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);
783 /* Type functions. */
786 type_run(const char *type)
788 struct dpif_backer *backer;
792 backer = shash_find_data(&all_dpif_backers, type);
794 /* This is not necessarily a problem, since backers are only
795 * created on demand. */
799 dpif_run(backer->dpif);
801 if (timer_expired(&backer->next_expiration)) {
802 int delay = expire(backer);
803 timer_set_duration(&backer->next_expiration, delay);
806 /* Check for port changes in the dpif. */
807 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
808 struct ofproto_dpif *ofproto = NULL;
809 struct dpif_port port;
811 /* Don't report on the datapath's device. */
812 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
816 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
817 &all_ofproto_dpifs) {
818 if (sset_contains(&ofproto->ports, devname)) {
823 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
824 /* The port was removed. If we know the datapath,
825 * report it through poll_set(). If we don't, it may be
826 * notifying us of a removal we initiated, so ignore it.
827 * If there's a pending ENOBUFS, let it stand, since
828 * everything will be reevaluated. */
829 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
830 sset_add(&ofproto->port_poll_set, devname);
831 ofproto->port_poll_errno = 0;
833 dpif_port_destroy(&port);
834 } else if (!ofproto) {
835 /* The port was added, but we don't know with which
836 * ofproto we should associate it. Delete it. */
837 dpif_port_del(backer->dpif, port.port_no);
843 if (error != EAGAIN) {
844 struct ofproto_dpif *ofproto;
846 /* There was some sort of error, so propagate it to all
847 * ofprotos that use this backer. */
848 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
849 &all_ofproto_dpifs) {
850 if (ofproto->backer == backer) {
851 sset_clear(&ofproto->port_poll_set);
852 ofproto->port_poll_errno = error;
861 type_run_fast(const char *type)
863 struct dpif_backer *backer;
866 backer = shash_find_data(&all_dpif_backers, type);
868 /* This is not necessarily a problem, since backers are only
869 * created on demand. */
873 /* Handle one or more batches of upcalls, until there's nothing left to do
874 * or until we do a fixed total amount of work.
876 * We do work in batches because it can be much cheaper to set up a number
877 * of flows and fire off their patches all at once. We do multiple batches
878 * because in some cases handling a packet can cause another packet to be
879 * queued almost immediately as part of the return flow. Both
880 * optimizations can make major improvements on some benchmarks and
881 * presumably for real traffic as well. */
883 while (work < FLOW_MISS_MAX_BATCH) {
884 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
895 type_wait(const char *type)
897 struct dpif_backer *backer;
899 backer = shash_find_data(&all_dpif_backers, type);
901 /* This is not necessarily a problem, since backers are only
902 * created on demand. */
906 timer_wait(&backer->next_expiration);
909 /* Basic life-cycle. */
911 static int add_internal_flows(struct ofproto_dpif *);
913 static struct ofproto *
916 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
921 dealloc(struct ofproto *ofproto_)
923 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
928 close_dpif_backer(struct dpif_backer *backer)
930 struct shash_node *node;
932 assert(backer->refcount > 0);
934 if (--backer->refcount) {
938 hmap_destroy(&backer->odp_to_ofport_map);
939 node = shash_find(&all_dpif_backers, backer->type);
941 shash_delete(&all_dpif_backers, node);
942 dpif_close(backer->dpif);
947 /* Datapath port slated for removal from datapath. */
949 struct list list_node;
954 open_dpif_backer(const char *type, struct dpif_backer **backerp)
956 struct dpif_backer *backer;
957 struct dpif_port_dump port_dump;
958 struct dpif_port port;
959 struct shash_node *node;
960 struct list garbage_list;
961 struct odp_garbage *garbage, *next;
967 backer = shash_find_data(&all_dpif_backers, type);
974 backer_name = xasprintf("ovs-%s", type);
976 /* Remove any existing datapaths, since we assume we're the only
977 * userspace controlling the datapath. */
979 dp_enumerate_names(type, &names);
980 SSET_FOR_EACH(name, &names) {
981 struct dpif *old_dpif;
983 /* Don't remove our backer if it exists. */
984 if (!strcmp(name, backer_name)) {
988 if (dpif_open(name, type, &old_dpif)) {
989 VLOG_WARN("couldn't open old datapath %s to remove it", name);
991 dpif_delete(old_dpif);
992 dpif_close(old_dpif);
995 sset_destroy(&names);
997 backer = xmalloc(sizeof *backer);
999 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1002 VLOG_ERR("failed to open datapath of type %s: %s", type,
1007 backer->type = xstrdup(type);
1008 backer->refcount = 1;
1009 hmap_init(&backer->odp_to_ofport_map);
1010 timer_set_duration(&backer->next_expiration, 1000);
1013 dpif_flow_flush(backer->dpif);
1015 /* Loop through the ports already on the datapath and remove any
1016 * that we don't need anymore. */
1017 list_init(&garbage_list);
1018 dpif_port_dump_start(&port_dump, backer->dpif);
1019 while (dpif_port_dump_next(&port_dump, &port)) {
1020 node = shash_find(&init_ofp_ports, port.name);
1021 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1022 garbage = xmalloc(sizeof *garbage);
1023 garbage->odp_port = port.port_no;
1024 list_push_front(&garbage_list, &garbage->list_node);
1027 dpif_port_dump_done(&port_dump);
1029 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1030 dpif_port_del(backer->dpif, garbage->odp_port);
1031 list_remove(&garbage->list_node);
1035 shash_add(&all_dpif_backers, type, backer);
1037 error = dpif_recv_set(backer->dpif, true);
1039 VLOG_ERR("failed to listen on datapath of type %s: %s",
1040 type, strerror(error));
1041 close_dpif_backer(backer);
1049 construct(struct ofproto *ofproto_)
1051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1052 struct shash_node *node, *next;
1057 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1062 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1063 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1065 ofproto->n_matches = 0;
1067 ofproto->netflow = NULL;
1068 ofproto->sflow = NULL;
1069 ofproto->stp = NULL;
1070 hmap_init(&ofproto->bundles);
1071 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1072 for (i = 0; i < MAX_MIRRORS; i++) {
1073 ofproto->mirrors[i] = NULL;
1075 ofproto->has_bonded_bundles = false;
1077 hmap_init(&ofproto->facets);
1078 hmap_init(&ofproto->subfacets);
1079 ofproto->governor = NULL;
1081 for (i = 0; i < N_TABLES; i++) {
1082 struct table_dpif *table = &ofproto->tables[i];
1084 table->catchall_table = NULL;
1085 table->other_table = NULL;
1086 table->basis = random_uint32();
1088 ofproto->need_revalidate = 0;
1089 tag_set_init(&ofproto->revalidate_set);
1091 list_init(&ofproto->completions);
1093 ofproto_dpif_unixctl_init();
1095 ofproto->has_mirrors = false;
1096 ofproto->has_bundle_action = false;
1098 hmap_init(&ofproto->vlandev_map);
1099 hmap_init(&ofproto->realdev_vid_map);
1101 sset_init(&ofproto->ports);
1102 sset_init(&ofproto->port_poll_set);
1103 ofproto->port_poll_errno = 0;
1105 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1106 const struct iface_hint *iface_hint = node->data;
1108 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1109 /* Check if the datapath already has this port. */
1110 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1111 sset_add(&ofproto->ports, node->name);
1114 free(iface_hint->br_name);
1115 free(iface_hint->br_type);
1116 shash_delete(&init_ofp_ports, node);
1120 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1121 hash_string(ofproto->up.name, 0));
1122 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1124 ofproto_init_tables(ofproto_, N_TABLES);
1125 error = add_internal_flows(ofproto);
1126 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1132 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1133 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1135 struct ofputil_flow_mod fm;
1138 match_init_catchall(&fm.match);
1140 match_set_reg(&fm.match, 0, id);
1141 fm.new_cookie = htonll(0);
1142 fm.cookie = htonll(0);
1143 fm.cookie_mask = htonll(0);
1144 fm.table_id = TBL_INTERNAL;
1145 fm.command = OFPFC_ADD;
1146 fm.idle_timeout = 0;
1147 fm.hard_timeout = 0;
1151 fm.ofpacts = ofpacts->data;
1152 fm.ofpacts_len = ofpacts->size;
1154 error = ofproto_flow_mod(&ofproto->up, &fm);
1156 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1157 id, ofperr_to_string(error));
1161 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1162 assert(*rulep != NULL);
1168 add_internal_flows(struct ofproto_dpif *ofproto)
1170 struct ofpact_controller *controller;
1171 uint64_t ofpacts_stub[128 / 8];
1172 struct ofpbuf ofpacts;
1176 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1179 controller = ofpact_put_CONTROLLER(&ofpacts);
1180 controller->max_len = UINT16_MAX;
1181 controller->controller_id = 0;
1182 controller->reason = OFPR_NO_MATCH;
1183 ofpact_pad(&ofpacts);
1185 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1190 ofpbuf_clear(&ofpacts);
1191 error = add_internal_flow(ofproto, id++, &ofpacts,
1192 &ofproto->no_packet_in_rule);
1197 complete_operations(struct ofproto_dpif *ofproto)
1199 struct dpif_completion *c, *next;
1201 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1202 ofoperation_complete(c->op, 0);
1203 list_remove(&c->list_node);
1209 destruct(struct ofproto *ofproto_)
1211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1212 struct rule_dpif *rule, *next_rule;
1213 struct oftable *table;
1216 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1217 complete_operations(ofproto);
1219 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1220 struct cls_cursor cursor;
1222 cls_cursor_init(&cursor, &table->cls, NULL);
1223 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1224 ofproto_rule_destroy(&rule->up);
1228 for (i = 0; i < MAX_MIRRORS; i++) {
1229 mirror_destroy(ofproto->mirrors[i]);
1232 netflow_destroy(ofproto->netflow);
1233 dpif_sflow_destroy(ofproto->sflow);
1234 hmap_destroy(&ofproto->bundles);
1235 mac_learning_destroy(ofproto->ml);
1237 hmap_destroy(&ofproto->facets);
1238 hmap_destroy(&ofproto->subfacets);
1239 governor_destroy(ofproto->governor);
1241 hmap_destroy(&ofproto->vlandev_map);
1242 hmap_destroy(&ofproto->realdev_vid_map);
1244 sset_destroy(&ofproto->ports);
1245 sset_destroy(&ofproto->port_poll_set);
1247 close_dpif_backer(ofproto->backer);
1251 run_fast(struct ofproto *ofproto_)
1253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1254 struct ofport_dpif *ofport;
1256 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1257 port_run_fast(ofport);
1264 run(struct ofproto *ofproto_)
1266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1267 struct ofport_dpif *ofport;
1268 struct ofbundle *bundle;
1272 complete_operations(ofproto);
1275 error = run_fast(ofproto_);
1280 if (ofproto->netflow) {
1281 if (netflow_run(ofproto->netflow)) {
1282 send_netflow_active_timeouts(ofproto);
1285 if (ofproto->sflow) {
1286 dpif_sflow_run(ofproto->sflow);
1289 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1292 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1297 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
1299 /* Now revalidate if there's anything to do. */
1300 if (ofproto->need_revalidate
1301 || !tag_set_is_empty(&ofproto->revalidate_set)) {
1302 struct tag_set revalidate_set = ofproto->revalidate_set;
1303 bool revalidate_all = ofproto->need_revalidate;
1304 struct facet *facet;
1306 switch (ofproto->need_revalidate) {
1307 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1308 case REV_STP: COVERAGE_INC(rev_stp); break;
1309 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1310 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1311 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1314 /* Clear the revalidation flags. */
1315 tag_set_init(&ofproto->revalidate_set);
1316 ofproto->need_revalidate = 0;
1318 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1320 || tag_set_intersects(&revalidate_set, facet->tags)) {
1321 facet_revalidate(facet);
1326 /* Check the consistency of a random facet, to aid debugging. */
1327 if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) {
1328 struct facet *facet;
1330 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1331 struct facet, hmap_node);
1332 if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) {
1333 if (!facet_check_consistency(facet)) {
1334 ofproto->need_revalidate = REV_INCONSISTENCY;
1339 if (ofproto->governor) {
1342 governor_run(ofproto->governor);
1344 /* If the governor has shrunk to its minimum size and the number of
1345 * subfacets has dwindled, then drop the governor entirely.
1347 * For hysteresis, the number of subfacets to drop the governor is
1348 * smaller than the number needed to trigger its creation. */
1349 n_subfacets = hmap_count(&ofproto->subfacets);
1350 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1351 && governor_is_idle(ofproto->governor)) {
1352 governor_destroy(ofproto->governor);
1353 ofproto->governor = NULL;
1361 wait(struct ofproto *ofproto_)
1363 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1364 struct ofport_dpif *ofport;
1365 struct ofbundle *bundle;
1367 if (!clogged && !list_is_empty(&ofproto->completions)) {
1368 poll_immediate_wake();
1371 dpif_wait(ofproto->backer->dpif);
1372 dpif_recv_wait(ofproto->backer->dpif);
1373 if (ofproto->sflow) {
1374 dpif_sflow_wait(ofproto->sflow);
1376 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
1377 poll_immediate_wake();
1379 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1382 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1383 bundle_wait(bundle);
1385 if (ofproto->netflow) {
1386 netflow_wait(ofproto->netflow);
1388 mac_learning_wait(ofproto->ml);
1390 if (ofproto->need_revalidate) {
1391 /* Shouldn't happen, but if it does just go around again. */
1392 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1393 poll_immediate_wake();
1395 if (ofproto->governor) {
1396 governor_wait(ofproto->governor);
1401 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1403 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1405 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1406 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1410 flush(struct ofproto *ofproto_)
1412 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1413 struct subfacet *subfacet, *next_subfacet;
1414 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1418 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1419 &ofproto->subfacets) {
1420 if (subfacet->path != SF_NOT_INSTALLED) {
1421 batch[n_batch++] = subfacet;
1422 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1423 subfacet_destroy_batch(ofproto, batch, n_batch);
1427 subfacet_destroy(subfacet);
1432 subfacet_destroy_batch(ofproto, batch, n_batch);
1437 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1438 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1440 *arp_match_ip = true;
1441 *actions = (OFPUTIL_A_OUTPUT |
1442 OFPUTIL_A_SET_VLAN_VID |
1443 OFPUTIL_A_SET_VLAN_PCP |
1444 OFPUTIL_A_STRIP_VLAN |
1445 OFPUTIL_A_SET_DL_SRC |
1446 OFPUTIL_A_SET_DL_DST |
1447 OFPUTIL_A_SET_NW_SRC |
1448 OFPUTIL_A_SET_NW_DST |
1449 OFPUTIL_A_SET_NW_TOS |
1450 OFPUTIL_A_SET_TP_SRC |
1451 OFPUTIL_A_SET_TP_DST |
1456 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1459 struct dpif_dp_stats s;
1461 strcpy(ots->name, "classifier");
1463 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1465 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1466 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1469 static struct ofport *
1472 struct ofport_dpif *port = xmalloc(sizeof *port);
1477 port_dealloc(struct ofport *port_)
1479 struct ofport_dpif *port = ofport_dpif_cast(port_);
1484 port_construct(struct ofport *port_)
1486 struct ofport_dpif *port = ofport_dpif_cast(port_);
1487 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1488 struct dpif_port dpif_port;
1491 ofproto->need_revalidate = REV_RECONFIGURE;
1492 port->bundle = NULL;
1494 port->tag = tag_create_random();
1495 port->may_enable = true;
1496 port->stp_port = NULL;
1497 port->stp_state = STP_DISABLED;
1498 hmap_init(&port->priorities);
1499 port->realdev_ofp_port = 0;
1500 port->vlandev_vid = 0;
1501 port->carrier_seq = netdev_get_carrier_resets(port->up.netdev);
1503 error = dpif_port_query_by_name(ofproto->backer->dpif,
1504 netdev_get_name(port->up.netdev),
1510 port->odp_port = dpif_port.port_no;
1512 /* Sanity-check that a mapping doesn't already exist. This
1513 * shouldn't happen. */
1514 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1515 VLOG_ERR("port %s already has an OpenFlow port number\n",
1520 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1521 hash_int(port->odp_port, 0));
1523 if (ofproto->sflow) {
1524 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1531 port_destruct(struct ofport *port_)
1533 struct ofport_dpif *port = ofport_dpif_cast(port_);
1534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1535 struct dpif_port dpif_port;
1537 if (!dpif_port_query_by_number(ofproto->backer->dpif,
1538 port->odp_port, &dpif_port)) {
1539 /* The underlying device is still there, so delete it. This
1540 * happens when the ofproto is being destroyed, since the caller
1541 * assumes that removal of attached ports will happen as part of
1543 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1544 dpif_port_destroy(&dpif_port);
1547 sset_find_and_delete(&ofproto->ports, netdev_get_name(port->up.netdev));
1548 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1549 ofproto->need_revalidate = REV_RECONFIGURE;
1550 bundle_remove(port_);
1551 set_cfm(port_, NULL);
1552 if (ofproto->sflow) {
1553 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1556 ofport_clear_priorities(port);
1557 hmap_destroy(&port->priorities);
1561 port_modified(struct ofport *port_)
1563 struct ofport_dpif *port = ofport_dpif_cast(port_);
1565 if (port->bundle && port->bundle->bond) {
1566 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1571 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1573 struct ofport_dpif *port = ofport_dpif_cast(port_);
1574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1575 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1577 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1578 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1579 OFPUTIL_PC_NO_PACKET_IN)) {
1580 ofproto->need_revalidate = REV_RECONFIGURE;
1582 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1583 bundle_update(port->bundle);
1589 set_sflow(struct ofproto *ofproto_,
1590 const struct ofproto_sflow_options *sflow_options)
1592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1593 struct dpif_sflow *ds = ofproto->sflow;
1595 if (sflow_options) {
1597 struct ofport_dpif *ofport;
1599 ds = ofproto->sflow = dpif_sflow_create();
1600 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1601 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1603 ofproto->need_revalidate = REV_RECONFIGURE;
1605 dpif_sflow_set_options(ds, sflow_options);
1608 dpif_sflow_destroy(ds);
1609 ofproto->need_revalidate = REV_RECONFIGURE;
1610 ofproto->sflow = NULL;
1617 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1619 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1626 struct ofproto_dpif *ofproto;
1628 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1629 ofproto->need_revalidate = REV_RECONFIGURE;
1630 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1633 if (cfm_configure(ofport->cfm, s)) {
1639 cfm_destroy(ofport->cfm);
1645 get_cfm_fault(const struct ofport *ofport_)
1647 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1649 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1653 get_cfm_opup(const struct ofport *ofport_)
1655 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1657 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1661 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1664 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1667 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1675 get_cfm_health(const struct ofport *ofport_)
1677 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1679 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1682 /* Spanning Tree. */
1685 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1687 struct ofproto_dpif *ofproto = ofproto_;
1688 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1689 struct ofport_dpif *ofport;
1691 ofport = stp_port_get_aux(sp);
1693 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1694 ofproto->up.name, port_num);
1696 struct eth_header *eth = pkt->l2;
1698 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1699 if (eth_addr_is_zero(eth->eth_src)) {
1700 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1701 "with unknown MAC", ofproto->up.name, port_num);
1703 send_packet(ofport, pkt);
1709 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1711 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1715 /* Only revalidate flows if the configuration changed. */
1716 if (!s != !ofproto->stp) {
1717 ofproto->need_revalidate = REV_RECONFIGURE;
1721 if (!ofproto->stp) {
1722 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1723 send_bpdu_cb, ofproto);
1724 ofproto->stp_last_tick = time_msec();
1727 stp_set_bridge_id(ofproto->stp, s->system_id);
1728 stp_set_bridge_priority(ofproto->stp, s->priority);
1729 stp_set_hello_time(ofproto->stp, s->hello_time);
1730 stp_set_max_age(ofproto->stp, s->max_age);
1731 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1733 struct ofport *ofport;
1735 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1736 set_stp_port(ofport, NULL);
1739 stp_destroy(ofproto->stp);
1740 ofproto->stp = NULL;
1747 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1753 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1754 s->designated_root = stp_get_designated_root(ofproto->stp);
1755 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1764 update_stp_port_state(struct ofport_dpif *ofport)
1766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1767 enum stp_state state;
1769 /* Figure out new state. */
1770 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1774 if (ofport->stp_state != state) {
1775 enum ofputil_port_state of_state;
1778 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1779 netdev_get_name(ofport->up.netdev),
1780 stp_state_name(ofport->stp_state),
1781 stp_state_name(state));
1782 if (stp_learn_in_state(ofport->stp_state)
1783 != stp_learn_in_state(state)) {
1784 /* xxx Learning action flows should also be flushed. */
1785 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1787 fwd_change = stp_forward_in_state(ofport->stp_state)
1788 != stp_forward_in_state(state);
1790 ofproto->need_revalidate = REV_STP;
1791 ofport->stp_state = state;
1792 ofport->stp_state_entered = time_msec();
1794 if (fwd_change && ofport->bundle) {
1795 bundle_update(ofport->bundle);
1798 /* Update the STP state bits in the OpenFlow port description. */
1799 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1800 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1801 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1802 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1803 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1805 ofproto_port_set_state(&ofport->up, of_state);
1809 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1810 * caller is responsible for assigning STP port numbers and ensuring
1811 * there are no duplicates. */
1813 set_stp_port(struct ofport *ofport_,
1814 const struct ofproto_port_stp_settings *s)
1816 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1818 struct stp_port *sp = ofport->stp_port;
1820 if (!s || !s->enable) {
1822 ofport->stp_port = NULL;
1823 stp_port_disable(sp);
1824 update_stp_port_state(ofport);
1827 } else if (sp && stp_port_no(sp) != s->port_num
1828 && ofport == stp_port_get_aux(sp)) {
1829 /* The port-id changed, so disable the old one if it's not
1830 * already in use by another port. */
1831 stp_port_disable(sp);
1834 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1835 stp_port_enable(sp);
1837 stp_port_set_aux(sp, ofport);
1838 stp_port_set_priority(sp, s->priority);
1839 stp_port_set_path_cost(sp, s->path_cost);
1841 update_stp_port_state(ofport);
1847 get_stp_port_status(struct ofport *ofport_,
1848 struct ofproto_port_stp_status *s)
1850 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1852 struct stp_port *sp = ofport->stp_port;
1854 if (!ofproto->stp || !sp) {
1860 s->port_id = stp_port_get_id(sp);
1861 s->state = stp_port_get_state(sp);
1862 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1863 s->role = stp_port_get_role(sp);
1864 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1870 stp_run(struct ofproto_dpif *ofproto)
1873 long long int now = time_msec();
1874 long long int elapsed = now - ofproto->stp_last_tick;
1875 struct stp_port *sp;
1878 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1879 ofproto->stp_last_tick = now;
1881 while (stp_get_changed_port(ofproto->stp, &sp)) {
1882 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1885 update_stp_port_state(ofport);
1889 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1890 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
1896 stp_wait(struct ofproto_dpif *ofproto)
1899 poll_timer_wait(1000);
1903 /* Returns true if STP should process 'flow'. */
1905 stp_should_process_flow(const struct flow *flow)
1907 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1911 stp_process_packet(const struct ofport_dpif *ofport,
1912 const struct ofpbuf *packet)
1914 struct ofpbuf payload = *packet;
1915 struct eth_header *eth = payload.data;
1916 struct stp_port *sp = ofport->stp_port;
1918 /* Sink packets on ports that have STP disabled when the bridge has
1920 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1924 /* Trim off padding on payload. */
1925 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1926 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1929 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1930 stp_received_bpdu(sp, payload.data, payload.size);
1934 static struct priority_to_dscp *
1935 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
1937 struct priority_to_dscp *pdscp;
1940 hash = hash_int(priority, 0);
1941 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
1942 if (pdscp->priority == priority) {
1950 ofport_clear_priorities(struct ofport_dpif *ofport)
1952 struct priority_to_dscp *pdscp, *next;
1954 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
1955 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1961 set_queues(struct ofport *ofport_,
1962 const struct ofproto_port_queue *qdscp_list,
1965 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1967 struct hmap new = HMAP_INITIALIZER(&new);
1970 for (i = 0; i < n_qdscp; i++) {
1971 struct priority_to_dscp *pdscp;
1975 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1976 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
1981 pdscp = get_priority(ofport, priority);
1983 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
1985 pdscp = xmalloc(sizeof *pdscp);
1986 pdscp->priority = priority;
1988 ofproto->need_revalidate = REV_RECONFIGURE;
1991 if (pdscp->dscp != dscp) {
1993 ofproto->need_revalidate = REV_RECONFIGURE;
1996 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
1999 if (!hmap_is_empty(&ofport->priorities)) {
2000 ofport_clear_priorities(ofport);
2001 ofproto->need_revalidate = REV_RECONFIGURE;
2004 hmap_swap(&new, &ofport->priorities);
2012 /* Expires all MAC learning entries associated with 'bundle' and forces its
2013 * ofproto to revalidate every flow.
2015 * Normally MAC learning entries are removed only from the ofproto associated
2016 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2017 * are removed from every ofproto. When patch ports and SLB bonds are in use
2018 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2019 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2020 * with the host from which it migrated. */
2022 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2024 struct ofproto_dpif *ofproto = bundle->ofproto;
2025 struct mac_learning *ml = ofproto->ml;
2026 struct mac_entry *mac, *next_mac;
2028 ofproto->need_revalidate = REV_RECONFIGURE;
2029 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2030 if (mac->port.p == bundle) {
2032 struct ofproto_dpif *o;
2034 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2036 struct mac_entry *e;
2038 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2041 tag_set_add(&o->revalidate_set, e->tag);
2042 mac_learning_expire(o->ml, e);
2048 mac_learning_expire(ml, mac);
2053 static struct ofbundle *
2054 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2056 struct ofbundle *bundle;
2058 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2059 &ofproto->bundles) {
2060 if (bundle->aux == aux) {
2067 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2068 * ones that are found to 'bundles'. */
2070 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2071 void **auxes, size_t n_auxes,
2072 struct hmapx *bundles)
2076 hmapx_init(bundles);
2077 for (i = 0; i < n_auxes; i++) {
2078 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2080 hmapx_add(bundles, bundle);
2086 bundle_update(struct ofbundle *bundle)
2088 struct ofport_dpif *port;
2090 bundle->floodable = true;
2091 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2092 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2093 || !stp_forward_in_state(port->stp_state)) {
2094 bundle->floodable = false;
2101 bundle_del_port(struct ofport_dpif *port)
2103 struct ofbundle *bundle = port->bundle;
2105 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2107 list_remove(&port->bundle_node);
2108 port->bundle = NULL;
2111 lacp_slave_unregister(bundle->lacp, port);
2114 bond_slave_unregister(bundle->bond, port);
2117 bundle_update(bundle);
2121 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2122 struct lacp_slave_settings *lacp,
2123 uint32_t bond_stable_id)
2125 struct ofport_dpif *port;
2127 port = get_ofp_port(bundle->ofproto, ofp_port);
2132 if (port->bundle != bundle) {
2133 bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2135 bundle_del_port(port);
2138 port->bundle = bundle;
2139 list_push_back(&bundle->ports, &port->bundle_node);
2140 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2141 || !stp_forward_in_state(port->stp_state)) {
2142 bundle->floodable = false;
2146 port->bundle->ofproto->need_revalidate = REV_RECONFIGURE;
2147 lacp_slave_register(bundle->lacp, port, lacp);
2150 port->bond_stable_id = bond_stable_id;
2156 bundle_destroy(struct ofbundle *bundle)
2158 struct ofproto_dpif *ofproto;
2159 struct ofport_dpif *port, *next_port;
2166 ofproto = bundle->ofproto;
2167 for (i = 0; i < MAX_MIRRORS; i++) {
2168 struct ofmirror *m = ofproto->mirrors[i];
2170 if (m->out == bundle) {
2172 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2173 || hmapx_find_and_delete(&m->dsts, bundle)) {
2174 ofproto->need_revalidate = REV_RECONFIGURE;
2179 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2180 bundle_del_port(port);
2183 bundle_flush_macs(bundle, true);
2184 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2186 free(bundle->trunks);
2187 lacp_destroy(bundle->lacp);
2188 bond_destroy(bundle->bond);
2193 bundle_set(struct ofproto *ofproto_, void *aux,
2194 const struct ofproto_bundle_settings *s)
2196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2197 bool need_flush = false;
2198 struct ofport_dpif *port;
2199 struct ofbundle *bundle;
2200 unsigned long *trunks;
2206 bundle_destroy(bundle_lookup(ofproto, aux));
2210 assert(s->n_slaves == 1 || s->bond != NULL);
2211 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2213 bundle = bundle_lookup(ofproto, aux);
2215 bundle = xmalloc(sizeof *bundle);
2217 bundle->ofproto = ofproto;
2218 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2219 hash_pointer(aux, 0));
2221 bundle->name = NULL;
2223 list_init(&bundle->ports);
2224 bundle->vlan_mode = PORT_VLAN_TRUNK;
2226 bundle->trunks = NULL;
2227 bundle->use_priority_tags = s->use_priority_tags;
2228 bundle->lacp = NULL;
2229 bundle->bond = NULL;
2231 bundle->floodable = true;
2233 bundle->src_mirrors = 0;
2234 bundle->dst_mirrors = 0;
2235 bundle->mirror_out = 0;
2238 if (!bundle->name || strcmp(s->name, bundle->name)) {
2240 bundle->name = xstrdup(s->name);
2245 if (!bundle->lacp) {
2246 ofproto->need_revalidate = REV_RECONFIGURE;
2247 bundle->lacp = lacp_create();
2249 lacp_configure(bundle->lacp, s->lacp);
2251 lacp_destroy(bundle->lacp);
2252 bundle->lacp = NULL;
2255 /* Update set of ports. */
2257 for (i = 0; i < s->n_slaves; i++) {
2258 if (!bundle_add_port(bundle, s->slaves[i],
2259 s->lacp ? &s->lacp_slaves[i] : NULL,
2260 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2264 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2265 struct ofport_dpif *next_port;
2267 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2268 for (i = 0; i < s->n_slaves; i++) {
2269 if (s->slaves[i] == port->up.ofp_port) {
2274 bundle_del_port(port);
2278 assert(list_size(&bundle->ports) <= s->n_slaves);
2280 if (list_is_empty(&bundle->ports)) {
2281 bundle_destroy(bundle);
2285 /* Set VLAN tagging mode */
2286 if (s->vlan_mode != bundle->vlan_mode
2287 || s->use_priority_tags != bundle->use_priority_tags) {
2288 bundle->vlan_mode = s->vlan_mode;
2289 bundle->use_priority_tags = s->use_priority_tags;
2294 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2295 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2297 if (vlan != bundle->vlan) {
2298 bundle->vlan = vlan;
2302 /* Get trunked VLANs. */
2303 switch (s->vlan_mode) {
2304 case PORT_VLAN_ACCESS:
2308 case PORT_VLAN_TRUNK:
2309 trunks = CONST_CAST(unsigned long *, s->trunks);
2312 case PORT_VLAN_NATIVE_UNTAGGED:
2313 case PORT_VLAN_NATIVE_TAGGED:
2314 if (vlan != 0 && (!s->trunks
2315 || !bitmap_is_set(s->trunks, vlan)
2316 || bitmap_is_set(s->trunks, 0))) {
2317 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2319 trunks = bitmap_clone(s->trunks, 4096);
2321 trunks = bitmap_allocate1(4096);
2323 bitmap_set1(trunks, vlan);
2324 bitmap_set0(trunks, 0);
2326 trunks = CONST_CAST(unsigned long *, s->trunks);
2333 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2334 free(bundle->trunks);
2335 if (trunks == s->trunks) {
2336 bundle->trunks = vlan_bitmap_clone(trunks);
2338 bundle->trunks = trunks;
2343 if (trunks != s->trunks) {
2348 if (!list_is_short(&bundle->ports)) {
2349 bundle->ofproto->has_bonded_bundles = true;
2351 if (bond_reconfigure(bundle->bond, s->bond)) {
2352 ofproto->need_revalidate = REV_RECONFIGURE;
2355 bundle->bond = bond_create(s->bond);
2356 ofproto->need_revalidate = REV_RECONFIGURE;
2359 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2360 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2364 bond_destroy(bundle->bond);
2365 bundle->bond = NULL;
2368 /* If we changed something that would affect MAC learning, un-learn
2369 * everything on this port and force flow revalidation. */
2371 bundle_flush_macs(bundle, false);
2378 bundle_remove(struct ofport *port_)
2380 struct ofport_dpif *port = ofport_dpif_cast(port_);
2381 struct ofbundle *bundle = port->bundle;
2384 bundle_del_port(port);
2385 if (list_is_empty(&bundle->ports)) {
2386 bundle_destroy(bundle);
2387 } else if (list_is_short(&bundle->ports)) {
2388 bond_destroy(bundle->bond);
2389 bundle->bond = NULL;
2395 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2397 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2398 struct ofport_dpif *port = port_;
2399 uint8_t ea[ETH_ADDR_LEN];
2402 error = netdev_get_etheraddr(port->up.netdev, ea);
2404 struct ofpbuf packet;
2407 ofpbuf_init(&packet, 0);
2408 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2410 memcpy(packet_pdu, pdu, pdu_size);
2412 send_packet(port, &packet);
2413 ofpbuf_uninit(&packet);
2415 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2416 "%s (%s)", port->bundle->name,
2417 netdev_get_name(port->up.netdev), strerror(error));
2422 bundle_send_learning_packets(struct ofbundle *bundle)
2424 struct ofproto_dpif *ofproto = bundle->ofproto;
2425 int error, n_packets, n_errors;
2426 struct mac_entry *e;
2428 error = n_packets = n_errors = 0;
2429 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2430 if (e->port.p != bundle) {
2431 struct ofpbuf *learning_packet;
2432 struct ofport_dpif *port;
2436 /* The assignment to "port" is unnecessary but makes "grep"ing for
2437 * struct ofport_dpif more effective. */
2438 learning_packet = bond_compose_learning_packet(bundle->bond,
2442 ret = send_packet(port, learning_packet);
2443 ofpbuf_delete(learning_packet);
2453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2454 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2455 "packets, last error was: %s",
2456 bundle->name, n_errors, n_packets, strerror(error));
2458 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2459 bundle->name, n_packets);
2464 bundle_run(struct ofbundle *bundle)
2467 lacp_run(bundle->lacp, send_pdu_cb);
2470 struct ofport_dpif *port;
2472 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2473 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2476 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
2477 lacp_status(bundle->lacp));
2478 if (bond_should_send_learning_packets(bundle->bond)) {
2479 bundle_send_learning_packets(bundle);
2485 bundle_wait(struct ofbundle *bundle)
2488 lacp_wait(bundle->lacp);
2491 bond_wait(bundle->bond);
2498 mirror_scan(struct ofproto_dpif *ofproto)
2502 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2503 if (!ofproto->mirrors[idx]) {
2510 static struct ofmirror *
2511 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2515 for (i = 0; i < MAX_MIRRORS; i++) {
2516 struct ofmirror *mirror = ofproto->mirrors[i];
2517 if (mirror && mirror->aux == aux) {
2525 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2527 mirror_update_dups(struct ofproto_dpif *ofproto)
2531 for (i = 0; i < MAX_MIRRORS; i++) {
2532 struct ofmirror *m = ofproto->mirrors[i];
2535 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2539 for (i = 0; i < MAX_MIRRORS; i++) {
2540 struct ofmirror *m1 = ofproto->mirrors[i];
2547 for (j = i + 1; j < MAX_MIRRORS; j++) {
2548 struct ofmirror *m2 = ofproto->mirrors[j];
2550 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2551 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2552 m2->dup_mirrors |= m1->dup_mirrors;
2559 mirror_set(struct ofproto *ofproto_, void *aux,
2560 const struct ofproto_mirror_settings *s)
2562 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2563 mirror_mask_t mirror_bit;
2564 struct ofbundle *bundle;
2565 struct ofmirror *mirror;
2566 struct ofbundle *out;
2567 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2568 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2571 mirror = mirror_lookup(ofproto, aux);
2573 mirror_destroy(mirror);
2579 idx = mirror_scan(ofproto);
2581 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2583 ofproto->up.name, MAX_MIRRORS, s->name);
2587 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2588 mirror->ofproto = ofproto;
2591 mirror->out_vlan = -1;
2592 mirror->name = NULL;
2595 if (!mirror->name || strcmp(s->name, mirror->name)) {
2597 mirror->name = xstrdup(s->name);
2600 /* Get the new configuration. */
2601 if (s->out_bundle) {
2602 out = bundle_lookup(ofproto, s->out_bundle);
2604 mirror_destroy(mirror);
2610 out_vlan = s->out_vlan;
2612 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2613 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2615 /* If the configuration has not changed, do nothing. */
2616 if (hmapx_equals(&srcs, &mirror->srcs)
2617 && hmapx_equals(&dsts, &mirror->dsts)
2618 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2619 && mirror->out == out
2620 && mirror->out_vlan == out_vlan)
2622 hmapx_destroy(&srcs);
2623 hmapx_destroy(&dsts);
2627 hmapx_swap(&srcs, &mirror->srcs);
2628 hmapx_destroy(&srcs);
2630 hmapx_swap(&dsts, &mirror->dsts);
2631 hmapx_destroy(&dsts);
2633 free(mirror->vlans);
2634 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2637 mirror->out_vlan = out_vlan;
2639 /* Update bundles. */
2640 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2641 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2642 if (hmapx_contains(&mirror->srcs, bundle)) {
2643 bundle->src_mirrors |= mirror_bit;
2645 bundle->src_mirrors &= ~mirror_bit;
2648 if (hmapx_contains(&mirror->dsts, bundle)) {
2649 bundle->dst_mirrors |= mirror_bit;
2651 bundle->dst_mirrors &= ~mirror_bit;
2654 if (mirror->out == bundle) {
2655 bundle->mirror_out |= mirror_bit;
2657 bundle->mirror_out &= ~mirror_bit;
2661 ofproto->need_revalidate = REV_RECONFIGURE;
2662 ofproto->has_mirrors = true;
2663 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2664 mirror_update_dups(ofproto);
2670 mirror_destroy(struct ofmirror *mirror)
2672 struct ofproto_dpif *ofproto;
2673 mirror_mask_t mirror_bit;
2674 struct ofbundle *bundle;
2681 ofproto = mirror->ofproto;
2682 ofproto->need_revalidate = REV_RECONFIGURE;
2683 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2685 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2686 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2687 bundle->src_mirrors &= ~mirror_bit;
2688 bundle->dst_mirrors &= ~mirror_bit;
2689 bundle->mirror_out &= ~mirror_bit;
2692 hmapx_destroy(&mirror->srcs);
2693 hmapx_destroy(&mirror->dsts);
2694 free(mirror->vlans);
2696 ofproto->mirrors[mirror->idx] = NULL;
2700 mirror_update_dups(ofproto);
2702 ofproto->has_mirrors = false;
2703 for (i = 0; i < MAX_MIRRORS; i++) {
2704 if (ofproto->mirrors[i]) {
2705 ofproto->has_mirrors = true;
2712 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2713 uint64_t *packets, uint64_t *bytes)
2715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2716 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2719 *packets = *bytes = UINT64_MAX;
2723 *packets = mirror->packet_count;
2724 *bytes = mirror->byte_count;
2730 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2733 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2734 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
2740 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2742 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2743 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2744 return bundle && bundle->mirror_out != 0;
2748 forward_bpdu_changed(struct ofproto *ofproto_)
2750 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2751 ofproto->need_revalidate = REV_RECONFIGURE;
2755 set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time)
2757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2758 mac_learning_set_idle_time(ofproto->ml, idle_time);
2763 static struct ofport_dpif *
2764 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2766 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2767 return ofport ? ofport_dpif_cast(ofport) : NULL;
2770 static struct ofport_dpif *
2771 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2773 return get_ofp_port(ofproto, odp_port_to_ofp_port(ofproto, odp_port));
2777 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2778 struct ofproto_port *ofproto_port,
2779 struct dpif_port *dpif_port)
2781 ofproto_port->name = dpif_port->name;
2782 ofproto_port->type = dpif_port->type;
2783 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2787 port_run_fast(struct ofport_dpif *ofport)
2789 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2790 struct ofpbuf packet;
2792 ofpbuf_init(&packet, 0);
2793 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2794 send_packet(ofport, &packet);
2795 ofpbuf_uninit(&packet);
2800 port_run(struct ofport_dpif *ofport)
2802 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2803 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2804 bool enable = netdev_get_carrier(ofport->up.netdev);
2806 ofport->carrier_seq = carrier_seq;
2808 port_run_fast(ofport);
2810 int cfm_opup = cfm_get_opup(ofport->cfm);
2812 cfm_run(ofport->cfm);
2813 enable = enable && !cfm_get_fault(ofport->cfm);
2815 if (cfm_opup >= 0) {
2816 enable = enable && cfm_opup;
2820 if (ofport->bundle) {
2821 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2822 if (carrier_changed) {
2823 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2827 if (ofport->may_enable != enable) {
2828 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2830 if (ofproto->has_bundle_action) {
2831 ofproto->need_revalidate = REV_PORT_TOGGLED;
2835 ofport->may_enable = enable;
2839 port_wait(struct ofport_dpif *ofport)
2842 cfm_wait(ofport->cfm);
2847 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2848 struct ofproto_port *ofproto_port)
2850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2851 struct dpif_port dpif_port;
2854 if (!sset_contains(&ofproto->ports, devname)) {
2857 error = dpif_port_query_by_name(ofproto->backer->dpif,
2858 devname, &dpif_port);
2860 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2866 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2869 uint32_t odp_port = UINT32_MAX;
2872 error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port);
2874 sset_add(&ofproto->ports, netdev_get_name(netdev));
2880 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
2882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2883 uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port);
2886 if (odp_port != OFPP_NONE) {
2887 error = dpif_port_del(ofproto->backer->dpif, odp_port);
2890 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2892 /* The caller is going to close ofport->up.netdev. If this is a
2893 * bonded port, then the bond is using that netdev, so remove it
2894 * from the bond. The client will need to reconfigure everything
2895 * after deleting ports, so then the slave will get re-added. */
2896 bundle_remove(&ofport->up);
2903 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2905 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2908 error = netdev_get_stats(ofport->up.netdev, stats);
2910 if (!error && ofport->odp_port == OVSP_LOCAL) {
2911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2913 /* ofproto->stats.tx_packets represents packets that we created
2914 * internally and sent to some port (e.g. packets sent with
2915 * send_packet()). Account for them as if they had come from
2916 * OFPP_LOCAL and got forwarded. */
2918 if (stats->rx_packets != UINT64_MAX) {
2919 stats->rx_packets += ofproto->stats.tx_packets;
2922 if (stats->rx_bytes != UINT64_MAX) {
2923 stats->rx_bytes += ofproto->stats.tx_bytes;
2926 /* ofproto->stats.rx_packets represents packets that were received on
2927 * some port and we processed internally and dropped (e.g. STP).
2928 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2930 if (stats->tx_packets != UINT64_MAX) {
2931 stats->tx_packets += ofproto->stats.rx_packets;
2934 if (stats->tx_bytes != UINT64_MAX) {
2935 stats->tx_bytes += ofproto->stats.rx_bytes;
2942 /* Account packets for LOCAL port. */
2944 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
2945 size_t tx_size, size_t rx_size)
2947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2950 ofproto->stats.rx_packets++;
2951 ofproto->stats.rx_bytes += rx_size;
2954 ofproto->stats.tx_packets++;
2955 ofproto->stats.tx_bytes += tx_size;
2959 struct port_dump_state {
2965 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2967 struct port_dump_state *state;
2969 *statep = state = xmalloc(sizeof *state);
2976 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2977 struct ofproto_port *port)
2979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2980 struct port_dump_state *state = state_;
2981 struct sset_node *node;
2983 while ((node = sset_at_position(&ofproto->ports, &state->bucket,
2987 error = port_query_by_name(ofproto_, node->name, port);
2988 if (error != ENODEV) {
2997 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2999 struct port_dump_state *state = state_;
3006 port_poll(const struct ofproto *ofproto_, char **devnamep)
3008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3010 if (ofproto->port_poll_errno) {
3011 int error = ofproto->port_poll_errno;
3012 ofproto->port_poll_errno = 0;
3016 if (sset_is_empty(&ofproto->port_poll_set)) {
3020 *devnamep = sset_pop(&ofproto->port_poll_set);
3025 port_poll_wait(const struct ofproto *ofproto_)
3027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3028 dpif_port_poll_wait(ofproto->backer->dpif);
3032 port_is_lacp_current(const struct ofport *ofport_)
3034 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3035 return (ofport->bundle && ofport->bundle->lacp
3036 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3040 /* Upcall handling. */
3042 /* Flow miss batching.
3044 * Some dpifs implement operations faster when you hand them off in a batch.
3045 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3046 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3047 * more packets, plus possibly installing the flow in the dpif.
3049 * So far we only batch the operations that affect flow setup time the most.
3050 * It's possible to batch more than that, but the benefit might be minimal. */
3052 struct hmap_node hmap_node;
3053 struct ofproto_dpif *ofproto;
3055 enum odp_key_fitness key_fitness;
3056 const struct nlattr *key;
3058 ovs_be16 initial_tci;
3059 struct list packets;
3060 enum dpif_upcall_type upcall_type;
3063 struct flow_miss_op {
3064 struct dpif_op dpif_op;
3065 struct subfacet *subfacet; /* Subfacet */
3066 void *garbage; /* Pointer to pass to free(), NULL if none. */
3067 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3070 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3071 * OpenFlow controller as necessary according to their individual
3072 * configurations. */
3074 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3075 const struct flow *flow)
3077 struct ofputil_packet_in pin;
3079 pin.packet = packet->data;
3080 pin.packet_len = packet->size;
3081 pin.reason = OFPR_NO_MATCH;
3082 pin.controller_id = 0;
3087 pin.send_len = 0; /* not used for flow table misses */
3089 flow_get_metadata(flow, &pin.fmd);
3091 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3094 static enum slow_path_reason
3095 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3096 const struct ofpbuf *packet)
3098 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
3104 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3106 cfm_process_heartbeat(ofport->cfm, packet);
3109 } else if (ofport->bundle && ofport->bundle->lacp
3110 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3112 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3115 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3117 stp_process_packet(ofport, packet);
3124 static struct flow_miss *
3125 flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash)
3127 struct flow_miss *miss;
3129 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3130 if (flow_equal(&miss->flow, flow)) {
3138 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3139 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3140 * 'miss' is associated with a subfacet the caller must also initialize the
3141 * returned op->subfacet, and if anything needs to be freed after processing
3142 * the op, the caller must initialize op->garbage also. */
3144 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3145 struct flow_miss_op *op)
3147 if (miss->flow.vlan_tci != miss->initial_tci) {
3148 /* This packet was received on a VLAN splinter port. We
3149 * added a VLAN to the packet to make the packet resemble
3150 * the flow, but the actions were composed assuming that
3151 * the packet contained no VLAN. So, we must remove the
3152 * VLAN header from the packet before trying to execute the
3154 eth_pop_vlan(packet);
3157 op->subfacet = NULL;
3159 op->dpif_op.type = DPIF_OP_EXECUTE;
3160 op->dpif_op.u.execute.key = miss->key;
3161 op->dpif_op.u.execute.key_len = miss->key_len;
3162 op->dpif_op.u.execute.packet = packet;
3165 /* Helper for handle_flow_miss_without_facet() and
3166 * handle_flow_miss_with_facet(). */
3168 handle_flow_miss_common(struct rule_dpif *rule,
3169 struct ofpbuf *packet, const struct flow *flow)
3171 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3173 ofproto->n_matches++;
3175 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3177 * Extra-special case for fail-open mode.
3179 * We are in fail-open mode and the packet matched the fail-open
3180 * rule, but we are connected to a controller too. We should send
3181 * the packet up to the controller in the hope that it will try to
3182 * set up a flow and thereby allow us to exit fail-open.
3184 * See the top-level comment in fail-open.c for more information.
3186 send_packet_in_miss(ofproto, packet, flow);
3190 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3191 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3192 * installing a datapath flow. The answer is usually "yes" (a return value of
3193 * true). However, for short flows the cost of bookkeeping is much higher than
3194 * the benefits, so when the datapath holds a large number of flows we impose
3195 * some heuristics to decide which flows are likely to be worth tracking. */
3197 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3198 struct flow_miss *miss, uint32_t hash)
3200 if (!ofproto->governor) {
3203 n_subfacets = hmap_count(&ofproto->subfacets);
3204 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3208 ofproto->governor = governor_create(ofproto->up.name);
3211 return governor_should_install_flow(ofproto->governor, hash,
3212 list_size(&miss->packets));
3215 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3216 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3217 * increment '*n_ops'. */
3219 handle_flow_miss_without_facet(struct flow_miss *miss,
3220 struct rule_dpif *rule,
3221 struct flow_miss_op *ops, size_t *n_ops)
3223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3224 long long int now = time_msec();
3225 struct action_xlate_ctx ctx;
3226 struct ofpbuf *packet;
3228 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3229 struct flow_miss_op *op = &ops[*n_ops];
3230 struct dpif_flow_stats stats;
3231 struct ofpbuf odp_actions;
3233 COVERAGE_INC(facet_suppress);
3235 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3237 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3238 rule_credit_stats(rule, &stats);
3240 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3242 ctx.resubmit_stats = &stats;
3243 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3246 if (odp_actions.size) {
3247 struct dpif_execute *execute = &op->dpif_op.u.execute;
3249 init_flow_miss_execute_op(miss, packet, op);
3250 execute->actions = odp_actions.data;
3251 execute->actions_len = odp_actions.size;
3252 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3256 ofpbuf_uninit(&odp_actions);
3261 /* Handles 'miss', which matches 'facet'. May add any required datapath
3262 * operations to 'ops', incrementing '*n_ops' for each new op.
3264 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3265 * This is really important only for new facets: if we just called time_msec()
3266 * here, then the new subfacet or its packets could look (occasionally) as
3267 * though it was used some time after the facet was used. That can make a
3268 * one-packet flow look like it has a nonzero duration, which looks odd in
3269 * e.g. NetFlow statistics. */
3271 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3273 struct flow_miss_op *ops, size_t *n_ops)
3275 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3276 enum subfacet_path want_path;
3277 struct subfacet *subfacet;
3278 struct ofpbuf *packet;
3280 subfacet = subfacet_create(facet,
3281 miss->key_fitness, miss->key, miss->key_len,
3282 miss->initial_tci, now);
3284 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3285 struct flow_miss_op *op = &ops[*n_ops];
3286 struct dpif_flow_stats stats;
3287 struct ofpbuf odp_actions;
3289 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3291 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3292 if (!subfacet->actions || subfacet->slow) {
3293 subfacet_make_actions(subfacet, packet, &odp_actions);
3296 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3297 subfacet_update_stats(subfacet, &stats);
3299 if (subfacet->actions_len) {
3300 struct dpif_execute *execute = &op->dpif_op.u.execute;
3302 init_flow_miss_execute_op(miss, packet, op);
3303 op->subfacet = subfacet;
3304 if (!subfacet->slow) {
3305 execute->actions = subfacet->actions;
3306 execute->actions_len = subfacet->actions_len;
3307 ofpbuf_uninit(&odp_actions);
3309 execute->actions = odp_actions.data;
3310 execute->actions_len = odp_actions.size;
3311 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3316 ofpbuf_uninit(&odp_actions);
3320 want_path = subfacet_want_path(subfacet->slow);
3321 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3322 struct flow_miss_op *op = &ops[(*n_ops)++];
3323 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3325 op->subfacet = subfacet;
3327 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3328 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3329 put->key = miss->key;
3330 put->key_len = miss->key_len;
3331 if (want_path == SF_FAST_PATH) {
3332 put->actions = subfacet->actions;
3333 put->actions_len = subfacet->actions_len;
3335 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3336 op->stub, sizeof op->stub,
3337 &put->actions, &put->actions_len);
3343 /* Handles flow miss 'miss'. May add any required datapath operations
3344 * to 'ops', incrementing '*n_ops' for each new op. */
3346 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3349 struct ofproto_dpif *ofproto = miss->ofproto;
3350 struct facet *facet;
3354 /* The caller must ensure that miss->hmap_node.hash contains
3355 * flow_hash(miss->flow, 0). */
3356 hash = miss->hmap_node.hash;
3358 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3360 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3362 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3363 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3367 facet = facet_create(rule, &miss->flow, hash);
3372 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3375 /* This function does post-processing on data returned from
3376 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the
3377 * rest of the upcall processing logic. In particular, if the extracted
3378 * in_port is a VLAN splinter port, it replaces flow->in_port by the "real"
3379 * port, sets flow->vlan_tci correctly for the VLAN of the VLAN splinter
3380 * port, and pushes a VLAN header onto 'packet' (if it is nonnull). The
3381 * caller must have called odp_flow_key_to_flow() and supply 'fitness' and
3382 * 'flow' from its output. The 'flow' argument must have had the "in_port"
3383 * member converted to the OpenFlow number.
3385 * Sets '*initial_tci' to the VLAN TCI with which the packet was really
3386 * received, that is, the actual VLAN TCI extracted by odp_flow_key_to_flow().
3387 * (This differs from the value returned in flow->vlan_tci only for packets
3388 * received on VLAN splinters.) */
3389 static enum odp_key_fitness
3390 ofproto_dpif_vsp_adjust(const struct ofproto_dpif *ofproto,
3391 enum odp_key_fitness fitness,
3392 struct flow *flow, ovs_be16 *initial_tci,
3393 struct ofpbuf *packet)
3395 if (fitness == ODP_FIT_ERROR) {
3398 *initial_tci = flow->vlan_tci;
3400 if (vsp_adjust_flow(ofproto, flow)) {
3402 /* Make the packet resemble the flow, so that it gets sent to an
3403 * OpenFlow controller properly, so that it looks correct for
3404 * sFlow, and so that flow_extract() will get the correct vlan_tci
3405 * if it is called on 'packet'.
3407 * The allocated space inside 'packet' probably also contains
3408 * 'key', that is, both 'packet' and 'key' are probably part of a
3409 * struct dpif_upcall (see the large comment on that structure
3410 * definition), so pushing data on 'packet' is in general not a
3411 * good idea since it could overwrite 'key' or free it as a side
3412 * effect. However, it's OK in this special case because we know
3413 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3414 * will just overwrite the 4-byte "struct nlattr", which is fine
3415 * since we don't need that header anymore. */
3416 eth_push_vlan(packet, flow->vlan_tci);
3419 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3420 if (fitness == ODP_FIT_PERFECT) {
3421 fitness = ODP_FIT_TOO_MUCH;
3429 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3432 struct dpif_upcall *upcall;
3433 struct flow_miss *miss;
3434 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3435 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3436 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3446 /* Construct the to-do list.
3448 * This just amounts to extracting the flow from each packet and sticking
3449 * the packets that have the same flow in the same "flow_miss" structure so
3450 * that we can process them together. */
3453 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3454 struct flow_miss *miss = &misses[n_misses];
3455 struct flow_miss *existing_miss;
3456 enum odp_key_fitness fitness;
3457 struct ofproto_dpif *ofproto;
3458 struct ofport_dpif *port;
3462 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3463 port = odp_port_to_ofport(backer, flow.in_port);
3465 /* Received packet on port for which we couldn't associate
3466 * an ofproto. This can happen if a port is removed while
3467 * traffic is being received. Print a rate-limited message
3468 * in case it happens frequently. */
3469 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3473 ofproto = ofproto_dpif_cast(port->up.ofproto);
3474 flow.in_port = port->up.ofp_port;
3476 /* Obtain metadata and check userspace/kernel agreement on flow match,
3477 * then set 'flow''s header pointers. */
3478 miss->key_fitness = ofproto_dpif_vsp_adjust(ofproto, fitness,
3479 &flow, &miss->initial_tci, upcall->packet);
3480 if (miss->key_fitness == ODP_FIT_ERROR) {
3483 flow_extract(upcall->packet, flow.skb_priority,
3484 &flow.tunnel, flow.in_port, &miss->flow);
3486 /* Add other packets to a to-do list. */
3487 hash = flow_hash(&miss->flow, 0);
3488 existing_miss = flow_miss_find(&todo, &miss->flow, hash);
3489 if (!existing_miss) {
3490 hmap_insert(&todo, &miss->hmap_node, hash);
3491 miss->ofproto = ofproto;
3492 miss->key = upcall->key;
3493 miss->key_len = upcall->key_len;
3494 miss->upcall_type = upcall->type;
3495 list_init(&miss->packets);
3499 miss = existing_miss;
3501 list_push_back(&miss->packets, &upcall->packet->list_node);
3504 /* Process each element in the to-do list, constructing the set of
3505 * operations to batch. */
3507 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3508 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3510 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3512 /* Execute batch. */
3513 for (i = 0; i < n_ops; i++) {
3514 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3516 dpif_operate(backer->dpif, dpif_ops, n_ops);
3518 /* Free memory and update facets. */
3519 for (i = 0; i < n_ops; i++) {
3520 struct flow_miss_op *op = &flow_miss_ops[i];
3522 switch (op->dpif_op.type) {
3523 case DPIF_OP_EXECUTE:
3526 case DPIF_OP_FLOW_PUT:
3527 if (!op->dpif_op.error) {
3528 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3532 case DPIF_OP_FLOW_DEL:
3538 hmap_destroy(&todo);
3541 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3542 classify_upcall(const struct dpif_upcall *upcall)
3544 union user_action_cookie cookie;
3546 /* First look at the upcall type. */
3547 switch (upcall->type) {
3548 case DPIF_UC_ACTION:
3554 case DPIF_N_UC_TYPES:
3556 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3560 /* "action" upcalls need a closer look. */
3561 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3562 switch (cookie.type) {
3563 case USER_ACTION_COOKIE_SFLOW:
3564 return SFLOW_UPCALL;
3566 case USER_ACTION_COOKIE_SLOW_PATH:
3569 case USER_ACTION_COOKIE_UNSPEC:
3571 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3577 handle_sflow_upcall(struct dpif_backer *backer,
3578 const struct dpif_upcall *upcall)
3580 struct ofproto_dpif *ofproto;
3581 union user_action_cookie cookie;
3582 enum odp_key_fitness fitness;
3583 struct ofport_dpif *port;
3584 ovs_be16 initial_tci;
3586 uint32_t odp_in_port;
3588 fitness = odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
3590 port = odp_port_to_ofport(backer, flow.in_port);
3595 ofproto = ofproto_dpif_cast(port->up.ofproto);
3596 if (!ofproto->sflow) {
3600 odp_in_port = flow.in_port;
3601 flow.in_port = port->up.ofp_port;
3602 fitness = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
3603 &initial_tci, upcall->packet);
3604 if (fitness == ODP_FIT_ERROR) {
3608 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3609 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3610 odp_in_port, &cookie);
3614 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3616 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3617 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3618 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3623 assert(max_batch <= FLOW_MISS_MAX_BATCH);
3626 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3627 struct dpif_upcall *upcall = &misses[n_misses];
3628 struct ofpbuf *buf = &miss_bufs[n_misses];
3631 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3632 sizeof miss_buf_stubs[n_misses]);
3633 error = dpif_recv(backer->dpif, upcall, buf);
3639 switch (classify_upcall(upcall)) {
3641 /* Handle it later. */
3646 handle_sflow_upcall(backer, upcall);
3656 /* Handle deferred MISS_UPCALL processing. */
3657 handle_miss_upcalls(backer, misses, n_misses);
3658 for (i = 0; i < n_misses; i++) {
3659 ofpbuf_uninit(&miss_bufs[i]);
3665 /* Flow expiration. */
3667 static int subfacet_max_idle(const struct ofproto_dpif *);
3668 static void update_stats(struct dpif_backer *);
3669 static void rule_expire(struct rule_dpif *);
3670 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3672 /* This function is called periodically by run(). Its job is to collect
3673 * updates for the flows that have been installed into the datapath, most
3674 * importantly when they last were used, and then use that information to
3675 * expire flows that have not been used recently.
3677 * Returns the number of milliseconds after which it should be called again. */
3679 expire(struct dpif_backer *backer)
3681 struct ofproto_dpif *ofproto;
3682 int max_idle = INT32_MAX;
3684 /* Update stats for each flow in the backer. */
3685 update_stats(backer);
3687 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3688 struct rule_dpif *rule, *next_rule;
3689 struct oftable *table;
3692 if (ofproto->backer != backer) {
3696 /* Expire subfacets that have been idle too long. */
3697 dp_max_idle = subfacet_max_idle(ofproto);
3698 expire_subfacets(ofproto, dp_max_idle);
3700 max_idle = MIN(max_idle, dp_max_idle);
3702 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3704 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
3705 struct cls_cursor cursor;
3707 cls_cursor_init(&cursor, &table->cls, NULL);
3708 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
3713 /* All outstanding data in existing flows has been accounted, so it's a
3714 * good time to do bond rebalancing. */
3715 if (ofproto->has_bonded_bundles) {
3716 struct ofbundle *bundle;
3718 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3720 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
3726 return MIN(max_idle, 1000);
3729 /* Updates flow table statistics given that the datapath just reported 'stats'
3730 * as 'subfacet''s statistics. */
3732 update_subfacet_stats(struct subfacet *subfacet,
3733 const struct dpif_flow_stats *stats)
3735 struct facet *facet = subfacet->facet;
3737 if (stats->n_packets >= subfacet->dp_packet_count) {
3738 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
3739 facet->packet_count += extra;
3741 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3744 if (stats->n_bytes >= subfacet->dp_byte_count) {
3745 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
3747 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3750 subfacet->dp_packet_count = stats->n_packets;
3751 subfacet->dp_byte_count = stats->n_bytes;
3753 facet->tcp_flags |= stats->tcp_flags;
3755 subfacet_update_time(subfacet, stats->used);
3756 if (facet->accounted_bytes < facet->byte_count) {
3758 facet_account(facet);
3759 facet->accounted_bytes = facet->byte_count;
3761 facet_push_stats(facet);
3764 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3765 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3767 delete_unexpected_flow(struct ofproto_dpif *ofproto,
3768 const struct nlattr *key, size_t key_len)
3770 if (!VLOG_DROP_WARN(&rl)) {
3774 odp_flow_key_format(key, key_len, &s);
3775 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
3779 COVERAGE_INC(facet_unexpected);
3780 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
3783 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3785 * This function also pushes statistics updates to rules which each facet
3786 * resubmits into. Generally these statistics will be accurate. However, if a
3787 * facet changes the rule it resubmits into at some time in between
3788 * update_stats() runs, it is possible that statistics accrued to the
3789 * old rule will be incorrectly attributed to the new rule. This could be
3790 * avoided by calling update_stats() whenever rules are created or
3791 * deleted. However, the performance impact of making so many calls to the
3792 * datapath do not justify the benefit of having perfectly accurate statistics.
3795 update_stats(struct dpif_backer *backer)
3797 const struct dpif_flow_stats *stats;
3798 struct dpif_flow_dump dump;
3799 const struct nlattr *key;
3802 dpif_flow_dump_start(&dump, backer->dpif);
3803 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
3805 struct subfacet *subfacet;
3806 enum odp_key_fitness fitness;
3807 struct ofproto_dpif *ofproto;
3808 struct ofport_dpif *port;
3811 fitness = odp_flow_key_to_flow(key, key_len, &flow);
3812 if (fitness == ODP_FIT_ERROR) {
3816 port = odp_port_to_ofport(backer, flow.in_port);
3818 /* This flow is for a port for which we couldn't associate an
3819 * ofproto. This can happen if a port is removed while
3820 * traffic is being received. Print a rate-limited message
3821 * in case it happens frequently. */
3823 "stats update for flow with unassociated port %"PRIu32,
3828 ofproto = ofproto_dpif_cast(port->up.ofproto);
3829 flow.in_port = port->up.ofp_port;
3830 key_hash = odp_flow_key_hash(key, key_len);
3832 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
3833 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3835 update_subfacet_stats(subfacet, stats);
3839 /* Stats are updated per-packet. */
3842 case SF_NOT_INSTALLED:
3844 delete_unexpected_flow(ofproto, key, key_len);
3848 dpif_flow_dump_done(&dump);
3851 /* Calculates and returns the number of milliseconds of idle time after which
3852 * subfacets should expire from the datapath. When a subfacet expires, we fold
3853 * its statistics into its facet, and when a facet's last subfacet expires, we
3854 * fold its statistic into its rule. */
3856 subfacet_max_idle(const struct ofproto_dpif *ofproto)
3859 * Idle time histogram.
3861 * Most of the time a switch has a relatively small number of subfacets.
3862 * When this is the case we might as well keep statistics for all of them
3863 * in userspace and to cache them in the kernel datapath for performance as
3866 * As the number of subfacets increases, the memory required to maintain
3867 * statistics about them in userspace and in the kernel becomes
3868 * significant. However, with a large number of subfacets it is likely
3869 * that only a few of them are "heavy hitters" that consume a large amount
3870 * of bandwidth. At this point, only heavy hitters are worth caching in
3871 * the kernel and maintaining in userspaces; other subfacets we can
3874 * The technique used to compute the idle time is to build a histogram with
3875 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3876 * that is installed in the kernel gets dropped in the appropriate bucket.
3877 * After the histogram has been built, we compute the cutoff so that only
3878 * the most-recently-used 1% of subfacets (but at least
3879 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3880 * the most-recently-used bucket of subfacets is kept, so actually an
3881 * arbitrary number of subfacets can be kept in any given expiration run
3882 * (though the next run will delete most of those unless they receive
3885 * This requires a second pass through the subfacets, in addition to the
3886 * pass made by update_stats(), because the former function never looks at
3887 * uninstallable subfacets.
3889 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3890 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3891 int buckets[N_BUCKETS] = { 0 };
3892 int total, subtotal, bucket;
3893 struct subfacet *subfacet;
3897 total = hmap_count(&ofproto->subfacets);
3898 if (total <= ofproto->up.flow_eviction_threshold) {
3899 return N_BUCKETS * BUCKET_WIDTH;
3902 /* Build histogram. */
3904 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
3905 long long int idle = now - subfacet->used;
3906 int bucket = (idle <= 0 ? 0
3907 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3908 : (unsigned int) idle / BUCKET_WIDTH);
3912 /* Find the first bucket whose flows should be expired. */
3913 subtotal = bucket = 0;
3915 subtotal += buckets[bucket++];
3916 } while (bucket < N_BUCKETS &&
3917 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
3919 if (VLOG_IS_DBG_ENABLED()) {
3923 ds_put_cstr(&s, "keep");
3924 for (i = 0; i < N_BUCKETS; i++) {
3926 ds_put_cstr(&s, ", drop");
3929 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3932 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
3936 return bucket * BUCKET_WIDTH;
3940 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
3942 /* Cutoff time for most flows. */
3943 long long int normal_cutoff = time_msec() - dp_max_idle;
3945 /* We really want to keep flows for special protocols around, so use a more
3946 * conservative cutoff. */
3947 long long int special_cutoff = time_msec() - 10000;
3949 struct subfacet *subfacet, *next_subfacet;
3950 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3954 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3955 &ofproto->subfacets) {
3956 long long int cutoff;
3958 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
3961 if (subfacet->used < cutoff) {
3962 if (subfacet->path != SF_NOT_INSTALLED) {
3963 batch[n_batch++] = subfacet;
3964 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3965 subfacet_destroy_batch(ofproto, batch, n_batch);
3969 subfacet_destroy(subfacet);
3975 subfacet_destroy_batch(ofproto, batch, n_batch);
3979 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3980 * then delete it entirely. */
3982 rule_expire(struct rule_dpif *rule)
3984 struct facet *facet, *next_facet;
3988 if (rule->up.pending) {
3989 /* We'll have to expire it later. */
3993 /* Has 'rule' expired? */
3995 if (rule->up.hard_timeout
3996 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
3997 reason = OFPRR_HARD_TIMEOUT;
3998 } else if (rule->up.idle_timeout
3999 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4000 reason = OFPRR_IDLE_TIMEOUT;
4005 COVERAGE_INC(ofproto_dpif_expired);
4007 /* Update stats. (This is a no-op if the rule expired due to an idle
4008 * timeout, because that only happens when the rule has no facets left.) */
4009 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4010 facet_remove(facet);
4013 /* Get rid of the rule. */
4014 ofproto_rule_expire(&rule->up, reason);
4019 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4021 * The caller must already have determined that no facet with an identical
4022 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4023 * the ofproto's classifier table.
4025 * 'hash' must be the return value of flow_hash(flow, 0).
4027 * The facet will initially have no subfacets. The caller should create (at
4028 * least) one subfacet with subfacet_create(). */
4029 static struct facet *
4030 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4033 struct facet *facet;
4035 facet = xzalloc(sizeof *facet);
4036 facet->used = time_msec();
4037 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4038 list_push_back(&rule->facets, &facet->list_node);
4040 facet->flow = *flow;
4041 list_init(&facet->subfacets);
4042 netflow_flow_init(&facet->nf_flow);
4043 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4049 facet_free(struct facet *facet)
4054 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4055 * 'packet', which arrived on 'in_port'.
4057 * Takes ownership of 'packet'. */
4059 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4060 const struct nlattr *odp_actions, size_t actions_len,
4061 struct ofpbuf *packet)
4063 struct odputil_keybuf keybuf;
4067 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4068 odp_flow_key_from_flow(&key, flow,
4069 ofp_port_to_odp_port(ofproto, flow->in_port));
4071 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4072 odp_actions, actions_len, packet);
4074 ofpbuf_delete(packet);
4078 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4080 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4081 * rule's statistics, via subfacet_uninstall().
4083 * - Removes 'facet' from its rule and from ofproto->facets.
4086 facet_remove(struct facet *facet)
4088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4089 struct subfacet *subfacet, *next_subfacet;
4091 assert(!list_is_empty(&facet->subfacets));
4093 /* First uninstall all of the subfacets to get final statistics. */
4094 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4095 subfacet_uninstall(subfacet);
4098 /* Flush the final stats to the rule.
4100 * This might require us to have at least one subfacet around so that we
4101 * can use its actions for accounting in facet_account(), which is why we
4102 * have uninstalled but not yet destroyed the subfacets. */
4103 facet_flush_stats(facet);
4105 /* Now we're really all done so destroy everything. */
4106 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4107 &facet->subfacets) {
4108 subfacet_destroy__(subfacet);
4110 hmap_remove(&ofproto->facets, &facet->hmap_node);
4111 list_remove(&facet->list_node);
4115 /* Feed information from 'facet' back into the learning table to keep it in
4116 * sync with what is actually flowing through the datapath. */
4118 facet_learn(struct facet *facet)
4120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4121 struct action_xlate_ctx ctx;
4123 if (!facet->has_learn
4124 && !facet->has_normal
4125 && (!facet->has_fin_timeout
4126 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4130 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4131 facet->flow.vlan_tci,
4132 facet->rule, facet->tcp_flags, NULL);
4133 ctx.may_learn = true;
4134 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4135 facet->rule->up.ofpacts_len);
4139 facet_account(struct facet *facet)
4141 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4142 struct subfacet *subfacet;
4143 const struct nlattr *a;
4148 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4151 n_bytes = facet->byte_count - facet->accounted_bytes;
4153 /* This loop feeds byte counters to bond_account() for rebalancing to use
4154 * as a basis. We also need to track the actual VLAN on which the packet
4155 * is going to be sent to ensure that it matches the one passed to
4156 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4159 * We use the actions from an arbitrary subfacet because they should all
4160 * be equally valid for our purpose. */
4161 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4162 struct subfacet, list_node);
4163 vlan_tci = facet->flow.vlan_tci;
4164 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4165 subfacet->actions, subfacet->actions_len) {
4166 const struct ovs_action_push_vlan *vlan;
4167 struct ofport_dpif *port;
4169 switch (nl_attr_type(a)) {
4170 case OVS_ACTION_ATTR_OUTPUT:
4171 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4172 if (port && port->bundle && port->bundle->bond) {
4173 bond_account(port->bundle->bond, &facet->flow,
4174 vlan_tci_to_vid(vlan_tci), n_bytes);
4178 case OVS_ACTION_ATTR_POP_VLAN:
4179 vlan_tci = htons(0);
4182 case OVS_ACTION_ATTR_PUSH_VLAN:
4183 vlan = nl_attr_get(a);
4184 vlan_tci = vlan->vlan_tci;
4190 /* Returns true if the only action for 'facet' is to send to the controller.
4191 * (We don't report NetFlow expiration messages for such facets because they
4192 * are just part of the control logic for the network, not real traffic). */
4194 facet_is_controller_flow(struct facet *facet)
4197 const struct rule *rule = &facet->rule->up;
4198 const struct ofpact *ofpacts = rule->ofpacts;
4199 size_t ofpacts_len = rule->ofpacts_len;
4201 if (ofpacts_len > 0 &&
4202 ofpacts->type == OFPACT_CONTROLLER &&
4203 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4210 /* Folds all of 'facet''s statistics into its rule. Also updates the
4211 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4212 * 'facet''s statistics in the datapath should have been zeroed and folded into
4213 * its packet and byte counts before this function is called. */
4215 facet_flush_stats(struct facet *facet)
4217 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4218 struct subfacet *subfacet;
4220 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4221 assert(!subfacet->dp_byte_count);
4222 assert(!subfacet->dp_packet_count);
4225 facet_push_stats(facet);
4226 if (facet->accounted_bytes < facet->byte_count) {
4227 facet_account(facet);
4228 facet->accounted_bytes = facet->byte_count;
4231 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4232 struct ofexpired expired;
4233 expired.flow = facet->flow;
4234 expired.packet_count = facet->packet_count;
4235 expired.byte_count = facet->byte_count;
4236 expired.used = facet->used;
4237 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4240 facet->rule->packet_count += facet->packet_count;
4241 facet->rule->byte_count += facet->byte_count;
4243 /* Reset counters to prevent double counting if 'facet' ever gets
4245 facet_reset_counters(facet);
4247 netflow_flow_clear(&facet->nf_flow);
4248 facet->tcp_flags = 0;
4251 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4252 * Returns it if found, otherwise a null pointer.
4254 * 'hash' must be the return value of flow_hash(flow, 0).
4256 * The returned facet might need revalidation; use facet_lookup_valid()
4257 * instead if that is important. */
4258 static struct facet *
4259 facet_find(struct ofproto_dpif *ofproto,
4260 const struct flow *flow, uint32_t hash)
4262 struct facet *facet;
4264 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4265 if (flow_equal(flow, &facet->flow)) {
4273 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4274 * Returns it if found, otherwise a null pointer.
4276 * 'hash' must be the return value of flow_hash(flow, 0).
4278 * The returned facet is guaranteed to be valid. */
4279 static struct facet *
4280 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4283 struct facet *facet;
4285 facet = facet_find(ofproto, flow, hash);
4287 && (ofproto->need_revalidate
4288 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) {
4289 facet_revalidate(facet);
4296 subfacet_path_to_string(enum subfacet_path path)
4299 case SF_NOT_INSTALLED:
4300 return "not installed";
4302 return "in fast path";
4304 return "in slow path";
4310 /* Returns the path in which a subfacet should be installed if its 'slow'
4311 * member has the specified value. */
4312 static enum subfacet_path
4313 subfacet_want_path(enum slow_path_reason slow)
4315 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4318 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4319 * supposing that its actions have been recalculated as 'want_actions' and that
4320 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4322 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4323 const struct ofpbuf *want_actions)
4325 enum subfacet_path want_path = subfacet_want_path(slow);
4326 return (want_path != subfacet->path
4327 || (want_path == SF_FAST_PATH
4328 && (subfacet->actions_len != want_actions->size
4329 || memcmp(subfacet->actions, want_actions->data,
4330 subfacet->actions_len))));
4334 facet_check_consistency(struct facet *facet)
4336 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4340 uint64_t odp_actions_stub[1024 / 8];
4341 struct ofpbuf odp_actions;
4343 struct rule_dpif *rule;
4344 struct subfacet *subfacet;
4345 bool may_log = false;
4348 /* Check the rule for consistency. */
4349 rule = rule_dpif_lookup(ofproto, &facet->flow);
4350 ok = rule == facet->rule;
4352 may_log = !VLOG_DROP_WARN(&rl);
4357 flow_format(&s, &facet->flow);
4358 ds_put_format(&s, ": facet associated with wrong rule (was "
4359 "table=%"PRIu8",", facet->rule->up.table_id);
4360 cls_rule_format(&facet->rule->up.cr, &s);
4361 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4363 cls_rule_format(&rule->up.cr, &s);
4364 ds_put_char(&s, ')');
4366 VLOG_WARN("%s", ds_cstr(&s));
4371 /* Check the datapath actions for consistency. */
4372 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4373 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4374 enum subfacet_path want_path;
4375 struct odputil_keybuf keybuf;
4376 struct action_xlate_ctx ctx;
4380 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4381 subfacet->initial_tci, rule, 0, NULL);
4382 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4385 if (subfacet->path == SF_NOT_INSTALLED) {
4386 /* This only happens if the datapath reported an error when we
4387 * tried to install the flow. Don't flag another error here. */
4391 want_path = subfacet_want_path(subfacet->slow);
4392 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4393 /* The actions for slow-path flows may legitimately vary from one
4394 * packet to the next. We're done. */
4398 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4402 /* Inconsistency! */
4404 may_log = !VLOG_DROP_WARN(&rl);
4408 /* Rate-limited, skip reporting. */
4413 subfacet_get_key(subfacet, &keybuf, &key);
4414 odp_flow_key_format(key.data, key.size, &s);
4416 ds_put_cstr(&s, ": inconsistency in subfacet");
4417 if (want_path != subfacet->path) {
4418 enum odp_key_fitness fitness = subfacet->key_fitness;
4420 ds_put_format(&s, " (%s, fitness=%s)",
4421 subfacet_path_to_string(subfacet->path),
4422 odp_key_fitness_to_string(fitness));
4423 ds_put_format(&s, " (should have been %s)",
4424 subfacet_path_to_string(want_path));
4425 } else if (want_path == SF_FAST_PATH) {
4426 ds_put_cstr(&s, " (actions were: ");
4427 format_odp_actions(&s, subfacet->actions,
4428 subfacet->actions_len);
4429 ds_put_cstr(&s, ") (correct actions: ");
4430 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4431 ds_put_char(&s, ')');
4433 ds_put_cstr(&s, " (actions: ");
4434 format_odp_actions(&s, subfacet->actions,
4435 subfacet->actions_len);
4436 ds_put_char(&s, ')');
4438 VLOG_WARN("%s", ds_cstr(&s));
4441 ofpbuf_uninit(&odp_actions);
4446 /* Re-searches the classifier for 'facet':
4448 * - If the rule found is different from 'facet''s current rule, moves
4449 * 'facet' to the new rule and recompiles its actions.
4451 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4452 * where it is and recompiles its actions anyway. */
4454 facet_revalidate(struct facet *facet)
4456 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4458 struct nlattr *odp_actions;
4461 struct actions *new_actions;
4463 struct action_xlate_ctx ctx;
4464 uint64_t odp_actions_stub[1024 / 8];
4465 struct ofpbuf odp_actions;
4467 struct rule_dpif *new_rule;
4468 struct subfacet *subfacet;
4471 COVERAGE_INC(facet_revalidate);
4473 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4475 /* Calculate new datapath actions.
4477 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4478 * emit a NetFlow expiration and, if so, we need to have the old state
4479 * around to properly compose it. */
4481 /* If the datapath actions changed or the installability changed,
4482 * then we need to talk to the datapath. */
4485 memset(&ctx, 0, sizeof ctx);
4486 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4487 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4488 enum slow_path_reason slow;
4490 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4491 subfacet->initial_tci, new_rule, 0, NULL);
4492 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4495 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4496 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4497 struct dpif_flow_stats stats;
4499 subfacet_install(subfacet,
4500 odp_actions.data, odp_actions.size, &stats, slow);
4501 subfacet_update_stats(subfacet, &stats);
4504 new_actions = xcalloc(list_size(&facet->subfacets),
4505 sizeof *new_actions);
4507 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4509 new_actions[i].actions_len = odp_actions.size;
4514 ofpbuf_uninit(&odp_actions);
4517 facet_flush_stats(facet);
4520 /* Update 'facet' now that we've taken care of all the old state. */
4521 facet->tags = ctx.tags;
4522 facet->nf_flow.output_iface = ctx.nf_output_iface;
4523 facet->has_learn = ctx.has_learn;
4524 facet->has_normal = ctx.has_normal;
4525 facet->has_fin_timeout = ctx.has_fin_timeout;
4526 facet->mirrors = ctx.mirrors;
4529 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4530 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4532 if (new_actions && new_actions[i].odp_actions) {
4533 free(subfacet->actions);
4534 subfacet->actions = new_actions[i].odp_actions;
4535 subfacet->actions_len = new_actions[i].actions_len;
4541 if (facet->rule != new_rule) {
4542 COVERAGE_INC(facet_changed_rule);
4543 list_remove(&facet->list_node);
4544 list_push_back(&new_rule->facets, &facet->list_node);
4545 facet->rule = new_rule;
4546 facet->used = new_rule->up.created;
4547 facet->prev_used = facet->used;
4551 /* Updates 'facet''s used time. Caller is responsible for calling
4552 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4554 facet_update_time(struct facet *facet, long long int used)
4556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4557 if (used > facet->used) {
4559 ofproto_rule_update_used(&facet->rule->up, used);
4560 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4565 facet_reset_counters(struct facet *facet)
4567 facet->packet_count = 0;
4568 facet->byte_count = 0;
4569 facet->prev_packet_count = 0;
4570 facet->prev_byte_count = 0;
4571 facet->accounted_bytes = 0;
4575 facet_push_stats(struct facet *facet)
4577 struct dpif_flow_stats stats;
4579 assert(facet->packet_count >= facet->prev_packet_count);
4580 assert(facet->byte_count >= facet->prev_byte_count);
4581 assert(facet->used >= facet->prev_used);
4583 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4584 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4585 stats.used = facet->used;
4586 stats.tcp_flags = 0;
4588 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4589 facet->prev_packet_count = facet->packet_count;
4590 facet->prev_byte_count = facet->byte_count;
4591 facet->prev_used = facet->used;
4593 flow_push_stats(facet->rule, &facet->flow, &stats);
4595 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4596 facet->mirrors, stats.n_packets, stats.n_bytes);
4601 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4603 rule->packet_count += stats->n_packets;
4604 rule->byte_count += stats->n_bytes;
4605 ofproto_rule_update_used(&rule->up, stats->used);
4608 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4609 * 'rule''s actions and mirrors. */
4611 flow_push_stats(struct rule_dpif *rule,
4612 const struct flow *flow, const struct dpif_flow_stats *stats)
4614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4615 struct action_xlate_ctx ctx;
4617 ofproto_rule_update_used(&rule->up, stats->used);
4619 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4621 ctx.resubmit_stats = stats;
4622 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4623 rule->up.ofpacts_len);
4628 static struct subfacet *
4629 subfacet_find(struct ofproto_dpif *ofproto,
4630 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4631 const struct flow *flow)
4633 struct subfacet *subfacet;
4635 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4636 &ofproto->subfacets) {
4638 ? (subfacet->key_len == key_len
4639 && !memcmp(key, subfacet->key, key_len))
4640 : flow_equal(flow, &subfacet->facet->flow)) {
4648 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4649 * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if
4650 * there is one, otherwise creates and returns a new subfacet.
4652 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4653 * which case the caller must populate the actions with
4654 * subfacet_make_actions(). */
4655 static struct subfacet *
4656 subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness,
4657 const struct nlattr *key, size_t key_len,
4658 ovs_be16 initial_tci, long long int now)
4660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4661 uint32_t key_hash = odp_flow_key_hash(key, key_len);
4662 struct subfacet *subfacet;
4664 if (list_is_empty(&facet->subfacets)) {
4665 subfacet = &facet->one_subfacet;
4667 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4670 if (subfacet->facet == facet) {
4674 /* This shouldn't happen. */
4675 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4676 subfacet_destroy(subfacet);
4679 subfacet = xmalloc(sizeof *subfacet);
4682 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
4683 list_push_back(&facet->subfacets, &subfacet->list_node);
4684 subfacet->facet = facet;
4685 subfacet->key_fitness = key_fitness;
4686 if (key_fitness != ODP_FIT_PERFECT) {
4687 subfacet->key = xmemdup(key, key_len);
4688 subfacet->key_len = key_len;
4690 subfacet->key = NULL;
4691 subfacet->key_len = 0;
4693 subfacet->used = now;
4694 subfacet->dp_packet_count = 0;
4695 subfacet->dp_byte_count = 0;
4696 subfacet->actions_len = 0;
4697 subfacet->actions = NULL;
4698 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
4701 subfacet->path = SF_NOT_INSTALLED;
4702 subfacet->initial_tci = initial_tci;
4707 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4708 * its facet within 'ofproto', and frees it. */
4710 subfacet_destroy__(struct subfacet *subfacet)
4712 struct facet *facet = subfacet->facet;
4713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4715 subfacet_uninstall(subfacet);
4716 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
4717 list_remove(&subfacet->list_node);
4718 free(subfacet->key);
4719 free(subfacet->actions);
4720 if (subfacet != &facet->one_subfacet) {
4725 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4726 * last remaining subfacet in its facet destroys the facet too. */
4728 subfacet_destroy(struct subfacet *subfacet)
4730 struct facet *facet = subfacet->facet;
4732 if (list_is_singleton(&facet->subfacets)) {
4733 /* facet_remove() needs at least one subfacet (it will remove it). */
4734 facet_remove(facet);
4736 subfacet_destroy__(subfacet);
4741 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
4742 struct subfacet **subfacets, int n)
4744 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
4745 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4746 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4747 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
4748 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4751 for (i = 0; i < n; i++) {
4752 ops[i].type = DPIF_OP_FLOW_DEL;
4753 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
4754 ops[i].u.flow_del.key = keys[i].data;
4755 ops[i].u.flow_del.key_len = keys[i].size;
4756 ops[i].u.flow_del.stats = &stats[i];
4760 dpif_operate(ofproto->backer->dpif, opsp, n);
4761 for (i = 0; i < n; i++) {
4762 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4763 subfacets[i]->path = SF_NOT_INSTALLED;
4764 subfacet_destroy(subfacets[i]);
4768 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4769 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4770 * for use as temporary storage. */
4772 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
4776 if (!subfacet->key) {
4777 struct ofproto_dpif *ofproto;
4778 struct flow *flow = &subfacet->facet->flow;
4780 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
4781 ofproto = ofproto_dpif_cast(subfacet->facet->rule->up.ofproto);
4782 odp_flow_key_from_flow(key, flow,
4783 ofp_port_to_odp_port(ofproto, flow->in_port));
4785 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
4789 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4790 * Translates the actions into 'odp_actions', which the caller must have
4791 * initialized and is responsible for uninitializing. */
4793 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
4794 struct ofpbuf *odp_actions)
4796 struct facet *facet = subfacet->facet;
4797 struct rule_dpif *rule = facet->rule;
4798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4800 struct action_xlate_ctx ctx;
4802 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
4804 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
4805 facet->tags = ctx.tags;
4806 facet->has_learn = ctx.has_learn;
4807 facet->has_normal = ctx.has_normal;
4808 facet->has_fin_timeout = ctx.has_fin_timeout;
4809 facet->nf_flow.output_iface = ctx.nf_output_iface;
4810 facet->mirrors = ctx.mirrors;
4812 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4813 if (subfacet->actions_len != odp_actions->size
4814 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
4815 free(subfacet->actions);
4816 subfacet->actions_len = odp_actions->size;
4817 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
4821 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4822 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4823 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4824 * since 'subfacet' was last updated.
4826 * Returns 0 if successful, otherwise a positive errno value. */
4828 subfacet_install(struct subfacet *subfacet,
4829 const struct nlattr *actions, size_t actions_len,
4830 struct dpif_flow_stats *stats,
4831 enum slow_path_reason slow)
4833 struct facet *facet = subfacet->facet;
4834 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4835 enum subfacet_path path = subfacet_want_path(slow);
4836 uint64_t slow_path_stub[128 / 8];
4837 struct odputil_keybuf keybuf;
4838 enum dpif_flow_put_flags flags;
4842 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
4844 flags |= DPIF_FP_ZERO_STATS;
4847 if (path == SF_SLOW_PATH) {
4848 compose_slow_path(ofproto, &facet->flow, slow,
4849 slow_path_stub, sizeof slow_path_stub,
4850 &actions, &actions_len);
4853 subfacet_get_key(subfacet, &keybuf, &key);
4854 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
4855 actions, actions_len, stats);
4858 subfacet_reset_dp_stats(subfacet, stats);
4862 subfacet->path = path;
4868 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
4870 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
4871 stats, subfacet->slow);
4874 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4876 subfacet_uninstall(struct subfacet *subfacet)
4878 if (subfacet->path != SF_NOT_INSTALLED) {
4879 struct rule_dpif *rule = subfacet->facet->rule;
4880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4881 struct odputil_keybuf keybuf;
4882 struct dpif_flow_stats stats;
4886 subfacet_get_key(subfacet, &keybuf, &key);
4887 error = dpif_flow_del(ofproto->backer->dpif,
4888 key.data, key.size, &stats);
4889 subfacet_reset_dp_stats(subfacet, &stats);
4891 subfacet_update_stats(subfacet, &stats);
4893 subfacet->path = SF_NOT_INSTALLED;
4895 assert(subfacet->dp_packet_count == 0);
4896 assert(subfacet->dp_byte_count == 0);
4900 /* Resets 'subfacet''s datapath statistics counters. This should be called
4901 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4902 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4903 * was reset in the datapath. 'stats' will be modified to include only
4904 * statistics new since 'subfacet' was last updated. */
4906 subfacet_reset_dp_stats(struct subfacet *subfacet,
4907 struct dpif_flow_stats *stats)
4910 && subfacet->dp_packet_count <= stats->n_packets
4911 && subfacet->dp_byte_count <= stats->n_bytes) {
4912 stats->n_packets -= subfacet->dp_packet_count;
4913 stats->n_bytes -= subfacet->dp_byte_count;
4916 subfacet->dp_packet_count = 0;
4917 subfacet->dp_byte_count = 0;
4920 /* Updates 'subfacet''s used time. The caller is responsible for calling
4921 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4923 subfacet_update_time(struct subfacet *subfacet, long long int used)
4925 if (used > subfacet->used) {
4926 subfacet->used = used;
4927 facet_update_time(subfacet->facet, used);
4931 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4933 * Because of the meaning of a subfacet's counters, it only makes sense to do
4934 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4935 * represents a packet that was sent by hand or if it represents statistics
4936 * that have been cleared out of the datapath. */
4938 subfacet_update_stats(struct subfacet *subfacet,
4939 const struct dpif_flow_stats *stats)
4941 if (stats->n_packets || stats->used > subfacet->used) {
4942 struct facet *facet = subfacet->facet;
4944 subfacet_update_time(subfacet, stats->used);
4945 facet->packet_count += stats->n_packets;
4946 facet->byte_count += stats->n_bytes;
4947 facet->tcp_flags |= stats->tcp_flags;
4948 facet_push_stats(facet);
4949 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4955 static struct rule_dpif *
4956 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
4958 struct rule_dpif *rule;
4960 rule = rule_dpif_lookup__(ofproto, flow, 0);
4965 return rule_dpif_miss_rule(ofproto, flow);
4968 static struct rule_dpif *
4969 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
4972 struct cls_rule *cls_rule;
4973 struct classifier *cls;
4975 if (table_id >= N_TABLES) {
4979 cls = &ofproto->up.tables[table_id].cls;
4980 if (flow->nw_frag & FLOW_NW_FRAG_ANY
4981 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4982 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
4983 * are unavailable. */
4984 struct flow ofpc_normal_flow = *flow;
4985 ofpc_normal_flow.tp_src = htons(0);
4986 ofpc_normal_flow.tp_dst = htons(0);
4987 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
4989 cls_rule = classifier_lookup(cls, flow);
4991 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
4994 static struct rule_dpif *
4995 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
4997 struct ofport_dpif *port;
4999 port = get_ofp_port(ofproto, flow->in_port);
5001 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5002 return ofproto->miss_rule;
5005 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5006 return ofproto->no_packet_in_rule;
5008 return ofproto->miss_rule;
5012 complete_operation(struct rule_dpif *rule)
5014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5016 rule_invalidate(rule);
5018 struct dpif_completion *c = xmalloc(sizeof *c);
5019 c->op = rule->up.pending;
5020 list_push_back(&ofproto->completions, &c->list_node);
5022 ofoperation_complete(rule->up.pending, 0);
5026 static struct rule *
5029 struct rule_dpif *rule = xmalloc(sizeof *rule);
5034 rule_dealloc(struct rule *rule_)
5036 struct rule_dpif *rule = rule_dpif_cast(rule_);
5041 rule_construct(struct rule *rule_)
5043 struct rule_dpif *rule = rule_dpif_cast(rule_);
5044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5045 struct rule_dpif *victim;
5048 rule->packet_count = 0;
5049 rule->byte_count = 0;
5051 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5052 if (victim && !list_is_empty(&victim->facets)) {
5053 struct facet *facet;
5055 rule->facets = victim->facets;
5056 list_moved(&rule->facets);
5057 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5058 /* XXX: We're only clearing our local counters here. It's possible
5059 * that quite a few packets are unaccounted for in the datapath
5060 * statistics. These will be accounted to the new rule instead of
5061 * cleared as required. This could be fixed by clearing out the
5062 * datapath statistics for this facet, but currently it doesn't
5064 facet_reset_counters(facet);
5068 /* Must avoid list_moved() in this case. */
5069 list_init(&rule->facets);
5072 table_id = rule->up.table_id;
5074 rule->tag = victim->tag;
5075 } else if (table_id == 0) {
5080 miniflow_expand(&rule->up.cr.match.flow, &flow);
5081 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5082 ofproto->tables[table_id].basis);
5085 complete_operation(rule);
5090 rule_destruct(struct rule *rule_)
5092 struct rule_dpif *rule = rule_dpif_cast(rule_);
5093 struct facet *facet, *next_facet;
5095 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5096 facet_revalidate(facet);
5099 complete_operation(rule);
5103 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5105 struct rule_dpif *rule = rule_dpif_cast(rule_);
5106 struct facet *facet;
5108 /* Start from historical data for 'rule' itself that are no longer tracked
5109 * in facets. This counts, for example, facets that have expired. */
5110 *packets = rule->packet_count;
5111 *bytes = rule->byte_count;
5113 /* Add any statistics that are tracked by facets. This includes
5114 * statistical data recently updated by ofproto_update_stats() as well as
5115 * stats for packets that were executed "by hand" via dpif_execute(). */
5116 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5117 *packets += facet->packet_count;
5118 *bytes += facet->byte_count;
5123 rule_execute(struct rule *rule_, const struct flow *flow,
5124 struct ofpbuf *packet)
5126 struct rule_dpif *rule = rule_dpif_cast(rule_);
5127 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5129 struct dpif_flow_stats stats;
5131 struct action_xlate_ctx ctx;
5132 uint64_t odp_actions_stub[1024 / 8];
5133 struct ofpbuf odp_actions;
5135 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5136 rule_credit_stats(rule, &stats);
5138 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5139 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5140 rule, stats.tcp_flags, packet);
5141 ctx.resubmit_stats = &stats;
5142 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5144 execute_odp_actions(ofproto, flow, odp_actions.data,
5145 odp_actions.size, packet);
5147 ofpbuf_uninit(&odp_actions);
5153 rule_modify_actions(struct rule *rule_)
5155 struct rule_dpif *rule = rule_dpif_cast(rule_);
5157 complete_operation(rule);
5160 /* Sends 'packet' out 'ofport'.
5161 * May modify 'packet'.
5162 * Returns 0 if successful, otherwise a positive errno value. */
5164 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5166 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5167 struct ofpbuf key, odp_actions;
5168 struct odputil_keybuf keybuf;
5173 flow_extract(packet, 0, NULL, 0, &flow);
5174 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5176 if (odp_port != ofport->odp_port) {
5177 eth_pop_vlan(packet);
5178 flow.vlan_tci = htons(0);
5181 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5182 odp_flow_key_from_flow(&key, &flow,
5183 ofp_port_to_odp_port(ofproto, flow.in_port));
5185 ofpbuf_init(&odp_actions, 32);
5186 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5188 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5189 error = dpif_execute(ofproto->backer->dpif,
5191 odp_actions.data, odp_actions.size,
5193 ofpbuf_uninit(&odp_actions);
5196 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5197 ofproto->up.name, odp_port, strerror(error));
5199 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5203 /* OpenFlow to datapath action translation. */
5205 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5206 struct action_xlate_ctx *);
5207 static void xlate_normal(struct action_xlate_ctx *);
5209 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5210 * The action will state 'slow' as the reason that the action is in the slow
5211 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5212 * dump-flows" output to see why a flow is in the slow path.)
5214 * The 'stub_size' bytes in 'stub' will be used to store the action.
5215 * 'stub_size' must be large enough for the action.
5217 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5220 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5221 enum slow_path_reason slow,
5222 uint64_t *stub, size_t stub_size,
5223 const struct nlattr **actionsp, size_t *actions_lenp)
5225 union user_action_cookie cookie;
5228 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5229 cookie.slow_path.unused = 0;
5230 cookie.slow_path.reason = slow;
5232 ofpbuf_use_stack(&buf, stub, stub_size);
5233 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5234 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX);
5235 odp_put_userspace_action(pid, &cookie, &buf);
5237 put_userspace_action(ofproto, &buf, flow, &cookie);
5239 *actionsp = buf.data;
5240 *actions_lenp = buf.size;
5244 put_userspace_action(const struct ofproto_dpif *ofproto,
5245 struct ofpbuf *odp_actions,
5246 const struct flow *flow,
5247 const union user_action_cookie *cookie)
5251 pid = dpif_port_get_pid(ofproto->backer->dpif,
5252 ofp_port_to_odp_port(ofproto, flow->in_port));
5254 return odp_put_userspace_action(pid, cookie, odp_actions);
5258 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5259 ovs_be16 vlan_tci, uint32_t odp_port,
5260 unsigned int n_outputs, union user_action_cookie *cookie)
5264 cookie->type = USER_ACTION_COOKIE_SFLOW;
5265 cookie->sflow.vlan_tci = vlan_tci;
5267 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5268 * port information") for the interpretation of cookie->output. */
5269 switch (n_outputs) {
5271 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5272 cookie->sflow.output = 0x40000000 | 256;
5276 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5278 cookie->sflow.output = ifindex;
5283 /* 0x80000000 means "multiple output ports. */
5284 cookie->sflow.output = 0x80000000 | n_outputs;
5289 /* Compose SAMPLE action for sFlow. */
5291 compose_sflow_action(const struct ofproto_dpif *ofproto,
5292 struct ofpbuf *odp_actions,
5293 const struct flow *flow,
5296 uint32_t probability;
5297 union user_action_cookie cookie;
5298 size_t sample_offset, actions_offset;
5301 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5305 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5307 /* Number of packets out of UINT_MAX to sample. */
5308 probability = dpif_sflow_get_probability(ofproto->sflow);
5309 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5311 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5312 compose_sflow_cookie(ofproto, htons(0), odp_port,
5313 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5314 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5316 nl_msg_end_nested(odp_actions, actions_offset);
5317 nl_msg_end_nested(odp_actions, sample_offset);
5318 return cookie_offset;
5321 /* SAMPLE action must be first action in any given list of actions.
5322 * At this point we do not have all information required to build it. So try to
5323 * build sample action as complete as possible. */
5325 add_sflow_action(struct action_xlate_ctx *ctx)
5327 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5329 &ctx->flow, OVSP_NONE);
5330 ctx->sflow_odp_port = 0;
5331 ctx->sflow_n_outputs = 0;
5334 /* Fix SAMPLE action according to data collected while composing ODP actions.
5335 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5336 * USERSPACE action's user-cookie which is required for sflow. */
5338 fix_sflow_action(struct action_xlate_ctx *ctx)
5340 const struct flow *base = &ctx->base_flow;
5341 union user_action_cookie *cookie;
5343 if (!ctx->user_cookie_offset) {
5347 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5349 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5351 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5352 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5356 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5359 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5360 uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5361 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5362 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5366 struct priority_to_dscp *pdscp;
5368 if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5369 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5371 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5372 xlate_report(ctx, "STP not in forwarding state, skipping output");
5376 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5378 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5379 ctx->flow.nw_tos |= pdscp->dscp;
5382 /* We may not have an ofport record for this port, but it doesn't hurt
5383 * to allow forwarding to it anyhow. Maybe such a port will appear
5384 * later and we're pre-populating the flow table. */
5387 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5388 ctx->flow.vlan_tci);
5389 if (out_port != odp_port) {
5390 ctx->flow.vlan_tci = htons(0);
5392 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5393 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5395 ctx->sflow_odp_port = odp_port;
5396 ctx->sflow_n_outputs++;
5397 ctx->nf_output_iface = ofp_port;
5398 ctx->flow.vlan_tci = flow_vlan_tci;
5399 ctx->flow.nw_tos = flow_nw_tos;
5403 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5405 compose_output_action__(ctx, ofp_port, true);
5409 xlate_table_action(struct action_xlate_ctx *ctx,
5410 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5412 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5413 struct ofproto_dpif *ofproto = ctx->ofproto;
5414 struct rule_dpif *rule;
5415 uint16_t old_in_port;
5416 uint8_t old_table_id;
5418 old_table_id = ctx->table_id;
5419 ctx->table_id = table_id;
5421 /* Look up a flow with 'in_port' as the input port. */
5422 old_in_port = ctx->flow.in_port;
5423 ctx->flow.in_port = in_port;
5424 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5427 if (table_id > 0 && table_id < N_TABLES) {
5428 struct table_dpif *table = &ofproto->tables[table_id];
5429 if (table->other_table) {
5430 ctx->tags |= (rule && rule->tag
5432 : rule_calculate_tag(&ctx->flow,
5433 &table->other_table->mask,
5438 /* Restore the original input port. Otherwise OFPP_NORMAL and
5439 * OFPP_IN_PORT will have surprising behavior. */
5440 ctx->flow.in_port = old_in_port;
5442 if (ctx->resubmit_hook) {
5443 ctx->resubmit_hook(ctx, rule);
5446 if (rule == NULL && may_packet_in) {
5448 * check if table configuration flags
5449 * OFPTC_TABLE_MISS_CONTROLLER, default.
5450 * OFPTC_TABLE_MISS_CONTINUE,
5451 * OFPTC_TABLE_MISS_DROP
5452 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5454 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5458 struct rule_dpif *old_rule = ctx->rule;
5460 if (ctx->resubmit_stats) {
5461 rule_credit_stats(rule, ctx->resubmit_stats);
5466 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5467 ctx->rule = old_rule;
5471 ctx->table_id = old_table_id;
5473 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5475 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5476 MAX_RESUBMIT_RECURSION);
5477 ctx->max_resubmit_trigger = true;
5482 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5483 const struct ofpact_resubmit *resubmit)
5488 in_port = resubmit->in_port;
5489 if (in_port == OFPP_IN_PORT) {
5490 in_port = ctx->flow.in_port;
5493 table_id = resubmit->table_id;
5494 if (table_id == 255) {
5495 table_id = ctx->table_id;
5498 xlate_table_action(ctx, in_port, table_id, false);
5502 flood_packets(struct action_xlate_ctx *ctx, bool all)
5504 struct ofport_dpif *ofport;
5506 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5507 uint16_t ofp_port = ofport->up.ofp_port;
5509 if (ofp_port == ctx->flow.in_port) {
5514 compose_output_action__(ctx, ofp_port, false);
5515 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5516 compose_output_action(ctx, ofp_port);
5520 ctx->nf_output_iface = NF_OUT_FLOOD;
5524 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5525 enum ofp_packet_in_reason reason,
5526 uint16_t controller_id)
5528 struct ofputil_packet_in pin;
5529 struct ofpbuf *packet;
5531 ctx->slow |= SLOW_CONTROLLER;
5536 packet = ofpbuf_clone(ctx->packet);
5538 if (packet->l2 && packet->l3) {
5539 struct eth_header *eh;
5541 eth_pop_vlan(packet);
5544 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5545 * LLC frame. Calculating the Ethernet type of these frames is more
5546 * trouble than seems appropriate for a simple assertion. */
5547 assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5548 || eh->eth_type == ctx->flow.dl_type);
5550 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5551 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5553 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5554 eth_push_vlan(packet, ctx->flow.vlan_tci);
5558 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5559 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5560 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5564 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5565 packet_set_tcp_port(packet, ctx->flow.tp_src,
5567 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5568 packet_set_udp_port(packet, ctx->flow.tp_src,
5575 pin.packet = packet->data;
5576 pin.packet_len = packet->size;
5577 pin.reason = reason;
5578 pin.controller_id = controller_id;
5579 pin.table_id = ctx->table_id;
5580 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5583 flow_get_metadata(&ctx->flow, &pin.fmd);
5585 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
5586 ofpbuf_delete(packet);
5590 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
5592 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
5593 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
5597 if (ctx->flow.nw_ttl > 1) {
5603 for (i = 0; i < ids->n_controllers; i++) {
5604 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
5608 /* Stop processing for current table. */
5614 xlate_output_action(struct action_xlate_ctx *ctx,
5615 uint16_t port, uint16_t max_len, bool may_packet_in)
5617 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
5619 ctx->nf_output_iface = NF_OUT_DROP;
5623 compose_output_action(ctx, ctx->flow.in_port);
5626 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
5632 flood_packets(ctx, false);
5635 flood_packets(ctx, true);
5637 case OFPP_CONTROLLER:
5638 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
5644 if (port != ctx->flow.in_port) {
5645 compose_output_action(ctx, port);
5647 xlate_report(ctx, "skipping output to input port");
5652 if (prev_nf_output_iface == NF_OUT_FLOOD) {
5653 ctx->nf_output_iface = NF_OUT_FLOOD;
5654 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
5655 ctx->nf_output_iface = prev_nf_output_iface;
5656 } else if (prev_nf_output_iface != NF_OUT_DROP &&
5657 ctx->nf_output_iface != NF_OUT_FLOOD) {
5658 ctx->nf_output_iface = NF_OUT_MULTI;
5663 xlate_output_reg_action(struct action_xlate_ctx *ctx,
5664 const struct ofpact_output_reg *or)
5666 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
5667 if (port <= UINT16_MAX) {
5668 xlate_output_action(ctx, port, or->max_len, false);
5673 xlate_enqueue_action(struct action_xlate_ctx *ctx,
5674 const struct ofpact_enqueue *enqueue)
5676 uint16_t ofp_port = enqueue->port;
5677 uint32_t queue_id = enqueue->queue;
5678 uint32_t flow_priority, priority;
5681 /* Translate queue to priority. */
5682 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5683 queue_id, &priority);
5685 /* Fall back to ordinary output action. */
5686 xlate_output_action(ctx, enqueue->port, 0, false);
5690 /* Check output port. */
5691 if (ofp_port == OFPP_IN_PORT) {
5692 ofp_port = ctx->flow.in_port;
5693 } else if (ofp_port == ctx->flow.in_port) {
5697 /* Add datapath actions. */
5698 flow_priority = ctx->flow.skb_priority;
5699 ctx->flow.skb_priority = priority;
5700 compose_output_action(ctx, ofp_port);
5701 ctx->flow.skb_priority = flow_priority;
5703 /* Update NetFlow output port. */
5704 if (ctx->nf_output_iface == NF_OUT_DROP) {
5705 ctx->nf_output_iface = ofp_port;
5706 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
5707 ctx->nf_output_iface = NF_OUT_MULTI;
5712 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
5714 uint32_t skb_priority;
5716 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
5717 queue_id, &skb_priority)) {
5718 ctx->flow.skb_priority = skb_priority;
5720 /* Couldn't translate queue to a priority. Nothing to do. A warning
5721 * has already been logged. */
5725 struct xlate_reg_state {
5731 xlate_autopath(struct action_xlate_ctx *ctx,
5732 const struct ofpact_autopath *ap)
5734 uint16_t ofp_port = ap->port;
5735 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
5737 if (!port || !port->bundle) {
5738 ofp_port = OFPP_NONE;
5739 } else if (port->bundle->bond) {
5740 /* Autopath does not support VLAN hashing. */
5741 struct ofport_dpif *slave = bond_choose_output_slave(
5742 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
5744 ofp_port = slave->up.ofp_port;
5747 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
5751 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
5753 struct ofproto_dpif *ofproto = ofproto_;
5754 struct ofport_dpif *port;
5764 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
5767 port = get_ofp_port(ofproto, ofp_port);
5768 return port ? port->may_enable : false;
5773 xlate_bundle_action(struct action_xlate_ctx *ctx,
5774 const struct ofpact_bundle *bundle)
5778 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
5779 if (bundle->dst.field) {
5780 nxm_reg_load(&bundle->dst, port, &ctx->flow);
5782 xlate_output_action(ctx, port, 0, false);
5787 xlate_learn_action(struct action_xlate_ctx *ctx,
5788 const struct ofpact_learn *learn)
5790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
5791 struct ofputil_flow_mod fm;
5792 uint64_t ofpacts_stub[1024 / 8];
5793 struct ofpbuf ofpacts;
5796 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5797 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
5799 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
5800 if (error && !VLOG_DROP_WARN(&rl)) {
5801 VLOG_WARN("learning action failed to modify flow table (%s)",
5802 ofperr_get_name(error));
5805 ofpbuf_uninit(&ofpacts);
5808 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5809 * means "infinite". */
5811 reduce_timeout(uint16_t max, uint16_t *timeout)
5813 if (max && (!*timeout || *timeout > max)) {
5819 xlate_fin_timeout(struct action_xlate_ctx *ctx,
5820 const struct ofpact_fin_timeout *oft)
5822 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
5823 struct rule_dpif *rule = ctx->rule;
5825 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
5826 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
5831 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
5833 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
5834 ? OFPUTIL_PC_NO_RECV_STP
5835 : OFPUTIL_PC_NO_RECV)) {
5839 /* Only drop packets here if both forwarding and learning are
5840 * disabled. If just learning is enabled, we need to have
5841 * OFPP_NORMAL and the learning action have a look at the packet
5842 * before we can drop it. */
5843 if (!stp_forward_in_state(port->stp_state)
5844 && !stp_learn_in_state(port->stp_state)) {
5852 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
5853 struct action_xlate_ctx *ctx)
5855 const struct ofport_dpif *port;
5856 bool was_evictable = true;
5857 const struct ofpact *a;
5859 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5860 if (port && !may_receive(port, ctx)) {
5861 /* Drop this flow. */
5866 /* Don't let the rule we're working on get evicted underneath us. */
5867 was_evictable = ctx->rule->up.evictable;
5868 ctx->rule->up.evictable = false;
5870 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
5871 struct ofpact_controller *controller;
5872 const struct ofpact_metadata *metadata;
5880 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
5881 ofpact_get_OUTPUT(a)->max_len, true);
5884 case OFPACT_CONTROLLER:
5885 controller = ofpact_get_CONTROLLER(a);
5886 execute_controller_action(ctx, controller->max_len,
5888 controller->controller_id);
5891 case OFPACT_ENQUEUE:
5892 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
5895 case OFPACT_SET_VLAN_VID:
5896 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
5897 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
5901 case OFPACT_SET_VLAN_PCP:
5902 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
5903 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
5908 case OFPACT_STRIP_VLAN:
5909 ctx->flow.vlan_tci = htons(0);
5912 case OFPACT_PUSH_VLAN:
5913 /* TODO:XXX 802.1AD(QinQ) */
5914 ctx->flow.vlan_tci = htons(VLAN_CFI);
5917 case OFPACT_SET_ETH_SRC:
5918 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
5922 case OFPACT_SET_ETH_DST:
5923 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
5927 case OFPACT_SET_IPV4_SRC:
5928 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
5931 case OFPACT_SET_IPV4_DST:
5932 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
5935 case OFPACT_SET_IPV4_DSCP:
5936 /* OpenFlow 1.0 only supports IPv4. */
5937 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5938 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5939 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
5943 case OFPACT_SET_L4_SRC_PORT:
5944 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
5947 case OFPACT_SET_L4_DST_PORT:
5948 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
5951 case OFPACT_RESUBMIT:
5952 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
5955 case OFPACT_SET_TUNNEL:
5956 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
5959 case OFPACT_SET_QUEUE:
5960 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
5963 case OFPACT_POP_QUEUE:
5964 ctx->flow.skb_priority = ctx->orig_skb_priority;
5967 case OFPACT_REG_MOVE:
5968 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
5971 case OFPACT_REG_LOAD:
5972 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
5975 case OFPACT_DEC_TTL:
5976 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
5982 /* Nothing to do. */
5985 case OFPACT_MULTIPATH:
5986 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
5989 case OFPACT_AUTOPATH:
5990 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
5994 ctx->ofproto->has_bundle_action = true;
5995 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
5998 case OFPACT_OUTPUT_REG:
5999 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6003 ctx->has_learn = true;
6004 if (ctx->may_learn) {
6005 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6013 case OFPACT_FIN_TIMEOUT:
6014 ctx->has_fin_timeout = true;
6015 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6018 case OFPACT_CLEAR_ACTIONS:
6020 * Nothing to do because writa-actions is not supported for now.
6021 * When writa-actions is supported, clear-actions also must
6022 * be supported at the same time.
6026 case OFPACT_WRITE_METADATA:
6027 metadata = ofpact_get_WRITE_METADATA(a);
6028 ctx->flow.metadata &= ~metadata->mask;
6029 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6032 case OFPACT_GOTO_TABLE: {
6033 /* TODO:XXX remove recursion */
6034 /* It is assumed that goto-table is last action */
6035 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6036 assert(ctx->table_id < ogt->table_id);
6037 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6044 /* We've let OFPP_NORMAL and the learning action look at the packet,
6045 * so drop it now if forwarding is disabled. */
6046 if (port && !stp_forward_in_state(port->stp_state)) {
6047 ofpbuf_clear(ctx->odp_actions);
6048 add_sflow_action(ctx);
6051 ctx->rule->up.evictable = was_evictable;
6056 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6057 struct ofproto_dpif *ofproto, const struct flow *flow,
6058 ovs_be16 initial_tci, struct rule_dpif *rule,
6059 uint8_t tcp_flags, const struct ofpbuf *packet)
6061 ctx->ofproto = ofproto;
6063 ctx->base_flow = ctx->flow;
6064 memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel);
6065 ctx->base_flow.vlan_tci = initial_tci;
6067 ctx->packet = packet;
6068 ctx->may_learn = packet != NULL;
6069 ctx->tcp_flags = tcp_flags;
6070 ctx->resubmit_hook = NULL;
6071 ctx->report_hook = NULL;
6072 ctx->resubmit_stats = NULL;
6075 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6076 * into datapath actions in 'odp_actions', using 'ctx'. */
6078 xlate_actions(struct action_xlate_ctx *ctx,
6079 const struct ofpact *ofpacts, size_t ofpacts_len,
6080 struct ofpbuf *odp_actions)
6082 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6083 * that in the future we always keep a copy of the original flow for
6084 * tracing purposes. */
6085 static bool hit_resubmit_limit;
6087 enum slow_path_reason special;
6089 COVERAGE_INC(ofproto_dpif_xlate);
6091 ofpbuf_clear(odp_actions);
6092 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6094 ctx->odp_actions = odp_actions;
6097 ctx->has_learn = false;
6098 ctx->has_normal = false;
6099 ctx->has_fin_timeout = false;
6100 ctx->nf_output_iface = NF_OUT_DROP;
6103 ctx->max_resubmit_trigger = false;
6104 ctx->orig_skb_priority = ctx->flow.skb_priority;
6108 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6109 /* Do this conditionally because the copy is expensive enough that it
6110 * shows up in profiles.
6112 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6113 * believe that I wasn't using it without initializing it if I kept it
6114 * in a local variable. */
6115 ctx->orig_flow = ctx->flow;
6118 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6119 switch (ctx->ofproto->up.frag_handling) {
6120 case OFPC_FRAG_NORMAL:
6121 /* We must pretend that transport ports are unavailable. */
6122 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6123 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6126 case OFPC_FRAG_DROP:
6129 case OFPC_FRAG_REASM:
6132 case OFPC_FRAG_NX_MATCH:
6133 /* Nothing to do. */
6136 case OFPC_INVALID_TTL_TO_CONTROLLER:
6141 special = process_special(ctx->ofproto, &ctx->flow, ctx->packet);
6143 ctx->slow |= special;
6145 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6146 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6148 add_sflow_action(ctx);
6149 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6151 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6152 if (!hit_resubmit_limit) {
6153 /* We didn't record the original flow. Make sure we do from
6155 hit_resubmit_limit = true;
6156 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6157 struct ds ds = DS_EMPTY_INITIALIZER;
6159 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6161 VLOG_ERR("Trace triggered by excessive resubmit "
6162 "recursion:\n%s", ds_cstr(&ds));
6167 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6168 ctx->odp_actions->data,
6169 ctx->odp_actions->size)) {
6170 ctx->slow |= SLOW_IN_BAND;
6172 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6174 compose_output_action(ctx, OFPP_LOCAL);
6177 if (ctx->ofproto->has_mirrors) {
6178 add_mirror_actions(ctx, &ctx->orig_flow);
6180 fix_sflow_action(ctx);
6184 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6185 * into datapath actions, using 'ctx', and discards the datapath actions. */
6187 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6188 const struct ofpact *ofpacts,
6191 uint64_t odp_actions_stub[1024 / 8];
6192 struct ofpbuf odp_actions;
6194 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6195 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6196 ofpbuf_uninit(&odp_actions);
6200 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6202 if (ctx->report_hook) {
6203 ctx->report_hook(ctx, s);
6207 /* OFPP_NORMAL implementation. */
6209 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6211 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6212 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6213 * the bundle on which the packet was received, returns the VLAN to which the
6216 * Both 'vid' and the return value are in the range 0...4095. */
6218 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6220 switch (in_bundle->vlan_mode) {
6221 case PORT_VLAN_ACCESS:
6222 return in_bundle->vlan;
6225 case PORT_VLAN_TRUNK:
6228 case PORT_VLAN_NATIVE_UNTAGGED:
6229 case PORT_VLAN_NATIVE_TAGGED:
6230 return vid ? vid : in_bundle->vlan;
6237 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6238 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6241 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6242 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6245 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6247 /* Allow any VID on the OFPP_NONE port. */
6248 if (in_bundle == &ofpp_none_bundle) {
6252 switch (in_bundle->vlan_mode) {
6253 case PORT_VLAN_ACCESS:
6256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6257 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6258 "packet received on port %s configured as VLAN "
6259 "%"PRIu16" access port",
6260 in_bundle->ofproto->up.name, vid,
6261 in_bundle->name, in_bundle->vlan);
6267 case PORT_VLAN_NATIVE_UNTAGGED:
6268 case PORT_VLAN_NATIVE_TAGGED:
6270 /* Port must always carry its native VLAN. */
6274 case PORT_VLAN_TRUNK:
6275 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6277 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6278 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6279 "received on port %s not configured for trunking "
6281 in_bundle->ofproto->up.name, vid,
6282 in_bundle->name, vid);
6294 /* Given 'vlan', the VLAN that a packet belongs to, and
6295 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6296 * that should be included in the 802.1Q header. (If the return value is 0,
6297 * then the 802.1Q header should only be included in the packet if there is a
6300 * Both 'vlan' and the return value are in the range 0...4095. */
6302 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6304 switch (out_bundle->vlan_mode) {
6305 case PORT_VLAN_ACCESS:
6308 case PORT_VLAN_TRUNK:
6309 case PORT_VLAN_NATIVE_TAGGED:
6312 case PORT_VLAN_NATIVE_UNTAGGED:
6313 return vlan == out_bundle->vlan ? 0 : vlan;
6321 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6324 struct ofport_dpif *port;
6326 ovs_be16 tci, old_tci;
6328 vid = output_vlan_to_vid(out_bundle, vlan);
6329 if (!out_bundle->bond) {
6330 port = ofbundle_get_a_port(out_bundle);
6332 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6335 /* No slaves enabled, so drop packet. */
6340 old_tci = ctx->flow.vlan_tci;
6342 if (tci || out_bundle->use_priority_tags) {
6343 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6345 tci |= htons(VLAN_CFI);
6348 ctx->flow.vlan_tci = tci;
6350 compose_output_action(ctx, port->up.ofp_port);
6351 ctx->flow.vlan_tci = old_tci;
6355 mirror_mask_ffs(mirror_mask_t mask)
6357 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6362 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6364 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6365 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6369 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6371 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6374 /* Returns an arbitrary interface within 'bundle'. */
6375 static struct ofport_dpif *
6376 ofbundle_get_a_port(const struct ofbundle *bundle)
6378 return CONTAINER_OF(list_front(&bundle->ports),
6379 struct ofport_dpif, bundle_node);
6383 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6385 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6389 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6391 struct ofproto_dpif *ofproto = ctx->ofproto;
6392 mirror_mask_t mirrors;
6393 struct ofbundle *in_bundle;
6396 const struct nlattr *a;
6399 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6400 ctx->packet != NULL, NULL);
6404 mirrors = in_bundle->src_mirrors;
6406 /* Drop frames on bundles reserved for mirroring. */
6407 if (in_bundle->mirror_out) {
6408 if (ctx->packet != NULL) {
6409 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6410 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6411 "%s, which is reserved exclusively for mirroring",
6412 ctx->ofproto->up.name, in_bundle->name);
6418 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6419 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6422 vlan = input_vid_to_vlan(in_bundle, vid);
6424 /* Look at the output ports to check for destination selections. */
6426 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6427 ctx->odp_actions->size) {
6428 enum ovs_action_attr type = nl_attr_type(a);
6429 struct ofport_dpif *ofport;
6431 if (type != OVS_ACTION_ATTR_OUTPUT) {
6435 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6436 if (ofport && ofport->bundle) {
6437 mirrors |= ofport->bundle->dst_mirrors;
6445 /* Restore the original packet before adding the mirror actions. */
6446 ctx->flow = *orig_flow;
6451 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6453 if (!vlan_is_mirrored(m, vlan)) {
6454 mirrors = zero_rightmost_1bit(mirrors);
6458 mirrors &= ~m->dup_mirrors;
6459 ctx->mirrors |= m->dup_mirrors;
6461 output_normal(ctx, m->out, vlan);
6462 } else if (vlan != m->out_vlan
6463 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6464 struct ofbundle *bundle;
6466 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6467 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6468 && !bundle->mirror_out) {
6469 output_normal(ctx, bundle, m->out_vlan);
6477 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6478 uint64_t packets, uint64_t bytes)
6484 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6487 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6490 /* In normal circumstances 'm' will not be NULL. However,
6491 * if mirrors are reconfigured, we can temporarily get out
6492 * of sync in facet_revalidate(). We could "correct" the
6493 * mirror list before reaching here, but doing that would
6494 * not properly account the traffic stats we've currently
6495 * accumulated for previous mirror configuration. */
6499 m->packet_count += packets;
6500 m->byte_count += bytes;
6504 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6505 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6506 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6508 is_gratuitous_arp(const struct flow *flow)
6510 return (flow->dl_type == htons(ETH_TYPE_ARP)
6511 && eth_addr_is_broadcast(flow->dl_dst)
6512 && (flow->nw_proto == ARP_OP_REPLY
6513 || (flow->nw_proto == ARP_OP_REQUEST
6514 && flow->nw_src == flow->nw_dst)));
6518 update_learning_table(struct ofproto_dpif *ofproto,
6519 const struct flow *flow, int vlan,
6520 struct ofbundle *in_bundle)
6522 struct mac_entry *mac;
6524 /* Don't learn the OFPP_NONE port. */
6525 if (in_bundle == &ofpp_none_bundle) {
6529 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6533 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6534 if (is_gratuitous_arp(flow)) {
6535 /* We don't want to learn from gratuitous ARP packets that are
6536 * reflected back over bond slaves so we lock the learning table. */
6537 if (!in_bundle->bond) {
6538 mac_entry_set_grat_arp_lock(mac);
6539 } else if (mac_entry_is_grat_arp_locked(mac)) {
6544 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
6545 /* The log messages here could actually be useful in debugging,
6546 * so keep the rate limit relatively high. */
6547 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
6548 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
6549 "on port %s in VLAN %d",
6550 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
6551 in_bundle->name, vlan);
6553 mac->port.p = in_bundle;
6554 tag_set_add(&ofproto->revalidate_set,
6555 mac_learning_changed(ofproto->ml, mac));
6559 static struct ofbundle *
6560 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
6561 bool warn, struct ofport_dpif **in_ofportp)
6563 struct ofport_dpif *ofport;
6565 /* Find the port and bundle for the received packet. */
6566 ofport = get_ofp_port(ofproto, in_port);
6568 *in_ofportp = ofport;
6570 if (ofport && ofport->bundle) {
6571 return ofport->bundle;
6574 /* Special-case OFPP_NONE, which a controller may use as the ingress
6575 * port for traffic that it is sourcing. */
6576 if (in_port == OFPP_NONE) {
6577 return &ofpp_none_bundle;
6580 /* Odd. A few possible reasons here:
6582 * - We deleted a port but there are still a few packets queued up
6585 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6586 * we don't know about.
6588 * - The ofproto client didn't configure the port as part of a bundle.
6589 * This is particularly likely to happen if a packet was received on the
6590 * port after it was created, but before the client had a chance to
6591 * configure its bundle.
6594 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6596 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
6597 "port %"PRIu16, ofproto->up.name, in_port);
6602 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6603 * dropped. Returns true if they may be forwarded, false if they should be
6606 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6607 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6609 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6610 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6611 * checked by input_vid_is_valid().
6613 * May also add tags to '*tags', although the current implementation only does
6614 * so in one special case.
6617 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
6620 struct ofproto_dpif *ofproto = ctx->ofproto;
6621 struct flow *flow = &ctx->flow;
6622 struct ofbundle *in_bundle = in_port->bundle;
6624 /* Drop frames for reserved multicast addresses
6625 * only if forward_bpdu option is absent. */
6626 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
6627 xlate_report(ctx, "packet has reserved destination MAC, dropping");
6631 if (in_bundle->bond) {
6632 struct mac_entry *mac;
6634 switch (bond_check_admissibility(in_bundle->bond, in_port,
6635 flow->dl_dst, &ctx->tags)) {
6640 xlate_report(ctx, "bonding refused admissibility, dropping");
6643 case BV_DROP_IF_MOVED:
6644 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
6645 if (mac && mac->port.p != in_bundle &&
6646 (!is_gratuitous_arp(flow)
6647 || mac_entry_is_grat_arp_locked(mac))) {
6648 xlate_report(ctx, "SLB bond thinks this packet looped back, "
6660 xlate_normal(struct action_xlate_ctx *ctx)
6662 struct ofport_dpif *in_port;
6663 struct ofbundle *in_bundle;
6664 struct mac_entry *mac;
6668 ctx->has_normal = true;
6670 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
6671 ctx->packet != NULL, &in_port);
6673 xlate_report(ctx, "no input bundle, dropping");
6677 /* Drop malformed frames. */
6678 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
6679 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
6680 if (ctx->packet != NULL) {
6681 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6682 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
6683 "VLAN tag received on port %s",
6684 ctx->ofproto->up.name, in_bundle->name);
6686 xlate_report(ctx, "partial VLAN tag, dropping");
6690 /* Drop frames on bundles reserved for mirroring. */
6691 if (in_bundle->mirror_out) {
6692 if (ctx->packet != NULL) {
6693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6694 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6695 "%s, which is reserved exclusively for mirroring",
6696 ctx->ofproto->up.name, in_bundle->name);
6698 xlate_report(ctx, "input port is mirror output port, dropping");
6703 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
6704 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6705 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
6708 vlan = input_vid_to_vlan(in_bundle, vid);
6710 /* Check other admissibility requirements. */
6711 if (in_port && !is_admissible(ctx, in_port, vlan)) {
6715 /* Learn source MAC. */
6716 if (ctx->may_learn) {
6717 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
6720 /* Determine output bundle. */
6721 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
6724 if (mac->port.p != in_bundle) {
6725 xlate_report(ctx, "forwarding to learned port");
6726 output_normal(ctx, mac->port.p, vlan);
6728 xlate_report(ctx, "learned port is input port, dropping");
6731 struct ofbundle *bundle;
6733 xlate_report(ctx, "no learned MAC for destination, flooding");
6734 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
6735 if (bundle != in_bundle
6736 && ofbundle_includes_vlan(bundle, vlan)
6737 && bundle->floodable
6738 && !bundle->mirror_out) {
6739 output_normal(ctx, bundle, vlan);
6742 ctx->nf_output_iface = NF_OUT_FLOOD;
6746 /* Optimized flow revalidation.
6748 * It's a difficult problem, in general, to tell which facets need to have
6749 * their actions recalculated whenever the OpenFlow flow table changes. We
6750 * don't try to solve that general problem: for most kinds of OpenFlow flow
6751 * table changes, we recalculate the actions for every facet. This is
6752 * relatively expensive, but it's good enough if the OpenFlow flow table
6753 * doesn't change very often.
6755 * However, we can expect one particular kind of OpenFlow flow table change to
6756 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6757 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6758 * table, we add a special case that applies to flow tables in which every rule
6759 * has the same form (that is, the same wildcards), except that the table is
6760 * also allowed to have a single "catch-all" flow that matches all packets. We
6761 * optimize this case by tagging all of the facets that resubmit into the table
6762 * and invalidating the same tag whenever a flow changes in that table. The
6763 * end result is that we revalidate just the facets that need it (and sometimes
6764 * a few more, but not all of the facets or even all of the facets that
6765 * resubmit to the table modified by MAC learning). */
6767 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6768 * into an OpenFlow table with the given 'basis'. */
6770 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
6773 if (minimask_is_catchall(mask)) {
6776 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
6777 return tag_create_deterministic(hash);
6781 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6782 * taggability of that table.
6784 * This function must be called after *each* change to a flow table. If you
6785 * skip calling it on some changes then the pointer comparisons at the end can
6786 * be invalid if you get unlucky. For example, if a flow removal causes a
6787 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6788 * different wildcards to be created with the same address, then this function
6789 * will incorrectly skip revalidation. */
6791 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
6793 struct table_dpif *table = &ofproto->tables[table_id];
6794 const struct oftable *oftable = &ofproto->up.tables[table_id];
6795 struct cls_table *catchall, *other;
6796 struct cls_table *t;
6798 catchall = other = NULL;
6800 switch (hmap_count(&oftable->cls.tables)) {
6802 /* We could tag this OpenFlow table but it would make the logic a
6803 * little harder and it's a corner case that doesn't seem worth it
6809 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
6810 if (cls_table_is_catchall(t)) {
6812 } else if (!other) {
6815 /* Indicate that we can't tag this by setting both tables to
6816 * NULL. (We know that 'catchall' is already NULL.) */
6823 /* Can't tag this table. */
6827 if (table->catchall_table != catchall || table->other_table != other) {
6828 table->catchall_table = catchall;
6829 table->other_table = other;
6830 ofproto->need_revalidate = REV_FLOW_TABLE;
6834 /* Given 'rule' that has changed in some way (either it is a rule being
6835 * inserted, a rule being deleted, or a rule whose actions are being
6836 * modified), marks facets for revalidation to ensure that packets will be
6837 * forwarded correctly according to the new state of the flow table.
6839 * This function must be called after *each* change to a flow table. See
6840 * the comment on table_update_taggable() for more information. */
6842 rule_invalidate(const struct rule_dpif *rule)
6844 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
6846 table_update_taggable(ofproto, rule->up.table_id);
6848 if (!ofproto->need_revalidate) {
6849 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
6851 if (table->other_table && rule->tag) {
6852 tag_set_add(&ofproto->revalidate_set, rule->tag);
6854 ofproto->need_revalidate = REV_FLOW_TABLE;
6860 set_frag_handling(struct ofproto *ofproto_,
6861 enum ofp_config_flags frag_handling)
6863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6865 if (frag_handling != OFPC_FRAG_REASM) {
6866 ofproto->need_revalidate = REV_RECONFIGURE;
6874 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
6875 const struct flow *flow,
6876 const struct ofpact *ofpacts, size_t ofpacts_len)
6878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6879 struct odputil_keybuf keybuf;
6880 struct dpif_flow_stats stats;
6884 struct action_xlate_ctx ctx;
6885 uint64_t odp_actions_stub[1024 / 8];
6886 struct ofpbuf odp_actions;
6888 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
6889 odp_flow_key_from_flow(&key, flow,
6890 ofp_port_to_odp_port(ofproto, flow->in_port));
6892 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
6894 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
6895 packet_get_tcp_flags(packet, flow), packet);
6896 ctx.resubmit_stats = &stats;
6898 ofpbuf_use_stub(&odp_actions,
6899 odp_actions_stub, sizeof odp_actions_stub);
6900 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
6901 dpif_execute(ofproto->backer->dpif, key.data, key.size,
6902 odp_actions.data, odp_actions.size, packet);
6903 ofpbuf_uninit(&odp_actions);
6911 set_netflow(struct ofproto *ofproto_,
6912 const struct netflow_options *netflow_options)
6914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6916 if (netflow_options) {
6917 if (!ofproto->netflow) {
6918 ofproto->netflow = netflow_create();
6920 return netflow_set_options(ofproto->netflow, netflow_options);
6922 netflow_destroy(ofproto->netflow);
6923 ofproto->netflow = NULL;
6929 get_netflow_ids(const struct ofproto *ofproto_,
6930 uint8_t *engine_type, uint8_t *engine_id)
6932 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
6934 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
6938 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
6940 if (!facet_is_controller_flow(facet) &&
6941 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
6942 struct subfacet *subfacet;
6943 struct ofexpired expired;
6945 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
6946 if (subfacet->path == SF_FAST_PATH) {
6947 struct dpif_flow_stats stats;
6949 subfacet_reinstall(subfacet, &stats);
6950 subfacet_update_stats(subfacet, &stats);
6954 expired.flow = facet->flow;
6955 expired.packet_count = facet->packet_count;
6956 expired.byte_count = facet->byte_count;
6957 expired.used = facet->used;
6958 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
6963 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
6965 struct facet *facet;
6967 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
6968 send_active_timeout(ofproto, facet);
6972 static struct ofproto_dpif *
6973 ofproto_dpif_lookup(const char *name)
6975 struct ofproto_dpif *ofproto;
6977 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
6978 hash_string(name, 0), &all_ofproto_dpifs) {
6979 if (!strcmp(ofproto->up.name, name)) {
6987 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
6988 const char *argv[], void *aux OVS_UNUSED)
6990 struct ofproto_dpif *ofproto;
6993 ofproto = ofproto_dpif_lookup(argv[1]);
6995 unixctl_command_reply_error(conn, "no such bridge");
6998 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7000 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7001 mac_learning_flush(ofproto->ml, &ofproto->revalidate_set);
7005 unixctl_command_reply(conn, "table successfully flushed");
7009 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7010 const char *argv[], void *aux OVS_UNUSED)
7012 struct ds ds = DS_EMPTY_INITIALIZER;
7013 const struct ofproto_dpif *ofproto;
7014 const struct mac_entry *e;
7016 ofproto = ofproto_dpif_lookup(argv[1]);
7018 unixctl_command_reply_error(conn, "no such bridge");
7022 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7023 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7024 struct ofbundle *bundle = e->port.p;
7025 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7026 ofbundle_get_a_port(bundle)->odp_port,
7027 e->vlan, ETH_ADDR_ARGS(e->mac),
7028 mac_entry_age(ofproto->ml, e));
7030 unixctl_command_reply(conn, ds_cstr(&ds));
7035 struct action_xlate_ctx ctx;
7041 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7042 const struct rule_dpif *rule)
7044 ds_put_char_multiple(result, '\t', level);
7046 ds_put_cstr(result, "No match\n");
7050 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7051 table_id, ntohll(rule->up.flow_cookie));
7052 cls_rule_format(&rule->up.cr, result);
7053 ds_put_char(result, '\n');
7055 ds_put_char_multiple(result, '\t', level);
7056 ds_put_cstr(result, "OpenFlow ");
7057 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7058 ds_put_char(result, '\n');
7062 trace_format_flow(struct ds *result, int level, const char *title,
7063 struct trace_ctx *trace)
7065 ds_put_char_multiple(result, '\t', level);
7066 ds_put_format(result, "%s: ", title);
7067 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7068 ds_put_cstr(result, "unchanged");
7070 flow_format(result, &trace->ctx.flow);
7071 trace->flow = trace->ctx.flow;
7073 ds_put_char(result, '\n');
7077 trace_format_regs(struct ds *result, int level, const char *title,
7078 struct trace_ctx *trace)
7082 ds_put_char_multiple(result, '\t', level);
7083 ds_put_format(result, "%s:", title);
7084 for (i = 0; i < FLOW_N_REGS; i++) {
7085 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7087 ds_put_char(result, '\n');
7091 trace_format_odp(struct ds *result, int level, const char *title,
7092 struct trace_ctx *trace)
7094 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7096 ds_put_char_multiple(result, '\t', level);
7097 ds_put_format(result, "%s: ", title);
7098 format_odp_actions(result, odp_actions->data, odp_actions->size);
7099 ds_put_char(result, '\n');
7103 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7105 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7106 struct ds *result = trace->result;
7108 ds_put_char(result, '\n');
7109 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7110 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7111 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7112 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7116 trace_report(struct action_xlate_ctx *ctx, const char *s)
7118 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7119 struct ds *result = trace->result;
7121 ds_put_char_multiple(result, '\t', ctx->recurse);
7122 ds_put_cstr(result, s);
7123 ds_put_char(result, '\n');
7127 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7128 void *aux OVS_UNUSED)
7130 const char *dpname = argv[1];
7131 struct ofproto_dpif *ofproto;
7132 struct ofpbuf odp_key;
7133 struct ofpbuf *packet;
7134 ovs_be16 initial_tci;
7140 ofpbuf_init(&odp_key, 0);
7143 ofproto = ofproto_dpif_lookup(dpname);
7145 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7149 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7150 /* ofproto/trace dpname flow [-generate] */
7151 const char *flow_s = argv[2];
7152 const char *generate_s = argv[3];
7154 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7155 * flow. We guess which type it is based on whether 'flow_s' contains
7156 * an '(', since a datapath flow always contains '(') but an
7157 * OpenFlow-like flow should not (in fact it's allowed but I believe
7158 * that's not documented anywhere).
7160 * An alternative would be to try to parse 'flow_s' both ways, but then
7161 * it would be tricky giving a sensible error message. After all, do
7162 * you just say "syntax error" or do you present both error messages?
7163 * Both choices seem lousy. */
7164 if (strchr(flow_s, '(')) {
7165 enum odp_key_fitness fitness;
7168 /* Convert string to datapath key. */
7169 ofpbuf_init(&odp_key, 0);
7170 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7172 unixctl_command_reply_error(conn, "Bad flow syntax");
7176 fitness = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
7177 flow.in_port = odp_port_to_ofp_port(ofproto, flow.in_port);
7179 /* Convert odp_key to flow. */
7180 error = ofproto_dpif_vsp_adjust(ofproto, fitness, &flow,
7181 &initial_tci, NULL);
7182 if (error == ODP_FIT_ERROR) {
7183 unixctl_command_reply_error(conn, "Invalid flow");
7189 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7191 unixctl_command_reply_error(conn, error_s);
7196 initial_tci = flow.vlan_tci;
7197 vsp_adjust_flow(ofproto, &flow);
7200 /* Generate a packet, if requested. */
7202 packet = ofpbuf_new(0);
7203 flow_compose(packet, &flow);
7205 } else if (argc == 6) {
7206 /* ofproto/trace dpname priority tun_id in_port packet */
7207 const char *priority_s = argv[2];
7208 const char *tun_id_s = argv[3];
7209 const char *in_port_s = argv[4];
7210 const char *packet_s = argv[5];
7211 uint32_t in_port = atoi(in_port_s);
7212 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7213 uint32_t priority = atoi(priority_s);
7216 msg = eth_from_hex(packet_s, &packet);
7218 unixctl_command_reply_error(conn, msg);
7222 ds_put_cstr(&result, "Packet: ");
7223 s = ofp_packet_to_string(packet->data, packet->size);
7224 ds_put_cstr(&result, s);
7227 flow_extract(packet, priority, NULL, in_port, &flow);
7228 flow.tunnel.tun_id = tun_id;
7229 initial_tci = flow.vlan_tci;
7231 unixctl_command_reply_error(conn, "Bad command syntax");
7235 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7236 unixctl_command_reply(conn, ds_cstr(&result));
7239 ds_destroy(&result);
7240 ofpbuf_delete(packet);
7241 ofpbuf_uninit(&odp_key);
7245 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7246 const struct ofpbuf *packet, ovs_be16 initial_tci,
7249 struct rule_dpif *rule;
7251 ds_put_cstr(ds, "Flow: ");
7252 flow_format(ds, flow);
7253 ds_put_char(ds, '\n');
7255 rule = rule_dpif_lookup(ofproto, flow);
7257 trace_format_rule(ds, 0, 0, rule);
7258 if (rule == ofproto->miss_rule) {
7259 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7260 } else if (rule == ofproto->no_packet_in_rule) {
7261 ds_put_cstr(ds, "\nNo match, packets dropped because "
7262 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7266 uint64_t odp_actions_stub[1024 / 8];
7267 struct ofpbuf odp_actions;
7269 struct trace_ctx trace;
7272 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7275 ofpbuf_use_stub(&odp_actions,
7276 odp_actions_stub, sizeof odp_actions_stub);
7277 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7278 rule, tcp_flags, packet);
7279 trace.ctx.resubmit_hook = trace_resubmit;
7280 trace.ctx.report_hook = trace_report;
7281 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7284 ds_put_char(ds, '\n');
7285 trace_format_flow(ds, 0, "Final flow", &trace);
7286 ds_put_cstr(ds, "Datapath actions: ");
7287 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7288 ofpbuf_uninit(&odp_actions);
7290 if (trace.ctx.slow) {
7291 enum slow_path_reason slow;
7293 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7294 "slow path because it:");
7295 for (slow = trace.ctx.slow; slow; ) {
7296 enum slow_path_reason bit = rightmost_1bit(slow);
7300 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7303 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7306 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7309 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7312 ds_put_cstr(ds, "\n\t (The datapath actions are "
7313 "incomplete--for complete actions, "
7314 "please supply a packet.)");
7317 case SLOW_CONTROLLER:
7318 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7319 "to the OpenFlow controller.");
7322 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7323 "than the datapath supports.");
7330 if (slow & ~SLOW_MATCH) {
7331 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7332 "the special slow-path processing.");
7339 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7340 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7343 unixctl_command_reply(conn, NULL);
7347 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7348 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7351 unixctl_command_reply(conn, NULL);
7354 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7355 * 'reply' describing the results. */
7357 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7359 struct facet *facet;
7363 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7364 if (!facet_check_consistency(facet)) {
7369 ofproto->need_revalidate = REV_INCONSISTENCY;
7373 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7374 ofproto->up.name, errors);
7376 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7381 ofproto_dpif_self_check(struct unixctl_conn *conn,
7382 int argc, const char *argv[], void *aux OVS_UNUSED)
7384 struct ds reply = DS_EMPTY_INITIALIZER;
7385 struct ofproto_dpif *ofproto;
7388 ofproto = ofproto_dpif_lookup(argv[1]);
7390 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7391 "ofproto/list for help)");
7394 ofproto_dpif_self_check__(ofproto, &reply);
7396 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7397 ofproto_dpif_self_check__(ofproto, &reply);
7401 unixctl_command_reply(conn, ds_cstr(&reply));
7405 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7406 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7407 * to destroy 'ofproto_shash' and free the returned value. */
7408 static const struct shash_node **
7409 get_ofprotos(struct shash *ofproto_shash)
7411 const struct ofproto_dpif *ofproto;
7413 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7414 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7415 shash_add_nocopy(ofproto_shash, name, ofproto);
7418 return shash_sort(ofproto_shash);
7422 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7423 const char *argv[] OVS_UNUSED,
7424 void *aux OVS_UNUSED)
7426 struct ds ds = DS_EMPTY_INITIALIZER;
7427 struct shash ofproto_shash;
7428 const struct shash_node **sorted_ofprotos;
7431 shash_init(&ofproto_shash);
7432 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7433 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7434 const struct shash_node *node = sorted_ofprotos[i];
7435 ds_put_format(&ds, "%s\n", node->name);
7438 shash_destroy(&ofproto_shash);
7439 free(sorted_ofprotos);
7441 unixctl_command_reply(conn, ds_cstr(&ds));
7446 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7448 struct dpif_dp_stats s;
7449 const struct shash_node **ports;
7452 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7454 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7455 dpif_name(ofproto->backer->dpif));
7456 /* xxx It would be better to show bridge-specific stats instead
7457 * xxx of dp ones. */
7459 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7460 s.n_hit, s.n_missed, s.n_lost);
7461 ds_put_format(ds, "\tflows: %zu\n",
7462 hmap_count(&ofproto->subfacets));
7464 ports = shash_sort(&ofproto->up.port_by_name);
7465 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7466 const struct shash_node *node = ports[i];
7467 struct ofport *ofport = node->data;
7468 const char *name = netdev_get_name(ofport->netdev);
7469 const char *type = netdev_get_type(ofport->netdev);
7471 ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port,
7472 ofp_port_to_odp_port(ofproto, ofport->ofp_port));
7473 if (strcmp(type, "system")) {
7474 struct netdev *netdev;
7477 ds_put_format(ds, " (%s", type);
7479 error = netdev_open(name, type, &netdev);
7484 error = netdev_get_config(netdev, &config);
7486 const struct smap_node **nodes;
7489 nodes = smap_sort(&config);
7490 for (i = 0; i < smap_count(&config); i++) {
7491 const struct smap_node *node = nodes[i];
7492 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7493 node->key, node->value);
7497 smap_destroy(&config);
7499 netdev_close(netdev);
7501 ds_put_char(ds, ')');
7503 ds_put_char(ds, '\n');
7509 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7510 const char *argv[], void *aux OVS_UNUSED)
7512 struct ds ds = DS_EMPTY_INITIALIZER;
7513 const struct ofproto_dpif *ofproto;
7517 for (i = 1; i < argc; i++) {
7518 ofproto = ofproto_dpif_lookup(argv[i]);
7520 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7521 "for help)", argv[i]);
7522 unixctl_command_reply_error(conn, ds_cstr(&ds));
7525 show_dp_format(ofproto, &ds);
7528 struct shash ofproto_shash;
7529 const struct shash_node **sorted_ofprotos;
7532 shash_init(&ofproto_shash);
7533 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7534 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7535 const struct shash_node *node = sorted_ofprotos[i];
7536 show_dp_format(node->data, &ds);
7539 shash_destroy(&ofproto_shash);
7540 free(sorted_ofprotos);
7543 unixctl_command_reply(conn, ds_cstr(&ds));
7548 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
7549 int argc OVS_UNUSED, const char *argv[],
7550 void *aux OVS_UNUSED)
7552 struct ds ds = DS_EMPTY_INITIALIZER;
7553 const struct ofproto_dpif *ofproto;
7554 struct subfacet *subfacet;
7556 ofproto = ofproto_dpif_lookup(argv[1]);
7558 unixctl_command_reply_error(conn, "no such bridge");
7562 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
7563 struct odputil_keybuf keybuf;
7566 subfacet_get_key(subfacet, &keybuf, &key);
7567 odp_flow_key_format(key.data, key.size, &ds);
7569 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
7570 subfacet->dp_packet_count, subfacet->dp_byte_count);
7571 if (subfacet->used) {
7572 ds_put_format(&ds, "%.3fs",
7573 (time_msec() - subfacet->used) / 1000.0);
7575 ds_put_format(&ds, "never");
7577 if (subfacet->facet->tcp_flags) {
7578 ds_put_cstr(&ds, ", flags:");
7579 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
7582 ds_put_cstr(&ds, ", actions:");
7583 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
7584 ds_put_char(&ds, '\n');
7587 unixctl_command_reply(conn, ds_cstr(&ds));
7592 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
7593 int argc OVS_UNUSED, const char *argv[],
7594 void *aux OVS_UNUSED)
7596 struct ds ds = DS_EMPTY_INITIALIZER;
7597 struct ofproto_dpif *ofproto;
7599 ofproto = ofproto_dpif_lookup(argv[1]);
7601 unixctl_command_reply_error(conn, "no such bridge");
7605 flush(&ofproto->up);
7607 unixctl_command_reply(conn, ds_cstr(&ds));
7612 ofproto_dpif_unixctl_init(void)
7614 static bool registered;
7620 unixctl_command_register(
7622 "bridge {tun_id in_port packet | odp_flow [-generate]}",
7623 2, 5, ofproto_unixctl_trace, NULL);
7624 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7625 ofproto_unixctl_fdb_flush, NULL);
7626 unixctl_command_register("fdb/show", "bridge", 1, 1,
7627 ofproto_unixctl_fdb_show, NULL);
7628 unixctl_command_register("ofproto/clog", "", 0, 0,
7629 ofproto_dpif_clog, NULL);
7630 unixctl_command_register("ofproto/unclog", "", 0, 0,
7631 ofproto_dpif_unclog, NULL);
7632 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7633 ofproto_dpif_self_check, NULL);
7634 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7635 ofproto_unixctl_dpif_dump_dps, NULL);
7636 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
7637 ofproto_unixctl_dpif_show, NULL);
7638 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7639 ofproto_unixctl_dpif_dump_flows, NULL);
7640 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7641 ofproto_unixctl_dpif_del_flows, NULL);
7644 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7646 * This is deprecated. It is only for compatibility with broken device drivers
7647 * in old versions of Linux that do not properly support VLANs when VLAN
7648 * devices are not used. When broken device drivers are no longer in
7649 * widespread use, we will delete these interfaces. */
7652 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
7654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
7655 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
7657 if (realdev_ofp_port == ofport->realdev_ofp_port
7658 && vid == ofport->vlandev_vid) {
7662 ofproto->need_revalidate = REV_RECONFIGURE;
7664 if (ofport->realdev_ofp_port) {
7667 if (realdev_ofp_port && ofport->bundle) {
7668 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7669 * themselves be part of a bundle. */
7670 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
7673 ofport->realdev_ofp_port = realdev_ofp_port;
7674 ofport->vlandev_vid = vid;
7676 if (realdev_ofp_port) {
7677 vsp_add(ofport, realdev_ofp_port, vid);
7684 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
7686 return hash_2words(realdev_ofp_port, vid);
7689 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7690 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7691 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7692 * it would return the port number of eth0.9.
7694 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7695 * function just returns its 'realdev_odp_port' argument. */
7697 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
7698 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
7700 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
7701 uint16_t realdev_ofp_port;
7702 int vid = vlan_tci_to_vid(vlan_tci);
7703 const struct vlan_splinter *vsp;
7705 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
7706 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
7707 hash_realdev_vid(realdev_ofp_port, vid),
7708 &ofproto->realdev_vid_map) {
7709 if (vsp->realdev_ofp_port == realdev_ofp_port
7710 && vsp->vid == vid) {
7711 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
7715 return realdev_odp_port;
7718 static struct vlan_splinter *
7719 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
7721 struct vlan_splinter *vsp;
7723 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
7724 &ofproto->vlandev_map) {
7725 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
7733 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7734 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7735 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7736 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7737 * eth0 and store 9 in '*vid'.
7739 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7740 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7743 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
7744 uint16_t vlandev_ofp_port, int *vid)
7746 if (!hmap_is_empty(&ofproto->vlandev_map)) {
7747 const struct vlan_splinter *vsp;
7749 vsp = vlandev_find(ofproto, vlandev_ofp_port);
7754 return vsp->realdev_ofp_port;
7760 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7761 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7762 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7763 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7764 * always the case unless VLAN splinters are enabled), returns false without
7765 * making any changes. */
7767 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
7772 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
7777 /* Cause the flow to be processed as if it came in on the real device with
7778 * the VLAN device's VLAN ID. */
7779 flow->in_port = realdev;
7780 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
7785 vsp_remove(struct ofport_dpif *port)
7787 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7788 struct vlan_splinter *vsp;
7790 vsp = vlandev_find(ofproto, port->up.ofp_port);
7792 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
7793 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
7796 port->realdev_ofp_port = 0;
7798 VLOG_ERR("missing vlan device record");
7803 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
7805 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
7807 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
7808 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
7809 == realdev_ofp_port)) {
7810 struct vlan_splinter *vsp;
7812 vsp = xmalloc(sizeof *vsp);
7813 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
7814 hash_int(port->up.ofp_port, 0));
7815 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
7816 hash_realdev_vid(realdev_ofp_port, vid));
7817 vsp->realdev_ofp_port = realdev_ofp_port;
7818 vsp->vlandev_ofp_port = port->up.ofp_port;
7821 port->realdev_ofp_port = realdev_ofp_port;
7823 VLOG_ERR("duplicate vlan device record");
7828 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
7830 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
7831 return ofport ? ofport->odp_port : OVSP_NONE;
7834 static struct ofport_dpif *
7835 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
7837 struct ofport_dpif *port;
7839 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
7840 hash_int(odp_port, 0),
7841 &backer->odp_to_ofport_map) {
7842 if (port->odp_port == odp_port) {
7851 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
7853 struct ofport_dpif *port;
7855 port = odp_port_to_ofport(ofproto->backer, odp_port);
7856 if (port && ofproto == ofproto_dpif_cast(port->up.ofproto)) {
7857 return port->up.ofp_port;
7863 const struct ofproto_class ofproto_dpif_class = {
7897 port_is_lacp_current,
7898 NULL, /* rule_choose_table */
7905 rule_modify_actions,
7914 get_cfm_remote_mpids,
7919 get_stp_port_status,
7926 is_mirror_output_bundle,
7927 forward_bpdu_changed,