4182 case OVS_ACTION_ATTR_POP_VLAN:
4183 vlan_tci = htons(0);
4186 case OVS_ACTION_ATTR_PUSH_VLAN:
4187 vlan = nl_attr_get(a);
4188 vlan_tci = vlan->vlan_tci;
4194 /* Returns true if the only action for 'facet' is to send to the controller.
4195 * (We don't report NetFlow expiration messages for such facets because they
4196 * are just part of the control logic for the network, not real traffic). */
4198 facet_is_controller_flow(struct facet *facet)
4201 const struct rule *rule = &facet->rule->up;
4202 const struct ofpact *ofpacts = rule->ofpacts;
4203 size_t ofpacts_len = rule->ofpacts_len;
4205 if (ofpacts_len > 0 &&
4206 ofpacts->type == OFPACT_CONTROLLER &&
4207 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4214 /* Folds all of 'facet''s statistics into its rule. Also updates the
4215 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4216 * 'facet''s statistics in the datapath should have been zeroed and folded into
4217 * its packet and byte counts before this function is called. */
4219 facet_flush_stats(struct facet *facet)
4221 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4222 struct subfacet *subfacet;
4224 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4225 assert(!subfacet->dp_byte_count);
4226 assert(!subfacet->dp_packet_count);
4229 facet_push_stats(facet);
4230 if (facet->accounted_bytes < facet->byte_count) {
4231 facet_account(facet);
4232 facet->accounted_bytes = facet->byte_count;
4235 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4236 struct ofexpired expired;
4237 expired.flow = facet->flow;
4238 expired.packet_count = facet->packet_count;
4239 expired.byte_count = facet->byte_count;
4240 expired.used = facet->used;
4241 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4244 facet->rule->packet_count += facet->packet_count;
4245 facet->rule->byte_count += facet->byte_count;
4247 /* Reset counters to prevent double counting if 'facet' ever gets
4249 facet_reset_counters(facet);
4251 netflow_flow_clear(&facet->nf_flow);
4252 facet->tcp_flags = 0;
4255 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4256 * Returns it if found, otherwise a null pointer.
4258 * 'hash' must be the return value of flow_hash(flow, 0).
4260 * The returned facet might need revalidation; use facet_lookup_valid()
4261 * instead if that is important. */
4262 static struct facet *
4263 facet_find(struct ofproto_dpif *ofproto,
4264 const struct flow *flow, uint32_t hash)
4266 struct facet *facet;
4268 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4269 if (flow_equal(flow, &facet->flow)) {
4277 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4278 * Returns it if found, otherwise a null pointer.
4280 * 'hash' must be the return value of flow_hash(flow, 0).
4282 * The returned facet is guaranteed to be valid. */
4283 static struct facet *
4284 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4287 struct facet *facet;
4289 facet = facet_find(ofproto, flow, hash);
4291 && (ofproto->need_revalidate
4292 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4293 facet_revalidate(facet);
4300 subfacet_path_to_string(enum subfacet_path path)
4303 case SF_NOT_INSTALLED:
4304 return "not installed";
4306 return "in fast path";
4308 return "in slow path";
4314 /* Returns the path in which a subfacet should be installed if its 'slow'
4315 * member has the specified value. */
4316 static enum subfacet_path
4317 subfacet_want_path(enum slow_path_reason slow)
4319 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4322 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4323 * supposing that its actions have been recalculated as 'want_actions' and that
4324 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4326 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4327 const struct ofpbuf *want_actions)
4329 enum subfacet_path want_path = subfacet_want_path(slow);
4330 return (want_path != subfacet->path
4331 || (want_path == SF_FAST_PATH
4332 && (subfacet->actions_len != want_actions->size
4333 || memcmp(subfacet->actions, want_actions->data,
4334 subfacet->actions_len))));
4338 facet_check_consistency(struct facet *facet)
4340 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4344 uint64_t odp_actions_stub[1024 / 8];
4345 struct ofpbuf odp_actions;
4347 struct rule_dpif *rule;
4348 struct subfacet *subfacet;
4349 bool may_log = false;
4352 /* Check the rule for consistency. */
4353 rule = rule_dpif_lookup(ofproto, &facet->flow);
4354 ok = rule == facet->rule;
4356 may_log = !VLOG_DROP_WARN(&rl);
4361 flow_format(&s, &facet->flow);
4362 ds_put_format(&s, ": facet associated with wrong rule (was "
4363 "table=%"PRIu8",", facet->rule->up.table_id);
4364 cls_rule_format(&facet->rule->up.cr, &s);
4365 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4367 cls_rule_format(&rule->up.cr, &s);
4368 ds_put_char(&s, ')');
4370 VLOG_WARN("%s", ds_cstr(&s));
4375 /* Check the datapath actions for consistency. */
4376 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4377 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4378 enum subfacet_path want_path;
4379 struct odputil_keybuf keybuf;
4380 struct action_xlate_ctx ctx;
4384 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4385 subfacet->initial_tci, rule, 0, NULL);
4386 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4389 if (subfacet->path == SF_NOT_INSTALLED) {
4390 /* This only happens if the datapath reported an error when we
4391 * tried to install the flow. Don't flag another error here. */
4395 want_path = subfacet_want_path(subfacet->slow);
4396 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4397 /* The actions for slow-path flows may legitimately vary from one
4398 * packet to the next. We're done. */
4402 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4406 /* Inconsistency! */
4408 may_log = !VLOG_DROP_WARN(&rl);
4412 /* Rate-limited, skip reporting. */
4417 subfacet_get_key(subfacet, &keybuf, &key);
4418 odp_flow_key_format(key.data, key.size, &s);
4420 ds_put_cstr(&s, ": inconsistency in subfacet");
4421 if (want_path != subfacet->path) {
4422 enum odp_key_fitness fitness = subfacet->key_fitness;
4424 ds_put_format(&s, " (%s, fitness=%s)",
4425 subfacet_path_to_string(subfacet->path),
4426 odp_key_fitness_to_string(fitness));
4427 ds_put_format(&s, " (should have been %s)",
4428 subfacet_path_to_string(want_path));
4429 } else if (want_path == SF_FAST_PATH) {
4430 ds_put_cstr(&s, " (actions were: ");
4431 format_odp_actions(&s, subfacet->actions,
4432 subfacet->actions_len);
4433 ds_put_cstr(&s, ") (correct actions: ");
4434 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4435 ds_put_char(&s, ')');
4437 ds_put_cstr(&s, " (actions: ");
4438 format_odp_actions(&s, subfacet->actions,
4439 subfacet->actions_len);
4440 ds_put_char(&s, ')');
4442 VLOG_WARN("%s", ds_cstr(&s));
4445 ofpbuf_uninit(&odp_actions);
4450 /* Re-searches the classifier for 'facet':
4452 * - If the rule found is different from 'facet''s current rule, moves
4453 * 'facet' to the new rule and recompiles its actions.
4455 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4456 * where it is and recompiles its actions anyway. */
4458 facet_revalidate(struct facet *facet)
4460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4462 struct nlattr *odp_actions;
4465 struct actions *new_actions;
4467 struct action_xlate_ctx ctx;
4468 uint64_t odp_actions_stub[1024 / 8];
4469 struct ofpbuf odp_actions;
4471 struct rule_dpif *new_rule;
4472 struct subfacet *subfacet;
4475 COVERAGE_INC(facet_revalidate);
4477 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4479 /* Calculate new datapath actions.
4481 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4482 * emit a NetFlow expiration and, if so, we need to have the old state
4483 * around to properly compose it. */
4485 /* If the datapath actions changed or the installability changed,
4486 * then we need to talk to the datapath. */
4489 memset(&ctx, 0, sizeof ctx);
4490 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4491 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4492 enum slow_path_reason slow;
4494 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4495 subfacet->initial_tci, new_rule, 0, NULL);
4496 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4499 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4500 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4501 struct dpif_flow_stats stats;
4503 subfacet_install(subfacet,
4504 odp_actions.data, odp_actions.size, &stats, slow);
4505 subfacet_update_stats(subfacet, &stats);
4508 new_actions = xcalloc(list_size(&facet->subfacets),
4509 sizeof *new_actions);
4511 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4513 new_actions[i].actions_len = odp_actions.size;
4518 ofpbuf_uninit(&odp_actions);
4521 facet_flush_stats(facet);
4524 /* Update 'facet' now that we've taken care of all the old state. */
4525 facet->tags = ctx.tags;
4526 facet->nf_flow.output_iface = ctx.nf_output_iface;
4527 facet->has_learn = ctx.has_learn;
4528 facet->has_normal = ctx.has_normal;
4529 facet->has_fin_timeout = ctx.has_fin_timeout;
4530 facet->mirrors = ctx.mirrors;
4533 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4534 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4536 if (new_actions && new_actions[i].odp_actions) {
4537 free(subfacet->actions);
4538 subfacet->actions = new_actions[i].odp_actions;
4539 subfacet->actions_len = new_actions[i].actions_len;
4545 if (facet->rule != new_rule) {
4546 COVERAGE_INC(facet_changed_rule);
4547 list_remove(&facet->list_node);
4548 list_push_back(&new_rule->facets, &facet->list_node);
4549 facet->rule = new_rule;
4550 facet->used = new_rule->up.created;
4551 facet->prev_used = facet->used;
4555 /* Updates 'facet''s used time. Caller is responsible for calling
4556 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4558 facet_update_time(struct facet *facet, long long int used)
4560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4561 if (used > facet->used) {
4563 ofproto_rule_update_used(&facet->rule->up, used);
4564 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4569 facet_reset_counters(struct facet *facet)
4571 facet->packet_count = 0;
4572 facet->byte_count = 0;
4573 facet->prev_packet_count = 0;
4574 facet->prev_byte_count = 0;
4575 facet->accounted_bytes = 0;
4579 facet_push_stats(struct facet *facet)
4581 struct dpif_flow_stats stats;
4583 assert(facet->packet_count >= facet->prev_packet_count);
4584 assert(facet->byte_count >= facet->prev_byte_count);
4585 assert(facet->used >= facet->prev_used);
4587 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4588 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4589 stats.used = facet->used;
4590 stats.tcp_flags = 0;
4592 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4593 facet->prev_packet_count = facet->packet_count;
4594 facet->prev_byte_count = facet->byte_count;
4595 facet->prev_used = facet->used;
4597 flow_push_stats(facet->rule, &facet->flow, &stats);
4599 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4600 facet->mirrors, stats.n_packets, stats.n_bytes);
4605 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4607 rule->packet_count += stats->n_packets;
4608 rule->byte_count += stats->n_bytes;
4609 ofproto_rule_update_used(&rule->up, stats->used);
4612 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4613 * 'rule''s actions and mirrors. */
4615 flow_push_stats(struct rule_dpif *rule,
4616 const struct flow *flow, const struct dpif_flow_stats *stats)
4618 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4619 struct action_xlate_ctx ctx;
4621 ofproto_rule_update_used(&rule->up, stats->used);
4623 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4625 ctx.resubmit_stats = stats;
4626 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4627 rule->up.ofpacts_len);
4632 static struct subfacet *
4633 subfacet_find(struct ofproto_dpif *ofproto,
4634 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4635 const struct flow *flow)
4637 struct subfacet *subfacet;
4639 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4640 &ofproto->subfacets) {
4642 ? (subfacet->key_len == key_len
4643 && !memcmp(key, subfacet->key, key_len))
4644 : flow_equal(flow, &subfacet->facet->flow)) {
4652 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4653 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4654 * there is one, otherwise creates and returns a new subfacet.
4656 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4657 * which case the caller must populate the actions with
4658 * subfacet_make_actions(). */
4659 static struct subfacet *
4660 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4661 const struct nlattr *key, size_t key_len,
4662 ovs_be16 initial_tci, long long int now)
4664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4665 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4666 struct subfacet *subfacet;
4668 if (list_is_empty(&facet->subfacets)) {
4669 subfacet = &facet->one_subfacet;
4671 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4674 if (subfacet->facet == facet) {
4678 /* This shouldn't happen. */
4679 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4680 subfacet_destroy(subfacet);
4683 subfacet = xmalloc(sizeof *subfacet);
4686 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4687 list_push_back(&facet->subfacets, &subfacet->list_node);
4688 subfacet->facet = facet;
4689 subfacet->key_fitness = key_fitness;
4690 if (key_fitness != ODP_FIT_PERFECT) {
4691 subfacet->key = xmemdup(key, key_len);
4692 subfacet->key_len = key_len;
4694 subfacet->key = NULL;
4695 subfacet->key_len = 0;
4697 subfacet->used = now;
4698 subfacet->dp_packet_count = 0;
4699 subfacet->dp_byte_count = 0;
4700 subfacet->actions_len = 0;
4701 subfacet->actions = NULL;
4702 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4705 subfacet->path = SF_NOT_INSTALLED;
4706 subfacet->initial_tci = initial_tci;
4711 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4712 * its facet within 'ofproto', and frees it. */
4714 subfacet_destroy__(struct subfacet *subfacet)
4716 struct facet *facet = subfacet->facet;
4717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4719 subfacet_uninstall(subfacet);
4720 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4721 list_remove(&subfacet->list_node);
4722 free(subfacet->key);
4723 free(subfacet->actions);
4724 if (subfacet != &facet->one_subfacet) {
4729 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4730 * last remaining subfacet in its facet destroys the facet too. */
4732 subfacet_destroy(struct subfacet *subfacet)
4734 struct facet *facet = subfacet->facet;
4736 if (list_is_singleton(&facet->subfacets)) {
4737 /* facet_remove() needs at least one subfacet (it will remove it). */
4738 facet_remove(facet);
4740 subfacet_destroy__(subfacet);
4745 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4746 struct subfacet **subfacets, int n)
4748 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4749 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4750 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4751 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4752 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4755 for (i = 0; i < n; i++) {
4756 ops[i].type = DPIF_OP_FLOW_DEL;
4757 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4758 ops[i].u.flow_del.key = keys[i].data;
4759 ops[i].u.flow_del.key_len = keys[i].size;
4760 ops[i].u.flow_del.stats = &stats[i];
4764 dpif_operate(ofproto->backer->dpif, opsp, n);
4765 for (i = 0; i < n; i++) {
4766 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4767 subfacets[i]->path = SF_NOT_INSTALLED;
4768 subfacet_destroy(subfacets[i]);
4772 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4773 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4774 * for use as temporary storage. */
4776 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4780 if (!subfacet->key) {
4781 struct ofproto_dpif *ofproto;
4782 struct flow *flow = &subfacet->facet->flow;
4784 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4785 ofproto = ofproto_dpif_cast(subfacet->facet->rule->up.ofproto);
4786 odp_flow_key_from_flow(key, flow,
4787 ofp_port_to_odp_port(ofproto, flow->in_port));
4789 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4793 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4794 * Translates the actions into 'odp_actions', which the caller must have
4795 * initialized and is responsible for uninitializing. */
4797 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4798 struct ofpbuf *odp_actions)
4800 struct facet *facet = subfacet->facet;
4801 struct rule_dpif *rule = facet->rule;
4802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4804 struct action_xlate_ctx ctx;
4806 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4808 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4809 facet->tags = ctx.tags;
4810 facet->has_learn = ctx.has_learn;
4811 facet->has_normal = ctx.has_normal;
4812 facet->has_fin_timeout = ctx.has_fin_timeout;
4813 facet->nf_flow.output_iface = ctx.nf_output_iface;
4814 facet->mirrors = ctx.mirrors;
4816 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4817 if (subfacet->actions_len != odp_actions->size
4818 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4819 free(subfacet->actions);
4820 subfacet->actions_len = odp_actions->size;
4821 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4825 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4826 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4827 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4828 * since 'subfacet' was last updated.
4830 * Returns 0 if successful, otherwise a positive errno value. */
4832 subfacet_install(struct subfacet *subfacet,
4833 const struct nlattr *actions, size_t actions_len,
4834 struct dpif_flow_stats *stats,
4835 enum slow_path_reason slow)
4837 struct facet *facet = subfacet->facet;
4838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4839 enum subfacet_path path = subfacet_want_path(slow);
4840 uint64_t slow_path_stub[128 / 8];
4841 struct odputil_keybuf keybuf;
4842 enum dpif_flow_put_flags flags;
4846 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4848 flags |= DPIF_FP_ZERO_STATS;
4851 if (path == SF_SLOW_PATH) {
4852 compose_slow_path(ofproto, &facet->flow, slow,
4853 slow_path_stub, sizeof slow_path_stub,
4854 &actions, &actions_len);
4857 subfacet_get_key(subfacet, &keybuf, &key);
4858 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4859 actions, actions_len, stats);
4862 subfacet_reset_dp_stats(subfacet, stats);
4866 subfacet->path = path;
4872 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4874 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4875 stats, subfacet->slow);
4878 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4880 subfacet_uninstall(struct subfacet *subfacet)
4882 if (subfacet->path != SF_NOT_INSTALLED) {
4883 struct rule_dpif *rule = subfacet->facet->rule;
4884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4885 struct odputil_keybuf keybuf;
4886 struct dpif_flow_stats stats;
4890 subfacet_get_key(subfacet, &keybuf, &key);
4891 error = dpif_flow_del(ofproto->backer->dpif,
4892 key.data, key.size, &stats);
4893 subfacet_reset_dp_stats(subfacet, &stats);
4895 subfacet_update_stats(subfacet, &stats);
4897 subfacet->path = SF_NOT_INSTALLED;
4899 assert(subfacet->dp_packet_count == 0);
4900 assert(subfacet->dp_byte_count == 0);
4904 /* Resets 'subfacet''s datapath statistics counters. This should be called
4905 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4906 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4907 * was reset in the datapath. 'stats' will be modified to include only
4908 * statistics new since 'subfacet' was last updated. */
4910 subfacet_reset_dp_stats(struct subfacet *subfacet,
4911 struct dpif_flow_stats *stats)
4914 && subfacet->dp_packet_count <= stats->n_packets
4915 && subfacet->dp_byte_count <= stats->n_bytes) {
4916 stats->n_packets -= subfacet->dp_packet_count;
4917 stats->n_bytes -= subfacet->dp_byte_count;
4920 subfacet->dp_packet_count = 0;
4921 subfacet->dp_byte_count = 0;
4924 /* Updates 'subfacet''s used time. The caller is responsible for calling
4925 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4927 subfacet_update_time(struct subfacet *subfacet, long long int used)
4929 if (used > subfacet->used) {
4930 subfacet->used = used;
4931 facet_update_time(subfacet->facet, used);
4935 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4937 * Because of the meaning of a subfacet's counters, it only makes sense to do
4938 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4939 * represents a packet that was sent by hand or if it represents statistics
4940 * that have been cleared out of the datapath. */
4942 subfacet_update_stats(struct subfacet *subfacet,
4943 const struct dpif_flow_stats *stats)
4945 if (stats->n_packets || stats->used > subfacet->used) {
4946 struct facet *facet = subfacet->facet;
4948 subfacet_update_time(subfacet, stats->used);
4949 facet->packet_count += stats->n_packets;
4950 facet->byte_count += stats->n_bytes;
4951 facet->tcp_flags |= stats->tcp_flags;
4952 facet_push_stats(facet);
4953 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4959 static struct rule_dpif *
4960 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4962 struct rule_dpif *rule;
4964 rule = rule_dpif_lookup__(ofproto, flow, 0);
4969 return rule_dpif_miss_rule(ofproto, flow);
4972 static struct rule_dpif *
4973 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4976 struct cls_rule *cls_rule;
4977 struct classifier *cls;
4979 if (table_id >= N_TABLES) {
4983 cls = &ofproto->up.tables[table_id].cls;
4984 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4985 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4986 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4987 * are unavailable. */
4988 struct flow ofpc_normal_flow = *flow;
4989 ofpc_normal_flow.tp_src = htons(0);
4990 ofpc_normal_flow.tp_dst = htons(0);
4991 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4993 cls_rule = classifier_lookup(cls, flow);
4995 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4998 static struct rule_dpif *
4999 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5001 struct ofport_dpif *port;
5003 port = get_ofp_port(ofproto, flow->in_port);
5005 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5006 return ofproto->miss_rule;
5009 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5010 return ofproto->no_packet_in_rule;
5012 return ofproto->miss_rule;
5016 complete_operation(struct rule_dpif *rule)
5018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5020 rule_invalidate(rule);
5022 struct dpif_completion *c = xmalloc(sizeof *c);
5023 c->op = rule->up.pending;
5024 list_push_back(&ofproto->completions, &c->list_node);
5026 ofoperation_complete(rule->up.pending, 0);
5030 static struct rule *
5033 struct rule_dpif *rule = xmalloc(sizeof *rule);
5038 rule_dealloc(struct rule *rule_)
5040 struct rule_dpif *rule = rule_dpif_cast(rule_);
5045 rule_construct(struct rule *rule_)
5047 struct rule_dpif *rule = rule_dpif_cast(rule_);
5048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5049 struct rule_dpif *victim;
5052 rule->packet_count = 0;
5053 rule->byte_count = 0;
5055 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5056 if (victim && !list_is_empty(&victim->facets)) {
5057 struct facet *facet;
5059 rule->facets = victim->facets;
5060 list_moved(&rule->facets);
5061 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5062 /* XXX: We're only clearing our local counters here. It's possible
5063 * that quite a few packets are unaccounted for in the datapath
5064 * statistics. These will be accounted to the new rule instead of
5065 * cleared as required. This could be fixed by clearing out the
5066 * datapath statistics for this facet, but currently it doesn't
5068 facet_reset_counters(facet);
5072 /* Must avoid list_moved() in this case. */
5073 list_init(&rule->facets);
5076 table_id = rule->up.table_id;
5078 rule->tag = victim->tag;
5079 } else if (table_id == 0) {
5084 miniflow_expand(&rule->up.cr.match.flow, &flow);
5085 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5086 ofproto->tables[table_id].basis);
5089 complete_operation(rule);
5094 rule_destruct(struct rule *rule_)
5096 struct rule_dpif *rule = rule_dpif_cast(rule_);
5097 struct facet *facet, *next_facet;
5099 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5100 facet_revalidate(facet);
5103 complete_operation(rule);
5107 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5109 struct rule_dpif *rule = rule_dpif_cast(rule_);
5110 struct facet *facet;
5112 /* Start from historical data for 'rule' itself that are no longer tracked
5113 * in facets. This counts, for example, facets that have expired. */
5114 *packets = rule->packet_count;
5115 *bytes = rule->byte_count;
5117 /* Add any statistics that are tracked by facets. This includes
5118 * statistical data recently updated by ofproto_update_stats() as well as
5119 * stats for packets that were executed "by hand" via dpif_execute(). */
5120 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5121 *packets += facet->packet_count;
5122 *bytes += facet->byte_count;
5127 rule_execute(struct rule *rule_, const struct flow *flow,
5128 struct ofpbuf *packet)
5130 struct rule_dpif *rule = rule_dpif_cast(rule_);
5131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5133 struct dpif_flow_stats stats;
5135 struct action_xlate_ctx ctx;
5136 uint64_t odp_actions_stub[1024 / 8];
5137 struct ofpbuf odp_actions;
5139 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5140 rule_credit_stats(rule, &stats);
5142 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5143 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5144 rule, stats.tcp_flags, packet);
5145 ctx.resubmit_stats = &stats;
5146 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5148 execute_odp_actions(ofproto, flow, odp_actions.data,
5149 odp_actions.size, packet);
5151 ofpbuf_uninit(&odp_actions);
5157 rule_modify_actions(struct rule *rule_)
5159 struct rule_dpif *rule = rule_dpif_cast(rule_);
5161 complete_operation(rule);
5164 /* Sends 'packet' out 'ofport'.
5165 * May modify 'packet'.
5166 * Returns 0 if successful, otherwise a positive errno value. */
5168 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5170 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5171 struct ofpbuf key, odp_actions;
5172 struct odputil_keybuf keybuf;
5177 flow_extract(packet, 0, NULL, OFPP_LOCAL, &flow);
5178 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5180 if (odp_port != ofport->odp_port) {
5181 eth_pop_vlan(packet);
5182 flow.vlan_tci = htons(0);
5185 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5186 odp_flow_key_from_flow(&key, &flow,
5187 ofp_port_to_odp_port(ofproto, flow.in_port));
5189 ofpbuf_init(&odp_actions, 32);
5190 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5192 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5193 error = dpif_execute(ofproto->backer->dpif,
5195 odp_actions.data, odp_actions.size,
5197 ofpbuf_uninit(&odp_actions);
5200 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5201 ofproto->up.name, odp_port, strerror(error));
5203 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5207 /* OpenFlow to datapath action translation. */
5209 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5210 struct action_xlate_ctx *);
5211 static void xlate_normal(struct action_xlate_ctx *);
5213 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5214 * The action will state 'slow' as the reason that the action is in the slow
5215 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5216 * dump-flows" output to see why a flow is in the slow path.)
5218 * The 'stub_size' bytes in 'stub' will be used to store the action.
5219 * 'stub_size' must be large enough for the action.
5221 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5224 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5225 enum slow_path_reason slow,
5226 uint64_t *stub, size_t stub_size,
5227 const struct nlattr **actionsp, size_t *actions_lenp)
5229 union user_action_cookie cookie;
5232 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5233 cookie.slow_path.unused = 0;
5234 cookie.slow_path.reason = slow;
5236 ofpbuf_use_stack(&buf, stub, stub_size);
5237 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5238 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5239 odp_put_userspace_action(pid, &cookie, &buf);
5241 put_userspace_action(ofproto, &buf, flow, &cookie);
5243 *actionsp = buf.data;
5244 *actions_lenp = buf.size;
5248 put_userspace_action(const struct ofproto_dpif *ofproto,
5249 struct ofpbuf *odp_actions,
5250 const struct flow *flow,
5251 const union user_action_cookie *cookie)
5255 pid = dpif_port_get_pid(ofproto->backer->dpif,
5256 ofp_port_to_odp_port(ofproto, flow->in_port));
5258 return odp_put_userspace_action(pid, cookie, odp_actions);
5262 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5263 ovs_be16 vlan_tci, uint32_t odp_port,
5264 unsigned int n_outputs, union user_action_cookie *cookie)
5268 cookie->type = USER_ACTION_COOKIE_SFLOW;
5269 cookie->sflow.vlan_tci = vlan_tci;
5271 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5272 * port information") for the interpretation of cookie->output. */
5273 switch (n_outputs) {
5275 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5276 cookie->sflow.output = 0x40000000 | 256;
5280 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5282 cookie->sflow.output = ifindex;
5287 /* 0x80000000 means "multiple output ports. */
5288 cookie->sflow.output = 0x80000000 | n_outputs;
5293 /* Compose SAMPLE action for sFlow. */
5295 compose_sflow_action(const struct ofproto_dpif *ofproto,
5296 struct ofpbuf *odp_actions,
5297 const struct flow *flow,
5300 uint32_t probability;
5301 union user_action_cookie cookie;
5302 size_t sample_offset, actions_offset;
5305 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5309 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5311 /* Number of packets out of UINT_MAX to sample. */
5312 probability = dpif_sflow_get_probability(ofproto->sflow);
5313 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5315 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5316 compose_sflow_cookie(ofproto, htons(0), odp_port,
5317 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5318 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5320 nl_msg_end_nested(odp_actions, actions_offset);
5321 nl_msg_end_nested(odp_actions, sample_offset);
5322 return cookie_offset;
5325 /* SAMPLE action must be first action in any given list of actions.
5326 * At this point we do not have all information required to build it. So try to
5327 * build sample action as complete as possible. */
5329 add_sflow_action(struct action_xlate_ctx *ctx)
5331 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5333 &ctx->flow, OVSP_NONE);
5334 ctx->sflow_odp_port = 0;
5335 ctx->sflow_n_outputs = 0;
5338 /* Fix SAMPLE action according to data collected while composing ODP actions.
5339 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5340 * USERSPACE action's user-cookie which is required for sflow. */
5342 fix_sflow_action(struct action_xlate_ctx *ctx)
5344 const struct flow *base = &ctx->base_flow;
5345 union user_action_cookie *cookie;
5347 if (!ctx->user_cookie_offset) {
5351 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5353 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5355 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5356 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5360 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5363 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5364 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5365 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5366 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5370 struct priority_to_dscp *pdscp;
5372 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5373 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5375 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5376 xlate_report(ctx, "STP not in forwarding state, skipping output");
5380 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5382 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5383 ctx->flow.nw_tos |= pdscp->dscp;
5386 /* We may not have an ofport record for this port, but it doesn't hurt
5387 * to allow forwarding to it anyhow. Maybe such a port will appear
5388 * later and we're pre-populating the flow table. */
5391 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5392 ctx->flow.vlan_tci);
5393 if (out_port != odp_port) {
5394 ctx->flow.vlan_tci = htons(0);
5396 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5397 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5399 ctx->sflow_odp_port = odp_port;
5400 ctx->sflow_n_outputs++;
5401 ctx->nf_output_iface = ofp_port;
5402 ctx->flow.vlan_tci = flow_vlan_tci;
5403 ctx->flow.nw_tos = flow_nw_tos;
5407 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5409 compose_output_action__(ctx, ofp_port, true);
5413 xlate_table_action(struct action_xlate_ctx *ctx,
5414 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5416 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5417 struct ofproto_dpif *ofproto = ctx->ofproto;
5418 struct rule_dpif *rule;
5419 uint16_t old_in_port;
5420 uint8_t old_table_id;
5422 old_table_id = ctx->table_id;
5423 ctx->table_id = table_id;
5425 /* Look up a flow with 'in_port' as the input port. */
5426 old_in_port = ctx->flow.in_port;
5427 ctx->flow.in_port = in_port;
5428 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5431 if (table_id > 0 && table_id < N_TABLES) {
5432 struct table_dpif *table = &ofproto->tables[table_id];
5433 if (table->other_table) {
5434 ctx->tags |= (rule && rule->tag
5436 : rule_calculate_tag(&ctx->flow,
5437 &table->other_table->mask,
5442 /* Restore the original input port. Otherwise OFPP_NORMAL and
5443 * OFPP_IN_PORT will have surprising behavior. */
5444 ctx->flow.in_port = old_in_port;
5446 if (ctx->resubmit_hook) {
5447 ctx->resubmit_hook(ctx, rule);
5450 if (rule == NULL && may_packet_in) {
5452 * check if table configuration flags
5453 * OFPTC_TABLE_MISS_CONTROLLER, default.
5454 * OFPTC_TABLE_MISS_CONTINUE,
5455 * OFPTC_TABLE_MISS_DROP
5456 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5458 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5462 struct rule_dpif *old_rule = ctx->rule;
5464 if (ctx->resubmit_stats) {
5465 rule_credit_stats(rule, ctx->resubmit_stats);
5470 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5471 ctx->rule = old_rule;
5475 ctx->table_id = old_table_id;
5477 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5479 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5480 MAX_RESUBMIT_RECURSION);
5481 ctx->max_resubmit_trigger = true;
5486 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5487 const struct ofpact_resubmit *resubmit)
5492 in_port = resubmit->in_port;
5493 if (in_port == OFPP_IN_PORT) {
5494 in_port = ctx->flow.in_port;
5497 table_id = resubmit->table_id;
5498 if (table_id == 255) {
5499 table_id = ctx->table_id;
5502 xlate_table_action(ctx, in_port, table_id, false);
5506 flood_packets(struct action_xlate_ctx *ctx, bool all)
5508 struct ofport_dpif *ofport;
5510 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5511 uint16_t ofp_port = ofport->up.ofp_port;
5513 if (ofp_port == ctx->flow.in_port) {
5518 compose_output_action__(ctx, ofp_port, false);
5519 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5520 compose_output_action(ctx, ofp_port);
5524 ctx->nf_output_iface = NF_OUT_FLOOD;
5528 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5529 enum ofp_packet_in_reason reason,
5530 uint16_t controller_id)
5532 struct ofputil_packet_in pin;
5533 struct ofpbuf *packet;
5535 ctx->slow |= SLOW_CONTROLLER;
5540 packet = ofpbuf_clone(ctx->packet);
5542 if (packet->l2 && packet->l3) {
5543 struct eth_header *eh;
5545 eth_pop_vlan(packet);
5548 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5549 * LLC frame. Calculating the Ethernet type of these frames is more
5550 * trouble than seems appropriate for a simple assertion. */
5551 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5552 || eh->eth_type == ctx->flow.dl_type);
5554 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5555 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5557 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5558 eth_push_vlan(packet, ctx->flow.vlan_tci);
5562 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5563 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5564 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5568 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5569 packet_set_tcp_port(packet, ctx->flow.tp_src,
5571 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5572 packet_set_udp_port(packet, ctx->flow.tp_src,
5579 pin.packet = packet->data;
5580 pin.packet_len = packet->size;
5581 pin.reason = reason;
5582 pin.controller_id = controller_id;
5583 pin.table_id = ctx->table_id;
5584 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5587 flow_get_metadata(&ctx->flow, &pin.fmd);
5589 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5590 ofpbuf_delete(packet);
5594 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5596 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5597 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5601 if (ctx->flow.nw_ttl > 1) {
5607 for (i = 0; i < ids->n_controllers; i++) {
5608 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5612 /* Stop processing for current table. */
5618 xlate_output_action(struct action_xlate_ctx *ctx,
5619 uint16_t port, uint16_t max_len, bool may_packet_in)
5621 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5623 ctx->nf_output_iface = NF_OUT_DROP;
5627 compose_output_action(ctx, ctx->flow.in_port);
5630 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5636 flood_packets(ctx, false);
5639 flood_packets(ctx, true);
5641 case OFPP_CONTROLLER:
5642 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5648 if (port != ctx->flow.in_port) {
5649 compose_output_action(ctx, port);
5651 xlate_report(ctx, "skipping output to input port");
5656 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5657 ctx->nf_output_iface = NF_OUT_FLOOD;
5658 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5659 ctx->nf_output_iface = prev_nf_output_iface;
5660 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5661 ctx->nf_output_iface != NF_OUT_FLOOD) {
5662 ctx->nf_output_iface = NF_OUT_MULTI;
5667 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5668 const struct ofpact_output_reg *or)
5670 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5671 if (port <= UINT16_MAX) {
5672 xlate_output_action(ctx, port, or->max_len, false);
5677 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5678 const struct ofpact_enqueue *enqueue)
5680 uint16_t ofp_port = enqueue->port;
5681 uint32_t queue_id = enqueue->queue;
5682 uint32_t flow_priority, priority;
5685 /* Translate queue to priority. */
5686 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5687 queue_id, &priority);
5689 /* Fall back to ordinary output action. */
5690 xlate_output_action(ctx, enqueue->port, 0, false);
5694 /* Check output port. */
5695 if (ofp_port == OFPP_IN_PORT) {
5696 ofp_port = ctx->flow.in_port;
5697 } else if (ofp_port == ctx->flow.in_port) {
5701 /* Add datapath actions. */
5702 flow_priority = ctx->flow.skb_priority;
5703 ctx->flow.skb_priority = priority;
5704 compose_output_action(ctx, ofp_port);
5705 ctx->flow.skb_priority = flow_priority;
5707 /* Update NetFlow output port. */
5708 if (ctx->nf_output_iface == NF_OUT_DROP) {
5709 ctx->nf_output_iface = ofp_port;
5710 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5711 ctx->nf_output_iface = NF_OUT_MULTI;
5716 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5718 uint32_t skb_priority;
5720 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5721 queue_id, &skb_priority)) {
5722 ctx->flow.skb_priority = skb_priority;
5724 /* Couldn't translate queue to a priority. Nothing to do. A warning
5725 * has already been logged. */
5729 struct xlate_reg_state {
5735 xlate_autopath(struct action_xlate_ctx *ctx,
5736 const struct ofpact_autopath *ap)
5738 uint16_t ofp_port = ap->port;
5739 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5741 if (!port || !port->bundle) {
5742 ofp_port = OFPP_NONE;
5743 } else if (port->bundle->bond) {
5744 /* Autopath does not support VLAN hashing. */
5745 struct ofport_dpif *slave = bond_choose_output_slave(
5746 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5748 ofp_port = slave->up.ofp_port;
5751 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5755 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5757 struct ofproto_dpif *ofproto = ofproto_;
5758 struct ofport_dpif *port;
5768 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5771 port = get_ofp_port(ofproto, ofp_port);
5772 return port ? port->may_enable : false;
5777 xlate_bundle_action(struct action_xlate_ctx *ctx,
5778 const struct ofpact_bundle *bundle)
5782 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5783 if (bundle->dst.field) {
5784 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5786 xlate_output_action(ctx, port, 0, false);
5791 xlate_learn_action(struct action_xlate_ctx *ctx,
5792 const struct ofpact_learn *learn)
5794 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5795 struct ofputil_flow_mod fm;
5796 uint64_t ofpacts_stub[1024 / 8];
5797 struct ofpbuf ofpacts;
5800 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5801 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5803 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5804 if (error && !VLOG_DROP_WARN(&rl)) {
5805 VLOG_WARN("learning action failed to modify flow table (%s)",
5806 ofperr_get_name(error));
5809 ofpbuf_uninit(&ofpacts);
5812 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5813 * means "infinite". */
5815 reduce_timeout(uint16_t max, uint16_t *timeout)
5817 if (max && (!*timeout || *timeout > max)) {
5823 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5824 const struct ofpact_fin_timeout *oft)
5826 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5827 struct rule_dpif *rule = ctx->rule;
5829 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5830 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5835 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5837 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5838 ? OFPUTIL_PC_NO_RECV_STP
5839 : OFPUTIL_PC_NO_RECV)) {
5843 /* Only drop packets here if both forwarding and learning are
5844 * disabled. If just learning is enabled, we need to have
5845 * OFPP_NORMAL and the learning action have a look at the packet
5846 * before we can drop it. */
5847 if (!stp_forward_in_state(port->stp_state)
5848 && !stp_learn_in_state(port->stp_state)) {
5856 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5857 struct action_xlate_ctx *ctx)
5859 const struct ofport_dpif *port;
5860 bool was_evictable = true;
5861 const struct ofpact *a;
5863 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5864 if (port && !may_receive(port, ctx)) {
5865 /* Drop this flow. */
5870 /* Don't let the rule we're working on get evicted underneath us. */
5871 was_evictable = ctx->rule->up.evictable;
5872 ctx->rule->up.evictable = false;
5874 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5875 struct ofpact_controller *controller;
5876 const struct ofpact_metadata *metadata;
5884 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5885 ofpact_get_OUTPUT(a)->max_len, true);
5888 case OFPACT_CONTROLLER:
5889 controller = ofpact_get_CONTROLLER(a);
5890 execute_controller_action(ctx, controller->max_len,
5892 controller->controller_id);
5895 case OFPACT_ENQUEUE:
5896 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5899 case OFPACT_SET_VLAN_VID:
5900 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5901 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5905 case OFPACT_SET_VLAN_PCP:
5906 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5907 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5912 case OFPACT_STRIP_VLAN:
5913 ctx->flow.vlan_tci = htons(0);
5916 case OFPACT_PUSH_VLAN:
5917 /* TODO:XXX 802.1AD(QinQ) */
5918 ctx->flow.vlan_tci = htons(VLAN_CFI);
5921 case OFPACT_SET_ETH_SRC:
5922 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5926 case OFPACT_SET_ETH_DST:
5927 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5931 case OFPACT_SET_IPV4_SRC:
5932 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5935 case OFPACT_SET_IPV4_DST:
5936 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5939 case OFPACT_SET_IPV4_DSCP:
5940 /* OpenFlow 1.0 only supports IPv4. */
5941 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5942 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5943 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5947 case OFPACT_SET_L4_SRC_PORT:
5948 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5951 case OFPACT_SET_L4_DST_PORT:
5952 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5955 case OFPACT_RESUBMIT:
5956 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5959 case OFPACT_SET_TUNNEL:
5960 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5963 case OFPACT_SET_QUEUE:
5964 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5967 case OFPACT_POP_QUEUE:
5968 ctx->flow.skb_priority = ctx->orig_skb_priority;
5971 case OFPACT_REG_MOVE:
5972 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5975 case OFPACT_REG_LOAD:
5976 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5979 case OFPACT_DEC_TTL:
5980 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5986 /* Nothing to do. */
5989 case OFPACT_MULTIPATH:
5990 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5993 case OFPACT_AUTOPATH:
5994 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5998 ctx->ofproto->has_bundle_action = true;
5999 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6002 case OFPACT_OUTPUT_REG:
6003 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6007 ctx->has_learn = true;
6008 if (ctx->may_learn) {
6009 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6017 case OFPACT_FIN_TIMEOUT:
6018 ctx->has_fin_timeout = true;
6019 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6022 case OFPACT_CLEAR_ACTIONS:
6024 * Nothing to do because writa-actions is not supported for now.
6025 * When writa-actions is supported, clear-actions also must
6026 * be supported at the same time.
6030 case OFPACT_WRITE_METADATA:
6031 metadata = ofpact_get_WRITE_METADATA(a);
6032 ctx->flow.metadata &= ~metadata->mask;
6033 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6036 case OFPACT_GOTO_TABLE: {
6037 /* TODO:XXX remove recursion */
6038 /* It is assumed that goto-table is last action */
6039 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6040 assert(ctx->table_id < ogt->table_id);
6041 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6048 /* We've let OFPP_NORMAL and the learning action look at the packet,
6049 * so drop it now if forwarding is disabled. */
6050 if (port && !stp_forward_in_state(port->stp_state)) {
6051 ofpbuf_clear(ctx->odp_actions);
6052 add_sflow_action(ctx);
6055 ctx->rule->up.evictable = was_evictable;
6060 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6061 struct ofproto_dpif *ofproto, const struct flow *flow,
6062 ovs_be16 initial_tci, struct rule_dpif *rule,
6063 uint8_t tcp_flags, const struct ofpbuf *packet)
6065 ctx->ofproto = ofproto;
6067 ctx->base_flow = ctx->flow;
6068 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6069 ctx->base_flow.vlan_tci = initial_tci;
6071 ctx->packet = packet;
6072 ctx->may_learn = packet != NULL;
6073 ctx->tcp_flags = tcp_flags;
6074 ctx->resubmit_hook = NULL;
6075 ctx->report_hook = NULL;
6076 ctx->resubmit_stats = NULL;
6079 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6080 * into datapath actions in 'odp_actions', using 'ctx'. */
6082 xlate_actions(struct action_xlate_ctx *ctx,
6083 const struct ofpact *ofpacts, size_t ofpacts_len,
6084 struct ofpbuf *odp_actions)
6086 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6087 * that in the future we always keep a copy of the original flow for
6088 * tracing purposes. */
6089 static bool hit_resubmit_limit;
6091 enum slow_path_reason special;
6093 COVERAGE_INC(ofproto_dpif_xlate);
6095 ofpbuf_clear(odp_actions);
6096 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6098 ctx->odp_actions = odp_actions;
6101 ctx->has_learn = false;
6102 ctx->has_normal = false;
6103 ctx->has_fin_timeout = false;
6104 ctx->nf_output_iface = NF_OUT_DROP;
6107 ctx->max_resubmit_trigger = false;
6108 ctx->orig_skb_priority = ctx->flow.skb_priority;
6112 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6113 /* Do this conditionally because the copy is expensive enough that it
6114 * shows up in profiles.
6116 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6117 * believe that I wasn't using it without initializing it if I kept it
6118 * in a local variable. */
6119 ctx->orig_flow = ctx->flow;
6122 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6123 switch (ctx->ofproto->up.frag_handling) {
6124 case OFPC_FRAG_NORMAL:
6125 /* We must pretend that transport ports are unavailable. */
6126 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6127 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6130 case OFPC_FRAG_DROP:
6133 case OFPC_FRAG_REASM:
6136 case OFPC_FRAG_NX_MATCH:
6137 /* Nothing to do. */
6140 case OFPC_INVALID_TTL_TO_CONTROLLER:
6145 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6147 ctx->slow |= special;
6149 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6150 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6152 add_sflow_action(ctx);
6153 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6155 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6156 if (!hit_resubmit_limit) {
6157 /* We didn't record the original flow. Make sure we do from
6159 hit_resubmit_limit = true;
6160 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6161 struct ds ds = DS_EMPTY_INITIALIZER;
6163 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6165 VLOG_ERR("Trace triggered by excessive resubmit "
6166 "recursion:\n%s", ds_cstr(&ds));
6171 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6172 ctx->odp_actions->data,
6173 ctx->odp_actions->size)) {
6174 ctx->slow |= SLOW_IN_BAND;
6176 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6178 compose_output_action(ctx, OFPP_LOCAL);
6181 if (ctx->ofproto->has_mirrors) {
6182 add_mirror_actions(ctx, &ctx->orig_flow);
6184 fix_sflow_action(ctx);
6188 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6189 * into datapath actions, using 'ctx', and discards the datapath actions. */
6191 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6192 const struct ofpact *ofpacts,
6195 uint64_t odp_actions_stub[1024 / 8];
6196 struct ofpbuf odp_actions;
6198 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6199 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6200 ofpbuf_uninit(&odp_actions);
6204 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6206 if (ctx->report_hook) {
6207 ctx->report_hook(ctx, s);
6211 /* OFPP_NORMAL implementation. */
6213 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6215 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6216 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6217 * the bundle on which the packet was received, returns the VLAN to which the
6220 * Both 'vid' and the return value are in the range 0...4095. */
6222 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6224 switch (in_bundle->vlan_mode) {
6225 case PORT_VLAN_ACCESS:
6226 return in_bundle->vlan;
6229 case PORT_VLAN_TRUNK:
6232 case PORT_VLAN_NATIVE_UNTAGGED:
6233 case PORT_VLAN_NATIVE_TAGGED:
6234 return vid ? vid : in_bundle->vlan;
6241 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6242 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6245 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6246 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6249 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6251 /* Allow any VID on the OFPP_NONE port. */
6252 if (in_bundle == &ofpp_none_bundle) {
6256 switch (in_bundle->vlan_mode) {
6257 case PORT_VLAN_ACCESS:
6260 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6261 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6262 "packet received on port %s configured as VLAN "
6263 "%"PRIu16" access port",
6264 in_bundle->ofproto->up.name, vid,
6265 in_bundle->name, in_bundle->vlan);
6271 case PORT_VLAN_NATIVE_UNTAGGED:
6272 case PORT_VLAN_NATIVE_TAGGED:
6274 /* Port must always carry its native VLAN. */
6278 case PORT_VLAN_TRUNK:
6279 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6281 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6282 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6283 "received on port %s not configured for trunking "
6285 in_bundle->ofproto->up.name, vid,
6286 in_bundle->name, vid);
6298 /* Given 'vlan', the VLAN that a packet belongs to, and
6299 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6300 * that should be included in the 802.1Q header. (If the return value is 0,
6301 * then the 802.1Q header should only be included in the packet if there is a
6304 * Both 'vlan' and the return value are in the range 0...4095. */
6306 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6308 switch (out_bundle->vlan_mode) {
6309 case PORT_VLAN_ACCESS:
6312 case PORT_VLAN_TRUNK:
6313 case PORT_VLAN_NATIVE_TAGGED:
6316 case PORT_VLAN_NATIVE_UNTAGGED:
6317 return vlan == out_bundle->vlan ? 0 : vlan;
6325 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6328 struct ofport_dpif *port;
6330 ovs_be16 tci, old_tci;
6332 vid = output_vlan_to_vid(out_bundle, vlan);
6333 if (!out_bundle->bond) {
6334 port = ofbundle_get_a_port(out_bundle);
6336 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6339 /* No slaves enabled, so drop packet. */
6344 old_tci = ctx->flow.vlan_tci;
6346 if (tci || out_bundle->use_priority_tags) {
6347 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6349 tci |= htons(VLAN_CFI);
6352 ctx->flow.vlan_tci = tci;
6354 compose_output_action(ctx, port->up.ofp_port);
6355 ctx->flow.vlan_tci = old_tci;
6359 mirror_mask_ffs(mirror_mask_t mask)
6361 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6366 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6368 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6369 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6373 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6375 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6378 /* Returns an arbitrary interface within 'bundle'. */
6379 static struct ofport_dpif *
6380 ofbundle_get_a_port(const struct ofbundle *bundle)
6382 return CONTAINER_OF(list_front(&bundle->ports),
6383 struct ofport_dpif, bundle_node);
6387 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6389 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6393 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6395 struct ofproto_dpif *ofproto = ctx->ofproto;
6396 mirror_mask_t mirrors;
6397 struct ofbundle *in_bundle;
6400 const struct nlattr *a;
6403 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6404 ctx->packet != NULL, NULL);
6408 mirrors = in_bundle->src_mirrors;
6410 /* Drop frames on bundles reserved for mirroring. */
6411 if (in_bundle->mirror_out) {
6412 if (ctx->packet != NULL) {
6413 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6414 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6415 "%s, which is reserved exclusively for mirroring",
6416 ctx->ofproto->up.name, in_bundle->name);
6422 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6423 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6426 vlan = input_vid_to_vlan(in_bundle, vid);
6428 /* Look at the output ports to check for destination selections. */
6430 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6431 ctx->odp_actions->size) {
6432 enum ovs_action_attr type = nl_attr_type(a);
6433 struct ofport_dpif *ofport;
6435 if (type != OVS_ACTION_ATTR_OUTPUT) {
6439 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6440 if (ofport && ofport->bundle) {
6441 mirrors |= ofport->bundle->dst_mirrors;
6449 /* Restore the original packet before adding the mirror actions. */
6450 ctx->flow = *orig_flow;
6455 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6457 if (!vlan_is_mirrored(m, vlan)) {
6458 mirrors = zero_rightmost_1bit(mirrors);
6462 mirrors &= ~m->dup_mirrors;
6463 ctx->mirrors |= m->dup_mirrors;
6465 output_normal(ctx, m->out, vlan);
6466 } else if (vlan != m->out_vlan
6467 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6468 struct ofbundle *bundle;
6470 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6471 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6472 && !bundle->mirror_out) {
6473 output_normal(ctx, bundle, m->out_vlan);
6481 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6482 uint64_t packets, uint64_t bytes)
6488 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6491 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6494 /* In normal circumstances 'm' will not be NULL. However,
6495 * if mirrors are reconfigured, we can temporarily get out
6496 * of sync in facet_revalidate(). We could "correct" the
6497 * mirror list before reaching here, but doing that would
6498 * not properly account the traffic stats we've currently
6499 * accumulated for previous mirror configuration. */
6503 m->packet_count += packets;
6504 m->byte_count += bytes;
6508 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6509 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6510 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6512 is_gratuitous_arp(const struct flow *flow)
6514 return (flow->dl_type == htons(ETH_TYPE_ARP)
6515 && eth_addr_is_broadcast(flow->dl_dst)
6516 && (flow->nw_proto == ARP_OP_REPLY
6517 || (flow->nw_proto == ARP_OP_REQUEST
6518 && flow->nw_src == flow->nw_dst)));
6522 update_learning_table(struct ofproto_dpif *ofproto,
6523 const struct flow *flow, int vlan,
6524 struct ofbundle *in_bundle)
6526 struct mac_entry *mac;
6528 /* Don't learn the OFPP_NONE port. */
6529 if (in_bundle == &ofpp_none_bundle) {
6533 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6537 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6538 if (is_gratuitous_arp(flow)) {
6539 /* We don't want to learn from gratuitous ARP packets that are
6540 * reflected back over bond slaves so we lock the learning table. */
6541 if (!in_bundle->bond) {
6542 mac_entry_set_grat_arp_lock(mac);
6543 } else if (mac_entry_is_grat_arp_locked(mac)) {
6548 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6549 /* The log messages here could actually be useful in debugging,
6550 * so keep the rate limit relatively high. */
6551 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6552 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6553 "on port %s in VLAN %d",
6554 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6555 in_bundle->name, vlan);
6557 mac->port.p = in_bundle;
6558 tag_set_add(&ofproto->revalidate_set,
6559 mac_learning_changed(ofproto->ml, mac));
6563 static struct ofbundle *
6564 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6565 bool warn, struct ofport_dpif **in_ofportp)
6567 struct ofport_dpif *ofport;
6569 /* Find the port and bundle for the received packet. */
6570 ofport = get_ofp_port(ofproto, in_port);
6572 *in_ofportp = ofport;
6574 if (ofport && ofport->bundle) {
6575 return ofport->bundle;
6578 /* Special-case OFPP_NONE, which a controller may use as the ingress
6579 * port for traffic that it is sourcing. */
6580 if (in_port == OFPP_NONE) {
6581 return &ofpp_none_bundle;
6584 /* Odd. A few possible reasons here:
6586 * - We deleted a port but there are still a few packets queued up
6589 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6590 * we don't know about.
6592 * - The ofproto client didn't configure the port as part of a bundle.
6593 * This is particularly likely to happen if a packet was received on the
6594 * port after it was created, but before the client had a chance to
6595 * configure its bundle.
6598 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6600 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6601 "port %"PRIu16, ofproto->up.name, in_port);
6606 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6607 * dropped. Returns true if they may be forwarded, false if they should be
6610 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6611 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6613 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6614 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6615 * checked by input_vid_is_valid().
6617 * May also add tags to '*tags', although the current implementation only does
6618 * so in one special case.
6621 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6624 struct ofproto_dpif *ofproto = ctx->ofproto;
6625 struct flow *flow = &ctx->flow;
6626 struct ofbundle *in_bundle = in_port->bundle;
6628 /* Drop frames for reserved multicast addresses
6629 * only if forward_bpdu option is absent. */
6630 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6631 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6635 if (in_bundle->bond) {
6636 struct mac_entry *mac;
6638 switch (bond_check_admissibility(in_bundle->bond, in_port,
6639 flow->dl_dst, &ctx->tags)) {
6644 xlate_report(ctx, "bonding refused admissibility, dropping");
6647 case BV_DROP_IF_MOVED:
6648 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6649 if (mac && mac->port.p != in_bundle &&
6650 (!is_gratuitous_arp(flow)
6651 || mac_entry_is_grat_arp_locked(mac))) {
6652 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6664 xlate_normal(struct action_xlate_ctx *ctx)
6666 struct ofport_dpif *in_port;
6667 struct ofbundle *in_bundle;
6668 struct mac_entry *mac;
6672 ctx->has_normal = true;
6674 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6675 ctx->packet != NULL, &in_port);
6677 xlate_report(ctx, "no input bundle, dropping");
6681 /* Drop malformed frames. */
6682 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6683 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6684 if (ctx->packet != NULL) {
6685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6686 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6687 "VLAN tag received on port %s",
6688 ctx->ofproto->up.name, in_bundle->name);
6690 xlate_report(ctx, "partial VLAN tag, dropping");
6694 /* Drop frames on bundles reserved for mirroring. */
6695 if (in_bundle->mirror_out) {
6696 if (ctx->packet != NULL) {
6697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6698 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6699 "%s, which is reserved exclusively for mirroring",
6700 ctx->ofproto->up.name, in_bundle->name);
6702 xlate_report(ctx, "input port is mirror output port, dropping");
6707 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6708 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6709 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6712 vlan = input_vid_to_vlan(in_bundle, vid);
6714 /* Check other admissibility requirements. */
6715 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6719 /* Learn source MAC. */
6720 if (ctx->may_learn) {
6721 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6724 /* Determine output bundle. */
6725 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6728 if (mac->port.p != in_bundle) {
6729 xlate_report(ctx, "forwarding to learned port");
6730 output_normal(ctx, mac->port.p, vlan);
6732 xlate_report(ctx, "learned port is input port, dropping");
6735 struct ofbundle *bundle;
6737 xlate_report(ctx, "no learned MAC for destination, flooding");
6738 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6739 if (bundle != in_bundle
6740 && ofbundle_includes_vlan(bundle, vlan)
6741 && bundle->floodable
6742 && !bundle->mirror_out) {
6743 output_normal(ctx, bundle, vlan);
6746 ctx->nf_output_iface = NF_OUT_FLOOD;
6750 /* Optimized flow revalidation.
6752 * It's a difficult problem, in general, to tell which facets need to have
6753 * their actions recalculated whenever the OpenFlow flow table changes. We
6754 * don't try to solve that general problem: for most kinds of OpenFlow flow
6755 * table changes, we recalculate the actions for every facet. This is
6756 * relatively expensive, but it's good enough if the OpenFlow flow table
6757 * doesn't change very often.
6759 * However, we can expect one particular kind of OpenFlow flow table change to
6760 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6761 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6762 * table, we add a special case that applies to flow tables in which every rule
6763 * has the same form (that is, the same wildcards), except that the table is
6764 * also allowed to have a single "catch-all" flow that matches all packets. We
6765 * optimize this case by tagging all of the facets that resubmit into the table
6766 * and invalidating the same tag whenever a flow changes in that table. The
6767 * end result is that we revalidate just the facets that need it (and sometimes
6768 * a few more, but not all of the facets or even all of the facets that
6769 * resubmit to the table modified by MAC learning). */
6771 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6772 * into an OpenFlow table with the given 'basis'. */
6774 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6777 if (minimask_is_catchall(mask)) {
6780 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6781 return tag_create_deterministic(hash);
6785 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6786 * taggability of that table.
6788 * This function must be called after *each* change to a flow table. If you
6789 * skip calling it on some changes then the pointer comparisons at the end can
6790 * be invalid if you get unlucky. For example, if a flow removal causes a
6791 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6792 * different wildcards to be created with the same address, then this function
6793 * will incorrectly skip revalidation. */
6795 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6797 struct table_dpif *table = &ofproto->tables[table_id];
6798 const struct oftable *oftable = &ofproto->up.tables[table_id];
6799 struct cls_table *catchall, *other;
6800 struct cls_table *t;
6802 catchall = other = NULL;
6804 switch (hmap_count(&oftable->cls.tables)) {
6806 /* We could tag this OpenFlow table but it would make the logic a
6807 * little harder and it's a corner case that doesn't seem worth it
6813 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6814 if (cls_table_is_catchall(t)) {
6816 } else if (!other) {
6819 /* Indicate that we can't tag this by setting both tables to
6820 * NULL. (We know that 'catchall' is already NULL.) */
6827 /* Can't tag this table. */
6831 if (table->catchall_table != catchall || table->other_table != other) {
6832 table->catchall_table = catchall;
6833 table->other_table = other;
6834 ofproto->need_revalidate = REV_FLOW_TABLE;
6838 /* Given 'rule' that has changed in some way (either it is a rule being
6839 * inserted, a rule being deleted, or a rule whose actions are being
6840 * modified), marks facets for revalidation to ensure that packets will be
6841 * forwarded correctly according to the new state of the flow table.
6843 * This function must be called after *each* change to a flow table. See
6844 * the comment on table_update_taggable() for more information. */
6846 rule_invalidate(const struct rule_dpif *rule)
6848 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6850 table_update_taggable(ofproto, rule->up.table_id);
6852 if (!ofproto->need_revalidate) {
6853 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6855 if (table->other_table && rule->tag) {
6856 tag_set_add(&ofproto->revalidate_set, rule->tag);
6858 ofproto->need_revalidate = REV_FLOW_TABLE;
6864 set_frag_handling(struct ofproto *ofproto_,
6865 enum ofp_config_flags frag_handling)
6867 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6869 if (frag_handling != OFPC_FRAG_REASM) {
6870 ofproto->need_revalidate = REV_RECONFIGURE;
6878 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6879 const struct flow *flow,
6880 const struct ofpact *ofpacts, size_t ofpacts_len)
6882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6883 struct odputil_keybuf keybuf;
6884 struct dpif_flow_stats stats;
6888 struct action_xlate_ctx ctx;
6889 uint64_t odp_actions_stub[1024 / 8];
6890 struct ofpbuf odp_actions;
6892 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6893 odp_flow_key_from_flow(&key, flow,
6894 ofp_port_to_odp_port(ofproto, flow->in_port));
6896 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6898 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6899 packet_get_tcp_flags(packet, flow), packet);
6900 ctx.resubmit_stats = &stats;
6902 ofpbuf_use_stub(&odp_actions,
6903 odp_actions_stub, sizeof odp_actions_stub);
6904 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6905 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6906 odp_actions.data, odp_actions.size, packet);
6907 ofpbuf_uninit(&odp_actions);
6915 set_netflow(struct ofproto *ofproto_,
6916 const struct netflow_options *netflow_options)
6918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6920 if (netflow_options) {
6921 if (!ofproto->netflow) {
6922 ofproto->netflow = netflow_create();
6924 return netflow_set_options(ofproto->netflow, netflow_options);
6926 netflow_destroy(ofproto->netflow);
6927 ofproto->netflow = NULL;
6933 get_netflow_ids(const struct ofproto *ofproto_,
6934 uint8_t *engine_type, uint8_t *engine_id)
6936 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6938 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6942 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6944 if (!facet_is_controller_flow(facet) &&
6945 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6946 struct subfacet *subfacet;
6947 struct ofexpired expired;
6949 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6950 if (subfacet->path == SF_FAST_PATH) {
6951 struct dpif_flow_stats stats;
6953 subfacet_reinstall(subfacet, &stats);
6954 subfacet_update_stats(subfacet, &stats);
6958 expired.flow = facet->flow;
6959 expired.packet_count = facet->packet_count;
6960 expired.byte_count = facet->byte_count;
6961 expired.used = facet->used;
6962 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6967 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6969 struct facet *facet;
6971 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6972 send_active_timeout(ofproto, facet);
6976 static struct ofproto_dpif *
6977 ofproto_dpif_lookup(const char *name)
6979 struct ofproto_dpif *ofproto;
6981 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6982 hash_string(name, 0), &all_ofproto_dpifs) {
6983 if (!strcmp(ofproto->up.name, name)) {
6991 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6992 const char *argv[], void *aux OVS_UNUSED)
6994 struct ofproto_dpif *ofproto;
6997 ofproto = ofproto_dpif_lookup(argv[1]);
6999 unixctl_command_reply_error(conn, "no such bridge");
7002 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7004 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7005 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7009 unixctl_command_reply(conn, "table successfully flushed");
7013 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7014 const char *argv[], void *aux OVS_UNUSED)
7016 struct ds ds = DS_EMPTY_INITIALIZER;
7017 const struct ofproto_dpif *ofproto;
7018 const struct mac_entry *e;
7020 ofproto = ofproto_dpif_lookup(argv[1]);
7022 unixctl_command_reply_error(conn, "no such bridge");
7026 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7027 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7028 struct ofbundle *bundle = e->port.p;
7029 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7030 ofbundle_get_a_port(bundle)->odp_port,
7031 e->vlan, ETH_ADDR_ARGS(e->mac),
7032 mac_entry_age(ofproto->ml, e));
7034 unixctl_command_reply(conn, ds_cstr(&ds));
7039 struct action_xlate_ctx ctx;
7045 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7046 const struct rule_dpif *rule)
7048 ds_put_char_multiple(result, '\t', level);
7050 ds_put_cstr(result, "No match\n");
7054 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7055 table_id, ntohll(rule->up.flow_cookie));
7056 cls_rule_format(&rule->up.cr, result);
7057 ds_put_char(result, '\n');
7059 ds_put_char_multiple(result, '\t', level);
7060 ds_put_cstr(result, "OpenFlow ");
7061 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7062 ds_put_char(result, '\n');
7066 trace_format_flow(struct ds *result, int level, const char *title,
7067 struct trace_ctx *trace)
7069 ds_put_char_multiple(result, '\t', level);
7070 ds_put_format(result, "%s: ", title);
7071 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7072 ds_put_cstr(result, "unchanged");
7074 flow_format(result, &trace->ctx.flow);
7075 trace->flow = trace->ctx.flow;
7077 ds_put_char(result, '\n');
7081 trace_format_regs(struct ds *result, int level, const char *title,
7082 struct trace_ctx *trace)
7086 ds_put_char_multiple(result, '\t', level);
7087 ds_put_format(result, "%s:", title);
7088 for (i = 0; i < FLOW_N_REGS; i++) {
7089 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7091 ds_put_char(result, '\n');
7095 trace_format_odp(struct ds *result, int level, const char *title,
7096 struct trace_ctx *trace)
7098 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7100 ds_put_char_multiple(result, '\t', level);
7101 ds_put_format(result, "%s: ", title);
7102 format_odp_actions(result, odp_actions->data, odp_actions->size);
7103 ds_put_char(result, '\n');
7107 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7109 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7110 struct ds *result = trace->result;
7112 ds_put_char(result, '\n');
7113 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7114 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7115 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7116 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7120 trace_report(struct action_xlate_ctx *ctx, const char *s)
7122 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7123 struct ds *result = trace->result;
7125 ds_put_char_multiple(result, '\t', ctx->recurse);
7126 ds_put_cstr(result, s);
7127 ds_put_char(result, '\n');
7131 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7132 void *aux OVS_UNUSED)
7134 const char *dpname = argv[1];
7135 struct ofproto_dpif *ofproto;
7136 struct ofpbuf odp_key;
7137 struct ofpbuf *packet;
7138 ovs_be16 initial_tci;
7144 ofpbuf_init(&odp_key, 0);
7147 ofproto = ofproto_dpif_lookup(dpname);
7149 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7153 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7154 /* ofproto/trace dpname flow [-generate] */
7155 const char *flow_s = argv[2];
7156 const char *generate_s = argv[3];
7158 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7159 * flow. We guess which type it is based on whether 'flow_s' contains
7160 * an '(', since a datapath flow always contains '(') but an
7161 * OpenFlow-like flow should not (in fact it's allowed but I believe
7162 * that's not documented anywhere).
7164 * An alternative would be to try to parse 'flow_s' both ways, but then
7165 * it would be tricky giving a sensible error message. After all, do
7166 * you just say "syntax error" or do you present both error messages?
7167 * Both choices seem lousy. */
7168 if (strchr(flow_s, '(')) {
7169 enum odp_key_fitness fitness;
7172 /* Convert string to datapath key. */
7173 ofpbuf_init(&odp_key, 0);
7174 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7176 unixctl_command_reply_error(conn, "Bad flow syntax");
7180 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7181 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7183 /* Convert odp_key to flow. */
7184 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7185 &initial_tci, NULL);
7186 if (error == ODP_FIT_ERROR) {
7187 unixctl_command_reply_error(conn, "Invalid flow");
7193 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7195 unixctl_command_reply_error(conn, error_s);
7200 initial_tci = flow.vlan_tci;
7201 vsp_adjust_flow(ofproto, &flow);
7204 /* Generate a packet, if requested. */
7206 packet = ofpbuf_new(0);
7207 flow_compose(packet, &flow);
7209 } else if (argc == 6) {
7210 /* ofproto/trace dpname priority tun_id in_port packet */
7211 const char *priority_s = argv[2];
7212 const char *tun_id_s = argv[3];
7213 const char *in_port_s = argv[4];
7214 const char *packet_s = argv[5];
7215 uint32_t in_port = atoi(in_port_s);
7216 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7217 uint32_t priority = atoi(priority_s);
7220 msg = eth_from_hex(packet_s, &packet);
7222 unixctl_command_reply_error(conn, msg);
7226 ds_put_cstr(&result, "Packet: ");
7227 s = ofp_packet_to_string(packet->data, packet->size);
7228 ds_put_cstr(&result, s);
7231 flow_extract(packet, priority, NULL, in_port, &flow);
7232 flow.tunnel.tun_id = tun_id;
7233 initial_tci = flow.vlan_tci;
7235 unixctl_command_reply_error(conn, "Bad command syntax");
7239 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7240 unixctl_command_reply(conn, ds_cstr(&result));
7243 ds_destroy(&result);
7244 ofpbuf_delete(packet);
7245 ofpbuf_uninit(&odp_key);
7249 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7250 const struct ofpbuf *packet, ovs_be16 initial_tci,
7253 struct rule_dpif *rule;
7255 ds_put_cstr(ds, "Flow: ");
7256 flow_format(ds, flow);
7257 ds_put_char(ds, '\n');
7259 rule = rule_dpif_lookup(ofproto, flow);
7261 trace_format_rule(ds, 0, 0, rule);
7262 if (rule == ofproto->miss_rule) {
7263 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7264 } else if (rule == ofproto->no_packet_in_rule) {
7265 ds_put_cstr(ds, "\nNo match, packets dropped because "
7266 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7270 uint64_t odp_actions_stub[1024 / 8];
7271 struct ofpbuf odp_actions;
7273 struct trace_ctx trace;
7276 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7279 ofpbuf_use_stub(&odp_actions,
7280 odp_actions_stub, sizeof odp_actions_stub);
7281 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7282 rule, tcp_flags, packet);
7283 trace.ctx.resubmit_hook = trace_resubmit;
7284 trace.ctx.report_hook = trace_report;
7285 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7288 ds_put_char(ds, '\n');
7289 trace_format_flow(ds, 0, "Final flow", &trace);
7290 ds_put_cstr(ds, "Datapath actions: ");
7291 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7292 ofpbuf_uninit(&odp_actions);
7294 if (trace.ctx.slow) {
7295 enum slow_path_reason slow;
7297 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7298 "slow path because it:");
7299 for (slow = trace.ctx.slow; slow; ) {
7300 enum slow_path_reason bit = rightmost_1bit(slow);
7304 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7307 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7310 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7313 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7316 ds_put_cstr(ds, "\n\t (The datapath actions are "
7317 "incomplete--for complete actions, "
7318 "please supply a packet.)");
7321 case SLOW_CONTROLLER:
7322 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7323 "to the OpenFlow controller.");
7326 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7327 "than the datapath supports.");
7334 if (slow & ~SLOW_MATCH) {
7335 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7336 "the special slow-path processing.");
7343 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7344 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7347 unixctl_command_reply(conn, NULL);
7351 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7352 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7355 unixctl_command_reply(conn, NULL);
7358 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7359 * 'reply' describing the results. */
7361 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7363 struct facet *facet;
7367 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7368 if (!facet_check_consistency(facet)) {
7373 ofproto->need_revalidate = REV_INCONSISTENCY;
7377 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7378 ofproto->up.name, errors);
7380 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7385 ofproto_dpif_self_check(struct unixctl_conn *conn,
7386 int argc, const char *argv[], void *aux OVS_UNUSED)
7388 struct ds reply = DS_EMPTY_INITIALIZER;
7389 struct ofproto_dpif *ofproto;
7392 ofproto = ofproto_dpif_lookup(argv[1]);
7394 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7395 "ofproto/list for help)");
7398 ofproto_dpif_self_check__(ofproto, &reply);
7400 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7401 ofproto_dpif_self_check__(ofproto, &reply);
7405 unixctl_command_reply(conn, ds_cstr(&reply));
7409 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7410 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7411 * to destroy 'ofproto_shash' and free the returned value. */
7412 static const struct shash_node **
7413 get_ofprotos(struct shash *ofproto_shash)
7415 const struct ofproto_dpif *ofproto;
7417 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7418 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7419 shash_add_nocopy(ofproto_shash, name, ofproto);
7422 return shash_sort(ofproto_shash);
7426 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7427 const char *argv[] OVS_UNUSED,
7428 void *aux OVS_UNUSED)
7430 struct ds ds = DS_EMPTY_INITIALIZER;
7431 struct shash ofproto_shash;
7432 const struct shash_node **sorted_ofprotos;
7435 shash_init(&ofproto_shash);
7436 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7437 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7438 const struct shash_node *node = sorted_ofprotos[i];
7439 ds_put_format(&ds, "%s\n", node->name);
7442 shash_destroy(&ofproto_shash);
7443 free(sorted_ofprotos);
7445 unixctl_command_reply(conn, ds_cstr(&ds));
7450 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7452 struct dpif_dp_stats s;
7453 const struct shash_node **ports;
7456 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7458 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7459 dpif_name(ofproto->backer->dpif));
7460 /* xxx It would be better to show bridge-specific stats instead
7461 * xxx of dp ones. */
7463 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7464 s.n_hit, s.n_missed, s.n_lost);
7465 ds_put_format(ds, "\tflows: %zu\n",
7466 hmap_count(&ofproto->subfacets));
7468 ports = shash_sort(&ofproto->up.port_by_name);
7469 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7470 const struct shash_node *node = ports[i];
7471 struct ofport *ofport = node->data;
7472 const char *name = netdev_get_name(ofport->netdev);
7473 const char *type = netdev_get_type(ofport->netdev);
7475 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7476 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7477 if (strcmp(type, "system")) {
7478 struct netdev *netdev;
7481 ds_put_format(ds, " (%s", type);
7483 error = netdev_open(name, type, &netdev);
7488 error = netdev_get_config(netdev, &config);
7490 const struct smap_node **nodes;
7493 nodes = smap_sort(&config);
7494 for (i = 0; i < smap_count(&config); i++) {
7495 const struct smap_node *node = nodes[i];
7496 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7497 node->key, node->value);
7501 smap_destroy(&config);
7503 netdev_close(netdev);
7505 ds_put_char(ds, ')');
7507 ds_put_char(ds, '\n');
7513 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7514 const char *argv[], void *aux OVS_UNUSED)
7516 struct ds ds = DS_EMPTY_INITIALIZER;
7517 const struct ofproto_dpif *ofproto;
7521 for (i = 1; i < argc; i++) {
7522 ofproto = ofproto_dpif_lookup(argv[i]);
7524 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7525 "for help)", argv[i]);
7526 unixctl_command_reply_error(conn, ds_cstr(&ds));
7529 show_dp_format(ofproto, &ds);
7532 struct shash ofproto_shash;
7533 const struct shash_node **sorted_ofprotos;
7536 shash_init(&ofproto_shash);
7537 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7538 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7539 const struct shash_node *node = sorted_ofprotos[i];
7540 show_dp_format(node->data, &ds);
7543 shash_destroy(&ofproto_shash);
7544 free(sorted_ofprotos);
7547 unixctl_command_reply(conn, ds_cstr(&ds));
7552 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7553 int argc OVS_UNUSED, const char *argv[],
7554 void *aux OVS_UNUSED)
7556 struct ds ds = DS_EMPTY_INITIALIZER;
7557 const struct ofproto_dpif *ofproto;
7558 struct subfacet *subfacet;
7560 ofproto = ofproto_dpif_lookup(argv[1]);
7562 unixctl_command_reply_error(conn, "no such bridge");
7566 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7567 struct odputil_keybuf keybuf;
7570 subfacet_get_key(subfacet, &keybuf, &key);
7571 odp_flow_key_format(key.data, key.size, &ds);
7573 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7574 subfacet->dp_packet_count, subfacet->dp_byte_count);
7575 if (subfacet->used) {
7576 ds_put_format(&ds, "%.3fs",
7577 (time_msec() - subfacet->used) / 1000.0);
7579 ds_put_format(&ds, "never");
7581 if (subfacet->facet->tcp_flags) {
7582 ds_put_cstr(&ds, ", flags:");
7583 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7586 ds_put_cstr(&ds, ", actions:");
7587 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7588 ds_put_char(&ds, '\n');
7591 unixctl_command_reply(conn, ds_cstr(&ds));
7596 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7597 int argc OVS_UNUSED, const char *argv[],
7598 void *aux OVS_UNUSED)
7600 struct ds ds = DS_EMPTY_INITIALIZER;
7601 struct ofproto_dpif *ofproto;
7603 ofproto = ofproto_dpif_lookup(argv[1]);
7605 unixctl_command_reply_error(conn, "no such bridge");
7609 flush(&ofproto->up);
7611 unixctl_command_reply(conn, ds_cstr(&ds));
7616 ofproto_dpif_unixctl_init(void)
7618 static bool registered;
7624 unixctl_command_register(
7626 "bridge {tun_id in_port packet | odp_flow [-generate]}",
7627 2, 5, ofproto_unixctl_trace, NULL);
7628 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7629 ofproto_unixctl_fdb_flush, NULL);
7630 unixctl_command_register("fdb/show", "bridge", 1, 1,
7631 ofproto_unixctl_fdb_show, NULL);
7632 unixctl_command_register("ofproto/clog", "", 0, 0,
7633 ofproto_dpif_clog, NULL);
7634 unixctl_command_register("ofproto/unclog", "", 0, 0,
7635 ofproto_dpif_unclog, NULL);
7636 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7637 ofproto_dpif_self_check, NULL);
7638 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7639 ofproto_unixctl_dpif_dump_dps, NULL);
7640 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7641 ofproto_unixctl_dpif_show, NULL);
7642 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7643 ofproto_unixctl_dpif_dump_flows, NULL);
7644 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7645 ofproto_unixctl_dpif_del_flows, NULL);
7648 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7650 * This is deprecated. It is only for compatibility with broken device drivers
7651 * in old versions of Linux that do not properly support VLANs when VLAN
7652 * devices are not used. When broken device drivers are no longer in
7653 * widespread use, we will delete these interfaces. */
7656 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7659 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7661 if (realdev_ofp_port == ofport->realdev_ofp_port
7662 && vid == ofport->vlandev_vid) {
7666 ofproto->need_revalidate = REV_RECONFIGURE;
7668 if (ofport->realdev_ofp_port) {
7671 if (realdev_ofp_port && ofport->bundle) {
7672 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7673 * themselves be part of a bundle. */
7674 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7677 ofport->realdev_ofp_port = realdev_ofp_port;
7678 ofport->vlandev_vid = vid;
7680 if (realdev_ofp_port) {
7681 vsp_add(ofport, realdev_ofp_port, vid);
7688 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7690 return hash_2words(realdev_ofp_port, vid);
7693 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7694 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7695 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7696 * it would return the port number of eth0.9.
7698 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7699 * function just returns its 'realdev_odp_port' argument. */
7701 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7702 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7704 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7705 uint16_t realdev_ofp_port;
7706 int vid = vlan_tci_to_vid(vlan_tci);
7707 const struct vlan_splinter *vsp;
7709 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7710 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7711 hash_realdev_vid(realdev_ofp_port, vid),
7712 &ofproto->realdev_vid_map) {
7713 if (vsp->realdev_ofp_port == realdev_ofp_port
7714 && vsp->vid == vid) {
7715 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7719 return realdev_odp_port;
7722 static struct vlan_splinter *
7723 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7725 struct vlan_splinter *vsp;
7727 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7728 &ofproto->vlandev_map) {
7729 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7737 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7738 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7739 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7740 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7741 * eth0 and store 9 in '*vid'.
7743 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7744 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7747 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7748 uint16_t vlandev_ofp_port, int *vid)
7750 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7751 const struct vlan_splinter *vsp;
7753 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7758 return vsp->realdev_ofp_port;
7764 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7765 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7766 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7767 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7768 * always the case unless VLAN splinters are enabled), returns false without
7769 * making any changes. */
7771 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7776 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7781 /* Cause the flow to be processed as if it came in on the real device with
7782 * the VLAN device's VLAN ID. */
7783 flow->in_port = realdev;
7784 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7789 vsp_remove(struct ofport_dpif *port)
7791 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7792 struct vlan_splinter *vsp;
7794 vsp = vlandev_find(ofproto, port->up.ofp_port);
7796 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7797 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7800 port->realdev_ofp_port = 0;
7802 VLOG_ERR("missing vlan device record");
7807 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7811 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7812 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7813 == realdev_ofp_port)) {
7814 struct vlan_splinter *vsp;
7816 vsp = xmalloc(sizeof *vsp);
7817 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7818 hash_int(port->up.ofp_port, 0));
7819 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7820 hash_realdev_vid(realdev_ofp_port, vid));
7821 vsp->realdev_ofp_port = realdev_ofp_port;
7822 vsp->vlandev_ofp_port = port->up.ofp_port;
7825 port->realdev_ofp_port = realdev_ofp_port;
7827 VLOG_ERR("duplicate vlan device record");
7832 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7834 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7835 return ofport ? ofport->odp_port : OVSP_NONE;
7838 static struct ofport_dpif *
7839 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7841 struct ofport_dpif *port;
7843 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7844 hash_int(odp_port, 0),
7845 &backer->odp_to_ofport_map) {
7846 if (port->odp_port == odp_port) {
7855 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7857 struct ofport_dpif *port;
7859 port = odp_port_to_ofport(ofproto->backer, odp_port);
7860 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7861 return port->up.ofp_port;
7867 const struct ofproto_class ofproto_dpif_class = {
7902 port_is_lacp_current,
7903 NULL, /* rule_choose_table */
7910 rule_modify_actions,
7919 get_cfm_remote_mpids,
7924 get_stp_port_status,
7931 is_mirror_output_bundle,
7932 forward_bpdu_changed,