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"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif
);
64 COVERAGE_DEFINE(ofproto_dpif_expired
);
65 COVERAGE_DEFINE(ofproto_dpif_xlate
);
66 COVERAGE_DEFINE(facet_changed_rule
);
67 COVERAGE_DEFINE(facet_revalidate
);
68 COVERAGE_DEFINE(facet_unexpected
);
69 COVERAGE_DEFINE(facet_suppress
);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES
= 255 };
77 enum { TBL_INTERNAL
= N_TABLES
- 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES
>= 2 && N_TABLES
<= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count
; /* Number of packets received. */
102 uint64_t byte_count
; /* Number of bytes received. */
104 tag_type tag
; /* Caches rule_calculate_tag() result. */
106 struct list facets
; /* List of "struct facet"s. */
109 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
111 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
114 static struct rule_dpif
*rule_dpif_lookup(struct ofproto_dpif
*,
115 const struct flow
*);
116 static struct rule_dpif
*rule_dpif_lookup__(struct ofproto_dpif
*,
119 static struct rule_dpif
*rule_dpif_miss_rule(struct ofproto_dpif
*ofproto
,
120 const struct flow
*flow
);
122 static void rule_credit_stats(struct rule_dpif
*,
123 const struct dpif_flow_stats
*);
124 static void flow_push_stats(struct facet
*, const struct dpif_flow_stats
*);
125 static tag_type
rule_calculate_tag(const struct flow
*,
126 const struct minimask
*, uint32_t basis
);
127 static void rule_invalidate(const struct rule_dpif
*);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t
;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t
) * CHAR_BIT
>= MAX_MIRRORS
);
134 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
135 size_t idx
; /* In ofproto's "mirrors" array. */
136 void *aux
; /* Key supplied by ofproto's client. */
137 char *name
; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs
; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts
; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans
; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle
*out
; /* Output port or NULL. */
146 int out_vlan
; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors
; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count
; /* Number of packets sent. */
151 int64_t byte_count
; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror
*);
155 static void update_mirror_stats(struct ofproto_dpif
*ofproto
,
156 mirror_mask_t mirrors
,
157 uint64_t packets
, uint64_t bytes
);
160 struct hmap_node hmap_node
; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
162 void *aux
; /* Key supplied by ofproto's client. */
163 char *name
; /* Identifier for log messages. */
166 struct list ports
; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode
; /* VLAN mode */
168 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp
*lacp
; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond
*bond
; /* Nonnull iff more than one port. */
173 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable
; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors
; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors
; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out
; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport
*);
185 static void bundle_update(struct ofbundle
*);
186 static void bundle_destroy(struct ofbundle
*);
187 static void bundle_del_port(struct ofport_dpif
*);
188 static void bundle_run(struct ofbundle
*);
189 static void bundle_wait(struct ofbundle
*);
190 static struct ofbundle
*lookup_input_bundle(const struct ofproto_dpif
*,
191 uint16_t in_port
, bool warn
,
192 struct ofport_dpif
**in_ofportp
);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle
= {
201 .vlan_mode
= PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif
*ofproto
);
205 static void stp_wait(struct ofproto_dpif
*ofproto
);
206 static int set_stp_port(struct ofport
*,
207 const struct ofproto_port_stp_settings
*);
209 static bool ofbundle_includes_vlan(const struct ofbundle
*, uint16_t vlan
);
211 struct action_xlate_ctx
{
212 /* action_xlate_ctx_init() initializes these members. */
215 struct ofproto_dpif
*ofproto
;
217 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
218 * this flow when actions change header fields. */
221 /* stack for the push and pop actions.
222 * Each stack element is of the type "union mf_subvalue". */
224 union mf_subvalue init_stack
[1024 / sizeof(union mf_subvalue
)];
226 /* The packet corresponding to 'flow', or a null pointer if we are
227 * revalidating without a packet to refer to. */
228 const struct ofpbuf
*packet
;
230 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
231 * actions update the flow table?
233 * We want to update these tables if we are actually processing a packet,
234 * or if we are accounting for packets that the datapath has processed, but
235 * not if we are just revalidating. */
238 /* The rule that we are currently translating, or NULL. */
239 struct rule_dpif
*rule
;
241 /* Union of the set of TCP flags seen so far in this flow. (Used only by
242 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
246 /* If nonnull, flow translation calls this function just before executing a
247 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
248 * when the recursion depth is exceeded.
250 * 'rule' is the rule being submitted into. It will be null if the
251 * resubmit or OFPP_TABLE action didn't find a matching rule.
253 * This is normally null so the client has to set it manually after
254 * calling action_xlate_ctx_init(). */
255 void (*resubmit_hook
)(struct action_xlate_ctx
*, struct rule_dpif
*rule
);
257 /* If nonnull, flow translation calls this function to report some
258 * significant decision, e.g. to explain why OFPP_NORMAL translation
259 * dropped a packet. */
260 void (*report_hook
)(struct action_xlate_ctx
*, const char *s
);
262 /* If nonnull, flow translation credits the specified statistics to each
263 * rule reached through a resubmit or OFPP_TABLE action.
265 * This is normally null so the client has to set it manually after
266 * calling action_xlate_ctx_init(). */
267 const struct dpif_flow_stats
*resubmit_stats
;
269 /* xlate_actions() initializes and uses these members. The client might want
270 * to look at them after it returns. */
272 struct ofpbuf
*odp_actions
; /* Datapath actions. */
273 tag_type tags
; /* Tags associated with actions. */
274 enum slow_path_reason slow
; /* 0 if fast path may be used. */
275 bool has_learn
; /* Actions include NXAST_LEARN? */
276 bool has_normal
; /* Actions output to OFPP_NORMAL? */
277 bool has_fin_timeout
; /* Actions include NXAST_FIN_TIMEOUT? */
278 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
279 mirror_mask_t mirrors
; /* Bitmap of associated mirrors. */
281 /* xlate_actions() initializes and uses these members, but the client has no
282 * reason to look at them. */
284 int recurse
; /* Recursion level, via xlate_table_action. */
285 bool max_resubmit_trigger
; /* Recursed too deeply during translation. */
286 struct flow base_flow
; /* Flow at the last commit. */
287 uint32_t orig_skb_priority
; /* Priority when packet arrived. */
288 uint8_t table_id
; /* OpenFlow table ID where flow was found. */
289 uint32_t sflow_n_outputs
; /* Number of output ports. */
290 uint32_t sflow_odp_port
; /* Output port for composing sFlow action. */
291 uint16_t user_cookie_offset
;/* Used for user_action_cookie fixup. */
292 bool exit
; /* No further actions should be processed. */
295 /* Initial values of fields of the packet that may be changed during
296 * flow processing and needed later. */
297 struct initial_vals
{
298 /* This is the value of vlan_tci in the packet as actually received from
299 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
300 * was received via a VLAN splinter. In that case, this value is 0
301 * (because the packet as actually received from the dpif had no 802.1Q
302 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
305 * This member should be removed when the VLAN splinters feature is no
310 static void action_xlate_ctx_init(struct action_xlate_ctx
*,
311 struct ofproto_dpif
*, const struct flow
*,
312 const struct initial_vals
*initial_vals
,
314 uint8_t tcp_flags
, const struct ofpbuf
*);
315 static void xlate_actions(struct action_xlate_ctx
*,
316 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
317 struct ofpbuf
*odp_actions
);
318 static void xlate_actions_for_side_effects(struct action_xlate_ctx
*,
319 const struct ofpact
*ofpacts
,
321 static void xlate_table_action(struct action_xlate_ctx
*, uint16_t in_port
,
322 uint8_t table_id
, bool may_packet_in
);
324 static size_t put_userspace_action(const struct ofproto_dpif
*,
325 struct ofpbuf
*odp_actions
,
327 const union user_action_cookie
*,
330 static void compose_slow_path(const struct ofproto_dpif
*, const struct flow
*,
331 enum slow_path_reason
,
332 uint64_t *stub
, size_t stub_size
,
333 const struct nlattr
**actionsp
,
334 size_t *actions_lenp
);
336 static void xlate_report(struct action_xlate_ctx
*ctx
, const char *s
);
338 /* A subfacet (see "struct subfacet" below) has three possible installation
341 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
342 * case just after the subfacet is created, just before the subfacet is
343 * destroyed, or if the datapath returns an error when we try to install a
346 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
348 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
349 * ofproto_dpif is installed in the datapath.
352 SF_NOT_INSTALLED
, /* No datapath flow for this subfacet. */
353 SF_FAST_PATH
, /* Full actions are installed. */
354 SF_SLOW_PATH
, /* Send-to-userspace action is installed. */
357 static const char *subfacet_path_to_string(enum subfacet_path
);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node
; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node
; /* In struct facet's 'facets' list. */
366 struct facet
*facet
; /* Owning facet. */
368 enum odp_key_fitness key_fitness
;
372 long long int used
; /* Time last used; time created if not used. */
373 long long int created
; /* Time created. */
375 uint64_t dp_packet_count
; /* Last known packet count in the datapath. */
376 uint64_t dp_byte_count
; /* Last known byte count in the datapath. */
380 * These should be essentially identical for every subfacet in a facet, but
381 * may differ in trivial ways due to VLAN splinters. */
382 size_t actions_len
; /* Number of bytes in actions[]. */
383 struct nlattr
*actions
; /* Datapath actions. */
385 enum slow_path_reason slow
; /* 0 if fast path may be used. */
386 enum subfacet_path path
; /* Installed in datapath? */
388 /* Initial values of the packet that may be needed later. */
389 struct initial_vals initial_vals
;
391 /* Datapath port the packet arrived on. This is needed to remove
392 * flows for ports that are no longer part of the bridge. Since the
393 * flow definition only has the OpenFlow port number and the port is
394 * no longer part of the bridge, we can't determine the datapath port
395 * number needed to delete the flow from the datapath. */
396 uint32_t odp_in_port
;
399 #define SUBFACET_DESTROY_MAX_BATCH 50
401 static struct subfacet
*subfacet_create(struct facet
*, struct flow_miss
*miss
,
403 static struct subfacet
*subfacet_find(struct ofproto_dpif
*,
404 const struct nlattr
*key
, size_t key_len
,
406 static void subfacet_destroy(struct subfacet
*);
407 static void subfacet_destroy__(struct subfacet
*);
408 static void subfacet_destroy_batch(struct ofproto_dpif
*,
409 struct subfacet
**, int n
);
410 static void subfacet_reset_dp_stats(struct subfacet
*,
411 struct dpif_flow_stats
*);
412 static void subfacet_update_time(struct subfacet
*, long long int used
);
413 static void subfacet_update_stats(struct subfacet
*,
414 const struct dpif_flow_stats
*);
415 static void subfacet_make_actions(struct subfacet
*,
416 const struct ofpbuf
*packet
,
417 struct ofpbuf
*odp_actions
);
418 static int subfacet_install(struct subfacet
*,
419 const struct nlattr
*actions
, size_t actions_len
,
420 struct dpif_flow_stats
*, enum slow_path_reason
);
421 static void subfacet_uninstall(struct subfacet
*);
423 static enum subfacet_path
subfacet_want_path(enum slow_path_reason
);
425 /* An exact-match instantiation of an OpenFlow flow.
427 * A facet associates a "struct flow", which represents the Open vSwitch
428 * userspace idea of an exact-match flow, with one or more subfacets. Each
429 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
430 * the facet. When the kernel module (or other dpif implementation) and Open
431 * vSwitch userspace agree on the definition of a flow key, there is exactly
432 * one subfacet per facet. If the dpif implementation supports more-specific
433 * flow matching than userspace, however, a facet can have more than one
434 * subfacet, each of which corresponds to some distinction in flow that
435 * userspace simply doesn't understand.
437 * Flow expiration works in terms of subfacets, so a facet must have at least
438 * one subfacet or it will never expire, leaking memory. */
441 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
442 struct list list_node
; /* In owning rule's 'facets' list. */
443 struct rule_dpif
*rule
; /* Owning rule. */
446 struct list subfacets
;
447 long long int used
; /* Time last used; time created if not used. */
454 * - Do include packets and bytes sent "by hand", e.g. with
457 * - Do include packets and bytes that were obtained from the datapath
458 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
459 * DPIF_FP_ZERO_STATS).
461 * - Do not include packets or bytes that can be obtained from the
462 * datapath for any existing subfacet.
464 uint64_t packet_count
; /* Number of packets received. */
465 uint64_t byte_count
; /* Number of bytes received. */
467 /* Resubmit statistics. */
468 uint64_t prev_packet_count
; /* Number of packets from last stats push. */
469 uint64_t prev_byte_count
; /* Number of bytes from last stats push. */
470 long long int prev_used
; /* Used time from last stats push. */
473 uint64_t accounted_bytes
; /* Bytes processed by facet_account(). */
474 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
475 uint8_t tcp_flags
; /* TCP flags seen for this 'rule'. */
477 /* Properties of datapath actions.
479 * Every subfacet has its own actions because actions can differ slightly
480 * between splintered and non-splintered subfacets due to the VLAN tag
481 * being initially different (present vs. absent). All of them have these
482 * properties in common so we just store one copy of them here. */
483 bool has_learn
; /* Actions include NXAST_LEARN? */
484 bool has_normal
; /* Actions output to OFPP_NORMAL? */
485 bool has_fin_timeout
; /* Actions include NXAST_FIN_TIMEOUT? */
486 tag_type tags
; /* Tags that would require revalidation. */
487 mirror_mask_t mirrors
; /* Bitmap of dependent mirrors. */
489 /* Storage for a single subfacet, to reduce malloc() time and space
490 * overhead. (A facet always has at least one subfacet and in the common
491 * case has exactly one subfacet. However, 'one_subfacet' may not
492 * always be valid, since it could have been removed after newer
493 * subfacets were pushed onto the 'subfacets' list.) */
494 struct subfacet one_subfacet
;
496 long long int learn_rl
; /* Rate limiter for facet_learn(). */
499 static struct facet
*facet_create(struct rule_dpif
*,
500 const struct flow
*, uint32_t hash
);
501 static void facet_remove(struct facet
*);
502 static void facet_free(struct facet
*);
504 static struct facet
*facet_find(struct ofproto_dpif
*,
505 const struct flow
*, uint32_t hash
);
506 static struct facet
*facet_lookup_valid(struct ofproto_dpif
*,
507 const struct flow
*, uint32_t hash
);
508 static void facet_revalidate(struct facet
*);
509 static bool facet_check_consistency(struct facet
*);
511 static void facet_flush_stats(struct facet
*);
513 static void facet_update_time(struct facet
*, long long int used
);
514 static void facet_reset_counters(struct facet
*);
515 static void facet_push_stats(struct facet
*);
516 static void facet_learn(struct facet
*);
517 static void facet_account(struct facet
*);
518 static void push_all_stats(void);
520 static struct subfacet
*facet_get_subfacet(struct facet
*);
522 static bool facet_is_controller_flow(struct facet
*);
525 struct hmap_node odp_port_node
; /* In dpif_backer's "odp_to_ofport_map". */
529 struct ofbundle
*bundle
; /* Bundle that contains this port, if any. */
530 struct list bundle_node
; /* In struct ofbundle's "ports" list. */
531 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
532 struct bfd
*bfd
; /* BFD, if any. */
533 tag_type tag
; /* Tag associated with this port. */
534 bool may_enable
; /* May be enabled in bonds. */
535 long long int carrier_seq
; /* Carrier status changes. */
536 struct tnl_port
*tnl_port
; /* Tunnel handle, or null. */
539 struct stp_port
*stp_port
; /* Spanning Tree Protocol, if any. */
540 enum stp_state stp_state
; /* Always STP_DISABLED if STP not in use. */
541 long long int stp_state_entered
;
543 struct hmap priorities
; /* Map of attached 'priority_to_dscp's. */
545 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
547 * This is deprecated. It is only for compatibility with broken device
548 * drivers in old versions of Linux that do not properly support VLANs when
549 * VLAN devices are not used. When broken device drivers are no longer in
550 * widespread use, we will delete these interfaces. */
551 uint16_t realdev_ofp_port
;
555 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
556 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
557 * traffic egressing the 'ofport' with that priority should be marked with. */
558 struct priority_to_dscp
{
559 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'priorities' map. */
560 uint32_t priority
; /* Priority of this queue (see struct flow). */
562 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
565 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
567 * This is deprecated. It is only for compatibility with broken device drivers
568 * in old versions of Linux that do not properly support VLANs when VLAN
569 * devices are not used. When broken device drivers are no longer in
570 * widespread use, we will delete these interfaces. */
571 struct vlan_splinter
{
572 struct hmap_node realdev_vid_node
;
573 struct hmap_node vlandev_node
;
574 uint16_t realdev_ofp_port
;
575 uint16_t vlandev_ofp_port
;
579 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif
*,
580 uint32_t realdev
, ovs_be16 vlan_tci
);
581 static bool vsp_adjust_flow(const struct ofproto_dpif
*, struct flow
*);
582 static void vsp_remove(struct ofport_dpif
*);
583 static void vsp_add(struct ofport_dpif
*, uint16_t realdev_ofp_port
, int vid
);
585 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif
*,
587 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif
*,
590 static struct ofport_dpif
*
591 ofport_dpif_cast(const struct ofport
*ofport
)
593 return ofport
? CONTAINER_OF(ofport
, struct ofport_dpif
, up
) : NULL
;
596 static void port_run(struct ofport_dpif
*);
597 static void port_run_fast(struct ofport_dpif
*);
598 static void port_wait(struct ofport_dpif
*);
599 static int set_cfm(struct ofport
*, const struct cfm_settings
*);
600 static void ofport_clear_priorities(struct ofport_dpif
*);
601 static void run_fast_rl(void);
603 struct dpif_completion
{
604 struct list list_node
;
605 struct ofoperation
*op
;
608 /* Extra information about a classifier table.
609 * Currently used just for optimized flow revalidation. */
611 /* If either of these is nonnull, then this table has a form that allows
612 * flows to be tagged to avoid revalidating most flows for the most common
613 * kinds of flow table changes. */
614 struct cls_table
*catchall_table
; /* Table that wildcards all fields. */
615 struct cls_table
*other_table
; /* Table with any other wildcard set. */
616 uint32_t basis
; /* Keeps each table's tags separate. */
619 /* Reasons that we might need to revalidate every facet, and corresponding
622 * A value of 0 means that there is no need to revalidate.
624 * It would be nice to have some cleaner way to integrate with coverage
625 * counters, but with only a few reasons I guess this is good enough for
627 enum revalidate_reason
{
628 REV_RECONFIGURE
= 1, /* Switch configuration changed. */
629 REV_STP
, /* Spanning tree protocol port status change. */
630 REV_PORT_TOGGLED
, /* Port enabled or disabled by CFM, LACP, ...*/
631 REV_FLOW_TABLE
, /* Flow table changed. */
632 REV_INCONSISTENCY
/* Facet self-check failed. */
634 COVERAGE_DEFINE(rev_reconfigure
);
635 COVERAGE_DEFINE(rev_stp
);
636 COVERAGE_DEFINE(rev_port_toggled
);
637 COVERAGE_DEFINE(rev_flow_table
);
638 COVERAGE_DEFINE(rev_inconsistency
);
640 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
641 * These are datapath flows which have no associated ofproto, if they did we
642 * would use facets. */
644 struct hmap_node hmap_node
;
649 /* All datapaths of a given type share a single dpif backer instance. */
654 struct timer next_expiration
;
655 struct hmap odp_to_ofport_map
; /* ODP port to ofport mapping. */
657 struct simap tnl_backers
; /* Set of dpif ports backing tunnels. */
659 /* Facet revalidation flags applying to facets which use this backer. */
660 enum revalidate_reason need_revalidate
; /* Revalidate every facet. */
661 struct tag_set revalidate_set
; /* Revalidate only matching facets. */
663 struct hmap drop_keys
; /* Set of dropped odp keys. */
666 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
667 static struct shash all_dpif_backers
= SHASH_INITIALIZER(&all_dpif_backers
);
669 static void drop_key_clear(struct dpif_backer
*);
670 static struct ofport_dpif
*
671 odp_port_to_ofport(const struct dpif_backer
*, uint32_t odp_port
);
673 static void dpif_stats_update_hit_count(struct ofproto_dpif
*ofproto
,
675 struct avg_subfacet_rates
{
676 double add_rate
; /* Moving average of new flows created per minute. */
677 double del_rate
; /* Moving average of flows deleted per minute. */
679 static void show_dp_rates(struct ds
*ds
, const char *heading
,
680 const struct avg_subfacet_rates
*rates
);
681 static void exp_mavg(double *avg
, int base
, double new);
683 struct ofproto_dpif
{
684 struct hmap_node all_ofproto_dpifs_node
; /* In 'all_ofproto_dpifs'. */
686 struct dpif_backer
*backer
;
688 /* Special OpenFlow rules. */
689 struct rule_dpif
*miss_rule
; /* Sends flow table misses to controller. */
690 struct rule_dpif
*no_packet_in_rule
; /* Drops flow table misses. */
696 struct netflow
*netflow
;
697 struct dpif_sflow
*sflow
;
698 struct dpif_ipfix
*ipfix
;
699 struct hmap bundles
; /* Contains "struct ofbundle"s. */
700 struct mac_learning
*ml
;
701 struct ofmirror
*mirrors
[MAX_MIRRORS
];
703 bool has_bonded_bundles
;
707 struct hmap subfacets
;
708 struct governor
*governor
;
709 long long int consistency_rl
;
712 struct table_dpif tables
[N_TABLES
];
714 /* Support for debugging async flow mods. */
715 struct list completions
;
717 bool has_bundle_action
; /* True when the first bundle action appears. */
718 struct netdev_stats stats
; /* To account packets generated and consumed in
723 long long int stp_last_tick
;
725 /* VLAN splinters. */
726 struct hmap realdev_vid_map
; /* (realdev,vid) -> vlandev. */
727 struct hmap vlandev_map
; /* vlandev -> (realdev,vid). */
730 struct sset ports
; /* Set of standard port names. */
731 struct sset ghost_ports
; /* Ports with no datapath port. */
732 struct sset port_poll_set
; /* Queued names for port_poll() reply. */
733 int port_poll_errno
; /* Last errno for port_poll() reply. */
735 /* Per ofproto's dpif stats. */
739 /* Subfacet statistics.
741 * These keep track of the total number of subfacets added and deleted and
742 * flow life span. They are useful for computing the flow rates stats
743 * exposed via "ovs-appctl dpif/show". The goal is to learn about
744 * traffic patterns in ways that we can use later to improve Open vSwitch
745 * performance in new situations. */
746 long long int created
; /* Time when it is created. */
747 unsigned int max_n_subfacet
; /* Maximum number of flows */
749 /* The average number of subfacets... */
750 struct avg_subfacet_rates hourly
; /* ...over the last hour. */
751 struct avg_subfacet_rates daily
; /* ...over the last day. */
752 long long int last_minute
; /* Last time 'hourly' was updated. */
754 /* Number of subfacets added or deleted since 'last_minute'. */
755 unsigned int subfacet_add_count
;
756 unsigned int subfacet_del_count
;
758 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
759 unsigned long long int total_subfacet_add_count
;
760 unsigned long long int total_subfacet_del_count
;
762 /* Sum of the number of milliseconds that each subfacet existed,
763 * over the subfacets that have been added and then later deleted. */
764 unsigned long long int total_subfacet_life_span
;
766 /* Incremented by the number of currently existing subfacets, each
767 * time we pull statistics from the kernel. */
768 unsigned long long int total_subfacet_count
;
770 /* Number of times we pull statistics from the kernel. */
771 unsigned long long int n_update_stats
;
773 static unsigned long long int avg_subfacet_life_span(
774 const struct ofproto_dpif
*);
775 static double avg_subfacet_count(const struct ofproto_dpif
*ofproto
);
776 static void update_moving_averages(struct ofproto_dpif
*ofproto
);
777 static void dpif_stats_update_hit_count(struct ofproto_dpif
*ofproto
,
779 static void update_max_subfacet_count(struct ofproto_dpif
*ofproto
);
781 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
782 * for debugging the asynchronous flow_mod implementation.) */
785 /* All existing ofproto_dpif instances, indexed by ->up.name. */
786 static struct hmap all_ofproto_dpifs
= HMAP_INITIALIZER(&all_ofproto_dpifs
);
788 static void ofproto_dpif_unixctl_init(void);
790 static struct ofproto_dpif
*
791 ofproto_dpif_cast(const struct ofproto
*ofproto
)
793 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
794 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
797 static struct ofport_dpif
*get_ofp_port(const struct ofproto_dpif
*,
799 static struct ofport_dpif
*get_odp_port(const struct ofproto_dpif
*,
801 static void ofproto_trace(struct ofproto_dpif
*, const struct flow
*,
802 const struct ofpbuf
*,
803 const struct initial_vals
*, struct ds
*);
805 /* Packet processing. */
806 static void update_learning_table(struct ofproto_dpif
*,
807 const struct flow
*, int vlan
,
810 #define FLOW_MISS_MAX_BATCH 50
811 static int handle_upcalls(struct dpif_backer
*, unsigned int max_batch
);
813 /* Flow expiration. */
814 static int expire(struct dpif_backer
*);
817 static void send_netflow_active_timeouts(struct ofproto_dpif
*);
820 static int send_packet(const struct ofport_dpif
*, struct ofpbuf
*packet
);
821 static size_t compose_sflow_action(const struct ofproto_dpif
*,
822 struct ofpbuf
*odp_actions
,
823 const struct flow
*, uint32_t odp_port
);
824 static void compose_ipfix_action(const struct ofproto_dpif
*,
825 struct ofpbuf
*odp_actions
,
826 const struct flow
*);
827 static void add_mirror_actions(struct action_xlate_ctx
*ctx
,
828 const struct flow
*flow
);
829 /* Global variables. */
830 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
832 /* Initial mappings of port to bridge mappings. */
833 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
835 /* Factory functions. */
838 init(const struct shash
*iface_hints
)
840 struct shash_node
*node
;
842 /* Make a local copy, since we don't own 'iface_hints' elements. */
843 SHASH_FOR_EACH(node
, iface_hints
) {
844 const struct iface_hint
*orig_hint
= node
->data
;
845 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
847 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
848 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
849 new_hint
->ofp_port
= orig_hint
->ofp_port
;
851 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
856 enumerate_types(struct sset
*types
)
858 dp_enumerate_types(types
);
862 enumerate_names(const char *type
, struct sset
*names
)
864 struct ofproto_dpif
*ofproto
;
867 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
868 if (strcmp(type
, ofproto
->up
.type
)) {
871 sset_add(names
, ofproto
->up
.name
);
878 del(const char *type
, const char *name
)
883 error
= dpif_open(name
, type
, &dpif
);
885 error
= dpif_delete(dpif
);
892 port_open_type(const char *datapath_type
, const char *port_type
)
894 return dpif_port_open_type(datapath_type
, port_type
);
897 /* Type functions. */
899 static struct ofproto_dpif
*
900 lookup_ofproto_dpif_by_port_name(const char *name
)
902 struct ofproto_dpif
*ofproto
;
904 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
905 if (sset_contains(&ofproto
->ports
, name
)) {
914 type_run(const char *type
)
916 static long long int push_timer
= LLONG_MIN
;
917 struct dpif_backer
*backer
;
921 backer
= shash_find_data(&all_dpif_backers
, type
);
923 /* This is not necessarily a problem, since backers are only
924 * created on demand. */
928 dpif_run(backer
->dpif
);
930 /* The most natural place to push facet statistics is when they're pulled
931 * from the datapath. However, when there are many flows in the datapath,
932 * this expensive operation can occur so frequently, that it reduces our
933 * ability to quickly set up flows. To reduce the cost, we push statistics
935 if (time_msec() > push_timer
) {
936 push_timer
= time_msec() + 2000;
940 if (backer
->need_revalidate
941 || !tag_set_is_empty(&backer
->revalidate_set
)) {
942 struct tag_set revalidate_set
= backer
->revalidate_set
;
943 bool need_revalidate
= backer
->need_revalidate
;
944 struct ofproto_dpif
*ofproto
;
945 struct simap_node
*node
;
946 struct simap tmp_backers
;
948 /* Handle tunnel garbage collection. */
949 simap_init(&tmp_backers
);
950 simap_swap(&backer
->tnl_backers
, &tmp_backers
);
952 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
953 struct ofport_dpif
*iter
;
955 if (backer
!= ofproto
->backer
) {
959 HMAP_FOR_EACH (iter
, up
.hmap_node
, &ofproto
->up
.ports
) {
962 if (!iter
->tnl_port
) {
966 dp_port
= netdev_vport_get_dpif_port(iter
->up
.netdev
);
967 node
= simap_find(&tmp_backers
, dp_port
);
969 simap_put(&backer
->tnl_backers
, dp_port
, node
->data
);
970 simap_delete(&tmp_backers
, node
);
971 node
= simap_find(&backer
->tnl_backers
, dp_port
);
973 node
= simap_find(&backer
->tnl_backers
, dp_port
);
975 uint32_t odp_port
= UINT32_MAX
;
977 if (!dpif_port_add(backer
->dpif
, iter
->up
.netdev
,
979 simap_put(&backer
->tnl_backers
, dp_port
, odp_port
);
980 node
= simap_find(&backer
->tnl_backers
, dp_port
);
985 iter
->odp_port
= node
? node
->data
: OVSP_NONE
;
986 if (tnl_port_reconfigure(&iter
->up
, iter
->odp_port
,
988 backer
->need_revalidate
= REV_RECONFIGURE
;
993 SIMAP_FOR_EACH (node
, &tmp_backers
) {
994 dpif_port_del(backer
->dpif
, node
->data
);
996 simap_destroy(&tmp_backers
);
998 switch (backer
->need_revalidate
) {
999 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
1000 case REV_STP
: COVERAGE_INC(rev_stp
); break;
1001 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
1002 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
1003 case REV_INCONSISTENCY
: COVERAGE_INC(rev_inconsistency
); break;
1006 if (backer
->need_revalidate
) {
1007 /* Clear the drop_keys in case we should now be accepting some
1008 * formerly dropped flows. */
1009 drop_key_clear(backer
);
1012 /* Clear the revalidation flags. */
1013 tag_set_init(&backer
->revalidate_set
);
1014 backer
->need_revalidate
= 0;
1016 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
1017 struct facet
*facet
, *next
;
1019 if (ofproto
->backer
!= backer
) {
1023 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &ofproto
->facets
) {
1025 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1026 facet_revalidate(facet
);
1033 if (timer_expired(&backer
->next_expiration
)) {
1034 int delay
= expire(backer
);
1035 timer_set_duration(&backer
->next_expiration
, delay
);
1038 /* Check for port changes in the dpif. */
1039 while ((error
= dpif_port_poll(backer
->dpif
, &devname
)) == 0) {
1040 struct ofproto_dpif
*ofproto
;
1041 struct dpif_port port
;
1043 /* Don't report on the datapath's device. */
1044 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
1048 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
1049 &all_ofproto_dpifs
) {
1050 if (simap_contains(&ofproto
->backer
->tnl_backers
, devname
)) {
1055 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
1056 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
1057 /* The port was removed. If we know the datapath,
1058 * report it through poll_set(). If we don't, it may be
1059 * notifying us of a removal we initiated, so ignore it.
1060 * If there's a pending ENOBUFS, let it stand, since
1061 * everything will be reevaluated. */
1062 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
1063 sset_add(&ofproto
->port_poll_set
, devname
);
1064 ofproto
->port_poll_errno
= 0;
1066 } else if (!ofproto
) {
1067 /* The port was added, but we don't know with which
1068 * ofproto we should associate it. Delete it. */
1069 dpif_port_del(backer
->dpif
, port
.port_no
);
1071 dpif_port_destroy(&port
);
1077 if (error
!= EAGAIN
) {
1078 struct ofproto_dpif
*ofproto
;
1080 /* There was some sort of error, so propagate it to all
1081 * ofprotos that use this backer. */
1082 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
1083 &all_ofproto_dpifs
) {
1084 if (ofproto
->backer
== backer
) {
1085 sset_clear(&ofproto
->port_poll_set
);
1086 ofproto
->port_poll_errno
= error
;
1095 dpif_backer_run_fast(struct dpif_backer
*backer
, int max_batch
)
1099 /* Handle one or more batches of upcalls, until there's nothing left to do
1100 * or until we do a fixed total amount of work.
1102 * We do work in batches because it can be much cheaper to set up a number
1103 * of flows and fire off their patches all at once. We do multiple batches
1104 * because in some cases handling a packet can cause another packet to be
1105 * queued almost immediately as part of the return flow. Both
1106 * optimizations can make major improvements on some benchmarks and
1107 * presumably for real traffic as well. */
1109 while (work
< max_batch
) {
1110 int retval
= handle_upcalls(backer
, max_batch
- work
);
1121 type_run_fast(const char *type
)
1123 struct dpif_backer
*backer
;
1125 backer
= shash_find_data(&all_dpif_backers
, type
);
1127 /* This is not necessarily a problem, since backers are only
1128 * created on demand. */
1132 return dpif_backer_run_fast(backer
, FLOW_MISS_MAX_BATCH
);
1138 static long long int port_rl
= LLONG_MIN
;
1139 static unsigned int backer_rl
= 0;
1141 if (time_msec() >= port_rl
) {
1142 struct ofproto_dpif
*ofproto
;
1143 struct ofport_dpif
*ofport
;
1145 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
1147 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1148 port_run_fast(ofport
);
1151 port_rl
= time_msec() + 200;
1154 /* XXX: We have to be careful not to do too much work in this function. If
1155 * we call dpif_backer_run_fast() too often, or with too large a batch,
1156 * performance improves signifcantly, but at a cost. It's possible for the
1157 * number of flows in the datapath to increase without bound, and for poll
1158 * loops to take 10s of seconds. The correct solution to this problem,
1159 * long term, is to separate flow miss handling into it's own thread so it
1160 * isn't affected by revalidations, and expirations. Until then, this is
1161 * the best we can do. */
1162 if (++backer_rl
>= 10) {
1163 struct shash_node
*node
;
1166 SHASH_FOR_EACH (node
, &all_dpif_backers
) {
1167 dpif_backer_run_fast(node
->data
, 1);
1173 type_wait(const char *type
)
1175 struct dpif_backer
*backer
;
1177 backer
= shash_find_data(&all_dpif_backers
, type
);
1179 /* This is not necessarily a problem, since backers are only
1180 * created on demand. */
1184 timer_wait(&backer
->next_expiration
);
1187 /* Basic life-cycle. */
1189 static int add_internal_flows(struct ofproto_dpif
*);
1191 static struct ofproto
*
1194 struct ofproto_dpif
*ofproto
= xmalloc(sizeof *ofproto
);
1195 return &ofproto
->up
;
1199 dealloc(struct ofproto
*ofproto_
)
1201 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1206 close_dpif_backer(struct dpif_backer
*backer
)
1208 struct shash_node
*node
;
1210 ovs_assert(backer
->refcount
> 0);
1212 if (--backer
->refcount
) {
1216 drop_key_clear(backer
);
1217 hmap_destroy(&backer
->drop_keys
);
1219 simap_destroy(&backer
->tnl_backers
);
1220 hmap_destroy(&backer
->odp_to_ofport_map
);
1221 node
= shash_find(&all_dpif_backers
, backer
->type
);
1223 shash_delete(&all_dpif_backers
, node
);
1224 dpif_close(backer
->dpif
);
1229 /* Datapath port slated for removal from datapath. */
1230 struct odp_garbage
{
1231 struct list list_node
;
1236 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
1238 struct dpif_backer
*backer
;
1239 struct dpif_port_dump port_dump
;
1240 struct dpif_port port
;
1241 struct shash_node
*node
;
1242 struct list garbage_list
;
1243 struct odp_garbage
*garbage
, *next
;
1249 backer
= shash_find_data(&all_dpif_backers
, type
);
1256 backer_name
= xasprintf("ovs-%s", type
);
1258 /* Remove any existing datapaths, since we assume we're the only
1259 * userspace controlling the datapath. */
1261 dp_enumerate_names(type
, &names
);
1262 SSET_FOR_EACH(name
, &names
) {
1263 struct dpif
*old_dpif
;
1265 /* Don't remove our backer if it exists. */
1266 if (!strcmp(name
, backer_name
)) {
1270 if (dpif_open(name
, type
, &old_dpif
)) {
1271 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
1273 dpif_delete(old_dpif
);
1274 dpif_close(old_dpif
);
1277 sset_destroy(&names
);
1279 backer
= xmalloc(sizeof *backer
);
1281 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
1284 VLOG_ERR("failed to open datapath of type %s: %s", type
,
1290 backer
->type
= xstrdup(type
);
1291 backer
->refcount
= 1;
1292 hmap_init(&backer
->odp_to_ofport_map
);
1293 hmap_init(&backer
->drop_keys
);
1294 timer_set_duration(&backer
->next_expiration
, 1000);
1295 backer
->need_revalidate
= 0;
1296 simap_init(&backer
->tnl_backers
);
1297 tag_set_init(&backer
->revalidate_set
);
1300 dpif_flow_flush(backer
->dpif
);
1302 /* Loop through the ports already on the datapath and remove any
1303 * that we don't need anymore. */
1304 list_init(&garbage_list
);
1305 dpif_port_dump_start(&port_dump
, backer
->dpif
);
1306 while (dpif_port_dump_next(&port_dump
, &port
)) {
1307 node
= shash_find(&init_ofp_ports
, port
.name
);
1308 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
1309 garbage
= xmalloc(sizeof *garbage
);
1310 garbage
->odp_port
= port
.port_no
;
1311 list_push_front(&garbage_list
, &garbage
->list_node
);
1314 dpif_port_dump_done(&port_dump
);
1316 LIST_FOR_EACH_SAFE (garbage
, next
, list_node
, &garbage_list
) {
1317 dpif_port_del(backer
->dpif
, garbage
->odp_port
);
1318 list_remove(&garbage
->list_node
);
1322 shash_add(&all_dpif_backers
, type
, backer
);
1324 error
= dpif_recv_set(backer
->dpif
, true);
1326 VLOG_ERR("failed to listen on datapath of type %s: %s",
1327 type
, strerror(error
));
1328 close_dpif_backer(backer
);
1336 construct(struct ofproto
*ofproto_
)
1338 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1339 struct shash_node
*node
, *next
;
1344 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1349 max_ports
= dpif_get_max_ports(ofproto
->backer
->dpif
);
1350 ofproto_init_max_ports(ofproto_
, MIN(max_ports
, OFPP_MAX
));
1352 ofproto
->n_matches
= 0;
1354 ofproto
->netflow
= NULL
;
1355 ofproto
->sflow
= NULL
;
1356 ofproto
->ipfix
= NULL
;
1357 ofproto
->stp
= NULL
;
1358 hmap_init(&ofproto
->bundles
);
1359 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1360 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
1361 ofproto
->mirrors
[i
] = NULL
;
1363 ofproto
->has_bonded_bundles
= false;
1365 hmap_init(&ofproto
->facets
);
1366 hmap_init(&ofproto
->subfacets
);
1367 ofproto
->governor
= NULL
;
1368 ofproto
->consistency_rl
= LLONG_MIN
;
1370 for (i
= 0; i
< N_TABLES
; i
++) {
1371 struct table_dpif
*table
= &ofproto
->tables
[i
];
1373 table
->catchall_table
= NULL
;
1374 table
->other_table
= NULL
;
1375 table
->basis
= random_uint32();
1378 list_init(&ofproto
->completions
);
1380 ofproto_dpif_unixctl_init();
1382 ofproto
->has_mirrors
= false;
1383 ofproto
->has_bundle_action
= false;
1385 hmap_init(&ofproto
->vlandev_map
);
1386 hmap_init(&ofproto
->realdev_vid_map
);
1388 sset_init(&ofproto
->ports
);
1389 sset_init(&ofproto
->ghost_ports
);
1390 sset_init(&ofproto
->port_poll_set
);
1391 ofproto
->port_poll_errno
= 0;
1393 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1394 struct iface_hint
*iface_hint
= node
->data
;
1396 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1397 /* Check if the datapath already has this port. */
1398 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1399 sset_add(&ofproto
->ports
, node
->name
);
1402 free(iface_hint
->br_name
);
1403 free(iface_hint
->br_type
);
1405 shash_delete(&init_ofp_ports
, node
);
1409 hmap_insert(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
,
1410 hash_string(ofproto
->up
.name
, 0));
1411 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1413 ofproto_init_tables(ofproto_
, N_TABLES
);
1414 error
= add_internal_flows(ofproto
);
1415 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1418 ofproto
->n_missed
= 0;
1420 ofproto
->max_n_subfacet
= 0;
1421 ofproto
->created
= time_msec();
1422 ofproto
->last_minute
= ofproto
->created
;
1423 memset(&ofproto
->hourly
, 0, sizeof ofproto
->hourly
);
1424 memset(&ofproto
->daily
, 0, sizeof ofproto
->daily
);
1425 ofproto
->subfacet_add_count
= 0;
1426 ofproto
->subfacet_del_count
= 0;
1427 ofproto
->total_subfacet_add_count
= 0;
1428 ofproto
->total_subfacet_del_count
= 0;
1429 ofproto
->total_subfacet_life_span
= 0;
1430 ofproto
->total_subfacet_count
= 0;
1431 ofproto
->n_update_stats
= 0;
1437 add_internal_flow(struct ofproto_dpif
*ofproto
, int id
,
1438 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1440 struct ofputil_flow_mod fm
;
1443 match_init_catchall(&fm
.match
);
1445 match_set_reg(&fm
.match
, 0, id
);
1446 fm
.new_cookie
= htonll(0);
1447 fm
.cookie
= htonll(0);
1448 fm
.cookie_mask
= htonll(0);
1449 fm
.table_id
= TBL_INTERNAL
;
1450 fm
.command
= OFPFC_ADD
;
1451 fm
.idle_timeout
= 0;
1452 fm
.hard_timeout
= 0;
1456 fm
.ofpacts
= ofpacts
->data
;
1457 fm
.ofpacts_len
= ofpacts
->size
;
1459 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
1461 VLOG_ERR_RL(&rl
, "failed to add internal flow %d (%s)",
1462 id
, ofperr_to_string(error
));
1466 *rulep
= rule_dpif_lookup__(ofproto
, &fm
.match
.flow
, TBL_INTERNAL
);
1467 ovs_assert(*rulep
!= NULL
);
1473 add_internal_flows(struct ofproto_dpif
*ofproto
)
1475 struct ofpact_controller
*controller
;
1476 uint64_t ofpacts_stub
[128 / 8];
1477 struct ofpbuf ofpacts
;
1481 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1484 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1485 controller
->max_len
= UINT16_MAX
;
1486 controller
->controller_id
= 0;
1487 controller
->reason
= OFPR_NO_MATCH
;
1488 ofpact_pad(&ofpacts
);
1490 error
= add_internal_flow(ofproto
, id
++, &ofpacts
, &ofproto
->miss_rule
);
1495 ofpbuf_clear(&ofpacts
);
1496 error
= add_internal_flow(ofproto
, id
++, &ofpacts
,
1497 &ofproto
->no_packet_in_rule
);
1502 complete_operations(struct ofproto_dpif
*ofproto
)
1504 struct dpif_completion
*c
, *next
;
1506 LIST_FOR_EACH_SAFE (c
, next
, list_node
, &ofproto
->completions
) {
1507 ofoperation_complete(c
->op
, 0);
1508 list_remove(&c
->list_node
);
1514 destruct(struct ofproto
*ofproto_
)
1516 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1517 struct rule_dpif
*rule
, *next_rule
;
1518 struct oftable
*table
;
1521 hmap_remove(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
);
1522 complete_operations(ofproto
);
1524 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1525 struct cls_cursor cursor
;
1527 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1528 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
1529 ofproto_rule_destroy(&rule
->up
);
1533 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
1534 mirror_destroy(ofproto
->mirrors
[i
]);
1537 netflow_destroy(ofproto
->netflow
);
1538 dpif_sflow_destroy(ofproto
->sflow
);
1539 hmap_destroy(&ofproto
->bundles
);
1540 mac_learning_destroy(ofproto
->ml
);
1542 hmap_destroy(&ofproto
->facets
);
1543 hmap_destroy(&ofproto
->subfacets
);
1544 governor_destroy(ofproto
->governor
);
1546 hmap_destroy(&ofproto
->vlandev_map
);
1547 hmap_destroy(&ofproto
->realdev_vid_map
);
1549 sset_destroy(&ofproto
->ports
);
1550 sset_destroy(&ofproto
->ghost_ports
);
1551 sset_destroy(&ofproto
->port_poll_set
);
1553 close_dpif_backer(ofproto
->backer
);
1557 run_fast(struct ofproto
*ofproto_
)
1559 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1560 struct ofport_dpif
*ofport
;
1562 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1563 port_run_fast(ofport
);
1570 run(struct ofproto
*ofproto_
)
1572 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1573 struct ofport_dpif
*ofport
;
1574 struct ofbundle
*bundle
;
1578 complete_operations(ofproto
);
1581 error
= run_fast(ofproto_
);
1586 if (ofproto
->netflow
) {
1587 if (netflow_run(ofproto
->netflow
)) {
1588 send_netflow_active_timeouts(ofproto
);
1591 if (ofproto
->sflow
) {
1592 dpif_sflow_run(ofproto
->sflow
);
1595 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1598 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1603 mac_learning_run(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
1605 /* Check the consistency of a random facet, to aid debugging. */
1606 if (time_msec() >= ofproto
->consistency_rl
1607 && !hmap_is_empty(&ofproto
->facets
)
1608 && !ofproto
->backer
->need_revalidate
) {
1609 struct facet
*facet
;
1611 ofproto
->consistency_rl
= time_msec() + 250;
1613 facet
= CONTAINER_OF(hmap_random_node(&ofproto
->facets
),
1614 struct facet
, hmap_node
);
1615 if (!tag_set_intersects(&ofproto
->backer
->revalidate_set
,
1617 if (!facet_check_consistency(facet
)) {
1618 ofproto
->backer
->need_revalidate
= REV_INCONSISTENCY
;
1623 if (ofproto
->governor
) {
1626 governor_run(ofproto
->governor
);
1628 /* If the governor has shrunk to its minimum size and the number of
1629 * subfacets has dwindled, then drop the governor entirely.
1631 * For hysteresis, the number of subfacets to drop the governor is
1632 * smaller than the number needed to trigger its creation. */
1633 n_subfacets
= hmap_count(&ofproto
->subfacets
);
1634 if (n_subfacets
* 4 < ofproto
->up
.flow_eviction_threshold
1635 && governor_is_idle(ofproto
->governor
)) {
1636 governor_destroy(ofproto
->governor
);
1637 ofproto
->governor
= NULL
;
1645 wait(struct ofproto
*ofproto_
)
1647 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1648 struct ofport_dpif
*ofport
;
1649 struct ofbundle
*bundle
;
1651 if (!clogged
&& !list_is_empty(&ofproto
->completions
)) {
1652 poll_immediate_wake();
1655 dpif_wait(ofproto
->backer
->dpif
);
1656 dpif_recv_wait(ofproto
->backer
->dpif
);
1657 if (ofproto
->sflow
) {
1658 dpif_sflow_wait(ofproto
->sflow
);
1660 if (!tag_set_is_empty(&ofproto
->backer
->revalidate_set
)) {
1661 poll_immediate_wake();
1663 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1666 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1667 bundle_wait(bundle
);
1669 if (ofproto
->netflow
) {
1670 netflow_wait(ofproto
->netflow
);
1672 mac_learning_wait(ofproto
->ml
);
1674 if (ofproto
->backer
->need_revalidate
) {
1675 /* Shouldn't happen, but if it does just go around again. */
1676 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1677 poll_immediate_wake();
1679 if (ofproto
->governor
) {
1680 governor_wait(ofproto
->governor
);
1685 get_memory_usage(const struct ofproto
*ofproto_
, struct simap
*usage
)
1687 const struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1689 simap_increase(usage
, "facets", hmap_count(&ofproto
->facets
));
1690 simap_increase(usage
, "subfacets", hmap_count(&ofproto
->subfacets
));
1694 flush(struct ofproto
*ofproto_
)
1696 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1697 struct subfacet
*subfacet
, *next_subfacet
;
1698 struct subfacet
*batch
[SUBFACET_DESTROY_MAX_BATCH
];
1702 HMAP_FOR_EACH_SAFE (subfacet
, next_subfacet
, hmap_node
,
1703 &ofproto
->subfacets
) {
1704 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
1705 batch
[n_batch
++] = subfacet
;
1706 if (n_batch
>= SUBFACET_DESTROY_MAX_BATCH
) {
1707 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
1711 subfacet_destroy(subfacet
);
1716 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
1721 get_features(struct ofproto
*ofproto_ OVS_UNUSED
,
1722 bool *arp_match_ip
, enum ofputil_action_bitmap
*actions
)
1724 *arp_match_ip
= true;
1725 *actions
= (OFPUTIL_A_OUTPUT
|
1726 OFPUTIL_A_SET_VLAN_VID
|
1727 OFPUTIL_A_SET_VLAN_PCP
|
1728 OFPUTIL_A_STRIP_VLAN
|
1729 OFPUTIL_A_SET_DL_SRC
|
1730 OFPUTIL_A_SET_DL_DST
|
1731 OFPUTIL_A_SET_NW_SRC
|
1732 OFPUTIL_A_SET_NW_DST
|
1733 OFPUTIL_A_SET_NW_TOS
|
1734 OFPUTIL_A_SET_TP_SRC
|
1735 OFPUTIL_A_SET_TP_DST
|
1740 get_tables(struct ofproto
*ofproto_
, struct ofp12_table_stats
*ots
)
1742 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1743 struct dpif_dp_stats s
;
1745 strcpy(ots
->name
, "classifier");
1747 dpif_get_dp_stats(ofproto
->backer
->dpif
, &s
);
1749 ots
->lookup_count
= htonll(s
.n_hit
+ s
.n_missed
);
1750 ots
->matched_count
= htonll(s
.n_hit
+ ofproto
->n_matches
);
1753 static struct ofport
*
1756 struct ofport_dpif
*port
= xmalloc(sizeof *port
);
1761 port_dealloc(struct ofport
*port_
)
1763 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1768 port_construct(struct ofport
*port_
)
1770 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1771 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1772 const struct netdev
*netdev
= port
->up
.netdev
;
1773 struct dpif_port dpif_port
;
1776 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1777 port
->bundle
= NULL
;
1780 port
->tag
= tag_create_random();
1781 port
->may_enable
= true;
1782 port
->stp_port
= NULL
;
1783 port
->stp_state
= STP_DISABLED
;
1784 port
->tnl_port
= NULL
;
1785 hmap_init(&port
->priorities
);
1786 port
->realdev_ofp_port
= 0;
1787 port
->vlandev_vid
= 0;
1788 port
->carrier_seq
= netdev_get_carrier_resets(netdev
);
1790 if (netdev_vport_is_patch(netdev
)) {
1791 /* By bailing out here, we don't submit the port to the sFlow module
1792 * to be considered for counter polling export. This is correct
1793 * because the patch port represents an interface that sFlow considers
1794 * to be "internal" to the switch as a whole, and therefore not an
1795 * candidate for counter polling. */
1796 port
->odp_port
= OVSP_NONE
;
1800 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
1801 netdev_vport_get_dpif_port(netdev
),
1807 port
->odp_port
= dpif_port
.port_no
;
1809 if (netdev_get_tunnel_config(netdev
)) {
1810 port
->tnl_port
= tnl_port_add(&port
->up
, port
->odp_port
);
1812 /* Sanity-check that a mapping doesn't already exist. This
1813 * shouldn't happen for non-tunnel ports. */
1814 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
1815 VLOG_ERR("port %s already has an OpenFlow port number",
1817 dpif_port_destroy(&dpif_port
);
1821 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
1822 hash_int(port
->odp_port
, 0));
1824 dpif_port_destroy(&dpif_port
);
1826 if (ofproto
->sflow
) {
1827 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
1834 port_destruct(struct ofport
*port_
)
1836 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1837 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1838 const char *dp_port_name
= netdev_vport_get_dpif_port(port
->up
.netdev
);
1839 const char *devname
= netdev_get_name(port
->up
.netdev
);
1841 if (dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
1842 /* The underlying device is still there, so delete it. This
1843 * happens when the ofproto is being destroyed, since the caller
1844 * assumes that removal of attached ports will happen as part of
1846 if (!port
->tnl_port
) {
1847 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
);
1849 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1852 if (port
->odp_port
!= OVSP_NONE
&& !port
->tnl_port
) {
1853 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
1856 tnl_port_del(port
->tnl_port
);
1857 sset_find_and_delete(&ofproto
->ports
, devname
);
1858 sset_find_and_delete(&ofproto
->ghost_ports
, devname
);
1859 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1860 bundle_remove(port_
);
1861 set_cfm(port_
, NULL
);
1862 if (ofproto
->sflow
) {
1863 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
1866 ofport_clear_priorities(port
);
1867 hmap_destroy(&port
->priorities
);
1871 port_modified(struct ofport
*port_
)
1873 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1875 if (port
->bundle
&& port
->bundle
->bond
) {
1876 bond_slave_set_netdev(port
->bundle
->bond
, port
, port
->up
.netdev
);
1881 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
1883 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1884 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1885 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
1887 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
1888 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
1889 OFPUTIL_PC_NO_PACKET_IN
)) {
1890 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1892 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
1893 bundle_update(port
->bundle
);
1899 set_sflow(struct ofproto
*ofproto_
,
1900 const struct ofproto_sflow_options
*sflow_options
)
1902 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1903 struct dpif_sflow
*ds
= ofproto
->sflow
;
1905 if (sflow_options
) {
1907 struct ofport_dpif
*ofport
;
1909 ds
= ofproto
->sflow
= dpif_sflow_create();
1910 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1911 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
1913 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1915 dpif_sflow_set_options(ds
, sflow_options
);
1918 dpif_sflow_destroy(ds
);
1919 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1920 ofproto
->sflow
= NULL
;
1928 struct ofproto
*ofproto_
,
1929 const struct ofproto_ipfix_bridge_exporter_options
*bridge_exporter_options
,
1930 const struct ofproto_ipfix_flow_exporter_options
*flow_exporters_options
,
1931 size_t n_flow_exporters_options
)
1933 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1934 struct dpif_ipfix
*di
= ofproto
->ipfix
;
1936 if (bridge_exporter_options
|| flow_exporters_options
) {
1938 di
= ofproto
->ipfix
= dpif_ipfix_create();
1940 dpif_ipfix_set_options(
1941 di
, bridge_exporter_options
, flow_exporters_options
,
1942 n_flow_exporters_options
);
1945 dpif_ipfix_destroy(di
);
1946 ofproto
->ipfix
= NULL
;
1953 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
1955 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1962 struct ofproto_dpif
*ofproto
;
1964 ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1965 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1966 ofport
->cfm
= cfm_create(netdev_get_name(ofport
->up
.netdev
));
1969 if (cfm_configure(ofport
->cfm
, s
)) {
1975 cfm_destroy(ofport
->cfm
);
1981 get_cfm_status(const struct ofport
*ofport_
,
1982 struct ofproto_cfm_status
*status
)
1984 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1987 status
->faults
= cfm_get_fault(ofport
->cfm
);
1988 status
->remote_opstate
= cfm_get_opup(ofport
->cfm
);
1989 status
->health
= cfm_get_health(ofport
->cfm
);
1990 cfm_get_remote_mpids(ofport
->cfm
, &status
->rmps
, &status
->n_rmps
);
1998 set_bfd(struct ofport
*ofport_
, const struct smap
*cfg
)
2000 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
2001 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2005 ofport
->bfd
= bfd_configure(old
, netdev_get_name(ofport
->up
.netdev
), cfg
);
2006 if (ofport
->bfd
!= old
) {
2007 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2014 get_bfd_status(struct ofport
*ofport_
, struct smap
*smap
)
2016 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2019 bfd_get_status(ofport
->bfd
, smap
);
2026 /* Spanning Tree. */
2029 send_bpdu_cb(struct ofpbuf
*pkt
, int port_num
, void *ofproto_
)
2031 struct ofproto_dpif
*ofproto
= ofproto_
;
2032 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
2033 struct ofport_dpif
*ofport
;
2035 ofport
= stp_port_get_aux(sp
);
2037 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
2038 ofproto
->up
.name
, port_num
);
2040 struct eth_header
*eth
= pkt
->l2
;
2042 netdev_get_etheraddr(ofport
->up
.netdev
, eth
->eth_src
);
2043 if (eth_addr_is_zero(eth
->eth_src
)) {
2044 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
2045 "with unknown MAC", ofproto
->up
.name
, port_num
);
2047 send_packet(ofport
, pkt
);
2053 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2055 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
2057 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2059 /* Only revalidate flows if the configuration changed. */
2060 if (!s
!= !ofproto
->stp
) {
2061 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2065 if (!ofproto
->stp
) {
2066 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
2067 send_bpdu_cb
, ofproto
);
2068 ofproto
->stp_last_tick
= time_msec();
2071 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
2072 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
2073 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
2074 stp_set_max_age(ofproto
->stp
, s
->max_age
);
2075 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
2077 struct ofport
*ofport
;
2079 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
2080 set_stp_port(ofport
, NULL
);
2083 stp_destroy(ofproto
->stp
);
2084 ofproto
->stp
= NULL
;
2091 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
2093 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2097 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
2098 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
2099 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
2108 update_stp_port_state(struct ofport_dpif
*ofport
)
2110 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2111 enum stp_state state
;
2113 /* Figure out new state. */
2114 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
2118 if (ofport
->stp_state
!= state
) {
2119 enum ofputil_port_state of_state
;
2122 VLOG_DBG_RL(&rl
, "port %s: STP state changed from %s to %s",
2123 netdev_get_name(ofport
->up
.netdev
),
2124 stp_state_name(ofport
->stp_state
),
2125 stp_state_name(state
));
2126 if (stp_learn_in_state(ofport
->stp_state
)
2127 != stp_learn_in_state(state
)) {
2128 /* xxx Learning action flows should also be flushed. */
2129 mac_learning_flush(ofproto
->ml
,
2130 &ofproto
->backer
->revalidate_set
);
2132 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
2133 != stp_forward_in_state(state
);
2135 ofproto
->backer
->need_revalidate
= REV_STP
;
2136 ofport
->stp_state
= state
;
2137 ofport
->stp_state_entered
= time_msec();
2139 if (fwd_change
&& ofport
->bundle
) {
2140 bundle_update(ofport
->bundle
);
2143 /* Update the STP state bits in the OpenFlow port description. */
2144 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
2145 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
2146 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
2147 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
2148 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
2150 ofproto_port_set_state(&ofport
->up
, of_state
);
2154 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2155 * caller is responsible for assigning STP port numbers and ensuring
2156 * there are no duplicates. */
2158 set_stp_port(struct ofport
*ofport_
,
2159 const struct ofproto_port_stp_settings
*s
)
2161 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2162 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2163 struct stp_port
*sp
= ofport
->stp_port
;
2165 if (!s
|| !s
->enable
) {
2167 ofport
->stp_port
= NULL
;
2168 stp_port_disable(sp
);
2169 update_stp_port_state(ofport
);
2172 } else if (sp
&& stp_port_no(sp
) != s
->port_num
2173 && ofport
== stp_port_get_aux(sp
)) {
2174 /* The port-id changed, so disable the old one if it's not
2175 * already in use by another port. */
2176 stp_port_disable(sp
);
2179 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
2180 stp_port_enable(sp
);
2182 stp_port_set_aux(sp
, ofport
);
2183 stp_port_set_priority(sp
, s
->priority
);
2184 stp_port_set_path_cost(sp
, s
->path_cost
);
2186 update_stp_port_state(ofport
);
2192 get_stp_port_status(struct ofport
*ofport_
,
2193 struct ofproto_port_stp_status
*s
)
2195 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2196 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2197 struct stp_port
*sp
= ofport
->stp_port
;
2199 if (!ofproto
->stp
|| !sp
) {
2205 s
->port_id
= stp_port_get_id(sp
);
2206 s
->state
= stp_port_get_state(sp
);
2207 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
2208 s
->role
= stp_port_get_role(sp
);
2209 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
2215 stp_run(struct ofproto_dpif
*ofproto
)
2218 long long int now
= time_msec();
2219 long long int elapsed
= now
- ofproto
->stp_last_tick
;
2220 struct stp_port
*sp
;
2223 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
2224 ofproto
->stp_last_tick
= now
;
2226 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
2227 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
2230 update_stp_port_state(ofport
);
2234 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
2235 mac_learning_flush(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
2241 stp_wait(struct ofproto_dpif
*ofproto
)
2244 poll_timer_wait(1000);
2248 /* Returns true if STP should process 'flow'. */
2250 stp_should_process_flow(const struct flow
*flow
)
2252 return eth_addr_equals(flow
->dl_dst
, eth_addr_stp
);
2256 stp_process_packet(const struct ofport_dpif
*ofport
,
2257 const struct ofpbuf
*packet
)
2259 struct ofpbuf payload
= *packet
;
2260 struct eth_header
*eth
= payload
.data
;
2261 struct stp_port
*sp
= ofport
->stp_port
;
2263 /* Sink packets on ports that have STP disabled when the bridge has
2265 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
2269 /* Trim off padding on payload. */
2270 if (payload
.size
> ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
2271 payload
.size
= ntohs(eth
->eth_type
) + ETH_HEADER_LEN
;
2274 if (ofpbuf_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
2275 stp_received_bpdu(sp
, payload
.data
, payload
.size
);
2279 static struct priority_to_dscp
*
2280 get_priority(const struct ofport_dpif
*ofport
, uint32_t priority
)
2282 struct priority_to_dscp
*pdscp
;
2285 hash
= hash_int(priority
, 0);
2286 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &ofport
->priorities
) {
2287 if (pdscp
->priority
== priority
) {
2295 ofport_clear_priorities(struct ofport_dpif
*ofport
)
2297 struct priority_to_dscp
*pdscp
, *next
;
2299 HMAP_FOR_EACH_SAFE (pdscp
, next
, hmap_node
, &ofport
->priorities
) {
2300 hmap_remove(&ofport
->priorities
, &pdscp
->hmap_node
);
2306 set_queues(struct ofport
*ofport_
,
2307 const struct ofproto_port_queue
*qdscp_list
,
2310 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2311 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2312 struct hmap
new = HMAP_INITIALIZER(&new);
2315 for (i
= 0; i
< n_qdscp
; i
++) {
2316 struct priority_to_dscp
*pdscp
;
2320 dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
2321 if (dpif_queue_to_priority(ofproto
->backer
->dpif
, qdscp_list
[i
].queue
,
2326 pdscp
= get_priority(ofport
, priority
);
2328 hmap_remove(&ofport
->priorities
, &pdscp
->hmap_node
);
2330 pdscp
= xmalloc(sizeof *pdscp
);
2331 pdscp
->priority
= priority
;
2333 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2336 if (pdscp
->dscp
!= dscp
) {
2338 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2341 hmap_insert(&new, &pdscp
->hmap_node
, hash_int(pdscp
->priority
, 0));
2344 if (!hmap_is_empty(&ofport
->priorities
)) {
2345 ofport_clear_priorities(ofport
);
2346 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2349 hmap_swap(&new, &ofport
->priorities
);
2357 /* Expires all MAC learning entries associated with 'bundle' and forces its
2358 * ofproto to revalidate every flow.
2360 * Normally MAC learning entries are removed only from the ofproto associated
2361 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2362 * are removed from every ofproto. When patch ports and SLB bonds are in use
2363 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2364 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2365 * with the host from which it migrated. */
2367 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
2369 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2370 struct mac_learning
*ml
= ofproto
->ml
;
2371 struct mac_entry
*mac
, *next_mac
;
2373 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2374 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
2375 if (mac
->port
.p
== bundle
) {
2377 struct ofproto_dpif
*o
;
2379 HMAP_FOR_EACH (o
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2381 struct mac_entry
*e
;
2383 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
,
2386 mac_learning_expire(o
->ml
, e
);
2392 mac_learning_expire(ml
, mac
);
2397 static struct ofbundle
*
2398 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
2400 struct ofbundle
*bundle
;
2402 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
2403 &ofproto
->bundles
) {
2404 if (bundle
->aux
== aux
) {
2411 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2412 * ones that are found to 'bundles'. */
2414 bundle_lookup_multiple(struct ofproto_dpif
*ofproto
,
2415 void **auxes
, size_t n_auxes
,
2416 struct hmapx
*bundles
)
2420 hmapx_init(bundles
);
2421 for (i
= 0; i
< n_auxes
; i
++) {
2422 struct ofbundle
*bundle
= bundle_lookup(ofproto
, auxes
[i
]);
2424 hmapx_add(bundles
, bundle
);
2430 bundle_update(struct ofbundle
*bundle
)
2432 struct ofport_dpif
*port
;
2434 bundle
->floodable
= true;
2435 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2436 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2437 || !stp_forward_in_state(port
->stp_state
)) {
2438 bundle
->floodable
= false;
2445 bundle_del_port(struct ofport_dpif
*port
)
2447 struct ofbundle
*bundle
= port
->bundle
;
2449 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2451 list_remove(&port
->bundle_node
);
2452 port
->bundle
= NULL
;
2455 lacp_slave_unregister(bundle
->lacp
, port
);
2458 bond_slave_unregister(bundle
->bond
, port
);
2461 bundle_update(bundle
);
2465 bundle_add_port(struct ofbundle
*bundle
, uint32_t ofp_port
,
2466 struct lacp_slave_settings
*lacp
)
2468 struct ofport_dpif
*port
;
2470 port
= get_ofp_port(bundle
->ofproto
, ofp_port
);
2475 if (port
->bundle
!= bundle
) {
2476 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2478 bundle_del_port(port
);
2481 port
->bundle
= bundle
;
2482 list_push_back(&bundle
->ports
, &port
->bundle_node
);
2483 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2484 || !stp_forward_in_state(port
->stp_state
)) {
2485 bundle
->floodable
= false;
2489 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2490 lacp_slave_register(bundle
->lacp
, port
, lacp
);
2497 bundle_destroy(struct ofbundle
*bundle
)
2499 struct ofproto_dpif
*ofproto
;
2500 struct ofport_dpif
*port
, *next_port
;
2507 ofproto
= bundle
->ofproto
;
2508 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2509 struct ofmirror
*m
= ofproto
->mirrors
[i
];
2511 if (m
->out
== bundle
) {
2513 } else if (hmapx_find_and_delete(&m
->srcs
, bundle
)
2514 || hmapx_find_and_delete(&m
->dsts
, bundle
)) {
2515 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2520 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2521 bundle_del_port(port
);
2524 bundle_flush_macs(bundle
, true);
2525 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
2527 free(bundle
->trunks
);
2528 lacp_destroy(bundle
->lacp
);
2529 bond_destroy(bundle
->bond
);
2534 bundle_set(struct ofproto
*ofproto_
, void *aux
,
2535 const struct ofproto_bundle_settings
*s
)
2537 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2538 bool need_flush
= false;
2539 struct ofport_dpif
*port
;
2540 struct ofbundle
*bundle
;
2541 unsigned long *trunks
;
2547 bundle_destroy(bundle_lookup(ofproto
, aux
));
2551 ovs_assert(s
->n_slaves
== 1 || s
->bond
!= NULL
);
2552 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_slaves
!= NULL
));
2554 bundle
= bundle_lookup(ofproto
, aux
);
2556 bundle
= xmalloc(sizeof *bundle
);
2558 bundle
->ofproto
= ofproto
;
2559 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
2560 hash_pointer(aux
, 0));
2562 bundle
->name
= NULL
;
2564 list_init(&bundle
->ports
);
2565 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
2567 bundle
->trunks
= NULL
;
2568 bundle
->use_priority_tags
= s
->use_priority_tags
;
2569 bundle
->lacp
= NULL
;
2570 bundle
->bond
= NULL
;
2572 bundle
->floodable
= true;
2574 bundle
->src_mirrors
= 0;
2575 bundle
->dst_mirrors
= 0;
2576 bundle
->mirror_out
= 0;
2579 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
2581 bundle
->name
= xstrdup(s
->name
);
2586 if (!bundle
->lacp
) {
2587 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2588 bundle
->lacp
= lacp_create();
2590 lacp_configure(bundle
->lacp
, s
->lacp
);
2592 lacp_destroy(bundle
->lacp
);
2593 bundle
->lacp
= NULL
;
2596 /* Update set of ports. */
2598 for (i
= 0; i
< s
->n_slaves
; i
++) {
2599 if (!bundle_add_port(bundle
, s
->slaves
[i
],
2600 s
->lacp
? &s
->lacp_slaves
[i
] : NULL
)) {
2604 if (!ok
|| list_size(&bundle
->ports
) != s
->n_slaves
) {
2605 struct ofport_dpif
*next_port
;
2607 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2608 for (i
= 0; i
< s
->n_slaves
; i
++) {
2609 if (s
->slaves
[i
] == port
->up
.ofp_port
) {
2614 bundle_del_port(port
);
2618 ovs_assert(list_size(&bundle
->ports
) <= s
->n_slaves
);
2620 if (list_is_empty(&bundle
->ports
)) {
2621 bundle_destroy(bundle
);
2625 /* Set VLAN tagging mode */
2626 if (s
->vlan_mode
!= bundle
->vlan_mode
2627 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
2628 bundle
->vlan_mode
= s
->vlan_mode
;
2629 bundle
->use_priority_tags
= s
->use_priority_tags
;
2634 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
2635 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
2637 if (vlan
!= bundle
->vlan
) {
2638 bundle
->vlan
= vlan
;
2642 /* Get trunked VLANs. */
2643 switch (s
->vlan_mode
) {
2644 case PORT_VLAN_ACCESS
:
2648 case PORT_VLAN_TRUNK
:
2649 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2652 case PORT_VLAN_NATIVE_UNTAGGED
:
2653 case PORT_VLAN_NATIVE_TAGGED
:
2654 if (vlan
!= 0 && (!s
->trunks
2655 || !bitmap_is_set(s
->trunks
, vlan
)
2656 || bitmap_is_set(s
->trunks
, 0))) {
2657 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2659 trunks
= bitmap_clone(s
->trunks
, 4096);
2661 trunks
= bitmap_allocate1(4096);
2663 bitmap_set1(trunks
, vlan
);
2664 bitmap_set0(trunks
, 0);
2666 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2673 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
2674 free(bundle
->trunks
);
2675 if (trunks
== s
->trunks
) {
2676 bundle
->trunks
= vlan_bitmap_clone(trunks
);
2678 bundle
->trunks
= trunks
;
2683 if (trunks
!= s
->trunks
) {
2688 if (!list_is_short(&bundle
->ports
)) {
2689 bundle
->ofproto
->has_bonded_bundles
= true;
2691 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
2692 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2695 bundle
->bond
= bond_create(s
->bond
);
2696 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2699 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2700 bond_slave_register(bundle
->bond
, port
, port
->up
.netdev
);
2703 bond_destroy(bundle
->bond
);
2704 bundle
->bond
= NULL
;
2707 /* If we changed something that would affect MAC learning, un-learn
2708 * everything on this port and force flow revalidation. */
2710 bundle_flush_macs(bundle
, false);
2717 bundle_remove(struct ofport
*port_
)
2719 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2720 struct ofbundle
*bundle
= port
->bundle
;
2723 bundle_del_port(port
);
2724 if (list_is_empty(&bundle
->ports
)) {
2725 bundle_destroy(bundle
);
2726 } else if (list_is_short(&bundle
->ports
)) {
2727 bond_destroy(bundle
->bond
);
2728 bundle
->bond
= NULL
;
2734 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
2736 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 10);
2737 struct ofport_dpif
*port
= port_
;
2738 uint8_t ea
[ETH_ADDR_LEN
];
2741 error
= netdev_get_etheraddr(port
->up
.netdev
, ea
);
2743 struct ofpbuf packet
;
2746 ofpbuf_init(&packet
, 0);
2747 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
2749 memcpy(packet_pdu
, pdu
, pdu_size
);
2751 send_packet(port
, &packet
);
2752 ofpbuf_uninit(&packet
);
2754 VLOG_ERR_RL(&rl
, "port %s: cannot obtain Ethernet address of iface "
2755 "%s (%s)", port
->bundle
->name
,
2756 netdev_get_name(port
->up
.netdev
), strerror(error
));
2761 bundle_send_learning_packets(struct ofbundle
*bundle
)
2763 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2764 int error
, n_packets
, n_errors
;
2765 struct mac_entry
*e
;
2767 error
= n_packets
= n_errors
= 0;
2768 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
2769 if (e
->port
.p
!= bundle
) {
2770 struct ofpbuf
*learning_packet
;
2771 struct ofport_dpif
*port
;
2775 /* The assignment to "port" is unnecessary but makes "grep"ing for
2776 * struct ofport_dpif more effective. */
2777 learning_packet
= bond_compose_learning_packet(bundle
->bond
,
2781 ret
= send_packet(port
, learning_packet
);
2782 ofpbuf_delete(learning_packet
);
2792 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2793 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2794 "packets, last error was: %s",
2795 bundle
->name
, n_errors
, n_packets
, strerror(error
));
2797 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2798 bundle
->name
, n_packets
);
2803 bundle_run(struct ofbundle
*bundle
)
2806 lacp_run(bundle
->lacp
, send_pdu_cb
);
2809 struct ofport_dpif
*port
;
2811 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2812 bond_slave_set_may_enable(bundle
->bond
, port
, port
->may_enable
);
2815 bond_run(bundle
->bond
, &bundle
->ofproto
->backer
->revalidate_set
,
2816 lacp_status(bundle
->lacp
));
2817 if (bond_should_send_learning_packets(bundle
->bond
)) {
2818 bundle_send_learning_packets(bundle
);
2824 bundle_wait(struct ofbundle
*bundle
)
2827 lacp_wait(bundle
->lacp
);
2830 bond_wait(bundle
->bond
);
2837 mirror_scan(struct ofproto_dpif
*ofproto
)
2841 for (idx
= 0; idx
< MAX_MIRRORS
; idx
++) {
2842 if (!ofproto
->mirrors
[idx
]) {
2849 static struct ofmirror
*
2850 mirror_lookup(struct ofproto_dpif
*ofproto
, void *aux
)
2854 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2855 struct ofmirror
*mirror
= ofproto
->mirrors
[i
];
2856 if (mirror
&& mirror
->aux
== aux
) {
2864 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2866 mirror_update_dups(struct ofproto_dpif
*ofproto
)
2870 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2871 struct ofmirror
*m
= ofproto
->mirrors
[i
];
2874 m
->dup_mirrors
= MIRROR_MASK_C(1) << i
;
2878 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2879 struct ofmirror
*m1
= ofproto
->mirrors
[i
];
2886 for (j
= i
+ 1; j
< MAX_MIRRORS
; j
++) {
2887 struct ofmirror
*m2
= ofproto
->mirrors
[j
];
2889 if (m2
&& m1
->out
== m2
->out
&& m1
->out_vlan
== m2
->out_vlan
) {
2890 m1
->dup_mirrors
|= MIRROR_MASK_C(1) << j
;
2891 m2
->dup_mirrors
|= m1
->dup_mirrors
;
2898 mirror_set(struct ofproto
*ofproto_
, void *aux
,
2899 const struct ofproto_mirror_settings
*s
)
2901 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2902 mirror_mask_t mirror_bit
;
2903 struct ofbundle
*bundle
;
2904 struct ofmirror
*mirror
;
2905 struct ofbundle
*out
;
2906 struct hmapx srcs
; /* Contains "struct ofbundle *"s. */
2907 struct hmapx dsts
; /* Contains "struct ofbundle *"s. */
2910 mirror
= mirror_lookup(ofproto
, aux
);
2912 mirror_destroy(mirror
);
2918 idx
= mirror_scan(ofproto
);
2920 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2922 ofproto
->up
.name
, MAX_MIRRORS
, s
->name
);
2926 mirror
= ofproto
->mirrors
[idx
] = xzalloc(sizeof *mirror
);
2927 mirror
->ofproto
= ofproto
;
2930 mirror
->out_vlan
= -1;
2931 mirror
->name
= NULL
;
2934 if (!mirror
->name
|| strcmp(s
->name
, mirror
->name
)) {
2936 mirror
->name
= xstrdup(s
->name
);
2939 /* Get the new configuration. */
2940 if (s
->out_bundle
) {
2941 out
= bundle_lookup(ofproto
, s
->out_bundle
);
2943 mirror_destroy(mirror
);
2949 out_vlan
= s
->out_vlan
;
2951 bundle_lookup_multiple(ofproto
, s
->srcs
, s
->n_srcs
, &srcs
);
2952 bundle_lookup_multiple(ofproto
, s
->dsts
, s
->n_dsts
, &dsts
);
2954 /* If the configuration has not changed, do nothing. */
2955 if (hmapx_equals(&srcs
, &mirror
->srcs
)
2956 && hmapx_equals(&dsts
, &mirror
->dsts
)
2957 && vlan_bitmap_equal(mirror
->vlans
, s
->src_vlans
)
2958 && mirror
->out
== out
2959 && mirror
->out_vlan
== out_vlan
)
2961 hmapx_destroy(&srcs
);
2962 hmapx_destroy(&dsts
);
2966 hmapx_swap(&srcs
, &mirror
->srcs
);
2967 hmapx_destroy(&srcs
);
2969 hmapx_swap(&dsts
, &mirror
->dsts
);
2970 hmapx_destroy(&dsts
);
2972 free(mirror
->vlans
);
2973 mirror
->vlans
= vlan_bitmap_clone(s
->src_vlans
);
2976 mirror
->out_vlan
= out_vlan
;
2978 /* Update bundles. */
2979 mirror_bit
= MIRROR_MASK_C(1) << mirror
->idx
;
2980 HMAP_FOR_EACH (bundle
, hmap_node
, &mirror
->ofproto
->bundles
) {
2981 if (hmapx_contains(&mirror
->srcs
, bundle
)) {
2982 bundle
->src_mirrors
|= mirror_bit
;
2984 bundle
->src_mirrors
&= ~mirror_bit
;
2987 if (hmapx_contains(&mirror
->dsts
, bundle
)) {
2988 bundle
->dst_mirrors
|= mirror_bit
;
2990 bundle
->dst_mirrors
&= ~mirror_bit
;
2993 if (mirror
->out
== bundle
) {
2994 bundle
->mirror_out
|= mirror_bit
;
2996 bundle
->mirror_out
&= ~mirror_bit
;
3000 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3001 ofproto
->has_mirrors
= true;
3002 mac_learning_flush(ofproto
->ml
,
3003 &ofproto
->backer
->revalidate_set
);
3004 mirror_update_dups(ofproto
);
3010 mirror_destroy(struct ofmirror
*mirror
)
3012 struct ofproto_dpif
*ofproto
;
3013 mirror_mask_t mirror_bit
;
3014 struct ofbundle
*bundle
;
3021 ofproto
= mirror
->ofproto
;
3022 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3023 mac_learning_flush(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
3025 mirror_bit
= MIRROR_MASK_C(1) << mirror
->idx
;
3026 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
3027 bundle
->src_mirrors
&= ~mirror_bit
;
3028 bundle
->dst_mirrors
&= ~mirror_bit
;
3029 bundle
->mirror_out
&= ~mirror_bit
;
3032 hmapx_destroy(&mirror
->srcs
);
3033 hmapx_destroy(&mirror
->dsts
);
3034 free(mirror
->vlans
);
3036 ofproto
->mirrors
[mirror
->idx
] = NULL
;
3040 mirror_update_dups(ofproto
);
3042 ofproto
->has_mirrors
= false;
3043 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3044 if (ofproto
->mirrors
[i
]) {
3045 ofproto
->has_mirrors
= true;
3052 mirror_get_stats(struct ofproto
*ofproto_
, void *aux
,
3053 uint64_t *packets
, uint64_t *bytes
)
3055 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3056 struct ofmirror
*mirror
= mirror_lookup(ofproto
, aux
);
3059 *packets
= *bytes
= UINT64_MAX
;
3065 *packets
= mirror
->packet_count
;
3066 *bytes
= mirror
->byte_count
;
3072 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
3074 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3075 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
3076 mac_learning_flush(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
3082 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
3084 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3085 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
3086 return bundle
&& bundle
->mirror_out
!= 0;
3090 forward_bpdu_changed(struct ofproto
*ofproto_
)
3092 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3093 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3097 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
3100 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3101 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
3102 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
3107 static struct ofport_dpif
*
3108 get_ofp_port(const struct ofproto_dpif
*ofproto
, uint16_t ofp_port
)
3110 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
3111 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
3114 static struct ofport_dpif
*
3115 get_odp_port(const struct ofproto_dpif
*ofproto
, uint32_t odp_port
)
3117 struct ofport_dpif
*port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
3118 return port
&& &ofproto
->up
== port
->up
.ofproto
? port
: NULL
;
3122 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
3123 struct ofproto_port
*ofproto_port
,
3124 struct dpif_port
*dpif_port
)
3126 ofproto_port
->name
= dpif_port
->name
;
3127 ofproto_port
->type
= dpif_port
->type
;
3128 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
3131 static struct ofport_dpif
*
3132 ofport_get_peer(const struct ofport_dpif
*ofport_dpif
)
3134 const struct ofproto_dpif
*ofproto
;
3137 peer
= netdev_vport_patch_peer(ofport_dpif
->up
.netdev
);
3142 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
3143 struct ofport
*ofport
;
3145 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, peer
);
3146 if (ofport
&& ofport
->ofproto
->ofproto_class
== &ofproto_dpif_class
) {
3147 return ofport_dpif_cast(ofport
);
3154 port_run_fast(struct ofport_dpif
*ofport
)
3156 if (ofport
->cfm
&& cfm_should_send_ccm(ofport
->cfm
)) {
3157 struct ofpbuf packet
;
3159 ofpbuf_init(&packet
, 0);
3160 cfm_compose_ccm(ofport
->cfm
, &packet
, ofport
->up
.pp
.hw_addr
);
3161 send_packet(ofport
, &packet
);
3162 ofpbuf_uninit(&packet
);
3165 if (ofport
->bfd
&& bfd_should_send_packet(ofport
->bfd
)) {
3166 struct ofpbuf packet
;
3168 ofpbuf_init(&packet
, 0);
3169 bfd_put_packet(ofport
->bfd
, &packet
, ofport
->up
.pp
.hw_addr
);
3170 send_packet(ofport
, &packet
);
3171 ofpbuf_uninit(&packet
);
3176 port_run(struct ofport_dpif
*ofport
)
3178 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
3179 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
3180 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
3182 ofport
->carrier_seq
= carrier_seq
;
3184 port_run_fast(ofport
);
3186 if (ofport
->tnl_port
3187 && tnl_port_reconfigure(&ofport
->up
, ofport
->odp_port
,
3188 &ofport
->tnl_port
)) {
3189 ofproto_dpif_cast(ofport
->up
.ofproto
)->backer
->need_revalidate
= true;
3193 int cfm_opup
= cfm_get_opup(ofport
->cfm
);
3195 cfm_run(ofport
->cfm
);
3196 enable
= enable
&& !cfm_get_fault(ofport
->cfm
);
3198 if (cfm_opup
>= 0) {
3199 enable
= enable
&& cfm_opup
;
3204 bfd_run(ofport
->bfd
);
3205 enable
= enable
&& bfd_forwarding(ofport
->bfd
);
3208 if (ofport
->bundle
) {
3209 enable
= enable
&& lacp_slave_may_enable(ofport
->bundle
->lacp
, ofport
);
3210 if (carrier_changed
) {
3211 lacp_slave_carrier_changed(ofport
->bundle
->lacp
, ofport
);
3215 if (ofport
->may_enable
!= enable
) {
3216 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3218 if (ofproto
->has_bundle_action
) {
3219 ofproto
->backer
->need_revalidate
= REV_PORT_TOGGLED
;
3223 ofport
->may_enable
= enable
;
3227 port_wait(struct ofport_dpif
*ofport
)
3230 cfm_wait(ofport
->cfm
);
3234 bfd_wait(ofport
->bfd
);
3239 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
3240 struct ofproto_port
*ofproto_port
)
3242 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3243 struct dpif_port dpif_port
;
3246 if (sset_contains(&ofproto
->ghost_ports
, devname
)) {
3247 const char *type
= netdev_get_type_from_name(devname
);
3249 /* We may be called before ofproto->up.port_by_name is populated with
3250 * the appropriate ofport. For this reason, we must get the name and
3251 * type from the netdev layer directly. */
3253 const struct ofport
*ofport
;
3255 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, devname
);
3256 ofproto_port
->ofp_port
= ofport
? ofport
->ofp_port
: OFPP_NONE
;
3257 ofproto_port
->name
= xstrdup(devname
);
3258 ofproto_port
->type
= xstrdup(type
);
3264 if (!sset_contains(&ofproto
->ports
, devname
)) {
3267 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
3268 devname
, &dpif_port
);
3270 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
3276 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
3278 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3279 const char *dp_port_name
= netdev_vport_get_dpif_port(netdev
);
3280 const char *devname
= netdev_get_name(netdev
);
3282 if (netdev_vport_is_patch(netdev
)) {
3283 sset_add(&ofproto
->ghost_ports
, netdev_get_name(netdev
));
3287 if (!dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
3288 uint32_t port_no
= UINT32_MAX
;
3291 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &port_no
);
3295 if (netdev_get_tunnel_config(netdev
)) {
3296 simap_put(&ofproto
->backer
->tnl_backers
, dp_port_name
, port_no
);
3300 if (netdev_get_tunnel_config(netdev
)) {
3301 sset_add(&ofproto
->ghost_ports
, devname
);
3303 sset_add(&ofproto
->ports
, devname
);
3309 port_del(struct ofproto
*ofproto_
, uint16_t ofp_port
)
3311 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3312 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
3319 sset_find_and_delete(&ofproto
->ghost_ports
,
3320 netdev_get_name(ofport
->up
.netdev
));
3321 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3322 if (!ofport
->tnl_port
) {
3323 error
= dpif_port_del(ofproto
->backer
->dpif
, ofport
->odp_port
);
3325 /* The caller is going to close ofport->up.netdev. If this is a
3326 * bonded port, then the bond is using that netdev, so remove it
3327 * from the bond. The client will need to reconfigure everything
3328 * after deleting ports, so then the slave will get re-added. */
3329 bundle_remove(&ofport
->up
);
3336 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
3338 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3343 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
3345 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
3346 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3348 /* ofproto->stats.tx_packets represents packets that we created
3349 * internally and sent to some port (e.g. packets sent with
3350 * send_packet()). Account for them as if they had come from
3351 * OFPP_LOCAL and got forwarded. */
3353 if (stats
->rx_packets
!= UINT64_MAX
) {
3354 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
3357 if (stats
->rx_bytes
!= UINT64_MAX
) {
3358 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
3361 /* ofproto->stats.rx_packets represents packets that were received on
3362 * some port and we processed internally and dropped (e.g. STP).
3363 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3365 if (stats
->tx_packets
!= UINT64_MAX
) {
3366 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
3369 if (stats
->tx_bytes
!= UINT64_MAX
) {
3370 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
3377 struct port_dump_state
{
3382 struct ofproto_port port
;
3387 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
3389 *statep
= xzalloc(sizeof(struct port_dump_state
));
3394 port_dump_next(const struct ofproto
*ofproto_
, void *state_
,
3395 struct ofproto_port
*port
)
3397 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3398 struct port_dump_state
*state
= state_
;
3399 const struct sset
*sset
;
3400 struct sset_node
*node
;
3402 if (state
->has_port
) {
3403 ofproto_port_destroy(&state
->port
);
3404 state
->has_port
= false;
3406 sset
= state
->ghost
? &ofproto
->ghost_ports
: &ofproto
->ports
;
3407 while ((node
= sset_at_position(sset
, &state
->bucket
, &state
->offset
))) {
3410 error
= port_query_by_name(ofproto_
, node
->name
, &state
->port
);
3412 *port
= state
->port
;
3413 state
->has_port
= true;
3415 } else if (error
!= ENODEV
) {
3420 if (!state
->ghost
) {
3421 state
->ghost
= true;
3424 return port_dump_next(ofproto_
, state_
, port
);
3431 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
3433 struct port_dump_state
*state
= state_
;
3435 if (state
->has_port
) {
3436 ofproto_port_destroy(&state
->port
);
3443 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
3445 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3447 if (ofproto
->port_poll_errno
) {
3448 int error
= ofproto
->port_poll_errno
;
3449 ofproto
->port_poll_errno
= 0;
3453 if (sset_is_empty(&ofproto
->port_poll_set
)) {
3457 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
3462 port_poll_wait(const struct ofproto
*ofproto_
)
3464 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3465 dpif_port_poll_wait(ofproto
->backer
->dpif
);
3469 port_is_lacp_current(const struct ofport
*ofport_
)
3471 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3472 return (ofport
->bundle
&& ofport
->bundle
->lacp
3473 ? lacp_slave_is_current(ofport
->bundle
->lacp
, ofport
)
3477 /* Upcall handling. */
3479 /* Flow miss batching.
3481 * Some dpifs implement operations faster when you hand them off in a batch.
3482 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3483 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3484 * more packets, plus possibly installing the flow in the dpif.
3486 * So far we only batch the operations that affect flow setup time the most.
3487 * It's possible to batch more than that, but the benefit might be minimal. */
3489 struct hmap_node hmap_node
;
3490 struct ofproto_dpif
*ofproto
;
3492 enum odp_key_fitness key_fitness
;
3493 const struct nlattr
*key
;
3495 struct initial_vals initial_vals
;
3496 struct list packets
;
3497 enum dpif_upcall_type upcall_type
;
3498 uint32_t odp_in_port
;
3501 struct flow_miss_op
{
3502 struct dpif_op dpif_op
;
3503 void *garbage
; /* Pointer to pass to free(), NULL if none. */
3504 uint64_t stub
[1024 / 8]; /* Temporary buffer. */
3507 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3508 * OpenFlow controller as necessary according to their individual
3509 * configurations. */
3511 send_packet_in_miss(struct ofproto_dpif
*ofproto
, const struct ofpbuf
*packet
,
3512 const struct flow
*flow
)
3514 struct ofputil_packet_in pin
;
3516 pin
.packet
= packet
->data
;
3517 pin
.packet_len
= packet
->size
;
3518 pin
.reason
= OFPR_NO_MATCH
;
3519 pin
.controller_id
= 0;
3524 pin
.send_len
= 0; /* not used for flow table misses */
3526 flow_get_metadata(flow
, &pin
.fmd
);
3528 connmgr_send_packet_in(ofproto
->up
.connmgr
, &pin
);
3531 static enum slow_path_reason
3532 process_special(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
3533 const struct ofport_dpif
*ofport
, const struct ofpbuf
*packet
)
3537 } else if (ofport
->cfm
&& cfm_should_process_flow(ofport
->cfm
, flow
)) {
3539 cfm_process_heartbeat(ofport
->cfm
, packet
);
3542 } else if (ofport
->bfd
&& bfd_should_process_flow(flow
)) {
3544 bfd_process_packet(ofport
->bfd
, flow
, packet
);
3547 } else if (ofport
->bundle
&& ofport
->bundle
->lacp
3548 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
3550 lacp_process_packet(ofport
->bundle
->lacp
, ofport
, packet
);
3553 } else if (ofproto
->stp
&& stp_should_process_flow(flow
)) {
3555 stp_process_packet(ofport
, packet
);
3563 static struct flow_miss
*
3564 flow_miss_find(struct hmap
*todo
, const struct ofproto_dpif
*ofproto
,
3565 const struct flow
*flow
, uint32_t hash
)
3567 struct flow_miss
*miss
;
3569 HMAP_FOR_EACH_WITH_HASH (miss
, hmap_node
, hash
, todo
) {
3570 if (miss
->ofproto
== ofproto
&& flow_equal(&miss
->flow
, flow
)) {
3578 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3579 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3580 * 'miss' is associated with a subfacet the caller must also initialize the
3581 * returned op->subfacet, and if anything needs to be freed after processing
3582 * the op, the caller must initialize op->garbage also. */
3584 init_flow_miss_execute_op(struct flow_miss
*miss
, struct ofpbuf
*packet
,
3585 struct flow_miss_op
*op
)
3587 if (miss
->flow
.vlan_tci
!= miss
->initial_vals
.vlan_tci
) {
3588 /* This packet was received on a VLAN splinter port. We
3589 * added a VLAN to the packet to make the packet resemble
3590 * the flow, but the actions were composed assuming that
3591 * the packet contained no VLAN. So, we must remove the
3592 * VLAN header from the packet before trying to execute the
3594 eth_pop_vlan(packet
);
3598 op
->dpif_op
.type
= DPIF_OP_EXECUTE
;
3599 op
->dpif_op
.u
.execute
.key
= miss
->key
;
3600 op
->dpif_op
.u
.execute
.key_len
= miss
->key_len
;
3601 op
->dpif_op
.u
.execute
.packet
= packet
;
3604 /* Helper for handle_flow_miss_without_facet() and
3605 * handle_flow_miss_with_facet(). */
3607 handle_flow_miss_common(struct rule_dpif
*rule
,
3608 struct ofpbuf
*packet
, const struct flow
*flow
)
3610 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3612 ofproto
->n_matches
++;
3614 if (rule
->up
.cr
.priority
== FAIL_OPEN_PRIORITY
) {
3616 * Extra-special case for fail-open mode.
3618 * We are in fail-open mode and the packet matched the fail-open
3619 * rule, but we are connected to a controller too. We should send
3620 * the packet up to the controller in the hope that it will try to
3621 * set up a flow and thereby allow us to exit fail-open.
3623 * See the top-level comment in fail-open.c for more information.
3625 send_packet_in_miss(ofproto
, packet
, flow
);
3629 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3630 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3631 * installing a datapath flow. The answer is usually "yes" (a return value of
3632 * true). However, for short flows the cost of bookkeeping is much higher than
3633 * the benefits, so when the datapath holds a large number of flows we impose
3634 * some heuristics to decide which flows are likely to be worth tracking. */
3636 flow_miss_should_make_facet(struct ofproto_dpif
*ofproto
,
3637 struct flow_miss
*miss
, uint32_t hash
)
3639 if (!ofproto
->governor
) {
3642 n_subfacets
= hmap_count(&ofproto
->subfacets
);
3643 if (n_subfacets
* 2 <= ofproto
->up
.flow_eviction_threshold
) {
3647 ofproto
->governor
= governor_create(ofproto
->up
.name
);
3650 return governor_should_install_flow(ofproto
->governor
, hash
,
3651 list_size(&miss
->packets
));
3654 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3655 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3656 * increment '*n_ops'. */
3658 handle_flow_miss_without_facet(struct flow_miss
*miss
,
3659 struct rule_dpif
*rule
,
3660 struct flow_miss_op
*ops
, size_t *n_ops
)
3662 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3663 long long int now
= time_msec();
3664 struct action_xlate_ctx ctx
;
3665 struct ofpbuf
*packet
;
3667 LIST_FOR_EACH (packet
, list_node
, &miss
->packets
) {
3668 struct flow_miss_op
*op
= &ops
[*n_ops
];
3669 struct dpif_flow_stats stats
;
3670 struct ofpbuf odp_actions
;
3672 COVERAGE_INC(facet_suppress
);
3674 ofpbuf_use_stub(&odp_actions
, op
->stub
, sizeof op
->stub
);
3676 dpif_flow_stats_extract(&miss
->flow
, packet
, now
, &stats
);
3677 rule_credit_stats(rule
, &stats
);
3679 action_xlate_ctx_init(&ctx
, ofproto
, &miss
->flow
,
3680 &miss
->initial_vals
, rule
, 0, packet
);
3681 ctx
.resubmit_stats
= &stats
;
3682 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
3685 if (odp_actions
.size
) {
3686 struct dpif_execute
*execute
= &op
->dpif_op
.u
.execute
;
3688 init_flow_miss_execute_op(miss
, packet
, op
);
3689 execute
->actions
= odp_actions
.data
;
3690 execute
->actions_len
= odp_actions
.size
;
3691 op
->garbage
= ofpbuf_get_uninit_pointer(&odp_actions
);
3695 ofpbuf_uninit(&odp_actions
);
3700 /* Handles 'miss', which matches 'facet'. May add any required datapath
3701 * operations to 'ops', incrementing '*n_ops' for each new op.
3703 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3704 * This is really important only for new facets: if we just called time_msec()
3705 * here, then the new subfacet or its packets could look (occasionally) as
3706 * though it was used some time after the facet was used. That can make a
3707 * one-packet flow look like it has a nonzero duration, which looks odd in
3708 * e.g. NetFlow statistics. */
3710 handle_flow_miss_with_facet(struct flow_miss
*miss
, struct facet
*facet
,
3712 struct flow_miss_op
*ops
, size_t *n_ops
)
3714 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
3715 enum subfacet_path want_path
;
3716 struct subfacet
*subfacet
;
3717 struct ofpbuf
*packet
;
3719 subfacet
= subfacet_create(facet
, miss
, now
);
3721 LIST_FOR_EACH (packet
, list_node
, &miss
->packets
) {
3722 struct flow_miss_op
*op
= &ops
[*n_ops
];
3723 struct dpif_flow_stats stats
;
3724 struct ofpbuf odp_actions
;
3726 handle_flow_miss_common(facet
->rule
, packet
, &miss
->flow
);
3728 ofpbuf_use_stub(&odp_actions
, op
->stub
, sizeof op
->stub
);
3729 if (!subfacet
->actions
|| subfacet
->slow
) {
3730 subfacet_make_actions(subfacet
, packet
, &odp_actions
);
3733 dpif_flow_stats_extract(&facet
->flow
, packet
, now
, &stats
);
3734 subfacet_update_stats(subfacet
, &stats
);
3736 if (subfacet
->actions_len
) {
3737 struct dpif_execute
*execute
= &op
->dpif_op
.u
.execute
;
3739 init_flow_miss_execute_op(miss
, packet
, op
);
3740 if (!subfacet
->slow
) {
3741 execute
->actions
= subfacet
->actions
;
3742 execute
->actions_len
= subfacet
->actions_len
;
3743 ofpbuf_uninit(&odp_actions
);
3745 execute
->actions
= odp_actions
.data
;
3746 execute
->actions_len
= odp_actions
.size
;
3747 op
->garbage
= ofpbuf_get_uninit_pointer(&odp_actions
);
3752 ofpbuf_uninit(&odp_actions
);
3756 want_path
= subfacet_want_path(subfacet
->slow
);
3757 if (miss
->upcall_type
== DPIF_UC_MISS
|| subfacet
->path
!= want_path
) {
3758 struct flow_miss_op
*op
= &ops
[(*n_ops
)++];
3759 struct dpif_flow_put
*put
= &op
->dpif_op
.u
.flow_put
;
3761 subfacet
->path
= want_path
;
3764 op
->dpif_op
.type
= DPIF_OP_FLOW_PUT
;
3765 put
->flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
3766 put
->key
= miss
->key
;
3767 put
->key_len
= miss
->key_len
;
3768 if (want_path
== SF_FAST_PATH
) {
3769 put
->actions
= subfacet
->actions
;
3770 put
->actions_len
= subfacet
->actions_len
;
3772 compose_slow_path(ofproto
, &facet
->flow
, subfacet
->slow
,
3773 op
->stub
, sizeof op
->stub
,
3774 &put
->actions
, &put
->actions_len
);
3780 /* Handles flow miss 'miss'. May add any required datapath operations
3781 * to 'ops', incrementing '*n_ops' for each new op. */
3783 handle_flow_miss(struct flow_miss
*miss
, struct flow_miss_op
*ops
,
3786 struct ofproto_dpif
*ofproto
= miss
->ofproto
;
3787 struct facet
*facet
;
3791 /* The caller must ensure that miss->hmap_node.hash contains
3792 * flow_hash(miss->flow, 0). */
3793 hash
= miss
->hmap_node
.hash
;
3795 facet
= facet_lookup_valid(ofproto
, &miss
->flow
, hash
);
3797 struct rule_dpif
*rule
= rule_dpif_lookup(ofproto
, &miss
->flow
);
3799 if (!flow_miss_should_make_facet(ofproto
, miss
, hash
)) {
3800 handle_flow_miss_without_facet(miss
, rule
, ops
, n_ops
);
3804 facet
= facet_create(rule
, &miss
->flow
, hash
);
3809 handle_flow_miss_with_facet(miss
, facet
, now
, ops
, n_ops
);
3812 static struct drop_key
*
3813 drop_key_lookup(const struct dpif_backer
*backer
, const struct nlattr
*key
,
3816 struct drop_key
*drop_key
;
3818 HMAP_FOR_EACH_WITH_HASH (drop_key
, hmap_node
, hash_bytes(key
, key_len
, 0),
3819 &backer
->drop_keys
) {
3820 if (drop_key
->key_len
== key_len
3821 && !memcmp(drop_key
->key
, key
, key_len
)) {
3829 drop_key_clear(struct dpif_backer
*backer
)
3831 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 15);
3832 struct drop_key
*drop_key
, *next
;
3834 HMAP_FOR_EACH_SAFE (drop_key
, next
, hmap_node
, &backer
->drop_keys
) {
3837 error
= dpif_flow_del(backer
->dpif
, drop_key
->key
, drop_key
->key_len
,
3839 if (error
&& !VLOG_DROP_WARN(&rl
)) {
3840 struct ds ds
= DS_EMPTY_INITIALIZER
;
3841 odp_flow_key_format(drop_key
->key
, drop_key
->key_len
, &ds
);
3842 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error
),
3847 hmap_remove(&backer
->drop_keys
, &drop_key
->hmap_node
);
3848 free(drop_key
->key
);
3853 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3854 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3855 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3856 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3857 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3858 * 'packet' ingressed.
3860 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3861 * 'flow''s in_port to OFPP_NONE.
3863 * This function does post-processing on data returned from
3864 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3865 * of the upcall processing logic. In particular, if the extracted in_port is
3866 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3867 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3868 * a VLAN header onto 'packet' (if it is nonnull).
3870 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3871 * to the VLAN TCI with which the packet was really received, that is, the
3872 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3873 * the value returned in flow->vlan_tci only for packets received on
3876 * Similarly, this function also includes some logic to help with tunnels. It
3877 * may modify 'flow' as necessary to make the tunneling implementation
3878 * transparent to the upcall processing logic.
3880 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3881 * or some other positive errno if there are other problems. */
3883 ofproto_receive(const struct dpif_backer
*backer
, struct ofpbuf
*packet
,
3884 const struct nlattr
*key
, size_t key_len
,
3885 struct flow
*flow
, enum odp_key_fitness
*fitnessp
,
3886 struct ofproto_dpif
**ofproto
, uint32_t *odp_in_port
,
3887 struct initial_vals
*initial_vals
)
3889 const struct ofport_dpif
*port
;
3890 enum odp_key_fitness fitness
;
3893 fitness
= odp_flow_key_to_flow(key
, key_len
, flow
);
3894 if (fitness
== ODP_FIT_ERROR
) {
3900 initial_vals
->vlan_tci
= flow
->vlan_tci
;
3904 *odp_in_port
= flow
->in_port
;
3907 port
= (tnl_port_should_receive(flow
)
3908 ? ofport_dpif_cast(tnl_port_receive(flow
))
3909 : odp_port_to_ofport(backer
, flow
->in_port
));
3910 flow
->in_port
= port
? port
->up
.ofp_port
: OFPP_NONE
;
3915 /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port
3916 * it's theoretically possible that we'll receive an ofport belonging to an
3917 * entirely different datapath. In practice, this can't happen because no
3918 * platforms has two separate datapaths which each support tunneling. */
3919 ovs_assert(ofproto_dpif_cast(port
->up
.ofproto
)->backer
== backer
);
3921 if (vsp_adjust_flow(ofproto_dpif_cast(port
->up
.ofproto
), flow
)) {
3923 /* Make the packet resemble the flow, so that it gets sent to
3924 * an OpenFlow controller properly, so that it looks correct
3925 * for sFlow, and so that flow_extract() will get the correct
3926 * vlan_tci if it is called on 'packet'.
3928 * The allocated space inside 'packet' probably also contains
3929 * 'key', that is, both 'packet' and 'key' are probably part of
3930 * a struct dpif_upcall (see the large comment on that
3931 * structure definition), so pushing data on 'packet' is in
3932 * general not a good idea since it could overwrite 'key' or
3933 * free it as a side effect. However, it's OK in this special
3934 * case because we know that 'packet' is inside a Netlink
3935 * attribute: pushing 4 bytes will just overwrite the 4-byte
3936 * "struct nlattr", which is fine since we don't need that
3937 * header anymore. */
3938 eth_push_vlan(packet
, flow
->vlan_tci
);
3940 /* We can't reproduce 'key' from 'flow'. */
3941 fitness
= fitness
== ODP_FIT_PERFECT
? ODP_FIT_TOO_MUCH
: fitness
;
3946 *ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
3951 *fitnessp
= fitness
;
3957 handle_miss_upcalls(struct dpif_backer
*backer
, struct dpif_upcall
*upcalls
,
3960 struct dpif_upcall
*upcall
;
3961 struct flow_miss
*miss
;
3962 struct flow_miss misses
[FLOW_MISS_MAX_BATCH
];
3963 struct flow_miss_op flow_miss_ops
[FLOW_MISS_MAX_BATCH
* 2];
3964 struct dpif_op
*dpif_ops
[FLOW_MISS_MAX_BATCH
* 2];
3974 /* Construct the to-do list.
3976 * This just amounts to extracting the flow from each packet and sticking
3977 * the packets that have the same flow in the same "flow_miss" structure so
3978 * that we can process them together. */
3981 for (upcall
= upcalls
; upcall
< &upcalls
[n_upcalls
]; upcall
++) {
3982 struct flow_miss
*miss
= &misses
[n_misses
];
3983 struct flow_miss
*existing_miss
;
3984 struct ofproto_dpif
*ofproto
;
3985 uint32_t odp_in_port
;
3990 error
= ofproto_receive(backer
, upcall
->packet
, upcall
->key
,
3991 upcall
->key_len
, &flow
, &miss
->key_fitness
,
3992 &ofproto
, &odp_in_port
, &miss
->initial_vals
);
3993 if (error
== ENODEV
) {
3994 struct drop_key
*drop_key
;
3996 /* Received packet on port for which we couldn't associate
3997 * an ofproto. This can happen if a port is removed while
3998 * traffic is being received. Print a rate-limited message
3999 * in case it happens frequently. Install a drop flow so
4000 * that future packets of the flow are inexpensively dropped
4002 VLOG_INFO_RL(&rl
, "received packet on unassociated port %"PRIu32
,
4005 drop_key
= drop_key_lookup(backer
, upcall
->key
, upcall
->key_len
);
4007 drop_key
= xmalloc(sizeof *drop_key
);
4008 drop_key
->key
= xmemdup(upcall
->key
, upcall
->key_len
);
4009 drop_key
->key_len
= upcall
->key_len
;
4011 hmap_insert(&backer
->drop_keys
, &drop_key
->hmap_node
,
4012 hash_bytes(drop_key
->key
, drop_key
->key_len
, 0));
4013 dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
4014 drop_key
->key
, drop_key
->key_len
, NULL
, 0, NULL
);
4022 ofproto
->n_missed
++;
4023 flow_extract(upcall
->packet
, flow
.skb_priority
, flow
.skb_mark
,
4024 &flow
.tunnel
, flow
.in_port
, &miss
->flow
);
4026 /* Add other packets to a to-do list. */
4027 hash
= flow_hash(&miss
->flow
, 0);
4028 existing_miss
= flow_miss_find(&todo
, ofproto
, &miss
->flow
, hash
);
4029 if (!existing_miss
) {
4030 hmap_insert(&todo
, &miss
->hmap_node
, hash
);
4031 miss
->ofproto
= ofproto
;
4032 miss
->key
= upcall
->key
;
4033 miss
->key_len
= upcall
->key_len
;
4034 miss
->upcall_type
= upcall
->type
;
4035 miss
->odp_in_port
= odp_in_port
;
4036 list_init(&miss
->packets
);
4040 miss
= existing_miss
;
4042 list_push_back(&miss
->packets
, &upcall
->packet
->list_node
);
4045 /* Process each element in the to-do list, constructing the set of
4046 * operations to batch. */
4048 HMAP_FOR_EACH (miss
, hmap_node
, &todo
) {
4049 handle_flow_miss(miss
, flow_miss_ops
, &n_ops
);
4051 ovs_assert(n_ops
<= ARRAY_SIZE(flow_miss_ops
));
4053 /* Execute batch. */
4054 for (i
= 0; i
< n_ops
; i
++) {
4055 dpif_ops
[i
] = &flow_miss_ops
[i
].dpif_op
;
4057 dpif_operate(backer
->dpif
, dpif_ops
, n_ops
);
4060 for (i
= 0; i
< n_ops
; i
++) {
4061 free(flow_miss_ops
[i
].garbage
);
4063 hmap_destroy(&todo
);
4066 static enum { SFLOW_UPCALL
, MISS_UPCALL
, BAD_UPCALL
, FLOW_SAMPLE_UPCALL
,
4068 classify_upcall(const struct dpif_upcall
*upcall
)
4070 size_t userdata_len
;
4071 union user_action_cookie cookie
;
4073 /* First look at the upcall type. */
4074 switch (upcall
->type
) {
4075 case DPIF_UC_ACTION
:
4081 case DPIF_N_UC_TYPES
:
4083 VLOG_WARN_RL(&rl
, "upcall has unexpected type %"PRIu32
, upcall
->type
);
4087 /* "action" upcalls need a closer look. */
4088 if (!upcall
->userdata
) {
4089 VLOG_WARN_RL(&rl
, "action upcall missing cookie");
4092 userdata_len
= nl_attr_get_size(upcall
->userdata
);
4093 if (userdata_len
< sizeof cookie
.type
4094 || userdata_len
> sizeof cookie
) {
4095 VLOG_WARN_RL(&rl
, "action upcall cookie has unexpected size %zu",
4099 memset(&cookie
, 0, sizeof cookie
);
4100 memcpy(&cookie
, nl_attr_get(upcall
->userdata
), userdata_len
);
4101 if (userdata_len
== sizeof cookie
.sflow
4102 && cookie
.type
== USER_ACTION_COOKIE_SFLOW
) {
4103 return SFLOW_UPCALL
;
4104 } else if (userdata_len
== sizeof cookie
.slow_path
4105 && cookie
.type
== USER_ACTION_COOKIE_SLOW_PATH
) {
4107 } else if (userdata_len
== sizeof cookie
.flow_sample
4108 && cookie
.type
== USER_ACTION_COOKIE_FLOW_SAMPLE
) {
4109 return FLOW_SAMPLE_UPCALL
;
4110 } else if (userdata_len
== sizeof cookie
.ipfix
4111 && cookie
.type
== USER_ACTION_COOKIE_IPFIX
) {
4112 return IPFIX_UPCALL
;
4114 VLOG_WARN_RL(&rl
, "invalid user cookie of type %"PRIu16
4115 " and size %zu", cookie
.type
, userdata_len
);
4121 handle_sflow_upcall(struct dpif_backer
*backer
,
4122 const struct dpif_upcall
*upcall
)
4124 struct ofproto_dpif
*ofproto
;
4125 union user_action_cookie cookie
;
4127 uint32_t odp_in_port
;
4129 if (ofproto_receive(backer
, upcall
->packet
, upcall
->key
, upcall
->key_len
,
4130 &flow
, NULL
, &ofproto
, &odp_in_port
, NULL
)
4131 || !ofproto
->sflow
) {
4135 memset(&cookie
, 0, sizeof cookie
);
4136 memcpy(&cookie
, nl_attr_get(upcall
->userdata
), sizeof cookie
.sflow
);
4137 dpif_sflow_received(ofproto
->sflow
, upcall
->packet
, &flow
,
4138 odp_in_port
, &cookie
);
4142 handle_flow_sample_upcall(struct dpif_backer
*backer
,
4143 const struct dpif_upcall
*upcall
)
4145 struct ofproto_dpif
*ofproto
;
4146 union user_action_cookie cookie
;
4149 if (ofproto_receive(backer
, upcall
->packet
, upcall
->key
, upcall
->key_len
,
4150 &flow
, NULL
, &ofproto
, NULL
, NULL
)
4151 || !ofproto
->ipfix
) {
4155 memset(&cookie
, 0, sizeof cookie
);
4156 memcpy(&cookie
, nl_attr_get(upcall
->userdata
), sizeof cookie
.flow_sample
);
4158 /* The flow reflects exactly the contents of the packet. Sample
4159 * the packet using it. */
4160 dpif_ipfix_flow_sample(ofproto
->ipfix
, upcall
->packet
, &flow
,
4161 cookie
.flow_sample
.collector_set_id
,
4162 cookie
.flow_sample
.probability
,
4163 cookie
.flow_sample
.obs_domain_id
,
4164 cookie
.flow_sample
.obs_point_id
);
4168 handle_ipfix_upcall(struct dpif_backer
*backer
,
4169 const struct dpif_upcall
*upcall
)
4171 struct ofproto_dpif
*ofproto
;
4174 if (ofproto_receive(backer
, upcall
->packet
, upcall
->key
, upcall
->key_len
,
4175 &flow
, NULL
, &ofproto
, NULL
, NULL
)
4176 || !ofproto
->ipfix
) {
4180 /* The flow reflects exactly the contents of the packet. Sample
4181 * the packet using it. */
4182 dpif_ipfix_bridge_sample(ofproto
->ipfix
, upcall
->packet
, &flow
);
4186 handle_upcalls(struct dpif_backer
*backer
, unsigned int max_batch
)
4188 struct dpif_upcall misses
[FLOW_MISS_MAX_BATCH
];
4189 struct ofpbuf miss_bufs
[FLOW_MISS_MAX_BATCH
];
4190 uint64_t miss_buf_stubs
[FLOW_MISS_MAX_BATCH
][4096 / 8];
4195 ovs_assert(max_batch
<= FLOW_MISS_MAX_BATCH
);
4198 for (n_processed
= 0; n_processed
< max_batch
; n_processed
++) {
4199 struct dpif_upcall
*upcall
= &misses
[n_misses
];
4200 struct ofpbuf
*buf
= &miss_bufs
[n_misses
];
4203 ofpbuf_use_stub(buf
, miss_buf_stubs
[n_misses
],
4204 sizeof miss_buf_stubs
[n_misses
]);
4205 error
= dpif_recv(backer
->dpif
, upcall
, buf
);
4211 switch (classify_upcall(upcall
)) {
4213 /* Handle it later. */
4218 handle_sflow_upcall(backer
, upcall
);
4222 case FLOW_SAMPLE_UPCALL
:
4223 handle_flow_sample_upcall(backer
, upcall
);
4228 handle_ipfix_upcall(backer
, upcall
);
4238 /* Handle deferred MISS_UPCALL processing. */
4239 handle_miss_upcalls(backer
, misses
, n_misses
);
4240 for (i
= 0; i
< n_misses
; i
++) {
4241 ofpbuf_uninit(&miss_bufs
[i
]);
4247 /* Flow expiration. */
4249 static int subfacet_max_idle(const struct ofproto_dpif
*);
4250 static void update_stats(struct dpif_backer
*);
4251 static void rule_expire(struct rule_dpif
*);
4252 static void expire_subfacets(struct ofproto_dpif
*, int dp_max_idle
);
4254 /* This function is called periodically by run(). Its job is to collect
4255 * updates for the flows that have been installed into the datapath, most
4256 * importantly when they last were used, and then use that information to
4257 * expire flows that have not been used recently.
4259 * Returns the number of milliseconds after which it should be called again. */
4261 expire(struct dpif_backer
*backer
)
4263 struct ofproto_dpif
*ofproto
;
4264 int max_idle
= INT32_MAX
;
4266 /* Periodically clear out the drop keys in an effort to keep them
4267 * relatively few. */
4268 drop_key_clear(backer
);
4270 /* Update stats for each flow in the backer. */
4271 update_stats(backer
);
4273 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
4274 struct rule
*rule
, *next_rule
;
4277 if (ofproto
->backer
!= backer
) {
4281 /* Keep track of the max number of flows per ofproto_dpif. */
4282 update_max_subfacet_count(ofproto
);
4284 /* Expire subfacets that have been idle too long. */
4285 dp_max_idle
= subfacet_max_idle(ofproto
);
4286 expire_subfacets(ofproto
, dp_max_idle
);
4288 max_idle
= MIN(max_idle
, dp_max_idle
);
4290 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4292 LIST_FOR_EACH_SAFE (rule
, next_rule
, expirable
,
4293 &ofproto
->up
.expirable
) {
4294 rule_expire(rule_dpif_cast(rule
));
4297 /* All outstanding data in existing flows has been accounted, so it's a
4298 * good time to do bond rebalancing. */
4299 if (ofproto
->has_bonded_bundles
) {
4300 struct ofbundle
*bundle
;
4302 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
4304 bond_rebalance(bundle
->bond
, &backer
->revalidate_set
);
4310 return MIN(max_idle
, 1000);
4313 /* Updates flow table statistics given that the datapath just reported 'stats'
4314 * as 'subfacet''s statistics. */
4316 update_subfacet_stats(struct subfacet
*subfacet
,
4317 const struct dpif_flow_stats
*stats
)
4319 struct facet
*facet
= subfacet
->facet
;
4321 if (stats
->n_packets
>= subfacet
->dp_packet_count
) {
4322 uint64_t extra
= stats
->n_packets
- subfacet
->dp_packet_count
;
4323 facet
->packet_count
+= extra
;
4325 VLOG_WARN_RL(&rl
, "unexpected packet count from the datapath");
4328 if (stats
->n_bytes
>= subfacet
->dp_byte_count
) {
4329 facet
->byte_count
+= stats
->n_bytes
- subfacet
->dp_byte_count
;
4331 VLOG_WARN_RL(&rl
, "unexpected byte count from datapath");
4334 subfacet
->dp_packet_count
= stats
->n_packets
;
4335 subfacet
->dp_byte_count
= stats
->n_bytes
;
4337 facet
->tcp_flags
|= stats
->tcp_flags
;
4339 subfacet_update_time(subfacet
, stats
->used
);
4340 if (facet
->accounted_bytes
< facet
->byte_count
) {
4342 facet_account(facet
);
4343 facet
->accounted_bytes
= facet
->byte_count
;
4347 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4348 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4350 delete_unexpected_flow(struct ofproto_dpif
*ofproto
,
4351 const struct nlattr
*key
, size_t key_len
)
4353 if (!VLOG_DROP_WARN(&rl
)) {
4357 odp_flow_key_format(key
, key_len
, &s
);
4358 VLOG_WARN("unexpected flow on %s: %s", ofproto
->up
.name
, ds_cstr(&s
));
4362 COVERAGE_INC(facet_unexpected
);
4363 dpif_flow_del(ofproto
->backer
->dpif
, key
, key_len
, NULL
);
4366 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4368 * This function also pushes statistics updates to rules which each facet
4369 * resubmits into. Generally these statistics will be accurate. However, if a
4370 * facet changes the rule it resubmits into at some time in between
4371 * update_stats() runs, it is possible that statistics accrued to the
4372 * old rule will be incorrectly attributed to the new rule. This could be
4373 * avoided by calling update_stats() whenever rules are created or
4374 * deleted. However, the performance impact of making so many calls to the
4375 * datapath do not justify the benefit of having perfectly accurate statistics.
4377 * In addition, this function maintains per ofproto flow hit counts. The patch
4378 * port is not treated specially. e.g. A packet ingress from br0 patched into
4379 * br1 will increase the hit count of br0 by 1, however, does not affect
4380 * the hit or miss counts of br1.
4383 update_stats(struct dpif_backer
*backer
)
4385 const struct dpif_flow_stats
*stats
;
4386 struct dpif_flow_dump dump
;
4387 const struct nlattr
*key
;
4388 struct ofproto_dpif
*ofproto
;
4391 dpif_flow_dump_start(&dump
, backer
->dpif
);
4392 while (dpif_flow_dump_next(&dump
, &key
, &key_len
, NULL
, NULL
, &stats
)) {
4394 struct subfacet
*subfacet
;
4395 struct ofport_dpif
*ofport
;
4398 if (ofproto_receive(backer
, NULL
, key
, key_len
, &flow
, NULL
, &ofproto
,
4403 ofproto
->total_subfacet_count
+= hmap_count(&ofproto
->subfacets
);
4404 ofproto
->n_update_stats
++;
4406 ofport
= get_ofp_port(ofproto
, flow
.in_port
);
4407 if (ofport
&& ofport
->tnl_port
) {
4408 netdev_vport_inc_rx(ofport
->up
.netdev
, stats
);
4411 key_hash
= odp_flow_key_hash(key
, key_len
);
4412 subfacet
= subfacet_find(ofproto
, key
, key_len
, key_hash
);
4413 switch (subfacet
? subfacet
->path
: SF_NOT_INSTALLED
) {
4415 /* Update ofproto_dpif's hit count. */
4416 if (stats
->n_packets
> subfacet
->dp_packet_count
) {
4417 uint64_t delta
= stats
->n_packets
- subfacet
->dp_packet_count
;
4418 dpif_stats_update_hit_count(ofproto
, delta
);
4421 update_subfacet_stats(subfacet
, stats
);
4425 /* Stats are updated per-packet. */
4428 case SF_NOT_INSTALLED
:
4430 delete_unexpected_flow(ofproto
, key
, key_len
);
4435 dpif_flow_dump_done(&dump
);
4437 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
4438 update_moving_averages(ofproto
);
4443 /* Calculates and returns the number of milliseconds of idle time after which
4444 * subfacets should expire from the datapath. When a subfacet expires, we fold
4445 * its statistics into its facet, and when a facet's last subfacet expires, we
4446 * fold its statistic into its rule. */
4448 subfacet_max_idle(const struct ofproto_dpif
*ofproto
)
4451 * Idle time histogram.
4453 * Most of the time a switch has a relatively small number of subfacets.
4454 * When this is the case we might as well keep statistics for all of them
4455 * in userspace and to cache them in the kernel datapath for performance as
4458 * As the number of subfacets increases, the memory required to maintain
4459 * statistics about them in userspace and in the kernel becomes
4460 * significant. However, with a large number of subfacets it is likely
4461 * that only a few of them are "heavy hitters" that consume a large amount
4462 * of bandwidth. At this point, only heavy hitters are worth caching in
4463 * the kernel and maintaining in userspaces; other subfacets we can
4466 * The technique used to compute the idle time is to build a histogram with
4467 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4468 * that is installed in the kernel gets dropped in the appropriate bucket.
4469 * After the histogram has been built, we compute the cutoff so that only
4470 * the most-recently-used 1% of subfacets (but at least
4471 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4472 * the most-recently-used bucket of subfacets is kept, so actually an
4473 * arbitrary number of subfacets can be kept in any given expiration run
4474 * (though the next run will delete most of those unless they receive
4477 * This requires a second pass through the subfacets, in addition to the
4478 * pass made by update_stats(), because the former function never looks at
4479 * uninstallable subfacets.
4481 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4482 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4483 int buckets
[N_BUCKETS
] = { 0 };
4484 int total
, subtotal
, bucket
;
4485 struct subfacet
*subfacet
;
4489 total
= hmap_count(&ofproto
->subfacets
);
4490 if (total
<= ofproto
->up
.flow_eviction_threshold
) {
4491 return N_BUCKETS
* BUCKET_WIDTH
;
4494 /* Build histogram. */
4496 HMAP_FOR_EACH (subfacet
, hmap_node
, &ofproto
->subfacets
) {
4497 long long int idle
= now
- subfacet
->used
;
4498 int bucket
= (idle
<= 0 ? 0
4499 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4500 : (unsigned int) idle
/ BUCKET_WIDTH
);
4504 /* Find the first bucket whose flows should be expired. */
4505 subtotal
= bucket
= 0;
4507 subtotal
+= buckets
[bucket
++];
4508 } while (bucket
< N_BUCKETS
&&
4509 subtotal
< MAX(ofproto
->up
.flow_eviction_threshold
, total
/ 100));
4511 if (VLOG_IS_DBG_ENABLED()) {
4515 ds_put_cstr(&s
, "keep");
4516 for (i
= 0; i
< N_BUCKETS
; i
++) {
4518 ds_put_cstr(&s
, ", drop");
4521 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4524 VLOG_INFO("%s: %s (msec:count)", ofproto
->up
.name
, ds_cstr(&s
));
4528 return bucket
* BUCKET_WIDTH
;
4532 expire_subfacets(struct ofproto_dpif
*ofproto
, int dp_max_idle
)
4534 /* Cutoff time for most flows. */
4535 long long int normal_cutoff
= time_msec() - dp_max_idle
;
4537 /* We really want to keep flows for special protocols around, so use a more
4538 * conservative cutoff. */
4539 long long int special_cutoff
= time_msec() - 10000;
4541 struct subfacet
*subfacet
, *next_subfacet
;
4542 struct subfacet
*batch
[SUBFACET_DESTROY_MAX_BATCH
];
4546 HMAP_FOR_EACH_SAFE (subfacet
, next_subfacet
, hmap_node
,
4547 &ofproto
->subfacets
) {
4548 long long int cutoff
;
4550 cutoff
= (subfacet
->slow
& (SLOW_CFM
| SLOW_BFD
| SLOW_LACP
| SLOW_STP
)
4553 if (subfacet
->used
< cutoff
) {
4554 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
4555 batch
[n_batch
++] = subfacet
;
4556 if (n_batch
>= SUBFACET_DESTROY_MAX_BATCH
) {
4557 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
4561 subfacet_destroy(subfacet
);
4567 subfacet_destroy_batch(ofproto
, batch
, n_batch
);
4571 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4572 * then delete it entirely. */
4574 rule_expire(struct rule_dpif
*rule
)
4576 struct facet
*facet
, *next_facet
;
4580 if (rule
->up
.pending
) {
4581 /* We'll have to expire it later. */
4585 /* Has 'rule' expired? */
4587 if (rule
->up
.hard_timeout
4588 && now
> rule
->up
.modified
+ rule
->up
.hard_timeout
* 1000) {
4589 reason
= OFPRR_HARD_TIMEOUT
;
4590 } else if (rule
->up
.idle_timeout
4591 && now
> rule
->up
.used
+ rule
->up
.idle_timeout
* 1000) {
4592 reason
= OFPRR_IDLE_TIMEOUT
;
4597 COVERAGE_INC(ofproto_dpif_expired
);
4599 /* Update stats. (This is a no-op if the rule expired due to an idle
4600 * timeout, because that only happens when the rule has no facets left.) */
4601 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4602 facet_remove(facet
);
4605 /* Get rid of the rule. */
4606 ofproto_rule_expire(&rule
->up
, reason
);
4611 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4613 * The caller must already have determined that no facet with an identical
4614 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4615 * the ofproto's classifier table.
4617 * 'hash' must be the return value of flow_hash(flow, 0).
4619 * The facet will initially have no subfacets. The caller should create (at
4620 * least) one subfacet with subfacet_create(). */
4621 static struct facet
*
4622 facet_create(struct rule_dpif
*rule
, const struct flow
*flow
, uint32_t hash
)
4624 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
4625 struct facet
*facet
;
4627 facet
= xzalloc(sizeof *facet
);
4628 facet
->used
= time_msec();
4629 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, hash
);
4630 list_push_back(&rule
->facets
, &facet
->list_node
);
4632 facet
->flow
= *flow
;
4633 list_init(&facet
->subfacets
);
4634 netflow_flow_init(&facet
->nf_flow
);
4635 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
4637 facet
->learn_rl
= time_msec() + 500;
4643 facet_free(struct facet
*facet
)
4648 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4649 * 'packet', which arrived on 'in_port'. */
4651 execute_odp_actions(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
4652 const struct nlattr
*odp_actions
, size_t actions_len
,
4653 struct ofpbuf
*packet
)
4655 struct odputil_keybuf keybuf
;
4659 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
4660 odp_flow_key_from_flow(&key
, flow
,
4661 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
4663 error
= dpif_execute(ofproto
->backer
->dpif
, key
.data
, key
.size
,
4664 odp_actions
, actions_len
, packet
);
4668 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4670 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4671 * rule's statistics, via subfacet_uninstall().
4673 * - Removes 'facet' from its rule and from ofproto->facets.
4676 facet_remove(struct facet
*facet
)
4678 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4679 struct subfacet
*subfacet
, *next_subfacet
;
4681 ovs_assert(!list_is_empty(&facet
->subfacets
));
4683 /* First uninstall all of the subfacets to get final statistics. */
4684 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4685 subfacet_uninstall(subfacet
);
4688 /* Flush the final stats to the rule.
4690 * This might require us to have at least one subfacet around so that we
4691 * can use its actions for accounting in facet_account(), which is why we
4692 * have uninstalled but not yet destroyed the subfacets. */
4693 facet_flush_stats(facet
);
4695 /* Now we're really all done so destroy everything. */
4696 LIST_FOR_EACH_SAFE (subfacet
, next_subfacet
, list_node
,
4697 &facet
->subfacets
) {
4698 subfacet_destroy__(subfacet
);
4700 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
4701 list_remove(&facet
->list_node
);
4705 /* Feed information from 'facet' back into the learning table to keep it in
4706 * sync with what is actually flowing through the datapath. */
4708 facet_learn(struct facet
*facet
)
4710 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4711 struct subfacet
*subfacet
= CONTAINER_OF(list_front(&facet
->subfacets
),
4712 struct subfacet
, list_node
);
4713 long long int now
= time_msec();
4714 struct action_xlate_ctx ctx
;
4716 if (!facet
->has_fin_timeout
&& now
< facet
->learn_rl
) {
4720 facet
->learn_rl
= now
+ 500;
4722 if (!facet
->has_learn
4723 && !facet
->has_normal
4724 && (!facet
->has_fin_timeout
4725 || !(facet
->tcp_flags
& (TCP_FIN
| TCP_RST
)))) {
4729 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
4730 &subfacet
->initial_vals
,
4731 facet
->rule
, facet
->tcp_flags
, NULL
);
4732 ctx
.may_learn
= true;
4733 xlate_actions_for_side_effects(&ctx
, facet
->rule
->up
.ofpacts
,
4734 facet
->rule
->up
.ofpacts_len
);
4738 facet_account(struct facet
*facet
)
4740 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4741 struct subfacet
*subfacet
= facet_get_subfacet(facet
);
4742 const struct nlattr
*a
;
4747 if (!facet
->has_normal
|| !ofproto
->has_bonded_bundles
) {
4750 n_bytes
= facet
->byte_count
- facet
->accounted_bytes
;
4752 /* This loop feeds byte counters to bond_account() for rebalancing to use
4753 * as a basis. We also need to track the actual VLAN on which the packet
4754 * is going to be sent to ensure that it matches the one passed to
4755 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4758 * We use the actions from an arbitrary subfacet because they should all
4759 * be equally valid for our purpose. */
4760 vlan_tci
= facet
->flow
.vlan_tci
;
4761 NL_ATTR_FOR_EACH_UNSAFE (a
, left
,
4762 subfacet
->actions
, subfacet
->actions_len
) {
4763 const struct ovs_action_push_vlan
*vlan
;
4764 struct ofport_dpif
*port
;
4766 switch (nl_attr_type(a
)) {
4767 case OVS_ACTION_ATTR_OUTPUT
:
4768 port
= get_odp_port(ofproto
, nl_attr_get_u32(a
));
4769 if (port
&& port
->bundle
&& port
->bundle
->bond
) {
4770 bond_account(port
->bundle
->bond
, &facet
->flow
,
4771 vlan_tci_to_vid(vlan_tci
), n_bytes
);
4775 case OVS_ACTION_ATTR_POP_VLAN
:
4776 vlan_tci
= htons(0);
4779 case OVS_ACTION_ATTR_PUSH_VLAN
:
4780 vlan
= nl_attr_get(a
);
4781 vlan_tci
= vlan
->vlan_tci
;
4787 /* Returns true if the only action for 'facet' is to send to the controller.
4788 * (We don't report NetFlow expiration messages for such facets because they
4789 * are just part of the control logic for the network, not real traffic). */
4791 facet_is_controller_flow(struct facet
*facet
)
4794 const struct rule
*rule
= &facet
->rule
->up
;
4795 const struct ofpact
*ofpacts
= rule
->ofpacts
;
4796 size_t ofpacts_len
= rule
->ofpacts_len
;
4798 if (ofpacts_len
> 0 &&
4799 ofpacts
->type
== OFPACT_CONTROLLER
&&
4800 ofpact_next(ofpacts
) >= ofpact_end(ofpacts
, ofpacts_len
)) {
4807 /* Folds all of 'facet''s statistics into its rule. Also updates the
4808 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4809 * 'facet''s statistics in the datapath should have been zeroed and folded into
4810 * its packet and byte counts before this function is called. */
4812 facet_flush_stats(struct facet
*facet
)
4814 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4815 struct subfacet
*subfacet
;
4817 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4818 ovs_assert(!subfacet
->dp_byte_count
);
4819 ovs_assert(!subfacet
->dp_packet_count
);
4822 facet_push_stats(facet
);
4823 if (facet
->accounted_bytes
< facet
->byte_count
) {
4824 facet_account(facet
);
4825 facet
->accounted_bytes
= facet
->byte_count
;
4828 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
4829 struct ofexpired expired
;
4830 expired
.flow
= facet
->flow
;
4831 expired
.packet_count
= facet
->packet_count
;
4832 expired
.byte_count
= facet
->byte_count
;
4833 expired
.used
= facet
->used
;
4834 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4837 facet
->rule
->packet_count
+= facet
->packet_count
;
4838 facet
->rule
->byte_count
+= facet
->byte_count
;
4840 /* Reset counters to prevent double counting if 'facet' ever gets
4842 facet_reset_counters(facet
);
4844 netflow_flow_clear(&facet
->nf_flow
);
4845 facet
->tcp_flags
= 0;
4848 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4849 * Returns it if found, otherwise a null pointer.
4851 * 'hash' must be the return value of flow_hash(flow, 0).
4853 * The returned facet might need revalidation; use facet_lookup_valid()
4854 * instead if that is important. */
4855 static struct facet
*
4856 facet_find(struct ofproto_dpif
*ofproto
,
4857 const struct flow
*flow
, uint32_t hash
)
4859 struct facet
*facet
;
4861 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, hash
, &ofproto
->facets
) {
4862 if (flow_equal(flow
, &facet
->flow
)) {
4870 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4871 * Returns it if found, otherwise a null pointer.
4873 * 'hash' must be the return value of flow_hash(flow, 0).
4875 * The returned facet is guaranteed to be valid. */
4876 static struct facet
*
4877 facet_lookup_valid(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
4880 struct facet
*facet
;
4882 facet
= facet_find(ofproto
, flow
, hash
);
4884 && (ofproto
->backer
->need_revalidate
4885 || tag_set_intersects(&ofproto
->backer
->revalidate_set
,
4887 facet_revalidate(facet
);
4889 /* facet_revalidate() may have destroyed 'facet'. */
4890 facet
= facet_find(ofproto
, flow
, hash
);
4896 /* Return a subfacet from 'facet'. A facet consists of one or more
4897 * subfacets, and this function returns one of them. */
4898 static struct subfacet
*facet_get_subfacet(struct facet
*facet
)
4900 return CONTAINER_OF(list_front(&facet
->subfacets
), struct subfacet
,
4905 subfacet_path_to_string(enum subfacet_path path
)
4908 case SF_NOT_INSTALLED
:
4909 return "not installed";
4911 return "in fast path";
4913 return "in slow path";
4919 /* Returns the path in which a subfacet should be installed if its 'slow'
4920 * member has the specified value. */
4921 static enum subfacet_path
4922 subfacet_want_path(enum slow_path_reason slow
)
4924 return slow
? SF_SLOW_PATH
: SF_FAST_PATH
;
4927 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4928 * supposing that its actions have been recalculated as 'want_actions' and that
4929 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4931 subfacet_should_install(struct subfacet
*subfacet
, enum slow_path_reason slow
,
4932 const struct ofpbuf
*want_actions
)
4934 enum subfacet_path want_path
= subfacet_want_path(slow
);
4935 return (want_path
!= subfacet
->path
4936 || (want_path
== SF_FAST_PATH
4937 && (subfacet
->actions_len
!= want_actions
->size
4938 || memcmp(subfacet
->actions
, want_actions
->data
,
4939 subfacet
->actions_len
))));
4943 facet_check_consistency(struct facet
*facet
)
4945 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 15);
4947 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
4949 uint64_t odp_actions_stub
[1024 / 8];
4950 struct ofpbuf odp_actions
;
4952 struct rule_dpif
*rule
;
4953 struct subfacet
*subfacet
;
4954 bool may_log
= false;
4957 /* Check the rule for consistency. */
4958 rule
= rule_dpif_lookup(ofproto
, &facet
->flow
);
4959 ok
= rule
== facet
->rule
;
4961 may_log
= !VLOG_DROP_WARN(&rl
);
4966 flow_format(&s
, &facet
->flow
);
4967 ds_put_format(&s
, ": facet associated with wrong rule (was "
4968 "table=%"PRIu8
",", facet
->rule
->up
.table_id
);
4969 cls_rule_format(&facet
->rule
->up
.cr
, &s
);
4970 ds_put_format(&s
, ") (should have been table=%"PRIu8
",",
4972 cls_rule_format(&rule
->up
.cr
, &s
);
4973 ds_put_char(&s
, ')');
4975 VLOG_WARN("%s", ds_cstr(&s
));
4980 /* Check the datapath actions for consistency. */
4981 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
4982 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
4983 enum subfacet_path want_path
;
4984 struct action_xlate_ctx ctx
;
4987 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
4988 &subfacet
->initial_vals
, rule
, 0, NULL
);
4989 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
4992 if (subfacet
->path
== SF_NOT_INSTALLED
) {
4993 /* This only happens if the datapath reported an error when we
4994 * tried to install the flow. Don't flag another error here. */
4998 want_path
= subfacet_want_path(subfacet
->slow
);
4999 if (want_path
== SF_SLOW_PATH
&& subfacet
->path
== SF_SLOW_PATH
) {
5000 /* The actions for slow-path flows may legitimately vary from one
5001 * packet to the next. We're done. */
5005 if (!subfacet_should_install(subfacet
, subfacet
->slow
, &odp_actions
)) {
5009 /* Inconsistency! */
5011 may_log
= !VLOG_DROP_WARN(&rl
);
5015 /* Rate-limited, skip reporting. */
5020 odp_flow_key_format(subfacet
->key
, subfacet
->key_len
, &s
);
5022 ds_put_cstr(&s
, ": inconsistency in subfacet");
5023 if (want_path
!= subfacet
->path
) {
5024 enum odp_key_fitness fitness
= subfacet
->key_fitness
;
5026 ds_put_format(&s
, " (%s, fitness=%s)",
5027 subfacet_path_to_string(subfacet
->path
),
5028 odp_key_fitness_to_string(fitness
));
5029 ds_put_format(&s
, " (should have been %s)",
5030 subfacet_path_to_string(want_path
));
5031 } else if (want_path
== SF_FAST_PATH
) {
5032 ds_put_cstr(&s
, " (actions were: ");
5033 format_odp_actions(&s
, subfacet
->actions
,
5034 subfacet
->actions_len
);
5035 ds_put_cstr(&s
, ") (correct actions: ");
5036 format_odp_actions(&s
, odp_actions
.data
, odp_actions
.size
);
5037 ds_put_char(&s
, ')');
5039 ds_put_cstr(&s
, " (actions: ");
5040 format_odp_actions(&s
, subfacet
->actions
,
5041 subfacet
->actions_len
);
5042 ds_put_char(&s
, ')');
5044 VLOG_WARN("%s", ds_cstr(&s
));
5047 ofpbuf_uninit(&odp_actions
);
5052 /* Re-searches the classifier for 'facet':
5054 * - If the rule found is different from 'facet''s current rule, moves
5055 * 'facet' to the new rule and recompiles its actions.
5057 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5058 * where it is and recompiles its actions anyway.
5060 * - If any of 'facet''s subfacets correspond to a new flow according to
5061 * ofproto_receive(), 'facet' is removed. */
5063 facet_revalidate(struct facet
*facet
)
5065 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
5067 struct nlattr
*odp_actions
;
5070 struct actions
*new_actions
;
5072 struct action_xlate_ctx ctx
;
5073 uint64_t odp_actions_stub
[1024 / 8];
5074 struct ofpbuf odp_actions
;
5076 struct rule_dpif
*new_rule
;
5077 struct subfacet
*subfacet
;
5080 COVERAGE_INC(facet_revalidate
);
5082 /* Check that child subfacets still correspond to this facet. Tunnel
5083 * configuration changes could cause a subfacet's OpenFlow in_port to
5085 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
5086 struct ofproto_dpif
*recv_ofproto
;
5087 struct flow recv_flow
;
5090 error
= ofproto_receive(ofproto
->backer
, NULL
, subfacet
->key
,
5091 subfacet
->key_len
, &recv_flow
, NULL
,
5092 &recv_ofproto
, NULL
, NULL
);
5094 || recv_ofproto
!= ofproto
5095 || memcmp(&recv_flow
, &facet
->flow
, sizeof recv_flow
)) {
5096 facet_remove(facet
);
5101 new_rule
= rule_dpif_lookup(ofproto
, &facet
->flow
);
5103 /* Calculate new datapath actions.
5105 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5106 * emit a NetFlow expiration and, if so, we need to have the old state
5107 * around to properly compose it. */
5109 /* If the datapath actions changed or the installability changed,
5110 * then we need to talk to the datapath. */
5113 memset(&ctx
, 0, sizeof ctx
);
5114 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
5115 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
5116 enum slow_path_reason slow
;
5118 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
5119 &subfacet
->initial_vals
, new_rule
, 0, NULL
);
5120 xlate_actions(&ctx
, new_rule
->up
.ofpacts
, new_rule
->up
.ofpacts_len
,
5123 slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
5124 if (subfacet_should_install(subfacet
, slow
, &odp_actions
)) {
5125 struct dpif_flow_stats stats
;
5127 subfacet_install(subfacet
,
5128 odp_actions
.data
, odp_actions
.size
, &stats
, slow
);
5129 subfacet_update_stats(subfacet
, &stats
);
5132 new_actions
= xcalloc(list_size(&facet
->subfacets
),
5133 sizeof *new_actions
);
5135 new_actions
[i
].odp_actions
= xmemdup(odp_actions
.data
,
5137 new_actions
[i
].actions_len
= odp_actions
.size
;
5142 ofpbuf_uninit(&odp_actions
);
5145 facet_flush_stats(facet
);
5148 /* Update 'facet' now that we've taken care of all the old state. */
5149 facet
->tags
= ctx
.tags
;
5150 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
5151 facet
->has_learn
= ctx
.has_learn
;
5152 facet
->has_normal
= ctx
.has_normal
;
5153 facet
->has_fin_timeout
= ctx
.has_fin_timeout
;
5154 facet
->mirrors
= ctx
.mirrors
;
5157 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
5158 subfacet
->slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
5160 if (new_actions
&& new_actions
[i
].odp_actions
) {
5161 free(subfacet
->actions
);
5162 subfacet
->actions
= new_actions
[i
].odp_actions
;
5163 subfacet
->actions_len
= new_actions
[i
].actions_len
;
5169 if (facet
->rule
!= new_rule
) {
5170 COVERAGE_INC(facet_changed_rule
);
5171 list_remove(&facet
->list_node
);
5172 list_push_back(&new_rule
->facets
, &facet
->list_node
);
5173 facet
->rule
= new_rule
;
5174 facet
->used
= new_rule
->up
.created
;
5175 facet
->prev_used
= facet
->used
;
5179 /* Updates 'facet''s used time. Caller is responsible for calling
5180 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5182 facet_update_time(struct facet
*facet
, long long int used
)
5184 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
5185 if (used
> facet
->used
) {
5187 ofproto_rule_update_used(&facet
->rule
->up
, used
);
5188 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
5193 facet_reset_counters(struct facet
*facet
)
5195 facet
->packet_count
= 0;
5196 facet
->byte_count
= 0;
5197 facet
->prev_packet_count
= 0;
5198 facet
->prev_byte_count
= 0;
5199 facet
->accounted_bytes
= 0;
5203 facet_push_stats(struct facet
*facet
)
5205 struct dpif_flow_stats stats
;
5207 ovs_assert(facet
->packet_count
>= facet
->prev_packet_count
);
5208 ovs_assert(facet
->byte_count
>= facet
->prev_byte_count
);
5209 ovs_assert(facet
->used
>= facet
->prev_used
);
5211 stats
.n_packets
= facet
->packet_count
- facet
->prev_packet_count
;
5212 stats
.n_bytes
= facet
->byte_count
- facet
->prev_byte_count
;
5213 stats
.used
= facet
->used
;
5214 stats
.tcp_flags
= 0;
5216 if (stats
.n_packets
|| stats
.n_bytes
|| facet
->used
> facet
->prev_used
) {
5217 facet
->prev_packet_count
= facet
->packet_count
;
5218 facet
->prev_byte_count
= facet
->byte_count
;
5219 facet
->prev_used
= facet
->used
;
5221 flow_push_stats(facet
, &stats
);
5223 update_mirror_stats(ofproto_dpif_cast(facet
->rule
->up
.ofproto
),
5224 facet
->mirrors
, stats
.n_packets
, stats
.n_bytes
);
5229 push_all_stats__(bool run_fast
)
5231 static long long int rl
= LLONG_MIN
;
5232 struct ofproto_dpif
*ofproto
;
5234 if (time_msec() < rl
) {
5238 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
5239 struct facet
*facet
;
5241 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
5242 facet_push_stats(facet
);
5249 rl
= time_msec() + 100;
5253 push_all_stats(void)
5255 push_all_stats__(true);
5259 rule_credit_stats(struct rule_dpif
*rule
, const struct dpif_flow_stats
*stats
)
5261 rule
->packet_count
+= stats
->n_packets
;
5262 rule
->byte_count
+= stats
->n_bytes
;
5263 ofproto_rule_update_used(&rule
->up
, stats
->used
);
5266 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5267 * into given 'facet->rule''s actions and mirrors. */
5269 flow_push_stats(struct facet
*facet
, const struct dpif_flow_stats
*stats
)
5271 struct rule_dpif
*rule
= facet
->rule
;
5272 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5273 struct subfacet
*subfacet
= facet_get_subfacet(facet
);
5274 struct action_xlate_ctx ctx
;
5276 ofproto_rule_update_used(&rule
->up
, stats
->used
);
5278 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
5279 &subfacet
->initial_vals
, rule
, 0, NULL
);
5280 ctx
.resubmit_stats
= stats
;
5281 xlate_actions_for_side_effects(&ctx
, rule
->up
.ofpacts
,
5282 rule
->up
.ofpacts_len
);
5287 static struct subfacet
*
5288 subfacet_find(struct ofproto_dpif
*ofproto
,
5289 const struct nlattr
*key
, size_t key_len
, uint32_t key_hash
)
5291 struct subfacet
*subfacet
;
5293 HMAP_FOR_EACH_WITH_HASH (subfacet
, hmap_node
, key_hash
,
5294 &ofproto
->subfacets
) {
5295 if (subfacet
->key_len
== key_len
5296 && !memcmp(key
, subfacet
->key
, key_len
)) {
5304 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5305 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5306 * existing subfacet if there is one, otherwise creates and returns a
5309 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5310 * which case the caller must populate the actions with
5311 * subfacet_make_actions(). */
5312 static struct subfacet
*
5313 subfacet_create(struct facet
*facet
, struct flow_miss
*miss
,
5316 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
5317 enum odp_key_fitness key_fitness
= miss
->key_fitness
;
5318 const struct nlattr
*key
= miss
->key
;
5319 size_t key_len
= miss
->key_len
;
5321 struct subfacet
*subfacet
;
5323 key_hash
= odp_flow_key_hash(key
, key_len
);
5325 if (list_is_empty(&facet
->subfacets
)) {
5326 subfacet
= &facet
->one_subfacet
;
5328 subfacet
= subfacet_find(ofproto
, key
, key_len
, key_hash
);
5330 if (subfacet
->facet
== facet
) {
5334 /* This shouldn't happen. */
5335 VLOG_ERR_RL(&rl
, "subfacet with wrong facet");
5336 subfacet_destroy(subfacet
);
5339 subfacet
= xmalloc(sizeof *subfacet
);
5342 hmap_insert(&ofproto
->subfacets
, &subfacet
->hmap_node
, key_hash
);
5343 list_push_back(&facet
->subfacets
, &subfacet
->list_node
);
5344 subfacet
->facet
= facet
;
5345 subfacet
->key_fitness
= key_fitness
;
5346 subfacet
->key
= xmemdup(key
, key_len
);
5347 subfacet
->key_len
= key_len
;
5348 subfacet
->used
= now
;
5349 subfacet
->created
= now
;
5350 subfacet
->dp_packet_count
= 0;
5351 subfacet
->dp_byte_count
= 0;
5352 subfacet
->actions_len
= 0;
5353 subfacet
->actions
= NULL
;
5354 subfacet
->slow
= (subfacet
->key_fitness
== ODP_FIT_TOO_LITTLE
5357 subfacet
->path
= SF_NOT_INSTALLED
;
5358 subfacet
->initial_vals
= miss
->initial_vals
;
5359 subfacet
->odp_in_port
= miss
->odp_in_port
;
5361 ofproto
->subfacet_add_count
++;
5365 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5366 * its facet within 'ofproto', and frees it. */
5368 subfacet_destroy__(struct subfacet
*subfacet
)
5370 struct facet
*facet
= subfacet
->facet
;
5371 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
5373 /* Update ofproto stats before uninstall the subfacet. */
5374 ofproto
->subfacet_del_count
++;
5375 ofproto
->total_subfacet_life_span
+= (time_msec() - subfacet
->created
);
5377 subfacet_uninstall(subfacet
);
5378 hmap_remove(&ofproto
->subfacets
, &subfacet
->hmap_node
);
5379 list_remove(&subfacet
->list_node
);
5380 free(subfacet
->key
);
5381 free(subfacet
->actions
);
5382 if (subfacet
!= &facet
->one_subfacet
) {
5387 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5388 * last remaining subfacet in its facet destroys the facet too. */
5390 subfacet_destroy(struct subfacet
*subfacet
)
5392 struct facet
*facet
= subfacet
->facet
;
5394 if (list_is_singleton(&facet
->subfacets
)) {
5395 /* facet_remove() needs at least one subfacet (it will remove it). */
5396 facet_remove(facet
);
5398 subfacet_destroy__(subfacet
);
5403 subfacet_destroy_batch(struct ofproto_dpif
*ofproto
,
5404 struct subfacet
**subfacets
, int n
)
5406 struct dpif_op ops
[SUBFACET_DESTROY_MAX_BATCH
];
5407 struct dpif_op
*opsp
[SUBFACET_DESTROY_MAX_BATCH
];
5408 struct dpif_flow_stats stats
[SUBFACET_DESTROY_MAX_BATCH
];
5411 for (i
= 0; i
< n
; i
++) {
5412 ops
[i
].type
= DPIF_OP_FLOW_DEL
;
5413 ops
[i
].u
.flow_del
.key
= subfacets
[i
]->key
;
5414 ops
[i
].u
.flow_del
.key_len
= subfacets
[i
]->key_len
;
5415 ops
[i
].u
.flow_del
.stats
= &stats
[i
];
5419 dpif_operate(ofproto
->backer
->dpif
, opsp
, n
);
5420 for (i
= 0; i
< n
; i
++) {
5421 subfacet_reset_dp_stats(subfacets
[i
], &stats
[i
]);
5422 subfacets
[i
]->path
= SF_NOT_INSTALLED
;
5423 subfacet_destroy(subfacets
[i
]);
5428 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5429 * Translates the actions into 'odp_actions', which the caller must have
5430 * initialized and is responsible for uninitializing. */
5432 subfacet_make_actions(struct subfacet
*subfacet
, const struct ofpbuf
*packet
,
5433 struct ofpbuf
*odp_actions
)
5435 struct facet
*facet
= subfacet
->facet
;
5436 struct rule_dpif
*rule
= facet
->rule
;
5437 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5439 struct action_xlate_ctx ctx
;
5441 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
,
5442 &subfacet
->initial_vals
, rule
, 0, packet
);
5443 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, odp_actions
);
5444 facet
->tags
= ctx
.tags
;
5445 facet
->has_learn
= ctx
.has_learn
;
5446 facet
->has_normal
= ctx
.has_normal
;
5447 facet
->has_fin_timeout
= ctx
.has_fin_timeout
;
5448 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
5449 facet
->mirrors
= ctx
.mirrors
;
5451 subfacet
->slow
= (subfacet
->slow
& SLOW_MATCH
) | ctx
.slow
;
5452 if (subfacet
->actions_len
!= odp_actions
->size
5453 || memcmp(subfacet
->actions
, odp_actions
->data
, odp_actions
->size
)) {
5454 free(subfacet
->actions
);
5455 subfacet
->actions_len
= odp_actions
->size
;
5456 subfacet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
5460 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5461 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5462 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5463 * since 'subfacet' was last updated.
5465 * Returns 0 if successful, otherwise a positive errno value. */
5467 subfacet_install(struct subfacet
*subfacet
,
5468 const struct nlattr
*actions
, size_t actions_len
,
5469 struct dpif_flow_stats
*stats
,
5470 enum slow_path_reason slow
)
5472 struct facet
*facet
= subfacet
->facet
;
5473 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(facet
->rule
->up
.ofproto
);
5474 enum subfacet_path path
= subfacet_want_path(slow
);
5475 uint64_t slow_path_stub
[128 / 8];
5476 enum dpif_flow_put_flags flags
;
5479 flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
5481 flags
|= DPIF_FP_ZERO_STATS
;
5484 if (path
== SF_SLOW_PATH
) {
5485 compose_slow_path(ofproto
, &facet
->flow
, slow
,
5486 slow_path_stub
, sizeof slow_path_stub
,
5487 &actions
, &actions_len
);
5490 ret
= dpif_flow_put(ofproto
->backer
->dpif
, flags
, subfacet
->key
,
5491 subfacet
->key_len
, actions
, actions_len
, stats
);
5494 subfacet_reset_dp_stats(subfacet
, stats
);
5498 subfacet
->path
= path
;
5504 subfacet_reinstall(struct subfacet
*subfacet
, struct dpif_flow_stats
*stats
)
5506 return subfacet_install(subfacet
, subfacet
->actions
, subfacet
->actions_len
,
5507 stats
, subfacet
->slow
);
5510 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5512 subfacet_uninstall(struct subfacet
*subfacet
)
5514 if (subfacet
->path
!= SF_NOT_INSTALLED
) {
5515 struct rule_dpif
*rule
= subfacet
->facet
->rule
;
5516 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5517 struct dpif_flow_stats stats
;
5520 error
= dpif_flow_del(ofproto
->backer
->dpif
, subfacet
->key
,
5521 subfacet
->key_len
, &stats
);
5522 subfacet_reset_dp_stats(subfacet
, &stats
);
5524 subfacet_update_stats(subfacet
, &stats
);
5526 subfacet
->path
= SF_NOT_INSTALLED
;
5528 ovs_assert(subfacet
->dp_packet_count
== 0);
5529 ovs_assert(subfacet
->dp_byte_count
== 0);
5533 /* Resets 'subfacet''s datapath statistics counters. This should be called
5534 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5535 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5536 * was reset in the datapath. 'stats' will be modified to include only
5537 * statistics new since 'subfacet' was last updated. */
5539 subfacet_reset_dp_stats(struct subfacet
*subfacet
,
5540 struct dpif_flow_stats
*stats
)
5543 && subfacet
->dp_packet_count
<= stats
->n_packets
5544 && subfacet
->dp_byte_count
<= stats
->n_bytes
) {
5545 stats
->n_packets
-= subfacet
->dp_packet_count
;
5546 stats
->n_bytes
-= subfacet
->dp_byte_count
;
5549 subfacet
->dp_packet_count
= 0;
5550 subfacet
->dp_byte_count
= 0;
5553 /* Updates 'subfacet''s used time. The caller is responsible for calling
5554 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5556 subfacet_update_time(struct subfacet
*subfacet
, long long int used
)
5558 if (used
> subfacet
->used
) {
5559 subfacet
->used
= used
;
5560 facet_update_time(subfacet
->facet
, used
);
5564 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5566 * Because of the meaning of a subfacet's counters, it only makes sense to do
5567 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5568 * represents a packet that was sent by hand or if it represents statistics
5569 * that have been cleared out of the datapath. */
5571 subfacet_update_stats(struct subfacet
*subfacet
,
5572 const struct dpif_flow_stats
*stats
)
5574 if (stats
->n_packets
|| stats
->used
> subfacet
->used
) {
5575 struct facet
*facet
= subfacet
->facet
;
5577 subfacet_update_time(subfacet
, stats
->used
);
5578 facet
->packet_count
+= stats
->n_packets
;
5579 facet
->byte_count
+= stats
->n_bytes
;
5580 facet
->tcp_flags
|= stats
->tcp_flags
;
5581 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
5587 static struct rule_dpif
*
5588 rule_dpif_lookup(struct ofproto_dpif
*ofproto
, const struct flow
*flow
)
5590 struct rule_dpif
*rule
;
5592 rule
= rule_dpif_lookup__(ofproto
, flow
, 0);
5597 return rule_dpif_miss_rule(ofproto
, flow
);
5600 static struct rule_dpif
*
5601 rule_dpif_lookup__(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
5604 struct cls_rule
*cls_rule
;
5605 struct classifier
*cls
;
5607 if (table_id
>= N_TABLES
) {
5611 cls
= &ofproto
->up
.tables
[table_id
].cls
;
5612 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
5613 && ofproto
->up
.frag_handling
== OFPC_FRAG_NORMAL
) {
5614 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5615 * are unavailable. */
5616 struct flow ofpc_normal_flow
= *flow
;
5617 ofpc_normal_flow
.tp_src
= htons(0);
5618 ofpc_normal_flow
.tp_dst
= htons(0);
5619 cls_rule
= classifier_lookup(cls
, &ofpc_normal_flow
);
5621 cls_rule
= classifier_lookup(cls
, flow
);
5623 return rule_dpif_cast(rule_from_cls_rule(cls_rule
));
5626 static struct rule_dpif
*
5627 rule_dpif_miss_rule(struct ofproto_dpif
*ofproto
, const struct flow
*flow
)
5629 struct ofport_dpif
*port
;
5631 port
= get_ofp_port(ofproto
, flow
->in_port
);
5633 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, flow
->in_port
);
5634 return ofproto
->miss_rule
;
5637 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_PACKET_IN
) {
5638 return ofproto
->no_packet_in_rule
;
5640 return ofproto
->miss_rule
;
5644 complete_operation(struct rule_dpif
*rule
)
5646 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5648 rule_invalidate(rule
);
5650 struct dpif_completion
*c
= xmalloc(sizeof *c
);
5651 c
->op
= rule
->up
.pending
;
5652 list_push_back(&ofproto
->completions
, &c
->list_node
);
5654 ofoperation_complete(rule
->up
.pending
, 0);
5658 static struct rule
*
5661 struct rule_dpif
*rule
= xmalloc(sizeof *rule
);
5666 rule_dealloc(struct rule
*rule_
)
5668 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5673 rule_construct(struct rule
*rule_
)
5675 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5676 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5677 struct rule_dpif
*victim
;
5680 rule
->packet_count
= 0;
5681 rule
->byte_count
= 0;
5683 victim
= rule_dpif_cast(ofoperation_get_victim(rule
->up
.pending
));
5684 if (victim
&& !list_is_empty(&victim
->facets
)) {
5685 struct facet
*facet
;
5687 rule
->facets
= victim
->facets
;
5688 list_moved(&rule
->facets
);
5689 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
5690 /* XXX: We're only clearing our local counters here. It's possible
5691 * that quite a few packets are unaccounted for in the datapath
5692 * statistics. These will be accounted to the new rule instead of
5693 * cleared as required. This could be fixed by clearing out the
5694 * datapath statistics for this facet, but currently it doesn't
5696 facet_reset_counters(facet
);
5700 /* Must avoid list_moved() in this case. */
5701 list_init(&rule
->facets
);
5704 table_id
= rule
->up
.table_id
;
5706 rule
->tag
= victim
->tag
;
5707 } else if (table_id
== 0) {
5712 miniflow_expand(&rule
->up
.cr
.match
.flow
, &flow
);
5713 rule
->tag
= rule_calculate_tag(&flow
, &rule
->up
.cr
.match
.mask
,
5714 ofproto
->tables
[table_id
].basis
);
5717 complete_operation(rule
);
5722 rule_destruct(struct rule
*rule_
)
5724 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5725 struct facet
*facet
, *next_facet
;
5727 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
5728 facet_revalidate(facet
);
5731 complete_operation(rule
);
5735 rule_get_stats(struct rule
*rule_
, uint64_t *packets
, uint64_t *bytes
)
5737 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5738 struct facet
*facet
;
5740 /* push_all_stats() can handle flow misses which, when using the learn
5741 * action, can cause rules to be added and deleted. This can corrupt our
5742 * caller's datastructures which assume that rule_get_stats() doesn't have
5743 * an impact on the flow table. To be safe, we disable miss handling. */
5744 push_all_stats__(false);
5746 /* Start from historical data for 'rule' itself that are no longer tracked
5747 * in facets. This counts, for example, facets that have expired. */
5748 *packets
= rule
->packet_count
;
5749 *bytes
= rule
->byte_count
;
5751 /* Add any statistics that are tracked by facets. This includes
5752 * statistical data recently updated by ofproto_update_stats() as well as
5753 * stats for packets that were executed "by hand" via dpif_execute(). */
5754 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
5755 *packets
+= facet
->packet_count
;
5756 *bytes
+= facet
->byte_count
;
5761 rule_dpif_execute(struct rule_dpif
*rule
, const struct flow
*flow
,
5762 struct ofpbuf
*packet
)
5764 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
5765 struct initial_vals initial_vals
;
5766 struct dpif_flow_stats stats
;
5767 struct action_xlate_ctx ctx
;
5768 uint64_t odp_actions_stub
[1024 / 8];
5769 struct ofpbuf odp_actions
;
5771 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
5772 rule_credit_stats(rule
, &stats
);
5774 initial_vals
.vlan_tci
= flow
->vlan_tci
;
5775 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
5776 action_xlate_ctx_init(&ctx
, ofproto
, flow
, &initial_vals
,
5777 rule
, stats
.tcp_flags
, packet
);
5778 ctx
.resubmit_stats
= &stats
;
5779 xlate_actions(&ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, &odp_actions
);
5781 execute_odp_actions(ofproto
, flow
, odp_actions
.data
,
5782 odp_actions
.size
, packet
);
5784 ofpbuf_uninit(&odp_actions
);
5788 rule_execute(struct rule
*rule
, const struct flow
*flow
,
5789 struct ofpbuf
*packet
)
5791 rule_dpif_execute(rule_dpif_cast(rule
), flow
, packet
);
5792 ofpbuf_delete(packet
);
5797 rule_modify_actions(struct rule
*rule_
)
5799 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
5801 complete_operation(rule
);
5804 /* Sends 'packet' out 'ofport'.
5805 * May modify 'packet'.
5806 * Returns 0 if successful, otherwise a positive errno value. */
5808 send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
5810 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
5811 uint64_t odp_actions_stub
[1024 / 8];
5812 struct ofpbuf key
, odp_actions
;
5813 struct odputil_keybuf keybuf
;
5818 flow_extract(packet
, 0, 0, NULL
, OFPP_LOCAL
, &flow
);
5819 if (netdev_vport_is_patch(ofport
->up
.netdev
)) {
5820 struct ofproto_dpif
*peer_ofproto
;
5821 struct dpif_flow_stats stats
;
5822 struct ofport_dpif
*peer
;
5823 struct rule_dpif
*rule
;
5825 peer
= ofport_get_peer(ofport
);
5830 dpif_flow_stats_extract(&flow
, packet
, time_msec(), &stats
);
5831 netdev_vport_inc_tx(ofport
->up
.netdev
, &stats
);
5832 netdev_vport_inc_rx(peer
->up
.netdev
, &stats
);
5834 flow
.in_port
= peer
->up
.ofp_port
;
5835 peer_ofproto
= ofproto_dpif_cast(peer
->up
.ofproto
);
5836 rule
= rule_dpif_lookup(peer_ofproto
, &flow
);
5837 rule_dpif_execute(rule
, &flow
, packet
);
5842 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
5844 if (ofport
->tnl_port
) {
5845 struct dpif_flow_stats stats
;
5847 odp_port
= tnl_port_send(ofport
->tnl_port
, &flow
);
5848 if (odp_port
== OVSP_NONE
) {
5852 dpif_flow_stats_extract(&flow
, packet
, time_msec(), &stats
);
5853 netdev_vport_inc_tx(ofport
->up
.netdev
, &stats
);
5854 odp_put_tunnel_action(&flow
.tunnel
, &odp_actions
);
5855 odp_put_skb_mark_action(flow
.skb_mark
, &odp_actions
);
5857 odp_port
= vsp_realdev_to_vlandev(ofproto
, ofport
->odp_port
,
5859 if (odp_port
!= ofport
->odp_port
) {
5860 eth_pop_vlan(packet
);
5861 flow
.vlan_tci
= htons(0);
5865 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
5866 odp_flow_key_from_flow(&key
, &flow
,
5867 ofp_port_to_odp_port(ofproto
, flow
.in_port
));
5869 compose_sflow_action(ofproto
, &odp_actions
, &flow
, odp_port
);
5870 compose_ipfix_action(ofproto
, &odp_actions
, &flow
);
5872 nl_msg_put_u32(&odp_actions
, OVS_ACTION_ATTR_OUTPUT
, odp_port
);
5873 error
= dpif_execute(ofproto
->backer
->dpif
,
5875 odp_actions
.data
, odp_actions
.size
,
5877 ofpbuf_uninit(&odp_actions
);
5880 VLOG_WARN_RL(&rl
, "%s: failed to send packet on port %"PRIu32
" (%s)",
5881 ofproto
->up
.name
, odp_port
, strerror(error
));
5884 ofproto
->stats
.tx_packets
++;
5885 ofproto
->stats
.tx_bytes
+= packet
->size
;
5889 /* OpenFlow to datapath action translation. */
5891 static bool may_receive(const struct ofport_dpif
*, struct action_xlate_ctx
*);
5892 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
5893 struct action_xlate_ctx
*);
5894 static void xlate_normal(struct action_xlate_ctx
*);
5896 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5897 * The action will state 'slow' as the reason that the action is in the slow
5898 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5899 * dump-flows" output to see why a flow is in the slow path.)
5901 * The 'stub_size' bytes in 'stub' will be used to store the action.
5902 * 'stub_size' must be large enough for the action.
5904 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5907 compose_slow_path(const struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
5908 enum slow_path_reason slow
,
5909 uint64_t *stub
, size_t stub_size
,
5910 const struct nlattr
**actionsp
, size_t *actions_lenp
)
5912 union user_action_cookie cookie
;
5915 cookie
.type
= USER_ACTION_COOKIE_SLOW_PATH
;
5916 cookie
.slow_path
.unused
= 0;
5917 cookie
.slow_path
.reason
= slow
;
5919 ofpbuf_use_stack(&buf
, stub
, stub_size
);
5920 if (slow
& (SLOW_CFM
| SLOW_BFD
| SLOW_LACP
| SLOW_STP
)) {
5921 uint32_t pid
= dpif_port_get_pid(ofproto
->backer
->dpif
, UINT32_MAX
);
5922 odp_put_userspace_action(pid
, &cookie
, sizeof cookie
.slow_path
, &buf
);
5924 put_userspace_action(ofproto
, &buf
, flow
, &cookie
,
5925 sizeof cookie
.slow_path
);
5927 *actionsp
= buf
.data
;
5928 *actions_lenp
= buf
.size
;
5932 put_userspace_action(const struct ofproto_dpif
*ofproto
,
5933 struct ofpbuf
*odp_actions
,
5934 const struct flow
*flow
,
5935 const union user_action_cookie
*cookie
,
5936 const size_t cookie_size
)
5940 pid
= dpif_port_get_pid(ofproto
->backer
->dpif
,
5941 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
5943 return odp_put_userspace_action(pid
, cookie
, cookie_size
, odp_actions
);
5946 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5947 * the number of packets out of UINT32_MAX to sample. The given
5948 * cookie is passed back in the callback for each sampled packet.
5951 compose_sample_action(const struct ofproto_dpif
*ofproto
,
5952 struct ofpbuf
*odp_actions
,
5953 const struct flow
*flow
,
5954 const uint32_t probability
,
5955 const union user_action_cookie
*cookie
,
5956 const size_t cookie_size
)
5958 size_t sample_offset
, actions_offset
;
5961 sample_offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
5963 nl_msg_put_u32(odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
, probability
);
5965 actions_offset
= nl_msg_start_nested(odp_actions
, OVS_SAMPLE_ATTR_ACTIONS
);
5966 cookie_offset
= put_userspace_action(ofproto
, odp_actions
, flow
, cookie
,
5969 nl_msg_end_nested(odp_actions
, actions_offset
);
5970 nl_msg_end_nested(odp_actions
, sample_offset
);
5971 return cookie_offset
;
5975 compose_sflow_cookie(const struct ofproto_dpif
*ofproto
,
5976 ovs_be16 vlan_tci
, uint32_t odp_port
,
5977 unsigned int n_outputs
, union user_action_cookie
*cookie
)
5981 cookie
->type
= USER_ACTION_COOKIE_SFLOW
;
5982 cookie
->sflow
.vlan_tci
= vlan_tci
;
5984 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5985 * port information") for the interpretation of cookie->output. */
5986 switch (n_outputs
) {
5988 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5989 cookie
->sflow
.output
= 0x40000000 | 256;
5993 ifindex
= dpif_sflow_odp_port_to_ifindex(ofproto
->sflow
, odp_port
);
5995 cookie
->sflow
.output
= ifindex
;
6000 /* 0x80000000 means "multiple output ports. */
6001 cookie
->sflow
.output
= 0x80000000 | n_outputs
;
6006 /* Compose SAMPLE action for sFlow bridge sampling. */
6008 compose_sflow_action(const struct ofproto_dpif
*ofproto
,
6009 struct ofpbuf
*odp_actions
,
6010 const struct flow
*flow
,
6013 uint32_t probability
;
6014 union user_action_cookie cookie
;
6016 if (!ofproto
->sflow
|| flow
->in_port
== OFPP_NONE
) {
6020 probability
= dpif_sflow_get_probability(ofproto
->sflow
);
6021 compose_sflow_cookie(ofproto
, htons(0), odp_port
,
6022 odp_port
== OVSP_NONE
? 0 : 1, &cookie
);
6024 return compose_sample_action(ofproto
, odp_actions
, flow
, probability
,
6025 &cookie
, sizeof cookie
.sflow
);
6029 compose_flow_sample_cookie(uint16_t probability
, uint32_t collector_set_id
,
6030 uint32_t obs_domain_id
, uint32_t obs_point_id
,
6031 union user_action_cookie
*cookie
)
6033 cookie
->type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
6034 cookie
->flow_sample
.probability
= probability
;
6035 cookie
->flow_sample
.collector_set_id
= collector_set_id
;
6036 cookie
->flow_sample
.obs_domain_id
= obs_domain_id
;
6037 cookie
->flow_sample
.obs_point_id
= obs_point_id
;
6041 compose_ipfix_cookie(union user_action_cookie
*cookie
)
6043 cookie
->type
= USER_ACTION_COOKIE_IPFIX
;
6046 /* Compose SAMPLE action for IPFIX bridge sampling. */
6048 compose_ipfix_action(const struct ofproto_dpif
*ofproto
,
6049 struct ofpbuf
*odp_actions
,
6050 const struct flow
*flow
)
6052 uint32_t probability
;
6053 union user_action_cookie cookie
;
6055 if (!ofproto
->ipfix
|| flow
->in_port
== OFPP_NONE
) {
6059 probability
= dpif_ipfix_get_bridge_exporter_probability(ofproto
->ipfix
);
6060 compose_ipfix_cookie(&cookie
);
6062 compose_sample_action(ofproto
, odp_actions
, flow
, probability
,
6063 &cookie
, sizeof cookie
.ipfix
);
6066 /* SAMPLE action for sFlow must be first action in any given list of
6067 * actions. At this point we do not have all information required to
6068 * build it. So try to build sample action as complete as possible. */
6070 add_sflow_action(struct action_xlate_ctx
*ctx
)
6072 ctx
->user_cookie_offset
= compose_sflow_action(ctx
->ofproto
,
6074 &ctx
->flow
, OVSP_NONE
);
6075 ctx
->sflow_odp_port
= 0;
6076 ctx
->sflow_n_outputs
= 0;
6079 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6080 * of actions, eventually after the SAMPLE action for sFlow. */
6082 add_ipfix_action(struct action_xlate_ctx
*ctx
)
6084 compose_ipfix_action(ctx
->ofproto
, ctx
->odp_actions
, &ctx
->flow
);
6087 /* Fix SAMPLE action according to data collected while composing ODP actions.
6088 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6089 * USERSPACE action's user-cookie which is required for sflow. */
6091 fix_sflow_action(struct action_xlate_ctx
*ctx
)
6093 const struct flow
*base
= &ctx
->base_flow
;
6094 union user_action_cookie
*cookie
;
6096 if (!ctx
->user_cookie_offset
) {
6100 cookie
= ofpbuf_at(ctx
->odp_actions
, ctx
->user_cookie_offset
,
6101 sizeof cookie
->sflow
);
6102 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
6104 compose_sflow_cookie(ctx
->ofproto
, base
->vlan_tci
,
6105 ctx
->sflow_odp_port
, ctx
->sflow_n_outputs
, cookie
);
6109 compose_output_action__(struct action_xlate_ctx
*ctx
, uint16_t ofp_port
,
6112 const struct ofport_dpif
*ofport
= get_ofp_port(ctx
->ofproto
, ofp_port
);
6113 ovs_be16 flow_vlan_tci
;
6114 uint32_t flow_skb_mark
;
6115 uint8_t flow_nw_tos
;
6116 struct priority_to_dscp
*pdscp
;
6117 uint32_t out_port
, odp_port
;
6119 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6120 * before traversing a patch port. */
6121 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 20);
6124 xlate_report(ctx
, "Nonexistent output port");
6126 } else if (ofport
->up
.pp
.config
& OFPUTIL_PC_NO_FWD
) {
6127 xlate_report(ctx
, "OFPPC_NO_FWD set, skipping output");
6129 } else if (check_stp
&& !stp_forward_in_state(ofport
->stp_state
)) {
6130 xlate_report(ctx
, "STP not in forwarding state, skipping output");
6134 if (netdev_vport_is_patch(ofport
->up
.netdev
)) {
6135 struct ofport_dpif
*peer
= ofport_get_peer(ofport
);
6136 struct flow old_flow
= ctx
->flow
;
6137 const struct ofproto_dpif
*peer_ofproto
;
6138 enum slow_path_reason special
;
6139 struct ofport_dpif
*in_port
;
6142 xlate_report(ctx
, "Nonexistent patch port peer");
6146 peer_ofproto
= ofproto_dpif_cast(peer
->up
.ofproto
);
6147 if (peer_ofproto
->backer
!= ctx
->ofproto
->backer
) {
6148 xlate_report(ctx
, "Patch port peer on a different datapath");
6152 ctx
->ofproto
= ofproto_dpif_cast(peer
->up
.ofproto
);
6153 ctx
->flow
.in_port
= peer
->up
.ofp_port
;
6154 ctx
->flow
.metadata
= htonll(0);
6155 memset(&ctx
->flow
.tunnel
, 0, sizeof ctx
->flow
.tunnel
);
6156 memset(ctx
->flow
.regs
, 0, sizeof ctx
->flow
.regs
);
6158 in_port
= get_ofp_port(ctx
->ofproto
, ctx
->flow
.in_port
);
6159 special
= process_special(ctx
->ofproto
, &ctx
->flow
, in_port
,
6162 ctx
->slow
|= special
;
6163 } else if (!in_port
|| may_receive(in_port
, ctx
)) {
6164 if (!in_port
|| stp_forward_in_state(in_port
->stp_state
)) {
6165 xlate_table_action(ctx
, ctx
->flow
.in_port
, 0, true);
6167 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6168 * learning action look at the packet, then drop it. */
6169 struct flow old_base_flow
= ctx
->base_flow
;
6170 size_t old_size
= ctx
->odp_actions
->size
;
6171 xlate_table_action(ctx
, ctx
->flow
.in_port
, 0, true);
6172 ctx
->base_flow
= old_base_flow
;
6173 ctx
->odp_actions
->size
= old_size
;
6177 ctx
->flow
= old_flow
;
6178 ctx
->ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
6180 if (ctx
->resubmit_stats
) {
6181 netdev_vport_inc_tx(ofport
->up
.netdev
, ctx
->resubmit_stats
);
6182 netdev_vport_inc_rx(peer
->up
.netdev
, ctx
->resubmit_stats
);
6188 flow_vlan_tci
= ctx
->flow
.vlan_tci
;
6189 flow_skb_mark
= ctx
->flow
.skb_mark
;
6190 flow_nw_tos
= ctx
->flow
.nw_tos
;
6192 pdscp
= get_priority(ofport
, ctx
->flow
.skb_priority
);
6194 ctx
->flow
.nw_tos
&= ~IP_DSCP_MASK
;
6195 ctx
->flow
.nw_tos
|= pdscp
->dscp
;
6198 if (ofport
->tnl_port
) {
6199 /* Save tunnel metadata so that changes made due to
6200 * the Logical (tunnel) Port are not visible for any further
6201 * matches, while explicit set actions on tunnel metadata are.
6203 struct flow_tnl flow_tnl
= ctx
->flow
.tunnel
;
6204 odp_port
= tnl_port_send(ofport
->tnl_port
, &ctx
->flow
);
6205 if (odp_port
== OVSP_NONE
) {
6206 xlate_report(ctx
, "Tunneling decided against output");
6207 goto out
; /* restore flow_nw_tos */
6210 if (ctx
->resubmit_stats
) {
6211 netdev_vport_inc_tx(ofport
->up
.netdev
, ctx
->resubmit_stats
);
6213 out_port
= odp_port
;
6214 commit_odp_tunnel_action(&ctx
->flow
, &ctx
->base_flow
,
6216 ctx
->flow
.tunnel
= flow_tnl
; /* Restore tunnel metadata */
6218 odp_port
= ofport
->odp_port
;
6219 out_port
= vsp_realdev_to_vlandev(ctx
->ofproto
, odp_port
,
6220 ctx
->flow
.vlan_tci
);
6221 if (out_port
!= odp_port
) {
6222 ctx
->flow
.vlan_tci
= htons(0);
6224 ctx
->flow
.skb_mark
&= ~IPSEC_MARK
;
6226 commit_odp_actions(&ctx
->flow
, &ctx
->base_flow
, ctx
->odp_actions
);
6227 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_OUTPUT
, out_port
);
6229 ctx
->sflow_odp_port
= odp_port
;
6230 ctx
->sflow_n_outputs
++;
6231 ctx
->nf_output_iface
= ofp_port
;
6234 ctx
->flow
.vlan_tci
= flow_vlan_tci
;
6235 ctx
->flow
.skb_mark
= flow_skb_mark
;
6237 ctx
->flow
.nw_tos
= flow_nw_tos
;
6241 compose_output_action(struct action_xlate_ctx
*ctx
, uint16_t ofp_port
)
6243 compose_output_action__(ctx
, ofp_port
, true);
6247 tag_the_flow(struct action_xlate_ctx
*ctx
, struct rule_dpif
*rule
)
6249 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
6250 uint8_t table_id
= ctx
->table_id
;
6252 if (table_id
> 0 && table_id
< N_TABLES
) {
6253 struct table_dpif
*table
= &ofproto
->tables
[table_id
];
6254 if (table
->other_table
) {
6255 ctx
->tags
|= (rule
&& rule
->tag
6257 : rule_calculate_tag(&ctx
->flow
,
6258 &table
->other_table
->mask
,
6264 /* Common rule processing in one place to avoid duplicating code. */
6265 static struct rule_dpif
*
6266 ctx_rule_hooks(struct action_xlate_ctx
*ctx
, struct rule_dpif
*rule
,
6269 if (ctx
->resubmit_hook
) {
6270 ctx
->resubmit_hook(ctx
, rule
);
6272 if (rule
== NULL
&& may_packet_in
) {
6274 * check if table configuration flags
6275 * OFPTC_TABLE_MISS_CONTROLLER, default.
6276 * OFPTC_TABLE_MISS_CONTINUE,
6277 * OFPTC_TABLE_MISS_DROP
6278 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6280 rule
= rule_dpif_miss_rule(ctx
->ofproto
, &ctx
->flow
);
6282 if (rule
&& ctx
->resubmit_stats
) {
6283 rule_credit_stats(rule
, ctx
->resubmit_stats
);
6289 xlate_table_action(struct action_xlate_ctx
*ctx
,
6290 uint16_t in_port
, uint8_t table_id
, bool may_packet_in
)
6292 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
6293 struct rule_dpif
*rule
;
6294 uint16_t old_in_port
= ctx
->flow
.in_port
;
6295 uint8_t old_table_id
= ctx
->table_id
;
6297 ctx
->table_id
= table_id
;
6299 /* Look up a flow with 'in_port' as the input port. */
6300 ctx
->flow
.in_port
= in_port
;
6301 rule
= rule_dpif_lookup__(ctx
->ofproto
, &ctx
->flow
, table_id
);
6303 tag_the_flow(ctx
, rule
);
6305 /* Restore the original input port. Otherwise OFPP_NORMAL and
6306 * OFPP_IN_PORT will have surprising behavior. */
6307 ctx
->flow
.in_port
= old_in_port
;
6309 rule
= ctx_rule_hooks(ctx
, rule
, may_packet_in
);
6312 struct rule_dpif
*old_rule
= ctx
->rule
;
6316 do_xlate_actions(rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, ctx
);
6317 ctx
->rule
= old_rule
;
6321 ctx
->table_id
= old_table_id
;
6323 static struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
6325 VLOG_ERR_RL(&recurse_rl
, "resubmit actions recursed over %d times",
6326 MAX_RESUBMIT_RECURSION
);
6327 ctx
->max_resubmit_trigger
= true;
6332 xlate_ofpact_resubmit(struct action_xlate_ctx
*ctx
,
6333 const struct ofpact_resubmit
*resubmit
)
6338 in_port
= resubmit
->in_port
;
6339 if (in_port
== OFPP_IN_PORT
) {
6340 in_port
= ctx
->flow
.in_port
;
6343 table_id
= resubmit
->table_id
;
6344 if (table_id
== 255) {
6345 table_id
= ctx
->table_id
;
6348 xlate_table_action(ctx
, in_port
, table_id
, false);
6352 flood_packets(struct action_xlate_ctx
*ctx
, bool all
)
6354 struct ofport_dpif
*ofport
;
6356 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ctx
->ofproto
->up
.ports
) {
6357 uint16_t ofp_port
= ofport
->up
.ofp_port
;
6359 if (ofp_port
== ctx
->flow
.in_port
) {
6364 compose_output_action__(ctx
, ofp_port
, false);
6365 } else if (!(ofport
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
)) {
6366 compose_output_action(ctx
, ofp_port
);
6370 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
6374 execute_controller_action(struct action_xlate_ctx
*ctx
, int len
,
6375 enum ofp_packet_in_reason reason
,
6376 uint16_t controller_id
)
6378 struct ofputil_packet_in pin
;
6379 struct ofpbuf
*packet
;
6381 ctx
->slow
|= SLOW_CONTROLLER
;
6386 packet
= ofpbuf_clone(ctx
->packet
);
6388 if (packet
->l2
&& packet
->l3
) {
6389 struct eth_header
*eh
;
6390 uint16_t mpls_depth
;
6392 eth_pop_vlan(packet
);
6395 memcpy(eh
->eth_src
, ctx
->flow
.dl_src
, sizeof eh
->eth_src
);
6396 memcpy(eh
->eth_dst
, ctx
->flow
.dl_dst
, sizeof eh
->eth_dst
);
6398 if (ctx
->flow
.vlan_tci
& htons(VLAN_CFI
)) {
6399 eth_push_vlan(packet
, ctx
->flow
.vlan_tci
);
6402 mpls_depth
= eth_mpls_depth(packet
);
6404 if (mpls_depth
< ctx
->flow
.mpls_depth
) {
6405 push_mpls(packet
, ctx
->flow
.dl_type
, ctx
->flow
.mpls_lse
);
6406 } else if (mpls_depth
> ctx
->flow
.mpls_depth
) {
6407 pop_mpls(packet
, ctx
->flow
.dl_type
);
6408 } else if (mpls_depth
) {
6409 set_mpls_lse(packet
, ctx
->flow
.mpls_lse
);
6413 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
6414 packet_set_ipv4(packet
, ctx
->flow
.nw_src
, ctx
->flow
.nw_dst
,
6415 ctx
->flow
.nw_tos
, ctx
->flow
.nw_ttl
);
6419 if (ctx
->flow
.nw_proto
== IPPROTO_TCP
) {
6420 packet_set_tcp_port(packet
, ctx
->flow
.tp_src
,
6422 } else if (ctx
->flow
.nw_proto
== IPPROTO_UDP
) {
6423 packet_set_udp_port(packet
, ctx
->flow
.tp_src
,
6430 pin
.packet
= packet
->data
;
6431 pin
.packet_len
= packet
->size
;
6432 pin
.reason
= reason
;
6433 pin
.controller_id
= controller_id
;
6434 pin
.table_id
= ctx
->table_id
;
6435 pin
.cookie
= ctx
->rule
? ctx
->rule
->up
.flow_cookie
: 0;
6438 flow_get_metadata(&ctx
->flow
, &pin
.fmd
);
6440 connmgr_send_packet_in(ctx
->ofproto
->up
.connmgr
, &pin
);
6441 ofpbuf_delete(packet
);
6445 execute_mpls_push_action(struct action_xlate_ctx
*ctx
, ovs_be16 eth_type
)
6447 ovs_assert(eth_type_mpls(eth_type
));
6449 if (ctx
->base_flow
.mpls_depth
) {
6450 ctx
->flow
.mpls_lse
&= ~htonl(MPLS_BOS_MASK
);
6451 ctx
->flow
.mpls_depth
++;
6456 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IPV6
)) {
6457 label
= htonl(0x2); /* IPV6 Explicit Null. */
6459 label
= htonl(0x0); /* IPV4 Explicit Null. */
6461 tc
= (ctx
->flow
.nw_tos
& IP_DSCP_MASK
) >> 2;
6462 ttl
= ctx
->flow
.nw_ttl
? ctx
->flow
.nw_ttl
: 0x40;
6463 ctx
->flow
.mpls_lse
= set_mpls_lse_values(ttl
, tc
, 1, label
);
6464 ctx
->flow
.mpls_depth
= 1;
6466 ctx
->flow
.dl_type
= eth_type
;
6470 execute_mpls_pop_action(struct action_xlate_ctx
*ctx
, ovs_be16 eth_type
)
6472 ovs_assert(eth_type_mpls(ctx
->flow
.dl_type
));
6473 ovs_assert(!eth_type_mpls(eth_type
));
6475 if (ctx
->flow
.mpls_depth
) {
6476 ctx
->flow
.mpls_depth
--;
6477 ctx
->flow
.mpls_lse
= htonl(0);
6478 if (!ctx
->flow
.mpls_depth
) {
6479 ctx
->flow
.dl_type
= eth_type
;
6485 compose_dec_ttl(struct action_xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
6487 if (ctx
->flow
.dl_type
!= htons(ETH_TYPE_IP
) &&
6488 ctx
->flow
.dl_type
!= htons(ETH_TYPE_IPV6
)) {
6492 if (ctx
->flow
.nw_ttl
> 1) {
6498 for (i
= 0; i
< ids
->n_controllers
; i
++) {
6499 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
6503 /* Stop processing for current table. */
6509 execute_set_mpls_ttl_action(struct action_xlate_ctx
*ctx
, uint8_t ttl
)
6511 if (!eth_type_mpls(ctx
->flow
.dl_type
)) {
6515 set_mpls_lse_ttl(&ctx
->flow
.mpls_lse
, ttl
);
6520 execute_dec_mpls_ttl_action(struct action_xlate_ctx
*ctx
)
6522 uint8_t ttl
= mpls_lse_to_ttl(ctx
->flow
.mpls_lse
);
6524 if (!eth_type_mpls(ctx
->flow
.dl_type
)) {
6530 set_mpls_lse_ttl(&ctx
->flow
.mpls_lse
, ttl
);
6533 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0);
6535 /* Stop processing for current table. */
6541 xlate_output_action(struct action_xlate_ctx
*ctx
,
6542 uint16_t port
, uint16_t max_len
, bool may_packet_in
)
6544 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
6546 ctx
->nf_output_iface
= NF_OUT_DROP
;
6550 compose_output_action(ctx
, ctx
->flow
.in_port
);
6553 xlate_table_action(ctx
, ctx
->flow
.in_port
, 0, may_packet_in
);
6559 flood_packets(ctx
, false);
6562 flood_packets(ctx
, true);
6564 case OFPP_CONTROLLER
:
6565 execute_controller_action(ctx
, max_len
, OFPR_ACTION
, 0);
6571 if (port
!= ctx
->flow
.in_port
) {
6572 compose_output_action(ctx
, port
);
6574 xlate_report(ctx
, "skipping output to input port");
6579 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
6580 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
6581 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
6582 ctx
->nf_output_iface
= prev_nf_output_iface
;
6583 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
6584 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
6585 ctx
->nf_output_iface
= NF_OUT_MULTI
;
6590 xlate_output_reg_action(struct action_xlate_ctx
*ctx
,
6591 const struct ofpact_output_reg
*or)
6593 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->flow
);
6594 if (port
<= UINT16_MAX
) {
6595 xlate_output_action(ctx
, port
, or->max_len
, false);
6600 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
6601 const struct ofpact_enqueue
*enqueue
)
6603 uint16_t ofp_port
= enqueue
->port
;
6604 uint32_t queue_id
= enqueue
->queue
;
6605 uint32_t flow_priority
, priority
;
6608 /* Translate queue to priority. */
6609 error
= dpif_queue_to_priority(ctx
->ofproto
->backer
->dpif
,
6610 queue_id
, &priority
);
6612 /* Fall back to ordinary output action. */
6613 xlate_output_action(ctx
, enqueue
->port
, 0, false);
6617 /* Check output port. */
6618 if (ofp_port
== OFPP_IN_PORT
) {
6619 ofp_port
= ctx
->flow
.in_port
;
6620 } else if (ofp_port
== ctx
->flow
.in_port
) {
6624 /* Add datapath actions. */
6625 flow_priority
= ctx
->flow
.skb_priority
;
6626 ctx
->flow
.skb_priority
= priority
;
6627 compose_output_action(ctx
, ofp_port
);
6628 ctx
->flow
.skb_priority
= flow_priority
;
6630 /* Update NetFlow output port. */
6631 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
6632 ctx
->nf_output_iface
= ofp_port
;
6633 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
6634 ctx
->nf_output_iface
= NF_OUT_MULTI
;
6639 xlate_set_queue_action(struct action_xlate_ctx
*ctx
, uint32_t queue_id
)
6641 uint32_t skb_priority
;
6643 if (!dpif_queue_to_priority(ctx
->ofproto
->backer
->dpif
,
6644 queue_id
, &skb_priority
)) {
6645 ctx
->flow
.skb_priority
= skb_priority
;
6647 /* Couldn't translate queue to a priority. Nothing to do. A warning
6648 * has already been logged. */
6653 slave_enabled_cb(uint16_t ofp_port
, void *ofproto_
)
6655 struct ofproto_dpif
*ofproto
= ofproto_
;
6656 struct ofport_dpif
*port
;
6666 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
6669 port
= get_ofp_port(ofproto
, ofp_port
);
6670 return port
? port
->may_enable
: false;
6675 xlate_bundle_action(struct action_xlate_ctx
*ctx
,
6676 const struct ofpact_bundle
*bundle
)
6680 port
= bundle_execute(bundle
, &ctx
->flow
, slave_enabled_cb
, ctx
->ofproto
);
6681 if (bundle
->dst
.field
) {
6682 nxm_reg_load(&bundle
->dst
, port
, &ctx
->flow
);
6684 xlate_output_action(ctx
, port
, 0, false);
6689 xlate_learn_action(struct action_xlate_ctx
*ctx
,
6690 const struct ofpact_learn
*learn
)
6692 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 1);
6693 struct ofputil_flow_mod fm
;
6694 uint64_t ofpacts_stub
[1024 / 8];
6695 struct ofpbuf ofpacts
;
6698 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
6699 learn_execute(learn
, &ctx
->flow
, &fm
, &ofpacts
);
6701 error
= ofproto_flow_mod(&ctx
->ofproto
->up
, &fm
);
6702 if (error
&& !VLOG_DROP_WARN(&rl
)) {
6703 VLOG_WARN("learning action failed to modify flow table (%s)",
6704 ofperr_get_name(error
));
6707 ofpbuf_uninit(&ofpacts
);
6710 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6711 * means "infinite". */
6713 reduce_timeout(uint16_t max
, uint16_t *timeout
)
6715 if (max
&& (!*timeout
|| *timeout
> max
)) {
6721 xlate_fin_timeout(struct action_xlate_ctx
*ctx
,
6722 const struct ofpact_fin_timeout
*oft
)
6724 if (ctx
->tcp_flags
& (TCP_FIN
| TCP_RST
) && ctx
->rule
) {
6725 struct rule_dpif
*rule
= ctx
->rule
;
6727 reduce_timeout(oft
->fin_idle_timeout
, &rule
->up
.idle_timeout
);
6728 reduce_timeout(oft
->fin_hard_timeout
, &rule
->up
.hard_timeout
);
6733 xlate_sample_action(struct action_xlate_ctx
*ctx
,
6734 const struct ofpact_sample
*os
)
6736 union user_action_cookie cookie
;
6737 /* Scale the probability from 16-bit to 32-bit while representing
6738 * the same percentage. */
6739 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
6741 commit_odp_actions(&ctx
->flow
, &ctx
->base_flow
, ctx
->odp_actions
);
6743 compose_flow_sample_cookie(os
->probability
, os
->collector_set_id
,
6744 os
->obs_domain_id
, os
->obs_point_id
, &cookie
);
6745 compose_sample_action(ctx
->ofproto
, ctx
->odp_actions
, &ctx
->flow
,
6746 probability
, &cookie
, sizeof cookie
.flow_sample
);
6750 may_receive(const struct ofport_dpif
*port
, struct action_xlate_ctx
*ctx
)
6752 if (port
->up
.pp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
6753 ? OFPUTIL_PC_NO_RECV_STP
6754 : OFPUTIL_PC_NO_RECV
)) {
6758 /* Only drop packets here if both forwarding and learning are
6759 * disabled. If just learning is enabled, we need to have
6760 * OFPP_NORMAL and the learning action have a look at the packet
6761 * before we can drop it. */
6762 if (!stp_forward_in_state(port
->stp_state
)
6763 && !stp_learn_in_state(port
->stp_state
)) {
6771 tunnel_ecn_ok(struct action_xlate_ctx
*ctx
)
6773 if (is_ip_any(&ctx
->base_flow
)
6774 && (ctx
->flow
.tunnel
.ip_tos
& IP_ECN_MASK
) == IP_ECN_CE
) {
6775 if ((ctx
->base_flow
.nw_tos
& IP_ECN_MASK
) == IP_ECN_NOT_ECT
) {
6776 VLOG_WARN_RL(&rl
, "dropping tunnel packet marked ECN CE"
6777 " but is not ECN capable");
6780 /* Set the ECN CE value in the tunneled packet. */
6781 ctx
->flow
.nw_tos
|= IP_ECN_CE
;
6789 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
6790 struct action_xlate_ctx
*ctx
)
6792 bool was_evictable
= true;
6793 const struct ofpact
*a
;
6796 /* Don't let the rule we're working on get evicted underneath us. */
6797 was_evictable
= ctx
->rule
->up
.evictable
;
6798 ctx
->rule
->up
.evictable
= false;
6801 do_xlate_actions_again
:
6802 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
6803 struct ofpact_controller
*controller
;
6804 const struct ofpact_metadata
*metadata
;
6812 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
6813 ofpact_get_OUTPUT(a
)->max_len
, true);
6816 case OFPACT_CONTROLLER
:
6817 controller
= ofpact_get_CONTROLLER(a
);
6818 execute_controller_action(ctx
, controller
->max_len
,
6820 controller
->controller_id
);
6823 case OFPACT_ENQUEUE
:
6824 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
));
6827 case OFPACT_SET_VLAN_VID
:
6828 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
6829 ctx
->flow
.vlan_tci
|= (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
)
6833 case OFPACT_SET_VLAN_PCP
:
6834 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
6835 ctx
->flow
.vlan_tci
|= htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
6840 case OFPACT_STRIP_VLAN
:
6841 ctx
->flow
.vlan_tci
= htons(0);
6844 case OFPACT_PUSH_VLAN
:
6845 /* XXX 802.1AD(QinQ) */
6846 ctx
->flow
.vlan_tci
= htons(VLAN_CFI
);
6849 case OFPACT_SET_ETH_SRC
:
6850 memcpy(ctx
->flow
.dl_src
, ofpact_get_SET_ETH_SRC(a
)->mac
,
6854 case OFPACT_SET_ETH_DST
:
6855 memcpy(ctx
->flow
.dl_dst
, ofpact_get_SET_ETH_DST(a
)->mac
,
6859 case OFPACT_SET_IPV4_SRC
:
6860 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
6861 ctx
->flow
.nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
6865 case OFPACT_SET_IPV4_DST
:
6866 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
6867 ctx
->flow
.nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
6871 case OFPACT_SET_IPV4_DSCP
:
6872 /* OpenFlow 1.0 only supports IPv4. */
6873 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_IP
)) {
6874 ctx
->flow
.nw_tos
&= ~IP_DSCP_MASK
;
6875 ctx
->flow
.nw_tos
|= ofpact_get_SET_IPV4_DSCP(a
)->dscp
;
6879 case OFPACT_SET_L4_SRC_PORT
:
6880 if (is_ip_any(&ctx
->flow
)) {
6881 ctx
->flow
.tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
6885 case OFPACT_SET_L4_DST_PORT
:
6886 if (is_ip_any(&ctx
->flow
)) {
6887 ctx
->flow
.tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
6891 case OFPACT_RESUBMIT
:
6892 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
));
6895 case OFPACT_SET_TUNNEL
:
6896 ctx
->flow
.tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
6899 case OFPACT_SET_QUEUE
:
6900 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
6903 case OFPACT_POP_QUEUE
:
6904 ctx
->flow
.skb_priority
= ctx
->orig_skb_priority
;
6907 case OFPACT_REG_MOVE
:
6908 nxm_execute_reg_move(ofpact_get_REG_MOVE(a
), &ctx
->flow
);
6911 case OFPACT_REG_LOAD
:
6912 nxm_execute_reg_load(ofpact_get_REG_LOAD(a
), &ctx
->flow
);
6915 case OFPACT_STACK_PUSH
:
6916 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), &ctx
->flow
,
6920 case OFPACT_STACK_POP
:
6921 nxm_execute_stack_pop(ofpact_get_STACK_POP(a
), &ctx
->flow
,
6925 case OFPACT_PUSH_MPLS
:
6926 execute_mpls_push_action(ctx
, ofpact_get_PUSH_MPLS(a
)->ethertype
);
6929 case OFPACT_POP_MPLS
:
6930 execute_mpls_pop_action(ctx
, ofpact_get_POP_MPLS(a
)->ethertype
);
6933 case OFPACT_SET_MPLS_TTL
:
6934 if (execute_set_mpls_ttl_action(ctx
, ofpact_get_SET_MPLS_TTL(a
)->ttl
)) {
6939 case OFPACT_DEC_MPLS_TTL
:
6940 if (execute_dec_mpls_ttl_action(ctx
)) {
6945 case OFPACT_DEC_TTL
:
6946 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
6952 /* Nothing to do. */
6955 case OFPACT_MULTIPATH
:
6956 multipath_execute(ofpact_get_MULTIPATH(a
), &ctx
->flow
);
6960 ctx
->ofproto
->has_bundle_action
= true;
6961 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
));
6964 case OFPACT_OUTPUT_REG
:
6965 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
));
6969 ctx
->has_learn
= true;
6970 if (ctx
->may_learn
) {
6971 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
6979 case OFPACT_FIN_TIMEOUT
:
6980 ctx
->has_fin_timeout
= true;
6981 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
6984 case OFPACT_CLEAR_ACTIONS
:
6986 * Nothing to do because writa-actions is not supported for now.
6987 * When writa-actions is supported, clear-actions also must
6988 * be supported at the same time.
6992 case OFPACT_WRITE_METADATA
:
6993 metadata
= ofpact_get_WRITE_METADATA(a
);
6994 ctx
->flow
.metadata
&= ~metadata
->mask
;
6995 ctx
->flow
.metadata
|= metadata
->metadata
& metadata
->mask
;
6998 case OFPACT_GOTO_TABLE
: {
6999 /* It is assumed that goto-table is the last action. */
7000 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
7001 struct rule_dpif
*rule
;
7003 ovs_assert(ctx
->table_id
< ogt
->table_id
);
7005 ctx
->table_id
= ogt
->table_id
;
7007 /* Look up a flow from the new table. */
7008 rule
= rule_dpif_lookup__(ctx
->ofproto
, &ctx
->flow
, ctx
->table_id
);
7010 tag_the_flow(ctx
, rule
);
7012 rule
= ctx_rule_hooks(ctx
, rule
, true);
7016 ctx
->rule
->up
.evictable
= was_evictable
;
7019 was_evictable
= rule
->up
.evictable
;
7020 rule
->up
.evictable
= false;
7022 /* Tail recursion removal. */
7023 ofpacts
= rule
->up
.ofpacts
;
7024 ofpacts_len
= rule
->up
.ofpacts_len
;
7025 goto do_xlate_actions_again
;
7031 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
7038 ctx
->rule
->up
.evictable
= was_evictable
;
7043 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
7044 struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
7045 const struct initial_vals
*initial_vals
,
7046 struct rule_dpif
*rule
,
7047 uint8_t tcp_flags
, const struct ofpbuf
*packet
)
7049 /* Flow initialization rules:
7050 * - 'base_flow' must match the kernel's view of the packet at the
7051 * time that action processing starts. 'flow' represents any
7052 * transformations we wish to make through actions.
7053 * - By default 'base_flow' and 'flow' are the same since the input
7054 * packet matches the output before any actions are applied.
7055 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7056 * of the received packet as seen by the kernel. If we later output
7057 * to another device without any modifications this will cause us to
7058 * insert a new tag since the original one was stripped off by the
7060 * - Tunnel metadata as received is retained in 'flow'. This allows
7061 * tunnel metadata matching also in later tables.
7062 * Since a kernel action for setting the tunnel metadata will only be
7063 * generated with actual tunnel output, changing the tunnel metadata
7064 * values in 'flow' (such as tun_id) will only have effect with a later
7065 * tunnel output action.
7066 * - Tunnel 'base_flow' is completely cleared since that is what the
7067 * kernel does. If we wish to maintain the original values an action
7068 * needs to be generated. */
7070 ctx
->ofproto
= ofproto
;
7072 ctx
->base_flow
= ctx
->flow
;
7073 memset(&ctx
->base_flow
.tunnel
, 0, sizeof ctx
->base_flow
.tunnel
);
7074 ctx
->base_flow
.vlan_tci
= initial_vals
->vlan_tci
;
7076 ctx
->packet
= packet
;
7077 ctx
->may_learn
= packet
!= NULL
;
7078 ctx
->tcp_flags
= tcp_flags
;
7079 ctx
->resubmit_hook
= NULL
;
7080 ctx
->report_hook
= NULL
;
7081 ctx
->resubmit_stats
= NULL
;
7084 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7085 * into datapath actions in 'odp_actions', using 'ctx'. */
7087 xlate_actions(struct action_xlate_ctx
*ctx
,
7088 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
7089 struct ofpbuf
*odp_actions
)
7091 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7092 * that in the future we always keep a copy of the original flow for
7093 * tracing purposes. */
7094 static bool hit_resubmit_limit
;
7096 enum slow_path_reason special
;
7097 struct ofport_dpif
*in_port
;
7098 struct flow orig_flow
;
7100 COVERAGE_INC(ofproto_dpif_xlate
);
7102 ofpbuf_clear(odp_actions
);
7103 ofpbuf_reserve(odp_actions
, NL_A_U32_SIZE
);
7105 ctx
->odp_actions
= odp_actions
;
7108 ctx
->has_learn
= false;
7109 ctx
->has_normal
= false;
7110 ctx
->has_fin_timeout
= false;
7111 ctx
->nf_output_iface
= NF_OUT_DROP
;
7114 ctx
->max_resubmit_trigger
= false;
7115 ctx
->orig_skb_priority
= ctx
->flow
.skb_priority
;
7119 ofpbuf_use_stub(&ctx
->stack
, ctx
->init_stack
, sizeof ctx
->init_stack
);
7121 if (ctx
->ofproto
->has_mirrors
|| hit_resubmit_limit
) {
7122 /* Do this conditionally because the copy is expensive enough that it
7123 * shows up in profiles. */
7124 orig_flow
= ctx
->flow
;
7127 if (ctx
->flow
.nw_frag
& FLOW_NW_FRAG_ANY
) {
7128 switch (ctx
->ofproto
->up
.frag_handling
) {
7129 case OFPC_FRAG_NORMAL
:
7130 /* We must pretend that transport ports are unavailable. */
7131 ctx
->flow
.tp_src
= ctx
->base_flow
.tp_src
= htons(0);
7132 ctx
->flow
.tp_dst
= ctx
->base_flow
.tp_dst
= htons(0);
7135 case OFPC_FRAG_DROP
:
7138 case OFPC_FRAG_REASM
:
7141 case OFPC_FRAG_NX_MATCH
:
7142 /* Nothing to do. */
7145 case OFPC_INVALID_TTL_TO_CONTROLLER
:
7150 in_port
= get_ofp_port(ctx
->ofproto
, ctx
->flow
.in_port
);
7151 special
= process_special(ctx
->ofproto
, &ctx
->flow
, in_port
, ctx
->packet
);
7153 ctx
->slow
|= special
;
7155 static struct vlog_rate_limit trace_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
7156 struct initial_vals initial_vals
;
7157 size_t sample_actions_len
;
7158 uint32_t local_odp_port
;
7160 initial_vals
.vlan_tci
= ctx
->base_flow
.vlan_tci
;
7162 add_sflow_action(ctx
);
7163 add_ipfix_action(ctx
);
7164 sample_actions_len
= ctx
->odp_actions
->size
;
7166 if (tunnel_ecn_ok(ctx
) && (!in_port
|| may_receive(in_port
, ctx
))) {
7167 do_xlate_actions(ofpacts
, ofpacts_len
, ctx
);
7169 /* We've let OFPP_NORMAL and the learning action look at the
7170 * packet, so drop it now if forwarding is disabled. */
7171 if (in_port
&& !stp_forward_in_state(in_port
->stp_state
)) {
7172 ctx
->odp_actions
->size
= sample_actions_len
;
7176 if (ctx
->max_resubmit_trigger
&& !ctx
->resubmit_hook
) {
7177 if (!hit_resubmit_limit
) {
7178 /* We didn't record the original flow. Make sure we do from
7180 hit_resubmit_limit
= true;
7181 } else if (!VLOG_DROP_ERR(&trace_rl
)) {
7182 struct ds ds
= DS_EMPTY_INITIALIZER
;
7184 ofproto_trace(ctx
->ofproto
, &orig_flow
, ctx
->packet
,
7185 &initial_vals
, &ds
);
7186 VLOG_ERR("Trace triggered by excessive resubmit "
7187 "recursion:\n%s", ds_cstr(&ds
));
7192 local_odp_port
= ofp_port_to_odp_port(ctx
->ofproto
, OFPP_LOCAL
);
7193 if (!connmgr_may_set_up_flow(ctx
->ofproto
->up
.connmgr
, &ctx
->flow
,
7195 ctx
->odp_actions
->data
,
7196 ctx
->odp_actions
->size
)) {
7197 ctx
->slow
|= SLOW_IN_BAND
;
7199 && connmgr_msg_in_hook(ctx
->ofproto
->up
.connmgr
, &ctx
->flow
,
7201 compose_output_action(ctx
, OFPP_LOCAL
);
7204 if (ctx
->ofproto
->has_mirrors
) {
7205 add_mirror_actions(ctx
, &orig_flow
);
7207 fix_sflow_action(ctx
);
7210 ofpbuf_uninit(&ctx
->stack
);
7213 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7214 * into datapath actions, using 'ctx', and discards the datapath actions. */
7216 xlate_actions_for_side_effects(struct action_xlate_ctx
*ctx
,
7217 const struct ofpact
*ofpacts
,
7220 uint64_t odp_actions_stub
[1024 / 8];
7221 struct ofpbuf odp_actions
;
7223 ofpbuf_use_stub(&odp_actions
, odp_actions_stub
, sizeof odp_actions_stub
);
7224 xlate_actions(ctx
, ofpacts
, ofpacts_len
, &odp_actions
);
7225 ofpbuf_uninit(&odp_actions
);
7229 xlate_report(struct action_xlate_ctx
*ctx
, const char *s
)
7231 if (ctx
->report_hook
) {
7232 ctx
->report_hook(ctx
, s
);
7236 /* OFPP_NORMAL implementation. */
7238 static struct ofport_dpif
*ofbundle_get_a_port(const struct ofbundle
*);
7240 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7241 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7242 * the bundle on which the packet was received, returns the VLAN to which the
7245 * Both 'vid' and the return value are in the range 0...4095. */
7247 input_vid_to_vlan(const struct ofbundle
*in_bundle
, uint16_t vid
)
7249 switch (in_bundle
->vlan_mode
) {
7250 case PORT_VLAN_ACCESS
:
7251 return in_bundle
->vlan
;
7254 case PORT_VLAN_TRUNK
:
7257 case PORT_VLAN_NATIVE_UNTAGGED
:
7258 case PORT_VLAN_NATIVE_TAGGED
:
7259 return vid
? vid
: in_bundle
->vlan
;
7266 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7267 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7270 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7271 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7274 input_vid_is_valid(uint16_t vid
, struct ofbundle
*in_bundle
, bool warn
)
7276 /* Allow any VID on the OFPP_NONE port. */
7277 if (in_bundle
== &ofpp_none_bundle
) {
7281 switch (in_bundle
->vlan_mode
) {
7282 case PORT_VLAN_ACCESS
:
7285 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7286 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" tagged "
7287 "packet received on port %s configured as VLAN "
7288 "%"PRIu16
" access port",
7289 in_bundle
->ofproto
->up
.name
, vid
,
7290 in_bundle
->name
, in_bundle
->vlan
);
7296 case PORT_VLAN_NATIVE_UNTAGGED
:
7297 case PORT_VLAN_NATIVE_TAGGED
:
7299 /* Port must always carry its native VLAN. */
7303 case PORT_VLAN_TRUNK
:
7304 if (!ofbundle_includes_vlan(in_bundle
, vid
)) {
7306 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7307 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" packet "
7308 "received on port %s not configured for trunking "
7310 in_bundle
->ofproto
->up
.name
, vid
,
7311 in_bundle
->name
, vid
);
7323 /* Given 'vlan', the VLAN that a packet belongs to, and
7324 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7325 * that should be included in the 802.1Q header. (If the return value is 0,
7326 * then the 802.1Q header should only be included in the packet if there is a
7329 * Both 'vlan' and the return value are in the range 0...4095. */
7331 output_vlan_to_vid(const struct ofbundle
*out_bundle
, uint16_t vlan
)
7333 switch (out_bundle
->vlan_mode
) {
7334 case PORT_VLAN_ACCESS
:
7337 case PORT_VLAN_TRUNK
:
7338 case PORT_VLAN_NATIVE_TAGGED
:
7341 case PORT_VLAN_NATIVE_UNTAGGED
:
7342 return vlan
== out_bundle
->vlan
? 0 : vlan
;
7350 output_normal(struct action_xlate_ctx
*ctx
, const struct ofbundle
*out_bundle
,
7353 struct ofport_dpif
*port
;
7355 ovs_be16 tci
, old_tci
;
7357 vid
= output_vlan_to_vid(out_bundle
, vlan
);
7358 if (!out_bundle
->bond
) {
7359 port
= ofbundle_get_a_port(out_bundle
);
7361 port
= bond_choose_output_slave(out_bundle
->bond
, &ctx
->flow
,
7364 /* No slaves enabled, so drop packet. */
7369 old_tci
= ctx
->flow
.vlan_tci
;
7371 if (tci
|| out_bundle
->use_priority_tags
) {
7372 tci
|= ctx
->flow
.vlan_tci
& htons(VLAN_PCP_MASK
);
7374 tci
|= htons(VLAN_CFI
);
7377 ctx
->flow
.vlan_tci
= tci
;
7379 compose_output_action(ctx
, port
->up
.ofp_port
);
7380 ctx
->flow
.vlan_tci
= old_tci
;
7384 mirror_mask_ffs(mirror_mask_t mask
)
7386 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask
));
7391 ofbundle_trunks_vlan(const struct ofbundle
*bundle
, uint16_t vlan
)
7393 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
7394 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
7398 ofbundle_includes_vlan(const struct ofbundle
*bundle
, uint16_t vlan
)
7400 return vlan
== bundle
->vlan
|| ofbundle_trunks_vlan(bundle
, vlan
);
7403 /* Returns an arbitrary interface within 'bundle'. */
7404 static struct ofport_dpif
*
7405 ofbundle_get_a_port(const struct ofbundle
*bundle
)
7407 return CONTAINER_OF(list_front(&bundle
->ports
),
7408 struct ofport_dpif
, bundle_node
);
7412 vlan_is_mirrored(const struct ofmirror
*m
, int vlan
)
7414 return !m
->vlans
|| bitmap_is_set(m
->vlans
, vlan
);
7418 add_mirror_actions(struct action_xlate_ctx
*ctx
, const struct flow
*orig_flow
)
7420 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
7421 mirror_mask_t mirrors
;
7422 struct ofbundle
*in_bundle
;
7425 const struct nlattr
*a
;
7428 in_bundle
= lookup_input_bundle(ctx
->ofproto
, orig_flow
->in_port
,
7429 ctx
->packet
!= NULL
, NULL
);
7433 mirrors
= in_bundle
->src_mirrors
;
7435 /* Drop frames on bundles reserved for mirroring. */
7436 if (in_bundle
->mirror_out
) {
7437 if (ctx
->packet
!= NULL
) {
7438 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7439 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
7440 "%s, which is reserved exclusively for mirroring",
7441 ctx
->ofproto
->up
.name
, in_bundle
->name
);
7447 vid
= vlan_tci_to_vid(orig_flow
->vlan_tci
);
7448 if (!input_vid_is_valid(vid
, in_bundle
, ctx
->packet
!= NULL
)) {
7451 vlan
= input_vid_to_vlan(in_bundle
, vid
);
7453 /* Look at the output ports to check for destination selections. */
7455 NL_ATTR_FOR_EACH (a
, left
, ctx
->odp_actions
->data
,
7456 ctx
->odp_actions
->size
) {
7457 enum ovs_action_attr type
= nl_attr_type(a
);
7458 struct ofport_dpif
*ofport
;
7460 if (type
!= OVS_ACTION_ATTR_OUTPUT
) {
7464 ofport
= get_odp_port(ofproto
, nl_attr_get_u32(a
));
7465 if (ofport
&& ofport
->bundle
) {
7466 mirrors
|= ofport
->bundle
->dst_mirrors
;
7474 /* Restore the original packet before adding the mirror actions. */
7475 ctx
->flow
= *orig_flow
;
7480 m
= ofproto
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
7482 if (!vlan_is_mirrored(m
, vlan
)) {
7483 mirrors
= zero_rightmost_1bit(mirrors
);
7487 mirrors
&= ~m
->dup_mirrors
;
7488 ctx
->mirrors
|= m
->dup_mirrors
;
7490 output_normal(ctx
, m
->out
, vlan
);
7491 } else if (vlan
!= m
->out_vlan
7492 && !eth_addr_is_reserved(orig_flow
->dl_dst
)) {
7493 struct ofbundle
*bundle
;
7495 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
7496 if (ofbundle_includes_vlan(bundle
, m
->out_vlan
)
7497 && !bundle
->mirror_out
) {
7498 output_normal(ctx
, bundle
, m
->out_vlan
);
7506 update_mirror_stats(struct ofproto_dpif
*ofproto
, mirror_mask_t mirrors
,
7507 uint64_t packets
, uint64_t bytes
)
7513 for (; mirrors
; mirrors
= zero_rightmost_1bit(mirrors
)) {
7516 m
= ofproto
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
7519 /* In normal circumstances 'm' will not be NULL. However,
7520 * if mirrors are reconfigured, we can temporarily get out
7521 * of sync in facet_revalidate(). We could "correct" the
7522 * mirror list before reaching here, but doing that would
7523 * not properly account the traffic stats we've currently
7524 * accumulated for previous mirror configuration. */
7528 m
->packet_count
+= packets
;
7529 m
->byte_count
+= bytes
;
7533 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7534 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7535 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7537 is_gratuitous_arp(const struct flow
*flow
)
7539 return (flow
->dl_type
== htons(ETH_TYPE_ARP
)
7540 && eth_addr_is_broadcast(flow
->dl_dst
)
7541 && (flow
->nw_proto
== ARP_OP_REPLY
7542 || (flow
->nw_proto
== ARP_OP_REQUEST
7543 && flow
->nw_src
== flow
->nw_dst
)));
7547 update_learning_table(struct ofproto_dpif
*ofproto
,
7548 const struct flow
*flow
, int vlan
,
7549 struct ofbundle
*in_bundle
)
7551 struct mac_entry
*mac
;
7553 /* Don't learn the OFPP_NONE port. */
7554 if (in_bundle
== &ofpp_none_bundle
) {
7558 if (!mac_learning_may_learn(ofproto
->ml
, flow
->dl_src
, vlan
)) {
7562 mac
= mac_learning_insert(ofproto
->ml
, flow
->dl_src
, vlan
);
7563 if (is_gratuitous_arp(flow
)) {
7564 /* We don't want to learn from gratuitous ARP packets that are
7565 * reflected back over bond slaves so we lock the learning table. */
7566 if (!in_bundle
->bond
) {
7567 mac_entry_set_grat_arp_lock(mac
);
7568 } else if (mac_entry_is_grat_arp_locked(mac
)) {
7573 if (mac_entry_is_new(mac
) || mac
->port
.p
!= in_bundle
) {
7574 /* The log messages here could actually be useful in debugging,
7575 * so keep the rate limit relatively high. */
7576 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
7577 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
7578 "on port %s in VLAN %d",
7579 ofproto
->up
.name
, ETH_ADDR_ARGS(flow
->dl_src
),
7580 in_bundle
->name
, vlan
);
7582 mac
->port
.p
= in_bundle
;
7583 tag_set_add(&ofproto
->backer
->revalidate_set
,
7584 mac_learning_changed(ofproto
->ml
, mac
));
7588 static struct ofbundle
*
7589 lookup_input_bundle(const struct ofproto_dpif
*ofproto
, uint16_t in_port
,
7590 bool warn
, struct ofport_dpif
**in_ofportp
)
7592 struct ofport_dpif
*ofport
;
7594 /* Find the port and bundle for the received packet. */
7595 ofport
= get_ofp_port(ofproto
, in_port
);
7597 *in_ofportp
= ofport
;
7599 if (ofport
&& ofport
->bundle
) {
7600 return ofport
->bundle
;
7603 /* Special-case OFPP_NONE, which a controller may use as the ingress
7604 * port for traffic that it is sourcing. */
7605 if (in_port
== OFPP_NONE
) {
7606 return &ofpp_none_bundle
;
7609 /* Odd. A few possible reasons here:
7611 * - We deleted a port but there are still a few packets queued up
7614 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7615 * we don't know about.
7617 * - The ofproto client didn't configure the port as part of a bundle.
7618 * This is particularly likely to happen if a packet was received on the
7619 * port after it was created, but before the client had a chance to
7620 * configure its bundle.
7623 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7625 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
7626 "port %"PRIu16
, ofproto
->up
.name
, in_port
);
7631 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7632 * dropped. Returns true if they may be forwarded, false if they should be
7635 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7636 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7638 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7639 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7640 * checked by input_vid_is_valid().
7642 * May also add tags to '*tags', although the current implementation only does
7643 * so in one special case.
7646 is_admissible(struct action_xlate_ctx
*ctx
, struct ofport_dpif
*in_port
,
7649 struct ofproto_dpif
*ofproto
= ctx
->ofproto
;
7650 struct flow
*flow
= &ctx
->flow
;
7651 struct ofbundle
*in_bundle
= in_port
->bundle
;
7653 /* Drop frames for reserved multicast addresses
7654 * only if forward_bpdu option is absent. */
7655 if (!ofproto
->up
.forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
7656 xlate_report(ctx
, "packet has reserved destination MAC, dropping");
7660 if (in_bundle
->bond
) {
7661 struct mac_entry
*mac
;
7663 switch (bond_check_admissibility(in_bundle
->bond
, in_port
,
7664 flow
->dl_dst
, &ctx
->tags
)) {
7669 xlate_report(ctx
, "bonding refused admissibility, dropping");
7672 case BV_DROP_IF_MOVED
:
7673 mac
= mac_learning_lookup(ofproto
->ml
, flow
->dl_src
, vlan
, NULL
);
7674 if (mac
&& mac
->port
.p
!= in_bundle
&&
7675 (!is_gratuitous_arp(flow
)
7676 || mac_entry_is_grat_arp_locked(mac
))) {
7677 xlate_report(ctx
, "SLB bond thinks this packet looped back, "
7689 xlate_normal(struct action_xlate_ctx
*ctx
)
7691 struct ofport_dpif
*in_port
;
7692 struct ofbundle
*in_bundle
;
7693 struct mac_entry
*mac
;
7697 ctx
->has_normal
= true;
7699 in_bundle
= lookup_input_bundle(ctx
->ofproto
, ctx
->flow
.in_port
,
7700 ctx
->packet
!= NULL
, &in_port
);
7702 xlate_report(ctx
, "no input bundle, dropping");
7706 /* Drop malformed frames. */
7707 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_VLAN
) &&
7708 !(ctx
->flow
.vlan_tci
& htons(VLAN_CFI
))) {
7709 if (ctx
->packet
!= NULL
) {
7710 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7711 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet with partial "
7712 "VLAN tag received on port %s",
7713 ctx
->ofproto
->up
.name
, in_bundle
->name
);
7715 xlate_report(ctx
, "partial VLAN tag, dropping");
7719 /* Drop frames on bundles reserved for mirroring. */
7720 if (in_bundle
->mirror_out
) {
7721 if (ctx
->packet
!= NULL
) {
7722 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
7723 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
7724 "%s, which is reserved exclusively for mirroring",
7725 ctx
->ofproto
->up
.name
, in_bundle
->name
);
7727 xlate_report(ctx
, "input port is mirror output port, dropping");
7732 vid
= vlan_tci_to_vid(ctx
->flow
.vlan_tci
);
7733 if (!input_vid_is_valid(vid
, in_bundle
, ctx
->packet
!= NULL
)) {
7734 xlate_report(ctx
, "disallowed VLAN VID for this input port, dropping");
7737 vlan
= input_vid_to_vlan(in_bundle
, vid
);
7739 /* Check other admissibility requirements. */
7740 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
7744 /* Learn source MAC. */
7745 if (ctx
->may_learn
) {
7746 update_learning_table(ctx
->ofproto
, &ctx
->flow
, vlan
, in_bundle
);
7749 /* Determine output bundle. */
7750 mac
= mac_learning_lookup(ctx
->ofproto
->ml
, ctx
->flow
.dl_dst
, vlan
,
7753 if (mac
->port
.p
!= in_bundle
) {
7754 xlate_report(ctx
, "forwarding to learned port");
7755 output_normal(ctx
, mac
->port
.p
, vlan
);
7757 xlate_report(ctx
, "learned port is input port, dropping");
7760 struct ofbundle
*bundle
;
7762 xlate_report(ctx
, "no learned MAC for destination, flooding");
7763 HMAP_FOR_EACH (bundle
, hmap_node
, &ctx
->ofproto
->bundles
) {
7764 if (bundle
!= in_bundle
7765 && ofbundle_includes_vlan(bundle
, vlan
)
7766 && bundle
->floodable
7767 && !bundle
->mirror_out
) {
7768 output_normal(ctx
, bundle
, vlan
);
7771 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
7775 /* Optimized flow revalidation.
7777 * It's a difficult problem, in general, to tell which facets need to have
7778 * their actions recalculated whenever the OpenFlow flow table changes. We
7779 * don't try to solve that general problem: for most kinds of OpenFlow flow
7780 * table changes, we recalculate the actions for every facet. This is
7781 * relatively expensive, but it's good enough if the OpenFlow flow table
7782 * doesn't change very often.
7784 * However, we can expect one particular kind of OpenFlow flow table change to
7785 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7786 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7787 * table, we add a special case that applies to flow tables in which every rule
7788 * has the same form (that is, the same wildcards), except that the table is
7789 * also allowed to have a single "catch-all" flow that matches all packets. We
7790 * optimize this case by tagging all of the facets that resubmit into the table
7791 * and invalidating the same tag whenever a flow changes in that table. The
7792 * end result is that we revalidate just the facets that need it (and sometimes
7793 * a few more, but not all of the facets or even all of the facets that
7794 * resubmit to the table modified by MAC learning). */
7796 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7797 * into an OpenFlow table with the given 'basis'. */
7799 rule_calculate_tag(const struct flow
*flow
, const struct minimask
*mask
,
7802 if (minimask_is_catchall(mask
)) {
7805 uint32_t hash
= flow_hash_in_minimask(flow
, mask
, secret
);
7806 return tag_create_deterministic(hash
);
7810 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7811 * taggability of that table.
7813 * This function must be called after *each* change to a flow table. If you
7814 * skip calling it on some changes then the pointer comparisons at the end can
7815 * be invalid if you get unlucky. For example, if a flow removal causes a
7816 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7817 * different wildcards to be created with the same address, then this function
7818 * will incorrectly skip revalidation. */
7820 table_update_taggable(struct ofproto_dpif
*ofproto
, uint8_t table_id
)
7822 struct table_dpif
*table
= &ofproto
->tables
[table_id
];
7823 const struct oftable
*oftable
= &ofproto
->up
.tables
[table_id
];
7824 struct cls_table
*catchall
, *other
;
7825 struct cls_table
*t
;
7827 catchall
= other
= NULL
;
7829 switch (hmap_count(&oftable
->cls
.tables
)) {
7831 /* We could tag this OpenFlow table but it would make the logic a
7832 * little harder and it's a corner case that doesn't seem worth it
7838 HMAP_FOR_EACH (t
, hmap_node
, &oftable
->cls
.tables
) {
7839 if (cls_table_is_catchall(t
)) {
7841 } else if (!other
) {
7844 /* Indicate that we can't tag this by setting both tables to
7845 * NULL. (We know that 'catchall' is already NULL.) */
7852 /* Can't tag this table. */
7856 if (table
->catchall_table
!= catchall
|| table
->other_table
!= other
) {
7857 table
->catchall_table
= catchall
;
7858 table
->other_table
= other
;
7859 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
7863 /* Given 'rule' that has changed in some way (either it is a rule being
7864 * inserted, a rule being deleted, or a rule whose actions are being
7865 * modified), marks facets for revalidation to ensure that packets will be
7866 * forwarded correctly according to the new state of the flow table.
7868 * This function must be called after *each* change to a flow table. See
7869 * the comment on table_update_taggable() for more information. */
7871 rule_invalidate(const struct rule_dpif
*rule
)
7873 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
7875 table_update_taggable(ofproto
, rule
->up
.table_id
);
7877 if (!ofproto
->backer
->need_revalidate
) {
7878 struct table_dpif
*table
= &ofproto
->tables
[rule
->up
.table_id
];
7880 if (table
->other_table
&& rule
->tag
) {
7881 tag_set_add(&ofproto
->backer
->revalidate_set
, rule
->tag
);
7883 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
7889 set_frag_handling(struct ofproto
*ofproto_
,
7890 enum ofp_config_flags frag_handling
)
7892 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
7893 if (frag_handling
!= OFPC_FRAG_REASM
) {
7894 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
7902 packet_out(struct ofproto
*ofproto_
, struct ofpbuf
*packet
,
7903 const struct flow
*flow
,
7904 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
7906 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
7907 struct initial_vals initial_vals
;
7908 struct odputil_keybuf keybuf
;
7909 struct dpif_flow_stats stats
;
7913 struct action_xlate_ctx ctx
;
7914 uint64_t odp_actions_stub
[1024 / 8];
7915 struct ofpbuf odp_actions
;
7917 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
7918 odp_flow_key_from_flow(&key
, flow
,
7919 ofp_port_to_odp_port(ofproto
, flow
->in_port
));
7921 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
7923 initial_vals
.vlan_tci
= flow
->vlan_tci
;
7924 action_xlate_ctx_init(&ctx
, ofproto
, flow
, &initial_vals
, NULL
,
7925 packet_get_tcp_flags(packet
, flow
), packet
);
7926 ctx
.resubmit_stats
= &stats
;
7928 ofpbuf_use_stub(&odp_actions
,
7929 odp_actions_stub
, sizeof odp_actions_stub
);
7930 xlate_actions(&ctx
, ofpacts
, ofpacts_len
, &odp_actions
);
7931 dpif_execute(ofproto
->backer
->dpif
, key
.data
, key
.size
,
7932 odp_actions
.data
, odp_actions
.size
, packet
);
7933 ofpbuf_uninit(&odp_actions
);
7941 set_netflow(struct ofproto
*ofproto_
,
7942 const struct netflow_options
*netflow_options
)
7944 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
7946 if (netflow_options
) {
7947 if (!ofproto
->netflow
) {
7948 ofproto
->netflow
= netflow_create();
7950 return netflow_set_options(ofproto
->netflow
, netflow_options
);
7952 netflow_destroy(ofproto
->netflow
);
7953 ofproto
->netflow
= NULL
;
7959 get_netflow_ids(const struct ofproto
*ofproto_
,
7960 uint8_t *engine_type
, uint8_t *engine_id
)
7962 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
7964 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
7968 send_active_timeout(struct ofproto_dpif
*ofproto
, struct facet
*facet
)
7970 if (!facet_is_controller_flow(facet
) &&
7971 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
7972 struct subfacet
*subfacet
;
7973 struct ofexpired expired
;
7975 LIST_FOR_EACH (subfacet
, list_node
, &facet
->subfacets
) {
7976 if (subfacet
->path
== SF_FAST_PATH
) {
7977 struct dpif_flow_stats stats
;
7979 subfacet_reinstall(subfacet
, &stats
);
7980 subfacet_update_stats(subfacet
, &stats
);
7984 expired
.flow
= facet
->flow
;
7985 expired
.packet_count
= facet
->packet_count
;
7986 expired
.byte_count
= facet
->byte_count
;
7987 expired
.used
= facet
->used
;
7988 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
7993 send_netflow_active_timeouts(struct ofproto_dpif
*ofproto
)
7995 struct facet
*facet
;
7997 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
7998 send_active_timeout(ofproto
, facet
);
8002 static struct ofproto_dpif
*
8003 ofproto_dpif_lookup(const char *name
)
8005 struct ofproto_dpif
*ofproto
;
8007 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_node
,
8008 hash_string(name
, 0), &all_ofproto_dpifs
) {
8009 if (!strcmp(ofproto
->up
.name
, name
)) {
8017 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
8018 const char *argv
[], void *aux OVS_UNUSED
)
8020 struct ofproto_dpif
*ofproto
;
8023 ofproto
= ofproto_dpif_lookup(argv
[1]);
8025 unixctl_command_reply_error(conn
, "no such bridge");
8028 mac_learning_flush(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
8030 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
8031 mac_learning_flush(ofproto
->ml
, &ofproto
->backer
->revalidate_set
);
8035 unixctl_command_reply(conn
, "table successfully flushed");
8039 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
8040 const char *argv
[], void *aux OVS_UNUSED
)
8042 struct ds ds
= DS_EMPTY_INITIALIZER
;
8043 const struct ofproto_dpif
*ofproto
;
8044 const struct mac_entry
*e
;
8046 ofproto
= ofproto_dpif_lookup(argv
[1]);
8048 unixctl_command_reply_error(conn
, "no such bridge");
8052 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
8053 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
8054 struct ofbundle
*bundle
= e
->port
.p
;
8055 ds_put_format(&ds
, "%5d %4d "ETH_ADDR_FMT
" %3d\n",
8056 ofbundle_get_a_port(bundle
)->odp_port
,
8057 e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
8058 mac_entry_age(ofproto
->ml
, e
));
8060 unixctl_command_reply(conn
, ds_cstr(&ds
));
8065 struct action_xlate_ctx ctx
;
8071 trace_format_rule(struct ds
*result
, uint8_t table_id
, int level
,
8072 const struct rule_dpif
*rule
)
8074 ds_put_char_multiple(result
, '\t', level
);
8076 ds_put_cstr(result
, "No match\n");
8080 ds_put_format(result
, "Rule: table=%"PRIu8
" cookie=%#"PRIx64
" ",
8081 table_id
, ntohll(rule
->up
.flow_cookie
));
8082 cls_rule_format(&rule
->up
.cr
, result
);
8083 ds_put_char(result
, '\n');
8085 ds_put_char_multiple(result
, '\t', level
);
8086 ds_put_cstr(result
, "OpenFlow ");
8087 ofpacts_format(rule
->up
.ofpacts
, rule
->up
.ofpacts_len
, result
);
8088 ds_put_char(result
, '\n');
8092 trace_format_flow(struct ds
*result
, int level
, const char *title
,
8093 struct trace_ctx
*trace
)
8095 ds_put_char_multiple(result
, '\t', level
);
8096 ds_put_format(result
, "%s: ", title
);
8097 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
8098 ds_put_cstr(result
, "unchanged");
8100 flow_format(result
, &trace
->ctx
.flow
);
8101 trace
->flow
= trace
->ctx
.flow
;
8103 ds_put_char(result
, '\n');
8107 trace_format_regs(struct ds
*result
, int level
, const char *title
,
8108 struct trace_ctx
*trace
)
8112 ds_put_char_multiple(result
, '\t', level
);
8113 ds_put_format(result
, "%s:", title
);
8114 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
8115 ds_put_format(result
, " reg%zu=0x%"PRIx32
, i
, trace
->flow
.regs
[i
]);
8117 ds_put_char(result
, '\n');
8121 trace_format_odp(struct ds
*result
, int level
, const char *title
,
8122 struct trace_ctx
*trace
)
8124 struct ofpbuf
*odp_actions
= trace
->ctx
.odp_actions
;
8126 ds_put_char_multiple(result
, '\t', level
);
8127 ds_put_format(result
, "%s: ", title
);
8128 format_odp_actions(result
, odp_actions
->data
, odp_actions
->size
);
8129 ds_put_char(result
, '\n');
8133 trace_resubmit(struct action_xlate_ctx
*ctx
, struct rule_dpif
*rule
)
8135 struct trace_ctx
*trace
= CONTAINER_OF(ctx
, struct trace_ctx
, ctx
);
8136 struct ds
*result
= trace
->result
;
8138 ds_put_char(result
, '\n');
8139 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
8140 trace_format_regs(result
, ctx
->recurse
+ 1, "Resubmitted regs", trace
);
8141 trace_format_odp(result
, ctx
->recurse
+ 1, "Resubmitted odp", trace
);
8142 trace_format_rule(result
, ctx
->table_id
, ctx
->recurse
+ 1, rule
);
8146 trace_report(struct action_xlate_ctx
*ctx
, const char *s
)
8148 struct trace_ctx
*trace
= CONTAINER_OF(ctx
, struct trace_ctx
, ctx
);
8149 struct ds
*result
= trace
->result
;
8151 ds_put_char_multiple(result
, '\t', ctx
->recurse
);
8152 ds_put_cstr(result
, s
);
8153 ds_put_char(result
, '\n');
8157 ofproto_unixctl_trace(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
8158 void *aux OVS_UNUSED
)
8160 const char *dpname
= argv
[1];
8161 struct ofproto_dpif
*ofproto
;
8162 struct ofpbuf odp_key
;
8163 struct ofpbuf
*packet
;
8164 struct initial_vals initial_vals
;
8170 ofpbuf_init(&odp_key
, 0);
8173 ofproto
= ofproto_dpif_lookup(dpname
);
8175 unixctl_command_reply_error(conn
, "Unknown ofproto (use ofproto/list "
8179 if (argc
== 3 || (argc
== 4 && !strcmp(argv
[3], "-generate"))) {
8180 /* ofproto/trace dpname flow [-generate] */
8181 const char *flow_s
= argv
[2];
8182 const char *generate_s
= argv
[3];
8184 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8185 * flow. We guess which type it is based on whether 'flow_s' contains
8186 * an '(', since a datapath flow always contains '(') but an
8187 * OpenFlow-like flow should not (in fact it's allowed but I believe
8188 * that's not documented anywhere).
8190 * An alternative would be to try to parse 'flow_s' both ways, but then
8191 * it would be tricky giving a sensible error message. After all, do
8192 * you just say "syntax error" or do you present both error messages?
8193 * Both choices seem lousy. */
8194 if (strchr(flow_s
, '(')) {
8197 /* Convert string to datapath key. */
8198 ofpbuf_init(&odp_key
, 0);
8199 error
= odp_flow_key_from_string(flow_s
, NULL
, &odp_key
);
8201 unixctl_command_reply_error(conn
, "Bad flow syntax");
8205 /* The user might have specified the wrong ofproto but within the
8206 * same backer. That's OK, ofproto_receive() can find the right
8208 if (ofproto_receive(ofproto
->backer
, NULL
, odp_key
.data
,
8209 odp_key
.size
, &flow
, NULL
, &ofproto
, NULL
,
8211 unixctl_command_reply_error(conn
, "Invalid flow");
8214 ds_put_format(&result
, "Bridge: %s\n", ofproto
->up
.name
);
8218 error_s
= parse_ofp_exact_flow(&flow
, argv
[2]);
8220 unixctl_command_reply_error(conn
, error_s
);
8225 initial_vals
.vlan_tci
= flow
.vlan_tci
;
8228 /* Generate a packet, if requested. */
8230 packet
= ofpbuf_new(0);
8231 flow_compose(packet
, &flow
);
8233 } else if (argc
== 7) {
8234 /* ofproto/trace dpname priority tun_id in_port mark packet */
8235 const char *priority_s
= argv
[2];
8236 const char *tun_id_s
= argv
[3];
8237 const char *in_port_s
= argv
[4];
8238 const char *mark_s
= argv
[5];
8239 const char *packet_s
= argv
[6];
8240 uint32_t in_port
= atoi(in_port_s
);
8241 ovs_be64 tun_id
= htonll(strtoull(tun_id_s
, NULL
, 0));
8242 uint32_t priority
= atoi(priority_s
);
8243 uint32_t mark
= atoi(mark_s
);
8246 msg
= eth_from_hex(packet_s
, &packet
);
8248 unixctl_command_reply_error(conn
, msg
);
8252 ds_put_cstr(&result
, "Packet: ");
8253 s
= ofp_packet_to_string(packet
->data
, packet
->size
);
8254 ds_put_cstr(&result
, s
);
8257 flow_extract(packet
, priority
, mark
, NULL
, in_port
, &flow
);
8258 flow
.tunnel
.tun_id
= tun_id
;
8259 initial_vals
.vlan_tci
= flow
.vlan_tci
;
8261 unixctl_command_reply_error(conn
, "Bad command syntax");
8265 ofproto_trace(ofproto
, &flow
, packet
, &initial_vals
, &result
);
8266 unixctl_command_reply(conn
, ds_cstr(&result
));
8269 ds_destroy(&result
);
8270 ofpbuf_delete(packet
);
8271 ofpbuf_uninit(&odp_key
);
8275 ofproto_trace(struct ofproto_dpif
*ofproto
, const struct flow
*flow
,
8276 const struct ofpbuf
*packet
,
8277 const struct initial_vals
*initial_vals
, struct ds
*ds
)
8279 struct rule_dpif
*rule
;
8281 ds_put_cstr(ds
, "Flow: ");
8282 flow_format(ds
, flow
);
8283 ds_put_char(ds
, '\n');
8285 rule
= rule_dpif_lookup(ofproto
, flow
);
8287 trace_format_rule(ds
, 0, 0, rule
);
8288 if (rule
== ofproto
->miss_rule
) {
8289 ds_put_cstr(ds
, "\nNo match, flow generates \"packet in\"s.\n");
8290 } else if (rule
== ofproto
->no_packet_in_rule
) {
8291 ds_put_cstr(ds
, "\nNo match, packets dropped because "
8292 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8296 uint64_t odp_actions_stub
[1024 / 8];
8297 struct ofpbuf odp_actions
;
8299 struct trace_ctx trace
;
8302 tcp_flags
= packet
? packet_get_tcp_flags(packet
, flow
) : 0;
8305 ofpbuf_use_stub(&odp_actions
,
8306 odp_actions_stub
, sizeof odp_actions_stub
);
8307 action_xlate_ctx_init(&trace
.ctx
, ofproto
, flow
, initial_vals
,
8308 rule
, tcp_flags
, packet
);
8309 trace
.ctx
.resubmit_hook
= trace_resubmit
;
8310 trace
.ctx
.report_hook
= trace_report
;
8311 xlate_actions(&trace
.ctx
, rule
->up
.ofpacts
, rule
->up
.ofpacts_len
,
8314 ds_put_char(ds
, '\n');
8315 trace_format_flow(ds
, 0, "Final flow", &trace
);
8316 ds_put_cstr(ds
, "Datapath actions: ");
8317 format_odp_actions(ds
, odp_actions
.data
, odp_actions
.size
);
8318 ofpbuf_uninit(&odp_actions
);
8320 if (trace
.ctx
.slow
) {
8321 enum slow_path_reason slow
;
8323 ds_put_cstr(ds
, "\nThis flow is handled by the userspace "
8324 "slow path because it:");
8325 for (slow
= trace
.ctx
.slow
; slow
; ) {
8326 enum slow_path_reason bit
= rightmost_1bit(slow
);
8330 ds_put_cstr(ds
, "\n\t- Consists of CFM packets.");
8333 ds_put_cstr(ds
, "\n\t- Consists of LACP packets.");
8336 ds_put_cstr(ds
, "\n\t- Consists of STP packets.");
8339 ds_put_cstr(ds
, "\n\t- Consists of BFD packets.");
8342 ds_put_cstr(ds
, "\n\t- Needs in-band special case "
8345 ds_put_cstr(ds
, "\n\t (The datapath actions are "
8346 "incomplete--for complete actions, "
8347 "please supply a packet.)");
8350 case SLOW_CONTROLLER
:
8351 ds_put_cstr(ds
, "\n\t- Sends \"packet-in\" messages "
8352 "to the OpenFlow controller.");
8355 ds_put_cstr(ds
, "\n\t- Needs more specific matching "
8356 "than the datapath supports.");
8363 if (slow
& ~SLOW_MATCH
) {
8364 ds_put_cstr(ds
, "\nThe datapath actions above do not reflect "
8365 "the special slow-path processing.");
8372 ofproto_dpif_clog(struct unixctl_conn
*conn OVS_UNUSED
, int argc OVS_UNUSED
,
8373 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
8376 unixctl_command_reply(conn
, NULL
);
8380 ofproto_dpif_unclog(struct unixctl_conn
*conn OVS_UNUSED
, int argc OVS_UNUSED
,
8381 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
8384 unixctl_command_reply(conn
, NULL
);
8387 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8388 * 'reply' describing the results. */
8390 ofproto_dpif_self_check__(struct ofproto_dpif
*ofproto
, struct ds
*reply
)
8392 struct facet
*facet
;
8396 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
8397 if (!facet_check_consistency(facet
)) {
8402 ofproto
->backer
->need_revalidate
= REV_INCONSISTENCY
;
8406 ds_put_format(reply
, "%s: self-check failed (%d errors)\n",
8407 ofproto
->up
.name
, errors
);
8409 ds_put_format(reply
, "%s: self-check passed\n", ofproto
->up
.name
);
8414 ofproto_dpif_self_check(struct unixctl_conn
*conn
,
8415 int argc
, const char *argv
[], void *aux OVS_UNUSED
)
8417 struct ds reply
= DS_EMPTY_INITIALIZER
;
8418 struct ofproto_dpif
*ofproto
;
8421 ofproto
= ofproto_dpif_lookup(argv
[1]);
8423 unixctl_command_reply_error(conn
, "Unknown ofproto (use "
8424 "ofproto/list for help)");
8427 ofproto_dpif_self_check__(ofproto
, &reply
);
8429 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
8430 ofproto_dpif_self_check__(ofproto
, &reply
);
8434 unixctl_command_reply(conn
, ds_cstr(&reply
));
8438 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8439 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8440 * to destroy 'ofproto_shash' and free the returned value. */
8441 static const struct shash_node
**
8442 get_ofprotos(struct shash
*ofproto_shash
)
8444 const struct ofproto_dpif
*ofproto
;
8446 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
8447 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
8448 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
8451 return shash_sort(ofproto_shash
);
8455 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
8456 const char *argv
[] OVS_UNUSED
,
8457 void *aux OVS_UNUSED
)
8459 struct ds ds
= DS_EMPTY_INITIALIZER
;
8460 struct shash ofproto_shash
;
8461 const struct shash_node
**sorted_ofprotos
;
8464 shash_init(&ofproto_shash
);
8465 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
8466 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
8467 const struct shash_node
*node
= sorted_ofprotos
[i
];
8468 ds_put_format(&ds
, "%s\n", node
->name
);
8471 shash_destroy(&ofproto_shash
);
8472 free(sorted_ofprotos
);
8474 unixctl_command_reply(conn
, ds_cstr(&ds
));
8479 show_dp_format(const struct ofproto_dpif
*ofproto
, struct ds
*ds
)
8481 const struct shash_node
**ports
;
8483 struct avg_subfacet_rates lifetime
;
8484 unsigned long long int minutes
;
8485 const int min_ms
= 60 * 1000; /* milliseconds in one minute. */
8487 minutes
= (time_msec() - ofproto
->created
) / min_ms
;
8490 lifetime
.add_rate
= (double)ofproto
->total_subfacet_add_count
8492 lifetime
.del_rate
= (double)ofproto
->total_subfacet_del_count
8495 lifetime
.add_rate
= 0.0;
8496 lifetime
.del_rate
= 0.0;
8499 ds_put_format(ds
, "%s (%s):\n", ofproto
->up
.name
,
8500 dpif_name(ofproto
->backer
->dpif
));
8502 "\tlookups: hit:%"PRIu64
" missed:%"PRIu64
"\n",
8503 ofproto
->n_hit
, ofproto
->n_missed
);
8504 ds_put_format(ds
, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8505 " life span: %llu(ms)\n",
8506 hmap_count(&ofproto
->subfacets
),
8507 avg_subfacet_count(ofproto
),
8508 ofproto
->max_n_subfacet
,
8509 avg_subfacet_life_span(ofproto
));
8510 if (minutes
>= 60) {
8511 show_dp_rates(ds
, "\t\thourly avg:", &ofproto
->hourly
);
8513 if (minutes
>= 60 * 24) {
8514 show_dp_rates(ds
, "\t\tdaily avg:", &ofproto
->daily
);
8516 show_dp_rates(ds
, "\t\toverall avg:", &lifetime
);
8518 ports
= shash_sort(&ofproto
->up
.port_by_name
);
8519 for (i
= 0; i
< shash_count(&ofproto
->up
.port_by_name
); i
++) {
8520 const struct shash_node
*node
= ports
[i
];
8521 struct ofport
*ofport
= node
->data
;
8522 const char *name
= netdev_get_name(ofport
->netdev
);
8523 const char *type
= netdev_get_type(ofport
->netdev
);
8526 ds_put_format(ds
, "\t%s %u/", name
, ofport
->ofp_port
);
8528 odp_port
= ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
);
8529 if (odp_port
!= OVSP_NONE
) {
8530 ds_put_format(ds
, "%"PRIu32
":", odp_port
);
8532 ds_put_cstr(ds
, "none:");
8535 if (strcmp(type
, "system")) {
8536 struct netdev
*netdev
;
8539 ds_put_format(ds
, " (%s", type
);
8541 error
= netdev_open(name
, type
, &netdev
);
8546 error
= netdev_get_config(netdev
, &config
);
8548 const struct smap_node
**nodes
;
8551 nodes
= smap_sort(&config
);
8552 for (i
= 0; i
< smap_count(&config
); i
++) {
8553 const struct smap_node
*node
= nodes
[i
];
8554 ds_put_format(ds
, "%c %s=%s", i
? ',' : ':',
8555 node
->key
, node
->value
);
8559 smap_destroy(&config
);
8561 netdev_close(netdev
);
8563 ds_put_char(ds
, ')');
8565 ds_put_char(ds
, '\n');
8571 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc
,
8572 const char *argv
[], void *aux OVS_UNUSED
)
8574 struct ds ds
= DS_EMPTY_INITIALIZER
;
8575 const struct ofproto_dpif
*ofproto
;
8579 for (i
= 1; i
< argc
; i
++) {
8580 ofproto
= ofproto_dpif_lookup(argv
[i
]);
8582 ds_put_format(&ds
, "Unknown bridge %s (use dpif/dump-dps "
8583 "for help)", argv
[i
]);
8584 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
8587 show_dp_format(ofproto
, &ds
);
8590 struct shash ofproto_shash
;
8591 const struct shash_node
**sorted_ofprotos
;
8594 shash_init(&ofproto_shash
);
8595 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
8596 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
8597 const struct shash_node
*node
= sorted_ofprotos
[i
];
8598 show_dp_format(node
->data
, &ds
);
8601 shash_destroy(&ofproto_shash
);
8602 free(sorted_ofprotos
);
8605 unixctl_command_reply(conn
, ds_cstr(&ds
));
8610 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
8611 int argc OVS_UNUSED
, const char *argv
[],
8612 void *aux OVS_UNUSED
)
8614 struct ds ds
= DS_EMPTY_INITIALIZER
;
8615 const struct ofproto_dpif
*ofproto
;
8616 struct subfacet
*subfacet
;
8618 ofproto
= ofproto_dpif_lookup(argv
[1]);
8620 unixctl_command_reply_error(conn
, "no such bridge");
8624 update_stats(ofproto
->backer
);
8626 HMAP_FOR_EACH (subfacet
, hmap_node
, &ofproto
->subfacets
) {
8627 odp_flow_key_format(subfacet
->key
, subfacet
->key_len
, &ds
);
8629 ds_put_format(&ds
, ", packets:%"PRIu64
", bytes:%"PRIu64
", used:",
8630 subfacet
->dp_packet_count
, subfacet
->dp_byte_count
);
8631 if (subfacet
->used
) {
8632 ds_put_format(&ds
, "%.3fs",
8633 (time_msec() - subfacet
->used
) / 1000.0);
8635 ds_put_format(&ds
, "never");
8637 if (subfacet
->facet
->tcp_flags
) {
8638 ds_put_cstr(&ds
, ", flags:");
8639 packet_format_tcp_flags(&ds
, subfacet
->facet
->tcp_flags
);
8642 ds_put_cstr(&ds
, ", actions:");
8643 if (subfacet
->slow
) {
8644 uint64_t slow_path_stub
[128 / 8];
8645 const struct nlattr
*actions
;
8648 compose_slow_path(ofproto
, &subfacet
->facet
->flow
, subfacet
->slow
,
8649 slow_path_stub
, sizeof slow_path_stub
,
8650 &actions
, &actions_len
);
8651 format_odp_actions(&ds
, actions
, actions_len
);
8653 format_odp_actions(&ds
, subfacet
->actions
, subfacet
->actions_len
);
8655 ds_put_char(&ds
, '\n');
8658 unixctl_command_reply(conn
, ds_cstr(&ds
));
8663 ofproto_unixctl_dpif_del_flows(struct unixctl_conn
*conn
,
8664 int argc OVS_UNUSED
, const char *argv
[],
8665 void *aux OVS_UNUSED
)
8667 struct ds ds
= DS_EMPTY_INITIALIZER
;
8668 struct ofproto_dpif
*ofproto
;
8670 ofproto
= ofproto_dpif_lookup(argv
[1]);
8672 unixctl_command_reply_error(conn
, "no such bridge");
8676 flush(&ofproto
->up
);
8678 unixctl_command_reply(conn
, ds_cstr(&ds
));
8683 ofproto_dpif_unixctl_init(void)
8685 static bool registered
;
8691 unixctl_command_register(
8693 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8694 2, 6, ofproto_unixctl_trace
, NULL
);
8695 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8696 ofproto_unixctl_fdb_flush
, NULL
);
8697 unixctl_command_register("fdb/show", "bridge", 1, 1,
8698 ofproto_unixctl_fdb_show
, NULL
);
8699 unixctl_command_register("ofproto/clog", "", 0, 0,
8700 ofproto_dpif_clog
, NULL
);
8701 unixctl_command_register("ofproto/unclog", "", 0, 0,
8702 ofproto_dpif_unclog
, NULL
);
8703 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8704 ofproto_dpif_self_check
, NULL
);
8705 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8706 ofproto_unixctl_dpif_dump_dps
, NULL
);
8707 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX
,
8708 ofproto_unixctl_dpif_show
, NULL
);
8709 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8710 ofproto_unixctl_dpif_dump_flows
, NULL
);
8711 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8712 ofproto_unixctl_dpif_del_flows
, NULL
);
8715 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8717 * This is deprecated. It is only for compatibility with broken device drivers
8718 * in old versions of Linux that do not properly support VLANs when VLAN
8719 * devices are not used. When broken device drivers are no longer in
8720 * widespread use, we will delete these interfaces. */
8723 set_realdev(struct ofport
*ofport_
, uint16_t realdev_ofp_port
, int vid
)
8725 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
8726 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
8728 if (realdev_ofp_port
== ofport
->realdev_ofp_port
8729 && vid
== ofport
->vlandev_vid
) {
8733 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
8735 if (ofport
->realdev_ofp_port
) {
8738 if (realdev_ofp_port
&& ofport
->bundle
) {
8739 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8740 * themselves be part of a bundle. */
8741 bundle_set(ofport
->up
.ofproto
, ofport
->bundle
, NULL
);
8744 ofport
->realdev_ofp_port
= realdev_ofp_port
;
8745 ofport
->vlandev_vid
= vid
;
8747 if (realdev_ofp_port
) {
8748 vsp_add(ofport
, realdev_ofp_port
, vid
);
8755 hash_realdev_vid(uint16_t realdev_ofp_port
, int vid
)
8757 return hash_2words(realdev_ofp_port
, vid
);
8760 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8761 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8762 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8763 * it would return the port number of eth0.9.
8765 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8766 * function just returns its 'realdev_odp_port' argument. */
8768 vsp_realdev_to_vlandev(const struct ofproto_dpif
*ofproto
,
8769 uint32_t realdev_odp_port
, ovs_be16 vlan_tci
)
8771 if (!hmap_is_empty(&ofproto
->realdev_vid_map
)) {
8772 uint16_t realdev_ofp_port
;
8773 int vid
= vlan_tci_to_vid(vlan_tci
);
8774 const struct vlan_splinter
*vsp
;
8776 realdev_ofp_port
= odp_port_to_ofp_port(ofproto
, realdev_odp_port
);
8777 HMAP_FOR_EACH_WITH_HASH (vsp
, realdev_vid_node
,
8778 hash_realdev_vid(realdev_ofp_port
, vid
),
8779 &ofproto
->realdev_vid_map
) {
8780 if (vsp
->realdev_ofp_port
== realdev_ofp_port
8781 && vsp
->vid
== vid
) {
8782 return ofp_port_to_odp_port(ofproto
, vsp
->vlandev_ofp_port
);
8786 return realdev_odp_port
;
8789 static struct vlan_splinter
*
8790 vlandev_find(const struct ofproto_dpif
*ofproto
, uint16_t vlandev_ofp_port
)
8792 struct vlan_splinter
*vsp
;
8794 HMAP_FOR_EACH_WITH_HASH (vsp
, vlandev_node
, hash_int(vlandev_ofp_port
, 0),
8795 &ofproto
->vlandev_map
) {
8796 if (vsp
->vlandev_ofp_port
== vlandev_ofp_port
) {
8804 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8805 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8806 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8807 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8808 * eth0 and store 9 in '*vid'.
8810 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8811 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8814 vsp_vlandev_to_realdev(const struct ofproto_dpif
*ofproto
,
8815 uint16_t vlandev_ofp_port
, int *vid
)
8817 if (!hmap_is_empty(&ofproto
->vlandev_map
)) {
8818 const struct vlan_splinter
*vsp
;
8820 vsp
= vlandev_find(ofproto
, vlandev_ofp_port
);
8825 return vsp
->realdev_ofp_port
;
8831 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8832 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8833 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8834 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8835 * always the case unless VLAN splinters are enabled), returns false without
8836 * making any changes. */
8838 vsp_adjust_flow(const struct ofproto_dpif
*ofproto
, struct flow
*flow
)
8843 realdev
= vsp_vlandev_to_realdev(ofproto
, flow
->in_port
, &vid
);
8848 /* Cause the flow to be processed as if it came in on the real device with
8849 * the VLAN device's VLAN ID. */
8850 flow
->in_port
= realdev
;
8851 flow
->vlan_tci
= htons((vid
& VLAN_VID_MASK
) | VLAN_CFI
);
8856 vsp_remove(struct ofport_dpif
*port
)
8858 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
8859 struct vlan_splinter
*vsp
;
8861 vsp
= vlandev_find(ofproto
, port
->up
.ofp_port
);
8863 hmap_remove(&ofproto
->vlandev_map
, &vsp
->vlandev_node
);
8864 hmap_remove(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
);
8867 port
->realdev_ofp_port
= 0;
8869 VLOG_ERR("missing vlan device record");
8874 vsp_add(struct ofport_dpif
*port
, uint16_t realdev_ofp_port
, int vid
)
8876 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
8878 if (!vsp_vlandev_to_realdev(ofproto
, port
->up
.ofp_port
, NULL
)
8879 && (vsp_realdev_to_vlandev(ofproto
, realdev_ofp_port
, htons(vid
))
8880 == realdev_ofp_port
)) {
8881 struct vlan_splinter
*vsp
;
8883 vsp
= xmalloc(sizeof *vsp
);
8884 hmap_insert(&ofproto
->vlandev_map
, &vsp
->vlandev_node
,
8885 hash_int(port
->up
.ofp_port
, 0));
8886 hmap_insert(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
,
8887 hash_realdev_vid(realdev_ofp_port
, vid
));
8888 vsp
->realdev_ofp_port
= realdev_ofp_port
;
8889 vsp
->vlandev_ofp_port
= port
->up
.ofp_port
;
8892 port
->realdev_ofp_port
= realdev_ofp_port
;
8894 VLOG_ERR("duplicate vlan device record");
8899 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, uint16_t ofp_port
)
8901 const struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
8902 return ofport
? ofport
->odp_port
: OVSP_NONE
;
8905 static struct ofport_dpif
*
8906 odp_port_to_ofport(const struct dpif_backer
*backer
, uint32_t odp_port
)
8908 struct ofport_dpif
*port
;
8910 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
,
8911 hash_int(odp_port
, 0),
8912 &backer
->odp_to_ofport_map
) {
8913 if (port
->odp_port
== odp_port
) {
8922 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, uint32_t odp_port
)
8924 struct ofport_dpif
*port
;
8926 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
8927 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
8928 return port
->up
.ofp_port
;
8933 static unsigned long long int
8934 avg_subfacet_life_span(const struct ofproto_dpif
*ofproto
)
8936 unsigned long long int dc
;
8937 unsigned long long int avg
;
8939 dc
= ofproto
->total_subfacet_del_count
+ ofproto
->subfacet_del_count
;
8940 avg
= dc
? ofproto
->total_subfacet_life_span
/ dc
: 0;
8946 avg_subfacet_count(const struct ofproto_dpif
*ofproto
)
8950 if (ofproto
->n_update_stats
) {
8951 avg_c
= (double)ofproto
->total_subfacet_count
8952 / ofproto
->n_update_stats
;
8959 show_dp_rates(struct ds
*ds
, const char *heading
,
8960 const struct avg_subfacet_rates
*rates
)
8962 ds_put_format(ds
, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8963 heading
, rates
->add_rate
, rates
->del_rate
);
8967 update_max_subfacet_count(struct ofproto_dpif
*ofproto
)
8969 ofproto
->max_n_subfacet
= MAX(ofproto
->max_n_subfacet
,
8970 hmap_count(&ofproto
->subfacets
));
8973 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8974 * most heavily weighted element. 'base' designates the rate of decay: after
8975 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8978 exp_mavg(double *avg
, int base
, double new)
8980 *avg
= (*avg
* (base
- 1) + new) / base
;
8984 update_moving_averages(struct ofproto_dpif
*ofproto
)
8986 const int min_ms
= 60 * 1000; /* milliseconds in one minute. */
8988 /* Update hourly averages on the minute boundaries. */
8989 if (time_msec() - ofproto
->last_minute
>= min_ms
) {
8990 exp_mavg(&ofproto
->hourly
.add_rate
, 60, ofproto
->subfacet_add_count
);
8991 exp_mavg(&ofproto
->hourly
.del_rate
, 60, ofproto
->subfacet_del_count
);
8993 /* Update daily averages on the hour boundaries. */
8994 if ((ofproto
->last_minute
- ofproto
->created
) / min_ms
% 60 == 59) {
8995 exp_mavg(&ofproto
->daily
.add_rate
, 24, ofproto
->hourly
.add_rate
);
8996 exp_mavg(&ofproto
->daily
.del_rate
, 24, ofproto
->hourly
.del_rate
);
8999 ofproto
->total_subfacet_add_count
+= ofproto
->subfacet_add_count
;
9000 ofproto
->total_subfacet_del_count
+= ofproto
->subfacet_del_count
;
9001 ofproto
->subfacet_add_count
= 0;
9002 ofproto
->subfacet_del_count
= 0;
9003 ofproto
->last_minute
+= min_ms
;
9008 dpif_stats_update_hit_count(struct ofproto_dpif
*ofproto
, uint64_t delta
)
9010 ofproto
->n_hit
+= delta
;
9013 const struct ofproto_class ofproto_dpif_class
= {
9048 port_is_lacp_current
,
9049 NULL
, /* rule_choose_table */
9056 rule_modify_actions
,
9070 get_stp_port_status
,
9077 is_mirror_output_bundle
,
9078 forward_bpdu_changed
,
9079 set_mac_table_config
,