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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
88 * - Do include packets and bytes from facets that have been deleted or
89 * whose own statistics have been folded into the rule.
91 * - Do include packets and bytes sent "by hand" that were accounted to
92 * the rule without any facet being involved (this is a rare corner
93 * case in rule_execute()).
95 * - Do not include packet or bytes that can be obtained from any facet's
96 * packet_count or byte_count member or that can be obtained from the
97 * datapath by, e.g., dpif_flow_get() for any subfacet.
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
102 tag_type tag; /* Caches rule_calculate_tag() result. */
104 struct list facets; /* List of "struct facet"s. */
107 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
109 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
112 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
113 const struct flow *);
114 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
117 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
118 const struct flow *flow);
120 static void rule_credit_stats(struct rule_dpif *,
121 const struct dpif_flow_stats *);
122 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* stack for the push and pop actions.
220 * Each stack element is of the type "union mf_subvalue". */
222 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
224 /* The packet corresponding to 'flow', or a null pointer if we are
225 * revalidating without a packet to refer to. */
226 const struct ofpbuf *packet;
228 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
229 * actions update the flow table?
231 * We want to update these tables if we are actually processing a packet,
232 * or if we are accounting for packets that the datapath has processed, but
233 * not if we are just revalidating. */
236 /* The rule that we are currently translating, or NULL. */
237 struct rule_dpif *rule;
239 /* Union of the set of TCP flags seen so far in this flow. (Used only by
240 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
244 /* If nonnull, flow translation calls this function just before executing a
245 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
246 * when the recursion depth is exceeded.
248 * 'rule' is the rule being submitted into. It will be null if the
249 * resubmit or OFPP_TABLE action didn't find a matching rule.
251 * This is normally null so the client has to set it manually after
252 * calling action_xlate_ctx_init(). */
253 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
255 /* If nonnull, flow translation calls this function to report some
256 * significant decision, e.g. to explain why OFPP_NORMAL translation
257 * dropped a packet. */
258 void (*report_hook)(struct action_xlate_ctx *, const char *s);
260 /* If nonnull, flow translation credits the specified statistics to each
261 * rule reached through a resubmit or OFPP_TABLE action.
263 * This is normally null so the client has to set it manually after
264 * calling action_xlate_ctx_init(). */
265 const struct dpif_flow_stats *resubmit_stats;
267 /* xlate_actions() initializes and uses these members. The client might want
268 * to look at them after it returns. */
270 struct ofpbuf *odp_actions; /* Datapath actions. */
271 tag_type tags; /* Tags associated with actions. */
272 enum slow_path_reason slow; /* 0 if fast path may be used. */
273 bool has_learn; /* Actions include NXAST_LEARN? */
274 bool has_normal; /* Actions output to OFPP_NORMAL? */
275 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
276 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
277 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
279 /* xlate_actions() initializes and uses these members, but the client has no
280 * reason to look at them. */
282 int recurse; /* Recursion level, via xlate_table_action. */
283 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
284 struct flow base_flow; /* Flow at the last commit. */
285 uint32_t orig_skb_priority; /* Priority when packet arrived. */
286 uint8_t table_id; /* OpenFlow table ID where flow was found. */
287 uint32_t sflow_n_outputs; /* Number of output ports. */
288 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
289 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
290 bool exit; /* No further actions should be processed. */
293 /* Initial values of fields of the packet that may be changed during
294 * flow processing and needed later. */
295 struct initial_vals {
296 /* This is the value of vlan_tci in the packet as actually received from
297 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
298 * was received via a VLAN splinter. In that case, this value is 0
299 * (because the packet as actually received from the dpif had no 802.1Q
300 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
303 * This member should be removed when the VLAN splinters feature is no
307 /* If received on a tunnel, the IP TOS value of the tunnel. */
308 uint8_t tunnel_ip_tos;
311 static void action_xlate_ctx_init(struct action_xlate_ctx *,
312 struct ofproto_dpif *, const struct flow *,
313 const struct initial_vals *initial_vals,
315 uint8_t tcp_flags, const struct ofpbuf *);
316 static void xlate_actions(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts, size_t ofpacts_len,
318 struct ofpbuf *odp_actions);
319 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
320 const struct ofpact *ofpacts,
322 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
323 uint8_t table_id, bool may_packet_in);
325 static size_t put_userspace_action(const struct ofproto_dpif *,
326 struct ofpbuf *odp_actions,
328 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 tag_type tag; /* Tag associated with this port. */
533 bool may_enable; /* May be enabled in bonds. */
534 long long int carrier_seq; /* Carrier status changes. */
535 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
538 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
539 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
540 long long int stp_state_entered;
542 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
544 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
546 * This is deprecated. It is only for compatibility with broken device
547 * drivers in old versions of Linux that do not properly support VLANs when
548 * VLAN devices are not used. When broken device drivers are no longer in
549 * widespread use, we will delete these interfaces. */
550 uint16_t realdev_ofp_port;
554 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
555 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
556 * traffic egressing the 'ofport' with that priority should be marked with. */
557 struct priority_to_dscp {
558 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
559 uint32_t priority; /* Priority of this queue (see struct flow). */
561 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
564 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
566 * This is deprecated. It is only for compatibility with broken device drivers
567 * in old versions of Linux that do not properly support VLANs when VLAN
568 * devices are not used. When broken device drivers are no longer in
569 * widespread use, we will delete these interfaces. */
570 struct vlan_splinter {
571 struct hmap_node realdev_vid_node;
572 struct hmap_node vlandev_node;
573 uint16_t realdev_ofp_port;
574 uint16_t vlandev_ofp_port;
578 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
579 uint32_t realdev, ovs_be16 vlan_tci);
580 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
581 static void vsp_remove(struct ofport_dpif *);
582 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
584 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
586 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
589 static struct ofport_dpif *
590 ofport_dpif_cast(const struct ofport *ofport)
592 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
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 hmap bundles; /* Contains "struct ofbundle"s. */
699 struct mac_learning *ml;
700 struct ofmirror *mirrors[MAX_MIRRORS];
702 bool has_bonded_bundles;
706 struct hmap subfacets;
707 struct governor *governor;
708 long long int consistency_rl;
711 struct table_dpif tables[N_TABLES];
713 /* Support for debugging async flow mods. */
714 struct list completions;
716 bool has_bundle_action; /* True when the first bundle action appears. */
717 struct netdev_stats stats; /* To account packets generated and consumed in
722 long long int stp_last_tick;
724 /* VLAN splinters. */
725 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
726 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
729 struct sset ports; /* Set of standard port names. */
730 struct sset ghost_ports; /* Ports with no datapath port. */
731 struct sset port_poll_set; /* Queued names for port_poll() reply. */
732 int port_poll_errno; /* Last errno for port_poll() reply. */
734 /* Per ofproto's dpif stats. */
738 /* Subfacet statistics.
740 * These keep track of the total number of subfacets added and deleted and
741 * flow life span. They are useful for computing the flow rates stats
742 * exposed via "ovs-appctl dpif/show". The goal is to learn about
743 * traffic patterns in ways that we can use later to improve Open vSwitch
744 * performance in new situations. */
745 long long int created; /* Time when it is created. */
746 unsigned int max_n_subfacet; /* Maximum number of flows */
748 /* The average number of subfacets... */
749 struct avg_subfacet_rates hourly; /* ...over the last hour. */
750 struct avg_subfacet_rates daily; /* ...over the last day. */
751 long long int last_minute; /* Last time 'hourly' was updated. */
753 /* Number of subfacets added or deleted since 'last_minute'. */
754 unsigned int subfacet_add_count;
755 unsigned int subfacet_del_count;
757 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
758 unsigned long long int total_subfacet_add_count;
759 unsigned long long int total_subfacet_del_count;
761 /* Sum of the number of milliseconds that each subfacet existed,
762 * over the subfacets that have been added and then later deleted. */
763 unsigned long long int total_subfacet_life_span;
765 /* Incremented by the number of currently existing subfacets, each
766 * time we pull statistics from the kernel. */
767 unsigned long long int total_subfacet_count;
769 /* Number of times we pull statistics from the kernel. */
770 unsigned long long int n_update_stats;
772 static unsigned long long int avg_subfacet_life_span(
773 const struct ofproto_dpif *);
774 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
775 static void update_moving_averages(struct ofproto_dpif *ofproto);
776 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
778 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
780 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
781 * for debugging the asynchronous flow_mod implementation.) */
784 /* All existing ofproto_dpif instances, indexed by ->up.name. */
785 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
787 static void ofproto_dpif_unixctl_init(void);
789 static struct ofproto_dpif *
790 ofproto_dpif_cast(const struct ofproto *ofproto)
792 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
793 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
796 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
798 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
800 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
801 const struct ofpbuf *,
802 const struct initial_vals *, struct ds *);
804 /* Packet processing. */
805 static void update_learning_table(struct ofproto_dpif *,
806 const struct flow *, int vlan,
809 #define FLOW_MISS_MAX_BATCH 50
810 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
812 /* Flow expiration. */
813 static int expire(struct dpif_backer *);
816 static void send_netflow_active_timeouts(struct ofproto_dpif *);
819 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
820 static size_t compose_sflow_action(const struct ofproto_dpif *,
821 struct ofpbuf *odp_actions,
822 const struct flow *, uint32_t odp_port);
823 static void add_mirror_actions(struct action_xlate_ctx *ctx,
824 const struct flow *flow);
825 /* Global variables. */
826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
828 /* Initial mappings of port to bridge mappings. */
829 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
831 /* Factory functions. */
834 init(const struct shash *iface_hints)
836 struct shash_node *node;
838 /* Make a local copy, since we don't own 'iface_hints' elements. */
839 SHASH_FOR_EACH(node, iface_hints) {
840 const struct iface_hint *orig_hint = node->data;
841 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
843 new_hint->br_name = xstrdup(orig_hint->br_name);
844 new_hint->br_type = xstrdup(orig_hint->br_type);
845 new_hint->ofp_port = orig_hint->ofp_port;
847 shash_add(&init_ofp_ports, node->name, new_hint);
852 enumerate_types(struct sset *types)
854 dp_enumerate_types(types);
858 enumerate_names(const char *type, struct sset *names)
860 struct ofproto_dpif *ofproto;
863 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
864 if (strcmp(type, ofproto->up.type)) {
867 sset_add(names, ofproto->up.name);
874 del(const char *type, const char *name)
879 error = dpif_open(name, type, &dpif);
881 error = dpif_delete(dpif);
888 port_open_type(const char *datapath_type, const char *port_type)
890 return dpif_port_open_type(datapath_type, port_type);
893 /* Type functions. */
895 static struct ofproto_dpif *
896 lookup_ofproto_dpif_by_port_name(const char *name)
898 struct ofproto_dpif *ofproto;
900 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
901 if (sset_contains(&ofproto->ports, name)) {
910 type_run(const char *type)
912 static long long int push_timer = LLONG_MIN;
913 struct dpif_backer *backer;
917 backer = shash_find_data(&all_dpif_backers, type);
919 /* This is not necessarily a problem, since backers are only
920 * created on demand. */
924 dpif_run(backer->dpif);
926 /* The most natural place to push facet statistics is when they're pulled
927 * from the datapath. However, when there are many flows in the datapath,
928 * this expensive operation can occur so frequently, that it reduces our
929 * ability to quickly set up flows. To reduce the cost, we push statistics
931 if (time_msec() > push_timer) {
932 push_timer = time_msec() + 2000;
936 if (backer->need_revalidate
937 || !tag_set_is_empty(&backer->revalidate_set)) {
938 struct tag_set revalidate_set = backer->revalidate_set;
939 bool need_revalidate = backer->need_revalidate;
940 struct ofproto_dpif *ofproto;
941 struct simap_node *node;
942 struct simap tmp_backers;
944 /* Handle tunnel garbage collection. */
945 simap_init(&tmp_backers);
946 simap_swap(&backer->tnl_backers, &tmp_backers);
948 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
949 struct ofport_dpif *iter;
951 if (backer != ofproto->backer) {
955 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
958 if (!iter->tnl_port) {
962 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
963 node = simap_find(&tmp_backers, dp_port);
965 simap_put(&backer->tnl_backers, dp_port, node->data);
966 simap_delete(&tmp_backers, node);
967 node = simap_find(&backer->tnl_backers, dp_port);
969 node = simap_find(&backer->tnl_backers, dp_port);
971 uint32_t odp_port = UINT32_MAX;
973 if (!dpif_port_add(backer->dpif, iter->up.netdev,
975 simap_put(&backer->tnl_backers, dp_port, odp_port);
976 node = simap_find(&backer->tnl_backers, dp_port);
981 iter->odp_port = node ? node->data : OVSP_NONE;
982 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
984 backer->need_revalidate = REV_RECONFIGURE;
989 SIMAP_FOR_EACH (node, &tmp_backers) {
990 dpif_port_del(backer->dpif, node->data);
992 simap_destroy(&tmp_backers);
994 switch (backer->need_revalidate) {
995 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
996 case REV_STP: COVERAGE_INC(rev_stp); break;
997 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
998 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
999 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1002 if (backer->need_revalidate) {
1003 /* Clear the drop_keys in case we should now be accepting some
1004 * formerly dropped flows. */
1005 drop_key_clear(backer);
1008 /* Clear the revalidation flags. */
1009 tag_set_init(&backer->revalidate_set);
1010 backer->need_revalidate = 0;
1012 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1013 struct facet *facet, *next;
1015 if (ofproto->backer != backer) {
1019 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1021 || tag_set_intersects(&revalidate_set, facet->tags)) {
1022 facet_revalidate(facet);
1029 if (timer_expired(&backer->next_expiration)) {
1030 int delay = expire(backer);
1031 timer_set_duration(&backer->next_expiration, delay);
1034 /* Check for port changes in the dpif. */
1035 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1036 struct ofproto_dpif *ofproto;
1037 struct dpif_port port;
1039 /* Don't report on the datapath's device. */
1040 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1044 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1045 &all_ofproto_dpifs) {
1046 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1051 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1052 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1053 /* The port was removed. If we know the datapath,
1054 * report it through poll_set(). If we don't, it may be
1055 * notifying us of a removal we initiated, so ignore it.
1056 * If there's a pending ENOBUFS, let it stand, since
1057 * everything will be reevaluated. */
1058 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1059 sset_add(&ofproto->port_poll_set, devname);
1060 ofproto->port_poll_errno = 0;
1062 } else if (!ofproto) {
1063 /* The port was added, but we don't know with which
1064 * ofproto we should associate it. Delete it. */
1065 dpif_port_del(backer->dpif, port.port_no);
1067 dpif_port_destroy(&port);
1073 if (error != EAGAIN) {
1074 struct ofproto_dpif *ofproto;
1076 /* There was some sort of error, so propagate it to all
1077 * ofprotos that use this backer. */
1078 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1079 &all_ofproto_dpifs) {
1080 if (ofproto->backer == backer) {
1081 sset_clear(&ofproto->port_poll_set);
1082 ofproto->port_poll_errno = error;
1091 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1095 /* Handle one or more batches of upcalls, until there's nothing left to do
1096 * or until we do a fixed total amount of work.
1098 * We do work in batches because it can be much cheaper to set up a number
1099 * of flows and fire off their patches all at once. We do multiple batches
1100 * because in some cases handling a packet can cause another packet to be
1101 * queued almost immediately as part of the return flow. Both
1102 * optimizations can make major improvements on some benchmarks and
1103 * presumably for real traffic as well. */
1105 while (work < max_batch) {
1106 int retval = handle_upcalls(backer, max_batch - work);
1117 type_run_fast(const char *type)
1119 struct dpif_backer *backer;
1121 backer = shash_find_data(&all_dpif_backers, type);
1123 /* This is not necessarily a problem, since backers are only
1124 * created on demand. */
1128 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1134 static long long int port_rl = LLONG_MIN;
1135 static unsigned int backer_rl = 0;
1137 if (time_msec() >= port_rl) {
1138 struct ofproto_dpif *ofproto;
1139 struct ofport_dpif *ofport;
1141 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1143 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1144 port_run_fast(ofport);
1147 port_rl = time_msec() + 200;
1150 /* XXX: We have to be careful not to do too much work in this function. If
1151 * we call dpif_backer_run_fast() too often, or with too large a batch,
1152 * performance improves signifcantly, but at a cost. It's possible for the
1153 * number of flows in the datapath to increase without bound, and for poll
1154 * loops to take 10s of seconds. The correct solution to this problem,
1155 * long term, is to separate flow miss handling into it's own thread so it
1156 * isn't affected by revalidations, and expirations. Until then, this is
1157 * the best we can do. */
1158 if (++backer_rl >= 10) {
1159 struct shash_node *node;
1162 SHASH_FOR_EACH (node, &all_dpif_backers) {
1163 dpif_backer_run_fast(node->data, 1);
1169 type_wait(const char *type)
1171 struct dpif_backer *backer;
1173 backer = shash_find_data(&all_dpif_backers, type);
1175 /* This is not necessarily a problem, since backers are only
1176 * created on demand. */
1180 timer_wait(&backer->next_expiration);
1183 /* Basic life-cycle. */
1185 static int add_internal_flows(struct ofproto_dpif *);
1187 static struct ofproto *
1190 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1191 return &ofproto->up;
1195 dealloc(struct ofproto *ofproto_)
1197 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1202 close_dpif_backer(struct dpif_backer *backer)
1204 struct shash_node *node;
1206 ovs_assert(backer->refcount > 0);
1208 if (--backer->refcount) {
1212 drop_key_clear(backer);
1213 hmap_destroy(&backer->drop_keys);
1215 simap_destroy(&backer->tnl_backers);
1216 hmap_destroy(&backer->odp_to_ofport_map);
1217 node = shash_find(&all_dpif_backers, backer->type);
1219 shash_delete(&all_dpif_backers, node);
1220 dpif_close(backer->dpif);
1225 /* Datapath port slated for removal from datapath. */
1226 struct odp_garbage {
1227 struct list list_node;
1232 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1234 struct dpif_backer *backer;
1235 struct dpif_port_dump port_dump;
1236 struct dpif_port port;
1237 struct shash_node *node;
1238 struct list garbage_list;
1239 struct odp_garbage *garbage, *next;
1245 backer = shash_find_data(&all_dpif_backers, type);
1252 backer_name = xasprintf("ovs-%s", type);
1254 /* Remove any existing datapaths, since we assume we're the only
1255 * userspace controlling the datapath. */
1257 dp_enumerate_names(type, &names);
1258 SSET_FOR_EACH(name, &names) {
1259 struct dpif *old_dpif;
1261 /* Don't remove our backer if it exists. */
1262 if (!strcmp(name, backer_name)) {
1266 if (dpif_open(name, type, &old_dpif)) {
1267 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1269 dpif_delete(old_dpif);
1270 dpif_close(old_dpif);
1273 sset_destroy(&names);
1275 backer = xmalloc(sizeof *backer);
1277 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1280 VLOG_ERR("failed to open datapath of type %s: %s", type,
1286 backer->type = xstrdup(type);
1287 backer->refcount = 1;
1288 hmap_init(&backer->odp_to_ofport_map);
1289 hmap_init(&backer->drop_keys);
1290 timer_set_duration(&backer->next_expiration, 1000);
1291 backer->need_revalidate = 0;
1292 simap_init(&backer->tnl_backers);
1293 tag_set_init(&backer->revalidate_set);
1296 dpif_flow_flush(backer->dpif);
1298 /* Loop through the ports already on the datapath and remove any
1299 * that we don't need anymore. */
1300 list_init(&garbage_list);
1301 dpif_port_dump_start(&port_dump, backer->dpif);
1302 while (dpif_port_dump_next(&port_dump, &port)) {
1303 node = shash_find(&init_ofp_ports, port.name);
1304 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1305 garbage = xmalloc(sizeof *garbage);
1306 garbage->odp_port = port.port_no;
1307 list_push_front(&garbage_list, &garbage->list_node);
1310 dpif_port_dump_done(&port_dump);
1312 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1313 dpif_port_del(backer->dpif, garbage->odp_port);
1314 list_remove(&garbage->list_node);
1318 shash_add(&all_dpif_backers, type, backer);
1320 error = dpif_recv_set(backer->dpif, true);
1322 VLOG_ERR("failed to listen on datapath of type %s: %s",
1323 type, strerror(error));
1324 close_dpif_backer(backer);
1332 construct(struct ofproto *ofproto_)
1334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1335 struct shash_node *node, *next;
1340 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1345 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1346 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1348 ofproto->n_matches = 0;
1350 ofproto->netflow = NULL;
1351 ofproto->sflow = NULL;
1352 ofproto->stp = NULL;
1353 hmap_init(&ofproto->bundles);
1354 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1355 for (i = 0; i < MAX_MIRRORS; i++) {
1356 ofproto->mirrors[i] = NULL;
1358 ofproto->has_bonded_bundles = false;
1360 hmap_init(&ofproto->facets);
1361 hmap_init(&ofproto->subfacets);
1362 ofproto->governor = NULL;
1363 ofproto->consistency_rl = LLONG_MIN;
1365 for (i = 0; i < N_TABLES; i++) {
1366 struct table_dpif *table = &ofproto->tables[i];
1368 table->catchall_table = NULL;
1369 table->other_table = NULL;
1370 table->basis = random_uint32();
1373 list_init(&ofproto->completions);
1375 ofproto_dpif_unixctl_init();
1377 ofproto->has_mirrors = false;
1378 ofproto->has_bundle_action = false;
1380 hmap_init(&ofproto->vlandev_map);
1381 hmap_init(&ofproto->realdev_vid_map);
1383 sset_init(&ofproto->ports);
1384 sset_init(&ofproto->ghost_ports);
1385 sset_init(&ofproto->port_poll_set);
1386 ofproto->port_poll_errno = 0;
1388 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1389 struct iface_hint *iface_hint = node->data;
1391 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1392 /* Check if the datapath already has this port. */
1393 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1394 sset_add(&ofproto->ports, node->name);
1397 free(iface_hint->br_name);
1398 free(iface_hint->br_type);
1400 shash_delete(&init_ofp_ports, node);
1404 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1405 hash_string(ofproto->up.name, 0));
1406 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1408 ofproto_init_tables(ofproto_, N_TABLES);
1409 error = add_internal_flows(ofproto);
1410 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1413 ofproto->n_missed = 0;
1415 ofproto->max_n_subfacet = 0;
1416 ofproto->created = time_msec();
1417 ofproto->last_minute = ofproto->created;
1418 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1419 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1420 ofproto->subfacet_add_count = 0;
1421 ofproto->subfacet_del_count = 0;
1422 ofproto->total_subfacet_add_count = 0;
1423 ofproto->total_subfacet_del_count = 0;
1424 ofproto->total_subfacet_life_span = 0;
1425 ofproto->total_subfacet_count = 0;
1426 ofproto->n_update_stats = 0;
1432 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1433 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1435 struct ofputil_flow_mod fm;
1438 match_init_catchall(&fm.match);
1440 match_set_reg(&fm.match, 0, id);
1441 fm.new_cookie = htonll(0);
1442 fm.cookie = htonll(0);
1443 fm.cookie_mask = htonll(0);
1444 fm.table_id = TBL_INTERNAL;
1445 fm.command = OFPFC_ADD;
1446 fm.idle_timeout = 0;
1447 fm.hard_timeout = 0;
1451 fm.ofpacts = ofpacts->data;
1452 fm.ofpacts_len = ofpacts->size;
1454 error = ofproto_flow_mod(&ofproto->up, &fm);
1456 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1457 id, ofperr_to_string(error));
1461 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1462 ovs_assert(*rulep != NULL);
1468 add_internal_flows(struct ofproto_dpif *ofproto)
1470 struct ofpact_controller *controller;
1471 uint64_t ofpacts_stub[128 / 8];
1472 struct ofpbuf ofpacts;
1476 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1479 controller = ofpact_put_CONTROLLER(&ofpacts);
1480 controller->max_len = UINT16_MAX;
1481 controller->controller_id = 0;
1482 controller->reason = OFPR_NO_MATCH;
1483 ofpact_pad(&ofpacts);
1485 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1490 ofpbuf_clear(&ofpacts);
1491 error = add_internal_flow(ofproto, id++, &ofpacts,
1492 &ofproto->no_packet_in_rule);
1497 complete_operations(struct ofproto_dpif *ofproto)
1499 struct dpif_completion *c, *next;
1501 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1502 ofoperation_complete(c->op, 0);
1503 list_remove(&c->list_node);
1509 destruct(struct ofproto *ofproto_)
1511 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1512 struct rule_dpif *rule, *next_rule;
1513 struct oftable *table;
1516 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1517 complete_operations(ofproto);
1519 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1520 struct cls_cursor cursor;
1522 cls_cursor_init(&cursor, &table->cls, NULL);
1523 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1524 ofproto_rule_destroy(&rule->up);
1528 for (i = 0; i < MAX_MIRRORS; i++) {
1529 mirror_destroy(ofproto->mirrors[i]);
1532 netflow_destroy(ofproto->netflow);
1533 dpif_sflow_destroy(ofproto->sflow);
1534 hmap_destroy(&ofproto->bundles);
1535 mac_learning_destroy(ofproto->ml);
1537 hmap_destroy(&ofproto->facets);
1538 hmap_destroy(&ofproto->subfacets);
1539 governor_destroy(ofproto->governor);
1541 hmap_destroy(&ofproto->vlandev_map);
1542 hmap_destroy(&ofproto->realdev_vid_map);
1544 sset_destroy(&ofproto->ports);
1545 sset_destroy(&ofproto->ghost_ports);
1546 sset_destroy(&ofproto->port_poll_set);
1548 close_dpif_backer(ofproto->backer);
1552 run_fast(struct ofproto *ofproto_)
1554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555 struct ofport_dpif *ofport;
1557 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1558 port_run_fast(ofport);
1565 run(struct ofproto *ofproto_)
1567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1568 struct ofport_dpif *ofport;
1569 struct ofbundle *bundle;
1573 complete_operations(ofproto);
1576 error = run_fast(ofproto_);
1581 if (ofproto->netflow) {
1582 if (netflow_run(ofproto->netflow)) {
1583 send_netflow_active_timeouts(ofproto);
1586 if (ofproto->sflow) {
1587 dpif_sflow_run(ofproto->sflow);
1590 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1593 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1598 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1600 /* Check the consistency of a random facet, to aid debugging. */
1601 if (time_msec() >= ofproto->consistency_rl
1602 && !hmap_is_empty(&ofproto->facets)
1603 && !ofproto->backer->need_revalidate) {
1604 struct facet *facet;
1606 ofproto->consistency_rl = time_msec() + 250;
1608 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1609 struct facet, hmap_node);
1610 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1612 if (!facet_check_consistency(facet)) {
1613 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1618 if (ofproto->governor) {
1621 governor_run(ofproto->governor);
1623 /* If the governor has shrunk to its minimum size and the number of
1624 * subfacets has dwindled, then drop the governor entirely.
1626 * For hysteresis, the number of subfacets to drop the governor is
1627 * smaller than the number needed to trigger its creation. */
1628 n_subfacets = hmap_count(&ofproto->subfacets);
1629 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1630 && governor_is_idle(ofproto->governor)) {
1631 governor_destroy(ofproto->governor);
1632 ofproto->governor = NULL;
1640 wait(struct ofproto *ofproto_)
1642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1643 struct ofport_dpif *ofport;
1644 struct ofbundle *bundle;
1646 if (!clogged && !list_is_empty(&ofproto->completions)) {
1647 poll_immediate_wake();
1650 dpif_wait(ofproto->backer->dpif);
1651 dpif_recv_wait(ofproto->backer->dpif);
1652 if (ofproto->sflow) {
1653 dpif_sflow_wait(ofproto->sflow);
1655 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1656 poll_immediate_wake();
1658 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1661 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1662 bundle_wait(bundle);
1664 if (ofproto->netflow) {
1665 netflow_wait(ofproto->netflow);
1667 mac_learning_wait(ofproto->ml);
1669 if (ofproto->backer->need_revalidate) {
1670 /* Shouldn't happen, but if it does just go around again. */
1671 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1672 poll_immediate_wake();
1674 if (ofproto->governor) {
1675 governor_wait(ofproto->governor);
1680 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1682 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1684 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1685 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1689 flush(struct ofproto *ofproto_)
1691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1692 struct subfacet *subfacet, *next_subfacet;
1693 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1697 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1698 &ofproto->subfacets) {
1699 if (subfacet->path != SF_NOT_INSTALLED) {
1700 batch[n_batch++] = subfacet;
1701 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1702 subfacet_destroy_batch(ofproto, batch, n_batch);
1706 subfacet_destroy(subfacet);
1711 subfacet_destroy_batch(ofproto, batch, n_batch);
1716 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1717 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1719 *arp_match_ip = true;
1720 *actions = (OFPUTIL_A_OUTPUT |
1721 OFPUTIL_A_SET_VLAN_VID |
1722 OFPUTIL_A_SET_VLAN_PCP |
1723 OFPUTIL_A_STRIP_VLAN |
1724 OFPUTIL_A_SET_DL_SRC |
1725 OFPUTIL_A_SET_DL_DST |
1726 OFPUTIL_A_SET_NW_SRC |
1727 OFPUTIL_A_SET_NW_DST |
1728 OFPUTIL_A_SET_NW_TOS |
1729 OFPUTIL_A_SET_TP_SRC |
1730 OFPUTIL_A_SET_TP_DST |
1735 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1737 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1738 struct dpif_dp_stats s;
1740 strcpy(ots->name, "classifier");
1742 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1744 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1745 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1748 static struct ofport *
1751 struct ofport_dpif *port = xmalloc(sizeof *port);
1756 port_dealloc(struct ofport *port_)
1758 struct ofport_dpif *port = ofport_dpif_cast(port_);
1763 port_construct(struct ofport *port_)
1765 struct ofport_dpif *port = ofport_dpif_cast(port_);
1766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1767 const struct netdev *netdev = port->up.netdev;
1768 struct dpif_port dpif_port;
1771 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1772 port->bundle = NULL;
1774 port->tag = tag_create_random();
1775 port->may_enable = true;
1776 port->stp_port = NULL;
1777 port->stp_state = STP_DISABLED;
1778 port->tnl_port = NULL;
1779 hmap_init(&port->priorities);
1780 port->realdev_ofp_port = 0;
1781 port->vlandev_vid = 0;
1782 port->carrier_seq = netdev_get_carrier_resets(netdev);
1784 if (netdev_vport_is_patch(netdev)) {
1785 /* XXX By bailing out here, we don't do required sFlow work. */
1786 port->odp_port = OVSP_NONE;
1790 error = dpif_port_query_by_name(ofproto->backer->dpif,
1791 netdev_vport_get_dpif_port(netdev),
1797 port->odp_port = dpif_port.port_no;
1799 if (netdev_get_tunnel_config(netdev)) {
1800 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1802 /* Sanity-check that a mapping doesn't already exist. This
1803 * shouldn't happen for non-tunnel ports. */
1804 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1805 VLOG_ERR("port %s already has an OpenFlow port number",
1807 dpif_port_destroy(&dpif_port);
1811 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1812 hash_int(port->odp_port, 0));
1814 dpif_port_destroy(&dpif_port);
1816 if (ofproto->sflow) {
1817 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1824 port_destruct(struct ofport *port_)
1826 struct ofport_dpif *port = ofport_dpif_cast(port_);
1827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1828 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1829 const char *devname = netdev_get_name(port->up.netdev);
1831 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1832 /* The underlying device is still there, so delete it. This
1833 * happens when the ofproto is being destroyed, since the caller
1834 * assumes that removal of attached ports will happen as part of
1836 if (!port->tnl_port) {
1837 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1839 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1842 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1843 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1846 tnl_port_del(port->tnl_port);
1847 sset_find_and_delete(&ofproto->ports, devname);
1848 sset_find_and_delete(&ofproto->ghost_ports, devname);
1849 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1850 bundle_remove(port_);
1851 set_cfm(port_, NULL);
1852 if (ofproto->sflow) {
1853 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1856 ofport_clear_priorities(port);
1857 hmap_destroy(&port->priorities);
1861 port_modified(struct ofport *port_)
1863 struct ofport_dpif *port = ofport_dpif_cast(port_);
1865 if (port->bundle && port->bundle->bond) {
1866 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1871 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1873 struct ofport_dpif *port = ofport_dpif_cast(port_);
1874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1875 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1877 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1878 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1879 OFPUTIL_PC_NO_PACKET_IN)) {
1880 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1883 bundle_update(port->bundle);
1889 set_sflow(struct ofproto *ofproto_,
1890 const struct ofproto_sflow_options *sflow_options)
1892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1893 struct dpif_sflow *ds = ofproto->sflow;
1895 if (sflow_options) {
1897 struct ofport_dpif *ofport;
1899 ds = ofproto->sflow = dpif_sflow_create();
1900 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1901 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1903 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1905 dpif_sflow_set_options(ds, sflow_options);
1908 dpif_sflow_destroy(ds);
1909 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1910 ofproto->sflow = NULL;
1917 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1919 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1926 struct ofproto_dpif *ofproto;
1928 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1929 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1930 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1933 if (cfm_configure(ofport->cfm, s)) {
1939 cfm_destroy(ofport->cfm);
1945 get_cfm_status(const struct ofport *ofport_,
1946 struct ofproto_cfm_status *status)
1948 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1951 status->faults = cfm_get_fault(ofport->cfm);
1952 status->remote_opstate = cfm_get_opup(ofport->cfm);
1953 status->health = cfm_get_health(ofport->cfm);
1954 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1961 /* Spanning Tree. */
1964 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1966 struct ofproto_dpif *ofproto = ofproto_;
1967 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1968 struct ofport_dpif *ofport;
1970 ofport = stp_port_get_aux(sp);
1972 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1973 ofproto->up.name, port_num);
1975 struct eth_header *eth = pkt->l2;
1977 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1978 if (eth_addr_is_zero(eth->eth_src)) {
1979 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1980 "with unknown MAC", ofproto->up.name, port_num);
1982 send_packet(ofport, pkt);
1988 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1990 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1994 /* Only revalidate flows if the configuration changed. */
1995 if (!s != !ofproto->stp) {
1996 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2000 if (!ofproto->stp) {
2001 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2002 send_bpdu_cb, ofproto);
2003 ofproto->stp_last_tick = time_msec();
2006 stp_set_bridge_id(ofproto->stp, s->system_id);
2007 stp_set_bridge_priority(ofproto->stp, s->priority);
2008 stp_set_hello_time(ofproto->stp, s->hello_time);
2009 stp_set_max_age(ofproto->stp, s->max_age);
2010 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2012 struct ofport *ofport;
2014 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2015 set_stp_port(ofport, NULL);
2018 stp_destroy(ofproto->stp);
2019 ofproto->stp = NULL;
2026 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2032 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2033 s->designated_root = stp_get_designated_root(ofproto->stp);
2034 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2043 update_stp_port_state(struct ofport_dpif *ofport)
2045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2046 enum stp_state state;
2048 /* Figure out new state. */
2049 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2053 if (ofport->stp_state != state) {
2054 enum ofputil_port_state of_state;
2057 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2058 netdev_get_name(ofport->up.netdev),
2059 stp_state_name(ofport->stp_state),
2060 stp_state_name(state));
2061 if (stp_learn_in_state(ofport->stp_state)
2062 != stp_learn_in_state(state)) {
2063 /* xxx Learning action flows should also be flushed. */
2064 mac_learning_flush(ofproto->ml,
2065 &ofproto->backer->revalidate_set);
2067 fwd_change = stp_forward_in_state(ofport->stp_state)
2068 != stp_forward_in_state(state);
2070 ofproto->backer->need_revalidate = REV_STP;
2071 ofport->stp_state = state;
2072 ofport->stp_state_entered = time_msec();
2074 if (fwd_change && ofport->bundle) {
2075 bundle_update(ofport->bundle);
2078 /* Update the STP state bits in the OpenFlow port description. */
2079 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2080 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2081 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2082 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2083 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2085 ofproto_port_set_state(&ofport->up, of_state);
2089 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2090 * caller is responsible for assigning STP port numbers and ensuring
2091 * there are no duplicates. */
2093 set_stp_port(struct ofport *ofport_,
2094 const struct ofproto_port_stp_settings *s)
2096 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2098 struct stp_port *sp = ofport->stp_port;
2100 if (!s || !s->enable) {
2102 ofport->stp_port = NULL;
2103 stp_port_disable(sp);
2104 update_stp_port_state(ofport);
2107 } else if (sp && stp_port_no(sp) != s->port_num
2108 && ofport == stp_port_get_aux(sp)) {
2109 /* The port-id changed, so disable the old one if it's not
2110 * already in use by another port. */
2111 stp_port_disable(sp);
2114 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2115 stp_port_enable(sp);
2117 stp_port_set_aux(sp, ofport);
2118 stp_port_set_priority(sp, s->priority);
2119 stp_port_set_path_cost(sp, s->path_cost);
2121 update_stp_port_state(ofport);
2127 get_stp_port_status(struct ofport *ofport_,
2128 struct ofproto_port_stp_status *s)
2130 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2131 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2132 struct stp_port *sp = ofport->stp_port;
2134 if (!ofproto->stp || !sp) {
2140 s->port_id = stp_port_get_id(sp);
2141 s->state = stp_port_get_state(sp);
2142 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2143 s->role = stp_port_get_role(sp);
2144 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2150 stp_run(struct ofproto_dpif *ofproto)
2153 long long int now = time_msec();
2154 long long int elapsed = now - ofproto->stp_last_tick;
2155 struct stp_port *sp;
2158 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2159 ofproto->stp_last_tick = now;
2161 while (stp_get_changed_port(ofproto->stp, &sp)) {
2162 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2165 update_stp_port_state(ofport);
2169 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2170 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2176 stp_wait(struct ofproto_dpif *ofproto)
2179 poll_timer_wait(1000);
2183 /* Returns true if STP should process 'flow'. */
2185 stp_should_process_flow(const struct flow *flow)
2187 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2191 stp_process_packet(const struct ofport_dpif *ofport,
2192 const struct ofpbuf *packet)
2194 struct ofpbuf payload = *packet;
2195 struct eth_header *eth = payload.data;
2196 struct stp_port *sp = ofport->stp_port;
2198 /* Sink packets on ports that have STP disabled when the bridge has
2200 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2204 /* Trim off padding on payload. */
2205 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2206 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2209 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2210 stp_received_bpdu(sp, payload.data, payload.size);
2214 static struct priority_to_dscp *
2215 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2217 struct priority_to_dscp *pdscp;
2220 hash = hash_int(priority, 0);
2221 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2222 if (pdscp->priority == priority) {
2230 ofport_clear_priorities(struct ofport_dpif *ofport)
2232 struct priority_to_dscp *pdscp, *next;
2234 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2235 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2241 set_queues(struct ofport *ofport_,
2242 const struct ofproto_port_queue *qdscp_list,
2245 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2247 struct hmap new = HMAP_INITIALIZER(&new);
2250 for (i = 0; i < n_qdscp; i++) {
2251 struct priority_to_dscp *pdscp;
2255 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2256 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2261 pdscp = get_priority(ofport, priority);
2263 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2265 pdscp = xmalloc(sizeof *pdscp);
2266 pdscp->priority = priority;
2268 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2271 if (pdscp->dscp != dscp) {
2273 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2276 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2279 if (!hmap_is_empty(&ofport->priorities)) {
2280 ofport_clear_priorities(ofport);
2281 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2284 hmap_swap(&new, &ofport->priorities);
2292 /* Expires all MAC learning entries associated with 'bundle' and forces its
2293 * ofproto to revalidate every flow.
2295 * Normally MAC learning entries are removed only from the ofproto associated
2296 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2297 * are removed from every ofproto. When patch ports and SLB bonds are in use
2298 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2299 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2300 * with the host from which it migrated. */
2302 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2304 struct ofproto_dpif *ofproto = bundle->ofproto;
2305 struct mac_learning *ml = ofproto->ml;
2306 struct mac_entry *mac, *next_mac;
2308 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2309 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2310 if (mac->port.p == bundle) {
2312 struct ofproto_dpif *o;
2314 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2316 struct mac_entry *e;
2318 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2321 mac_learning_expire(o->ml, e);
2327 mac_learning_expire(ml, mac);
2332 static struct ofbundle *
2333 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2335 struct ofbundle *bundle;
2337 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2338 &ofproto->bundles) {
2339 if (bundle->aux == aux) {
2346 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2347 * ones that are found to 'bundles'. */
2349 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2350 void **auxes, size_t n_auxes,
2351 struct hmapx *bundles)
2355 hmapx_init(bundles);
2356 for (i = 0; i < n_auxes; i++) {
2357 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2359 hmapx_add(bundles, bundle);
2365 bundle_update(struct ofbundle *bundle)
2367 struct ofport_dpif *port;
2369 bundle->floodable = true;
2370 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2371 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2372 || !stp_forward_in_state(port->stp_state)) {
2373 bundle->floodable = false;
2380 bundle_del_port(struct ofport_dpif *port)
2382 struct ofbundle *bundle = port->bundle;
2384 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2386 list_remove(&port->bundle_node);
2387 port->bundle = NULL;
2390 lacp_slave_unregister(bundle->lacp, port);
2393 bond_slave_unregister(bundle->bond, port);
2396 bundle_update(bundle);
2400 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2401 struct lacp_slave_settings *lacp)
2403 struct ofport_dpif *port;
2405 port = get_ofp_port(bundle->ofproto, ofp_port);
2410 if (port->bundle != bundle) {
2411 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2413 bundle_del_port(port);
2416 port->bundle = bundle;
2417 list_push_back(&bundle->ports, &port->bundle_node);
2418 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2419 || !stp_forward_in_state(port->stp_state)) {
2420 bundle->floodable = false;
2424 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2425 lacp_slave_register(bundle->lacp, port, lacp);
2432 bundle_destroy(struct ofbundle *bundle)
2434 struct ofproto_dpif *ofproto;
2435 struct ofport_dpif *port, *next_port;
2442 ofproto = bundle->ofproto;
2443 for (i = 0; i < MAX_MIRRORS; i++) {
2444 struct ofmirror *m = ofproto->mirrors[i];
2446 if (m->out == bundle) {
2448 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2449 || hmapx_find_and_delete(&m->dsts, bundle)) {
2450 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2455 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2456 bundle_del_port(port);
2459 bundle_flush_macs(bundle, true);
2460 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2462 free(bundle->trunks);
2463 lacp_destroy(bundle->lacp);
2464 bond_destroy(bundle->bond);
2469 bundle_set(struct ofproto *ofproto_, void *aux,
2470 const struct ofproto_bundle_settings *s)
2472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2473 bool need_flush = false;
2474 struct ofport_dpif *port;
2475 struct ofbundle *bundle;
2476 unsigned long *trunks;
2482 bundle_destroy(bundle_lookup(ofproto, aux));
2486 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2487 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2489 bundle = bundle_lookup(ofproto, aux);
2491 bundle = xmalloc(sizeof *bundle);
2493 bundle->ofproto = ofproto;
2494 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2495 hash_pointer(aux, 0));
2497 bundle->name = NULL;
2499 list_init(&bundle->ports);
2500 bundle->vlan_mode = PORT_VLAN_TRUNK;
2502 bundle->trunks = NULL;
2503 bundle->use_priority_tags = s->use_priority_tags;
2504 bundle->lacp = NULL;
2505 bundle->bond = NULL;
2507 bundle->floodable = true;
2509 bundle->src_mirrors = 0;
2510 bundle->dst_mirrors = 0;
2511 bundle->mirror_out = 0;
2514 if (!bundle->name || strcmp(s->name, bundle->name)) {
2516 bundle->name = xstrdup(s->name);
2521 if (!bundle->lacp) {
2522 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2523 bundle->lacp = lacp_create();
2525 lacp_configure(bundle->lacp, s->lacp);
2527 lacp_destroy(bundle->lacp);
2528 bundle->lacp = NULL;
2531 /* Update set of ports. */
2533 for (i = 0; i < s->n_slaves; i++) {
2534 if (!bundle_add_port(bundle, s->slaves[i],
2535 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2539 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2540 struct ofport_dpif *next_port;
2542 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2543 for (i = 0; i < s->n_slaves; i++) {
2544 if (s->slaves[i] == port->up.ofp_port) {
2549 bundle_del_port(port);
2553 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2555 if (list_is_empty(&bundle->ports)) {
2556 bundle_destroy(bundle);
2560 /* Set VLAN tagging mode */
2561 if (s->vlan_mode != bundle->vlan_mode
2562 || s->use_priority_tags != bundle->use_priority_tags) {
2563 bundle->vlan_mode = s->vlan_mode;
2564 bundle->use_priority_tags = s->use_priority_tags;
2569 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2570 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2572 if (vlan != bundle->vlan) {
2573 bundle->vlan = vlan;
2577 /* Get trunked VLANs. */
2578 switch (s->vlan_mode) {
2579 case PORT_VLAN_ACCESS:
2583 case PORT_VLAN_TRUNK:
2584 trunks = CONST_CAST(unsigned long *, s->trunks);
2587 case PORT_VLAN_NATIVE_UNTAGGED:
2588 case PORT_VLAN_NATIVE_TAGGED:
2589 if (vlan != 0 && (!s->trunks
2590 || !bitmap_is_set(s->trunks, vlan)
2591 || bitmap_is_set(s->trunks, 0))) {
2592 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2594 trunks = bitmap_clone(s->trunks, 4096);
2596 trunks = bitmap_allocate1(4096);
2598 bitmap_set1(trunks, vlan);
2599 bitmap_set0(trunks, 0);
2601 trunks = CONST_CAST(unsigned long *, s->trunks);
2608 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2609 free(bundle->trunks);
2610 if (trunks == s->trunks) {
2611 bundle->trunks = vlan_bitmap_clone(trunks);
2613 bundle->trunks = trunks;
2618 if (trunks != s->trunks) {
2623 if (!list_is_short(&bundle->ports)) {
2624 bundle->ofproto->has_bonded_bundles = true;
2626 if (bond_reconfigure(bundle->bond, s->bond)) {
2627 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2630 bundle->bond = bond_create(s->bond);
2631 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2634 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2635 bond_slave_register(bundle->bond, port, port->up.netdev);
2638 bond_destroy(bundle->bond);
2639 bundle->bond = NULL;
2642 /* If we changed something that would affect MAC learning, un-learn
2643 * everything on this port and force flow revalidation. */
2645 bundle_flush_macs(bundle, false);
2652 bundle_remove(struct ofport *port_)
2654 struct ofport_dpif *port = ofport_dpif_cast(port_);
2655 struct ofbundle *bundle = port->bundle;
2658 bundle_del_port(port);
2659 if (list_is_empty(&bundle->ports)) {
2660 bundle_destroy(bundle);
2661 } else if (list_is_short(&bundle->ports)) {
2662 bond_destroy(bundle->bond);
2663 bundle->bond = NULL;
2669 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2671 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2672 struct ofport_dpif *port = port_;
2673 uint8_t ea[ETH_ADDR_LEN];
2676 error = netdev_get_etheraddr(port->up.netdev, ea);
2678 struct ofpbuf packet;
2681 ofpbuf_init(&packet, 0);
2682 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2684 memcpy(packet_pdu, pdu, pdu_size);
2686 send_packet(port, &packet);
2687 ofpbuf_uninit(&packet);
2689 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2690 "%s (%s)", port->bundle->name,
2691 netdev_get_name(port->up.netdev), strerror(error));
2696 bundle_send_learning_packets(struct ofbundle *bundle)
2698 struct ofproto_dpif *ofproto = bundle->ofproto;
2699 int error, n_packets, n_errors;
2700 struct mac_entry *e;
2702 error = n_packets = n_errors = 0;
2703 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2704 if (e->port.p != bundle) {
2705 struct ofpbuf *learning_packet;
2706 struct ofport_dpif *port;
2710 /* The assignment to "port" is unnecessary but makes "grep"ing for
2711 * struct ofport_dpif more effective. */
2712 learning_packet = bond_compose_learning_packet(bundle->bond,
2716 ret = send_packet(port, learning_packet);
2717 ofpbuf_delete(learning_packet);
2727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2728 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2729 "packets, last error was: %s",
2730 bundle->name, n_errors, n_packets, strerror(error));
2732 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2733 bundle->name, n_packets);
2738 bundle_run(struct ofbundle *bundle)
2741 lacp_run(bundle->lacp, send_pdu_cb);
2744 struct ofport_dpif *port;
2746 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2747 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2750 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2751 lacp_status(bundle->lacp));
2752 if (bond_should_send_learning_packets(bundle->bond)) {
2753 bundle_send_learning_packets(bundle);
2759 bundle_wait(struct ofbundle *bundle)
2762 lacp_wait(bundle->lacp);
2765 bond_wait(bundle->bond);
2772 mirror_scan(struct ofproto_dpif *ofproto)
2776 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2777 if (!ofproto->mirrors[idx]) {
2784 static struct ofmirror *
2785 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2789 for (i = 0; i < MAX_MIRRORS; i++) {
2790 struct ofmirror *mirror = ofproto->mirrors[i];
2791 if (mirror && mirror->aux == aux) {
2799 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2801 mirror_update_dups(struct ofproto_dpif *ofproto)
2805 for (i = 0; i < MAX_MIRRORS; i++) {
2806 struct ofmirror *m = ofproto->mirrors[i];
2809 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2813 for (i = 0; i < MAX_MIRRORS; i++) {
2814 struct ofmirror *m1 = ofproto->mirrors[i];
2821 for (j = i + 1; j < MAX_MIRRORS; j++) {
2822 struct ofmirror *m2 = ofproto->mirrors[j];
2824 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2825 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2826 m2->dup_mirrors |= m1->dup_mirrors;
2833 mirror_set(struct ofproto *ofproto_, void *aux,
2834 const struct ofproto_mirror_settings *s)
2836 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2837 mirror_mask_t mirror_bit;
2838 struct ofbundle *bundle;
2839 struct ofmirror *mirror;
2840 struct ofbundle *out;
2841 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2842 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2845 mirror = mirror_lookup(ofproto, aux);
2847 mirror_destroy(mirror);
2853 idx = mirror_scan(ofproto);
2855 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2857 ofproto->up.name, MAX_MIRRORS, s->name);
2861 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2862 mirror->ofproto = ofproto;
2865 mirror->out_vlan = -1;
2866 mirror->name = NULL;
2869 if (!mirror->name || strcmp(s->name, mirror->name)) {
2871 mirror->name = xstrdup(s->name);
2874 /* Get the new configuration. */
2875 if (s->out_bundle) {
2876 out = bundle_lookup(ofproto, s->out_bundle);
2878 mirror_destroy(mirror);
2884 out_vlan = s->out_vlan;
2886 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2887 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2889 /* If the configuration has not changed, do nothing. */
2890 if (hmapx_equals(&srcs, &mirror->srcs)
2891 && hmapx_equals(&dsts, &mirror->dsts)
2892 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2893 && mirror->out == out
2894 && mirror->out_vlan == out_vlan)
2896 hmapx_destroy(&srcs);
2897 hmapx_destroy(&dsts);
2901 hmapx_swap(&srcs, &mirror->srcs);
2902 hmapx_destroy(&srcs);
2904 hmapx_swap(&dsts, &mirror->dsts);
2905 hmapx_destroy(&dsts);
2907 free(mirror->vlans);
2908 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2911 mirror->out_vlan = out_vlan;
2913 /* Update bundles. */
2914 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2915 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2916 if (hmapx_contains(&mirror->srcs, bundle)) {
2917 bundle->src_mirrors |= mirror_bit;
2919 bundle->src_mirrors &= ~mirror_bit;
2922 if (hmapx_contains(&mirror->dsts, bundle)) {
2923 bundle->dst_mirrors |= mirror_bit;
2925 bundle->dst_mirrors &= ~mirror_bit;
2928 if (mirror->out == bundle) {
2929 bundle->mirror_out |= mirror_bit;
2931 bundle->mirror_out &= ~mirror_bit;
2935 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2936 ofproto->has_mirrors = true;
2937 mac_learning_flush(ofproto->ml,
2938 &ofproto->backer->revalidate_set);
2939 mirror_update_dups(ofproto);
2945 mirror_destroy(struct ofmirror *mirror)
2947 struct ofproto_dpif *ofproto;
2948 mirror_mask_t mirror_bit;
2949 struct ofbundle *bundle;
2956 ofproto = mirror->ofproto;
2957 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2958 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2960 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2961 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2962 bundle->src_mirrors &= ~mirror_bit;
2963 bundle->dst_mirrors &= ~mirror_bit;
2964 bundle->mirror_out &= ~mirror_bit;
2967 hmapx_destroy(&mirror->srcs);
2968 hmapx_destroy(&mirror->dsts);
2969 free(mirror->vlans);
2971 ofproto->mirrors[mirror->idx] = NULL;
2975 mirror_update_dups(ofproto);
2977 ofproto->has_mirrors = false;
2978 for (i = 0; i < MAX_MIRRORS; i++) {
2979 if (ofproto->mirrors[i]) {
2980 ofproto->has_mirrors = true;
2987 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2988 uint64_t *packets, uint64_t *bytes)
2990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2991 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2994 *packets = *bytes = UINT64_MAX;
3000 *packets = mirror->packet_count;
3001 *bytes = mirror->byte_count;
3007 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3010 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3011 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3017 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3019 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3020 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3021 return bundle && bundle->mirror_out != 0;
3025 forward_bpdu_changed(struct ofproto *ofproto_)
3027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3028 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3032 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3036 mac_learning_set_idle_time(ofproto->ml, idle_time);
3037 mac_learning_set_max_entries(ofproto->ml, max_entries);
3042 static struct ofport_dpif *
3043 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3045 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3046 return ofport ? ofport_dpif_cast(ofport) : NULL;
3049 static struct ofport_dpif *
3050 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3052 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3053 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3057 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3058 struct ofproto_port *ofproto_port,
3059 struct dpif_port *dpif_port)
3061 ofproto_port->name = dpif_port->name;
3062 ofproto_port->type = dpif_port->type;
3063 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3066 static struct ofport_dpif *
3067 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3069 const struct ofproto_dpif *ofproto;
3072 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3077 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3078 struct ofport *ofport;
3080 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3081 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3082 return ofport_dpif_cast(ofport);
3089 port_run_fast(struct ofport_dpif *ofport)
3091 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3092 struct ofpbuf packet;
3094 ofpbuf_init(&packet, 0);
3095 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3096 send_packet(ofport, &packet);
3097 ofpbuf_uninit(&packet);
3102 port_run(struct ofport_dpif *ofport)
3104 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3105 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3106 bool enable = netdev_get_carrier(ofport->up.netdev);
3108 ofport->carrier_seq = carrier_seq;
3110 port_run_fast(ofport);
3112 if (ofport->tnl_port
3113 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3114 &ofport->tnl_port)) {
3115 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3119 int cfm_opup = cfm_get_opup(ofport->cfm);
3121 cfm_run(ofport->cfm);
3122 enable = enable && !cfm_get_fault(ofport->cfm);
3124 if (cfm_opup >= 0) {
3125 enable = enable && cfm_opup;
3129 if (ofport->bundle) {
3130 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3131 if (carrier_changed) {
3132 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3136 if (ofport->may_enable != enable) {
3137 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3139 if (ofproto->has_bundle_action) {
3140 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3144 ofport->may_enable = enable;
3148 port_wait(struct ofport_dpif *ofport)
3151 cfm_wait(ofport->cfm);
3156 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3157 struct ofproto_port *ofproto_port)
3159 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3160 struct dpif_port dpif_port;
3163 if (sset_contains(&ofproto->ghost_ports, devname)) {
3164 const char *type = netdev_get_type_from_name(devname);
3166 /* We may be called before ofproto->up.port_by_name is populated with
3167 * the appropriate ofport. For this reason, we must get the name and
3168 * type from the netdev layer directly. */
3170 const struct ofport *ofport;
3172 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3173 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3174 ofproto_port->name = xstrdup(devname);
3175 ofproto_port->type = xstrdup(type);
3181 if (!sset_contains(&ofproto->ports, devname)) {
3184 error = dpif_port_query_by_name(ofproto->backer->dpif,
3185 devname, &dpif_port);
3187 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3193 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3196 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3197 const char *devname = netdev_get_name(netdev);
3199 if (netdev_vport_is_patch(netdev)) {
3200 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3204 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3205 uint32_t port_no = UINT32_MAX;
3208 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3212 if (netdev_get_tunnel_config(netdev)) {
3213 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3217 if (netdev_get_tunnel_config(netdev)) {
3218 sset_add(&ofproto->ghost_ports, devname);
3220 sset_add(&ofproto->ports, devname);
3226 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3228 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3229 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3236 sset_find_and_delete(&ofproto->ghost_ports,
3237 netdev_get_name(ofport->up.netdev));
3238 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3239 if (!ofport->tnl_port) {
3240 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3242 /* The caller is going to close ofport->up.netdev. If this is a
3243 * bonded port, then the bond is using that netdev, so remove it
3244 * from the bond. The client will need to reconfigure everything
3245 * after deleting ports, so then the slave will get re-added. */
3246 bundle_remove(&ofport->up);
3253 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3255 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3260 error = netdev_get_stats(ofport->up.netdev, stats);
3262 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3265 /* ofproto->stats.tx_packets represents packets that we created
3266 * internally and sent to some port (e.g. packets sent with
3267 * send_packet()). Account for them as if they had come from
3268 * OFPP_LOCAL and got forwarded. */
3270 if (stats->rx_packets != UINT64_MAX) {
3271 stats->rx_packets += ofproto->stats.tx_packets;
3274 if (stats->rx_bytes != UINT64_MAX) {
3275 stats->rx_bytes += ofproto->stats.tx_bytes;
3278 /* ofproto->stats.rx_packets represents packets that were received on
3279 * some port and we processed internally and dropped (e.g. STP).
3280 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3282 if (stats->tx_packets != UINT64_MAX) {
3283 stats->tx_packets += ofproto->stats.rx_packets;
3286 if (stats->tx_bytes != UINT64_MAX) {
3287 stats->tx_bytes += ofproto->stats.rx_bytes;
3294 /* Account packets for LOCAL port. */
3296 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3297 size_t tx_size, size_t rx_size)
3299 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3302 ofproto->stats.rx_packets++;
3303 ofproto->stats.rx_bytes += rx_size;
3306 ofproto->stats.tx_packets++;
3307 ofproto->stats.tx_bytes += tx_size;
3311 struct port_dump_state {
3316 struct ofproto_port port;
3321 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3323 *statep = xzalloc(sizeof(struct port_dump_state));
3328 port_dump_next(const struct ofproto *ofproto_, void *state_,
3329 struct ofproto_port *port)
3331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3332 struct port_dump_state *state = state_;
3333 const struct sset *sset;
3334 struct sset_node *node;
3336 if (state->has_port) {
3337 ofproto_port_destroy(&state->port);
3338 state->has_port = false;
3340 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3341 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3344 error = port_query_by_name(ofproto_, node->name, &state->port);
3346 *port = state->port;
3347 state->has_port = true;
3349 } else if (error != ENODEV) {
3354 if (!state->ghost) {
3355 state->ghost = true;
3358 return port_dump_next(ofproto_, state_, port);
3365 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3367 struct port_dump_state *state = state_;
3369 if (state->has_port) {
3370 ofproto_port_destroy(&state->port);
3377 port_poll(const struct ofproto *ofproto_, char **devnamep)
3379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3381 if (ofproto->port_poll_errno) {
3382 int error = ofproto->port_poll_errno;
3383 ofproto->port_poll_errno = 0;
3387 if (sset_is_empty(&ofproto->port_poll_set)) {
3391 *devnamep = sset_pop(&ofproto->port_poll_set);
3396 port_poll_wait(const struct ofproto *ofproto_)
3398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3399 dpif_port_poll_wait(ofproto->backer->dpif);
3403 port_is_lacp_current(const struct ofport *ofport_)
3405 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3406 return (ofport->bundle && ofport->bundle->lacp
3407 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3411 /* Upcall handling. */
3413 /* Flow miss batching.
3415 * Some dpifs implement operations faster when you hand them off in a batch.
3416 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3417 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3418 * more packets, plus possibly installing the flow in the dpif.
3420 * So far we only batch the operations that affect flow setup time the most.
3421 * It's possible to batch more than that, but the benefit might be minimal. */
3423 struct hmap_node hmap_node;
3424 struct ofproto_dpif *ofproto;
3426 enum odp_key_fitness key_fitness;
3427 const struct nlattr *key;
3429 struct initial_vals initial_vals;
3430 struct list packets;
3431 enum dpif_upcall_type upcall_type;
3432 uint32_t odp_in_port;
3435 struct flow_miss_op {
3436 struct dpif_op dpif_op;
3437 void *garbage; /* Pointer to pass to free(), NULL if none. */
3438 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3441 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3442 * OpenFlow controller as necessary according to their individual
3443 * configurations. */
3445 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3446 const struct flow *flow)
3448 struct ofputil_packet_in pin;
3450 pin.packet = packet->data;
3451 pin.packet_len = packet->size;
3452 pin.reason = OFPR_NO_MATCH;
3453 pin.controller_id = 0;
3458 pin.send_len = 0; /* not used for flow table misses */
3460 flow_get_metadata(flow, &pin.fmd);
3462 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3465 static enum slow_path_reason
3466 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3467 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3471 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3473 cfm_process_heartbeat(ofport->cfm, packet);
3476 } else if (ofport->bundle && ofport->bundle->lacp
3477 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3479 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3482 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3484 stp_process_packet(ofport, packet);
3492 static struct flow_miss *
3493 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3494 const struct flow *flow, uint32_t hash)
3496 struct flow_miss *miss;
3498 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3499 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3507 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3508 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3509 * 'miss' is associated with a subfacet the caller must also initialize the
3510 * returned op->subfacet, and if anything needs to be freed after processing
3511 * the op, the caller must initialize op->garbage also. */
3513 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3514 struct flow_miss_op *op)
3516 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3517 /* This packet was received on a VLAN splinter port. We
3518 * added a VLAN to the packet to make the packet resemble
3519 * the flow, but the actions were composed assuming that
3520 * the packet contained no VLAN. So, we must remove the
3521 * VLAN header from the packet before trying to execute the
3523 eth_pop_vlan(packet);
3527 op->dpif_op.type = DPIF_OP_EXECUTE;
3528 op->dpif_op.u.execute.key = miss->key;
3529 op->dpif_op.u.execute.key_len = miss->key_len;
3530 op->dpif_op.u.execute.packet = packet;
3533 /* Helper for handle_flow_miss_without_facet() and
3534 * handle_flow_miss_with_facet(). */
3536 handle_flow_miss_common(struct rule_dpif *rule,
3537 struct ofpbuf *packet, const struct flow *flow)
3539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3541 ofproto->n_matches++;
3543 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3545 * Extra-special case for fail-open mode.
3547 * We are in fail-open mode and the packet matched the fail-open
3548 * rule, but we are connected to a controller too. We should send
3549 * the packet up to the controller in the hope that it will try to
3550 * set up a flow and thereby allow us to exit fail-open.
3552 * See the top-level comment in fail-open.c for more information.
3554 send_packet_in_miss(ofproto, packet, flow);
3558 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3559 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3560 * installing a datapath flow. The answer is usually "yes" (a return value of
3561 * true). However, for short flows the cost of bookkeeping is much higher than
3562 * the benefits, so when the datapath holds a large number of flows we impose
3563 * some heuristics to decide which flows are likely to be worth tracking. */
3565 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3566 struct flow_miss *miss, uint32_t hash)
3568 if (!ofproto->governor) {
3571 n_subfacets = hmap_count(&ofproto->subfacets);
3572 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3576 ofproto->governor = governor_create(ofproto->up.name);
3579 return governor_should_install_flow(ofproto->governor, hash,
3580 list_size(&miss->packets));
3583 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3584 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3585 * increment '*n_ops'. */
3587 handle_flow_miss_without_facet(struct flow_miss *miss,
3588 struct rule_dpif *rule,
3589 struct flow_miss_op *ops, size_t *n_ops)
3591 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3592 long long int now = time_msec();
3593 struct action_xlate_ctx ctx;
3594 struct ofpbuf *packet;
3596 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3597 struct flow_miss_op *op = &ops[*n_ops];
3598 struct dpif_flow_stats stats;
3599 struct ofpbuf odp_actions;
3601 COVERAGE_INC(facet_suppress);
3603 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3605 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3606 rule_credit_stats(rule, &stats);
3608 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3609 &miss->initial_vals, rule, 0, packet);
3610 ctx.resubmit_stats = &stats;
3611 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3614 if (odp_actions.size) {
3615 struct dpif_execute *execute = &op->dpif_op.u.execute;
3617 init_flow_miss_execute_op(miss, packet, op);
3618 execute->actions = odp_actions.data;
3619 execute->actions_len = odp_actions.size;
3620 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3624 ofpbuf_uninit(&odp_actions);
3629 /* Handles 'miss', which matches 'facet'. May add any required datapath
3630 * operations to 'ops', incrementing '*n_ops' for each new op.
3632 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3633 * This is really important only for new facets: if we just called time_msec()
3634 * here, then the new subfacet or its packets could look (occasionally) as
3635 * though it was used some time after the facet was used. That can make a
3636 * one-packet flow look like it has a nonzero duration, which looks odd in
3637 * e.g. NetFlow statistics. */
3639 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3641 struct flow_miss_op *ops, size_t *n_ops)
3643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3644 enum subfacet_path want_path;
3645 struct subfacet *subfacet;
3646 struct ofpbuf *packet;
3648 subfacet = subfacet_create(facet, miss, now);
3650 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3651 struct flow_miss_op *op = &ops[*n_ops];
3652 struct dpif_flow_stats stats;
3653 struct ofpbuf odp_actions;
3655 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3657 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3658 if (!subfacet->actions || subfacet->slow) {
3659 subfacet_make_actions(subfacet, packet, &odp_actions);
3662 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3663 subfacet_update_stats(subfacet, &stats);
3665 if (subfacet->actions_len) {
3666 struct dpif_execute *execute = &op->dpif_op.u.execute;
3668 init_flow_miss_execute_op(miss, packet, op);
3669 if (!subfacet->slow) {
3670 execute->actions = subfacet->actions;
3671 execute->actions_len = subfacet->actions_len;
3672 ofpbuf_uninit(&odp_actions);
3674 execute->actions = odp_actions.data;
3675 execute->actions_len = odp_actions.size;
3676 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3681 ofpbuf_uninit(&odp_actions);
3685 want_path = subfacet_want_path(subfacet->slow);
3686 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3687 struct flow_miss_op *op = &ops[(*n_ops)++];
3688 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3690 subfacet->path = want_path;
3693 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3694 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3695 put->key = miss->key;
3696 put->key_len = miss->key_len;
3697 if (want_path == SF_FAST_PATH) {
3698 put->actions = subfacet->actions;
3699 put->actions_len = subfacet->actions_len;
3701 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3702 op->stub, sizeof op->stub,
3703 &put->actions, &put->actions_len);
3709 /* Handles flow miss 'miss'. May add any required datapath operations
3710 * to 'ops', incrementing '*n_ops' for each new op. */
3712 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3715 struct ofproto_dpif *ofproto = miss->ofproto;
3716 struct facet *facet;
3720 /* The caller must ensure that miss->hmap_node.hash contains
3721 * flow_hash(miss->flow, 0). */
3722 hash = miss->hmap_node.hash;
3724 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3726 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3728 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3729 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3733 facet = facet_create(rule, &miss->flow, hash);
3738 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3741 static struct drop_key *
3742 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3745 struct drop_key *drop_key;
3747 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3748 &backer->drop_keys) {
3749 if (drop_key->key_len == key_len
3750 && !memcmp(drop_key->key, key, key_len)) {
3758 drop_key_clear(struct dpif_backer *backer)
3760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3761 struct drop_key *drop_key, *next;
3763 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3766 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3768 if (error && !VLOG_DROP_WARN(&rl)) {
3769 struct ds ds = DS_EMPTY_INITIALIZER;
3770 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3771 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3776 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3777 free(drop_key->key);
3782 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3783 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3784 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3785 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3786 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3787 * 'packet' ingressed.
3789 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3790 * 'flow''s in_port to OFPP_NONE.
3792 * This function does post-processing on data returned from
3793 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3794 * of the upcall processing logic. In particular, if the extracted in_port is
3795 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3796 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3797 * a VLAN header onto 'packet' (if it is nonnull).
3799 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3800 * to the VLAN TCI with which the packet was really received, that is, the
3801 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3802 * the value returned in flow->vlan_tci only for packets received on
3803 * VLAN splinters.) Also, if received on an IP tunnel, sets
3804 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3806 * Similarly, this function also includes some logic to help with tunnels. It
3807 * may modify 'flow' as necessary to make the tunneling implementation
3808 * transparent to the upcall processing logic.
3810 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3811 * or some other positive errno if there are other problems. */
3813 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3814 const struct nlattr *key, size_t key_len,
3815 struct flow *flow, enum odp_key_fitness *fitnessp,
3816 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3817 struct initial_vals *initial_vals)
3819 const struct ofport_dpif *port;
3820 enum odp_key_fitness fitness;
3823 fitness = odp_flow_key_to_flow(key, key_len, flow);
3824 if (fitness == ODP_FIT_ERROR) {
3830 initial_vals->vlan_tci = flow->vlan_tci;
3831 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3835 *odp_in_port = flow->in_port;
3838 if (tnl_port_should_receive(flow)) {
3839 const struct ofport *ofport = tnl_port_receive(flow);
3841 flow->in_port = OFPP_NONE;
3844 port = ofport_dpif_cast(ofport);
3846 /* We can't reproduce 'key' from 'flow'. */
3847 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3849 /* XXX: Since the tunnel module is not scoped per backer, it's
3850 * theoretically possible that we'll receive an ofport belonging to an
3851 * entirely different datapath. In practice, this can't happen because
3852 * no platforms has two separate datapaths which each support
3854 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3856 port = odp_port_to_ofport(backer, flow->in_port);
3858 flow->in_port = OFPP_NONE;
3862 flow->in_port = port->up.ofp_port;
3863 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3865 /* Make the packet resemble the flow, so that it gets sent to
3866 * an OpenFlow controller properly, so that it looks correct
3867 * for sFlow, and so that flow_extract() will get the correct
3868 * vlan_tci if it is called on 'packet'.
3870 * The allocated space inside 'packet' probably also contains
3871 * 'key', that is, both 'packet' and 'key' are probably part of
3872 * a struct dpif_upcall (see the large comment on that
3873 * structure definition), so pushing data on 'packet' is in
3874 * general not a good idea since it could overwrite 'key' or
3875 * free it as a side effect. However, it's OK in this special
3876 * case because we know that 'packet' is inside a Netlink
3877 * attribute: pushing 4 bytes will just overwrite the 4-byte
3878 * "struct nlattr", which is fine since we don't need that
3879 * header anymore. */
3880 eth_push_vlan(packet, flow->vlan_tci);
3882 /* We can't reproduce 'key' from 'flow'. */
3883 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3889 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3894 *fitnessp = fitness;
3900 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3903 struct dpif_upcall *upcall;
3904 struct flow_miss *miss;
3905 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3906 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3907 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3917 /* Construct the to-do list.
3919 * This just amounts to extracting the flow from each packet and sticking
3920 * the packets that have the same flow in the same "flow_miss" structure so
3921 * that we can process them together. */
3924 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3925 struct flow_miss *miss = &misses[n_misses];
3926 struct flow_miss *existing_miss;
3927 struct ofproto_dpif *ofproto;
3928 uint32_t odp_in_port;
3933 error = ofproto_receive(backer, upcall->packet, upcall->key,
3934 upcall->key_len, &flow, &miss->key_fitness,
3935 &ofproto, &odp_in_port, &miss->initial_vals);
3936 if (error == ENODEV) {
3937 struct drop_key *drop_key;
3939 /* Received packet on port for which we couldn't associate
3940 * an ofproto. This can happen if a port is removed while
3941 * traffic is being received. Print a rate-limited message
3942 * in case it happens frequently. Install a drop flow so
3943 * that future packets of the flow are inexpensively dropped
3945 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3948 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3950 drop_key = xmalloc(sizeof *drop_key);
3951 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3952 drop_key->key_len = upcall->key_len;
3954 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3955 hash_bytes(drop_key->key, drop_key->key_len, 0));
3956 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3957 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3965 ofproto->n_missed++;
3966 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3967 &flow.tunnel, flow.in_port, &miss->flow);
3969 /* Add other packets to a to-do list. */
3970 hash = flow_hash(&miss->flow, 0);
3971 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3972 if (!existing_miss) {
3973 hmap_insert(&todo, &miss->hmap_node, hash);
3974 miss->ofproto = ofproto;
3975 miss->key = upcall->key;
3976 miss->key_len = upcall->key_len;
3977 miss->upcall_type = upcall->type;
3978 miss->odp_in_port = odp_in_port;
3979 list_init(&miss->packets);
3983 miss = existing_miss;
3985 list_push_back(&miss->packets, &upcall->packet->list_node);
3988 /* Process each element in the to-do list, constructing the set of
3989 * operations to batch. */
3991 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3992 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3994 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3996 /* Execute batch. */
3997 for (i = 0; i < n_ops; i++) {
3998 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4000 dpif_operate(backer->dpif, dpif_ops, n_ops);
4003 for (i = 0; i < n_ops; i++) {
4004 free(flow_miss_ops[i].garbage);
4006 hmap_destroy(&todo);
4009 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
4010 classify_upcall(const struct dpif_upcall *upcall)
4012 union user_action_cookie cookie;
4014 /* First look at the upcall type. */
4015 switch (upcall->type) {
4016 case DPIF_UC_ACTION:
4022 case DPIF_N_UC_TYPES:
4024 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4028 /* "action" upcalls need a closer look. */
4029 if (!upcall->userdata) {
4030 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4033 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
4034 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4035 nl_attr_get_size(upcall->userdata));
4038 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
4039 switch (cookie.type) {
4040 case USER_ACTION_COOKIE_SFLOW:
4041 return SFLOW_UPCALL;
4043 case USER_ACTION_COOKIE_SLOW_PATH:
4046 case USER_ACTION_COOKIE_UNSPEC:
4048 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
4049 nl_attr_get_u64(upcall->userdata));
4055 handle_sflow_upcall(struct dpif_backer *backer,
4056 const struct dpif_upcall *upcall)
4058 struct ofproto_dpif *ofproto;
4059 union user_action_cookie cookie;
4061 uint32_t odp_in_port;
4063 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4064 &flow, NULL, &ofproto, &odp_in_port, NULL)
4065 || !ofproto->sflow) {
4069 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
4070 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4071 odp_in_port, &cookie);
4075 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4077 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4078 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4079 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4084 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4087 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4088 struct dpif_upcall *upcall = &misses[n_misses];
4089 struct ofpbuf *buf = &miss_bufs[n_misses];
4092 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4093 sizeof miss_buf_stubs[n_misses]);
4094 error = dpif_recv(backer->dpif, upcall, buf);
4100 switch (classify_upcall(upcall)) {
4102 /* Handle it later. */
4107 handle_sflow_upcall(backer, upcall);
4117 /* Handle deferred MISS_UPCALL processing. */
4118 handle_miss_upcalls(backer, misses, n_misses);
4119 for (i = 0; i < n_misses; i++) {
4120 ofpbuf_uninit(&miss_bufs[i]);
4126 /* Flow expiration. */
4128 static int subfacet_max_idle(const struct ofproto_dpif *);
4129 static void update_stats(struct dpif_backer *);
4130 static void rule_expire(struct rule_dpif *);
4131 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4133 /* This function is called periodically by run(). Its job is to collect
4134 * updates for the flows that have been installed into the datapath, most
4135 * importantly when they last were used, and then use that information to
4136 * expire flows that have not been used recently.
4138 * Returns the number of milliseconds after which it should be called again. */
4140 expire(struct dpif_backer *backer)
4142 struct ofproto_dpif *ofproto;
4143 int max_idle = INT32_MAX;
4145 /* Periodically clear out the drop keys in an effort to keep them
4146 * relatively few. */
4147 drop_key_clear(backer);
4149 /* Update stats for each flow in the backer. */
4150 update_stats(backer);
4152 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4153 struct rule *rule, *next_rule;
4156 if (ofproto->backer != backer) {
4160 /* Keep track of the max number of flows per ofproto_dpif. */
4161 update_max_subfacet_count(ofproto);
4163 /* Expire subfacets that have been idle too long. */
4164 dp_max_idle = subfacet_max_idle(ofproto);
4165 expire_subfacets(ofproto, dp_max_idle);
4167 max_idle = MIN(max_idle, dp_max_idle);
4169 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4171 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4172 &ofproto->up.expirable) {
4173 rule_expire(rule_dpif_cast(rule));
4176 /* All outstanding data in existing flows has been accounted, so it's a
4177 * good time to do bond rebalancing. */
4178 if (ofproto->has_bonded_bundles) {
4179 struct ofbundle *bundle;
4181 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4183 bond_rebalance(bundle->bond, &backer->revalidate_set);
4189 return MIN(max_idle, 1000);
4192 /* Updates flow table statistics given that the datapath just reported 'stats'
4193 * as 'subfacet''s statistics. */
4195 update_subfacet_stats(struct subfacet *subfacet,
4196 const struct dpif_flow_stats *stats)
4198 struct facet *facet = subfacet->facet;
4200 if (stats->n_packets >= subfacet->dp_packet_count) {
4201 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4202 facet->packet_count += extra;
4204 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4207 if (stats->n_bytes >= subfacet->dp_byte_count) {
4208 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4210 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4213 subfacet->dp_packet_count = stats->n_packets;
4214 subfacet->dp_byte_count = stats->n_bytes;
4216 facet->tcp_flags |= stats->tcp_flags;
4218 subfacet_update_time(subfacet, stats->used);
4219 if (facet->accounted_bytes < facet->byte_count) {
4221 facet_account(facet);
4222 facet->accounted_bytes = facet->byte_count;
4226 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4227 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4229 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4230 const struct nlattr *key, size_t key_len)
4232 if (!VLOG_DROP_WARN(&rl)) {
4236 odp_flow_key_format(key, key_len, &s);
4237 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4241 COVERAGE_INC(facet_unexpected);
4242 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4245 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4247 * This function also pushes statistics updates to rules which each facet
4248 * resubmits into. Generally these statistics will be accurate. However, if a
4249 * facet changes the rule it resubmits into at some time in between
4250 * update_stats() runs, it is possible that statistics accrued to the
4251 * old rule will be incorrectly attributed to the new rule. This could be
4252 * avoided by calling update_stats() whenever rules are created or
4253 * deleted. However, the performance impact of making so many calls to the
4254 * datapath do not justify the benefit of having perfectly accurate statistics.
4256 * In addition, this function maintains per ofproto flow hit counts. The patch
4257 * port is not treated specially. e.g. A packet ingress from br0 patched into
4258 * br1 will increase the hit count of br0 by 1, however, does not affect
4259 * the hit or miss counts of br1.
4262 update_stats(struct dpif_backer *backer)
4264 const struct dpif_flow_stats *stats;
4265 struct dpif_flow_dump dump;
4266 const struct nlattr *key;
4267 struct ofproto_dpif *ofproto;
4270 dpif_flow_dump_start(&dump, backer->dpif);
4271 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4273 struct subfacet *subfacet;
4274 struct ofport_dpif *ofport;
4277 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4282 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4283 ofproto->n_update_stats++;
4285 ofport = get_ofp_port(ofproto, flow.in_port);
4286 if (ofport && ofport->tnl_port) {
4287 netdev_vport_inc_rx(ofport->up.netdev, stats);
4290 key_hash = odp_flow_key_hash(key, key_len);
4291 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4292 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4294 /* Update ofproto_dpif's hit count. */
4295 if (stats->n_packets > subfacet->dp_packet_count) {
4296 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4297 dpif_stats_update_hit_count(ofproto, delta);
4300 update_subfacet_stats(subfacet, stats);
4304 /* Stats are updated per-packet. */
4307 case SF_NOT_INSTALLED:
4309 delete_unexpected_flow(ofproto, key, key_len);
4314 dpif_flow_dump_done(&dump);
4316 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4317 update_moving_averages(ofproto);
4322 /* Calculates and returns the number of milliseconds of idle time after which
4323 * subfacets should expire from the datapath. When a subfacet expires, we fold
4324 * its statistics into its facet, and when a facet's last subfacet expires, we
4325 * fold its statistic into its rule. */
4327 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4330 * Idle time histogram.
4332 * Most of the time a switch has a relatively small number of subfacets.
4333 * When this is the case we might as well keep statistics for all of them
4334 * in userspace and to cache them in the kernel datapath for performance as
4337 * As the number of subfacets increases, the memory required to maintain
4338 * statistics about them in userspace and in the kernel becomes
4339 * significant. However, with a large number of subfacets it is likely
4340 * that only a few of them are "heavy hitters" that consume a large amount
4341 * of bandwidth. At this point, only heavy hitters are worth caching in
4342 * the kernel and maintaining in userspaces; other subfacets we can
4345 * The technique used to compute the idle time is to build a histogram with
4346 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4347 * that is installed in the kernel gets dropped in the appropriate bucket.
4348 * After the histogram has been built, we compute the cutoff so that only
4349 * the most-recently-used 1% of subfacets (but at least
4350 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4351 * the most-recently-used bucket of subfacets is kept, so actually an
4352 * arbitrary number of subfacets can be kept in any given expiration run
4353 * (though the next run will delete most of those unless they receive
4356 * This requires a second pass through the subfacets, in addition to the
4357 * pass made by update_stats(), because the former function never looks at
4358 * uninstallable subfacets.
4360 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4361 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4362 int buckets[N_BUCKETS] = { 0 };
4363 int total, subtotal, bucket;
4364 struct subfacet *subfacet;
4368 total = hmap_count(&ofproto->subfacets);
4369 if (total <= ofproto->up.flow_eviction_threshold) {
4370 return N_BUCKETS * BUCKET_WIDTH;
4373 /* Build histogram. */
4375 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4376 long long int idle = now - subfacet->used;
4377 int bucket = (idle <= 0 ? 0
4378 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4379 : (unsigned int) idle / BUCKET_WIDTH);
4383 /* Find the first bucket whose flows should be expired. */
4384 subtotal = bucket = 0;
4386 subtotal += buckets[bucket++];
4387 } while (bucket < N_BUCKETS &&
4388 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4390 if (VLOG_IS_DBG_ENABLED()) {
4394 ds_put_cstr(&s, "keep");
4395 for (i = 0; i < N_BUCKETS; i++) {
4397 ds_put_cstr(&s, ", drop");
4400 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4403 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4407 return bucket * BUCKET_WIDTH;
4411 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4413 /* Cutoff time for most flows. */
4414 long long int normal_cutoff = time_msec() - dp_max_idle;
4416 /* We really want to keep flows for special protocols around, so use a more
4417 * conservative cutoff. */
4418 long long int special_cutoff = time_msec() - 10000;
4420 struct subfacet *subfacet, *next_subfacet;
4421 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4425 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4426 &ofproto->subfacets) {
4427 long long int cutoff;
4429 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4432 if (subfacet->used < cutoff) {
4433 if (subfacet->path != SF_NOT_INSTALLED) {
4434 batch[n_batch++] = subfacet;
4435 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4436 subfacet_destroy_batch(ofproto, batch, n_batch);
4440 subfacet_destroy(subfacet);
4446 subfacet_destroy_batch(ofproto, batch, n_batch);
4450 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4451 * then delete it entirely. */
4453 rule_expire(struct rule_dpif *rule)
4455 struct facet *facet, *next_facet;
4459 if (rule->up.pending) {
4460 /* We'll have to expire it later. */
4464 /* Has 'rule' expired? */
4466 if (rule->up.hard_timeout
4467 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4468 reason = OFPRR_HARD_TIMEOUT;
4469 } else if (rule->up.idle_timeout
4470 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4471 reason = OFPRR_IDLE_TIMEOUT;
4476 COVERAGE_INC(ofproto_dpif_expired);
4478 /* Update stats. (This is a no-op if the rule expired due to an idle
4479 * timeout, because that only happens when the rule has no facets left.) */
4480 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4481 facet_remove(facet);
4484 /* Get rid of the rule. */
4485 ofproto_rule_expire(&rule->up, reason);
4490 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4492 * The caller must already have determined that no facet with an identical
4493 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4494 * the ofproto's classifier table.
4496 * 'hash' must be the return value of flow_hash(flow, 0).
4498 * The facet will initially have no subfacets. The caller should create (at
4499 * least) one subfacet with subfacet_create(). */
4500 static struct facet *
4501 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4504 struct facet *facet;
4506 facet = xzalloc(sizeof *facet);
4507 facet->used = time_msec();
4508 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4509 list_push_back(&rule->facets, &facet->list_node);
4511 facet->flow = *flow;
4512 list_init(&facet->subfacets);
4513 netflow_flow_init(&facet->nf_flow);
4514 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4516 facet->learn_rl = time_msec() + 500;
4522 facet_free(struct facet *facet)
4527 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4528 * 'packet', which arrived on 'in_port'. */
4530 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4531 const struct nlattr *odp_actions, size_t actions_len,
4532 struct ofpbuf *packet)
4534 struct odputil_keybuf keybuf;
4538 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4539 odp_flow_key_from_flow(&key, flow,
4540 ofp_port_to_odp_port(ofproto, flow->in_port));
4542 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4543 odp_actions, actions_len, packet);
4547 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4549 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4550 * rule's statistics, via subfacet_uninstall().
4552 * - Removes 'facet' from its rule and from ofproto->facets.
4555 facet_remove(struct facet *facet)
4557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4558 struct subfacet *subfacet, *next_subfacet;
4560 ovs_assert(!list_is_empty(&facet->subfacets));
4562 /* First uninstall all of the subfacets to get final statistics. */
4563 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4564 subfacet_uninstall(subfacet);
4567 /* Flush the final stats to the rule.
4569 * This might require us to have at least one subfacet around so that we
4570 * can use its actions for accounting in facet_account(), which is why we
4571 * have uninstalled but not yet destroyed the subfacets. */
4572 facet_flush_stats(facet);
4574 /* Now we're really all done so destroy everything. */
4575 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4576 &facet->subfacets) {
4577 subfacet_destroy__(subfacet);
4579 hmap_remove(&ofproto->facets, &facet->hmap_node);
4580 list_remove(&facet->list_node);
4584 /* Feed information from 'facet' back into the learning table to keep it in
4585 * sync with what is actually flowing through the datapath. */
4587 facet_learn(struct facet *facet)
4589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4590 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4591 struct subfacet, list_node);
4592 long long int now = time_msec();
4593 struct action_xlate_ctx ctx;
4595 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4599 facet->learn_rl = now + 500;
4601 if (!facet->has_learn
4602 && !facet->has_normal
4603 && (!facet->has_fin_timeout
4604 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4608 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4609 &subfacet->initial_vals,
4610 facet->rule, facet->tcp_flags, NULL);
4611 ctx.may_learn = true;
4612 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4613 facet->rule->up.ofpacts_len);
4617 facet_account(struct facet *facet)
4619 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4620 struct subfacet *subfacet = facet_get_subfacet(facet);
4621 const struct nlattr *a;
4626 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4629 n_bytes = facet->byte_count - facet->accounted_bytes;
4631 /* This loop feeds byte counters to bond_account() for rebalancing to use
4632 * as a basis. We also need to track the actual VLAN on which the packet
4633 * is going to be sent to ensure that it matches the one passed to
4634 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4637 * We use the actions from an arbitrary subfacet because they should all
4638 * be equally valid for our purpose. */
4639 vlan_tci = facet->flow.vlan_tci;
4640 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4641 subfacet->actions, subfacet->actions_len) {
4642 const struct ovs_action_push_vlan *vlan;
4643 struct ofport_dpif *port;
4645 switch (nl_attr_type(a)) {
4646 case OVS_ACTION_ATTR_OUTPUT:
4647 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4648 if (port && port->bundle && port->bundle->bond) {
4649 bond_account(port->bundle->bond, &facet->flow,
4650 vlan_tci_to_vid(vlan_tci), n_bytes);
4654 case OVS_ACTION_ATTR_POP_VLAN:
4655 vlan_tci = htons(0);
4658 case OVS_ACTION_ATTR_PUSH_VLAN:
4659 vlan = nl_attr_get(a);
4660 vlan_tci = vlan->vlan_tci;
4666 /* Returns true if the only action for 'facet' is to send to the controller.
4667 * (We don't report NetFlow expiration messages for such facets because they
4668 * are just part of the control logic for the network, not real traffic). */
4670 facet_is_controller_flow(struct facet *facet)
4673 const struct rule *rule = &facet->rule->up;
4674 const struct ofpact *ofpacts = rule->ofpacts;
4675 size_t ofpacts_len = rule->ofpacts_len;
4677 if (ofpacts_len > 0 &&
4678 ofpacts->type == OFPACT_CONTROLLER &&
4679 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4686 /* Folds all of 'facet''s statistics into its rule. Also updates the
4687 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4688 * 'facet''s statistics in the datapath should have been zeroed and folded into
4689 * its packet and byte counts before this function is called. */
4691 facet_flush_stats(struct facet *facet)
4693 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4694 struct subfacet *subfacet;
4696 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4697 ovs_assert(!subfacet->dp_byte_count);
4698 ovs_assert(!subfacet->dp_packet_count);
4701 facet_push_stats(facet);
4702 if (facet->accounted_bytes < facet->byte_count) {
4703 facet_account(facet);
4704 facet->accounted_bytes = facet->byte_count;
4707 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4708 struct ofexpired expired;
4709 expired.flow = facet->flow;
4710 expired.packet_count = facet->packet_count;
4711 expired.byte_count = facet->byte_count;
4712 expired.used = facet->used;
4713 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4716 facet->rule->packet_count += facet->packet_count;
4717 facet->rule->byte_count += facet->byte_count;
4719 /* Reset counters to prevent double counting if 'facet' ever gets
4721 facet_reset_counters(facet);
4723 netflow_flow_clear(&facet->nf_flow);
4724 facet->tcp_flags = 0;
4727 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4728 * Returns it if found, otherwise a null pointer.
4730 * 'hash' must be the return value of flow_hash(flow, 0).
4732 * The returned facet might need revalidation; use facet_lookup_valid()
4733 * instead if that is important. */
4734 static struct facet *
4735 facet_find(struct ofproto_dpif *ofproto,
4736 const struct flow *flow, uint32_t hash)
4738 struct facet *facet;
4740 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4741 if (flow_equal(flow, &facet->flow)) {
4749 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4750 * Returns it if found, otherwise a null pointer.
4752 * 'hash' must be the return value of flow_hash(flow, 0).
4754 * The returned facet is guaranteed to be valid. */
4755 static struct facet *
4756 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4759 struct facet *facet;
4761 facet = facet_find(ofproto, flow, hash);
4763 && (ofproto->backer->need_revalidate
4764 || tag_set_intersects(&ofproto->backer->revalidate_set,
4766 facet_revalidate(facet);
4768 /* facet_revalidate() may have destroyed 'facet'. */
4769 facet = facet_find(ofproto, flow, hash);
4775 /* Return a subfacet from 'facet'. A facet consists of one or more
4776 * subfacets, and this function returns one of them. */
4777 static struct subfacet *facet_get_subfacet(struct facet *facet)
4779 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4784 subfacet_path_to_string(enum subfacet_path path)
4787 case SF_NOT_INSTALLED:
4788 return "not installed";
4790 return "in fast path";
4792 return "in slow path";
4798 /* Returns the path in which a subfacet should be installed if its 'slow'
4799 * member has the specified value. */
4800 static enum subfacet_path
4801 subfacet_want_path(enum slow_path_reason slow)
4803 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4806 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4807 * supposing that its actions have been recalculated as 'want_actions' and that
4808 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4810 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4811 const struct ofpbuf *want_actions)
4813 enum subfacet_path want_path = subfacet_want_path(slow);
4814 return (want_path != subfacet->path
4815 || (want_path == SF_FAST_PATH
4816 && (subfacet->actions_len != want_actions->size
4817 || memcmp(subfacet->actions, want_actions->data,
4818 subfacet->actions_len))));
4822 facet_check_consistency(struct facet *facet)
4824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4828 uint64_t odp_actions_stub[1024 / 8];
4829 struct ofpbuf odp_actions;
4831 struct rule_dpif *rule;
4832 struct subfacet *subfacet;
4833 bool may_log = false;
4836 /* Check the rule for consistency. */
4837 rule = rule_dpif_lookup(ofproto, &facet->flow);
4838 ok = rule == facet->rule;
4840 may_log = !VLOG_DROP_WARN(&rl);
4845 flow_format(&s, &facet->flow);
4846 ds_put_format(&s, ": facet associated with wrong rule (was "
4847 "table=%"PRIu8",", facet->rule->up.table_id);
4848 cls_rule_format(&facet->rule->up.cr, &s);
4849 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4851 cls_rule_format(&rule->up.cr, &s);
4852 ds_put_char(&s, ')');
4854 VLOG_WARN("%s", ds_cstr(&s));
4859 /* Check the datapath actions for consistency. */
4860 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4861 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4862 enum subfacet_path want_path;
4863 struct action_xlate_ctx ctx;
4866 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4867 &subfacet->initial_vals, rule, 0, NULL);
4868 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4871 if (subfacet->path == SF_NOT_INSTALLED) {
4872 /* This only happens if the datapath reported an error when we
4873 * tried to install the flow. Don't flag another error here. */
4877 want_path = subfacet_want_path(subfacet->slow);
4878 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4879 /* The actions for slow-path flows may legitimately vary from one
4880 * packet to the next. We're done. */
4884 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4888 /* Inconsistency! */
4890 may_log = !VLOG_DROP_WARN(&rl);
4894 /* Rate-limited, skip reporting. */
4899 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4901 ds_put_cstr(&s, ": inconsistency in subfacet");
4902 if (want_path != subfacet->path) {
4903 enum odp_key_fitness fitness = subfacet->key_fitness;
4905 ds_put_format(&s, " (%s, fitness=%s)",
4906 subfacet_path_to_string(subfacet->path),
4907 odp_key_fitness_to_string(fitness));
4908 ds_put_format(&s, " (should have been %s)",
4909 subfacet_path_to_string(want_path));
4910 } else if (want_path == SF_FAST_PATH) {
4911 ds_put_cstr(&s, " (actions were: ");
4912 format_odp_actions(&s, subfacet->actions,
4913 subfacet->actions_len);
4914 ds_put_cstr(&s, ") (correct actions: ");
4915 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4916 ds_put_char(&s, ')');
4918 ds_put_cstr(&s, " (actions: ");
4919 format_odp_actions(&s, subfacet->actions,
4920 subfacet->actions_len);
4921 ds_put_char(&s, ')');
4923 VLOG_WARN("%s", ds_cstr(&s));
4926 ofpbuf_uninit(&odp_actions);
4931 /* Re-searches the classifier for 'facet':
4933 * - If the rule found is different from 'facet''s current rule, moves
4934 * 'facet' to the new rule and recompiles its actions.
4936 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4937 * where it is and recompiles its actions anyway.
4939 * - If any of 'facet''s subfacets correspond to a new flow according to
4940 * ofproto_receive(), 'facet' is removed. */
4942 facet_revalidate(struct facet *facet)
4944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4946 struct nlattr *odp_actions;
4949 struct actions *new_actions;
4951 struct action_xlate_ctx ctx;
4952 uint64_t odp_actions_stub[1024 / 8];
4953 struct ofpbuf odp_actions;
4955 struct rule_dpif *new_rule;
4956 struct subfacet *subfacet;
4959 COVERAGE_INC(facet_revalidate);
4961 /* Check that child subfacets still correspond to this facet. Tunnel
4962 * configuration changes could cause a subfacet's OpenFlow in_port to
4964 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4965 struct ofproto_dpif *recv_ofproto;
4966 struct flow recv_flow;
4969 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4970 subfacet->key_len, &recv_flow, NULL,
4971 &recv_ofproto, NULL, NULL);
4973 || recv_ofproto != ofproto
4974 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4975 facet_remove(facet);
4980 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4982 /* Calculate new datapath actions.
4984 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4985 * emit a NetFlow expiration and, if so, we need to have the old state
4986 * around to properly compose it. */
4988 /* If the datapath actions changed or the installability changed,
4989 * then we need to talk to the datapath. */
4992 memset(&ctx, 0, sizeof ctx);
4993 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4994 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4995 enum slow_path_reason slow;
4997 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4998 &subfacet->initial_vals, new_rule, 0, NULL);
4999 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5002 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5003 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5004 struct dpif_flow_stats stats;
5006 subfacet_install(subfacet,
5007 odp_actions.data, odp_actions.size, &stats, slow);
5008 subfacet_update_stats(subfacet, &stats);
5011 new_actions = xcalloc(list_size(&facet->subfacets),
5012 sizeof *new_actions);
5014 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5016 new_actions[i].actions_len = odp_actions.size;
5021 ofpbuf_uninit(&odp_actions);
5024 facet_flush_stats(facet);
5027 /* Update 'facet' now that we've taken care of all the old state. */
5028 facet->tags = ctx.tags;
5029 facet->nf_flow.output_iface = ctx.nf_output_iface;
5030 facet->has_learn = ctx.has_learn;
5031 facet->has_normal = ctx.has_normal;
5032 facet->has_fin_timeout = ctx.has_fin_timeout;
5033 facet->mirrors = ctx.mirrors;
5036 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5037 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5039 if (new_actions && new_actions[i].odp_actions) {
5040 free(subfacet->actions);
5041 subfacet->actions = new_actions[i].odp_actions;
5042 subfacet->actions_len = new_actions[i].actions_len;
5048 if (facet->rule != new_rule) {
5049 COVERAGE_INC(facet_changed_rule);
5050 list_remove(&facet->list_node);
5051 list_push_back(&new_rule->facets, &facet->list_node);
5052 facet->rule = new_rule;
5053 facet->used = new_rule->up.created;
5054 facet->prev_used = facet->used;
5058 /* Updates 'facet''s used time. Caller is responsible for calling
5059 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5061 facet_update_time(struct facet *facet, long long int used)
5063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5064 if (used > facet->used) {
5066 ofproto_rule_update_used(&facet->rule->up, used);
5067 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5072 facet_reset_counters(struct facet *facet)
5074 facet->packet_count = 0;
5075 facet->byte_count = 0;
5076 facet->prev_packet_count = 0;
5077 facet->prev_byte_count = 0;
5078 facet->accounted_bytes = 0;
5082 facet_push_stats(struct facet *facet)
5084 struct dpif_flow_stats stats;
5086 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5087 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5088 ovs_assert(facet->used >= facet->prev_used);
5090 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5091 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5092 stats.used = facet->used;
5093 stats.tcp_flags = 0;
5095 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5096 facet->prev_packet_count = facet->packet_count;
5097 facet->prev_byte_count = facet->byte_count;
5098 facet->prev_used = facet->used;
5100 flow_push_stats(facet, &stats);
5102 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5103 facet->mirrors, stats.n_packets, stats.n_bytes);
5108 push_all_stats__(bool run_fast)
5110 static long long int rl = LLONG_MIN;
5111 struct ofproto_dpif *ofproto;
5113 if (time_msec() < rl) {
5117 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5118 struct facet *facet;
5120 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5121 facet_push_stats(facet);
5128 rl = time_msec() + 100;
5132 push_all_stats(void)
5134 push_all_stats__(true);
5138 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5140 rule->packet_count += stats->n_packets;
5141 rule->byte_count += stats->n_bytes;
5142 ofproto_rule_update_used(&rule->up, stats->used);
5145 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5146 * into given 'facet->rule''s actions and mirrors. */
5148 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5150 struct rule_dpif *rule = facet->rule;
5151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5152 struct subfacet *subfacet = facet_get_subfacet(facet);
5153 struct action_xlate_ctx ctx;
5155 ofproto_rule_update_used(&rule->up, stats->used);
5157 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5158 &subfacet->initial_vals, rule, 0, NULL);
5159 ctx.resubmit_stats = stats;
5160 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5161 rule->up.ofpacts_len);
5166 static struct subfacet *
5167 subfacet_find(struct ofproto_dpif *ofproto,
5168 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5170 struct subfacet *subfacet;
5172 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5173 &ofproto->subfacets) {
5174 if (subfacet->key_len == key_len
5175 && !memcmp(key, subfacet->key, key_len)) {
5183 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5184 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5185 * existing subfacet if there is one, otherwise creates and returns a
5188 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5189 * which case the caller must populate the actions with
5190 * subfacet_make_actions(). */
5191 static struct subfacet *
5192 subfacet_create(struct facet *facet, struct flow_miss *miss,
5195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5196 enum odp_key_fitness key_fitness = miss->key_fitness;
5197 const struct nlattr *key = miss->key;
5198 size_t key_len = miss->key_len;
5200 struct subfacet *subfacet;
5202 key_hash = odp_flow_key_hash(key, key_len);
5204 if (list_is_empty(&facet->subfacets)) {
5205 subfacet = &facet->one_subfacet;
5207 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5209 if (subfacet->facet == facet) {
5213 /* This shouldn't happen. */
5214 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5215 subfacet_destroy(subfacet);
5218 subfacet = xmalloc(sizeof *subfacet);
5221 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5222 list_push_back(&facet->subfacets, &subfacet->list_node);
5223 subfacet->facet = facet;
5224 subfacet->key_fitness = key_fitness;
5225 subfacet->key = xmemdup(key, key_len);
5226 subfacet->key_len = key_len;
5227 subfacet->used = now;
5228 subfacet->created = now;
5229 subfacet->dp_packet_count = 0;
5230 subfacet->dp_byte_count = 0;
5231 subfacet->actions_len = 0;
5232 subfacet->actions = NULL;
5233 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5236 subfacet->path = SF_NOT_INSTALLED;
5237 subfacet->initial_vals = miss->initial_vals;
5238 subfacet->odp_in_port = miss->odp_in_port;
5240 ofproto->subfacet_add_count++;
5244 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5245 * its facet within 'ofproto', and frees it. */
5247 subfacet_destroy__(struct subfacet *subfacet)
5249 struct facet *facet = subfacet->facet;
5250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5252 /* Update ofproto stats before uninstall the subfacet. */
5253 ofproto->subfacet_del_count++;
5254 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5256 subfacet_uninstall(subfacet);
5257 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5258 list_remove(&subfacet->list_node);
5259 free(subfacet->key);
5260 free(subfacet->actions);
5261 if (subfacet != &facet->one_subfacet) {
5266 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5267 * last remaining subfacet in its facet destroys the facet too. */
5269 subfacet_destroy(struct subfacet *subfacet)
5271 struct facet *facet = subfacet->facet;
5273 if (list_is_singleton(&facet->subfacets)) {
5274 /* facet_remove() needs at least one subfacet (it will remove it). */
5275 facet_remove(facet);
5277 subfacet_destroy__(subfacet);
5282 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5283 struct subfacet **subfacets, int n)
5285 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5286 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5287 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5290 for (i = 0; i < n; i++) {
5291 ops[i].type = DPIF_OP_FLOW_DEL;
5292 ops[i].u.flow_del.key = subfacets[i]->key;
5293 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5294 ops[i].u.flow_del.stats = &stats[i];
5298 dpif_operate(ofproto->backer->dpif, opsp, n);
5299 for (i = 0; i < n; i++) {
5300 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5301 subfacets[i]->path = SF_NOT_INSTALLED;
5302 subfacet_destroy(subfacets[i]);
5307 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5308 * Translates the actions into 'odp_actions', which the caller must have
5309 * initialized and is responsible for uninitializing. */
5311 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5312 struct ofpbuf *odp_actions)
5314 struct facet *facet = subfacet->facet;
5315 struct rule_dpif *rule = facet->rule;
5316 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5318 struct action_xlate_ctx ctx;
5320 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5321 &subfacet->initial_vals, rule, 0, packet);
5322 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5323 facet->tags = ctx.tags;
5324 facet->has_learn = ctx.has_learn;
5325 facet->has_normal = ctx.has_normal;
5326 facet->has_fin_timeout = ctx.has_fin_timeout;
5327 facet->nf_flow.output_iface = ctx.nf_output_iface;
5328 facet->mirrors = ctx.mirrors;
5330 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5331 if (subfacet->actions_len != odp_actions->size
5332 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5333 free(subfacet->actions);
5334 subfacet->actions_len = odp_actions->size;
5335 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5339 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5340 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5341 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5342 * since 'subfacet' was last updated.
5344 * Returns 0 if successful, otherwise a positive errno value. */
5346 subfacet_install(struct subfacet *subfacet,
5347 const struct nlattr *actions, size_t actions_len,
5348 struct dpif_flow_stats *stats,
5349 enum slow_path_reason slow)
5351 struct facet *facet = subfacet->facet;
5352 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5353 enum subfacet_path path = subfacet_want_path(slow);
5354 uint64_t slow_path_stub[128 / 8];
5355 enum dpif_flow_put_flags flags;
5358 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5360 flags |= DPIF_FP_ZERO_STATS;
5363 if (path == SF_SLOW_PATH) {
5364 compose_slow_path(ofproto, &facet->flow, slow,
5365 slow_path_stub, sizeof slow_path_stub,
5366 &actions, &actions_len);
5369 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5370 subfacet->key_len, actions, actions_len, stats);
5373 subfacet_reset_dp_stats(subfacet, stats);
5377 subfacet->path = path;
5383 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5385 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5386 stats, subfacet->slow);
5389 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5391 subfacet_uninstall(struct subfacet *subfacet)
5393 if (subfacet->path != SF_NOT_INSTALLED) {
5394 struct rule_dpif *rule = subfacet->facet->rule;
5395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5396 struct dpif_flow_stats stats;
5399 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5400 subfacet->key_len, &stats);
5401 subfacet_reset_dp_stats(subfacet, &stats);
5403 subfacet_update_stats(subfacet, &stats);
5405 subfacet->path = SF_NOT_INSTALLED;
5407 ovs_assert(subfacet->dp_packet_count == 0);
5408 ovs_assert(subfacet->dp_byte_count == 0);
5412 /* Resets 'subfacet''s datapath statistics counters. This should be called
5413 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5414 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5415 * was reset in the datapath. 'stats' will be modified to include only
5416 * statistics new since 'subfacet' was last updated. */
5418 subfacet_reset_dp_stats(struct subfacet *subfacet,
5419 struct dpif_flow_stats *stats)
5422 && subfacet->dp_packet_count <= stats->n_packets
5423 && subfacet->dp_byte_count <= stats->n_bytes) {
5424 stats->n_packets -= subfacet->dp_packet_count;
5425 stats->n_bytes -= subfacet->dp_byte_count;
5428 subfacet->dp_packet_count = 0;
5429 subfacet->dp_byte_count = 0;
5432 /* Updates 'subfacet''s used time. The caller is responsible for calling
5433 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5435 subfacet_update_time(struct subfacet *subfacet, long long int used)
5437 if (used > subfacet->used) {
5438 subfacet->used = used;
5439 facet_update_time(subfacet->facet, used);
5443 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5445 * Because of the meaning of a subfacet's counters, it only makes sense to do
5446 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5447 * represents a packet that was sent by hand or if it represents statistics
5448 * that have been cleared out of the datapath. */
5450 subfacet_update_stats(struct subfacet *subfacet,
5451 const struct dpif_flow_stats *stats)
5453 if (stats->n_packets || stats->used > subfacet->used) {
5454 struct facet *facet = subfacet->facet;
5456 subfacet_update_time(subfacet, stats->used);
5457 facet->packet_count += stats->n_packets;
5458 facet->byte_count += stats->n_bytes;
5459 facet->tcp_flags |= stats->tcp_flags;
5460 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5466 static struct rule_dpif *
5467 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5469 struct rule_dpif *rule;
5471 rule = rule_dpif_lookup__(ofproto, flow, 0);
5476 return rule_dpif_miss_rule(ofproto, flow);
5479 static struct rule_dpif *
5480 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5483 struct cls_rule *cls_rule;
5484 struct classifier *cls;
5486 if (table_id >= N_TABLES) {
5490 cls = &ofproto->up.tables[table_id].cls;
5491 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5492 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5493 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5494 * are unavailable. */
5495 struct flow ofpc_normal_flow = *flow;
5496 ofpc_normal_flow.tp_src = htons(0);
5497 ofpc_normal_flow.tp_dst = htons(0);
5498 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5500 cls_rule = classifier_lookup(cls, flow);
5502 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5505 static struct rule_dpif *
5506 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5508 struct ofport_dpif *port;
5510 port = get_ofp_port(ofproto, flow->in_port);
5512 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5513 return ofproto->miss_rule;
5516 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5517 return ofproto->no_packet_in_rule;
5519 return ofproto->miss_rule;
5523 complete_operation(struct rule_dpif *rule)
5525 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5527 rule_invalidate(rule);
5529 struct dpif_completion *c = xmalloc(sizeof *c);
5530 c->op = rule->up.pending;
5531 list_push_back(&ofproto->completions, &c->list_node);
5533 ofoperation_complete(rule->up.pending, 0);
5537 static struct rule *
5540 struct rule_dpif *rule = xmalloc(sizeof *rule);
5545 rule_dealloc(struct rule *rule_)
5547 struct rule_dpif *rule = rule_dpif_cast(rule_);
5552 rule_construct(struct rule *rule_)
5554 struct rule_dpif *rule = rule_dpif_cast(rule_);
5555 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5556 struct rule_dpif *victim;
5559 rule->packet_count = 0;
5560 rule->byte_count = 0;
5562 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5563 if (victim && !list_is_empty(&victim->facets)) {
5564 struct facet *facet;
5566 rule->facets = victim->facets;
5567 list_moved(&rule->facets);
5568 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5569 /* XXX: We're only clearing our local counters here. It's possible
5570 * that quite a few packets are unaccounted for in the datapath
5571 * statistics. These will be accounted to the new rule instead of
5572 * cleared as required. This could be fixed by clearing out the
5573 * datapath statistics for this facet, but currently it doesn't
5575 facet_reset_counters(facet);
5579 /* Must avoid list_moved() in this case. */
5580 list_init(&rule->facets);
5583 table_id = rule->up.table_id;
5585 rule->tag = victim->tag;
5586 } else if (table_id == 0) {
5591 miniflow_expand(&rule->up.cr.match.flow, &flow);
5592 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5593 ofproto->tables[table_id].basis);
5596 complete_operation(rule);
5601 rule_destruct(struct rule *rule_)
5603 struct rule_dpif *rule = rule_dpif_cast(rule_);
5604 struct facet *facet, *next_facet;
5606 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5607 facet_revalidate(facet);
5610 complete_operation(rule);
5614 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5616 struct rule_dpif *rule = rule_dpif_cast(rule_);
5617 struct facet *facet;
5619 /* push_all_stats() can handle flow misses which, when using the learn
5620 * action, can cause rules to be added and deleted. This can corrupt our
5621 * caller's datastructures which assume that rule_get_stats() doesn't have
5622 * an impact on the flow table. To be safe, we disable miss handling. */
5623 push_all_stats__(false);
5625 /* Start from historical data for 'rule' itself that are no longer tracked
5626 * in facets. This counts, for example, facets that have expired. */
5627 *packets = rule->packet_count;
5628 *bytes = rule->byte_count;
5630 /* Add any statistics that are tracked by facets. This includes
5631 * statistical data recently updated by ofproto_update_stats() as well as
5632 * stats for packets that were executed "by hand" via dpif_execute(). */
5633 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5634 *packets += facet->packet_count;
5635 *bytes += facet->byte_count;
5640 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5641 struct ofpbuf *packet)
5643 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5644 struct initial_vals initial_vals;
5645 struct dpif_flow_stats stats;
5646 struct action_xlate_ctx ctx;
5647 uint64_t odp_actions_stub[1024 / 8];
5648 struct ofpbuf odp_actions;
5650 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5651 rule_credit_stats(rule, &stats);
5653 initial_vals.vlan_tci = flow->vlan_tci;
5654 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5655 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5656 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5657 rule, stats.tcp_flags, packet);
5658 ctx.resubmit_stats = &stats;
5659 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5661 execute_odp_actions(ofproto, flow, odp_actions.data,
5662 odp_actions.size, packet);
5664 ofpbuf_uninit(&odp_actions);
5668 rule_execute(struct rule *rule, const struct flow *flow,
5669 struct ofpbuf *packet)
5671 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5672 ofpbuf_delete(packet);
5677 rule_modify_actions(struct rule *rule_)
5679 struct rule_dpif *rule = rule_dpif_cast(rule_);
5681 complete_operation(rule);
5684 /* Sends 'packet' out 'ofport'.
5685 * May modify 'packet'.
5686 * Returns 0 if successful, otherwise a positive errno value. */
5688 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5690 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5691 uint64_t odp_actions_stub[1024 / 8];
5692 struct ofpbuf key, odp_actions;
5693 struct odputil_keybuf keybuf;
5698 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5699 if (netdev_vport_is_patch(ofport->up.netdev)) {
5700 struct ofproto_dpif *peer_ofproto;
5701 struct dpif_flow_stats stats;
5702 struct ofport_dpif *peer;
5703 struct rule_dpif *rule;
5705 peer = ofport_get_peer(ofport);
5710 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5711 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5712 netdev_vport_inc_rx(peer->up.netdev, &stats);
5714 flow.in_port = peer->up.ofp_port;
5715 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5716 rule = rule_dpif_lookup(peer_ofproto, &flow);
5717 rule_dpif_execute(rule, &flow, packet);
5722 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5724 if (ofport->tnl_port) {
5725 struct dpif_flow_stats stats;
5727 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5728 if (odp_port == OVSP_NONE) {
5732 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5733 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5734 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5735 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5737 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5739 if (odp_port != ofport->odp_port) {
5740 eth_pop_vlan(packet);
5741 flow.vlan_tci = htons(0);
5745 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5746 odp_flow_key_from_flow(&key, &flow,
5747 ofp_port_to_odp_port(ofproto, flow.in_port));
5749 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5751 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5752 error = dpif_execute(ofproto->backer->dpif,
5754 odp_actions.data, odp_actions.size,
5756 ofpbuf_uninit(&odp_actions);
5759 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5760 ofproto->up.name, odp_port, strerror(error));
5762 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5766 /* OpenFlow to datapath action translation. */
5768 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5769 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5770 struct action_xlate_ctx *);
5771 static void xlate_normal(struct action_xlate_ctx *);
5773 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5774 * The action will state 'slow' as the reason that the action is in the slow
5775 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5776 * dump-flows" output to see why a flow is in the slow path.)
5778 * The 'stub_size' bytes in 'stub' will be used to store the action.
5779 * 'stub_size' must be large enough for the action.
5781 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5784 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5785 enum slow_path_reason slow,
5786 uint64_t *stub, size_t stub_size,
5787 const struct nlattr **actionsp, size_t *actions_lenp)
5789 union user_action_cookie cookie;
5792 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5793 cookie.slow_path.unused = 0;
5794 cookie.slow_path.reason = slow;
5796 ofpbuf_use_stack(&buf, stub, stub_size);
5797 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5798 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5799 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5801 put_userspace_action(ofproto, &buf, flow, &cookie);
5803 *actionsp = buf.data;
5804 *actions_lenp = buf.size;
5808 put_userspace_action(const struct ofproto_dpif *ofproto,
5809 struct ofpbuf *odp_actions,
5810 const struct flow *flow,
5811 const union user_action_cookie *cookie)
5815 pid = dpif_port_get_pid(ofproto->backer->dpif,
5816 ofp_port_to_odp_port(ofproto, flow->in_port));
5818 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5822 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5823 ovs_be16 vlan_tci, uint32_t odp_port,
5824 unsigned int n_outputs, union user_action_cookie *cookie)
5828 cookie->type = USER_ACTION_COOKIE_SFLOW;
5829 cookie->sflow.vlan_tci = vlan_tci;
5831 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5832 * port information") for the interpretation of cookie->output. */
5833 switch (n_outputs) {
5835 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5836 cookie->sflow.output = 0x40000000 | 256;
5840 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5842 cookie->sflow.output = ifindex;
5847 /* 0x80000000 means "multiple output ports. */
5848 cookie->sflow.output = 0x80000000 | n_outputs;
5853 /* Compose SAMPLE action for sFlow. */
5855 compose_sflow_action(const struct ofproto_dpif *ofproto,
5856 struct ofpbuf *odp_actions,
5857 const struct flow *flow,
5860 uint32_t probability;
5861 union user_action_cookie cookie;
5862 size_t sample_offset, actions_offset;
5865 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5869 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5871 /* Number of packets out of UINT_MAX to sample. */
5872 probability = dpif_sflow_get_probability(ofproto->sflow);
5873 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5875 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5876 compose_sflow_cookie(ofproto, htons(0), odp_port,
5877 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5878 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5880 nl_msg_end_nested(odp_actions, actions_offset);
5881 nl_msg_end_nested(odp_actions, sample_offset);
5882 return cookie_offset;
5885 /* SAMPLE action must be first action in any given list of actions.
5886 * At this point we do not have all information required to build it. So try to
5887 * build sample action as complete as possible. */
5889 add_sflow_action(struct action_xlate_ctx *ctx)
5891 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5893 &ctx->flow, OVSP_NONE);
5894 ctx->sflow_odp_port = 0;
5895 ctx->sflow_n_outputs = 0;
5898 /* Fix SAMPLE action according to data collected while composing ODP actions.
5899 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5900 * USERSPACE action's user-cookie which is required for sflow. */
5902 fix_sflow_action(struct action_xlate_ctx *ctx)
5904 const struct flow *base = &ctx->base_flow;
5905 union user_action_cookie *cookie;
5907 if (!ctx->user_cookie_offset) {
5911 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5913 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5915 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5916 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5920 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5923 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5924 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5925 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5926 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5927 struct priority_to_dscp *pdscp;
5928 uint32_t out_port, odp_port;
5930 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5931 * before traversing a patch port. */
5932 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
5935 xlate_report(ctx, "Nonexistent output port");
5937 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5938 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5940 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5941 xlate_report(ctx, "STP not in forwarding state, skipping output");
5945 if (netdev_vport_is_patch(ofport->up.netdev)) {
5946 struct ofport_dpif *peer = ofport_get_peer(ofport);
5947 struct flow old_flow = ctx->flow;
5948 const struct ofproto_dpif *peer_ofproto;
5949 enum slow_path_reason special;
5950 struct ofport_dpif *in_port;
5953 xlate_report(ctx, "Nonexistent patch port peer");
5957 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5958 if (peer_ofproto->backer != ctx->ofproto->backer) {
5959 xlate_report(ctx, "Patch port peer on a different datapath");
5963 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5964 ctx->flow.in_port = peer->up.ofp_port;
5965 ctx->flow.metadata = htonll(0);
5966 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5967 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5969 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5970 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5973 ctx->slow |= special;
5974 } else if (!in_port || may_receive(in_port, ctx)) {
5975 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5976 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5978 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5979 * learning action look at the packet, then drop it. */
5980 struct flow old_base_flow = ctx->base_flow;
5981 size_t old_size = ctx->odp_actions->size;
5982 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5983 ctx->base_flow = old_base_flow;
5984 ctx->odp_actions->size = old_size;
5988 ctx->flow = old_flow;
5989 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5991 if (ctx->resubmit_stats) {
5992 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5993 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5999 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6001 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6002 ctx->flow.nw_tos |= pdscp->dscp;
6005 if (ofport->tnl_port) {
6006 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6007 if (odp_port == OVSP_NONE) {
6008 xlate_report(ctx, "Tunneling decided against output");
6012 if (ctx->resubmit_stats) {
6013 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6015 out_port = odp_port;
6016 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6019 odp_port = ofport->odp_port;
6020 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6021 ctx->flow.vlan_tci);
6022 if (out_port != odp_port) {
6023 ctx->flow.vlan_tci = htons(0);
6025 ctx->flow.skb_mark &= ~IPSEC_MARK;
6027 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6028 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6030 ctx->sflow_odp_port = odp_port;
6031 ctx->sflow_n_outputs++;
6032 ctx->nf_output_iface = ofp_port;
6033 ctx->flow.tunnel.tun_id = flow_tun_id;
6034 ctx->flow.vlan_tci = flow_vlan_tci;
6035 ctx->flow.nw_tos = flow_nw_tos;
6039 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6041 compose_output_action__(ctx, ofp_port, true);
6045 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6047 struct ofproto_dpif *ofproto = ctx->ofproto;
6048 uint8_t table_id = ctx->table_id;
6050 if (table_id > 0 && table_id < N_TABLES) {
6051 struct table_dpif *table = &ofproto->tables[table_id];
6052 if (table->other_table) {
6053 ctx->tags |= (rule && rule->tag
6055 : rule_calculate_tag(&ctx->flow,
6056 &table->other_table->mask,
6062 /* Common rule processing in one place to avoid duplicating code. */
6063 static struct rule_dpif *
6064 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6067 if (ctx->resubmit_hook) {
6068 ctx->resubmit_hook(ctx, rule);
6070 if (rule == NULL && may_packet_in) {
6072 * check if table configuration flags
6073 * OFPTC_TABLE_MISS_CONTROLLER, default.
6074 * OFPTC_TABLE_MISS_CONTINUE,
6075 * OFPTC_TABLE_MISS_DROP
6076 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6078 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6080 if (rule && ctx->resubmit_stats) {
6081 rule_credit_stats(rule, ctx->resubmit_stats);
6087 xlate_table_action(struct action_xlate_ctx *ctx,
6088 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6090 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6091 struct rule_dpif *rule;
6092 uint16_t old_in_port = ctx->flow.in_port;
6093 uint8_t old_table_id = ctx->table_id;
6095 ctx->table_id = table_id;
6097 /* Look up a flow with 'in_port' as the input port. */
6098 ctx->flow.in_port = in_port;
6099 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6101 tag_the_flow(ctx, rule);
6103 /* Restore the original input port. Otherwise OFPP_NORMAL and
6104 * OFPP_IN_PORT will have surprising behavior. */
6105 ctx->flow.in_port = old_in_port;
6107 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6110 struct rule_dpif *old_rule = ctx->rule;
6114 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6115 ctx->rule = old_rule;
6119 ctx->table_id = old_table_id;
6121 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6123 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6124 MAX_RESUBMIT_RECURSION);
6125 ctx->max_resubmit_trigger = true;
6130 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6131 const struct ofpact_resubmit *resubmit)
6136 in_port = resubmit->in_port;
6137 if (in_port == OFPP_IN_PORT) {
6138 in_port = ctx->flow.in_port;
6141 table_id = resubmit->table_id;
6142 if (table_id == 255) {
6143 table_id = ctx->table_id;
6146 xlate_table_action(ctx, in_port, table_id, false);
6150 flood_packets(struct action_xlate_ctx *ctx, bool all)
6152 struct ofport_dpif *ofport;
6154 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6155 uint16_t ofp_port = ofport->up.ofp_port;
6157 if (ofp_port == ctx->flow.in_port) {
6162 compose_output_action__(ctx, ofp_port, false);
6163 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6164 compose_output_action(ctx, ofp_port);
6168 ctx->nf_output_iface = NF_OUT_FLOOD;
6172 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6173 enum ofp_packet_in_reason reason,
6174 uint16_t controller_id)
6176 struct ofputil_packet_in pin;
6177 struct ofpbuf *packet;
6179 ctx->slow |= SLOW_CONTROLLER;
6184 packet = ofpbuf_clone(ctx->packet);
6186 if (packet->l2 && packet->l3) {
6187 struct eth_header *eh;
6188 uint16_t mpls_depth;
6190 eth_pop_vlan(packet);
6193 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6194 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6196 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6197 eth_push_vlan(packet, ctx->flow.vlan_tci);
6200 mpls_depth = eth_mpls_depth(packet);
6202 if (mpls_depth < ctx->flow.mpls_depth) {
6203 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6204 } else if (mpls_depth > ctx->flow.mpls_depth) {
6205 pop_mpls(packet, ctx->flow.dl_type);
6206 } else if (mpls_depth) {
6207 set_mpls_lse(packet, ctx->flow.mpls_lse);
6211 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6212 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6213 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6217 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6218 packet_set_tcp_port(packet, ctx->flow.tp_src,
6220 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6221 packet_set_udp_port(packet, ctx->flow.tp_src,
6228 pin.packet = packet->data;
6229 pin.packet_len = packet->size;
6230 pin.reason = reason;
6231 pin.controller_id = controller_id;
6232 pin.table_id = ctx->table_id;
6233 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6236 flow_get_metadata(&ctx->flow, &pin.fmd);
6238 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6239 ofpbuf_delete(packet);
6243 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6245 ovs_assert(eth_type_mpls(eth_type));
6247 if (ctx->base_flow.mpls_depth) {
6248 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6249 ctx->flow.mpls_depth++;
6254 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6255 label = htonl(0x2); /* IPV6 Explicit Null. */
6257 label = htonl(0x0); /* IPV4 Explicit Null. */
6259 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6260 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6261 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6262 ctx->flow.mpls_depth = 1;
6264 ctx->flow.dl_type = eth_type;
6268 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6270 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6271 ovs_assert(!eth_type_mpls(eth_type));
6273 if (ctx->flow.mpls_depth) {
6274 ctx->flow.mpls_depth--;
6275 ctx->flow.mpls_lse = htonl(0);
6276 if (!ctx->flow.mpls_depth) {
6277 ctx->flow.dl_type = eth_type;
6283 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6285 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6286 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6290 if (ctx->flow.nw_ttl > 1) {
6296 for (i = 0; i < ids->n_controllers; i++) {
6297 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6301 /* Stop processing for current table. */
6307 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6309 if (!eth_type_mpls(ctx->flow.dl_type)) {
6313 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6318 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6320 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6322 if (!eth_type_mpls(ctx->flow.dl_type)) {
6328 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6331 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6333 /* Stop processing for current table. */
6339 xlate_output_action(struct action_xlate_ctx *ctx,
6340 uint16_t port, uint16_t max_len, bool may_packet_in)
6342 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6344 ctx->nf_output_iface = NF_OUT_DROP;
6348 compose_output_action(ctx, ctx->flow.in_port);
6351 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6357 flood_packets(ctx, false);
6360 flood_packets(ctx, true);
6362 case OFPP_CONTROLLER:
6363 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6369 if (port != ctx->flow.in_port) {
6370 compose_output_action(ctx, port);
6372 xlate_report(ctx, "skipping output to input port");
6377 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6378 ctx->nf_output_iface = NF_OUT_FLOOD;
6379 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6380 ctx->nf_output_iface = prev_nf_output_iface;
6381 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6382 ctx->nf_output_iface != NF_OUT_FLOOD) {
6383 ctx->nf_output_iface = NF_OUT_MULTI;
6388 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6389 const struct ofpact_output_reg *or)
6391 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6392 if (port <= UINT16_MAX) {
6393 xlate_output_action(ctx, port, or->max_len, false);
6398 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6399 const struct ofpact_enqueue *enqueue)
6401 uint16_t ofp_port = enqueue->port;
6402 uint32_t queue_id = enqueue->queue;
6403 uint32_t flow_priority, priority;
6406 /* Translate queue to priority. */
6407 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6408 queue_id, &priority);
6410 /* Fall back to ordinary output action. */
6411 xlate_output_action(ctx, enqueue->port, 0, false);
6415 /* Check output port. */
6416 if (ofp_port == OFPP_IN_PORT) {
6417 ofp_port = ctx->flow.in_port;
6418 } else if (ofp_port == ctx->flow.in_port) {
6422 /* Add datapath actions. */
6423 flow_priority = ctx->flow.skb_priority;
6424 ctx->flow.skb_priority = priority;
6425 compose_output_action(ctx, ofp_port);
6426 ctx->flow.skb_priority = flow_priority;
6428 /* Update NetFlow output port. */
6429 if (ctx->nf_output_iface == NF_OUT_DROP) {
6430 ctx->nf_output_iface = ofp_port;
6431 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6432 ctx->nf_output_iface = NF_OUT_MULTI;
6437 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6439 uint32_t skb_priority;
6441 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6442 queue_id, &skb_priority)) {
6443 ctx->flow.skb_priority = skb_priority;
6445 /* Couldn't translate queue to a priority. Nothing to do. A warning
6446 * has already been logged. */
6451 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6453 struct ofproto_dpif *ofproto = ofproto_;
6454 struct ofport_dpif *port;
6464 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6467 port = get_ofp_port(ofproto, ofp_port);
6468 return port ? port->may_enable : false;
6473 xlate_bundle_action(struct action_xlate_ctx *ctx,
6474 const struct ofpact_bundle *bundle)
6478 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6479 if (bundle->dst.field) {
6480 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6482 xlate_output_action(ctx, port, 0, false);
6487 xlate_learn_action(struct action_xlate_ctx *ctx,
6488 const struct ofpact_learn *learn)
6490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6491 struct ofputil_flow_mod fm;
6492 uint64_t ofpacts_stub[1024 / 8];
6493 struct ofpbuf ofpacts;
6496 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6497 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6499 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6500 if (error && !VLOG_DROP_WARN(&rl)) {
6501 VLOG_WARN("learning action failed to modify flow table (%s)",
6502 ofperr_get_name(error));
6505 ofpbuf_uninit(&ofpacts);
6508 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6509 * means "infinite". */
6511 reduce_timeout(uint16_t max, uint16_t *timeout)
6513 if (max && (!*timeout || *timeout > max)) {
6519 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6520 const struct ofpact_fin_timeout *oft)
6522 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6523 struct rule_dpif *rule = ctx->rule;
6525 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6526 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6531 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6533 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6534 ? OFPUTIL_PC_NO_RECV_STP
6535 : OFPUTIL_PC_NO_RECV)) {
6539 /* Only drop packets here if both forwarding and learning are
6540 * disabled. If just learning is enabled, we need to have
6541 * OFPP_NORMAL and the learning action have a look at the packet
6542 * before we can drop it. */
6543 if (!stp_forward_in_state(port->stp_state)
6544 && !stp_learn_in_state(port->stp_state)) {
6552 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6554 if (is_ip_any(&ctx->base_flow)
6555 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6556 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6557 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6558 " but is not ECN capable");
6561 /* Set the ECN CE value in the tunneled packet. */
6562 ctx->flow.nw_tos |= IP_ECN_CE;
6570 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6571 struct action_xlate_ctx *ctx)
6573 bool was_evictable = true;
6574 const struct ofpact *a;
6577 /* Don't let the rule we're working on get evicted underneath us. */
6578 was_evictable = ctx->rule->up.evictable;
6579 ctx->rule->up.evictable = false;
6582 do_xlate_actions_again:
6583 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6584 struct ofpact_controller *controller;
6585 const struct ofpact_metadata *metadata;
6593 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6594 ofpact_get_OUTPUT(a)->max_len, true);
6597 case OFPACT_CONTROLLER:
6598 controller = ofpact_get_CONTROLLER(a);
6599 execute_controller_action(ctx, controller->max_len,
6601 controller->controller_id);
6604 case OFPACT_ENQUEUE:
6605 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6608 case OFPACT_SET_VLAN_VID:
6609 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6610 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6614 case OFPACT_SET_VLAN_PCP:
6615 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6616 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6621 case OFPACT_STRIP_VLAN:
6622 ctx->flow.vlan_tci = htons(0);
6625 case OFPACT_PUSH_VLAN:
6626 /* XXX 802.1AD(QinQ) */
6627 ctx->flow.vlan_tci = htons(VLAN_CFI);
6630 case OFPACT_SET_ETH_SRC:
6631 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6635 case OFPACT_SET_ETH_DST:
6636 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6640 case OFPACT_SET_IPV4_SRC:
6641 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6642 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6646 case OFPACT_SET_IPV4_DST:
6647 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6648 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6652 case OFPACT_SET_IPV4_DSCP:
6653 /* OpenFlow 1.0 only supports IPv4. */
6654 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6655 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6656 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6660 case OFPACT_SET_L4_SRC_PORT:
6661 if (is_ip_any(&ctx->flow)) {
6662 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6666 case OFPACT_SET_L4_DST_PORT:
6667 if (is_ip_any(&ctx->flow)) {
6668 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6672 case OFPACT_RESUBMIT:
6673 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6676 case OFPACT_SET_TUNNEL:
6677 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6680 case OFPACT_SET_QUEUE:
6681 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6684 case OFPACT_POP_QUEUE:
6685 ctx->flow.skb_priority = ctx->orig_skb_priority;
6688 case OFPACT_REG_MOVE:
6689 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6692 case OFPACT_REG_LOAD:
6693 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6696 case OFPACT_STACK_PUSH:
6697 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6701 case OFPACT_STACK_POP:
6702 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6706 case OFPACT_PUSH_MPLS:
6707 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6710 case OFPACT_POP_MPLS:
6711 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6714 case OFPACT_SET_MPLS_TTL:
6715 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6720 case OFPACT_DEC_MPLS_TTL:
6721 if (execute_dec_mpls_ttl_action(ctx)) {
6726 case OFPACT_DEC_TTL:
6727 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6733 /* Nothing to do. */
6736 case OFPACT_MULTIPATH:
6737 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6741 ctx->ofproto->has_bundle_action = true;
6742 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6745 case OFPACT_OUTPUT_REG:
6746 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6750 ctx->has_learn = true;
6751 if (ctx->may_learn) {
6752 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6760 case OFPACT_FIN_TIMEOUT:
6761 ctx->has_fin_timeout = true;
6762 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6765 case OFPACT_CLEAR_ACTIONS:
6767 * Nothing to do because writa-actions is not supported for now.
6768 * When writa-actions is supported, clear-actions also must
6769 * be supported at the same time.
6773 case OFPACT_WRITE_METADATA:
6774 metadata = ofpact_get_WRITE_METADATA(a);
6775 ctx->flow.metadata &= ~metadata->mask;
6776 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6779 case OFPACT_GOTO_TABLE: {
6780 /* It is assumed that goto-table is the last action. */
6781 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6782 struct rule_dpif *rule;
6784 ovs_assert(ctx->table_id < ogt->table_id);
6786 ctx->table_id = ogt->table_id;
6788 /* Look up a flow from the new table. */
6789 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6791 tag_the_flow(ctx, rule);
6793 rule = ctx_rule_hooks(ctx, rule, true);
6797 ctx->rule->up.evictable = was_evictable;
6800 was_evictable = rule->up.evictable;
6801 rule->up.evictable = false;
6803 /* Tail recursion removal. */
6804 ofpacts = rule->up.ofpacts;
6805 ofpacts_len = rule->up.ofpacts_len;
6806 goto do_xlate_actions_again;
6815 ctx->rule->up.evictable = was_evictable;
6820 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6821 struct ofproto_dpif *ofproto, const struct flow *flow,
6822 const struct initial_vals *initial_vals,
6823 struct rule_dpif *rule,
6824 uint8_t tcp_flags, const struct ofpbuf *packet)
6826 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6828 /* Flow initialization rules:
6829 * - 'base_flow' must match the kernel's view of the packet at the
6830 * time that action processing starts. 'flow' represents any
6831 * transformations we wish to make through actions.
6832 * - By default 'base_flow' and 'flow' are the same since the input
6833 * packet matches the output before any actions are applied.
6834 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6835 * of the received packet as seen by the kernel. If we later output
6836 * to another device without any modifications this will cause us to
6837 * insert a new tag since the original one was stripped off by the
6839 * - Tunnel 'flow' is largely cleared when transitioning between
6840 * the input and output stages since it does not make sense to output
6841 * a packet with the exact headers that it was received with (i.e.
6842 * the destination IP is us). The one exception is the tun_id, which
6843 * is preserved to allow use in later resubmit lookups and loads into
6845 * - Tunnel 'base_flow' is completely cleared since that is what the
6846 * kernel does. If we wish to maintain the original values an action
6847 * needs to be generated. */
6849 ctx->ofproto = ofproto;
6851 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6852 ctx->base_flow = ctx->flow;
6853 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6854 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6855 ctx->flow.tunnel.tun_id = initial_tun_id;
6857 ctx->packet = packet;
6858 ctx->may_learn = packet != NULL;
6859 ctx->tcp_flags = tcp_flags;
6860 ctx->resubmit_hook = NULL;
6861 ctx->report_hook = NULL;
6862 ctx->resubmit_stats = NULL;
6865 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6866 * into datapath actions in 'odp_actions', using 'ctx'. */
6868 xlate_actions(struct action_xlate_ctx *ctx,
6869 const struct ofpact *ofpacts, size_t ofpacts_len,
6870 struct ofpbuf *odp_actions)
6872 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6873 * that in the future we always keep a copy of the original flow for
6874 * tracing purposes. */
6875 static bool hit_resubmit_limit;
6877 enum slow_path_reason special;
6878 struct ofport_dpif *in_port;
6879 struct flow orig_flow;
6881 COVERAGE_INC(ofproto_dpif_xlate);
6883 ofpbuf_clear(odp_actions);
6884 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6886 ctx->odp_actions = odp_actions;
6889 ctx->has_learn = false;
6890 ctx->has_normal = false;
6891 ctx->has_fin_timeout = false;
6892 ctx->nf_output_iface = NF_OUT_DROP;
6895 ctx->max_resubmit_trigger = false;
6896 ctx->orig_skb_priority = ctx->flow.skb_priority;
6900 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6902 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6903 /* Do this conditionally because the copy is expensive enough that it
6904 * shows up in profiles. */
6905 orig_flow = ctx->flow;
6908 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6909 switch (ctx->ofproto->up.frag_handling) {
6910 case OFPC_FRAG_NORMAL:
6911 /* We must pretend that transport ports are unavailable. */
6912 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6913 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6916 case OFPC_FRAG_DROP:
6919 case OFPC_FRAG_REASM:
6922 case OFPC_FRAG_NX_MATCH:
6923 /* Nothing to do. */
6926 case OFPC_INVALID_TTL_TO_CONTROLLER:
6931 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6932 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6934 ctx->slow |= special;
6936 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6937 struct initial_vals initial_vals;
6938 uint32_t local_odp_port;
6940 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6941 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6943 add_sflow_action(ctx);
6945 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6946 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6948 /* We've let OFPP_NORMAL and the learning action look at the
6949 * packet, so drop it now if forwarding is disabled. */
6950 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6951 ofpbuf_clear(ctx->odp_actions);
6952 add_sflow_action(ctx);
6956 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6957 if (!hit_resubmit_limit) {
6958 /* We didn't record the original flow. Make sure we do from
6960 hit_resubmit_limit = true;
6961 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6962 struct ds ds = DS_EMPTY_INITIALIZER;
6964 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6965 &initial_vals, &ds);
6966 VLOG_ERR("Trace triggered by excessive resubmit "
6967 "recursion:\n%s", ds_cstr(&ds));
6972 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6973 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6975 ctx->odp_actions->data,
6976 ctx->odp_actions->size)) {
6977 ctx->slow |= SLOW_IN_BAND;
6979 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6981 compose_output_action(ctx, OFPP_LOCAL);
6984 if (ctx->ofproto->has_mirrors) {
6985 add_mirror_actions(ctx, &orig_flow);
6987 fix_sflow_action(ctx);
6990 ofpbuf_uninit(&ctx->stack);
6993 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6994 * into datapath actions, using 'ctx', and discards the datapath actions. */
6996 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6997 const struct ofpact *ofpacts,
7000 uint64_t odp_actions_stub[1024 / 8];
7001 struct ofpbuf odp_actions;
7003 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7004 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7005 ofpbuf_uninit(&odp_actions);
7009 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7011 if (ctx->report_hook) {
7012 ctx->report_hook(ctx, s);
7016 /* OFPP_NORMAL implementation. */
7018 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7020 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7021 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7022 * the bundle on which the packet was received, returns the VLAN to which the
7025 * Both 'vid' and the return value are in the range 0...4095. */
7027 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7029 switch (in_bundle->vlan_mode) {
7030 case PORT_VLAN_ACCESS:
7031 return in_bundle->vlan;
7034 case PORT_VLAN_TRUNK:
7037 case PORT_VLAN_NATIVE_UNTAGGED:
7038 case PORT_VLAN_NATIVE_TAGGED:
7039 return vid ? vid : in_bundle->vlan;
7046 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7047 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7050 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7051 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7054 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7056 /* Allow any VID on the OFPP_NONE port. */
7057 if (in_bundle == &ofpp_none_bundle) {
7061 switch (in_bundle->vlan_mode) {
7062 case PORT_VLAN_ACCESS:
7065 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7066 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7067 "packet received on port %s configured as VLAN "
7068 "%"PRIu16" access port",
7069 in_bundle->ofproto->up.name, vid,
7070 in_bundle->name, in_bundle->vlan);
7076 case PORT_VLAN_NATIVE_UNTAGGED:
7077 case PORT_VLAN_NATIVE_TAGGED:
7079 /* Port must always carry its native VLAN. */
7083 case PORT_VLAN_TRUNK:
7084 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7086 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7087 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7088 "received on port %s not configured for trunking "
7090 in_bundle->ofproto->up.name, vid,
7091 in_bundle->name, vid);
7103 /* Given 'vlan', the VLAN that a packet belongs to, and
7104 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7105 * that should be included in the 802.1Q header. (If the return value is 0,
7106 * then the 802.1Q header should only be included in the packet if there is a
7109 * Both 'vlan' and the return value are in the range 0...4095. */
7111 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7113 switch (out_bundle->vlan_mode) {
7114 case PORT_VLAN_ACCESS:
7117 case PORT_VLAN_TRUNK:
7118 case PORT_VLAN_NATIVE_TAGGED:
7121 case PORT_VLAN_NATIVE_UNTAGGED:
7122 return vlan == out_bundle->vlan ? 0 : vlan;
7130 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7133 struct ofport_dpif *port;
7135 ovs_be16 tci, old_tci;
7137 vid = output_vlan_to_vid(out_bundle, vlan);
7138 if (!out_bundle->bond) {
7139 port = ofbundle_get_a_port(out_bundle);
7141 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7144 /* No slaves enabled, so drop packet. */
7149 old_tci = ctx->flow.vlan_tci;
7151 if (tci || out_bundle->use_priority_tags) {
7152 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7154 tci |= htons(VLAN_CFI);
7157 ctx->flow.vlan_tci = tci;
7159 compose_output_action(ctx, port->up.ofp_port);
7160 ctx->flow.vlan_tci = old_tci;
7164 mirror_mask_ffs(mirror_mask_t mask)
7166 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7171 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7173 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7174 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7178 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7180 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7183 /* Returns an arbitrary interface within 'bundle'. */
7184 static struct ofport_dpif *
7185 ofbundle_get_a_port(const struct ofbundle *bundle)
7187 return CONTAINER_OF(list_front(&bundle->ports),
7188 struct ofport_dpif, bundle_node);
7192 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7194 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7198 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7200 struct ofproto_dpif *ofproto = ctx->ofproto;
7201 mirror_mask_t mirrors;
7202 struct ofbundle *in_bundle;
7205 const struct nlattr *a;
7208 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7209 ctx->packet != NULL, NULL);
7213 mirrors = in_bundle->src_mirrors;
7215 /* Drop frames on bundles reserved for mirroring. */
7216 if (in_bundle->mirror_out) {
7217 if (ctx->packet != NULL) {
7218 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7219 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7220 "%s, which is reserved exclusively for mirroring",
7221 ctx->ofproto->up.name, in_bundle->name);
7227 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7228 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7231 vlan = input_vid_to_vlan(in_bundle, vid);
7233 /* Look at the output ports to check for destination selections. */
7235 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7236 ctx->odp_actions->size) {
7237 enum ovs_action_attr type = nl_attr_type(a);
7238 struct ofport_dpif *ofport;
7240 if (type != OVS_ACTION_ATTR_OUTPUT) {
7244 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7245 if (ofport && ofport->bundle) {
7246 mirrors |= ofport->bundle->dst_mirrors;
7254 /* Restore the original packet before adding the mirror actions. */
7255 ctx->flow = *orig_flow;
7260 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7262 if (!vlan_is_mirrored(m, vlan)) {
7263 mirrors = zero_rightmost_1bit(mirrors);
7267 mirrors &= ~m->dup_mirrors;
7268 ctx->mirrors |= m->dup_mirrors;
7270 output_normal(ctx, m->out, vlan);
7271 } else if (vlan != m->out_vlan
7272 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7273 struct ofbundle *bundle;
7275 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7276 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7277 && !bundle->mirror_out) {
7278 output_normal(ctx, bundle, m->out_vlan);
7286 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7287 uint64_t packets, uint64_t bytes)
7293 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7296 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7299 /* In normal circumstances 'm' will not be NULL. However,
7300 * if mirrors are reconfigured, we can temporarily get out
7301 * of sync in facet_revalidate(). We could "correct" the
7302 * mirror list before reaching here, but doing that would
7303 * not properly account the traffic stats we've currently
7304 * accumulated for previous mirror configuration. */
7308 m->packet_count += packets;
7309 m->byte_count += bytes;
7313 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7314 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7315 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7317 is_gratuitous_arp(const struct flow *flow)
7319 return (flow->dl_type == htons(ETH_TYPE_ARP)
7320 && eth_addr_is_broadcast(flow->dl_dst)
7321 && (flow->nw_proto == ARP_OP_REPLY
7322 || (flow->nw_proto == ARP_OP_REQUEST
7323 && flow->nw_src == flow->nw_dst)));
7327 update_learning_table(struct ofproto_dpif *ofproto,
7328 const struct flow *flow, int vlan,
7329 struct ofbundle *in_bundle)
7331 struct mac_entry *mac;
7333 /* Don't learn the OFPP_NONE port. */
7334 if (in_bundle == &ofpp_none_bundle) {
7338 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7342 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7343 if (is_gratuitous_arp(flow)) {
7344 /* We don't want to learn from gratuitous ARP packets that are
7345 * reflected back over bond slaves so we lock the learning table. */
7346 if (!in_bundle->bond) {
7347 mac_entry_set_grat_arp_lock(mac);
7348 } else if (mac_entry_is_grat_arp_locked(mac)) {
7353 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7354 /* The log messages here could actually be useful in debugging,
7355 * so keep the rate limit relatively high. */
7356 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7357 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7358 "on port %s in VLAN %d",
7359 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7360 in_bundle->name, vlan);
7362 mac->port.p = in_bundle;
7363 tag_set_add(&ofproto->backer->revalidate_set,
7364 mac_learning_changed(ofproto->ml, mac));
7368 static struct ofbundle *
7369 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7370 bool warn, struct ofport_dpif **in_ofportp)
7372 struct ofport_dpif *ofport;
7374 /* Find the port and bundle for the received packet. */
7375 ofport = get_ofp_port(ofproto, in_port);
7377 *in_ofportp = ofport;
7379 if (ofport && ofport->bundle) {
7380 return ofport->bundle;
7383 /* Special-case OFPP_NONE, which a controller may use as the ingress
7384 * port for traffic that it is sourcing. */
7385 if (in_port == OFPP_NONE) {
7386 return &ofpp_none_bundle;
7389 /* Odd. A few possible reasons here:
7391 * - We deleted a port but there are still a few packets queued up
7394 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7395 * we don't know about.
7397 * - The ofproto client didn't configure the port as part of a bundle.
7398 * This is particularly likely to happen if a packet was received on the
7399 * port after it was created, but before the client had a chance to
7400 * configure its bundle.
7403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7405 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7406 "port %"PRIu16, ofproto->up.name, in_port);
7411 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7412 * dropped. Returns true if they may be forwarded, false if they should be
7415 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7416 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7418 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7419 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7420 * checked by input_vid_is_valid().
7422 * May also add tags to '*tags', although the current implementation only does
7423 * so in one special case.
7426 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7429 struct ofproto_dpif *ofproto = ctx->ofproto;
7430 struct flow *flow = &ctx->flow;
7431 struct ofbundle *in_bundle = in_port->bundle;
7433 /* Drop frames for reserved multicast addresses
7434 * only if forward_bpdu option is absent. */
7435 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7436 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7440 if (in_bundle->bond) {
7441 struct mac_entry *mac;
7443 switch (bond_check_admissibility(in_bundle->bond, in_port,
7444 flow->dl_dst, &ctx->tags)) {
7449 xlate_report(ctx, "bonding refused admissibility, dropping");
7452 case BV_DROP_IF_MOVED:
7453 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7454 if (mac && mac->port.p != in_bundle &&
7455 (!is_gratuitous_arp(flow)
7456 || mac_entry_is_grat_arp_locked(mac))) {
7457 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7469 xlate_normal(struct action_xlate_ctx *ctx)
7471 struct ofport_dpif *in_port;
7472 struct ofbundle *in_bundle;
7473 struct mac_entry *mac;
7477 ctx->has_normal = true;
7479 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7480 ctx->packet != NULL, &in_port);
7482 xlate_report(ctx, "no input bundle, dropping");
7486 /* Drop malformed frames. */
7487 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7488 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7489 if (ctx->packet != NULL) {
7490 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7491 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7492 "VLAN tag received on port %s",
7493 ctx->ofproto->up.name, in_bundle->name);
7495 xlate_report(ctx, "partial VLAN tag, dropping");
7499 /* Drop frames on bundles reserved for mirroring. */
7500 if (in_bundle->mirror_out) {
7501 if (ctx->packet != NULL) {
7502 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7503 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7504 "%s, which is reserved exclusively for mirroring",
7505 ctx->ofproto->up.name, in_bundle->name);
7507 xlate_report(ctx, "input port is mirror output port, dropping");
7512 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7513 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7514 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7517 vlan = input_vid_to_vlan(in_bundle, vid);
7519 /* Check other admissibility requirements. */
7520 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7524 /* Learn source MAC. */
7525 if (ctx->may_learn) {
7526 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7529 /* Determine output bundle. */
7530 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7533 if (mac->port.p != in_bundle) {
7534 xlate_report(ctx, "forwarding to learned port");
7535 output_normal(ctx, mac->port.p, vlan);
7537 xlate_report(ctx, "learned port is input port, dropping");
7540 struct ofbundle *bundle;
7542 xlate_report(ctx, "no learned MAC for destination, flooding");
7543 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7544 if (bundle != in_bundle
7545 && ofbundle_includes_vlan(bundle, vlan)
7546 && bundle->floodable
7547 && !bundle->mirror_out) {
7548 output_normal(ctx, bundle, vlan);
7551 ctx->nf_output_iface = NF_OUT_FLOOD;
7555 /* Optimized flow revalidation.
7557 * It's a difficult problem, in general, to tell which facets need to have
7558 * their actions recalculated whenever the OpenFlow flow table changes. We
7559 * don't try to solve that general problem: for most kinds of OpenFlow flow
7560 * table changes, we recalculate the actions for every facet. This is
7561 * relatively expensive, but it's good enough if the OpenFlow flow table
7562 * doesn't change very often.
7564 * However, we can expect one particular kind of OpenFlow flow table change to
7565 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7566 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7567 * table, we add a special case that applies to flow tables in which every rule
7568 * has the same form (that is, the same wildcards), except that the table is
7569 * also allowed to have a single "catch-all" flow that matches all packets. We
7570 * optimize this case by tagging all of the facets that resubmit into the table
7571 * and invalidating the same tag whenever a flow changes in that table. The
7572 * end result is that we revalidate just the facets that need it (and sometimes
7573 * a few more, but not all of the facets or even all of the facets that
7574 * resubmit to the table modified by MAC learning). */
7576 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7577 * into an OpenFlow table with the given 'basis'. */
7579 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7582 if (minimask_is_catchall(mask)) {
7585 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7586 return tag_create_deterministic(hash);
7590 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7591 * taggability of that table.
7593 * This function must be called after *each* change to a flow table. If you
7594 * skip calling it on some changes then the pointer comparisons at the end can
7595 * be invalid if you get unlucky. For example, if a flow removal causes a
7596 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7597 * different wildcards to be created with the same address, then this function
7598 * will incorrectly skip revalidation. */
7600 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7602 struct table_dpif *table = &ofproto->tables[table_id];
7603 const struct oftable *oftable = &ofproto->up.tables[table_id];
7604 struct cls_table *catchall, *other;
7605 struct cls_table *t;
7607 catchall = other = NULL;
7609 switch (hmap_count(&oftable->cls.tables)) {
7611 /* We could tag this OpenFlow table but it would make the logic a
7612 * little harder and it's a corner case that doesn't seem worth it
7618 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7619 if (cls_table_is_catchall(t)) {
7621 } else if (!other) {
7624 /* Indicate that we can't tag this by setting both tables to
7625 * NULL. (We know that 'catchall' is already NULL.) */
7632 /* Can't tag this table. */
7636 if (table->catchall_table != catchall || table->other_table != other) {
7637 table->catchall_table = catchall;
7638 table->other_table = other;
7639 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7643 /* Given 'rule' that has changed in some way (either it is a rule being
7644 * inserted, a rule being deleted, or a rule whose actions are being
7645 * modified), marks facets for revalidation to ensure that packets will be
7646 * forwarded correctly according to the new state of the flow table.
7648 * This function must be called after *each* change to a flow table. See
7649 * the comment on table_update_taggable() for more information. */
7651 rule_invalidate(const struct rule_dpif *rule)
7653 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7655 table_update_taggable(ofproto, rule->up.table_id);
7657 if (!ofproto->backer->need_revalidate) {
7658 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7660 if (table->other_table && rule->tag) {
7661 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7663 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7669 set_frag_handling(struct ofproto *ofproto_,
7670 enum ofp_config_flags frag_handling)
7672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7673 if (frag_handling != OFPC_FRAG_REASM) {
7674 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7682 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7683 const struct flow *flow,
7684 const struct ofpact *ofpacts, size_t ofpacts_len)
7686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7687 struct initial_vals initial_vals;
7688 struct odputil_keybuf keybuf;
7689 struct dpif_flow_stats stats;
7693 struct action_xlate_ctx ctx;
7694 uint64_t odp_actions_stub[1024 / 8];
7695 struct ofpbuf odp_actions;
7697 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7698 odp_flow_key_from_flow(&key, flow,
7699 ofp_port_to_odp_port(ofproto, flow->in_port));
7701 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7703 initial_vals.vlan_tci = flow->vlan_tci;
7704 initial_vals.tunnel_ip_tos = 0;
7705 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7706 packet_get_tcp_flags(packet, flow), packet);
7707 ctx.resubmit_stats = &stats;
7709 ofpbuf_use_stub(&odp_actions,
7710 odp_actions_stub, sizeof odp_actions_stub);
7711 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7712 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7713 odp_actions.data, odp_actions.size, packet);
7714 ofpbuf_uninit(&odp_actions);
7722 set_netflow(struct ofproto *ofproto_,
7723 const struct netflow_options *netflow_options)
7725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7727 if (netflow_options) {
7728 if (!ofproto->netflow) {
7729 ofproto->netflow = netflow_create();
7731 return netflow_set_options(ofproto->netflow, netflow_options);
7733 netflow_destroy(ofproto->netflow);
7734 ofproto->netflow = NULL;
7740 get_netflow_ids(const struct ofproto *ofproto_,
7741 uint8_t *engine_type, uint8_t *engine_id)
7743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7745 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7749 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7751 if (!facet_is_controller_flow(facet) &&
7752 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7753 struct subfacet *subfacet;
7754 struct ofexpired expired;
7756 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7757 if (subfacet->path == SF_FAST_PATH) {
7758 struct dpif_flow_stats stats;
7760 subfacet_reinstall(subfacet, &stats);
7761 subfacet_update_stats(subfacet, &stats);
7765 expired.flow = facet->flow;
7766 expired.packet_count = facet->packet_count;
7767 expired.byte_count = facet->byte_count;
7768 expired.used = facet->used;
7769 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7774 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7776 struct facet *facet;
7778 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7779 send_active_timeout(ofproto, facet);
7783 static struct ofproto_dpif *
7784 ofproto_dpif_lookup(const char *name)
7786 struct ofproto_dpif *ofproto;
7788 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7789 hash_string(name, 0), &all_ofproto_dpifs) {
7790 if (!strcmp(ofproto->up.name, name)) {
7798 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7799 const char *argv[], void *aux OVS_UNUSED)
7801 struct ofproto_dpif *ofproto;
7804 ofproto = ofproto_dpif_lookup(argv[1]);
7806 unixctl_command_reply_error(conn, "no such bridge");
7809 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7811 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7812 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7816 unixctl_command_reply(conn, "table successfully flushed");
7820 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7821 const char *argv[], void *aux OVS_UNUSED)
7823 struct ds ds = DS_EMPTY_INITIALIZER;
7824 const struct ofproto_dpif *ofproto;
7825 const struct mac_entry *e;
7827 ofproto = ofproto_dpif_lookup(argv[1]);
7829 unixctl_command_reply_error(conn, "no such bridge");
7833 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7834 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7835 struct ofbundle *bundle = e->port.p;
7836 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7837 ofbundle_get_a_port(bundle)->odp_port,
7838 e->vlan, ETH_ADDR_ARGS(e->mac),
7839 mac_entry_age(ofproto->ml, e));
7841 unixctl_command_reply(conn, ds_cstr(&ds));
7846 struct action_xlate_ctx ctx;
7852 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7853 const struct rule_dpif *rule)
7855 ds_put_char_multiple(result, '\t', level);
7857 ds_put_cstr(result, "No match\n");
7861 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7862 table_id, ntohll(rule->up.flow_cookie));
7863 cls_rule_format(&rule->up.cr, result);
7864 ds_put_char(result, '\n');
7866 ds_put_char_multiple(result, '\t', level);
7867 ds_put_cstr(result, "OpenFlow ");
7868 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7869 ds_put_char(result, '\n');
7873 trace_format_flow(struct ds *result, int level, const char *title,
7874 struct trace_ctx *trace)
7876 ds_put_char_multiple(result, '\t', level);
7877 ds_put_format(result, "%s: ", title);
7878 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7879 ds_put_cstr(result, "unchanged");
7881 flow_format(result, &trace->ctx.flow);
7882 trace->flow = trace->ctx.flow;
7884 ds_put_char(result, '\n');
7888 trace_format_regs(struct ds *result, int level, const char *title,
7889 struct trace_ctx *trace)
7893 ds_put_char_multiple(result, '\t', level);
7894 ds_put_format(result, "%s:", title);
7895 for (i = 0; i < FLOW_N_REGS; i++) {
7896 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7898 ds_put_char(result, '\n');
7902 trace_format_odp(struct ds *result, int level, const char *title,
7903 struct trace_ctx *trace)
7905 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7907 ds_put_char_multiple(result, '\t', level);
7908 ds_put_format(result, "%s: ", title);
7909 format_odp_actions(result, odp_actions->data, odp_actions->size);
7910 ds_put_char(result, '\n');
7914 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7916 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7917 struct ds *result = trace->result;
7919 ds_put_char(result, '\n');
7920 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7921 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7922 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7923 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7927 trace_report(struct action_xlate_ctx *ctx, const char *s)
7929 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7930 struct ds *result = trace->result;
7932 ds_put_char_multiple(result, '\t', ctx->recurse);
7933 ds_put_cstr(result, s);
7934 ds_put_char(result, '\n');
7938 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7939 void *aux OVS_UNUSED)
7941 const char *dpname = argv[1];
7942 struct ofproto_dpif *ofproto;
7943 struct ofpbuf odp_key;
7944 struct ofpbuf *packet;
7945 struct initial_vals initial_vals;
7951 ofpbuf_init(&odp_key, 0);
7954 ofproto = ofproto_dpif_lookup(dpname);
7956 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7960 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7961 /* ofproto/trace dpname flow [-generate] */
7962 const char *flow_s = argv[2];
7963 const char *generate_s = argv[3];
7965 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7966 * flow. We guess which type it is based on whether 'flow_s' contains
7967 * an '(', since a datapath flow always contains '(') but an
7968 * OpenFlow-like flow should not (in fact it's allowed but I believe
7969 * that's not documented anywhere).
7971 * An alternative would be to try to parse 'flow_s' both ways, but then
7972 * it would be tricky giving a sensible error message. After all, do
7973 * you just say "syntax error" or do you present both error messages?
7974 * Both choices seem lousy. */
7975 if (strchr(flow_s, '(')) {
7978 /* Convert string to datapath key. */
7979 ofpbuf_init(&odp_key, 0);
7980 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7982 unixctl_command_reply_error(conn, "Bad flow syntax");
7986 /* The user might have specified the wrong ofproto but within the
7987 * same backer. That's OK, ofproto_receive() can find the right
7989 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7990 odp_key.size, &flow, NULL, &ofproto, NULL,
7992 unixctl_command_reply_error(conn, "Invalid flow");
7995 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7999 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8001 unixctl_command_reply_error(conn, error_s);
8006 initial_vals.vlan_tci = flow.vlan_tci;
8007 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8010 /* Generate a packet, if requested. */
8012 packet = ofpbuf_new(0);
8013 flow_compose(packet, &flow);
8015 } else if (argc == 7) {
8016 /* ofproto/trace dpname priority tun_id in_port mark packet */
8017 const char *priority_s = argv[2];
8018 const char *tun_id_s = argv[3];
8019 const char *in_port_s = argv[4];
8020 const char *mark_s = argv[5];
8021 const char *packet_s = argv[6];
8022 uint32_t in_port = atoi(in_port_s);
8023 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8024 uint32_t priority = atoi(priority_s);
8025 uint32_t mark = atoi(mark_s);
8028 msg = eth_from_hex(packet_s, &packet);
8030 unixctl_command_reply_error(conn, msg);
8034 ds_put_cstr(&result, "Packet: ");
8035 s = ofp_packet_to_string(packet->data, packet->size);
8036 ds_put_cstr(&result, s);
8039 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8040 flow.tunnel.tun_id = tun_id;
8041 initial_vals.vlan_tci = flow.vlan_tci;
8042 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8044 unixctl_command_reply_error(conn, "Bad command syntax");
8048 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8049 unixctl_command_reply(conn, ds_cstr(&result));
8052 ds_destroy(&result);
8053 ofpbuf_delete(packet);
8054 ofpbuf_uninit(&odp_key);
8058 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8059 const struct ofpbuf *packet,
8060 const struct initial_vals *initial_vals, struct ds *ds)
8062 struct rule_dpif *rule;
8064 ds_put_cstr(ds, "Flow: ");
8065 flow_format(ds, flow);
8066 ds_put_char(ds, '\n');
8068 rule = rule_dpif_lookup(ofproto, flow);
8070 trace_format_rule(ds, 0, 0, rule);
8071 if (rule == ofproto->miss_rule) {
8072 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8073 } else if (rule == ofproto->no_packet_in_rule) {
8074 ds_put_cstr(ds, "\nNo match, packets dropped because "
8075 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8079 uint64_t odp_actions_stub[1024 / 8];
8080 struct ofpbuf odp_actions;
8082 struct trace_ctx trace;
8085 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8088 ofpbuf_use_stub(&odp_actions,
8089 odp_actions_stub, sizeof odp_actions_stub);
8090 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8091 rule, tcp_flags, packet);
8092 trace.ctx.resubmit_hook = trace_resubmit;
8093 trace.ctx.report_hook = trace_report;
8094 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8097 ds_put_char(ds, '\n');
8098 trace_format_flow(ds, 0, "Final flow", &trace);
8099 ds_put_cstr(ds, "Datapath actions: ");
8100 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8101 ofpbuf_uninit(&odp_actions);
8103 if (trace.ctx.slow) {
8104 enum slow_path_reason slow;
8106 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8107 "slow path because it:");
8108 for (slow = trace.ctx.slow; slow; ) {
8109 enum slow_path_reason bit = rightmost_1bit(slow);
8113 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8116 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8119 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8122 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8125 ds_put_cstr(ds, "\n\t (The datapath actions are "
8126 "incomplete--for complete actions, "
8127 "please supply a packet.)");
8130 case SLOW_CONTROLLER:
8131 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8132 "to the OpenFlow controller.");
8135 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8136 "than the datapath supports.");
8143 if (slow & ~SLOW_MATCH) {
8144 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8145 "the special slow-path processing.");
8152 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8153 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8156 unixctl_command_reply(conn, NULL);
8160 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8161 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8164 unixctl_command_reply(conn, NULL);
8167 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8168 * 'reply' describing the results. */
8170 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8172 struct facet *facet;
8176 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8177 if (!facet_check_consistency(facet)) {
8182 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8186 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8187 ofproto->up.name, errors);
8189 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8194 ofproto_dpif_self_check(struct unixctl_conn *conn,
8195 int argc, const char *argv[], void *aux OVS_UNUSED)
8197 struct ds reply = DS_EMPTY_INITIALIZER;
8198 struct ofproto_dpif *ofproto;
8201 ofproto = ofproto_dpif_lookup(argv[1]);
8203 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8204 "ofproto/list for help)");
8207 ofproto_dpif_self_check__(ofproto, &reply);
8209 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8210 ofproto_dpif_self_check__(ofproto, &reply);
8214 unixctl_command_reply(conn, ds_cstr(&reply));
8218 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8219 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8220 * to destroy 'ofproto_shash' and free the returned value. */
8221 static const struct shash_node **
8222 get_ofprotos(struct shash *ofproto_shash)
8224 const struct ofproto_dpif *ofproto;
8226 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8227 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8228 shash_add_nocopy(ofproto_shash, name, ofproto);
8231 return shash_sort(ofproto_shash);
8235 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8236 const char *argv[] OVS_UNUSED,
8237 void *aux OVS_UNUSED)
8239 struct ds ds = DS_EMPTY_INITIALIZER;
8240 struct shash ofproto_shash;
8241 const struct shash_node **sorted_ofprotos;
8244 shash_init(&ofproto_shash);
8245 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8246 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8247 const struct shash_node *node = sorted_ofprotos[i];
8248 ds_put_format(&ds, "%s\n", node->name);
8251 shash_destroy(&ofproto_shash);
8252 free(sorted_ofprotos);
8254 unixctl_command_reply(conn, ds_cstr(&ds));
8259 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8261 const struct shash_node **ports;
8263 struct avg_subfacet_rates lifetime;
8264 unsigned long long int minutes;
8265 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8267 minutes = (time_msec() - ofproto->created) / min_ms;
8270 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8272 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8275 lifetime.add_rate = 0.0;
8276 lifetime.del_rate = 0.0;
8279 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8280 dpif_name(ofproto->backer->dpif));
8282 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8283 ofproto->n_hit, ofproto->n_missed);
8284 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8285 " life span: %llu(ms)\n",
8286 hmap_count(&ofproto->subfacets),
8287 avg_subfacet_count(ofproto),
8288 ofproto->max_n_subfacet,
8289 avg_subfacet_life_span(ofproto));
8290 if (minutes >= 60) {
8291 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8293 if (minutes >= 60 * 24) {
8294 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8296 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8298 ports = shash_sort(&ofproto->up.port_by_name);
8299 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8300 const struct shash_node *node = ports[i];
8301 struct ofport *ofport = node->data;
8302 const char *name = netdev_get_name(ofport->netdev);
8303 const char *type = netdev_get_type(ofport->netdev);
8306 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8308 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8309 if (odp_port != OVSP_NONE) {
8310 ds_put_format(ds, "%"PRIu32":", odp_port);
8312 ds_put_cstr(ds, "none:");
8315 if (strcmp(type, "system")) {
8316 struct netdev *netdev;
8319 ds_put_format(ds, " (%s", type);
8321 error = netdev_open(name, type, &netdev);
8326 error = netdev_get_config(netdev, &config);
8328 const struct smap_node **nodes;
8331 nodes = smap_sort(&config);
8332 for (i = 0; i < smap_count(&config); i++) {
8333 const struct smap_node *node = nodes[i];
8334 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8335 node->key, node->value);
8339 smap_destroy(&config);
8341 netdev_close(netdev);
8343 ds_put_char(ds, ')');
8345 ds_put_char(ds, '\n');
8351 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8352 const char *argv[], void *aux OVS_UNUSED)
8354 struct ds ds = DS_EMPTY_INITIALIZER;
8355 const struct ofproto_dpif *ofproto;
8359 for (i = 1; i < argc; i++) {
8360 ofproto = ofproto_dpif_lookup(argv[i]);
8362 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8363 "for help)", argv[i]);
8364 unixctl_command_reply_error(conn, ds_cstr(&ds));
8367 show_dp_format(ofproto, &ds);
8370 struct shash ofproto_shash;
8371 const struct shash_node **sorted_ofprotos;
8374 shash_init(&ofproto_shash);
8375 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8376 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8377 const struct shash_node *node = sorted_ofprotos[i];
8378 show_dp_format(node->data, &ds);
8381 shash_destroy(&ofproto_shash);
8382 free(sorted_ofprotos);
8385 unixctl_command_reply(conn, ds_cstr(&ds));
8390 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8391 int argc OVS_UNUSED, const char *argv[],
8392 void *aux OVS_UNUSED)
8394 struct ds ds = DS_EMPTY_INITIALIZER;
8395 const struct ofproto_dpif *ofproto;
8396 struct subfacet *subfacet;
8398 ofproto = ofproto_dpif_lookup(argv[1]);
8400 unixctl_command_reply_error(conn, "no such bridge");
8404 update_stats(ofproto->backer);
8406 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8407 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8409 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8410 subfacet->dp_packet_count, subfacet->dp_byte_count);
8411 if (subfacet->used) {
8412 ds_put_format(&ds, "%.3fs",
8413 (time_msec() - subfacet->used) / 1000.0);
8415 ds_put_format(&ds, "never");
8417 if (subfacet->facet->tcp_flags) {
8418 ds_put_cstr(&ds, ", flags:");
8419 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8422 ds_put_cstr(&ds, ", actions:");
8423 if (subfacet->slow) {
8424 uint64_t slow_path_stub[128 / 8];
8425 const struct nlattr *actions;
8428 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8429 slow_path_stub, sizeof slow_path_stub,
8430 &actions, &actions_len);
8431 format_odp_actions(&ds, actions, actions_len);
8433 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8435 ds_put_char(&ds, '\n');
8438 unixctl_command_reply(conn, ds_cstr(&ds));
8443 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8444 int argc OVS_UNUSED, const char *argv[],
8445 void *aux OVS_UNUSED)
8447 struct ds ds = DS_EMPTY_INITIALIZER;
8448 struct ofproto_dpif *ofproto;
8450 ofproto = ofproto_dpif_lookup(argv[1]);
8452 unixctl_command_reply_error(conn, "no such bridge");
8456 flush(&ofproto->up);
8458 unixctl_command_reply(conn, ds_cstr(&ds));
8463 ofproto_dpif_unixctl_init(void)
8465 static bool registered;
8471 unixctl_command_register(
8473 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8474 2, 6, ofproto_unixctl_trace, NULL);
8475 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8476 ofproto_unixctl_fdb_flush, NULL);
8477 unixctl_command_register("fdb/show", "bridge", 1, 1,
8478 ofproto_unixctl_fdb_show, NULL);
8479 unixctl_command_register("ofproto/clog", "", 0, 0,
8480 ofproto_dpif_clog, NULL);
8481 unixctl_command_register("ofproto/unclog", "", 0, 0,
8482 ofproto_dpif_unclog, NULL);
8483 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8484 ofproto_dpif_self_check, NULL);
8485 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8486 ofproto_unixctl_dpif_dump_dps, NULL);
8487 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8488 ofproto_unixctl_dpif_show, NULL);
8489 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8490 ofproto_unixctl_dpif_dump_flows, NULL);
8491 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8492 ofproto_unixctl_dpif_del_flows, NULL);
8495 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8497 * This is deprecated. It is only for compatibility with broken device drivers
8498 * in old versions of Linux that do not properly support VLANs when VLAN
8499 * devices are not used. When broken device drivers are no longer in
8500 * widespread use, we will delete these interfaces. */
8503 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8506 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8508 if (realdev_ofp_port == ofport->realdev_ofp_port
8509 && vid == ofport->vlandev_vid) {
8513 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8515 if (ofport->realdev_ofp_port) {
8518 if (realdev_ofp_port && ofport->bundle) {
8519 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8520 * themselves be part of a bundle. */
8521 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8524 ofport->realdev_ofp_port = realdev_ofp_port;
8525 ofport->vlandev_vid = vid;
8527 if (realdev_ofp_port) {
8528 vsp_add(ofport, realdev_ofp_port, vid);
8535 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8537 return hash_2words(realdev_ofp_port, vid);
8540 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8541 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8542 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8543 * it would return the port number of eth0.9.
8545 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8546 * function just returns its 'realdev_odp_port' argument. */
8548 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8549 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8551 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8552 uint16_t realdev_ofp_port;
8553 int vid = vlan_tci_to_vid(vlan_tci);
8554 const struct vlan_splinter *vsp;
8556 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8557 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8558 hash_realdev_vid(realdev_ofp_port, vid),
8559 &ofproto->realdev_vid_map) {
8560 if (vsp->realdev_ofp_port == realdev_ofp_port
8561 && vsp->vid == vid) {
8562 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8566 return realdev_odp_port;
8569 static struct vlan_splinter *
8570 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8572 struct vlan_splinter *vsp;
8574 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8575 &ofproto->vlandev_map) {
8576 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8584 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8585 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8586 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8587 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8588 * eth0 and store 9 in '*vid'.
8590 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8591 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8594 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8595 uint16_t vlandev_ofp_port, int *vid)
8597 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8598 const struct vlan_splinter *vsp;
8600 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8605 return vsp->realdev_ofp_port;
8611 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8612 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8613 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8614 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8615 * always the case unless VLAN splinters are enabled), returns false without
8616 * making any changes. */
8618 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8623 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8628 /* Cause the flow to be processed as if it came in on the real device with
8629 * the VLAN device's VLAN ID. */
8630 flow->in_port = realdev;
8631 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8636 vsp_remove(struct ofport_dpif *port)
8638 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8639 struct vlan_splinter *vsp;
8641 vsp = vlandev_find(ofproto, port->up.ofp_port);
8643 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8644 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8647 port->realdev_ofp_port = 0;
8649 VLOG_ERR("missing vlan device record");
8654 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8656 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8658 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8659 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8660 == realdev_ofp_port)) {
8661 struct vlan_splinter *vsp;
8663 vsp = xmalloc(sizeof *vsp);
8664 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8665 hash_int(port->up.ofp_port, 0));
8666 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8667 hash_realdev_vid(realdev_ofp_port, vid));
8668 vsp->realdev_ofp_port = realdev_ofp_port;
8669 vsp->vlandev_ofp_port = port->up.ofp_port;
8672 port->realdev_ofp_port = realdev_ofp_port;
8674 VLOG_ERR("duplicate vlan device record");
8679 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8681 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8682 return ofport ? ofport->odp_port : OVSP_NONE;
8685 static struct ofport_dpif *
8686 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8688 struct ofport_dpif *port;
8690 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8691 hash_int(odp_port, 0),
8692 &backer->odp_to_ofport_map) {
8693 if (port->odp_port == odp_port) {
8702 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8704 struct ofport_dpif *port;
8706 port = odp_port_to_ofport(ofproto->backer, odp_port);
8707 if (port && &ofproto->up == port->up.ofproto) {
8708 return port->up.ofp_port;
8713 static unsigned long long int
8714 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8716 unsigned long long int dc;
8717 unsigned long long int avg;
8719 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8720 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8726 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8730 if (ofproto->n_update_stats) {
8731 avg_c = (double)ofproto->total_subfacet_count
8732 / ofproto->n_update_stats;
8739 show_dp_rates(struct ds *ds, const char *heading,
8740 const struct avg_subfacet_rates *rates)
8742 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8743 heading, rates->add_rate, rates->del_rate);
8747 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8749 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8750 hmap_count(&ofproto->subfacets));
8753 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8754 * most heavily weighted element. 'base' designates the rate of decay: after
8755 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8758 exp_mavg(double *avg, int base, double new)
8760 *avg = (*avg * (base - 1) + new) / base;
8764 update_moving_averages(struct ofproto_dpif *ofproto)
8766 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8768 /* Update hourly averages on the minute boundaries. */
8769 if (time_msec() - ofproto->last_minute >= min_ms) {
8770 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8771 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8773 /* Update daily averages on the hour boundaries. */
8774 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8775 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8776 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8779 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8780 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8781 ofproto->subfacet_add_count = 0;
8782 ofproto->subfacet_del_count = 0;
8783 ofproto->last_minute += min_ms;
8788 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8790 ofproto->n_hit += delta;
8793 const struct ofproto_class ofproto_dpif_class = {
8828 port_is_lacp_current,
8829 NULL, /* rule_choose_table */
8836 rule_modify_actions,
8847 get_stp_port_status,
8854 is_mirror_output_bundle,
8855 forward_bpdu_changed,
8856 set_mac_table_config,