2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
54 #include "unaligned.h"
56 #include "vlan-bitmap.h"
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif
);
61 COVERAGE_DEFINE(ofproto_dpif_expired
);
62 COVERAGE_DEFINE(ofproto_dpif_xlate
);
63 COVERAGE_DEFINE(facet_changed_rule
);
64 COVERAGE_DEFINE(facet_revalidate
);
65 COVERAGE_DEFINE(facet_unexpected
);
66 COVERAGE_DEFINE(facet_suppress
);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
72 /* Number of implemented OpenFlow tables. */
73 enum { N_TABLES
= 255 };
74 enum { TBL_INTERNAL
= N_TABLES
- 1 }; /* Used for internal hidden rules. */
75 BUILD_ASSERT_DECL(N_TABLES
>= 2 && N_TABLES
<= 255);
86 * - Do include packets and bytes from facets that have been deleted or
87 * whose own statistics have been folded into the rule.
89 * - Do include packets and bytes sent "by hand" that were accounted to
90 * the rule without any facet being involved (this is a rare corner
91 * case in rule_execute()).
93 * - Do not include packet or bytes that can be obtained from any facet's
94 * packet_count or byte_count member or that can be obtained from the
95 * datapath by, e.g., dpif_flow_get() for any subfacet.
97 uint64_t packet_count
; /* Number of packets received. */
98 uint64_t byte_count
; /* Number of bytes received. */
100 tag_type tag
; /* Caches rule_calculate_tag() result. */
102 struct list facets
; /* List of "struct facet"s. */
105 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
107 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
110 static struct rule_dpif
*rule_dpif_lookup(struct ofproto_dpif
*,
111 const struct flow
*);
112 static struct rule_dpif
*rule_dpif_lookup__(struct ofproto_dpif
*,
115 static struct rule_dpif
*rule_dpif_miss_rule(struct ofproto_dpif
*ofproto
,
116 const struct flow
*flow
);
118 static void rule_credit_stats(struct rule_dpif
*,
119 const struct dpif_flow_stats
*);
120 static void flow_push_stats(struct rule_dpif
*, const struct flow
*,
121 const struct dpif_flow_stats
*);
122 static tag_type
rule_calculate_tag(const struct flow
*,
123 const struct minimask
*, uint32_t basis
);
124 static void rule_invalidate(const struct rule_dpif
*);
126 #define MAX_MIRRORS 32
127 typedef uint32_t mirror_mask_t
;
128 #define MIRROR_MASK_C(X) UINT32_C(X)
129 BUILD_ASSERT_DECL(sizeof(mirror_mask_t
) * CHAR_BIT
>= MAX_MIRRORS
);
131 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
132 size_t idx
; /* In ofproto's "mirrors" array. */
133 void *aux
; /* Key supplied by ofproto's client. */
134 char *name
; /* Identifier for log messages. */
136 /* Selection criteria. */
137 struct hmapx srcs
; /* Contains "struct ofbundle *"s. */
138 struct hmapx dsts
; /* Contains "struct ofbundle *"s. */
139 unsigned long *vlans
; /* Bitmap of chosen VLANs, NULL selects all. */
141 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
142 struct ofbundle
*out
; /* Output port or NULL. */
143 int out_vlan
; /* Output VLAN or -1. */
144 mirror_mask_t dup_mirrors
; /* Bitmap of mirrors with the same output. */
147 int64_t packet_count
; /* Number of packets sent. */
148 int64_t byte_count
; /* Number of bytes sent. */
151 static void mirror_destroy(struct ofmirror
*);
152 static void update_mirror_stats(struct ofproto_dpif
*ofproto
,
153 mirror_mask_t mirrors
,
154 uint64_t packets
, uint64_t bytes
);
157 struct hmap_node hmap_node
; /* In struct ofproto's "bundles" hmap. */
158 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
159 void *aux
; /* Key supplied by ofproto's client. */
160 char *name
; /* Identifier for log messages. */
163 struct list ports
; /* Contains "struct ofport"s. */
164 enum port_vlan_mode vlan_mode
; /* VLAN mode */
165 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
166 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
167 * NULL if all VLANs are trunked. */
168 struct lacp
*lacp
; /* LACP if LACP is enabled, otherwise NULL. */
169 struct bond
*bond
; /* Nonnull iff more than one port. */
170 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
173 bool floodable
; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
175 /* Port mirroring info. */
176 mirror_mask_t src_mirrors
; /* Mirrors triggered when packet received. */
177 mirror_mask_t dst_mirrors
; /* Mirrors triggered when packet sent. */
178 mirror_mask_t mirror_out
; /* Mirrors that output to this bundle. */
181 static void bundle_remove(struct ofport
*);
182 static void bundle_update(struct ofbundle
*);
183 static void bundle_destroy(struct ofbundle
*);
184 static void bundle_del_port(struct ofport_dpif
*);
185 static void bundle_run(struct ofbundle
*);
186 static void bundle_wait(struct ofbundle
*);
187 static struct ofbundle
*lookup_input_bundle(const struct ofproto_dpif
*,
188 uint16_t in_port
, bool warn
,
189 struct ofport_dpif
**in_ofportp
);
191 /* A controller may use OFPP_NONE as the ingress port to indicate that
192 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
193 * when an input bundle is needed for validation (e.g., mirroring or
194 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
195 * any 'port' structs, so care must be taken when dealing with it. */
196 static struct ofbundle ofpp_none_bundle
= {
198 .vlan_mode
= PORT_VLAN_TRUNK
201 static void stp_run(struct ofproto_dpif
*ofproto
);
202 static void stp_wait(struct ofproto_dpif
*ofproto
);
203 static int set_stp_port(struct ofport
*,
204 const struct ofproto_port_stp_settings
*);
206 static bool ofbundle_includes_vlan(const struct ofbundle
*, uint16_t vlan
);
208 struct action_xlate_ctx
{
209 /* action_xlate_ctx_init() initializes these members. */
212 struct ofproto_dpif
*ofproto
;
214 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
215 * this flow when actions change header fields. */
218 /* The packet corresponding to 'flow', or a null pointer if we are
219 * revalidating without a packet to refer to. */
220 const struct ofpbuf
*packet
;
222 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
223 * actions update the flow table?
225 * We want to update these tables if we are actually processing a packet,
226 * or if we are accounting for packets that the datapath has processed, but
227 * not if we are just revalidating. */
230 /* The rule that we are currently translating, or NULL. */
231 struct rule_dpif
*rule
;
233 /* Union of the set of TCP flags seen so far in this flow. (Used only by
234 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
238 /* If nonnull, flow translation calls this function just before executing a
239 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
240 * when the recursion depth is exceeded.
242 * 'rule' is the rule being submitted into. It will be null if the
243 * resubmit or OFPP_TABLE action didn't find a matching rule.
245 * This is normally null so the client has to set it manually after
246 * calling action_xlate_ctx_init(). */
247 void (*resubmit_hook
)(struct action_xlate_ctx
*, struct rule_dpif
*rule
);
249 /* If nonnull, flow translation calls this function to report some
250 * significant decision, e.g. to explain why OFPP_NORMAL translation
251 * dropped a packet. */
252 void (*report_hook
)(struct action_xlate_ctx
*, const char *s
);
254 /* If nonnull, flow translation credits the specified statistics to each
255 * rule reached through a resubmit or OFPP_TABLE action.
257 * This is normally null so the client has to set it manually after
258 * calling action_xlate_ctx_init(). */
259 const struct dpif_flow_stats
*resubmit_stats
;
261 /* xlate_actions() initializes and uses these members. The client might want
262 * to look at them after it returns. */
264 struct ofpbuf
*odp_actions
; /* Datapath actions. */
265 tag_type tags
; /* Tags associated with actions. */
266 enum slow_path_reason slow
; /* 0 if fast path may be used. */
267 bool has_learn
; /* Actions include NXAST_LEARN? */
268 bool has_normal
; /* Actions output to OFPP_NORMAL? */
269 bool has_fin_timeout
; /* Actions include NXAST_FIN_TIMEOUT? */
270 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
271 mirror_mask_t mirrors
; /* Bitmap of associated mirrors. */
273 /* xlate_actions() initializes and uses these members, but the client has no
274 * reason to look at them. */
276 int recurse
; /* Recursion level, via xlate_table_action. */
277 bool max_resubmit_trigger
; /* Recursed too deeply during translation. */
278 struct flow base_flow
; /* Flow at the last commit. */
279 uint32_t orig_skb_priority
; /* Priority when packet arrived. */
280 uint8_t table_id
; /* OpenFlow table ID where flow was found. */
281 uint32_t sflow_n_outputs
; /* Number of output ports. */
282 uint32_t sflow_odp_port
; /* Output port for composing sFlow action. */
283 uint16_t user_cookie_offset
;/* Used for user_action_cookie fixup. */
284 bool exit
; /* No further actions should be processed. */
285 struct flow orig_flow
; /* Copy of original flow. */
288 static void action_xlate_ctx_init(struct action_xlate_ctx
*,
289 struct ofproto_dpif
*, const struct flow
*,
290 ovs_be16 initial_tci
, struct rule_dpif
*,
291 uint8_t tcp_flags
, const struct ofpbuf
*);
292 static void xlate_actions(struct action_xlate_ctx
*,
293 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
294 struct ofpbuf
*odp_actions
);
295 static void xlate_actions_for_side_effects(struct action_xlate_ctx
*,
296 const struct ofpact
*ofpacts
,
299 static size_t put_userspace_action(const struct ofproto_dpif
*,
300 struct ofpbuf
*odp_actions
,
302 const union user_action_cookie
*);
304 static void compose_slow_path(const struct ofproto_dpif
*, const struct flow
*,
305 enum slow_path_reason
,
306 uint64_t *stub
, size_t stub_size
,
307 const struct nlattr
**actionsp
,
308 size_t *actions_lenp
);
310 static void xlate_report(struct action_xlate_ctx
*ctx
, const char *s
);
312 /* A subfacet (see "struct subfacet" below) has three possible installation
315 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
316 * case just after the subfacet is created, just before the subfacet is
317 * destroyed, or if the datapath returns an error when we try to install a
320 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
322 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
323 * ofproto_dpif is installed in the datapath.
326 SF_NOT_INSTALLED
, /* No datapath flow for this subfacet. */
327 SF_FAST_PATH
, /* Full actions are installed. */
328 SF_SLOW_PATH
, /* Send-to-userspace action is installed. */
331 static const char *subfacet_path_to_string(enum subfacet_path
);
333 /* A dpif flow and actions associated with a facet.
335 * See also the large comment on struct facet. */
338 struct hmap_node hmap_node
; /* In struct ofproto_dpif 'subfacets' list. */
339 struct list list_node
; /* In struct facet's 'facets' list. */
340 struct facet
*facet
; /* Owning facet. */
344 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
345 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
346 * regenerate the ODP flow key from ->facet->flow. */
347 enum odp_key_fitness key_fitness
;
351 long long int used
; /* Time last used; time created if not used. */
353 uint64_t dp_packet_count
; /* Last known packet count in the datapath. */
354 uint64_t dp_byte_count
; /* Last known byte count in the datapath. */
358 * These should be essentially identical for every subfacet in a facet, but
359 * may differ in trivial ways due to VLAN splinters. */
360 size_t actions_len
; /* Number of bytes in actions[]. */
361 struct nlattr
*actions
; /* Datapath actions. */
363 enum slow_path_reason slow
; /* 0 if fast path may be used. */
364 enum subfacet_path path
; /* Installed in datapath? */
366 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
367 * splinters can cause it to differ. This value should be removed when
368 * the VLAN splinters feature is no longer needed. */
369 ovs_be16 initial_tci
; /* Initial VLAN TCI value. */
371 /* Datapath port the packet arrived on. This is needed to remove
372 * flows for ports that are no longer part of the bridge. Since the
373 * flow definition only has the OpenFlow port number and the port is
374 * no longer part of the bridge, we can't determine the datapath port
375 * number needed to delete the flow from the datapath. */
376 uint32_t odp_in_port
;
379 #define SUBFACET_DESTROY_MAX_BATCH 50
381 static struct subfacet
*subfacet_create(struct facet
*, struct flow_miss
*miss
,
383 static struct subfacet
*subfacet_find(struct ofproto_dpif
*,
384 const struct nlattr
*key
, size_t key_len
,
386 const struct flow
*flow
);
387 static void subfacet_destroy(struct subfacet
*);
388 static void subfacet_destroy__(struct subfacet
*);
389 static void subfacet_destroy_batch(struct ofproto_dpif
*,
390 struct subfacet
**, int n
);
391 static void subfacet_get_key(struct subfacet
*, struct odputil_keybuf
*,
393 static void subfacet_reset_dp_stats(struct subfacet
*,
394 struct dpif_flow_stats
*);
395 static void subfacet_update_time(struct subfacet
*, long long int used
);
396 static void subfacet_update_stats(struct subfacet
*,
397 const struct dpif_flow_stats
*);
398 static void subfacet_make_actions(struct subfacet
*,
399 const struct ofpbuf
*packet
,
400 struct ofpbuf
*odp_actions
);
401 static int subfacet_install(struct subfacet
*,
402 const struct nlattr
*actions
, size_t actions_len
,
403 struct dpif_flow_stats
*, enum slow_path_reason
);
404 static void subfacet_uninstall(struct subfacet
*);
406 static enum subfacet_path
subfacet_want_path(enum slow_path_reason
);
408 /* An exact-match instantiation of an OpenFlow flow.
410 * A facet associates a "struct flow", which represents the Open vSwitch
411 * userspace idea of an exact-match flow, with one or more subfacets. Each
412 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
413 * the facet. When the kernel module (or other dpif implementation) and Open
414 * vSwitch userspace agree on the definition of a flow key, there is exactly
415 * one subfacet per facet. If the dpif implementation supports more-specific
416 * flow matching than userspace, however, a facet can have more than one
417 * subfacet, each of which corresponds to some distinction in flow that
418 * userspace simply doesn't understand.
420 * Flow expiration works in terms of subfacets, so a facet must have at least
421 * one subfacet or it will never expire, leaking memory. */
424 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
425 struct list list_node
; /* In owning rule's 'facets' list. */
426 struct rule_dpif
*rule
; /* Owning rule. */
429 struct list subfacets
;
430 long long int used
; /* Time last used; time created if not used. */
437 * - Do include packets and bytes sent "by hand", e.g. with
440 * - Do include packets and bytes that were obtained from the datapath
441 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
442 * DPIF_FP_ZERO_STATS).
444 * - Do not include packets or bytes that can be obtained from the
445 * datapath for any existing subfacet.
447 uint64_t packet_count
; /* Number of packets received. */
448 uint64_t byte_count
; /* Number of bytes received. */
450 /* Resubmit statistics. */
451 uint64_t prev_packet_count
; /* Number of packets from last stats push. */
452 uint64_t prev_byte_count
; /* Number of bytes from last stats push. */
453 long long int prev_used
; /* Used time from last stats push. */
456 uint64_t accounted_bytes
; /* Bytes processed by facet_account(). */
457 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
458 uint8_t tcp_flags
; /* TCP flags seen for this 'rule'. */
460 /* Properties of datapath actions.
462 * Every subfacet has its own actions because actions can differ slightly
463 * between splintered and non-splintered subfacets due to the VLAN tag
464 * being initially different (present vs. absent). All of them have these
465 * properties in common so we just store one copy of them here. */
466 bool has_learn
; /* Actions include NXAST_LEARN? */
467 bool has_normal
; /* Actions output to OFPP_NORMAL? */
468 bool has_fin_timeout
; /* Actions include NXAST_FIN_TIMEOUT? */
469 tag_type tags
; /* Tags that would require revalidation. */
470 mirror_mask_t mirrors
; /* Bitmap of dependent mirrors. */
472 /* Storage for a single subfacet, to reduce malloc() time and space
473 * overhead. (A facet always has at least one subfacet and in the common
474 * case has exactly one subfacet.) */
475 struct subfacet one_subfacet
;
478 static struct facet
*facet_create(struct rule_dpif
*,
479 const struct flow
*, uint32_t hash
);
480 static void facet_remove(struct facet
*);
481 static void facet_free(struct facet
*);
483 static struct facet
*facet_find(struct ofproto_dpif
*,
484 const struct flow
*, uint32_t hash
);
485 static struct facet
*facet_lookup_valid(struct ofproto_dpif
*,
486 const struct flow
*, uint32_t hash
);
487 static void facet_revalidate(struct facet
*);
488 static bool facet_check_consistency(struct facet
*);
490 static void facet_flush_stats(struct facet
*);
492 static void facet_update_time(struct facet
*, long long int used
);
493 static void facet_reset_counters(struct facet
*);
494 static void facet_push_stats(struct facet
*);
495 static void facet_learn(struct facet
*);
496 static void facet_account(struct facet
*);
498 static bool facet_is_controller_flow(struct facet
*);
501 struct hmap_node odp_port_node
; /* In dpif_backer's "odp_to_ofport_map". */
505 struct ofbundle
*bundle
; /* Bundle that contains this port, if any. */
506 struct list bundle_node
; /* In struct ofbundle's "ports" list. */
507 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
508 tag_type tag
; /* Tag associated with this port. */
509 uint32_t bond_stable_id
; /* stable_id to use as bond slave, or 0. */
510 bool may_enable
; /* May be enabled in bonds. */
511 long long int carrier_seq
; /* Carrier status changes. */
514 struct stp_port
*stp_port
; /* Spanning Tree Protocol, if any. */
515 enum stp_state stp_state
; /* Always STP_DISABLED if STP not in use. */
516 long long int stp_state_entered
;
518 struct hmap priorities
; /* Map of attached 'priority_to_dscp's. */
520 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
522 * This is deprecated. It is only for compatibility with broken device
523 * drivers in old versions of Linux that do not properly support VLANs when
524 * VLAN devices are not used. When broken device drivers are no longer in
525 * widespread use, we will delete these interfaces. */
526 uint16_t realdev_ofp_port
;
530 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
531 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
532 * traffic egressing the 'ofport' with that priority should be marked with. */
533 struct priority_to_dscp
{
534 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'priorities' map. */
535 uint32_t priority
; /* Priority of this queue (see struct flow). */
537 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device drivers
543 * in old versions of Linux that do not properly support VLANs when VLAN
544 * devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 struct vlan_splinter
{
547 struct hmap_node realdev_vid_node
;
548 struct hmap_node vlandev_node
;
549 uint16_t realdev_ofp_port
;
550 uint16_t vlandev_ofp_port
;
554 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif
*,
555 uint32_t realdev
, ovs_be16 vlan_tci
);
556 static bool vsp_adjust_flow(const struct ofproto_dpif
*, struct flow
*);
557 static void vsp_remove(struct ofport_dpif
*);
558 static void vsp_add(struct ofport_dpif
*, uint16_t realdev_ofp_port
, int vid
);
560 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif
*,
562 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif
*,
565 static struct ofport_dpif
*
566 ofport_dpif_cast(const struct ofport
*ofport
)
568 ovs_assert(ofport
->ofproto
->ofproto_class
== &ofproto_dpif_class
);
569 return ofport
? CONTAINER_OF(ofport
, struct ofport_dpif
, up
) : NULL
;
572 static void port_run(struct ofport_dpif
*);
573 static void port_run_fast(struct ofport_dpif
*);
574 static void port_wait(struct ofport_dpif
*);
575 static int set_cfm(struct ofport
*, const struct cfm_settings
*);
576 static void ofport_clear_priorities(struct ofport_dpif
*);
578 struct dpif_completion
{
579 struct list list_node
;
580 struct ofoperation
*op
;
583 /* Extra information about a classifier table.
584 * Currently used just for optimized flow revalidation. */
586 /* If either of these is nonnull, then this table has a form that allows
587 * flows to be tagged to avoid revalidating most flows for the most common
588 * kinds of flow table changes. */
589 struct cls_table
*catchall_table
; /* Table that wildcards all fields. */
590 struct cls_table
*other_table
; /* Table with any other wildcard set. */
591 uint32_t basis
; /* Keeps each table's tags separate. */
594 /* Reasons that we might need to revalidate every facet, and corresponding
597 * A value of 0 means that there is no need to revalidate.
599 * It would be nice to have some cleaner way to integrate with coverage
600 * counters, but with only a few reasons I guess this is good enough for
602 enum revalidate_reason
{
603 REV_RECONFIGURE
= 1, /* Switch configuration changed. */
604 REV_STP
, /* Spanning tree protocol port status change. */
605 REV_PORT_TOGGLED
, /* Port enabled or disabled by CFM, LACP, ...*/
606 REV_FLOW_TABLE
, /* Flow table changed. */
607 REV_INCONSISTENCY
/* Facet self-check failed. */
609 COVERAGE_DEFINE(rev_reconfigure
);
610 COVERAGE_DEFINE(rev_stp
);
611 COVERAGE_DEFINE(rev_port_toggled
);
612 COVERAGE_DEFINE(rev_flow_table
);
613 COVERAGE_DEFINE(rev_inconsistency
);
615 /* All datapaths of a given type share a single dpif backer instance. */
620 struct timer next_expiration
;
621 struct hmap odp_to_ofport_map
; /* ODP port to ofport mapping. */
624 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
625 static struct shash all_dpif_backers
= SHASH_INITIALIZER(&all_dpif_backers
);
627 static struct ofport_dpif
*
628 odp_port_to_ofport(const struct dpif_backer
*, uint32_t odp_port
);
630 struct ofproto_dpif
{
631 struct hmap_node all_ofproto_dpifs_node
; /* In 'all_ofproto_dpifs'. */
633 struct dpif_backer
*backer
;
635 /* Special OpenFlow rules. */
636 struct rule_dpif
*miss_rule
; /* Sends flow table misses to controller. */
637 struct rule_dpif
*no_packet_in_rule
; /* Drops flow table misses. */
643 struct netflow
*netflow
;
644 struct dpif_sflow
*sflow
;
645 struct hmap bundles
; /* Contains "struct ofbundle"s. */
646 struct mac_learning
*ml
;
647 struct ofmirror
*mirrors
[MAX_MIRRORS
];
649 bool has_bonded_bundles
;
653 struct hmap subfacets
;
654 struct governor
*governor
;
657 struct table_dpif tables
[N_TABLES
];
658 enum revalidate_reason need_revalidate
;
659 struct tag_set revalidate_set
;
661 /* Support for debugging async flow mods. */
662 struct list completions
;
664 bool has_bundle_action
; /* True when the first bundle action appears. */
665 struct netdev_stats stats
; /* To account packets generated and consumed in
670 long long int stp_last_tick
;
672 /* VLAN splinters. */
673 struct hmap realdev_vid_map
; /* (realdev,vid) -> vlandev. */
674 struct hmap vlandev_map
; /* vlandev -> (realdev,vid). */
677 struct sset ports
; /* Set of port names. */
678 struct sset port_poll_set
; /* Queued names for port_poll() reply. */
679 int port_poll_errno
; /* Last errno for port_poll() reply. */
682 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
683 * for debugging the asynchronous flow_mod implementation.) */
686 /* All existing ofproto_dpif instances, indexed by ->up.name. */
687 static struct hmap all_ofproto_dpifs
= HMAP_INITIALIZER(&all_ofproto_dpifs
);
689 static void ofproto_dpif_unixctl_init(void);
691 static struct ofproto_dpif
*
692 ofproto_dpif_cast(const struct ofproto
*ofproto
)
694 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
695 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
698 static struct ofport_dpif
*get_ofp_port(const struct ofproto_dpif
*,
700 static struct ofport_dpif
*get_odp_port(const struct ofproto_dpif
*,
702 static void ofproto_trace(struct ofproto_dpif
*, const struct flow
*,
703 const struct ofpbuf
*, ovs_be16 initial_tci
,
706 /* Packet processing. */
707 static void update_learning_table(struct ofproto_dpif
*,
708 const struct flow
*, int vlan
,
711 #define FLOW_MISS_MAX_BATCH 50
712 static int handle_upcalls(struct dpif_backer
*, unsigned int max_batch
);
714 /* Flow expiration. */
715 static int expire(struct dpif_backer
*);
718 static void send_netflow_active_timeouts(struct ofproto_dpif
*);
721 static int send_packet(const struct ofport_dpif
*, struct ofpbuf
*packet
);
722 static size_t compose_sflow_action(const struct ofproto_dpif
*,
723 struct ofpbuf
*odp_actions
,
724 const struct flow
*, uint32_t odp_port
);
725 static void add_mirror_actions(struct action_xlate_ctx
*ctx
,
726 const struct flow
*flow
);
727 /* Global variables. */
728 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
730 /* Initial mappings of port to bridge mappings. */
731 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
733 /* Factory functions. */
736 init(const struct shash
*iface_hints
)
738 struct shash_node
*node
;
740 /* Make a local copy, since we don't own 'iface_hints' elements. */
741 SHASH_FOR_EACH(node
, iface_hints
) {
742 const struct iface_hint
*orig_hint
= node
->data
;
743 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
745 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
746 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
747 new_hint
->ofp_port
= orig_hint
->ofp_port
;
749 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
754 enumerate_types(struct sset
*types
)
756 dp_enumerate_types(types
);
760 enumerate_names(const char *type
, struct sset
*names
)
762 struct ofproto_dpif
*ofproto
;
765 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
766 if (strcmp(type
, ofproto
->up
.type
)) {
769 sset_add(names
, ofproto
->up
.name
);
776 del(const char *type
, const char *name
)
781 error
= dpif_open(name
, type
, &dpif
);
783 error
= dpif_delete(dpif
);
790 port_open_type(const char *datapath_type
, const char *port_type
)
792 return dpif_port_open_type(datapath_type
, port_type
);
795 /* Type functions. */
797 static struct ofproto_dpif
*
798 lookup_ofproto_dpif_by_port_name(const char *name
)
800 struct ofproto_dpif
*ofproto
;
802 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
803 if (sset_contains(&ofproto
->ports
, name
)) {
812 type_run(const char *type
)
814 struct dpif_backer
*backer
;
818 backer
= shash_find_data(&all_dpif_backers
, type
);
820 /* This is not necessarily a problem, since backers are only
821 * created on demand. */
825 dpif_run(backer
->dpif
);
827 if (timer_expired(&backer
->next_expiration
)) {
828 int delay
= expire(backer
);
829 timer_set_duration(&backer
->next_expiration
, delay
);
832 /* Check for port changes in the dpif. */
833 while ((error
= dpif_port_poll(backer
->dpif
, &devname
)) == 0) {
834 struct ofproto_dpif
*ofproto
;
835 struct dpif_port port
;
837 /* Don't report on the datapath's device. */
838 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
842 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
843 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
844 /* The port was removed. If we know the datapath,
845 * report it through poll_set(). If we don't, it may be
846 * notifying us of a removal we initiated, so ignore it.
847 * If there's a pending ENOBUFS, let it stand, since
848 * everything will be reevaluated. */
849 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
850 sset_add(&ofproto
->port_poll_set
, devname
);
851 ofproto
->port_poll_errno
= 0;
853 } else if (!ofproto
) {
854 /* The port was added, but we don't know with which
855 * ofproto we should associate it. Delete it. */
856 dpif_port_del(backer
->dpif
, port
.port_no
);
858 dpif_port_destroy(&port
);
864 if (error
!= EAGAIN
) {
865 struct ofproto_dpif
*ofproto
;
867 /* There was some sort of error, so propagate it to all
868 * ofprotos that use this backer. */
869 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
870 &all_ofproto_dpifs
) {
871 if (ofproto
->backer
== backer
) {
872 sset_clear(&ofproto
->port_poll_set
);
873 ofproto
->port_poll_errno
= error
;
882 type_run_fast(const char *type
)
884 struct dpif_backer
*backer
;
887 backer
= shash_find_data(&all_dpif_backers
, type
);
889 /* This is not necessarily a problem, since backers are only
890 * created on demand. */
894 /* Handle one or more batches of upcalls, until there's nothing left to do
895 * or until we do a fixed total amount of work.
897 * We do work in batches because it can be much cheaper to set up a number
898 * of flows and fire off their patches all at once. We do multiple batches
899 * because in some cases handling a packet can cause another packet to be
900 * queued almost immediately as part of the return flow. Both
901 * optimizations can make major improvements on some benchmarks and
902 * presumably for real traffic as well. */
904 while (work
< FLOW_MISS_MAX_BATCH
) {
905 int retval
= handle_upcalls(backer
, FLOW_MISS_MAX_BATCH
- work
);
916 type_wait(const char *type
)
918 struct dpif_backer
*backer
;
920 backer
= shash_find_data(&all_dpif_backers
, type
);
922 /* This is not necessarily a problem, since backers are only
923 * created on demand. */
927 timer_wait(&backer
->next_expiration
);
930 /* Basic life-cycle. */
932 static int add_internal_flows(struct ofproto_dpif
*);
934 static struct ofproto
*
937 struct ofproto_dpif
*ofproto
= xmalloc(sizeof *ofproto
);
942 dealloc(struct ofproto
*ofproto_
)
944 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
949 close_dpif_backer(struct dpif_backer
*backer
)
951 struct shash_node
*node
;
953 ovs_assert(backer
->refcount
> 0);
955 if (--backer
->refcount
) {
959 hmap_destroy(&backer
->odp_to_ofport_map
);
960 node
= shash_find(&all_dpif_backers
, backer
->type
);
962 shash_delete(&all_dpif_backers
, node
);
963 dpif_close(backer
->dpif
);
968 /* Datapath port slated for removal from datapath. */
970 struct list list_node
;
975 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
977 struct dpif_backer
*backer
;
978 struct dpif_port_dump port_dump
;
979 struct dpif_port port
;
980 struct shash_node
*node
;
981 struct list garbage_list
;
982 struct odp_garbage
*garbage
, *next
;
988 backer
= shash_find_data(&all_dpif_backers
, type
);
995 backer_name
= xasprintf("ovs-%s", type
);
997 /* Remove any existing datapaths, since we assume we're the only
998 * userspace controlling the datapath. */
1000 dp_enumerate_names(type
, &names
);
1001 SSET_FOR_EACH(name
, &names
) {
1002 struct dpif
*old_dpif
;
1004 /* Don't remove our backer if it exists. */
1005 if (!strcmp(name
, backer_name
)) {
1009 if (dpif_open(name
, type
, &old_dpif
)) {
1010 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
1012 dpif_delete(old_dpif
);
1013 dpif_close(old_dpif
);
1016 sset_destroy(&names
);
1018 backer
= xmalloc(sizeof *backer
);
1020 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
1023 VLOG_ERR("failed to open datapath of type %s: %s", type
,
1028 backer
->type
= xstrdup(type
);
1029 backer
->refcount
= 1;
1030 hmap_init(&backer
->odp_to_ofport_map
);
1031 timer_set_duration(&backer
->next_expiration
, 1000);
1034 dpif_flow_flush(backer
->dpif
);
1036 /* Loop through the ports already on the datapath and remove any
1037 * that we don't need anymore. */
1038 list_init(&garbage_list
);
1039 dpif_port_dump_start(&port_dump
, backer
->dpif
);
1040 while (dpif_port_dump_next(&port_dump
, &port
)) {
1041 node
= shash_find(&init_ofp_ports
, port
.name
);
1042 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
1043 garbage
= xmalloc(sizeof *garbage
);
1044 garbage
->odp_port
= port
.port_no
;
1045 list_push_front(&garbage_list
, &garbage
->list_node
);
1048 dpif_port_dump_done(&port_dump
);
1050 LIST_FOR_EACH_SAFE (garbage
, next
, list_node
, &garbage_list
) {
1051 dpif_port_del(backer
->dpif
, garbage
->odp_port
);
1052 list_remove(&garbage
->list_node
);
1056 shash_add(&all_dpif_backers
, type
, backer
);
1058 error
= dpif_recv_set(backer
->dpif
, true);
1060 VLOG_ERR("failed to listen on datapath of type %s: %s",
1061 type
, strerror(error
));
1062 close_dpif_backer(backer
);
1070 construct(struct ofproto
*ofproto_
)
1072 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1073 struct shash_node
*node
, *next
;
1078 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1083 max_ports
= dpif_get_max_ports(ofproto
->backer
->dpif
);
1084 ofproto_init_max_ports(ofproto_
, MIN(max_ports
, OFPP_MAX
));
1086 ofproto
->n_matches
= 0;
1088 ofproto
->netflow
= NULL
;
1089 ofproto
->sflow
= NULL
;
1090 ofproto
->stp
= NULL
;
1091 hmap_init(&ofproto
->bundles
);
1092 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1093 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
1094 ofproto
->mirrors
[i
] = NULL
;
1096 ofproto
->has_bonded_bundles
= false;
1098 hmap_init(&ofproto
->facets
);
1099 hmap_init(&ofproto
->subfacets
);
1100 ofproto
->governor
= NULL
;
1102 for (i
= 0; i
< N_TABLES
; i
++) {
1103 struct table_dpif
*table
= &ofproto
->tables
[i
];
1105 table
->catchall_table
= NULL
;
1106 table
->other_table
= NULL
;
1107 table
->basis
= random_uint32();
1109 ofproto
->need_revalidate
= 0;
1110 tag_set_init(&ofproto
->revalidate_set
);
1112 list_init(&ofproto
->completions
);
1114 ofproto_dpif_unixctl_init();
1116 ofproto
->has_mirrors
= false;
1117 ofproto
->has_bundle_action
= false;
1119 hmap_init(&ofproto
->vlandev_map
);
1120 hmap_init(&ofproto
->realdev_vid_map
);
1122 sset_init(&ofproto
->ports
);
1123 sset_init(&ofproto
->port_poll_set
);
1124 ofproto
->port_poll_errno
= 0;
1126 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1127 struct iface_hint
*iface_hint
= node
->data
;
1129 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1130 /* Check if the datapath already has this port. */
1131 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1132 sset_add(&ofproto
->ports
, node
->name
);
1135 free(iface_hint
->br_name
);
1136 free(iface_hint
->br_type
);
1138 shash_delete(&init_ofp_ports
, node
);
1142 hmap_insert(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
,
1143 hash_string(ofproto
->up
.name
, 0));
1144 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1146 ofproto_init_tables(ofproto_
, N_TABLES
);
1147 error
= add_internal_flows(ofproto
);
1148 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1154 add_internal_flow(struct ofproto_dpif
*ofproto
, int id
,
1155 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1157 struct ofputil_flow_mod fm
;
1160 match_init_catchall(&fm
.match
);
1162 match_set_reg(&fm
.match
, 0, id
);
1163 fm
.new_cookie
= htonll(0);
1164 fm
.cookie
= htonll(0);
1165 fm
.cookie_mask
= htonll(0);
1166 fm
.table_id
= TBL_INTERNAL
;
1167 fm
.command
= OFPFC_ADD
;
1168 fm
.idle_timeout
= 0;
1169 fm
.hard_timeout
= 0;
1173 fm
.ofpacts
= ofpacts
->data
;
1174 fm
.ofpacts_len
= ofpacts
->size
;
1176 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
1178 VLOG_ERR_RL(&rl
, "failed to add internal flow %d (%s)",
1179 id
, ofperr_to_string(error
));
1183 *rulep
= rule_dpif_lookup__(ofproto
, &fm
.match
.flow
, TBL_INTERNAL
);
1184 ovs_assert(*rulep
!= NULL
);
1190 add_internal_flows(struct ofproto_dpif
*ofproto
)
1192 struct ofpact_controller
*controller
;
1193 uint64_t ofpacts_stub
[128 / 8];
1194 struct ofpbuf ofpacts
;
1198 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1201 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1202 controller
->max_len
= UINT16_MAX
;
1203 controller
->controller_id
= 0;
1204 controller
->reason
= OFPR_NO_MATCH
;
1205 ofpact_pad(&ofpacts
);
1207 error
= add_internal_flow(ofproto
, id
++, &ofpacts
, &ofproto
->miss_rule
);
1212 ofpbuf_clear(&ofpacts
);
1213 error
= add_internal_flow(ofproto
, id
++, &ofpacts
,
1214 &ofproto
->no_packet_in_rule
);
1219 complete_operations(struct ofproto_dpif
*ofproto
)
1221 struct dpif_completion
*c
, *next
;
1223 LIST_FOR_EACH_SAFE (c
, next
, list_node
, &ofproto
->completions
) {
1224 ofoperation_complete(c
->op
, 0);
1225 list_remove(&c
->list_node
);
1231 destruct(struct ofproto
*ofproto_
)
1233 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1234 struct rule_dpif
*rule
, *next_rule
;
1235 struct oftable
*table
;
1238 hmap_remove(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
);
1239 complete_operations(ofproto
);
1241 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1242 struct cls_cursor cursor
;
1244 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1245 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
1246 ofproto_rule_destroy(&rule
->up
);
1250 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
1251 mirror_destroy(ofproto
->mirrors
[i
]);
1254 netflow_destroy(ofproto
->netflow
);
1255 dpif_sflow_destroy(ofproto
->sflow
);
1256 hmap_destroy(&ofproto
->bundles
);
1257 mac_learning_destroy(ofproto
->ml
);
1259 hmap_destroy(&ofproto
->facets
);
1260 hmap_destroy(&ofproto
->subfacets
);
1261 governor_destroy(ofproto
->governor
);
1263 hmap_destroy(&ofproto
->vlandev_map
);
1264 hmap_destroy(&ofproto
->realdev_vid_map
);
1266 sset_destroy(&ofproto
->ports
);
1267 sset_destroy(&ofproto
->port_poll_set
);
1269 close_dpif_backer(ofproto
->backer
);
1273 run_fast(struct ofproto
*ofproto_
)
1275 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1276 struct ofport_dpif
*ofport
;
1278 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1279 port_run_fast(ofport
);
1286 run(struct ofproto
*ofproto_
)
1288 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1289 struct ofport_dpif
*ofport
;
1290 struct ofbundle
*bundle
;
1294 complete_operations(ofproto
);
1297 error
= run_fast(ofproto_
);
1302 if (ofproto
->netflow
) {
1303 if (netflow_run(ofproto
->netflow
)) {
1304 send_netflow_active_timeouts(ofproto
);
1307 if (ofproto
->sflow
) {
1308 dpif_sflow_run(ofproto
->sflow
);
1311 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1314 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1319 mac_learning_run(ofproto
->ml
, &ofproto
->revalidate_set
);
1321 /* Now revalidate if there's anything to do. */
1322 if (ofproto
->need_revalidate
1323 || !tag_set_is_empty(&ofproto
->revalidate_set
)) {
1324 struct tag_set revalidate_set
= ofproto
->revalidate_set
;
1325 bool revalidate_all
= ofproto
->need_revalidate
;
1326 struct facet
*facet
;
1328 switch (ofproto
->need_revalidate
) {
1329 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
1330 case REV_STP
: COVERAGE_INC(rev_stp
); break;
1331 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
1332 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
1333 case REV_INCONSISTENCY
: COVERAGE_INC(rev_inconsistency
); break;
1336 /* Clear the revalidation flags. */
1337 tag_set_init(&ofproto
->revalidate_set
);
1338 ofproto
->need_revalidate
= 0;
1340 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
1342 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1343 facet_revalidate(facet
);
1348 /* Check the consistency of a random facet, to aid debugging. */
1349 if (!hmap_is_empty(&ofproto
->facets
) && !ofproto
->need_revalidate
) {
1350 struct facet
*facet
;
1352 facet
= CONTAINER_OF(hmap_random_node(&ofproto
->facets
),
1353 struct facet
, hmap_node
);
1354 if (!tag_set_intersects(&ofproto
->revalidate_set
, facet
->tags
)) {
1355 if (!facet_check_consistency(facet
)) {
1356 ofproto
->need_revalidate
= REV_INCONSISTENCY
;
1361 if (ofproto
->governor
) {
1364 governor_run(ofproto
->governor
);
1366 /* If the governor has shrunk to its minimum size and the number of
1367 * subfacets has dwindled, then drop the governor entirely.
1369 * For hysteresis, the number of subfacets to drop the governor is
1370 * smaller than the number needed to trigger its creation. */
1371 n_subfacets
= hmap_count(&ofproto
->subfacets
);
1372 if (n_subfacets
* 4 < ofproto
->up
.flow_eviction_threshold
1373 && governor_is_idle(ofproto
->governor
)) {
1374 governor_destroy(ofproto
->governor
);
1375 ofproto
->governor
= NULL
;
1383 wait(struct ofproto
*ofproto_
)
1385 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1386 struct ofport_dpif
*ofport
;
1387 struct ofbundle
*bundle
;
1389 if (!clogged
&& !list_is_empty(&ofproto
->completions
)) {
1390 poll_immediate_wake();
1393 dpif_wait(ofproto
->backer
->dpif
);
1394 dpif_recv_wait(ofproto
->backer
->dpif
);
1395 if (ofproto
->sflow
) {
1396 dpif_sflow_wait(ofproto
->sflow
);
1398 if (!tag_set_is_empty(&ofproto
->revalidate_set
)) {
1399 poll_immediate_wake();
1401 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1404 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1405 bundle_wait(bundle
);
1407 if (ofproto
->netflow
) {
1408 netflow_wait(ofproto
->netflow
);
1410 mac_learning_wait(ofproto
->ml
);
1412 if (ofproto
->need_revalidate
) {
1413 /* Shouldn't happen, but if it does just go around again. */
1414 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1415 poll_immediate_wake();
1417 if (ofproto
->governor
) {
1418 governor_wait(ofproto
->governor
);
1423 get_memory_usage(const struct ofproto
*ofproto_
, struct simap
*usage
)
1425 const struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1427 simap_increase(usage
, "facets", hmap_count(&ofproto
->facets
));
1428 simap_increase(usage
, "subfacets", hmap_count(&ofproto
->subfacets
));
1432 flush(struct ofproto
*ofproto_
)
1434 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1435 struct subfacet
*subfacet
, *next_subfacet
;
1436 struct subfacet
*batch
[SUBFACET_DESTROY_MAX_BATCH
];
1440 HMAP_FOR_EACH_SAFE (subfacet
, next_subfacet
, hmap_node
,
1441 &ofproto
->subfacets
) {
1442 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
1443 batch
[n_batch
++] = subfacet
;
1444 if (n_batch
>= SUBFACET_DESTROY_MAX_BATCH
) {
1445 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
1449 subfacet_destroy(subfacet
);
1454 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
1459 get_features(struct ofproto
*ofproto_ OVS_UNUSED
,
1460 bool *arp_match_ip
, enum ofputil_action_bitmap
*actions
)
1462 *arp_match_ip
= true;
1463 *actions
= (OFPUTIL_A_OUTPUT
|
1464 OFPUTIL_A_SET_VLAN_VID
|
1465 OFPUTIL_A_SET_VLAN_PCP
|
1466 OFPUTIL_A_STRIP_VLAN
|
1467 OFPUTIL_A_SET_DL_SRC
|
1468 OFPUTIL_A_SET_DL_DST
|
1469 OFPUTIL_A_SET_NW_SRC
|
1470 OFPUTIL_A_SET_NW_DST
|
1471 OFPUTIL_A_SET_NW_TOS
|
1472 OFPUTIL_A_SET_TP_SRC
|
1473 OFPUTIL_A_SET_TP_DST
|
1478 get_tables(struct ofproto
*ofproto_
, struct ofp12_table_stats
*ots
)
1480 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1481 struct dpif_dp_stats s
;
1483 strcpy(ots
->name
, "classifier");
1485 dpif_get_dp_stats(ofproto
->backer
->dpif
, &s
);
1487 ots
->lookup_count
= htonll(s
.n_hit
+ s
.n_missed
);
1488 ots
->matched_count
= htonll(s
.n_hit
+ ofproto
->n_matches
);
1491 static struct ofport
*
1494 struct ofport_dpif
*port
= xmalloc(sizeof *port
);
1499 port_dealloc(struct ofport
*port_
)
1501 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1506 port_construct(struct ofport
*port_
)
1508 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1509 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1510 struct dpif_port dpif_port
;
1513 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1514 port
->bundle
= NULL
;
1516 port
->tag
= tag_create_random();
1517 port
->may_enable
= true;
1518 port
->stp_port
= NULL
;
1519 port
->stp_state
= STP_DISABLED
;
1520 hmap_init(&port
->priorities
);
1521 port
->realdev_ofp_port
= 0;
1522 port
->vlandev_vid
= 0;
1523 port
->carrier_seq
= netdev_get_carrier_resets(port
->up
.netdev
);
1525 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
1526 netdev_get_name(port
->up
.netdev
),
1532 port
->odp_port
= dpif_port
.port_no
;
1534 /* Sanity-check that a mapping doesn't already exist. This
1535 * shouldn't happen. */
1536 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
1537 VLOG_ERR("port %s already has an OpenFlow port number\n",
1542 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
1543 hash_int(port
->odp_port
, 0));
1545 if (ofproto
->sflow
) {
1546 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
1553 port_destruct(struct ofport
*port_
)
1555 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1556 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1557 const char *devname
= netdev_get_name(port
->up
.netdev
);
1559 if (dpif_port_exists(ofproto
->backer
->dpif
, devname
)) {
1560 /* The underlying device is still there, so delete it. This
1561 * happens when the ofproto is being destroyed, since the caller
1562 * assumes that removal of attached ports will happen as part of
1564 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
);
1567 sset_find_and_delete(&ofproto
->ports
, devname
);
1568 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
1569 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1570 bundle_remove(port_
);
1571 set_cfm(port_
, NULL
);
1572 if (ofproto
->sflow
) {
1573 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
1576 ofport_clear_priorities(port
);
1577 hmap_destroy(&port
->priorities
);
1581 port_modified(struct ofport
*port_
)
1583 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1585 if (port
->bundle
&& port
->bundle
->bond
) {
1586 bond_slave_set_netdev(port
->bundle
->bond
, port
, port
->up
.netdev
);
1591 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
1593 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1594 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1595 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
1597 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
1598 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
1599 OFPUTIL_PC_NO_PACKET_IN
)) {
1600 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1602 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
1603 bundle_update(port
->bundle
);
1609 set_sflow(struct ofproto
*ofproto_
,
1610 const struct ofproto_sflow_options
*sflow_options
)
1612 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1613 struct dpif_sflow
*ds
= ofproto
->sflow
;
1615 if (sflow_options
) {
1617 struct ofport_dpif
*ofport
;
1619 ds
= ofproto
->sflow
= dpif_sflow_create();
1620 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1621 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
1623 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1625 dpif_sflow_set_options(ds
, sflow_options
);
1628 dpif_sflow_destroy(ds
);
1629 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1630 ofproto
->sflow
= NULL
;
1637 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
1639 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1646 struct ofproto_dpif
*ofproto
;
1648 ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1649 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1650 ofport
->cfm
= cfm_create(netdev_get_name(ofport
->up
.netdev
));
1653 if (cfm_configure(ofport
->cfm
, s
)) {
1659 cfm_destroy(ofport
->cfm
);
1665 get_cfm_fault(const struct ofport
*ofport_
)
1667 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1669 return ofport
->cfm
? cfm_get_fault(ofport
->cfm
) : -1;
1673 get_cfm_opup(const struct ofport
*ofport_
)
1675 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1677 return ofport
->cfm
? cfm_get_opup(ofport
->cfm
) : -1;
1681 get_cfm_remote_mpids(const struct ofport
*ofport_
, const uint64_t **rmps
,
1684 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1687 cfm_get_remote_mpids(ofport
->cfm
, rmps
, n_rmps
);
1695 get_cfm_health(const struct ofport
*ofport_
)
1697 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1699 return ofport
->cfm
? cfm_get_health(ofport
->cfm
) : -1;
1702 /* Spanning Tree. */
1705 send_bpdu_cb(struct ofpbuf
*pkt
, int port_num
, void *ofproto_
)
1707 struct ofproto_dpif
*ofproto
= ofproto_
;
1708 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
1709 struct ofport_dpif
*ofport
;
1711 ofport
= stp_port_get_aux(sp
);
1713 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
1714 ofproto
->up
.name
, port_num
);
1716 struct eth_header
*eth
= pkt
->l2
;
1718 netdev_get_etheraddr(ofport
->up
.netdev
, eth
->eth_src
);
1719 if (eth_addr_is_zero(eth
->eth_src
)) {
1720 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
1721 "with unknown MAC", ofproto
->up
.name
, port_num
);
1723 send_packet(ofport
, pkt
);
1729 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1731 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
1733 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1735 /* Only revalidate flows if the configuration changed. */
1736 if (!s
!= !ofproto
->stp
) {
1737 ofproto
->need_revalidate
= REV_RECONFIGURE
;
1741 if (!ofproto
->stp
) {
1742 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
1743 send_bpdu_cb
, ofproto
);
1744 ofproto
->stp_last_tick
= time_msec();
1747 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
1748 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
1749 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
1750 stp_set_max_age(ofproto
->stp
, s
->max_age
);
1751 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
1753 struct ofport
*ofport
;
1755 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
1756 set_stp_port(ofport
, NULL
);
1759 stp_destroy(ofproto
->stp
);
1760 ofproto
->stp
= NULL
;
1767 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
1769 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1773 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
1774 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
1775 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
1784 update_stp_port_state(struct ofport_dpif
*ofport
)
1786 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1787 enum stp_state state
;
1789 /* Figure out new state. */
1790 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
1794 if (ofport
->stp_state
!= state
) {
1795 enum ofputil_port_state of_state
;
1798 VLOG_DBG_RL(&rl
, "port %s: STP state changed from %s to %s",
1799 netdev_get_name(ofport
->up
.netdev
),
1800 stp_state_name(ofport
->stp_state
),
1801 stp_state_name(state
));
1802 if (stp_learn_in_state(ofport
->stp_state
)
1803 != stp_learn_in_state(state
)) {
1804 /* xxx Learning action flows should also be flushed. */
1805 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
1807 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
1808 != stp_forward_in_state(state
);
1810 ofproto
->need_revalidate
= REV_STP
;
1811 ofport
->stp_state
= state
;
1812 ofport
->stp_state_entered
= time_msec();
1814 if (fwd_change
&& ofport
->bundle
) {
1815 bundle_update(ofport
->bundle
);
1818 /* Update the STP state bits in the OpenFlow port description. */
1819 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
1820 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
1821 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
1822 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
1823 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
1825 ofproto_port_set_state(&ofport
->up
, of_state
);
1829 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1830 * caller is responsible for assigning STP port numbers and ensuring
1831 * there are no duplicates. */
1833 set_stp_port(struct ofport
*ofport_
,
1834 const struct ofproto_port_stp_settings
*s
)
1836 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1837 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1838 struct stp_port
*sp
= ofport
->stp_port
;
1840 if (!s
|| !s
->enable
) {
1842 ofport
->stp_port
= NULL
;
1843 stp_port_disable(sp
);
1844 update_stp_port_state(ofport
);
1847 } else if (sp
&& stp_port_no(sp
) != s
->port_num
1848 && ofport
== stp_port_get_aux(sp
)) {
1849 /* The port-id changed, so disable the old one if it's not
1850 * already in use by another port. */
1851 stp_port_disable(sp
);
1854 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
1855 stp_port_enable(sp
);
1857 stp_port_set_aux(sp
, ofport
);
1858 stp_port_set_priority(sp
, s
->priority
);
1859 stp_port_set_path_cost(sp
, s
->path_cost
);
1861 update_stp_port_state(ofport
);
1867 get_stp_port_status(struct ofport
*ofport_
,
1868 struct ofproto_port_stp_status
*s
)
1870 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1871 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1872 struct stp_port
*sp
= ofport
->stp_port
;
1874 if (!ofproto
->stp
|| !sp
) {
1880 s
->port_id
= stp_port_get_id(sp
);
1881 s
->state
= stp_port_get_state(sp
);
1882 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
1883 s
->role
= stp_port_get_role(sp
);
1884 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
1890 stp_run(struct ofproto_dpif
*ofproto
)
1893 long long int now
= time_msec();
1894 long long int elapsed
= now
- ofproto
->stp_last_tick
;
1895 struct stp_port
*sp
;
1898 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
1899 ofproto
->stp_last_tick
= now
;
1901 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
1902 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
1905 update_stp_port_state(ofport
);
1909 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
1910 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
1916 stp_wait(struct ofproto_dpif
*ofproto
)
1919 poll_timer_wait(1000);
1923 /* Returns true if STP should process 'flow'. */
1925 stp_should_process_flow(const struct flow
*flow
)
1927 return eth_addr_equals(flow
->dl_dst
, eth_addr_stp
);
1931 stp_process_packet(const struct ofport_dpif
*ofport
,
1932 const struct ofpbuf
*packet
)
1934 struct ofpbuf payload
= *packet
;
1935 struct eth_header
*eth
= payload
.data
;
1936 struct stp_port
*sp
= ofport
->stp_port
;
1938 /* Sink packets on ports that have STP disabled when the bridge has
1940 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
1944 /* Trim off padding on payload. */
1945 if (payload
.size
> ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1946 payload
.size
= ntohs(eth
->eth_type
) + ETH_HEADER_LEN
;
1949 if (ofpbuf_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
1950 stp_received_bpdu(sp
, payload
.data
, payload
.size
);
1954 static struct priority_to_dscp
*
1955 get_priority(const struct ofport_dpif
*ofport
, uint32_t priority
)
1957 struct priority_to_dscp
*pdscp
;
1960 hash
= hash_int(priority
, 0);
1961 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &ofport
->priorities
) {
1962 if (pdscp
->priority
== priority
) {
1970 ofport_clear_priorities(struct ofport_dpif
*ofport
)
1972 struct priority_to_dscp
*pdscp
, *next
;
1974 HMAP_FOR_EACH_SAFE (pdscp
, next
, hmap_node
, &ofport
->priorities
) {
1975 hmap_remove(&ofport
->priorities
, &pdscp
->hmap_node
);
1981 set_queues(struct ofport
*ofport_
,
1982 const struct ofproto_port_queue
*qdscp_list
,
1985 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1986 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1987 struct hmap
new = HMAP_INITIALIZER(&new);
1990 for (i
= 0; i
< n_qdscp
; i
++) {
1991 struct priority_to_dscp
*pdscp
;
1995 dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
1996 if (dpif_queue_to_priority(ofproto
->backer
->dpif
, qdscp_list
[i
].queue
,
2001 pdscp
= get_priority(ofport
, priority
);
2003 hmap_remove(&ofport
->priorities
, &pdscp
->hmap_node
);
2005 pdscp
= xmalloc(sizeof *pdscp
);
2006 pdscp
->priority
= priority
;
2008 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2011 if (pdscp
->dscp
!= dscp
) {
2013 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2016 hmap_insert(&new, &pdscp
->hmap_node
, hash_int(pdscp
->priority
, 0));
2019 if (!hmap_is_empty(&ofport
->priorities
)) {
2020 ofport_clear_priorities(ofport
);
2021 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2024 hmap_swap(&new, &ofport
->priorities
);
2032 /* Expires all MAC learning entries associated with 'bundle' and forces its
2033 * ofproto to revalidate every flow.
2035 * Normally MAC learning entries are removed only from the ofproto associated
2036 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2037 * are removed from every ofproto. When patch ports and SLB bonds are in use
2038 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2039 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2040 * with the host from which it migrated. */
2042 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
2044 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2045 struct mac_learning
*ml
= ofproto
->ml
;
2046 struct mac_entry
*mac
, *next_mac
;
2048 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2049 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
2050 if (mac
->port
.p
== bundle
) {
2052 struct ofproto_dpif
*o
;
2054 HMAP_FOR_EACH (o
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2056 struct mac_entry
*e
;
2058 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
,
2061 tag_set_add(&o
->revalidate_set
, e
->tag
);
2062 mac_learning_expire(o
->ml
, e
);
2068 mac_learning_expire(ml
, mac
);
2073 static struct ofbundle
*
2074 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
2076 struct ofbundle
*bundle
;
2078 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
2079 &ofproto
->bundles
) {
2080 if (bundle
->aux
== aux
) {
2087 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2088 * ones that are found to 'bundles'. */
2090 bundle_lookup_multiple(struct ofproto_dpif
*ofproto
,
2091 void **auxes
, size_t n_auxes
,
2092 struct hmapx
*bundles
)
2096 hmapx_init(bundles
);
2097 for (i
= 0; i
< n_auxes
; i
++) {
2098 struct ofbundle
*bundle
= bundle_lookup(ofproto
, auxes
[i
]);
2100 hmapx_add(bundles
, bundle
);
2106 bundle_update(struct ofbundle
*bundle
)
2108 struct ofport_dpif
*port
;
2110 bundle
->floodable
= true;
2111 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2112 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2113 || !stp_forward_in_state(port
->stp_state
)) {
2114 bundle
->floodable
= false;
2121 bundle_del_port(struct ofport_dpif
*port
)
2123 struct ofbundle
*bundle
= port
->bundle
;
2125 bundle
->ofproto
->need_revalidate
= REV_RECONFIGURE
;
2127 list_remove(&port
->bundle_node
);
2128 port
->bundle
= NULL
;
2131 lacp_slave_unregister(bundle
->lacp
, port
);
2134 bond_slave_unregister(bundle
->bond
, port
);
2137 bundle_update(bundle
);
2141 bundle_add_port(struct ofbundle
*bundle
, uint32_t ofp_port
,
2142 struct lacp_slave_settings
*lacp
,
2143 uint32_t bond_stable_id
)
2145 struct ofport_dpif
*port
;
2147 port
= get_ofp_port(bundle
->ofproto
, ofp_port
);
2152 if (port
->bundle
!= bundle
) {
2153 bundle
->ofproto
->need_revalidate
= REV_RECONFIGURE
;
2155 bundle_del_port(port
);
2158 port
->bundle
= bundle
;
2159 list_push_back(&bundle
->ports
, &port
->bundle_node
);
2160 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2161 || !stp_forward_in_state(port
->stp_state
)) {
2162 bundle
->floodable
= false;
2166 port
->bundle
->ofproto
->need_revalidate
= REV_RECONFIGURE
;
2167 lacp_slave_register(bundle
->lacp
, port
, lacp
);
2170 port
->bond_stable_id
= bond_stable_id
;
2176 bundle_destroy(struct ofbundle
*bundle
)
2178 struct ofproto_dpif
*ofproto
;
2179 struct ofport_dpif
*port
, *next_port
;
2186 ofproto
= bundle
->ofproto
;
2187 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2188 struct ofmirror
*m
= ofproto
->mirrors
[i
];
2190 if (m
->out
== bundle
) {
2192 } else if (hmapx_find_and_delete(&m
->srcs
, bundle
)
2193 || hmapx_find_and_delete(&m
->dsts
, bundle
)) {
2194 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2199 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2200 bundle_del_port(port
);
2203 bundle_flush_macs(bundle
, true);
2204 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
2206 free(bundle
->trunks
);
2207 lacp_destroy(bundle
->lacp
);
2208 bond_destroy(bundle
->bond
);
2213 bundle_set(struct ofproto
*ofproto_
, void *aux
,
2214 const struct ofproto_bundle_settings
*s
)
2216 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2217 bool need_flush
= false;
2218 struct ofport_dpif
*port
;
2219 struct ofbundle
*bundle
;
2220 unsigned long *trunks
;
2226 bundle_destroy(bundle_lookup(ofproto
, aux
));
2230 ovs_assert(s
->n_slaves
== 1 || s
->bond
!= NULL
);
2231 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_slaves
!= NULL
));
2233 bundle
= bundle_lookup(ofproto
, aux
);
2235 bundle
= xmalloc(sizeof *bundle
);
2237 bundle
->ofproto
= ofproto
;
2238 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
2239 hash_pointer(aux
, 0));
2241 bundle
->name
= NULL
;
2243 list_init(&bundle
->ports
);
2244 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
2246 bundle
->trunks
= NULL
;
2247 bundle
->use_priority_tags
= s
->use_priority_tags
;
2248 bundle
->lacp
= NULL
;
2249 bundle
->bond
= NULL
;
2251 bundle
->floodable
= true;
2253 bundle
->src_mirrors
= 0;
2254 bundle
->dst_mirrors
= 0;
2255 bundle
->mirror_out
= 0;
2258 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
2260 bundle
->name
= xstrdup(s
->name
);
2265 if (!bundle
->lacp
) {
2266 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2267 bundle
->lacp
= lacp_create();
2269 lacp_configure(bundle
->lacp
, s
->lacp
);
2271 lacp_destroy(bundle
->lacp
);
2272 bundle
->lacp
= NULL
;
2275 /* Update set of ports. */
2277 for (i
= 0; i
< s
->n_slaves
; i
++) {
2278 if (!bundle_add_port(bundle
, s
->slaves
[i
],
2279 s
->lacp
? &s
->lacp_slaves
[i
] : NULL
,
2280 s
->bond_stable_ids
? s
->bond_stable_ids
[i
] : 0)) {
2284 if (!ok
|| list_size(&bundle
->ports
) != s
->n_slaves
) {
2285 struct ofport_dpif
*next_port
;
2287 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2288 for (i
= 0; i
< s
->n_slaves
; i
++) {
2289 if (s
->slaves
[i
] == port
->up
.ofp_port
) {
2294 bundle_del_port(port
);
2298 ovs_assert(list_size(&bundle
->ports
) <= s
->n_slaves
);
2300 if (list_is_empty(&bundle
->ports
)) {
2301 bundle_destroy(bundle
);
2305 /* Set VLAN tagging mode */
2306 if (s
->vlan_mode
!= bundle
->vlan_mode
2307 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
2308 bundle
->vlan_mode
= s
->vlan_mode
;
2309 bundle
->use_priority_tags
= s
->use_priority_tags
;
2314 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
2315 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
2317 if (vlan
!= bundle
->vlan
) {
2318 bundle
->vlan
= vlan
;
2322 /* Get trunked VLANs. */
2323 switch (s
->vlan_mode
) {
2324 case PORT_VLAN_ACCESS
:
2328 case PORT_VLAN_TRUNK
:
2329 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2332 case PORT_VLAN_NATIVE_UNTAGGED
:
2333 case PORT_VLAN_NATIVE_TAGGED
:
2334 if (vlan
!= 0 && (!s
->trunks
2335 || !bitmap_is_set(s
->trunks
, vlan
)
2336 || bitmap_is_set(s
->trunks
, 0))) {
2337 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2339 trunks
= bitmap_clone(s
->trunks
, 4096);
2341 trunks
= bitmap_allocate1(4096);
2343 bitmap_set1(trunks
, vlan
);
2344 bitmap_set0(trunks
, 0);
2346 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2353 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
2354 free(bundle
->trunks
);
2355 if (trunks
== s
->trunks
) {
2356 bundle
->trunks
= vlan_bitmap_clone(trunks
);
2358 bundle
->trunks
= trunks
;
2363 if (trunks
!= s
->trunks
) {
2368 if (!list_is_short(&bundle
->ports
)) {
2369 bundle
->ofproto
->has_bonded_bundles
= true;
2371 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
2372 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2375 bundle
->bond
= bond_create(s
->bond
);
2376 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2379 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2380 bond_slave_register(bundle
->bond
, port
, port
->bond_stable_id
,
2384 bond_destroy(bundle
->bond
);
2385 bundle
->bond
= NULL
;
2388 /* If we changed something that would affect MAC learning, un-learn
2389 * everything on this port and force flow revalidation. */
2391 bundle_flush_macs(bundle
, false);
2398 bundle_remove(struct ofport
*port_
)
2400 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2401 struct ofbundle
*bundle
= port
->bundle
;
2404 bundle_del_port(port
);
2405 if (list_is_empty(&bundle
->ports
)) {
2406 bundle_destroy(bundle
);
2407 } else if (list_is_short(&bundle
->ports
)) {
2408 bond_destroy(bundle
->bond
);
2409 bundle
->bond
= NULL
;
2415 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
2417 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 10);
2418 struct ofport_dpif
*port
= port_
;
2419 uint8_t ea
[ETH_ADDR_LEN
];
2422 error
= netdev_get_etheraddr(port
->up
.netdev
, ea
);
2424 struct ofpbuf packet
;
2427 ofpbuf_init(&packet
, 0);
2428 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
2430 memcpy(packet_pdu
, pdu
, pdu_size
);
2432 send_packet(port
, &packet
);
2433 ofpbuf_uninit(&packet
);
2435 VLOG_ERR_RL(&rl
, "port %s: cannot obtain Ethernet address of iface "
2436 "%s (%s)", port
->bundle
->name
,
2437 netdev_get_name(port
->up
.netdev
), strerror(error
));
2442 bundle_send_learning_packets(struct ofbundle
*bundle
)
2444 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2445 int error
, n_packets
, n_errors
;
2446 struct mac_entry
*e
;
2448 error
= n_packets
= n_errors
= 0;
2449 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
2450 if (e
->port
.p
!= bundle
) {
2451 struct ofpbuf
*learning_packet
;
2452 struct ofport_dpif
*port
;
2456 /* The assignment to "port" is unnecessary but makes "grep"ing for
2457 * struct ofport_dpif more effective. */
2458 learning_packet
= bond_compose_learning_packet(bundle
->bond
,
2462 ret
= send_packet(port
, learning_packet
);
2463 ofpbuf_delete(learning_packet
);
2473 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2474 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2475 "packets, last error was: %s",
2476 bundle
->name
, n_errors
, n_packets
, strerror(error
));
2478 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2479 bundle
->name
, n_packets
);
2484 bundle_run(struct ofbundle
*bundle
)
2487 lacp_run(bundle
->lacp
, send_pdu_cb
);
2490 struct ofport_dpif
*port
;
2492 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2493 bond_slave_set_may_enable(bundle
->bond
, port
, port
->may_enable
);
2496 bond_run(bundle
->bond
, &bundle
->ofproto
->revalidate_set
,
2497 lacp_status(bundle
->lacp
));
2498 if (bond_should_send_learning_packets(bundle
->bond
)) {
2499 bundle_send_learning_packets(bundle
);
2505 bundle_wait(struct ofbundle
*bundle
)
2508 lacp_wait(bundle
->lacp
);
2511 bond_wait(bundle
->bond
);
2518 mirror_scan(struct ofproto_dpif
*ofproto
)
2522 for (idx
= 0; idx
< MAX_MIRRORS
; idx
++) {
2523 if (!ofproto
->mirrors
[idx
]) {
2530 static struct ofmirror
*
2531 mirror_lookup(struct ofproto_dpif
*ofproto
, void *aux
)
2535 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2536 struct ofmirror
*mirror
= ofproto
->mirrors
[i
];
2537 if (mirror
&& mirror
->aux
== aux
) {
2545 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2547 mirror_update_dups(struct ofproto_dpif
*ofproto
)
2551 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2552 struct ofmirror
*m
= ofproto
->mirrors
[i
];
2555 m
->dup_mirrors
= MIRROR_MASK_C(1) << i
;
2559 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2560 struct ofmirror
*m1
= ofproto
->mirrors
[i
];
2567 for (j
= i
+ 1; j
< MAX_MIRRORS
; j
++) {
2568 struct ofmirror
*m2
= ofproto
->mirrors
[j
];
2570 if (m2
&& m1
->out
== m2
->out
&& m1
->out_vlan
== m2
->out_vlan
) {
2571 m1
->dup_mirrors
|= MIRROR_MASK_C(1) << j
;
2572 m2
->dup_mirrors
|= m1
->dup_mirrors
;
2579 mirror_set(struct ofproto
*ofproto_
, void *aux
,
2580 const struct ofproto_mirror_settings
*s
)
2582 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2583 mirror_mask_t mirror_bit
;
2584 struct ofbundle
*bundle
;
2585 struct ofmirror
*mirror
;
2586 struct ofbundle
*out
;
2587 struct hmapx srcs
; /* Contains "struct ofbundle *"s. */
2588 struct hmapx dsts
; /* Contains "struct ofbundle *"s. */
2591 mirror
= mirror_lookup(ofproto
, aux
);
2593 mirror_destroy(mirror
);
2599 idx
= mirror_scan(ofproto
);
2601 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2603 ofproto
->up
.name
, MAX_MIRRORS
, s
->name
);
2607 mirror
= ofproto
->mirrors
[idx
] = xzalloc(sizeof *mirror
);
2608 mirror
->ofproto
= ofproto
;
2611 mirror
->out_vlan
= -1;
2612 mirror
->name
= NULL
;
2615 if (!mirror
->name
|| strcmp(s
->name
, mirror
->name
)) {
2617 mirror
->name
= xstrdup(s
->name
);
2620 /* Get the new configuration. */
2621 if (s
->out_bundle
) {
2622 out
= bundle_lookup(ofproto
, s
->out_bundle
);
2624 mirror_destroy(mirror
);
2630 out_vlan
= s
->out_vlan
;
2632 bundle_lookup_multiple(ofproto
, s
->srcs
, s
->n_srcs
, &srcs
);
2633 bundle_lookup_multiple(ofproto
, s
->dsts
, s
->n_dsts
, &dsts
);
2635 /* If the configuration has not changed, do nothing. */
2636 if (hmapx_equals(&srcs
, &mirror
->srcs
)
2637 && hmapx_equals(&dsts
, &mirror
->dsts
)
2638 && vlan_bitmap_equal(mirror
->vlans
, s
->src_vlans
)
2639 && mirror
->out
== out
2640 && mirror
->out_vlan
== out_vlan
)
2642 hmapx_destroy(&srcs
);
2643 hmapx_destroy(&dsts
);
2647 hmapx_swap(&srcs
, &mirror
->srcs
);
2648 hmapx_destroy(&srcs
);
2650 hmapx_swap(&dsts
, &mirror
->dsts
);
2651 hmapx_destroy(&dsts
);
2653 free(mirror
->vlans
);
2654 mirror
->vlans
= vlan_bitmap_clone(s
->src_vlans
);
2657 mirror
->out_vlan
= out_vlan
;
2659 /* Update bundles. */
2660 mirror_bit
= MIRROR_MASK_C(1) << mirror
->idx
;
2661 HMAP_FOR_EACH (bundle
, hmap_node
, &mirror
->ofproto
->bundles
) {
2662 if (hmapx_contains(&mirror
->srcs
, bundle
)) {
2663 bundle
->src_mirrors
|= mirror_bit
;
2665 bundle
->src_mirrors
&= ~mirror_bit
;
2668 if (hmapx_contains(&mirror
->dsts
, bundle
)) {
2669 bundle
->dst_mirrors
|= mirror_bit
;
2671 bundle
->dst_mirrors
&= ~mirror_bit
;
2674 if (mirror
->out
== bundle
) {
2675 bundle
->mirror_out
|= mirror_bit
;
2677 bundle
->mirror_out
&= ~mirror_bit
;
2681 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2682 ofproto
->has_mirrors
= true;
2683 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
2684 mirror_update_dups(ofproto
);
2690 mirror_destroy(struct ofmirror
*mirror
)
2692 struct ofproto_dpif
*ofproto
;
2693 mirror_mask_t mirror_bit
;
2694 struct ofbundle
*bundle
;
2701 ofproto
= mirror
->ofproto
;
2702 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2703 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
2705 mirror_bit
= MIRROR_MASK_C(1) << mirror
->idx
;
2706 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
2707 bundle
->src_mirrors
&= ~mirror_bit
;
2708 bundle
->dst_mirrors
&= ~mirror_bit
;
2709 bundle
->mirror_out
&= ~mirror_bit
;
2712 hmapx_destroy(&mirror
->srcs
);
2713 hmapx_destroy(&mirror
->dsts
);
2714 free(mirror
->vlans
);
2716 ofproto
->mirrors
[mirror
->idx
] = NULL
;
2720 mirror_update_dups(ofproto
);
2722 ofproto
->has_mirrors
= false;
2723 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2724 if (ofproto
->mirrors
[i
]) {
2725 ofproto
->has_mirrors
= true;
2732 mirror_get_stats(struct ofproto
*ofproto_
, void *aux
,
2733 uint64_t *packets
, uint64_t *bytes
)
2735 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2736 struct ofmirror
*mirror
= mirror_lookup(ofproto
, aux
);
2739 *packets
= *bytes
= UINT64_MAX
;
2743 *packets
= mirror
->packet_count
;
2744 *bytes
= mirror
->byte_count
;
2750 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
2752 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2753 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
2754 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
2760 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
2762 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2763 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
2764 return bundle
&& bundle
->mirror_out
!= 0;
2768 forward_bpdu_changed(struct ofproto
*ofproto_
)
2770 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2771 ofproto
->need_revalidate
= REV_RECONFIGURE
;
2775 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
2778 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2779 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
2780 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
2785 static struct ofport_dpif
*
2786 get_ofp_port(const struct ofproto_dpif
*ofproto
, uint16_t ofp_port
)
2788 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
2789 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
2792 static struct ofport_dpif
*
2793 get_odp_port(const struct ofproto_dpif
*ofproto
, uint32_t odp_port
)
2795 struct ofport_dpif
*port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
2796 return port
&& &ofproto
->up
== port
->up
.ofproto
? port
: NULL
;
2800 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
2801 struct ofproto_port
*ofproto_port
,
2802 struct dpif_port
*dpif_port
)
2804 ofproto_port
->name
= dpif_port
->name
;
2805 ofproto_port
->type
= dpif_port
->type
;
2806 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
2810 port_run_fast(struct ofport_dpif
*ofport
)
2812 if (ofport
->cfm
&& cfm_should_send_ccm(ofport
->cfm
)) {
2813 struct ofpbuf packet
;
2815 ofpbuf_init(&packet
, 0);
2816 cfm_compose_ccm(ofport
->cfm
, &packet
, ofport
->up
.pp
.hw_addr
);
2817 send_packet(ofport
, &packet
);
2818 ofpbuf_uninit(&packet
);
2823 port_run(struct ofport_dpif
*ofport
)
2825 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
2826 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
2827 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
2829 ofport
->carrier_seq
= carrier_seq
;
2831 port_run_fast(ofport
);
2833 int cfm_opup
= cfm_get_opup(ofport
->cfm
);
2835 cfm_run(ofport
->cfm
);
2836 enable
= enable
&& !cfm_get_fault(ofport
->cfm
);
2838 if (cfm_opup
>= 0) {
2839 enable
= enable
&& cfm_opup
;
2843 if (ofport
->bundle
) {
2844 enable
= enable
&& lacp_slave_may_enable(ofport
->bundle
->lacp
, ofport
);
2845 if (carrier_changed
) {
2846 lacp_slave_carrier_changed(ofport
->bundle
->lacp
, ofport
);
2850 if (ofport
->may_enable
!= enable
) {
2851 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2853 if (ofproto
->has_bundle_action
) {
2854 ofproto
->need_revalidate
= REV_PORT_TOGGLED
;
2858 ofport
->may_enable
= enable
;
2862 port_wait(struct ofport_dpif
*ofport
)
2865 cfm_wait(ofport
->cfm
);
2870 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
2871 struct ofproto_port
*ofproto_port
)
2873 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2874 struct dpif_port dpif_port
;
2877 if (!sset_contains(&ofproto
->ports
, devname
)) {
2880 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
2881 devname
, &dpif_port
);
2883 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
2889 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
2891 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2892 uint32_t odp_port
= UINT32_MAX
;
2895 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &odp_port
);
2897 sset_add(&ofproto
->ports
, netdev_get_name(netdev
));
2903 port_del(struct ofproto
*ofproto_
, uint16_t ofp_port
)
2905 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2906 uint32_t odp_port
= ofp_port_to_odp_port(ofproto
, ofp_port
);
2909 if (odp_port
!= OFPP_NONE
) {
2910 error
= dpif_port_del(ofproto
->backer
->dpif
, odp_port
);
2913 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
2915 /* The caller is going to close ofport->up.netdev. If this is a
2916 * bonded port, then the bond is using that netdev, so remove it
2917 * from the bond. The client will need to reconfigure everything
2918 * after deleting ports, so then the slave will get re-added. */
2919 bundle_remove(&ofport
->up
);
2926 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
2928 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2931 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
2933 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
2934 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2936 /* ofproto->stats.tx_packets represents packets that we created
2937 * internally and sent to some port (e.g. packets sent with
2938 * send_packet()). Account for them as if they had come from
2939 * OFPP_LOCAL and got forwarded. */
2941 if (stats
->rx_packets
!= UINT64_MAX
) {
2942 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
2945 if (stats
->rx_bytes
!= UINT64_MAX
) {
2946 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
2949 /* ofproto->stats.rx_packets represents packets that were received on
2950 * some port and we processed internally and dropped (e.g. STP).
2951 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2953 if (stats
->tx_packets
!= UINT64_MAX
) {
2954 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
2957 if (stats
->tx_bytes
!= UINT64_MAX
) {
2958 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
2965 /* Account packets for LOCAL port. */
2967 ofproto_update_local_port_stats(const struct ofproto
*ofproto_
,
2968 size_t tx_size
, size_t rx_size
)
2970 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2973 ofproto
->stats
.rx_packets
++;
2974 ofproto
->stats
.rx_bytes
+= rx_size
;
2977 ofproto
->stats
.tx_packets
++;
2978 ofproto
->stats
.tx_bytes
+= tx_size
;
2982 struct port_dump_state
{
2988 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
2990 struct port_dump_state
*state
;
2992 *statep
= state
= xmalloc(sizeof *state
);
2999 port_dump_next(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
,
3000 struct ofproto_port
*port
)
3002 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3003 struct port_dump_state
*state
= state_
;
3004 struct sset_node
*node
;
3006 while ((node
= sset_at_position(&ofproto
->ports
, &state
->bucket
,
3010 error
= port_query_by_name(ofproto_
, node
->name
, port
);
3011 if (error
!= ENODEV
) {
3020 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
3022 struct port_dump_state
*state
= state_
;
3029 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
3031 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3033 if (ofproto
->port_poll_errno
) {
3034 int error
= ofproto
->port_poll_errno
;
3035 ofproto
->port_poll_errno
= 0;
3039 if (sset_is_empty(&ofproto
->port_poll_set
)) {
3043 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
3048 port_poll_wait(const struct ofproto
*ofproto_
)
3050 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3051 dpif_port_poll_wait(ofproto
->backer
->dpif
);
3055 port_is_lacp_current(const struct ofport
*ofport_
)
3057 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3058 return (ofport
->bundle
&& ofport
->bundle
->lacp
3059 ? lacp_slave_is_current(ofport
->bundle
->lacp
, ofport
)
3063 /* Upcall handling. */
3065 /* Flow miss batching.
3067 * Some dpifs implement operations faster when you hand them off in a batch.
3068 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3069 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3070 * more packets, plus possibly installing the flow in the dpif.
3072 * So far we only batch the operations that affect flow setup time the most.
3073 * It's possible to batch more than that, but the benefit might be minimal. */
3075 struct hmap_node hmap_node
;
3076 struct ofproto_dpif
*ofproto
;
3078 enum odp_key_fitness key_fitness
;
3079 const struct nlattr
*key
;
3081 ovs_be16 initial_tci
;
3082 struct list packets
;
3083 enum dpif_upcall_type upcall_type
;
3084 uint32_t odp_in_port
;
3087 struct flow_miss_op
{
3088 struct dpif_op dpif_op
;
3089 struct subfacet
*subfacet
; /* Subfacet */
3090 void *garbage
; /* Pointer to pass to free(), NULL if none. */
3091 uint64_t stub
[1024 / 8]; /* Temporary buffer. */
3094 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3095 * OpenFlow controller as necessary according to their individual
3096 * configurations. */
3098 send_packet_in_miss(struct ofproto_dpif
*ofproto
, const struct ofpbuf
*packet
,
3099 const struct flow
*flow
)
3101 struct ofputil_packet_in pin
;
3103 pin
.packet
= packet
->data
;
3104 pin
.packet_len
= packet
->size
;
3105 pin
.reason
= OFPR_NO_MATCH
;
3106 pin
.controller_id
= 0;
3111 pin
.send_len
= 0; /* not used for flow table misses */
3113 flow_get_metadata(flow
, &pin
.fmd
);
3115 connmgr_send_packet_in(ofproto
->up
.connmgr
, &pin
);
3118 static enum slow_path_reason
3119 process_special(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
3120 const struct ofpbuf
*packet
)
3122 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, flow
->in_port
);
3128 if (ofport
->cfm
&& cfm_should_process_flow(ofport
->cfm
, flow
)) {
3130 cfm_process_heartbeat(ofport
->cfm
, packet
);
3133 } else if (ofport
->bundle
&& ofport
->bundle
->lacp
3134 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
3136 lacp_process_packet(ofport
->bundle
->lacp
, ofport
, packet
);
3139 } else if (ofproto
->stp
&& stp_should_process_flow(flow
)) {
3141 stp_process_packet(ofport
, packet
);
3148 static struct flow_miss
*
3149 flow_miss_find(struct hmap
*todo
, const struct flow
*flow
, uint32_t hash
)
3151 struct flow_miss
*miss
;
3153 HMAP_FOR_EACH_WITH_HASH (miss
, hmap_node
, hash
, todo
) {
3154 if (flow_equal(&miss
->flow
, flow
)) {
3162 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3163 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3164 * 'miss' is associated with a subfacet the caller must also initialize the
3165 * returned op->subfacet, and if anything needs to be freed after processing
3166 * the op, the caller must initialize op->garbage also. */
3168 init_flow_miss_execute_op(struct flow_miss
*miss
, struct ofpbuf
*packet
,
3169 struct flow_miss_op
*op
)
3171 if (miss
->flow
.vlan_tci
!= miss
->initial_tci
) {
3172 /* This packet was received on a VLAN splinter port. We
3173 * added a VLAN to the packet to make the packet resemble
3174 * the flow, but the actions were composed assuming that
3175 * the packet contained no VLAN. So, we must remove the
3176 * VLAN header from the packet before trying to execute the
3178 eth_pop_vlan(packet
);
3181 op
->subfacet
= NULL
;
3183 op
->dpif_op
.type
= DPIF_OP_EXECUTE
;
3184 op
->dpif_op
.u
.execute
.key
= miss
->key
;
3185 op
->dpif_op
.u
.execute
.key_len
= miss
->key_len
;
3186 op
->dpif_op
.u
.execute
.packet
= packet
;
3189 /* Helper for handle_flow_miss_without_facet() and
3190 * handle_flow_miss_with_facet(). */
3192 handle_flow_miss_common(struct rule_dpif
*rule
,
3193 struct ofpbuf
*packet
, const struct flow
*flow
)
3195 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3197 ofproto
->n_matches
++;
3199 if (rule
->up
.cr
.priority
== FAIL_OPEN_PRIORITY
) {
3201 * Extra-special case for fail-open mode.
3203 * We are in fail-open mode and the packet matched the fail-open
3204 * rule, but we are connected to a controller too. We should send
3205 * the packet up to the controller in the hope that it will try to
3206 * set up a flow and thereby allow us to exit fail-open.
3208 * See the top-level comment in fail-open.c for more information.
3210 send_packet_in_miss(ofproto
, packet
, flow
);
3214 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3215 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3216 * installing a datapath flow. The answer is usually "yes" (a return value of
3217 * true). However, for short flows the cost of bookkeeping is much higher than
3218 * the benefits, so when the datapath holds a large number of flows we impose
3219 * some heuristics to decide which flows are likely to be worth tracking. */
3221 flow_miss_should_make_facet(struct ofproto_dpif
*ofproto
,
3222 struct flow_miss
*miss
, uint32_t hash
)
3224 if (!ofproto
->governor
) {
3227 n_subfacets
= hmap_count(&ofproto
->subfacets
);
3228 if (n_subfacets
* 2 <= ofproto
->up
.flow_eviction_threshold
) {
3232 ofproto
->governor
= governor_create(ofproto
->up
.name
);
3235 return governor_should_install_flow(ofproto
->governor
, hash
,
3236 list_size(&miss
->packets
));
3239 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3240 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3241 * increment '*n_ops'. */
3243 handle_flow_miss_without_facet(struct flow_miss
*miss
,
3244 struct rule_dpif
*rule
,
3245 struct flow_miss_op
*ops
, size_t *n_ops
)
3247 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3248 long long int now
= time_msec();
3249 struct action_xlate_ctx ctx
;
3250 struct ofpbuf
*packet
;
3252 LIST_FOR_EACH (packet
, list_node
, &miss
->packets
) {
3253 struct flow_miss_op
*op
= &ops
[*n_ops
];
3254 struct dpif_flow_stats stats
;
3255 struct ofpbuf odp_actions
;
3257 COVERAGE_INC(facet_suppress
);
3259 ofpbuf_use_stub(&odp_actions
, op
->stub
, sizeof op
->stub
);
3261 dpif_flow_stats_extract(&miss
->flow
, packet
, now
, &stats
);
3262 rule_credit_stats(rule
, &stats
);
3264 action_xlate_ctx_init(&ctx
, ofproto
, &miss
->flow
, miss
->initial_tci
,
3266 ctx
.resubmit_stats
= &stats
;
3267 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
3270 if (odp_actions
.size
) {
3271 struct dpif_execute
*execute
= &op
->dpif_op
.u
.execute
;
3273 init_flow_miss_execute_op(miss
, packet
, op
);
3274 execute
->actions
= odp_actions
.data
;
3275 execute
->actions_len
= odp_actions
.size
;
3276 op
->garbage
= ofpbuf_get_uninit_pointer(&odp_actions
);
3280 ofpbuf_uninit(&odp_actions
);
3285 /* Handles 'miss', which matches 'facet'. May add any required datapath
3286 * operations to 'ops', incrementing '*n_ops' for each new op.
3288 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3289 * This is really important only for new facets: if we just called time_msec()
3290 * here, then the new subfacet or its packets could look (occasionally) as
3291 * though it was used some time after the facet was used. That can make a
3292 * one-packet flow look like it has a nonzero duration, which looks odd in
3293 * e.g. NetFlow statistics. */
3295 handle_flow_miss_with_facet(struct flow_miss
*miss
, struct facet
*facet
,
3297 struct flow_miss_op
*ops
, size_t *n_ops
)
3299 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
3300 enum subfacet_path want_path
;
3301 struct subfacet
*subfacet
;
3302 struct ofpbuf
*packet
;
3304 subfacet
= subfacet_create(facet
, miss
, now
);
3306 LIST_FOR_EACH (packet
, list_node
, &miss
->packets
) {
3307 struct flow_miss_op
*op
= &ops
[*n_ops
];
3308 struct dpif_flow_stats stats
;
3309 struct ofpbuf odp_actions
;
3311 handle_flow_miss_common(facet
->rule
, packet
, &miss
->flow
);
3313 ofpbuf_use_stub(&odp_actions
, op
->stub
, sizeof op
->stub
);
3314 if (!subfacet
->actions
|| subfacet
->slow
) {
3315 subfacet_make_actions(subfacet
, packet
, &odp_actions
);
3318 dpif_flow_stats_extract(&facet
->flow
, packet
, now
, &stats
);
3319 subfacet_update_stats(subfacet
, &stats
);
3321 if (subfacet
->actions_len
) {
3322 struct dpif_execute
*execute
= &op
->dpif_op
.u
.execute
;
3324 init_flow_miss_execute_op(miss
, packet
, op
);
3325 op
->subfacet
= subfacet
;
3326 if (!subfacet
->slow
) {
3327 execute
->actions
= subfacet
->actions
;
3328 execute
->actions_len
= subfacet
->actions_len
;
3329 ofpbuf_uninit(&odp_actions
);
3331 execute
->actions
= odp_actions
.data
;
3332 execute
->actions_len
= odp_actions
.size
;
3333 op
->garbage
= ofpbuf_get_uninit_pointer(&odp_actions
);
3338 ofpbuf_uninit(&odp_actions
);
3342 want_path
= subfacet_want_path(subfacet
->slow
);
3343 if (miss
->upcall_type
== DPIF_UC_MISS
|| subfacet
->path
!= want_path
) {
3344 struct flow_miss_op
*op
= &ops
[(*n_ops
)++];
3345 struct dpif_flow_put
*put
= &op
->dpif_op
.u
.flow_put
;
3347 op
->subfacet
= subfacet
;
3349 op
->dpif_op
.type
= DPIF_OP_FLOW_PUT
;
3350 put
->flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
3351 put
->key
= miss
->key
;
3352 put
->key_len
= miss
->key_len
;
3353 if (want_path
== SF_FAST_PATH
) {
3354 put
->actions
= subfacet
->actions
;
3355 put
->actions_len
= subfacet
->actions_len
;
3357 compose_slow_path(ofproto
, &facet
->flow
, subfacet
->slow
,
3358 op
->stub
, sizeof op
->stub
,
3359 &put
->actions
, &put
->actions_len
);
3365 /* Handles flow miss 'miss'. May add any required datapath operations
3366 * to 'ops', incrementing '*n_ops' for each new op. */
3368 handle_flow_miss(struct flow_miss
*miss
, struct flow_miss_op
*ops
,
3371 struct ofproto_dpif
*ofproto
= miss
->ofproto
;
3372 struct facet
*facet
;
3376 /* The caller must ensure that miss->hmap_node.hash contains
3377 * flow_hash(miss->flow, 0). */
3378 hash
= miss
->hmap_node
.hash
;
3380 facet
= facet_lookup_valid(ofproto
, &miss
->flow
, hash
);
3382 struct rule_dpif
*rule
= rule_dpif_lookup(ofproto
, &miss
->flow
);
3384 if (!flow_miss_should_make_facet(ofproto
, miss
, hash
)) {
3385 handle_flow_miss_without_facet(miss
, rule
, ops
, n_ops
);
3389 facet
= facet_create(rule
, &miss
->flow
, hash
);
3394 handle_flow_miss_with_facet(miss
, facet
, now
, ops
, n_ops
);
3397 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3398 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3399 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3400 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3401 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3402 * 'packet' ingressed.
3404 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3405 * 'flow''s in_port to OFPP_NONE.
3407 * This function does post-processing on data returned from
3408 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3409 * of the upcall processing logic. In particular, if the extracted in_port is
3410 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3411 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3412 * a VLAN header onto 'packet' (if it is nonnull).
3414 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3415 * packet was really received, that is, the actual VLAN TCI extracted by
3416 * odp_flow_key_to_flow(). (This differs from the value returned in
3417 * flow->vlan_tci only for packets received on VLAN splinters.)
3419 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3420 * or some other positive errno if there are other problems. */
3422 ofproto_receive(const struct dpif_backer
*backer
, struct ofpbuf
*packet
,
3423 const struct nlattr
*key
, size_t key_len
,
3424 struct flow
*flow
, enum odp_key_fitness
*fitnessp
,
3425 struct ofproto_dpif
**ofproto
, uint32_t *odp_in_port
,
3426 ovs_be16
*initial_tci
)
3428 const struct ofport_dpif
*port
;
3429 enum odp_key_fitness fitness
;
3432 fitness
= odp_flow_key_to_flow(key
, key_len
, flow
);
3433 if (fitness
== ODP_FIT_ERROR
) {
3439 *initial_tci
= flow
->vlan_tci
;
3443 *odp_in_port
= flow
->in_port
;
3446 port
= odp_port_to_ofport(backer
, flow
->in_port
);
3448 flow
->in_port
= OFPP_NONE
;
3449 error
= ofproto
? ENODEV
: 0;
3454 *ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
3457 flow
->in_port
= port
->up
.ofp_port
;
3458 if (vsp_adjust_flow(ofproto_dpif_cast(port
->up
.ofproto
), flow
)) {
3460 /* Make the packet resemble the flow, so that it gets sent to an
3461 * OpenFlow controller properly, so that it looks correct for
3462 * sFlow, and so that flow_extract() will get the correct vlan_tci
3463 * if it is called on 'packet'.
3465 * The allocated space inside 'packet' probably also contains
3466 * 'key', that is, both 'packet' and 'key' are probably part of a
3467 * struct dpif_upcall (see the large comment on that structure
3468 * definition), so pushing data on 'packet' is in general not a
3469 * good idea since it could overwrite 'key' or free it as a side
3470 * effect. However, it's OK in this special case because we know
3471 * that 'packet' is inside a Netlink attribute: pushing 4 bytes
3472 * will just overwrite the 4-byte "struct nlattr", which is fine
3473 * since we don't need that header anymore. */
3474 eth_push_vlan(packet
, flow
->vlan_tci
);
3477 /* Let the caller know that we can't reproduce 'key' from 'flow'. */
3478 if (fitness
== ODP_FIT_PERFECT
) {
3479 fitness
= ODP_FIT_TOO_MUCH
;
3486 *fitnessp
= fitness
;
3492 handle_miss_upcalls(struct dpif_backer
*backer
, struct dpif_upcall
*upcalls
,
3495 struct dpif_upcall
*upcall
;
3496 struct flow_miss
*miss
;
3497 struct flow_miss misses
[FLOW_MISS_MAX_BATCH
];
3498 struct flow_miss_op flow_miss_ops
[FLOW_MISS_MAX_BATCH
* 2];
3499 struct dpif_op
*dpif_ops
[FLOW_MISS_MAX_BATCH
* 2];
3509 /* Construct the to-do list.
3511 * This just amounts to extracting the flow from each packet and sticking
3512 * the packets that have the same flow in the same "flow_miss" structure so
3513 * that we can process them together. */
3516 for (upcall
= upcalls
; upcall
< &upcalls
[n_upcalls
]; upcall
++) {
3517 struct flow_miss
*miss
= &misses
[n_misses
];
3518 struct flow_miss
*existing_miss
;
3519 struct ofproto_dpif
*ofproto
;
3520 uint32_t odp_in_port
;
3525 error
= ofproto_receive(backer
, upcall
->packet
, upcall
->key
,
3526 upcall
->key_len
, &flow
, &miss
->key_fitness
,
3527 &ofproto
, &odp_in_port
, &miss
->initial_tci
);
3528 if (error
== ENODEV
) {
3529 /* Received packet on port for which we couldn't associate
3530 * an ofproto. This can happen if a port is removed while
3531 * traffic is being received. Print a rate-limited message
3532 * in case it happens frequently. */
3533 VLOG_INFO_RL(&rl
, "received packet on unassociated port %"PRIu32
,
3539 flow_extract(upcall
->packet
, flow
.skb_priority
, flow
.skb_mark
,
3540 &flow
.tunnel
, flow
.in_port
, &miss
->flow
);
3542 /* Add other packets to a to-do list. */
3543 hash
= flow_hash(&miss
->flow
, 0);
3544 existing_miss
= flow_miss_find(&todo
, &miss
->flow
, hash
);
3545 if (!existing_miss
) {
3546 hmap_insert(&todo
, &miss
->hmap_node
, hash
);
3547 miss
->ofproto
= ofproto
;
3548 miss
->key
= upcall
->key
;
3549 miss
->key_len
= upcall
->key_len
;
3550 miss
->upcall_type
= upcall
->type
;
3551 miss
->odp_in_port
= odp_in_port
;
3552 list_init(&miss
->packets
);
3556 miss
= existing_miss
;
3558 list_push_back(&miss
->packets
, &upcall
->packet
->list_node
);
3561 /* Process each element in the to-do list, constructing the set of
3562 * operations to batch. */
3564 HMAP_FOR_EACH (miss
, hmap_node
, &todo
) {
3565 handle_flow_miss(miss
, flow_miss_ops
, &n_ops
);
3567 ovs_assert(n_ops
<= ARRAY_SIZE(flow_miss_ops
));
3569 /* Execute batch. */
3570 for (i
= 0; i
< n_ops
; i
++) {
3571 dpif_ops
[i
] = &flow_miss_ops
[i
].dpif_op
;
3573 dpif_operate(backer
->dpif
, dpif_ops
, n_ops
);
3575 /* Free memory and update facets. */
3576 for (i
= 0; i
< n_ops
; i
++) {
3577 struct flow_miss_op
*op
= &flow_miss_ops
[i
];
3579 switch (op
->dpif_op
.type
) {
3580 case DPIF_OP_EXECUTE
:
3583 case DPIF_OP_FLOW_PUT
:
3584 if (!op
->dpif_op
.error
) {
3585 op
->subfacet
->path
= subfacet_want_path(op
->subfacet
->slow
);
3589 case DPIF_OP_FLOW_DEL
:
3595 hmap_destroy(&todo
);
3598 static enum { SFLOW_UPCALL
, MISS_UPCALL
, BAD_UPCALL
}
3599 classify_upcall(const struct dpif_upcall
*upcall
)
3601 union user_action_cookie cookie
;
3603 /* First look at the upcall type. */
3604 switch (upcall
->type
) {
3605 case DPIF_UC_ACTION
:
3611 case DPIF_N_UC_TYPES
:
3613 VLOG_WARN_RL(&rl
, "upcall has unexpected type %"PRIu32
, upcall
->type
);
3617 /* "action" upcalls need a closer look. */
3618 memcpy(&cookie
, &upcall
->userdata
, sizeof(cookie
));
3619 switch (cookie
.type
) {
3620 case USER_ACTION_COOKIE_SFLOW
:
3621 return SFLOW_UPCALL
;
3623 case USER_ACTION_COOKIE_SLOW_PATH
:
3626 case USER_ACTION_COOKIE_UNSPEC
:
3628 VLOG_WARN_RL(&rl
, "invalid user cookie : 0x%"PRIx64
, upcall
->userdata
);
3634 handle_sflow_upcall(struct dpif_backer
*backer
,
3635 const struct dpif_upcall
*upcall
)
3637 struct ofproto_dpif
*ofproto
;
3638 union user_action_cookie cookie
;
3640 uint32_t odp_in_port
;
3642 if (ofproto_receive(backer
, upcall
->packet
, upcall
->key
, upcall
->key_len
,
3643 &flow
, NULL
, &ofproto
, &odp_in_port
, NULL
)
3644 || !ofproto
->sflow
) {
3648 memcpy(&cookie
, &upcall
->userdata
, sizeof(cookie
));
3649 dpif_sflow_received(ofproto
->sflow
, upcall
->packet
, &flow
,
3650 odp_in_port
, &cookie
);
3654 handle_upcalls(struct dpif_backer
*backer
, unsigned int max_batch
)
3656 struct dpif_upcall misses
[FLOW_MISS_MAX_BATCH
];
3657 struct ofpbuf miss_bufs
[FLOW_MISS_MAX_BATCH
];
3658 uint64_t miss_buf_stubs
[FLOW_MISS_MAX_BATCH
][4096 / 8];
3663 ovs_assert(max_batch
<= FLOW_MISS_MAX_BATCH
);
3666 for (n_processed
= 0; n_processed
< max_batch
; n_processed
++) {
3667 struct dpif_upcall
*upcall
= &misses
[n_misses
];
3668 struct ofpbuf
*buf
= &miss_bufs
[n_misses
];
3671 ofpbuf_use_stub(buf
, miss_buf_stubs
[n_misses
],
3672 sizeof miss_buf_stubs
[n_misses
]);
3673 error
= dpif_recv(backer
->dpif
, upcall
, buf
);
3679 switch (classify_upcall(upcall
)) {
3681 /* Handle it later. */
3686 handle_sflow_upcall(backer
, upcall
);
3696 /* Handle deferred MISS_UPCALL processing. */
3697 handle_miss_upcalls(backer
, misses
, n_misses
);
3698 for (i
= 0; i
< n_misses
; i
++) {
3699 ofpbuf_uninit(&miss_bufs
[i
]);
3705 /* Flow expiration. */
3707 static int subfacet_max_idle(const struct ofproto_dpif
*);
3708 static void update_stats(struct dpif_backer
*);
3709 static void rule_expire(struct rule_dpif
*);
3710 static void expire_subfacets(struct ofproto_dpif
*, int dp_max_idle
);
3712 /* This function is called periodically by run(). Its job is to collect
3713 * updates for the flows that have been installed into the datapath, most
3714 * importantly when they last were used, and then use that information to
3715 * expire flows that have not been used recently.
3717 * Returns the number of milliseconds after which it should be called again. */
3719 expire(struct dpif_backer
*backer
)
3721 struct ofproto_dpif
*ofproto
;
3722 int max_idle
= INT32_MAX
;
3724 /* Update stats for each flow in the backer. */
3725 update_stats(backer
);
3727 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
3728 struct rule_dpif
*rule
, *next_rule
;
3729 struct oftable
*table
;
3732 if (ofproto
->backer
!= backer
) {
3736 /* Expire subfacets that have been idle too long. */
3737 dp_max_idle
= subfacet_max_idle(ofproto
);
3738 expire_subfacets(ofproto
, dp_max_idle
);
3740 max_idle
= MIN(max_idle
, dp_max_idle
);
3742 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3744 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
3745 struct cls_cursor cursor
;
3747 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
3748 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
3753 /* All outstanding data in existing flows has been accounted, so it's a
3754 * good time to do bond rebalancing. */
3755 if (ofproto
->has_bonded_bundles
) {
3756 struct ofbundle
*bundle
;
3758 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
3760 bond_rebalance(bundle
->bond
, &ofproto
->revalidate_set
);
3766 return MIN(max_idle
, 1000);
3769 /* Updates flow table statistics given that the datapath just reported 'stats'
3770 * as 'subfacet''s statistics. */
3772 update_subfacet_stats(struct subfacet
*subfacet
,
3773 const struct dpif_flow_stats
*stats
)
3775 struct facet
*facet
= subfacet
->facet
;
3777 if (stats
->n_packets
>= subfacet
->dp_packet_count
) {
3778 uint64_t extra
= stats
->n_packets
- subfacet
->dp_packet_count
;
3779 facet
->packet_count
+= extra
;
3781 VLOG_WARN_RL(&rl
, "unexpected packet count from the datapath");
3784 if (stats
->n_bytes
>= subfacet
->dp_byte_count
) {
3785 facet
->byte_count
+= stats
->n_bytes
- subfacet
->dp_byte_count
;
3787 VLOG_WARN_RL(&rl
, "unexpected byte count from datapath");
3790 subfacet
->dp_packet_count
= stats
->n_packets
;
3791 subfacet
->dp_byte_count
= stats
->n_bytes
;
3793 facet
->tcp_flags
|= stats
->tcp_flags
;
3795 subfacet_update_time(subfacet
, stats
->used
);
3796 if (facet
->accounted_bytes
< facet
->byte_count
) {
3798 facet_account(facet
);
3799 facet
->accounted_bytes
= facet
->byte_count
;
3801 facet_push_stats(facet
);
3804 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3805 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
3807 delete_unexpected_flow(struct ofproto_dpif
*ofproto
,
3808 const struct nlattr
*key
, size_t key_len
)
3810 if (!VLOG_DROP_WARN(&rl
)) {
3814 odp_flow_key_format(key
, key_len
, &s
);
3815 VLOG_WARN("unexpected flow on %s: %s", ofproto
->up
.name
, ds_cstr(&s
));
3819 COVERAGE_INC(facet_unexpected
);
3820 dpif_flow_del(ofproto
->backer
->dpif
, key
, key_len
, NULL
);
3823 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3825 * This function also pushes statistics updates to rules which each facet
3826 * resubmits into. Generally these statistics will be accurate. However, if a
3827 * facet changes the rule it resubmits into at some time in between
3828 * update_stats() runs, it is possible that statistics accrued to the
3829 * old rule will be incorrectly attributed to the new rule. This could be
3830 * avoided by calling update_stats() whenever rules are created or
3831 * deleted. However, the performance impact of making so many calls to the
3832 * datapath do not justify the benefit of having perfectly accurate statistics.
3835 update_stats(struct dpif_backer
*backer
)
3837 const struct dpif_flow_stats
*stats
;
3838 struct dpif_flow_dump dump
;
3839 const struct nlattr
*key
;
3842 dpif_flow_dump_start(&dump
, backer
->dpif
);
3843 while (dpif_flow_dump_next(&dump
, &key
, &key_len
, NULL
, NULL
, &stats
)) {
3845 struct subfacet
*subfacet
;
3846 enum odp_key_fitness fitness
;
3847 struct ofproto_dpif
*ofproto
;
3848 struct ofport_dpif
*port
;
3851 fitness
= odp_flow_key_to_flow(key
, key_len
, &flow
);
3852 if (fitness
== ODP_FIT_ERROR
) {
3856 port
= odp_port_to_ofport(backer
, flow
.in_port
);
3858 /* This flow is for a port for which we couldn't associate an
3859 * ofproto. This can happen if a port is removed while
3860 * traffic is being received. Ignore this flow, since it
3861 * will get timed out. */
3865 ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
3866 flow
.in_port
= port
->up
.ofp_port
;
3867 key_hash
= odp_flow_key_hash(key
, key_len
);
3869 subfacet
= subfacet_find(ofproto
, key
, key_len
, key_hash
, &flow
);
3870 switch (subfacet
? subfacet
->path
: SF_NOT_INSTALLED
) {
3872 update_subfacet_stats(subfacet
, stats
);
3876 /* Stats are updated per-packet. */
3879 case SF_NOT_INSTALLED
:
3881 delete_unexpected_flow(ofproto
, key
, key_len
);
3885 dpif_flow_dump_done(&dump
);
3888 /* Calculates and returns the number of milliseconds of idle time after which
3889 * subfacets should expire from the datapath. When a subfacet expires, we fold
3890 * its statistics into its facet, and when a facet's last subfacet expires, we
3891 * fold its statistic into its rule. */
3893 subfacet_max_idle(const struct ofproto_dpif
*ofproto
)
3896 * Idle time histogram.
3898 * Most of the time a switch has a relatively small number of subfacets.
3899 * When this is the case we might as well keep statistics for all of them
3900 * in userspace and to cache them in the kernel datapath for performance as
3903 * As the number of subfacets increases, the memory required to maintain
3904 * statistics about them in userspace and in the kernel becomes
3905 * significant. However, with a large number of subfacets it is likely
3906 * that only a few of them are "heavy hitters" that consume a large amount
3907 * of bandwidth. At this point, only heavy hitters are worth caching in
3908 * the kernel and maintaining in userspaces; other subfacets we can
3911 * The technique used to compute the idle time is to build a histogram with
3912 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
3913 * that is installed in the kernel gets dropped in the appropriate bucket.
3914 * After the histogram has been built, we compute the cutoff so that only
3915 * the most-recently-used 1% of subfacets (but at least
3916 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
3917 * the most-recently-used bucket of subfacets is kept, so actually an
3918 * arbitrary number of subfacets can be kept in any given expiration run
3919 * (though the next run will delete most of those unless they receive
3922 * This requires a second pass through the subfacets, in addition to the
3923 * pass made by update_stats(), because the former function never looks at
3924 * uninstallable subfacets.
3926 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
3927 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
3928 int buckets
[N_BUCKETS
] = { 0 };
3929 int total
, subtotal
, bucket
;
3930 struct subfacet
*subfacet
;
3934 total
= hmap_count(&ofproto
->subfacets
);
3935 if (total
<= ofproto
->up
.flow_eviction_threshold
) {
3936 return N_BUCKETS
* BUCKET_WIDTH
;
3939 /* Build histogram. */
3941 HMAP_FOR_EACH (subfacet
, hmap_node
, &ofproto
->subfacets
) {
3942 long long int idle
= now
- subfacet
->used
;
3943 int bucket
= (idle
<= 0 ? 0
3944 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
3945 : (unsigned int) idle
/ BUCKET_WIDTH
);
3949 /* Find the first bucket whose flows should be expired. */
3950 subtotal
= bucket
= 0;
3952 subtotal
+= buckets
[bucket
++];
3953 } while (bucket
< N_BUCKETS
&&
3954 subtotal
< MAX(ofproto
->up
.flow_eviction_threshold
, total
/ 100));
3956 if (VLOG_IS_DBG_ENABLED()) {
3960 ds_put_cstr(&s
, "keep");
3961 for (i
= 0; i
< N_BUCKETS
; i
++) {
3963 ds_put_cstr(&s
, ", drop");
3966 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
3969 VLOG_INFO("%s: %s (msec:count)", ofproto
->up
.name
, ds_cstr(&s
));
3973 return bucket
* BUCKET_WIDTH
;
3977 expire_subfacets(struct ofproto_dpif
*ofproto
, int dp_max_idle
)
3979 /* Cutoff time for most flows. */
3980 long long int normal_cutoff
= time_msec() - dp_max_idle
;
3982 /* We really want to keep flows for special protocols around, so use a more
3983 * conservative cutoff. */
3984 long long int special_cutoff
= time_msec() - 10000;
3986 struct subfacet
*subfacet
, *next_subfacet
;
3987 struct subfacet
*batch
[SUBFACET_DESTROY_MAX_BATCH
];
3991 HMAP_FOR_EACH_SAFE (subfacet
, next_subfacet
, hmap_node
,
3992 &ofproto
->subfacets
) {
3993 long long int cutoff
;
3995 cutoff
= (subfacet
->slow
& (SLOW_CFM
| SLOW_LACP
| SLOW_STP
)
3998 if (subfacet
->used
< cutoff
) {
3999 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
4000 batch
[n_batch
++] = subfacet
;
4001 if (n_batch
>= SUBFACET_DESTROY_MAX_BATCH
) {
4002 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
4006 subfacet_destroy(subfacet
);
4012 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
4016 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4017 * then delete it entirely. */
4019 rule_expire(struct rule_dpif
*rule
)
4021 struct facet
*facet
, *next_facet
;
4025 if (rule
->up
.pending
) {
4026 /* We'll have to expire it later. */
4030 /* Has 'rule' expired? */
4032 if (rule
->up
.hard_timeout
4033 && now
> rule
->up
.modified
+ rule
->up
.hard_timeout
* 1000) {
4034 reason
= OFPRR_HARD_TIMEOUT
;
4035 } else if (rule
->up
.idle_timeout
4036 && now
> rule
->up
.used
+ rule
->up
.idle_timeout
* 1000) {
4037 reason
= OFPRR_IDLE_TIMEOUT
;
4042 COVERAGE_INC(ofproto_dpif_expired
);
4044 /* Update stats. (This is a no-op if the rule expired due to an idle
4045 * timeout, because that only happens when the rule has no facets left.) */
4046 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4047 facet_remove(facet
);
4050 /* Get rid of the rule. */
4051 ofproto_rule_expire(&rule
->up
, reason
);
4056 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4058 * The caller must already have determined that no facet with an identical
4059 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4060 * the ofproto's classifier table.
4062 * 'hash' must be the return value of flow_hash(flow, 0).
4064 * The facet will initially have no subfacets. The caller should create (at
4065 * least) one subfacet with subfacet_create(). */
4066 static struct facet
*
4067 facet_create(struct rule_dpif
*rule
, const struct flow
*flow
, uint32_t hash
)
4069 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
4070 struct facet
*facet
;
4072 facet
= xzalloc(sizeof *facet
);
4073 facet
->used
= time_msec();
4074 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, hash
);
4075 list_push_back(&rule
->facets
, &facet
->list_node
);
4077 facet
->flow
= *flow
;
4078 list_init(&facet
->subfacets
);
4079 netflow_flow_init(&facet
->nf_flow
);
4080 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
4086 facet_free(struct facet
*facet
)
4091 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4092 * 'packet', which arrived on 'in_port'.
4094 * Takes ownership of 'packet'. */
4096 execute_odp_actions(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
4097 const struct nlattr
*odp_actions
, size_t actions_len
,
4098 struct ofpbuf
*packet
)
4100 struct odputil_keybuf keybuf
;
4104 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
4105 odp_flow_key_from_flow(&key
, flow
,
4106 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
4108 error
= dpif_execute(ofproto
->backer
->dpif
, key
.data
, key
.size
,
4109 odp_actions
, actions_len
, packet
);
4111 ofpbuf_delete(packet
);
4115 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4117 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4118 * rule's statistics, via subfacet_uninstall().
4120 * - Removes 'facet' from its rule and from ofproto->facets.
4123 facet_remove(struct facet
*facet
)
4125 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4126 struct subfacet
*subfacet
, *next_subfacet
;
4128 ovs_assert(!list_is_empty(&facet
->subfacets
));
4130 /* First uninstall all of the subfacets to get final statistics. */
4131 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4132 subfacet_uninstall(subfacet
);
4135 /* Flush the final stats to the rule.
4137 * This might require us to have at least one subfacet around so that we
4138 * can use its actions for accounting in facet_account(), which is why we
4139 * have uninstalled but not yet destroyed the subfacets. */
4140 facet_flush_stats(facet
);
4142 /* Now we're really all done so destroy everything. */
4143 LIST_FOR_EACH_SAFE (subfacet
, next_subfacet
, list_node
,
4144 &facet
->subfacets
) {
4145 subfacet_destroy__(subfacet
);
4147 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
4148 list_remove(&facet
->list_node
);
4152 /* Feed information from 'facet' back into the learning table to keep it in
4153 * sync with what is actually flowing through the datapath. */
4155 facet_learn(struct facet
*facet
)
4157 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4158 struct action_xlate_ctx ctx
;
4160 if (!facet
->has_learn
4161 && !facet
->has_normal
4162 && (!facet
->has_fin_timeout
4163 || !(facet
->tcp_flags
& (TCP_FIN
| TCP_RST
)))) {
4167 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
4168 facet
->flow
.vlan_tci
,
4169 facet
->rule
, facet
->tcp_flags
, NULL
);
4170 ctx
.may_learn
= true;
4171 xlate_actions_for_side_effects(&ctx
, facet
->rule
->up
.ofpacts
,
4172 facet
->rule
->up
.ofpacts_len
);
4176 facet_account(struct facet
*facet
)
4178 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4179 struct subfacet
*subfacet
;
4180 const struct nlattr
*a
;
4185 if (!facet
->has_normal
|| !ofproto
->has_bonded_bundles
) {
4188 n_bytes
= facet
->byte_count
- facet
->accounted_bytes
;
4190 /* This loop feeds byte counters to bond_account() for rebalancing to use
4191 * as a basis. We also need to track the actual VLAN on which the packet
4192 * is going to be sent to ensure that it matches the one passed to
4193 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4196 * We use the actions from an arbitrary subfacet because they should all
4197 * be equally valid for our purpose. */
4198 subfacet
= CONTAINER_OF(list_front(&facet
->subfacets
),
4199 struct subfacet
, list_node
);
4200 vlan_tci
= facet
->flow
.vlan_tci
;
4201 NL_ATTR_FOR_EACH_UNSAFE (a
, left
,
4202 subfacet
->actions
, subfacet
->actions_len
) {
4203 const struct ovs_action_push_vlan
*vlan
;
4204 struct ofport_dpif
*port
;
4206 switch (nl_attr_type(a
)) {
4207 case OVS_ACTION_ATTR_OUTPUT
:
4208 port
= get_odp_port(ofproto
, nl_attr_get_u32(a
));
4209 if (port
&& port
->bundle
&& port
->bundle
->bond
) {
4210 bond_account(port
->bundle
->bond
, &facet
->flow
,
4211 vlan_tci_to_vid(vlan_tci
), n_bytes
);
4215 case OVS_ACTION_ATTR_POP_VLAN
:
4216 vlan_tci
= htons(0);
4219 case OVS_ACTION_ATTR_PUSH_VLAN
:
4220 vlan
= nl_attr_get(a
);
4221 vlan_tci
= vlan
->vlan_tci
;
4227 /* Returns true if the only action for 'facet' is to send to the controller.
4228 * (We don't report NetFlow expiration messages for such facets because they
4229 * are just part of the control logic for the network, not real traffic). */
4231 facet_is_controller_flow(struct facet
*facet
)
4234 const struct rule
*rule
= &facet
->rule
->up
;
4235 const struct ofpact
*ofpacts
= rule
->ofpacts
;
4236 size_t ofpacts_len
= rule
->ofpacts_len
;
4238 if (ofpacts_len
> 0 &&
4239 ofpacts
->type
== OFPACT_CONTROLLER
&&
4240 ofpact_next(ofpacts
) >= ofpact_end(ofpacts
, ofpacts_len
)) {
4247 /* Folds all of 'facet''s statistics into its rule. Also updates the
4248 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4249 * 'facet''s statistics in the datapath should have been zeroed and folded into
4250 * its packet and byte counts before this function is called. */
4252 facet_flush_stats(struct facet
*facet
)
4254 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4255 struct subfacet
*subfacet
;
4257 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4258 ovs_assert(!subfacet
->dp_byte_count
);
4259 ovs_assert(!subfacet
->dp_packet_count
);
4262 facet_push_stats(facet
);
4263 if (facet
->accounted_bytes
< facet
->byte_count
) {
4264 facet_account(facet
);
4265 facet
->accounted_bytes
= facet
->byte_count
;
4268 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
4269 struct ofexpired expired
;
4270 expired
.flow
= facet
->flow
;
4271 expired
.packet_count
= facet
->packet_count
;
4272 expired
.byte_count
= facet
->byte_count
;
4273 expired
.used
= facet
->used
;
4274 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4277 facet
->rule
->packet_count
+= facet
->packet_count
;
4278 facet
->rule
->byte_count
+= facet
->byte_count
;
4280 /* Reset counters to prevent double counting if 'facet' ever gets
4282 facet_reset_counters(facet
);
4284 netflow_flow_clear(&facet
->nf_flow
);
4285 facet
->tcp_flags
= 0;
4288 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4289 * Returns it if found, otherwise a null pointer.
4291 * 'hash' must be the return value of flow_hash(flow, 0).
4293 * The returned facet might need revalidation; use facet_lookup_valid()
4294 * instead if that is important. */
4295 static struct facet
*
4296 facet_find(struct ofproto_dpif
*ofproto
,
4297 const struct flow
*flow
, uint32_t hash
)
4299 struct facet
*facet
;
4301 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, hash
, &ofproto
->facets
) {
4302 if (flow_equal(flow
, &facet
->flow
)) {
4310 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4311 * Returns it if found, otherwise a null pointer.
4313 * 'hash' must be the return value of flow_hash(flow, 0).
4315 * The returned facet is guaranteed to be valid. */
4316 static struct facet
*
4317 facet_lookup_valid(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
4320 struct facet
*facet
;
4322 facet
= facet_find(ofproto
, flow
, hash
);
4324 && (ofproto
->need_revalidate
4325 || tag_set_intersects(&ofproto
->revalidate_set
, facet
->tags
))) {
4326 facet_revalidate(facet
);
4333 subfacet_path_to_string(enum subfacet_path path
)
4336 case SF_NOT_INSTALLED
:
4337 return "not installed";
4339 return "in fast path";
4341 return "in slow path";
4347 /* Returns the path in which a subfacet should be installed if its 'slow'
4348 * member has the specified value. */
4349 static enum subfacet_path
4350 subfacet_want_path(enum slow_path_reason slow
)
4352 return slow
? SF_SLOW_PATH
: SF_FAST_PATH
;
4355 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4356 * supposing that its actions have been recalculated as 'want_actions' and that
4357 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4359 subfacet_should_install(struct subfacet
*subfacet
, enum slow_path_reason slow
,
4360 const struct ofpbuf
*want_actions
)
4362 enum subfacet_path want_path
= subfacet_want_path(slow
);
4363 return (want_path
!= subfacet
->path
4364 || (want_path
== SF_FAST_PATH
4365 && (subfacet
->actions_len
!= want_actions
->size
4366 || memcmp(subfacet
->actions
, want_actions
->data
,
4367 subfacet
->actions_len
))));
4371 facet_check_consistency(struct facet
*facet
)
4373 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 15);
4375 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4377 uint64_t odp_actions_stub
[1024 / 8];
4378 struct ofpbuf odp_actions
;
4380 struct rule_dpif
*rule
;
4381 struct subfacet
*subfacet
;
4382 bool may_log
= false;
4385 /* Check the rule for consistency. */
4386 rule
= rule_dpif_lookup(ofproto
, &facet
->flow
);
4387 ok
= rule
== facet
->rule
;
4389 may_log
= !VLOG_DROP_WARN(&rl
);
4394 flow_format(&s
, &facet
->flow
);
4395 ds_put_format(&s
, ": facet associated with wrong rule (was "
4396 "table=%"PRIu8
",", facet
->rule
->up
.table_id
);
4397 cls_rule_format(&facet
->rule
->up
.cr
, &s
);
4398 ds_put_format(&s
, ") (should have been table=%"PRIu8
",",
4400 cls_rule_format(&rule
->up
.cr
, &s
);
4401 ds_put_char(&s
, ')');
4403 VLOG_WARN("%s", ds_cstr(&s
));
4408 /* Check the datapath actions for consistency. */
4409 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
4410 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4411 enum subfacet_path want_path
;
4412 struct odputil_keybuf keybuf
;
4413 struct action_xlate_ctx ctx
;
4417 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
4418 subfacet
->initial_tci
, rule
, 0, NULL
);
4419 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
4422 if (subfacet
->path
== SF_NOT_INSTALLED
) {
4423 /* This only happens if the datapath reported an error when we
4424 * tried to install the flow. Don't flag another error here. */
4428 want_path
= subfacet_want_path(subfacet
->slow
);
4429 if (want_path
== SF_SLOW_PATH
&& subfacet
->path
== SF_SLOW_PATH
) {
4430 /* The actions for slow-path flows may legitimately vary from one
4431 * packet to the next. We're done. */
4435 if (!subfacet_should_install(subfacet
, subfacet
->slow
, &odp_actions
)) {
4439 /* Inconsistency! */
4441 may_log
= !VLOG_DROP_WARN(&rl
);
4445 /* Rate-limited, skip reporting. */
4450 subfacet_get_key(subfacet
, &keybuf
, &key
);
4451 odp_flow_key_format(key
.data
, key
.size
, &s
);
4453 ds_put_cstr(&s
, ": inconsistency in subfacet");
4454 if (want_path
!= subfacet
->path
) {
4455 enum odp_key_fitness fitness
= subfacet
->key_fitness
;
4457 ds_put_format(&s
, " (%s, fitness=%s)",
4458 subfacet_path_to_string(subfacet
->path
),
4459 odp_key_fitness_to_string(fitness
));
4460 ds_put_format(&s
, " (should have been %s)",
4461 subfacet_path_to_string(want_path
));
4462 } else if (want_path
== SF_FAST_PATH
) {
4463 ds_put_cstr(&s
, " (actions were: ");
4464 format_odp_actions(&s
, subfacet
->actions
,
4465 subfacet
->actions_len
);
4466 ds_put_cstr(&s
, ") (correct actions: ");
4467 format_odp_actions(&s
, odp_actions
.data
, odp_actions
.size
);
4468 ds_put_char(&s
, ')');
4470 ds_put_cstr(&s
, " (actions: ");
4471 format_odp_actions(&s
, subfacet
->actions
,
4472 subfacet
->actions_len
);
4473 ds_put_char(&s
, ')');
4475 VLOG_WARN("%s", ds_cstr(&s
));
4478 ofpbuf_uninit(&odp_actions
);
4483 /* Re-searches the classifier for 'facet':
4485 * - If the rule found is different from 'facet''s current rule, moves
4486 * 'facet' to the new rule and recompiles its actions.
4488 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4489 * where it is and recompiles its actions anyway. */
4491 facet_revalidate(struct facet
*facet
)
4493 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4495 struct nlattr
*odp_actions
;
4498 struct actions
*new_actions
;
4500 struct action_xlate_ctx ctx
;
4501 uint64_t odp_actions_stub
[1024 / 8];
4502 struct ofpbuf odp_actions
;
4504 struct rule_dpif
*new_rule
;
4505 struct subfacet
*subfacet
;
4508 COVERAGE_INC(facet_revalidate
);
4510 new_rule
= rule_dpif_lookup(ofproto
, &facet
->flow
);
4512 /* Calculate new datapath actions.
4514 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4515 * emit a NetFlow expiration and, if so, we need to have the old state
4516 * around to properly compose it. */
4518 /* If the datapath actions changed or the installability changed,
4519 * then we need to talk to the datapath. */
4522 memset(&ctx
, 0, sizeof ctx
);
4523 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
4524 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4525 enum slow_path_reason slow
;
4527 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
4528 subfacet
->initial_tci
, new_rule
, 0, NULL
);
4529 xlate_actions(&ctx
, new_rule
->up
.ofpacts
, new_rule
->up
.ofpacts_len
,
4532 slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
4533 if (subfacet_should_install(subfacet
, slow
, &odp_actions
)) {
4534 struct dpif_flow_stats stats
;
4536 subfacet_install(subfacet
,
4537 odp_actions
.data
, odp_actions
.size
, &stats
, slow
);
4538 subfacet_update_stats(subfacet
, &stats
);
4541 new_actions
= xcalloc(list_size(&facet
->subfacets
),
4542 sizeof *new_actions
);
4544 new_actions
[i
].odp_actions
= xmemdup(odp_actions
.data
,
4546 new_actions
[i
].actions_len
= odp_actions
.size
;
4551 ofpbuf_uninit(&odp_actions
);
4554 facet_flush_stats(facet
);
4557 /* Update 'facet' now that we've taken care of all the old state. */
4558 facet
->tags
= ctx
.tags
;
4559 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
4560 facet
->has_learn
= ctx
.has_learn
;
4561 facet
->has_normal
= ctx
.has_normal
;
4562 facet
->has_fin_timeout
= ctx
.has_fin_timeout
;
4563 facet
->mirrors
= ctx
.mirrors
;
4566 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4567 subfacet
->slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
4569 if (new_actions
&& new_actions
[i
].odp_actions
) {
4570 free(subfacet
->actions
);
4571 subfacet
->actions
= new_actions
[i
].odp_actions
;
4572 subfacet
->actions_len
= new_actions
[i
].actions_len
;
4578 if (facet
->rule
!= new_rule
) {
4579 COVERAGE_INC(facet_changed_rule
);
4580 list_remove(&facet
->list_node
);
4581 list_push_back(&new_rule
->facets
, &facet
->list_node
);
4582 facet
->rule
= new_rule
;
4583 facet
->used
= new_rule
->up
.created
;
4584 facet
->prev_used
= facet
->used
;
4588 /* Updates 'facet''s used time. Caller is responsible for calling
4589 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4591 facet_update_time(struct facet
*facet
, long long int used
)
4593 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4594 if (used
> facet
->used
) {
4596 ofproto_rule_update_used(&facet
->rule
->up
, used
);
4597 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
4602 facet_reset_counters(struct facet
*facet
)
4604 facet
->packet_count
= 0;
4605 facet
->byte_count
= 0;
4606 facet
->prev_packet_count
= 0;
4607 facet
->prev_byte_count
= 0;
4608 facet
->accounted_bytes
= 0;
4612 facet_push_stats(struct facet
*facet
)
4614 struct dpif_flow_stats stats
;
4616 ovs_assert(facet
->packet_count
>= facet
->prev_packet_count
);
4617 ovs_assert(facet
->byte_count
>= facet
->prev_byte_count
);
4618 ovs_assert(facet
->used
>= facet
->prev_used
);
4620 stats
.n_packets
= facet
->packet_count
- facet
->prev_packet_count
;
4621 stats
.n_bytes
= facet
->byte_count
- facet
->prev_byte_count
;
4622 stats
.used
= facet
->used
;
4623 stats
.tcp_flags
= 0;
4625 if (stats
.n_packets
|| stats
.n_bytes
|| facet
->used
> facet
->prev_used
) {
4626 facet
->prev_packet_count
= facet
->packet_count
;
4627 facet
->prev_byte_count
= facet
->byte_count
;
4628 facet
->prev_used
= facet
->used
;
4630 flow_push_stats(facet
->rule
, &facet
->flow
, &stats
);
4632 update_mirror_stats(ofproto_dpif_cast(facet
->rule
->up
.ofproto
),
4633 facet
->mirrors
, stats
.n_packets
, stats
.n_bytes
);
4638 rule_credit_stats(struct rule_dpif
*rule
, const struct dpif_flow_stats
*stats
)
4640 rule
->packet_count
+= stats
->n_packets
;
4641 rule
->byte_count
+= stats
->n_bytes
;
4642 ofproto_rule_update_used(&rule
->up
, stats
->used
);
4645 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4646 * 'rule''s actions and mirrors. */
4648 flow_push_stats(struct rule_dpif
*rule
,
4649 const struct flow
*flow
, const struct dpif_flow_stats
*stats
)
4651 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
4652 struct action_xlate_ctx ctx
;
4654 ofproto_rule_update_used(&rule
->up
, stats
->used
);
4656 action_xlate_ctx_init(&ctx
, ofproto
, flow
, flow
->vlan_tci
, rule
,
4658 ctx
.resubmit_stats
= stats
;
4659 xlate_actions_for_side_effects(&ctx
, rule
->up
.ofpacts
,
4660 rule
->up
.ofpacts_len
);
4665 static struct subfacet
*
4666 subfacet_find(struct ofproto_dpif
*ofproto
,
4667 const struct nlattr
*key
, size_t key_len
, uint32_t key_hash
,
4668 const struct flow
*flow
)
4670 struct subfacet
*subfacet
;
4672 HMAP_FOR_EACH_WITH_HASH (subfacet
, hmap_node
, key_hash
,
4673 &ofproto
->subfacets
) {
4675 ? (subfacet
->key_len
== key_len
4676 && !memcmp(key
, subfacet
->key
, key_len
))
4677 : flow_equal(flow
, &subfacet
->facet
->flow
)) {
4685 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4686 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4687 * existing subfacet if there is one, otherwise creates and returns a
4690 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4691 * which case the caller must populate the actions with
4692 * subfacet_make_actions(). */
4693 static struct subfacet
*
4694 subfacet_create(struct facet
*facet
, struct flow_miss
*miss
,
4697 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4698 enum odp_key_fitness key_fitness
= miss
->key_fitness
;
4699 const struct nlattr
*key
= miss
->key
;
4700 size_t key_len
= miss
->key_len
;
4702 struct subfacet
*subfacet
;
4704 key_hash
= odp_flow_key_hash(key
, key_len
);
4706 if (list_is_empty(&facet
->subfacets
)) {
4707 subfacet
= &facet
->one_subfacet
;
4709 subfacet
= subfacet_find(ofproto
, key
, key_len
, key_hash
,
4712 if (subfacet
->facet
== facet
) {
4716 /* This shouldn't happen. */
4717 VLOG_ERR_RL(&rl
, "subfacet with wrong facet");
4718 subfacet_destroy(subfacet
);
4721 subfacet
= xmalloc(sizeof *subfacet
);
4724 hmap_insert(&ofproto
->subfacets
, &subfacet
->hmap_node
, key_hash
);
4725 list_push_back(&facet
->subfacets
, &subfacet
->list_node
);
4726 subfacet
->facet
= facet
;
4727 subfacet
->key_fitness
= key_fitness
;
4728 if (key_fitness
!= ODP_FIT_PERFECT
) {
4729 subfacet
->key
= xmemdup(key
, key_len
);
4730 subfacet
->key_len
= key_len
;
4732 subfacet
->key
= NULL
;
4733 subfacet
->key_len
= 0;
4735 subfacet
->used
= now
;
4736 subfacet
->dp_packet_count
= 0;
4737 subfacet
->dp_byte_count
= 0;
4738 subfacet
->actions_len
= 0;
4739 subfacet
->actions
= NULL
;
4740 subfacet
->slow
= (subfacet
->key_fitness
== ODP_FIT_TOO_LITTLE
4743 subfacet
->path
= SF_NOT_INSTALLED
;
4744 subfacet
->initial_tci
= miss
->initial_tci
;
4745 subfacet
->odp_in_port
= miss
->odp_in_port
;
4750 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4751 * its facet within 'ofproto', and frees it. */
4753 subfacet_destroy__(struct subfacet
*subfacet
)
4755 struct facet
*facet
= subfacet
->facet
;
4756 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4758 subfacet_uninstall(subfacet
);
4759 hmap_remove(&ofproto
->subfacets
, &subfacet
->hmap_node
);
4760 list_remove(&subfacet
->list_node
);
4761 free(subfacet
->key
);
4762 free(subfacet
->actions
);
4763 if (subfacet
!= &facet
->one_subfacet
) {
4768 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4769 * last remaining subfacet in its facet destroys the facet too. */
4771 subfacet_destroy(struct subfacet
*subfacet
)
4773 struct facet
*facet
= subfacet
->facet
;
4775 if (list_is_singleton(&facet
->subfacets
)) {
4776 /* facet_remove() needs at least one subfacet (it will remove it). */
4777 facet_remove(facet
);
4779 subfacet_destroy__(subfacet
);
4784 subfacet_destroy_batch(struct ofproto_dpif
*ofproto
,
4785 struct subfacet
**subfacets
, int n
)
4787 struct odputil_keybuf keybufs
[SUBFACET_DESTROY_MAX_BATCH
];
4788 struct dpif_op ops
[SUBFACET_DESTROY_MAX_BATCH
];
4789 struct dpif_op
*opsp
[SUBFACET_DESTROY_MAX_BATCH
];
4790 struct ofpbuf keys
[SUBFACET_DESTROY_MAX_BATCH
];
4791 struct dpif_flow_stats stats
[SUBFACET_DESTROY_MAX_BATCH
];
4794 for (i
= 0; i
< n
; i
++) {
4795 ops
[i
].type
= DPIF_OP_FLOW_DEL
;
4796 subfacet_get_key(subfacets
[i
], &keybufs
[i
], &keys
[i
]);
4797 ops
[i
].u
.flow_del
.key
= keys
[i
].data
;
4798 ops
[i
].u
.flow_del
.key_len
= keys
[i
].size
;
4799 ops
[i
].u
.flow_del
.stats
= &stats
[i
];
4803 dpif_operate(ofproto
->backer
->dpif
, opsp
, n
);
4804 for (i
= 0; i
< n
; i
++) {
4805 subfacet_reset_dp_stats(subfacets
[i
], &stats
[i
]);
4806 subfacets
[i
]->path
= SF_NOT_INSTALLED
;
4807 subfacet_destroy(subfacets
[i
]);
4811 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
4812 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
4813 * for use as temporary storage. */
4815 subfacet_get_key(struct subfacet
*subfacet
, struct odputil_keybuf
*keybuf
,
4819 if (!subfacet
->key
) {
4820 struct flow
*flow
= &subfacet
->facet
->flow
;
4822 ofpbuf_use_stack(key
, keybuf
, sizeof *keybuf
);
4823 odp_flow_key_from_flow(key
, flow
, subfacet
->odp_in_port
);
4825 ofpbuf_use_const(key
, subfacet
->key
, subfacet
->key_len
);
4829 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
4830 * Translates the actions into 'odp_actions', which the caller must have
4831 * initialized and is responsible for uninitializing. */
4833 subfacet_make_actions(struct subfacet
*subfacet
, const struct ofpbuf
*packet
,
4834 struct ofpbuf
*odp_actions
)
4836 struct facet
*facet
= subfacet
->facet
;
4837 struct rule_dpif
*rule
= facet
->rule
;
4838 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
4840 struct action_xlate_ctx ctx
;
4842 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
, subfacet
->initial_tci
,
4844 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, odp_actions
);
4845 facet
->tags
= ctx
.tags
;
4846 facet
->has_learn
= ctx
.has_learn
;
4847 facet
->has_normal
= ctx
.has_normal
;
4848 facet
->has_fin_timeout
= ctx
.has_fin_timeout
;
4849 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
4850 facet
->mirrors
= ctx
.mirrors
;
4852 subfacet
->slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
4853 if (subfacet
->actions_len
!= odp_actions
->size
4854 || memcmp(subfacet
->actions
, odp_actions
->data
, odp_actions
->size
)) {
4855 free(subfacet
->actions
);
4856 subfacet
->actions_len
= odp_actions
->size
;
4857 subfacet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
4861 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4862 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
4863 * in the datapath will be zeroed and 'stats' will be updated with traffic new
4864 * since 'subfacet' was last updated.
4866 * Returns 0 if successful, otherwise a positive errno value. */
4868 subfacet_install(struct subfacet
*subfacet
,
4869 const struct nlattr
*actions
, size_t actions_len
,
4870 struct dpif_flow_stats
*stats
,
4871 enum slow_path_reason slow
)
4873 struct facet
*facet
= subfacet
->facet
;
4874 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4875 enum subfacet_path path
= subfacet_want_path(slow
);
4876 uint64_t slow_path_stub
[128 / 8];
4877 struct odputil_keybuf keybuf
;
4878 enum dpif_flow_put_flags flags
;
4882 flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
4884 flags
|= DPIF_FP_ZERO_STATS
;
4887 if (path
== SF_SLOW_PATH
) {
4888 compose_slow_path(ofproto
, &facet
->flow
, slow
,
4889 slow_path_stub
, sizeof slow_path_stub
,
4890 &actions
, &actions_len
);
4893 subfacet_get_key(subfacet
, &keybuf
, &key
);
4894 ret
= dpif_flow_put(ofproto
->backer
->dpif
, flags
, key
.data
, key
.size
,
4895 actions
, actions_len
, stats
);
4898 subfacet_reset_dp_stats(subfacet
, stats
);
4902 subfacet
->path
= path
;
4908 subfacet_reinstall(struct subfacet
*subfacet
, struct dpif_flow_stats
*stats
)
4910 return subfacet_install(subfacet
, subfacet
->actions
, subfacet
->actions_len
,
4911 stats
, subfacet
->slow
);
4914 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4916 subfacet_uninstall(struct subfacet
*subfacet
)
4918 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
4919 struct rule_dpif
*rule
= subfacet
->facet
->rule
;
4920 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
4921 struct odputil_keybuf keybuf
;
4922 struct dpif_flow_stats stats
;
4926 subfacet_get_key(subfacet
, &keybuf
, &key
);
4927 error
= dpif_flow_del(ofproto
->backer
->dpif
,
4928 key
.data
, key
.size
, &stats
);
4929 subfacet_reset_dp_stats(subfacet
, &stats
);
4931 subfacet_update_stats(subfacet
, &stats
);
4933 subfacet
->path
= SF_NOT_INSTALLED
;
4935 ovs_assert(subfacet
->dp_packet_count
== 0);
4936 ovs_assert(subfacet
->dp_byte_count
== 0);
4940 /* Resets 'subfacet''s datapath statistics counters. This should be called
4941 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
4942 * non-null, it should contain the statistics returned by dpif when 'subfacet'
4943 * was reset in the datapath. 'stats' will be modified to include only
4944 * statistics new since 'subfacet' was last updated. */
4946 subfacet_reset_dp_stats(struct subfacet
*subfacet
,
4947 struct dpif_flow_stats
*stats
)
4950 && subfacet
->dp_packet_count
<= stats
->n_packets
4951 && subfacet
->dp_byte_count
<= stats
->n_bytes
) {
4952 stats
->n_packets
-= subfacet
->dp_packet_count
;
4953 stats
->n_bytes
-= subfacet
->dp_byte_count
;
4956 subfacet
->dp_packet_count
= 0;
4957 subfacet
->dp_byte_count
= 0;
4960 /* Updates 'subfacet''s used time. The caller is responsible for calling
4961 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
4963 subfacet_update_time(struct subfacet
*subfacet
, long long int used
)
4965 if (used
> subfacet
->used
) {
4966 subfacet
->used
= used
;
4967 facet_update_time(subfacet
->facet
, used
);
4971 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4973 * Because of the meaning of a subfacet's counters, it only makes sense to do
4974 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4975 * represents a packet that was sent by hand or if it represents statistics
4976 * that have been cleared out of the datapath. */
4978 subfacet_update_stats(struct subfacet
*subfacet
,
4979 const struct dpif_flow_stats
*stats
)
4981 if (stats
->n_packets
|| stats
->used
> subfacet
->used
) {
4982 struct facet
*facet
= subfacet
->facet
;
4984 subfacet_update_time(subfacet
, stats
->used
);
4985 facet
->packet_count
+= stats
->n_packets
;
4986 facet
->byte_count
+= stats
->n_bytes
;
4987 facet
->tcp_flags
|= stats
->tcp_flags
;
4988 facet_push_stats(facet
);
4989 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
4995 static struct rule_dpif
*
4996 rule_dpif_lookup(struct ofproto_dpif
*ofproto
, const struct flow
*flow
)
4998 struct rule_dpif
*rule
;
5000 rule
= rule_dpif_lookup__(ofproto
, flow
, 0);
5005 return rule_dpif_miss_rule(ofproto
, flow
);
5008 static struct rule_dpif
*
5009 rule_dpif_lookup__(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
5012 struct cls_rule
*cls_rule
;
5013 struct classifier
*cls
;
5015 if (table_id
>= N_TABLES
) {
5019 cls
= &ofproto
->up
.tables
[table_id
].cls
;
5020 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
5021 && ofproto
->up
.frag_handling
== OFPC_FRAG_NORMAL
) {
5022 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5023 * are unavailable. */
5024 struct flow ofpc_normal_flow
= *flow
;
5025 ofpc_normal_flow
.tp_src
= htons(0);
5026 ofpc_normal_flow
.tp_dst
= htons(0);
5027 cls_rule
= classifier_lookup(cls
, &ofpc_normal_flow
);
5029 cls_rule
= classifier_lookup(cls
, flow
);
5031 return rule_dpif_cast(rule_from_cls_rule(cls_rule
));
5034 static struct rule_dpif
*
5035 rule_dpif_miss_rule(struct ofproto_dpif
*ofproto
, const struct flow
*flow
)
5037 struct ofport_dpif
*port
;
5039 port
= get_ofp_port(ofproto
, flow
->in_port
);
5041 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, flow
->in_port
);
5042 return ofproto
->miss_rule
;
5045 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_PACKET_IN
) {
5046 return ofproto
->no_packet_in_rule
;
5048 return ofproto
->miss_rule
;
5052 complete_operation(struct rule_dpif
*rule
)
5054 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5056 rule_invalidate(rule
);
5058 struct dpif_completion
*c
= xmalloc(sizeof *c
);
5059 c
->op
= rule
->up
.pending
;
5060 list_push_back(&ofproto
->completions
, &c
->list_node
);
5062 ofoperation_complete(rule
->up
.pending
, 0);
5066 static struct rule
*
5069 struct rule_dpif
*rule
= xmalloc(sizeof *rule
);
5074 rule_dealloc(struct rule
*rule_
)
5076 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5081 rule_construct(struct rule
*rule_
)
5083 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5084 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5085 struct rule_dpif
*victim
;
5088 rule
->packet_count
= 0;
5089 rule
->byte_count
= 0;
5091 victim
= rule_dpif_cast(ofoperation_get_victim(rule
->up
.pending
));
5092 if (victim
&& !list_is_empty(&victim
->facets
)) {
5093 struct facet
*facet
;
5095 rule
->facets
= victim
->facets
;
5096 list_moved(&rule
->facets
);
5097 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
5098 /* XXX: We're only clearing our local counters here. It's possible
5099 * that quite a few packets are unaccounted for in the datapath
5100 * statistics. These will be accounted to the new rule instead of
5101 * cleared as required. This could be fixed by clearing out the
5102 * datapath statistics for this facet, but currently it doesn't
5104 facet_reset_counters(facet
);
5108 /* Must avoid list_moved() in this case. */
5109 list_init(&rule
->facets
);
5112 table_id
= rule
->up
.table_id
;
5114 rule
->tag
= victim
->tag
;
5115 } else if (table_id
== 0) {
5120 miniflow_expand(&rule
->up
.cr
.match
.flow
, &flow
);
5121 rule
->tag
= rule_calculate_tag(&flow
, &rule
->up
.cr
.match
.mask
,
5122 ofproto
->tables
[table_id
].basis
);
5125 complete_operation(rule
);
5130 rule_destruct(struct rule
*rule_
)
5132 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5133 struct facet
*facet
, *next_facet
;
5135 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
5136 facet_revalidate(facet
);
5139 complete_operation(rule
);
5143 rule_get_stats(struct rule
*rule_
, uint64_t *packets
, uint64_t *bytes
)
5145 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5146 struct facet
*facet
;
5148 /* Start from historical data for 'rule' itself that are no longer tracked
5149 * in facets. This counts, for example, facets that have expired. */
5150 *packets
= rule
->packet_count
;
5151 *bytes
= rule
->byte_count
;
5153 /* Add any statistics that are tracked by facets. This includes
5154 * statistical data recently updated by ofproto_update_stats() as well as
5155 * stats for packets that were executed "by hand" via dpif_execute(). */
5156 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
5157 *packets
+= facet
->packet_count
;
5158 *bytes
+= facet
->byte_count
;
5163 rule_execute(struct rule
*rule_
, const struct flow
*flow
,
5164 struct ofpbuf
*packet
)
5166 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5167 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5169 struct dpif_flow_stats stats
;
5171 struct action_xlate_ctx ctx
;
5172 uint64_t odp_actions_stub
[1024 / 8];
5173 struct ofpbuf odp_actions
;
5175 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
5176 rule_credit_stats(rule
, &stats
);
5178 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
5179 action_xlate_ctx_init(&ctx
, ofproto
, flow
, flow
->vlan_tci
,
5180 rule
, stats
.tcp_flags
, packet
);
5181 ctx
.resubmit_stats
= &stats
;
5182 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, &odp_actions
);
5184 execute_odp_actions(ofproto
, flow
, odp_actions
.data
,
5185 odp_actions
.size
, packet
);
5187 ofpbuf_uninit(&odp_actions
);
5193 rule_modify_actions(struct rule
*rule_
)
5195 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5197 complete_operation(rule
);
5200 /* Sends 'packet' out 'ofport'.
5201 * May modify 'packet'.
5202 * Returns 0 if successful, otherwise a positive errno value. */
5204 send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
5206 const struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
5207 struct ofpbuf key
, odp_actions
;
5208 struct odputil_keybuf keybuf
;
5213 flow_extract(packet
, 0, 0, NULL
, OFPP_LOCAL
, &flow
);
5214 odp_port
= vsp_realdev_to_vlandev(ofproto
, ofport
->odp_port
,
5216 if (odp_port
!= ofport
->odp_port
) {
5217 eth_pop_vlan(packet
);
5218 flow
.vlan_tci
= htons(0);
5221 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
5222 odp_flow_key_from_flow(&key
, &flow
,
5223 ofp_port_to_odp_port(ofproto
, flow
.in_port
));
5225 ofpbuf_init(&odp_actions
, 32);
5226 compose_sflow_action(ofproto
, &odp_actions
, &flow
, odp_port
);
5228 nl_msg_put_u32(&odp_actions
, OVS_ACTION_ATTR_OUTPUT
, odp_port
);
5229 error
= dpif_execute(ofproto
->backer
->dpif
,
5231 odp_actions
.data
, odp_actions
.size
,
5233 ofpbuf_uninit(&odp_actions
);
5236 VLOG_WARN_RL(&rl
, "%s: failed to send packet on port %"PRIu32
" (%s)",
5237 ofproto
->up
.name
, odp_port
, strerror(error
));
5239 ofproto_update_local_port_stats(ofport
->up
.ofproto
, packet
->size
, 0);
5243 /* OpenFlow to datapath action translation. */
5245 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
5246 struct action_xlate_ctx
*);
5247 static void xlate_normal(struct action_xlate_ctx
*);
5249 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5250 * The action will state 'slow' as the reason that the action is in the slow
5251 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5252 * dump-flows" output to see why a flow is in the slow path.)
5254 * The 'stub_size' bytes in 'stub' will be used to store the action.
5255 * 'stub_size' must be large enough for the action.
5257 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5260 compose_slow_path(const struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
5261 enum slow_path_reason slow
,
5262 uint64_t *stub
, size_t stub_size
,
5263 const struct nlattr
**actionsp
, size_t *actions_lenp
)
5265 union user_action_cookie cookie
;
5268 cookie
.type
= USER_ACTION_COOKIE_SLOW_PATH
;
5269 cookie
.slow_path
.unused
= 0;
5270 cookie
.slow_path
.reason
= slow
;
5272 ofpbuf_use_stack(&buf
, stub
, stub_size
);
5273 if (slow
& (SLOW_CFM
| SLOW_LACP
| SLOW_STP
)) {
5274 uint32_t pid
= dpif_port_get_pid(ofproto
->backer
->dpif
, UINT32_MAX
);
5275 odp_put_userspace_action(pid
, &cookie
, &buf
);
5277 put_userspace_action(ofproto
, &buf
, flow
, &cookie
);
5279 *actionsp
= buf
.data
;
5280 *actions_lenp
= buf
.size
;
5284 put_userspace_action(const struct ofproto_dpif
*ofproto
,
5285 struct ofpbuf
*odp_actions
,
5286 const struct flow
*flow
,
5287 const union user_action_cookie
*cookie
)
5291 pid
= dpif_port_get_pid(ofproto
->backer
->dpif
,
5292 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
5294 return odp_put_userspace_action(pid
, cookie
, odp_actions
);
5298 compose_sflow_cookie(const struct ofproto_dpif
*ofproto
,
5299 ovs_be16 vlan_tci
, uint32_t odp_port
,
5300 unsigned int n_outputs
, union user_action_cookie
*cookie
)
5304 cookie
->type
= USER_ACTION_COOKIE_SFLOW
;
5305 cookie
->sflow
.vlan_tci
= vlan_tci
;
5307 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5308 * port information") for the interpretation of cookie->output. */
5309 switch (n_outputs
) {
5311 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5312 cookie
->sflow
.output
= 0x40000000 | 256;
5316 ifindex
= dpif_sflow_odp_port_to_ifindex(ofproto
->sflow
, odp_port
);
5318 cookie
->sflow
.output
= ifindex
;
5323 /* 0x80000000 means "multiple output ports. */
5324 cookie
->sflow
.output
= 0x80000000 | n_outputs
;
5329 /* Compose SAMPLE action for sFlow. */
5331 compose_sflow_action(const struct ofproto_dpif
*ofproto
,
5332 struct ofpbuf
*odp_actions
,
5333 const struct flow
*flow
,
5336 uint32_t probability
;
5337 union user_action_cookie cookie
;
5338 size_t sample_offset
, actions_offset
;
5341 if (!ofproto
->sflow
|| flow
->in_port
== OFPP_NONE
) {
5345 sample_offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
5347 /* Number of packets out of UINT_MAX to sample. */
5348 probability
= dpif_sflow_get_probability(ofproto
->sflow
);
5349 nl_msg_put_u32(odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
, probability
);
5351 actions_offset
= nl_msg_start_nested(odp_actions
, OVS_SAMPLE_ATTR_ACTIONS
);
5352 compose_sflow_cookie(ofproto
, htons(0), odp_port
,
5353 odp_port
== OVSP_NONE
? 0 : 1, &cookie
);
5354 cookie_offset
= put_userspace_action(ofproto
, odp_actions
, flow
, &cookie
);
5356 nl_msg_end_nested(odp_actions
, actions_offset
);
5357 nl_msg_end_nested(odp_actions
, sample_offset
);
5358 return cookie_offset
;
5361 /* SAMPLE action must be first action in any given list of actions.
5362 * At this point we do not have all information required to build it. So try to
5363 * build sample action as complete as possible. */
5365 add_sflow_action(struct action_xlate_ctx
*ctx
)
5367 ctx
->user_cookie_offset
= compose_sflow_action(ctx
->ofproto
,
5369 &ctx
->flow
, OVSP_NONE
);
5370 ctx
->sflow_odp_port
= 0;
5371 ctx
->sflow_n_outputs
= 0;
5374 /* Fix SAMPLE action according to data collected while composing ODP actions.
5375 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5376 * USERSPACE action's user-cookie which is required for sflow. */
5378 fix_sflow_action(struct action_xlate_ctx
*ctx
)
5380 const struct flow
*base
= &ctx
->base_flow
;
5381 union user_action_cookie
*cookie
;
5383 if (!ctx
->user_cookie_offset
) {
5387 cookie
= ofpbuf_at(ctx
->odp_actions
, ctx
->user_cookie_offset
,
5389 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
5391 compose_sflow_cookie(ctx
->ofproto
, base
->vlan_tci
,
5392 ctx
->sflow_odp_port
, ctx
->sflow_n_outputs
, cookie
);
5396 compose_output_action__(struct action_xlate_ctx
*ctx
, uint16_t ofp_port
,
5399 const struct ofport_dpif
*ofport
= get_ofp_port(ctx
->ofproto
, ofp_port
);
5400 uint32_t odp_port
= ofp_port_to_odp_port(ctx
->ofproto
, ofp_port
);
5401 ovs_be16 flow_vlan_tci
= ctx
->flow
.vlan_tci
;
5402 uint8_t flow_nw_tos
= ctx
->flow
.nw_tos
;
5403 struct priority_to_dscp
*pdscp
;
5407 xlate_report(ctx
, "Nonexistent output port");
5409 } else if (ofport
->up
.pp
.config
& OFPUTIL_PC_NO_FWD
) {
5410 xlate_report(ctx
, "OFPPC_NO_FWD set, skipping output");
5412 } else if (check_stp
&& !stp_forward_in_state(ofport
->stp_state
)) {
5413 xlate_report(ctx
, "STP not in forwarding state, skipping output");
5417 pdscp
= get_priority(ofport
, ctx
->flow
.skb_priority
);
5419 ctx
->flow
.nw_tos
&= ~IP_DSCP_MASK
;
5420 ctx
->flow
.nw_tos
|= pdscp
->dscp
;
5423 out_port
= vsp_realdev_to_vlandev(ctx
->ofproto
, odp_port
,
5424 ctx
->flow
.vlan_tci
);
5425 if (out_port
!= odp_port
) {
5426 ctx
->flow
.vlan_tci
= htons(0);
5428 commit_odp_actions(&ctx
->flow
, &ctx
->base_flow
, ctx
->odp_actions
);
5429 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_OUTPUT
, out_port
);
5431 ctx
->sflow_odp_port
= odp_port
;
5432 ctx
->sflow_n_outputs
++;
5433 ctx
->nf_output_iface
= ofp_port
;
5434 ctx
->flow
.vlan_tci
= flow_vlan_tci
;
5435 ctx
->flow
.nw_tos
= flow_nw_tos
;
5439 compose_output_action(struct action_xlate_ctx
*ctx
, uint16_t ofp_port
)
5441 compose_output_action__(ctx
, ofp_port
, true);
5445 xlate_table_action(struct action_xlate_ctx
*ctx
,
5446 uint16_t in_port
, uint8_t table_id
, bool may_packet_in
)
5448 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
5449 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
5450 struct rule_dpif
*rule
;
5451 uint16_t old_in_port
;
5452 uint8_t old_table_id
;
5454 old_table_id
= ctx
->table_id
;
5455 ctx
->table_id
= table_id
;
5457 /* Look up a flow with 'in_port' as the input port. */
5458 old_in_port
= ctx
->flow
.in_port
;
5459 ctx
->flow
.in_port
= in_port
;
5460 rule
= rule_dpif_lookup__(ofproto
, &ctx
->flow
, table_id
);
5463 if (table_id
> 0 && table_id
< N_TABLES
) {
5464 struct table_dpif
*table
= &ofproto
->tables
[table_id
];
5465 if (table
->other_table
) {
5466 ctx
->tags
|= (rule
&& rule
->tag
5468 : rule_calculate_tag(&ctx
->flow
,
5469 &table
->other_table
->mask
,
5474 /* Restore the original input port. Otherwise OFPP_NORMAL and
5475 * OFPP_IN_PORT will have surprising behavior. */
5476 ctx
->flow
.in_port
= old_in_port
;
5478 if (ctx
->resubmit_hook
) {
5479 ctx
->resubmit_hook(ctx
, rule
);
5482 if (rule
== NULL
&& may_packet_in
) {
5484 * check if table configuration flags
5485 * OFPTC_TABLE_MISS_CONTROLLER, default.
5486 * OFPTC_TABLE_MISS_CONTINUE,
5487 * OFPTC_TABLE_MISS_DROP
5488 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5490 rule
= rule_dpif_miss_rule(ofproto
, &ctx
->flow
);
5494 struct rule_dpif
*old_rule
= ctx
->rule
;
5496 if (ctx
->resubmit_stats
) {
5497 rule_credit_stats(rule
, ctx
->resubmit_stats
);
5502 do_xlate_actions(rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, ctx
);
5503 ctx
->rule
= old_rule
;
5507 ctx
->table_id
= old_table_id
;
5509 static struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
5511 VLOG_ERR_RL(&recurse_rl
, "resubmit actions recursed over %d times",
5512 MAX_RESUBMIT_RECURSION
);
5513 ctx
->max_resubmit_trigger
= true;
5518 xlate_ofpact_resubmit(struct action_xlate_ctx
*ctx
,
5519 const struct ofpact_resubmit
*resubmit
)
5524 in_port
= resubmit
->in_port
;
5525 if (in_port
== OFPP_IN_PORT
) {
5526 in_port
= ctx
->flow
.in_port
;
5529 table_id
= resubmit
->table_id
;
5530 if (table_id
== 255) {
5531 table_id
= ctx
->table_id
;
5534 xlate_table_action(ctx
, in_port
, table_id
, false);
5538 flood_packets(struct action_xlate_ctx
*ctx
, bool all
)
5540 struct ofport_dpif
*ofport
;
5542 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ctx
->ofproto
->up
.ports
) {
5543 uint16_t ofp_port
= ofport
->up
.ofp_port
;
5545 if (ofp_port
== ctx
->flow
.in_port
) {
5550 compose_output_action__(ctx
, ofp_port
, false);
5551 } else if (!(ofport
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
)) {
5552 compose_output_action(ctx
, ofp_port
);
5556 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
5560 execute_controller_action(struct action_xlate_ctx
*ctx
, int len
,
5561 enum ofp_packet_in_reason reason
,
5562 uint16_t controller_id
)
5564 struct ofputil_packet_in pin
;
5565 struct ofpbuf
*packet
;
5567 ctx
->slow
|= SLOW_CONTROLLER
;
5572 packet
= ofpbuf_clone(ctx
->packet
);
5574 if (packet
->l2
&& packet
->l3
) {
5575 struct eth_header
*eh
;
5577 eth_pop_vlan(packet
);
5580 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5581 * LLC frame. Calculating the Ethernet type of these frames is more
5582 * trouble than seems appropriate for a simple assertion. */
5583 ovs_assert(ntohs(eh
->eth_type
) < ETH_TYPE_MIN
5584 || eh
->eth_type
== ctx
->flow
.dl_type
);
5586 memcpy(eh
->eth_src
, ctx
->flow
.dl_src
, sizeof eh
->eth_src
);
5587 memcpy(eh
->eth_dst
, ctx
->flow
.dl_dst
, sizeof eh
->eth_dst
);
5589 if (ctx
->flow
.vlan_tci
& htons(VLAN_CFI
)) {
5590 eth_push_vlan(packet
, ctx
->flow
.vlan_tci
);
5594 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
5595 packet_set_ipv4(packet
, ctx
->flow
.nw_src
, ctx
->flow
.nw_dst
,
5596 ctx
->flow
.nw_tos
, ctx
->flow
.nw_ttl
);
5600 if (ctx
->flow
.nw_proto
== IPPROTO_TCP
) {
5601 packet_set_tcp_port(packet
, ctx
->flow
.tp_src
,
5603 } else if (ctx
->flow
.nw_proto
== IPPROTO_UDP
) {
5604 packet_set_udp_port(packet
, ctx
->flow
.tp_src
,
5611 pin
.packet
= packet
->data
;
5612 pin
.packet_len
= packet
->size
;
5613 pin
.reason
= reason
;
5614 pin
.controller_id
= controller_id
;
5615 pin
.table_id
= ctx
->table_id
;
5616 pin
.cookie
= ctx
->rule
? ctx
->rule
->up
.flow_cookie
: 0;
5619 flow_get_metadata(&ctx
->flow
, &pin
.fmd
);
5621 connmgr_send_packet_in(ctx
->ofproto
->up
.connmgr
, &pin
);
5622 ofpbuf_delete(packet
);
5626 compose_dec_ttl(struct action_xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
5628 if (ctx
->flow
.dl_type
!= htons(ETH_TYPE_IP
) &&
5629 ctx
->flow
.dl_type
!= htons(ETH_TYPE_IPV6
)) {
5633 if (ctx
->flow
.nw_ttl
> 1) {
5639 for (i
= 0; i
< ids
->n_controllers
; i
++) {
5640 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5644 /* Stop processing for current table. */
5650 xlate_output_action(struct action_xlate_ctx
*ctx
,
5651 uint16_t port
, uint16_t max_len
, bool may_packet_in
)
5653 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
5655 ctx
->nf_output_iface
= NF_OUT_DROP
;
5659 compose_output_action(ctx
, ctx
->flow
.in_port
);
5662 xlate_table_action(ctx
, ctx
->flow
.in_port
, 0, may_packet_in
);
5668 flood_packets(ctx
, false);
5671 flood_packets(ctx
, true);
5673 case OFPP_CONTROLLER
:
5674 execute_controller_action(ctx
, max_len
, OFPR_ACTION
, 0);
5680 if (port
!= ctx
->flow
.in_port
) {
5681 compose_output_action(ctx
, port
);
5683 xlate_report(ctx
, "skipping output to input port");
5688 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
5689 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
5690 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5691 ctx
->nf_output_iface
= prev_nf_output_iface
;
5692 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
5693 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5694 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5699 xlate_output_reg_action(struct action_xlate_ctx
*ctx
,
5700 const struct ofpact_output_reg
*or)
5702 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->flow
);
5703 if (port
<= UINT16_MAX
) {
5704 xlate_output_action(ctx
, port
, or->max_len
, false);
5709 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
5710 const struct ofpact_enqueue
*enqueue
)
5712 uint16_t ofp_port
= enqueue
->port
;
5713 uint32_t queue_id
= enqueue
->queue
;
5714 uint32_t flow_priority
, priority
;
5717 /* Translate queue to priority. */
5718 error
= dpif_queue_to_priority(ctx
->ofproto
->backer
->dpif
,
5719 queue_id
, &priority
);
5721 /* Fall back to ordinary output action. */
5722 xlate_output_action(ctx
, enqueue
->port
, 0, false);
5726 /* Check output port. */
5727 if (ofp_port
== OFPP_IN_PORT
) {
5728 ofp_port
= ctx
->flow
.in_port
;
5729 } else if (ofp_port
== ctx
->flow
.in_port
) {
5733 /* Add datapath actions. */
5734 flow_priority
= ctx
->flow
.skb_priority
;
5735 ctx
->flow
.skb_priority
= priority
;
5736 compose_output_action(ctx
, ofp_port
);
5737 ctx
->flow
.skb_priority
= flow_priority
;
5739 /* Update NetFlow output port. */
5740 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5741 ctx
->nf_output_iface
= ofp_port
;
5742 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5743 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5748 xlate_set_queue_action(struct action_xlate_ctx
*ctx
, uint32_t queue_id
)
5750 uint32_t skb_priority
;
5752 if (!dpif_queue_to_priority(ctx
->ofproto
->backer
->dpif
,
5753 queue_id
, &skb_priority
)) {
5754 ctx
->flow
.skb_priority
= skb_priority
;
5756 /* Couldn't translate queue to a priority. Nothing to do. A warning
5757 * has already been logged. */
5761 struct xlate_reg_state
{
5767 xlate_autopath(struct action_xlate_ctx
*ctx
,
5768 const struct ofpact_autopath
*ap
)
5770 uint16_t ofp_port
= ap
->port
;
5771 struct ofport_dpif
*port
= get_ofp_port(ctx
->ofproto
, ofp_port
);
5773 if (!port
|| !port
->bundle
) {
5774 ofp_port
= OFPP_NONE
;
5775 } else if (port
->bundle
->bond
) {
5776 /* Autopath does not support VLAN hashing. */
5777 struct ofport_dpif
*slave
= bond_choose_output_slave(
5778 port
->bundle
->bond
, &ctx
->flow
, 0, &ctx
->tags
);
5780 ofp_port
= slave
->up
.ofp_port
;
5783 nxm_reg_load(&ap
->dst
, ofp_port
, &ctx
->flow
);
5787 slave_enabled_cb(uint16_t ofp_port
, void *ofproto_
)
5789 struct ofproto_dpif
*ofproto
= ofproto_
;
5790 struct ofport_dpif
*port
;
5800 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
5803 port
= get_ofp_port(ofproto
, ofp_port
);
5804 return port
? port
->may_enable
: false;
5809 xlate_bundle_action(struct action_xlate_ctx
*ctx
,
5810 const struct ofpact_bundle
*bundle
)
5814 port
= bundle_execute(bundle
, &ctx
->flow
, slave_enabled_cb
, ctx
->ofproto
);
5815 if (bundle
->dst
.field
) {
5816 nxm_reg_load(&bundle
->dst
, port
, &ctx
->flow
);
5818 xlate_output_action(ctx
, port
, 0, false);
5823 xlate_learn_action(struct action_xlate_ctx
*ctx
,
5824 const struct ofpact_learn
*learn
)
5826 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 1);
5827 struct ofputil_flow_mod fm
;
5828 uint64_t ofpacts_stub
[1024 / 8];
5829 struct ofpbuf ofpacts
;
5832 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
5833 learn_execute(learn
, &ctx
->flow
, &fm
, &ofpacts
);
5835 error
= ofproto_flow_mod(&ctx
->ofproto
->up
, &fm
);
5836 if (error
&& !VLOG_DROP_WARN(&rl
)) {
5837 VLOG_WARN("learning action failed to modify flow table (%s)",
5838 ofperr_get_name(error
));
5841 ofpbuf_uninit(&ofpacts
);
5844 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5845 * means "infinite". */
5847 reduce_timeout(uint16_t max
, uint16_t *timeout
)
5849 if (max
&& (!*timeout
|| *timeout
> max
)) {
5855 xlate_fin_timeout(struct action_xlate_ctx
*ctx
,
5856 const struct ofpact_fin_timeout
*oft
)
5858 if (ctx
->tcp_flags
& (TCP_FIN
| TCP_RST
) && ctx
->rule
) {
5859 struct rule_dpif
*rule
= ctx
->rule
;
5861 reduce_timeout(oft
->fin_idle_timeout
, &rule
->up
.idle_timeout
);
5862 reduce_timeout(oft
->fin_hard_timeout
, &rule
->up
.hard_timeout
);
5867 may_receive(const struct ofport_dpif
*port
, struct action_xlate_ctx
*ctx
)
5869 if (port
->up
.pp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
5870 ? OFPUTIL_PC_NO_RECV_STP
5871 : OFPUTIL_PC_NO_RECV
)) {
5875 /* Only drop packets here if both forwarding and learning are
5876 * disabled. If just learning is enabled, we need to have
5877 * OFPP_NORMAL and the learning action have a look at the packet
5878 * before we can drop it. */
5879 if (!stp_forward_in_state(port
->stp_state
)
5880 && !stp_learn_in_state(port
->stp_state
)) {
5888 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
5889 struct action_xlate_ctx
*ctx
)
5891 const struct ofport_dpif
*port
;
5892 bool was_evictable
= true;
5893 const struct ofpact
*a
;
5895 port
= get_ofp_port(ctx
->ofproto
, ctx
->flow
.in_port
);
5896 if (port
&& !may_receive(port
, ctx
)) {
5897 /* Drop this flow. */
5902 /* Don't let the rule we're working on get evicted underneath us. */
5903 was_evictable
= ctx
->rule
->up
.evictable
;
5904 ctx
->rule
->up
.evictable
= false;
5906 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5907 struct ofpact_controller
*controller
;
5908 const struct ofpact_metadata
*metadata
;
5916 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
5917 ofpact_get_OUTPUT(a
)->max_len
, true);
5920 case OFPACT_CONTROLLER
:
5921 controller
= ofpact_get_CONTROLLER(a
);
5922 execute_controller_action(ctx
, controller
->max_len
,
5924 controller
->controller_id
);
5927 case OFPACT_ENQUEUE
:
5928 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
));
5931 case OFPACT_SET_VLAN_VID
:
5932 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
5933 ctx
->flow
.vlan_tci
|= (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
)
5937 case OFPACT_SET_VLAN_PCP
:
5938 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
5939 ctx
->flow
.vlan_tci
|= htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
5944 case OFPACT_STRIP_VLAN
:
5945 ctx
->flow
.vlan_tci
= htons(0);
5948 case OFPACT_PUSH_VLAN
:
5949 /* XXX 802.1AD(QinQ) */
5950 ctx
->flow
.vlan_tci
= htons(VLAN_CFI
);
5953 case OFPACT_SET_ETH_SRC
:
5954 memcpy(ctx
->flow
.dl_src
, ofpact_get_SET_ETH_SRC(a
)->mac
,
5958 case OFPACT_SET_ETH_DST
:
5959 memcpy(ctx
->flow
.dl_dst
, ofpact_get_SET_ETH_DST(a
)->mac
,
5963 case OFPACT_SET_IPV4_SRC
:
5964 ctx
->flow
.nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
5967 case OFPACT_SET_IPV4_DST
:
5968 ctx
->flow
.nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
5971 case OFPACT_SET_IPV4_DSCP
:
5972 /* OpenFlow 1.0 only supports IPv4. */
5973 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
5974 ctx
->flow
.nw_tos
&= ~IP_DSCP_MASK
;
5975 ctx
->flow
.nw_tos
|= ofpact_get_SET_IPV4_DSCP(a
)->dscp
;
5979 case OFPACT_SET_L4_SRC_PORT
:
5980 ctx
->flow
.tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
5983 case OFPACT_SET_L4_DST_PORT
:
5984 ctx
->flow
.tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
5987 case OFPACT_RESUBMIT
:
5988 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
));
5991 case OFPACT_SET_TUNNEL
:
5992 ctx
->flow
.tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
5995 case OFPACT_SET_QUEUE
:
5996 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
5999 case OFPACT_POP_QUEUE
:
6000 ctx
->flow
.skb_priority
= ctx
->orig_skb_priority
;
6003 case OFPACT_REG_MOVE
:
6004 nxm_execute_reg_move(ofpact_get_REG_MOVE(a
), &ctx
->flow
);
6007 case OFPACT_REG_LOAD
:
6008 nxm_execute_reg_load(ofpact_get_REG_LOAD(a
), &ctx
->flow
);
6011 case OFPACT_DEC_TTL
:
6012 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
6018 /* Nothing to do. */
6021 case OFPACT_MULTIPATH
:
6022 multipath_execute(ofpact_get_MULTIPATH(a
), &ctx
->flow
);
6025 case OFPACT_AUTOPATH
:
6026 xlate_autopath(ctx
, ofpact_get_AUTOPATH(a
));
6030 ctx
->ofproto
->has_bundle_action
= true;
6031 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
));
6034 case OFPACT_OUTPUT_REG
:
6035 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
));
6039 ctx
->has_learn
= true;
6040 if (ctx
->may_learn
) {
6041 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
6049 case OFPACT_FIN_TIMEOUT
:
6050 ctx
->has_fin_timeout
= true;
6051 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
6054 case OFPACT_CLEAR_ACTIONS
:
6056 * Nothing to do because writa-actions is not supported for now.
6057 * When writa-actions is supported, clear-actions also must
6058 * be supported at the same time.
6062 case OFPACT_WRITE_METADATA
:
6063 metadata
= ofpact_get_WRITE_METADATA(a
);
6064 ctx
->flow
.metadata
&= ~metadata
->mask
;
6065 ctx
->flow
.metadata
|= metadata
->metadata
& metadata
->mask
;
6068 case OFPACT_GOTO_TABLE
: {
6069 /* XXX remove recursion */
6070 /* It is assumed that goto-table is last action */
6071 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
6072 ovs_assert(ctx
->table_id
< ogt
->table_id
);
6073 xlate_table_action(ctx
, ctx
->flow
.in_port
, ogt
->table_id
, true);
6080 /* We've let OFPP_NORMAL and the learning action look at the packet,
6081 * so drop it now if forwarding is disabled. */
6082 if (port
&& !stp_forward_in_state(port
->stp_state
)) {
6083 ofpbuf_clear(ctx
->odp_actions
);
6084 add_sflow_action(ctx
);
6087 ctx
->rule
->up
.evictable
= was_evictable
;
6092 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
6093 struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
6094 ovs_be16 initial_tci
, struct rule_dpif
*rule
,
6095 uint8_t tcp_flags
, const struct ofpbuf
*packet
)
6097 ovs_be64 initial_tun_id
= flow
->tunnel
.tun_id
;
6099 /* Flow initialization rules:
6100 * - 'base_flow' must match the kernel's view of the packet at the
6101 * time that action processing starts. 'flow' represents any
6102 * transformations we wish to make through actions.
6103 * - By default 'base_flow' and 'flow' are the same since the input
6104 * packet matches the output before any actions are applied.
6105 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6106 * of the received packet as seen by the kernel. If we later output
6107 * to another device without any modifications this will cause us to
6108 * insert a new tag since the original one was stripped off by the
6110 * - Tunnel 'flow' is largely cleared when transitioning between
6111 * the input and output stages since it does not make sense to output
6112 * a packet with the exact headers that it was received with (i.e.
6113 * the destination IP is us). The one exception is the tun_id, which
6114 * is preserved to allow use in later resubmit lookups and loads into
6116 * - Tunnel 'base_flow' is completely cleared since that is what the
6117 * kernel does. If we wish to maintain the original values an action
6118 * needs to be generated. */
6120 ctx
->ofproto
= ofproto
;
6122 memset(&ctx
->flow
.tunnel
, 0, sizeof ctx
->flow
.tunnel
);
6123 ctx
->base_flow
= ctx
->flow
;
6124 ctx
->base_flow
.vlan_tci
= initial_tci
;
6125 ctx
->flow
.tunnel
.tun_id
= initial_tun_id
;
6127 ctx
->packet
= packet
;
6128 ctx
->may_learn
= packet
!= NULL
;
6129 ctx
->tcp_flags
= tcp_flags
;
6130 ctx
->resubmit_hook
= NULL
;
6131 ctx
->report_hook
= NULL
;
6132 ctx
->resubmit_stats
= NULL
;
6135 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6136 * into datapath actions in 'odp_actions', using 'ctx'. */
6138 xlate_actions(struct action_xlate_ctx
*ctx
,
6139 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
6140 struct ofpbuf
*odp_actions
)
6142 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6143 * that in the future we always keep a copy of the original flow for
6144 * tracing purposes. */
6145 static bool hit_resubmit_limit
;
6147 enum slow_path_reason special
;
6149 COVERAGE_INC(ofproto_dpif_xlate
);
6151 ofpbuf_clear(odp_actions
);
6152 ofpbuf_reserve(odp_actions
, NL_A_U32_SIZE
);
6154 ctx
->odp_actions
= odp_actions
;
6157 ctx
->has_learn
= false;
6158 ctx
->has_normal
= false;
6159 ctx
->has_fin_timeout
= false;
6160 ctx
->nf_output_iface
= NF_OUT_DROP
;
6163 ctx
->max_resubmit_trigger
= false;
6164 ctx
->orig_skb_priority
= ctx
->flow
.skb_priority
;
6168 if (ctx
->ofproto
->has_mirrors
|| hit_resubmit_limit
) {
6169 /* Do this conditionally because the copy is expensive enough that it
6170 * shows up in profiles.
6172 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6173 * believe that I wasn't using it without initializing it if I kept it
6174 * in a local variable. */
6175 ctx
->orig_flow
= ctx
->flow
;
6178 if (ctx
->flow
.nw_frag
& FLOW_NW_FRAG_ANY
) {
6179 switch (ctx
->ofproto
->up
.frag_handling
) {
6180 case OFPC_FRAG_NORMAL
:
6181 /* We must pretend that transport ports are unavailable. */
6182 ctx
->flow
.tp_src
= ctx
->base_flow
.tp_src
= htons(0);
6183 ctx
->flow
.tp_dst
= ctx
->base_flow
.tp_dst
= htons(0);
6186 case OFPC_FRAG_DROP
:
6189 case OFPC_FRAG_REASM
:
6192 case OFPC_FRAG_NX_MATCH
:
6193 /* Nothing to do. */
6196 case OFPC_INVALID_TTL_TO_CONTROLLER
:
6201 special
= process_special(ctx
->ofproto
, &ctx
->flow
, ctx
->packet
);
6203 ctx
->slow
|= special
;
6205 static struct vlog_rate_limit trace_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
6206 ovs_be16 initial_tci
= ctx
->base_flow
.vlan_tci
;
6207 uint32_t local_odp_port
;
6209 add_sflow_action(ctx
);
6210 do_xlate_actions(ofpacts
, ofpacts_len
, ctx
);
6212 if (ctx
->max_resubmit_trigger
&& !ctx
->resubmit_hook
) {
6213 if (!hit_resubmit_limit
) {
6214 /* We didn't record the original flow. Make sure we do from
6216 hit_resubmit_limit
= true;
6217 } else if (!VLOG_DROP_ERR(&trace_rl
)) {
6218 struct ds ds
= DS_EMPTY_INITIALIZER
;
6220 ofproto_trace(ctx
->ofproto
, &ctx
->orig_flow
, ctx
->packet
,
6222 VLOG_ERR("Trace triggered by excessive resubmit "
6223 "recursion:\n%s", ds_cstr(&ds
));
6228 local_odp_port
= ofp_port_to_odp_port(ctx
->ofproto
, OFPP_LOCAL
);
6229 if (!connmgr_may_set_up_flow(ctx
->ofproto
->up
.connmgr
, &ctx
->flow
,
6231 ctx
->odp_actions
->data
,
6232 ctx
->odp_actions
->size
)) {
6233 ctx
->slow
|= SLOW_IN_BAND
;
6235 && connmgr_msg_in_hook(ctx
->ofproto
->up
.connmgr
, &ctx
->flow
,
6237 compose_output_action(ctx
, OFPP_LOCAL
);
6240 if (ctx
->ofproto
->has_mirrors
) {
6241 add_mirror_actions(ctx
, &ctx
->orig_flow
);
6243 fix_sflow_action(ctx
);
6247 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6248 * into datapath actions, using 'ctx', and discards the datapath actions. */
6250 xlate_actions_for_side_effects(struct action_xlate_ctx
*ctx
,
6251 const struct ofpact
*ofpacts
,
6254 uint64_t odp_actions_stub
[1024 / 8];
6255 struct ofpbuf odp_actions
;
6257 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
6258 xlate_actions(ctx
, ofpacts
, ofpacts_len
, &odp_actions
);
6259 ofpbuf_uninit(&odp_actions
);
6263 xlate_report(struct action_xlate_ctx
*ctx
, const char *s
)
6265 if (ctx
->report_hook
) {
6266 ctx
->report_hook(ctx
, s
);
6270 /* OFPP_NORMAL implementation. */
6272 static struct ofport_dpif
*ofbundle_get_a_port(const struct ofbundle
*);
6274 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6275 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6276 * the bundle on which the packet was received, returns the VLAN to which the
6279 * Both 'vid' and the return value are in the range 0...4095. */
6281 input_vid_to_vlan(const struct ofbundle
*in_bundle
, uint16_t vid
)
6283 switch (in_bundle
->vlan_mode
) {
6284 case PORT_VLAN_ACCESS
:
6285 return in_bundle
->vlan
;
6288 case PORT_VLAN_TRUNK
:
6291 case PORT_VLAN_NATIVE_UNTAGGED
:
6292 case PORT_VLAN_NATIVE_TAGGED
:
6293 return vid
? vid
: in_bundle
->vlan
;
6300 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6301 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6304 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6305 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6308 input_vid_is_valid(uint16_t vid
, struct ofbundle
*in_bundle
, bool warn
)
6310 /* Allow any VID on the OFPP_NONE port. */
6311 if (in_bundle
== &ofpp_none_bundle
) {
6315 switch (in_bundle
->vlan_mode
) {
6316 case PORT_VLAN_ACCESS
:
6319 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6320 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" tagged "
6321 "packet received on port %s configured as VLAN "
6322 "%"PRIu16
" access port",
6323 in_bundle
->ofproto
->up
.name
, vid
,
6324 in_bundle
->name
, in_bundle
->vlan
);
6330 case PORT_VLAN_NATIVE_UNTAGGED
:
6331 case PORT_VLAN_NATIVE_TAGGED
:
6333 /* Port must always carry its native VLAN. */
6337 case PORT_VLAN_TRUNK
:
6338 if (!ofbundle_includes_vlan(in_bundle
, vid
)) {
6340 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6341 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" packet "
6342 "received on port %s not configured for trunking "
6344 in_bundle
->ofproto
->up
.name
, vid
,
6345 in_bundle
->name
, vid
);
6357 /* Given 'vlan', the VLAN that a packet belongs to, and
6358 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6359 * that should be included in the 802.1Q header. (If the return value is 0,
6360 * then the 802.1Q header should only be included in the packet if there is a
6363 * Both 'vlan' and the return value are in the range 0...4095. */
6365 output_vlan_to_vid(const struct ofbundle
*out_bundle
, uint16_t vlan
)
6367 switch (out_bundle
->vlan_mode
) {
6368 case PORT_VLAN_ACCESS
:
6371 case PORT_VLAN_TRUNK
:
6372 case PORT_VLAN_NATIVE_TAGGED
:
6375 case PORT_VLAN_NATIVE_UNTAGGED
:
6376 return vlan
== out_bundle
->vlan
? 0 : vlan
;
6384 output_normal(struct action_xlate_ctx
*ctx
, const struct ofbundle
*out_bundle
,
6387 struct ofport_dpif
*port
;
6389 ovs_be16 tci
, old_tci
;
6391 vid
= output_vlan_to_vid(out_bundle
, vlan
);
6392 if (!out_bundle
->bond
) {
6393 port
= ofbundle_get_a_port(out_bundle
);
6395 port
= bond_choose_output_slave(out_bundle
->bond
, &ctx
->flow
,
6398 /* No slaves enabled, so drop packet. */
6403 old_tci
= ctx
->flow
.vlan_tci
;
6405 if (tci
|| out_bundle
->use_priority_tags
) {
6406 tci
|= ctx
->flow
.vlan_tci
& htons(VLAN_PCP_MASK
);
6408 tci
|= htons(VLAN_CFI
);
6411 ctx
->flow
.vlan_tci
= tci
;
6413 compose_output_action(ctx
, port
->up
.ofp_port
);
6414 ctx
->flow
.vlan_tci
= old_tci
;
6418 mirror_mask_ffs(mirror_mask_t mask
)
6420 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask
));
6425 ofbundle_trunks_vlan(const struct ofbundle
*bundle
, uint16_t vlan
)
6427 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
6428 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
6432 ofbundle_includes_vlan(const struct ofbundle
*bundle
, uint16_t vlan
)
6434 return vlan
== bundle
->vlan
|| ofbundle_trunks_vlan(bundle
, vlan
);
6437 /* Returns an arbitrary interface within 'bundle'. */
6438 static struct ofport_dpif
*
6439 ofbundle_get_a_port(const struct ofbundle
*bundle
)
6441 return CONTAINER_OF(list_front(&bundle
->ports
),
6442 struct ofport_dpif
, bundle_node
);
6446 vlan_is_mirrored(const struct ofmirror
*m
, int vlan
)
6448 return !m
->vlans
|| bitmap_is_set(m
->vlans
, vlan
);
6452 add_mirror_actions(struct action_xlate_ctx
*ctx
, const struct flow
*orig_flow
)
6454 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
6455 mirror_mask_t mirrors
;
6456 struct ofbundle
*in_bundle
;
6459 const struct nlattr
*a
;
6462 in_bundle
= lookup_input_bundle(ctx
->ofproto
, orig_flow
->in_port
,
6463 ctx
->packet
!= NULL
, NULL
);
6467 mirrors
= in_bundle
->src_mirrors
;
6469 /* Drop frames on bundles reserved for mirroring. */
6470 if (in_bundle
->mirror_out
) {
6471 if (ctx
->packet
!= NULL
) {
6472 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6473 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
6474 "%s, which is reserved exclusively for mirroring",
6475 ctx
->ofproto
->up
.name
, in_bundle
->name
);
6481 vid
= vlan_tci_to_vid(orig_flow
->vlan_tci
);
6482 if (!input_vid_is_valid(vid
, in_bundle
, ctx
->packet
!= NULL
)) {
6485 vlan
= input_vid_to_vlan(in_bundle
, vid
);
6487 /* Look at the output ports to check for destination selections. */
6489 NL_ATTR_FOR_EACH (a
, left
, ctx
->odp_actions
->data
,
6490 ctx
->odp_actions
->size
) {
6491 enum ovs_action_attr type
= nl_attr_type(a
);
6492 struct ofport_dpif
*ofport
;
6494 if (type
!= OVS_ACTION_ATTR_OUTPUT
) {
6498 ofport
= get_odp_port(ofproto
, nl_attr_get_u32(a
));
6499 if (ofport
&& ofport
->bundle
) {
6500 mirrors
|= ofport
->bundle
->dst_mirrors
;
6508 /* Restore the original packet before adding the mirror actions. */
6509 ctx
->flow
= *orig_flow
;
6514 m
= ofproto
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
6516 if (!vlan_is_mirrored(m
, vlan
)) {
6517 mirrors
= zero_rightmost_1bit(mirrors
);
6521 mirrors
&= ~m
->dup_mirrors
;
6522 ctx
->mirrors
|= m
->dup_mirrors
;
6524 output_normal(ctx
, m
->out
, vlan
);
6525 } else if (vlan
!= m
->out_vlan
6526 && !eth_addr_is_reserved(orig_flow
->dl_dst
)) {
6527 struct ofbundle
*bundle
;
6529 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
6530 if (ofbundle_includes_vlan(bundle
, m
->out_vlan
)
6531 && !bundle
->mirror_out
) {
6532 output_normal(ctx
, bundle
, m
->out_vlan
);
6540 update_mirror_stats(struct ofproto_dpif
*ofproto
, mirror_mask_t mirrors
,
6541 uint64_t packets
, uint64_t bytes
)
6547 for (; mirrors
; mirrors
= zero_rightmost_1bit(mirrors
)) {
6550 m
= ofproto
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
6553 /* In normal circumstances 'm' will not be NULL. However,
6554 * if mirrors are reconfigured, we can temporarily get out
6555 * of sync in facet_revalidate(). We could "correct" the
6556 * mirror list before reaching here, but doing that would
6557 * not properly account the traffic stats we've currently
6558 * accumulated for previous mirror configuration. */
6562 m
->packet_count
+= packets
;
6563 m
->byte_count
+= bytes
;
6567 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6568 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6569 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6571 is_gratuitous_arp(const struct flow
*flow
)
6573 return (flow
->dl_type
== htons(ETH_TYPE_ARP
)
6574 && eth_addr_is_broadcast(flow
->dl_dst
)
6575 && (flow
->nw_proto
== ARP_OP_REPLY
6576 || (flow
->nw_proto
== ARP_OP_REQUEST
6577 && flow
->nw_src
== flow
->nw_dst
)));
6581 update_learning_table(struct ofproto_dpif
*ofproto
,
6582 const struct flow
*flow
, int vlan
,
6583 struct ofbundle
*in_bundle
)
6585 struct mac_entry
*mac
;
6587 /* Don't learn the OFPP_NONE port. */
6588 if (in_bundle
== &ofpp_none_bundle
) {
6592 if (!mac_learning_may_learn(ofproto
->ml
, flow
->dl_src
, vlan
)) {
6596 mac
= mac_learning_insert(ofproto
->ml
, flow
->dl_src
, vlan
);
6597 if (is_gratuitous_arp(flow
)) {
6598 /* We don't want to learn from gratuitous ARP packets that are
6599 * reflected back over bond slaves so we lock the learning table. */
6600 if (!in_bundle
->bond
) {
6601 mac_entry_set_grat_arp_lock(mac
);
6602 } else if (mac_entry_is_grat_arp_locked(mac
)) {
6607 if (mac_entry_is_new(mac
) || mac
->port
.p
!= in_bundle
) {
6608 /* The log messages here could actually be useful in debugging,
6609 * so keep the rate limit relatively high. */
6610 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
6611 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
6612 "on port %s in VLAN %d",
6613 ofproto
->up
.name
, ETH_ADDR_ARGS(flow
->dl_src
),
6614 in_bundle
->name
, vlan
);
6616 mac
->port
.p
= in_bundle
;
6617 tag_set_add(&ofproto
->revalidate_set
,
6618 mac_learning_changed(ofproto
->ml
, mac
));
6622 static struct ofbundle
*
6623 lookup_input_bundle(const struct ofproto_dpif
*ofproto
, uint16_t in_port
,
6624 bool warn
, struct ofport_dpif
**in_ofportp
)
6626 struct ofport_dpif
*ofport
;
6628 /* Find the port and bundle for the received packet. */
6629 ofport
= get_ofp_port(ofproto
, in_port
);
6631 *in_ofportp
= ofport
;
6633 if (ofport
&& ofport
->bundle
) {
6634 return ofport
->bundle
;
6637 /* Special-case OFPP_NONE, which a controller may use as the ingress
6638 * port for traffic that it is sourcing. */
6639 if (in_port
== OFPP_NONE
) {
6640 return &ofpp_none_bundle
;
6643 /* Odd. A few possible reasons here:
6645 * - We deleted a port but there are still a few packets queued up
6648 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
6649 * we don't know about.
6651 * - The ofproto client didn't configure the port as part of a bundle.
6652 * This is particularly likely to happen if a packet was received on the
6653 * port after it was created, but before the client had a chance to
6654 * configure its bundle.
6657 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6659 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
6660 "port %"PRIu16
, ofproto
->up
.name
, in_port
);
6665 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
6666 * dropped. Returns true if they may be forwarded, false if they should be
6669 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
6670 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
6672 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
6673 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
6674 * checked by input_vid_is_valid().
6676 * May also add tags to '*tags', although the current implementation only does
6677 * so in one special case.
6680 is_admissible(struct action_xlate_ctx
*ctx
, struct ofport_dpif
*in_port
,
6683 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
6684 struct flow
*flow
= &ctx
->flow
;
6685 struct ofbundle
*in_bundle
= in_port
->bundle
;
6687 /* Drop frames for reserved multicast addresses
6688 * only if forward_bpdu option is absent. */
6689 if (!ofproto
->up
.forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
6690 xlate_report(ctx
, "packet has reserved destination MAC, dropping");
6694 if (in_bundle
->bond
) {
6695 struct mac_entry
*mac
;
6697 switch (bond_check_admissibility(in_bundle
->bond
, in_port
,
6698 flow
->dl_dst
, &ctx
->tags
)) {
6703 xlate_report(ctx
, "bonding refused admissibility, dropping");
6706 case BV_DROP_IF_MOVED
:
6707 mac
= mac_learning_lookup(ofproto
->ml
, flow
->dl_src
, vlan
, NULL
);
6708 if (mac
&& mac
->port
.p
!= in_bundle
&&
6709 (!is_gratuitous_arp(flow
)
6710 || mac_entry_is_grat_arp_locked(mac
))) {
6711 xlate_report(ctx
, "SLB bond thinks this packet looped back, "
6723 xlate_normal(struct action_xlate_ctx
*ctx
)
6725 struct ofport_dpif
*in_port
;
6726 struct ofbundle
*in_bundle
;
6727 struct mac_entry
*mac
;
6731 ctx
->has_normal
= true;
6733 in_bundle
= lookup_input_bundle(ctx
->ofproto
, ctx
->flow
.in_port
,
6734 ctx
->packet
!= NULL
, &in_port
);
6736 xlate_report(ctx
, "no input bundle, dropping");
6740 /* Drop malformed frames. */
6741 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_VLAN
) &&
6742 !(ctx
->flow
.vlan_tci
& htons(VLAN_CFI
))) {
6743 if (ctx
->packet
!= NULL
) {
6744 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6745 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet with partial "
6746 "VLAN tag received on port %s",
6747 ctx
->ofproto
->up
.name
, in_bundle
->name
);
6749 xlate_report(ctx
, "partial VLAN tag, dropping");
6753 /* Drop frames on bundles reserved for mirroring. */
6754 if (in_bundle
->mirror_out
) {
6755 if (ctx
->packet
!= NULL
) {
6756 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
6757 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
6758 "%s, which is reserved exclusively for mirroring",
6759 ctx
->ofproto
->up
.name
, in_bundle
->name
);
6761 xlate_report(ctx
, "input port is mirror output port, dropping");
6766 vid
= vlan_tci_to_vid(ctx
->flow
.vlan_tci
);
6767 if (!input_vid_is_valid(vid
, in_bundle
, ctx
->packet
!= NULL
)) {
6768 xlate_report(ctx
, "disallowed VLAN VID for this input port, dropping");
6771 vlan
= input_vid_to_vlan(in_bundle
, vid
);
6773 /* Check other admissibility requirements. */
6774 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
6778 /* Learn source MAC. */
6779 if (ctx
->may_learn
) {
6780 update_learning_table(ctx
->ofproto
, &ctx
->flow
, vlan
, in_bundle
);
6783 /* Determine output bundle. */
6784 mac
= mac_learning_lookup(ctx
->ofproto
->ml
, ctx
->flow
.dl_dst
, vlan
,
6787 if (mac
->port
.p
!= in_bundle
) {
6788 xlate_report(ctx
, "forwarding to learned port");
6789 output_normal(ctx
, mac
->port
.p
, vlan
);
6791 xlate_report(ctx
, "learned port is input port, dropping");
6794 struct ofbundle
*bundle
;
6796 xlate_report(ctx
, "no learned MAC for destination, flooding");
6797 HMAP_FOR_EACH (bundle
, hmap_node
, &ctx
->ofproto
->bundles
) {
6798 if (bundle
!= in_bundle
6799 && ofbundle_includes_vlan(bundle
, vlan
)
6800 && bundle
->floodable
6801 && !bundle
->mirror_out
) {
6802 output_normal(ctx
, bundle
, vlan
);
6805 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
6809 /* Optimized flow revalidation.
6811 * It's a difficult problem, in general, to tell which facets need to have
6812 * their actions recalculated whenever the OpenFlow flow table changes. We
6813 * don't try to solve that general problem: for most kinds of OpenFlow flow
6814 * table changes, we recalculate the actions for every facet. This is
6815 * relatively expensive, but it's good enough if the OpenFlow flow table
6816 * doesn't change very often.
6818 * However, we can expect one particular kind of OpenFlow flow table change to
6819 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
6820 * of CPU on revalidating every facet whenever MAC learning modifies the flow
6821 * table, we add a special case that applies to flow tables in which every rule
6822 * has the same form (that is, the same wildcards), except that the table is
6823 * also allowed to have a single "catch-all" flow that matches all packets. We
6824 * optimize this case by tagging all of the facets that resubmit into the table
6825 * and invalidating the same tag whenever a flow changes in that table. The
6826 * end result is that we revalidate just the facets that need it (and sometimes
6827 * a few more, but not all of the facets or even all of the facets that
6828 * resubmit to the table modified by MAC learning). */
6830 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
6831 * into an OpenFlow table with the given 'basis'. */
6833 rule_calculate_tag(const struct flow
*flow
, const struct minimask
*mask
,
6836 if (minimask_is_catchall(mask
)) {
6839 uint32_t hash
= flow_hash_in_minimask(flow
, mask
, secret
);
6840 return tag_create_deterministic(hash
);
6844 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
6845 * taggability of that table.
6847 * This function must be called after *each* change to a flow table. If you
6848 * skip calling it on some changes then the pointer comparisons at the end can
6849 * be invalid if you get unlucky. For example, if a flow removal causes a
6850 * cls_table to be destroyed and then a flow insertion causes a cls_table with
6851 * different wildcards to be created with the same address, then this function
6852 * will incorrectly skip revalidation. */
6854 table_update_taggable(struct ofproto_dpif
*ofproto
, uint8_t table_id
)
6856 struct table_dpif
*table
= &ofproto
->tables
[table_id
];
6857 const struct oftable
*oftable
= &ofproto
->up
.tables
[table_id
];
6858 struct cls_table
*catchall
, *other
;
6859 struct cls_table
*t
;
6861 catchall
= other
= NULL
;
6863 switch (hmap_count(&oftable
->cls
.tables
)) {
6865 /* We could tag this OpenFlow table but it would make the logic a
6866 * little harder and it's a corner case that doesn't seem worth it
6872 HMAP_FOR_EACH (t
, hmap_node
, &oftable
->cls
.tables
) {
6873 if (cls_table_is_catchall(t
)) {
6875 } else if (!other
) {
6878 /* Indicate that we can't tag this by setting both tables to
6879 * NULL. (We know that 'catchall' is already NULL.) */
6886 /* Can't tag this table. */
6890 if (table
->catchall_table
!= catchall
|| table
->other_table
!= other
) {
6891 table
->catchall_table
= catchall
;
6892 table
->other_table
= other
;
6893 ofproto
->need_revalidate
= REV_FLOW_TABLE
;
6897 /* Given 'rule' that has changed in some way (either it is a rule being
6898 * inserted, a rule being deleted, or a rule whose actions are being
6899 * modified), marks facets for revalidation to ensure that packets will be
6900 * forwarded correctly according to the new state of the flow table.
6902 * This function must be called after *each* change to a flow table. See
6903 * the comment on table_update_taggable() for more information. */
6905 rule_invalidate(const struct rule_dpif
*rule
)
6907 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
6909 table_update_taggable(ofproto
, rule
->up
.table_id
);
6911 if (!ofproto
->need_revalidate
) {
6912 struct table_dpif
*table
= &ofproto
->tables
[rule
->up
.table_id
];
6914 if (table
->other_table
&& rule
->tag
) {
6915 tag_set_add(&ofproto
->revalidate_set
, rule
->tag
);
6917 ofproto
->need_revalidate
= REV_FLOW_TABLE
;
6923 set_frag_handling(struct ofproto
*ofproto_
,
6924 enum ofp_config_flags frag_handling
)
6926 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6928 if (frag_handling
!= OFPC_FRAG_REASM
) {
6929 ofproto
->need_revalidate
= REV_RECONFIGURE
;
6937 packet_out(struct ofproto
*ofproto_
, struct ofpbuf
*packet
,
6938 const struct flow
*flow
,
6939 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
6941 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6942 struct odputil_keybuf keybuf
;
6943 struct dpif_flow_stats stats
;
6947 struct action_xlate_ctx ctx
;
6948 uint64_t odp_actions_stub
[1024 / 8];
6949 struct ofpbuf odp_actions
;
6951 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
6952 odp_flow_key_from_flow(&key
, flow
,
6953 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
6955 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
6957 action_xlate_ctx_init(&ctx
, ofproto
, flow
, flow
->vlan_tci
, NULL
,
6958 packet_get_tcp_flags(packet
, flow
), packet
);
6959 ctx
.resubmit_stats
= &stats
;
6961 ofpbuf_use_stub(&odp_actions
,
6962 odp_actions_stub
, sizeof odp_actions_stub
);
6963 xlate_actions(&ctx
, ofpacts
, ofpacts_len
, &odp_actions
);
6964 dpif_execute(ofproto
->backer
->dpif
, key
.data
, key
.size
,
6965 odp_actions
.data
, odp_actions
.size
, packet
);
6966 ofpbuf_uninit(&odp_actions
);
6974 set_netflow(struct ofproto
*ofproto_
,
6975 const struct netflow_options
*netflow_options
)
6977 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6979 if (netflow_options
) {
6980 if (!ofproto
->netflow
) {
6981 ofproto
->netflow
= netflow_create();
6983 return netflow_set_options(ofproto
->netflow
, netflow_options
);
6985 netflow_destroy(ofproto
->netflow
);
6986 ofproto
->netflow
= NULL
;
6992 get_netflow_ids(const struct ofproto
*ofproto_
,
6993 uint8_t *engine_type
, uint8_t *engine_id
)
6995 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
6997 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
7001 send_active_timeout(struct ofproto_dpif
*ofproto
, struct facet
*facet
)
7003 if (!facet_is_controller_flow(facet
) &&
7004 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
7005 struct subfacet
*subfacet
;
7006 struct ofexpired expired
;
7008 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
7009 if (subfacet
->path
== SF_FAST_PATH
) {
7010 struct dpif_flow_stats stats
;
7012 subfacet_reinstall(subfacet
, &stats
);
7013 subfacet_update_stats(subfacet
, &stats
);
7017 expired
.flow
= facet
->flow
;
7018 expired
.packet_count
= facet
->packet_count
;
7019 expired
.byte_count
= facet
->byte_count
;
7020 expired
.used
= facet
->used
;
7021 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
7026 send_netflow_active_timeouts(struct ofproto_dpif
*ofproto
)
7028 struct facet
*facet
;
7030 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
7031 send_active_timeout(ofproto
, facet
);
7035 static struct ofproto_dpif
*
7036 ofproto_dpif_lookup(const char *name
)
7038 struct ofproto_dpif
*ofproto
;
7040 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_node
,
7041 hash_string(name
, 0), &all_ofproto_dpifs
) {
7042 if (!strcmp(ofproto
->up
.name
, name
)) {
7050 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
7051 const char *argv
[], void *aux OVS_UNUSED
)
7053 struct ofproto_dpif
*ofproto
;
7056 ofproto
= ofproto_dpif_lookup(argv
[1]);
7058 unixctl_command_reply_error(conn
, "no such bridge");
7061 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
7063 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
7064 mac_learning_flush(ofproto
->ml
, &ofproto
->revalidate_set
);
7068 unixctl_command_reply(conn
, "table successfully flushed");
7072 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
7073 const char *argv
[], void *aux OVS_UNUSED
)
7075 struct ds ds
= DS_EMPTY_INITIALIZER
;
7076 const struct ofproto_dpif
*ofproto
;
7077 const struct mac_entry
*e
;
7079 ofproto
= ofproto_dpif_lookup(argv
[1]);
7081 unixctl_command_reply_error(conn
, "no such bridge");
7085 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
7086 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
7087 struct ofbundle
*bundle
= e
->port
.p
;
7088 ds_put_format(&ds
, "%5d %4d "ETH_ADDR_FMT
" %3d\n",
7089 ofbundle_get_a_port(bundle
)->odp_port
,
7090 e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
7091 mac_entry_age(ofproto
->ml
, e
));
7093 unixctl_command_reply(conn
, ds_cstr(&ds
));
7098 struct action_xlate_ctx ctx
;
7104 trace_format_rule(struct ds
*result
, uint8_t table_id
, int level
,
7105 const struct rule_dpif
*rule
)
7107 ds_put_char_multiple(result
, '\t', level
);
7109 ds_put_cstr(result
, "No match\n");
7113 ds_put_format(result
, "Rule: table=%"PRIu8
" cookie=%#"PRIx64
" ",
7114 table_id
, ntohll(rule
->up
.flow_cookie
));
7115 cls_rule_format(&rule
->up
.cr
, result
);
7116 ds_put_char(result
, '\n');
7118 ds_put_char_multiple(result
, '\t', level
);
7119 ds_put_cstr(result
, "OpenFlow ");
7120 ofpacts_format(rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, result
);
7121 ds_put_char(result
, '\n');
7125 trace_format_flow(struct ds
*result
, int level
, const char *title
,
7126 struct trace_ctx
*trace
)
7128 ds_put_char_multiple(result
, '\t', level
);
7129 ds_put_format(result
, "%s: ", title
);
7130 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
7131 ds_put_cstr(result
, "unchanged");
7133 flow_format(result
, &trace
->ctx
.flow
);
7134 trace
->flow
= trace
->ctx
.flow
;
7136 ds_put_char(result
, '\n');
7140 trace_format_regs(struct ds
*result
, int level
, const char *title
,
7141 struct trace_ctx
*trace
)
7145 ds_put_char_multiple(result
, '\t', level
);
7146 ds_put_format(result
, "%s:", title
);
7147 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
7148 ds_put_format(result
, " reg%zu=0x%"PRIx32
, i
, trace
->flow
.regs
[i
]);
7150 ds_put_char(result
, '\n');
7154 trace_format_odp(struct ds
*result
, int level
, const char *title
,
7155 struct trace_ctx
*trace
)
7157 struct ofpbuf
*odp_actions
= trace
->ctx
.odp_actions
;
7159 ds_put_char_multiple(result
, '\t', level
);
7160 ds_put_format(result
, "%s: ", title
);
7161 format_odp_actions(result
, odp_actions
->data
, odp_actions
->size
);
7162 ds_put_char(result
, '\n');
7166 trace_resubmit(struct action_xlate_ctx
*ctx
, struct rule_dpif
*rule
)
7168 struct trace_ctx
*trace
= CONTAINER_OF(ctx
, struct trace_ctx
, ctx
);
7169 struct ds
*result
= trace
->result
;
7171 ds_put_char(result
, '\n');
7172 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
7173 trace_format_regs(result
, ctx
->recurse
+ 1, "Resubmitted regs", trace
);
7174 trace_format_odp(result
, ctx
->recurse
+ 1, "Resubmitted odp", trace
);
7175 trace_format_rule(result
, ctx
->table_id
, ctx
->recurse
+ 1, rule
);
7179 trace_report(struct action_xlate_ctx
*ctx
, const char *s
)
7181 struct trace_ctx
*trace
= CONTAINER_OF(ctx
, struct trace_ctx
, ctx
);
7182 struct ds
*result
= trace
->result
;
7184 ds_put_char_multiple(result
, '\t', ctx
->recurse
);
7185 ds_put_cstr(result
, s
);
7186 ds_put_char(result
, '\n');
7190 ofproto_unixctl_trace(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
7191 void *aux OVS_UNUSED
)
7193 const char *dpname
= argv
[1];
7194 struct ofproto_dpif
*ofproto
;
7195 struct ofpbuf odp_key
;
7196 struct ofpbuf
*packet
;
7197 ovs_be16 initial_tci
;
7203 ofpbuf_init(&odp_key
, 0);
7206 ofproto
= ofproto_dpif_lookup(dpname
);
7208 unixctl_command_reply_error(conn
, "Unknown ofproto (use ofproto/list "
7212 if (argc
== 3 || (argc
== 4 && !strcmp(argv
[3], "-generate"))) {
7213 /* ofproto/trace dpname flow [-generate] */
7214 const char *flow_s
= argv
[2];
7215 const char *generate_s
= argv
[3];
7217 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7218 * flow. We guess which type it is based on whether 'flow_s' contains
7219 * an '(', since a datapath flow always contains '(') but an
7220 * OpenFlow-like flow should not (in fact it's allowed but I believe
7221 * that's not documented anywhere).
7223 * An alternative would be to try to parse 'flow_s' both ways, but then
7224 * it would be tricky giving a sensible error message. After all, do
7225 * you just say "syntax error" or do you present both error messages?
7226 * Both choices seem lousy. */
7227 if (strchr(flow_s
, '(')) {
7230 /* Convert string to datapath key. */
7231 ofpbuf_init(&odp_key
, 0);
7232 error
= odp_flow_key_from_string(flow_s
, NULL
, &odp_key
);
7234 unixctl_command_reply_error(conn
, "Bad flow syntax");
7238 /* XXX: Since we allow the user to specify an ofproto, it's
7239 * possible they will specify a different ofproto than the one the
7240 * port actually belongs too. Ideally we should simply remove the
7241 * ability to specify the ofproto. */
7242 if (ofproto_receive(ofproto
->backer
, NULL
, odp_key
.data
,
7243 odp_key
.size
, &flow
, NULL
, NULL
, NULL
,
7245 unixctl_command_reply_error(conn
, "Invalid flow");
7251 error_s
= parse_ofp_exact_flow(&flow
, argv
[2]);
7253 unixctl_command_reply_error(conn
, error_s
);
7258 initial_tci
= flow
.vlan_tci
;
7261 /* Generate a packet, if requested. */
7263 packet
= ofpbuf_new(0);
7264 flow_compose(packet
, &flow
);
7266 } else if (argc
== 7) {
7267 /* ofproto/trace dpname priority tun_id in_port mark packet */
7268 const char *priority_s
= argv
[2];
7269 const char *tun_id_s
= argv
[3];
7270 const char *in_port_s
= argv
[4];
7271 const char *mark_s
= argv
[5];
7272 const char *packet_s
= argv
[6];
7273 uint32_t in_port
= atoi(in_port_s
);
7274 ovs_be64 tun_id
= htonll(strtoull(tun_id_s
, NULL
, 0));
7275 uint32_t priority
= atoi(priority_s
);
7276 uint32_t mark
= atoi(mark_s
);
7279 msg
= eth_from_hex(packet_s
, &packet
);
7281 unixctl_command_reply_error(conn
, msg
);
7285 ds_put_cstr(&result
, "Packet: ");
7286 s
= ofp_packet_to_string(packet
->data
, packet
->size
);
7287 ds_put_cstr(&result
, s
);
7290 flow_extract(packet
, priority
, mark
, NULL
, in_port
, &flow
);
7291 flow
.tunnel
.tun_id
= tun_id
;
7292 initial_tci
= flow
.vlan_tci
;
7294 unixctl_command_reply_error(conn
, "Bad command syntax");
7298 ofproto_trace(ofproto
, &flow
, packet
, initial_tci
, &result
);
7299 unixctl_command_reply(conn
, ds_cstr(&result
));
7302 ds_destroy(&result
);
7303 ofpbuf_delete(packet
);
7304 ofpbuf_uninit(&odp_key
);
7308 ofproto_trace(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
7309 const struct ofpbuf
*packet
, ovs_be16 initial_tci
,
7312 struct rule_dpif
*rule
;
7314 ds_put_cstr(ds
, "Flow: ");
7315 flow_format(ds
, flow
);
7316 ds_put_char(ds
, '\n');
7318 rule
= rule_dpif_lookup(ofproto
, flow
);
7320 trace_format_rule(ds
, 0, 0, rule
);
7321 if (rule
== ofproto
->miss_rule
) {
7322 ds_put_cstr(ds
, "\nNo match, flow generates \"packet in\"s.\n");
7323 } else if (rule
== ofproto
->no_packet_in_rule
) {
7324 ds_put_cstr(ds
, "\nNo match, packets dropped because "
7325 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7329 uint64_t odp_actions_stub
[1024 / 8];
7330 struct ofpbuf odp_actions
;
7332 struct trace_ctx trace
;
7335 tcp_flags
= packet
? packet_get_tcp_flags(packet
, flow
) : 0;
7338 ofpbuf_use_stub(&odp_actions
,
7339 odp_actions_stub
, sizeof odp_actions_stub
);
7340 action_xlate_ctx_init(&trace
.ctx
, ofproto
, flow
, initial_tci
,
7341 rule
, tcp_flags
, packet
);
7342 trace
.ctx
.resubmit_hook
= trace_resubmit
;
7343 trace
.ctx
.report_hook
= trace_report
;
7344 xlate_actions(&trace
.ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
7347 ds_put_char(ds
, '\n');
7348 trace_format_flow(ds
, 0, "Final flow", &trace
);
7349 ds_put_cstr(ds
, "Datapath actions: ");
7350 format_odp_actions(ds
, odp_actions
.data
, odp_actions
.size
);
7351 ofpbuf_uninit(&odp_actions
);
7353 if (trace
.ctx
.slow
) {
7354 enum slow_path_reason slow
;
7356 ds_put_cstr(ds
, "\nThis flow is handled by the userspace "
7357 "slow path because it:");
7358 for (slow
= trace
.ctx
.slow
; slow
; ) {
7359 enum slow_path_reason bit
= rightmost_1bit(slow
);
7363 ds_put_cstr(ds
, "\n\t- Consists of CFM packets.");
7366 ds_put_cstr(ds
, "\n\t- Consists of LACP packets.");
7369 ds_put_cstr(ds
, "\n\t- Consists of STP packets.");
7372 ds_put_cstr(ds
, "\n\t- Needs in-band special case "
7375 ds_put_cstr(ds
, "\n\t (The datapath actions are "
7376 "incomplete--for complete actions, "
7377 "please supply a packet.)");
7380 case SLOW_CONTROLLER
:
7381 ds_put_cstr(ds
, "\n\t- Sends \"packet-in\" messages "
7382 "to the OpenFlow controller.");
7385 ds_put_cstr(ds
, "\n\t- Needs more specific matching "
7386 "than the datapath supports.");
7393 if (slow
& ~SLOW_MATCH
) {
7394 ds_put_cstr(ds
, "\nThe datapath actions above do not reflect "
7395 "the special slow-path processing.");
7402 ofproto_dpif_clog(struct unixctl_conn
*conn OVS_UNUSED
, int argc OVS_UNUSED
,
7403 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
7406 unixctl_command_reply(conn
, NULL
);
7410 ofproto_dpif_unclog(struct unixctl_conn
*conn OVS_UNUSED
, int argc OVS_UNUSED
,
7411 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
7414 unixctl_command_reply(conn
, NULL
);
7417 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7418 * 'reply' describing the results. */
7420 ofproto_dpif_self_check__(struct ofproto_dpif
*ofproto
, struct ds
*reply
)
7422 struct facet
*facet
;
7426 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
7427 if (!facet_check_consistency(facet
)) {
7432 ofproto
->need_revalidate
= REV_INCONSISTENCY
;
7436 ds_put_format(reply
, "%s: self-check failed (%d errors)\n",
7437 ofproto
->up
.name
, errors
);
7439 ds_put_format(reply
, "%s: self-check passed\n", ofproto
->up
.name
);
7444 ofproto_dpif_self_check(struct unixctl_conn
*conn
,
7445 int argc
, const char *argv
[], void *aux OVS_UNUSED
)
7447 struct ds reply
= DS_EMPTY_INITIALIZER
;
7448 struct ofproto_dpif
*ofproto
;
7451 ofproto
= ofproto_dpif_lookup(argv
[1]);
7453 unixctl_command_reply_error(conn
, "Unknown ofproto (use "
7454 "ofproto/list for help)");
7457 ofproto_dpif_self_check__(ofproto
, &reply
);
7459 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
7460 ofproto_dpif_self_check__(ofproto
, &reply
);
7464 unixctl_command_reply(conn
, ds_cstr(&reply
));
7468 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7469 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7470 * to destroy 'ofproto_shash' and free the returned value. */
7471 static const struct shash_node
**
7472 get_ofprotos(struct shash
*ofproto_shash
)
7474 const struct ofproto_dpif
*ofproto
;
7476 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
7477 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
7478 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
7481 return shash_sort(ofproto_shash
);
7485 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
7486 const char *argv
[] OVS_UNUSED
,
7487 void *aux OVS_UNUSED
)
7489 struct ds ds
= DS_EMPTY_INITIALIZER
;
7490 struct shash ofproto_shash
;
7491 const struct shash_node
**sorted_ofprotos
;
7494 shash_init(&ofproto_shash
);
7495 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
7496 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
7497 const struct shash_node
*node
= sorted_ofprotos
[i
];
7498 ds_put_format(&ds
, "%s\n", node
->name
);
7501 shash_destroy(&ofproto_shash
);
7502 free(sorted_ofprotos
);
7504 unixctl_command_reply(conn
, ds_cstr(&ds
));
7509 show_dp_format(const struct ofproto_dpif
*ofproto
, struct ds
*ds
)
7511 struct dpif_dp_stats s
;
7512 const struct shash_node
**ports
;
7515 dpif_get_dp_stats(ofproto
->backer
->dpif
, &s
);
7517 ds_put_format(ds
, "%s (%s):\n", ofproto
->up
.name
,
7518 dpif_name(ofproto
->backer
->dpif
));
7519 /* xxx It would be better to show bridge-specific stats instead
7520 * xxx of dp ones. */
7522 "\tlookups: hit:%"PRIu64
" missed:%"PRIu64
" lost:%"PRIu64
"\n",
7523 s
.n_hit
, s
.n_missed
, s
.n_lost
);
7524 ds_put_format(ds
, "\tflows: %zu\n",
7525 hmap_count(&ofproto
->subfacets
));
7527 ports
= shash_sort(&ofproto
->up
.port_by_name
);
7528 for (i
= 0; i
< shash_count(&ofproto
->up
.port_by_name
); i
++) {
7529 const struct shash_node
*node
= ports
[i
];
7530 struct ofport
*ofport
= node
->data
;
7531 const char *name
= netdev_get_name(ofport
->netdev
);
7532 const char *type
= netdev_get_type(ofport
->netdev
);
7534 ds_put_format(ds
, "\t%s %u/%u:", name
, ofport
->ofp_port
,
7535 ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
));
7536 if (strcmp(type
, "system")) {
7537 struct netdev
*netdev
;
7540 ds_put_format(ds
, " (%s", type
);
7542 error
= netdev_open(name
, type
, &netdev
);
7547 error
= netdev_get_config(netdev
, &config
);
7549 const struct smap_node
**nodes
;
7552 nodes
= smap_sort(&config
);
7553 for (i
= 0; i
< smap_count(&config
); i
++) {
7554 const struct smap_node
*node
= nodes
[i
];
7555 ds_put_format(ds
, "%c %s=%s", i
? ',' : ':',
7556 node
->key
, node
->value
);
7560 smap_destroy(&config
);
7562 netdev_close(netdev
);
7564 ds_put_char(ds
, ')');
7566 ds_put_char(ds
, '\n');
7572 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc
,
7573 const char *argv
[], void *aux OVS_UNUSED
)
7575 struct ds ds
= DS_EMPTY_INITIALIZER
;
7576 const struct ofproto_dpif
*ofproto
;
7580 for (i
= 1; i
< argc
; i
++) {
7581 ofproto
= ofproto_dpif_lookup(argv
[i
]);
7583 ds_put_format(&ds
, "Unknown bridge %s (use dpif/dump-dps "
7584 "for help)", argv
[i
]);
7585 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
7588 show_dp_format(ofproto
, &ds
);
7591 struct shash ofproto_shash
;
7592 const struct shash_node
**sorted_ofprotos
;
7595 shash_init(&ofproto_shash
);
7596 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
7597 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
7598 const struct shash_node
*node
= sorted_ofprotos
[i
];
7599 show_dp_format(node
->data
, &ds
);
7602 shash_destroy(&ofproto_shash
);
7603 free(sorted_ofprotos
);
7606 unixctl_command_reply(conn
, ds_cstr(&ds
));
7611 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
7612 int argc OVS_UNUSED
, const char *argv
[],
7613 void *aux OVS_UNUSED
)
7615 struct ds ds
= DS_EMPTY_INITIALIZER
;
7616 const struct ofproto_dpif
*ofproto
;
7617 struct subfacet
*subfacet
;
7619 ofproto
= ofproto_dpif_lookup(argv
[1]);
7621 unixctl_command_reply_error(conn
, "no such bridge");
7625 HMAP_FOR_EACH (subfacet
, hmap_node
, &ofproto
->subfacets
) {
7626 struct odputil_keybuf keybuf
;
7629 subfacet_get_key(subfacet
, &keybuf
, &key
);
7630 odp_flow_key_format(key
.data
, key
.size
, &ds
);
7632 ds_put_format(&ds
, ", packets:%"PRIu64
", bytes:%"PRIu64
", used:",
7633 subfacet
->dp_packet_count
, subfacet
->dp_byte_count
);
7634 if (subfacet
->used
) {
7635 ds_put_format(&ds
, "%.3fs",
7636 (time_msec() - subfacet
->used
) / 1000.0);
7638 ds_put_format(&ds
, "never");
7640 if (subfacet
->facet
->tcp_flags
) {
7641 ds_put_cstr(&ds
, ", flags:");
7642 packet_format_tcp_flags(&ds
, subfacet
->facet
->tcp_flags
);
7645 ds_put_cstr(&ds
, ", actions:");
7646 format_odp_actions(&ds
, subfacet
->actions
, subfacet
->actions_len
);
7647 ds_put_char(&ds
, '\n');
7650 unixctl_command_reply(conn
, ds_cstr(&ds
));
7655 ofproto_unixctl_dpif_del_flows(struct unixctl_conn
*conn
,
7656 int argc OVS_UNUSED
, const char *argv
[],
7657 void *aux OVS_UNUSED
)
7659 struct ds ds
= DS_EMPTY_INITIALIZER
;
7660 struct ofproto_dpif
*ofproto
;
7662 ofproto
= ofproto_dpif_lookup(argv
[1]);
7664 unixctl_command_reply_error(conn
, "no such bridge");
7668 flush(&ofproto
->up
);
7670 unixctl_command_reply(conn
, ds_cstr(&ds
));
7675 ofproto_dpif_unixctl_init(void)
7677 static bool registered
;
7683 unixctl_command_register(
7685 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
7686 2, 6, ofproto_unixctl_trace
, NULL
);
7687 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
7688 ofproto_unixctl_fdb_flush
, NULL
);
7689 unixctl_command_register("fdb/show", "bridge", 1, 1,
7690 ofproto_unixctl_fdb_show
, NULL
);
7691 unixctl_command_register("ofproto/clog", "", 0, 0,
7692 ofproto_dpif_clog
, NULL
);
7693 unixctl_command_register("ofproto/unclog", "", 0, 0,
7694 ofproto_dpif_unclog
, NULL
);
7695 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
7696 ofproto_dpif_self_check
, NULL
);
7697 unixctl_command_register("dpif/dump-dps", "", 0, 0,
7698 ofproto_unixctl_dpif_dump_dps
, NULL
);
7699 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX
,
7700 ofproto_unixctl_dpif_show
, NULL
);
7701 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
7702 ofproto_unixctl_dpif_dump_flows
, NULL
);
7703 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
7704 ofproto_unixctl_dpif_del_flows
, NULL
);
7707 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7709 * This is deprecated. It is only for compatibility with broken device drivers
7710 * in old versions of Linux that do not properly support VLANs when VLAN
7711 * devices are not used. When broken device drivers are no longer in
7712 * widespread use, we will delete these interfaces. */
7715 set_realdev(struct ofport
*ofport_
, uint16_t realdev_ofp_port
, int vid
)
7717 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
7718 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
7720 if (realdev_ofp_port
== ofport
->realdev_ofp_port
7721 && vid
== ofport
->vlandev_vid
) {
7725 ofproto
->need_revalidate
= REV_RECONFIGURE
;
7727 if (ofport
->realdev_ofp_port
) {
7730 if (realdev_ofp_port
&& ofport
->bundle
) {
7731 /* vlandevs are enslaved to their realdevs, so they are not allowed to
7732 * themselves be part of a bundle. */
7733 bundle_set(ofport
->up
.ofproto
, ofport
->bundle
, NULL
);
7736 ofport
->realdev_ofp_port
= realdev_ofp_port
;
7737 ofport
->vlandev_vid
= vid
;
7739 if (realdev_ofp_port
) {
7740 vsp_add(ofport
, realdev_ofp_port
, vid
);
7747 hash_realdev_vid(uint16_t realdev_ofp_port
, int vid
)
7749 return hash_2words(realdev_ofp_port
, vid
);
7752 /* Returns the ODP port number of the Linux VLAN device that corresponds to
7753 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
7754 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
7755 * it would return the port number of eth0.9.
7757 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
7758 * function just returns its 'realdev_odp_port' argument. */
7760 vsp_realdev_to_vlandev(const struct ofproto_dpif
*ofproto
,
7761 uint32_t realdev_odp_port
, ovs_be16 vlan_tci
)
7763 if (!hmap_is_empty(&ofproto
->realdev_vid_map
)) {
7764 uint16_t realdev_ofp_port
;
7765 int vid
= vlan_tci_to_vid(vlan_tci
);
7766 const struct vlan_splinter
*vsp
;
7768 realdev_ofp_port
= odp_port_to_ofp_port(ofproto
, realdev_odp_port
);
7769 HMAP_FOR_EACH_WITH_HASH (vsp
, realdev_vid_node
,
7770 hash_realdev_vid(realdev_ofp_port
, vid
),
7771 &ofproto
->realdev_vid_map
) {
7772 if (vsp
->realdev_ofp_port
== realdev_ofp_port
7773 && vsp
->vid
== vid
) {
7774 return ofp_port_to_odp_port(ofproto
, vsp
->vlandev_ofp_port
);
7778 return realdev_odp_port
;
7781 static struct vlan_splinter
*
7782 vlandev_find(const struct ofproto_dpif
*ofproto
, uint16_t vlandev_ofp_port
)
7784 struct vlan_splinter
*vsp
;
7786 HMAP_FOR_EACH_WITH_HASH (vsp
, vlandev_node
, hash_int(vlandev_ofp_port
, 0),
7787 &ofproto
->vlandev_map
) {
7788 if (vsp
->vlandev_ofp_port
== vlandev_ofp_port
) {
7796 /* Returns the OpenFlow port number of the "real" device underlying the Linux
7797 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
7798 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
7799 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
7800 * eth0 and store 9 in '*vid'.
7802 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
7803 * VLAN device. Unless VLAN splinters are enabled, this is what this function
7806 vsp_vlandev_to_realdev(const struct ofproto_dpif
*ofproto
,
7807 uint16_t vlandev_ofp_port
, int *vid
)
7809 if (!hmap_is_empty(&ofproto
->vlandev_map
)) {
7810 const struct vlan_splinter
*vsp
;
7812 vsp
= vlandev_find(ofproto
, vlandev_ofp_port
);
7817 return vsp
->realdev_ofp_port
;
7823 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
7824 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
7825 * 'flow->in_port' to the "real" device backing the VLAN device, sets
7826 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
7827 * always the case unless VLAN splinters are enabled), returns false without
7828 * making any changes. */
7830 vsp_adjust_flow(const struct ofproto_dpif
*ofproto
, struct flow
*flow
)
7835 realdev
= vsp_vlandev_to_realdev(ofproto
, flow
->in_port
, &vid
);
7840 /* Cause the flow to be processed as if it came in on the real device with
7841 * the VLAN device's VLAN ID. */
7842 flow
->in_port
= realdev
;
7843 flow
->vlan_tci
= htons((vid
& VLAN_VID_MASK
) | VLAN_CFI
);
7848 vsp_remove(struct ofport_dpif
*port
)
7850 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
7851 struct vlan_splinter
*vsp
;
7853 vsp
= vlandev_find(ofproto
, port
->up
.ofp_port
);
7855 hmap_remove(&ofproto
->vlandev_map
, &vsp
->vlandev_node
);
7856 hmap_remove(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
);
7859 port
->realdev_ofp_port
= 0;
7861 VLOG_ERR("missing vlan device record");
7866 vsp_add(struct ofport_dpif
*port
, uint16_t realdev_ofp_port
, int vid
)
7868 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
7870 if (!vsp_vlandev_to_realdev(ofproto
, port
->up
.ofp_port
, NULL
)
7871 && (vsp_realdev_to_vlandev(ofproto
, realdev_ofp_port
, htons(vid
))
7872 == realdev_ofp_port
)) {
7873 struct vlan_splinter
*vsp
;
7875 vsp
= xmalloc(sizeof *vsp
);
7876 hmap_insert(&ofproto
->vlandev_map
, &vsp
->vlandev_node
,
7877 hash_int(port
->up
.ofp_port
, 0));
7878 hmap_insert(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
,
7879 hash_realdev_vid(realdev_ofp_port
, vid
));
7880 vsp
->realdev_ofp_port
= realdev_ofp_port
;
7881 vsp
->vlandev_ofp_port
= port
->up
.ofp_port
;
7884 port
->realdev_ofp_port
= realdev_ofp_port
;
7886 VLOG_ERR("duplicate vlan device record");
7891 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, uint16_t ofp_port
)
7893 const struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
7894 return ofport
? ofport
->odp_port
: OVSP_NONE
;
7897 static struct ofport_dpif
*
7898 odp_port_to_ofport(const struct dpif_backer
*backer
, uint32_t odp_port
)
7900 struct ofport_dpif
*port
;
7902 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
,
7903 hash_int(odp_port
, 0),
7904 &backer
->odp_to_ofport_map
) {
7905 if (port
->odp_port
== odp_port
) {
7914 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, uint32_t odp_port
)
7916 struct ofport_dpif
*port
;
7918 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
7919 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
7920 return port
->up
.ofp_port
;
7926 const struct ofproto_class ofproto_dpif_class
= {
7961 port_is_lacp_current
,
7962 NULL
, /* rule_choose_table */
7969 rule_modify_actions
,
7978 get_cfm_remote_mpids
,
7983 get_stp_port_status
,
7990 is_mirror_output_bundle
,
7991 forward_bpdu_changed
,
7992 set_mac_table_config
,