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-ipfix.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
89 * - Do include packets and bytes from facets that have been deleted or
90 * whose own statistics have been folded into the rule.
92 * - Do include packets and bytes sent "by hand" that were accounted to
93 * the rule without any facet being involved (this is a rare corner
94 * case in rule_execute()).
96 * - Do not include packet or bytes that can be obtained from any facet's
97 * packet_count or byte_count member or that can be obtained from the
98 * datapath by, e.g., dpif_flow_get() for any subfacet.
100 uint64_t packet_count; /* Number of packets received. */
101 uint64_t byte_count; /* Number of bytes received. */
103 tag_type tag; /* Caches rule_calculate_tag() result. */
105 struct list facets; /* List of "struct facet"s. */
108 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
110 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
113 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
114 const struct flow *);
115 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
118 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
119 const struct flow *flow);
121 static void rule_credit_stats(struct rule_dpif *,
122 const struct dpif_flow_stats *);
123 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
210 struct action_xlate_ctx {
211 /* action_xlate_ctx_init() initializes these members. */
214 struct ofproto_dpif *ofproto;
216 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
217 * this flow when actions change header fields. */
220 /* stack for the push and pop actions.
221 * Each stack element is of the type "union mf_subvalue". */
223 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
225 /* The packet corresponding to 'flow', or a null pointer if we are
226 * revalidating without a packet to refer to. */
227 const struct ofpbuf *packet;
229 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
230 * actions update the flow table?
232 * We want to update these tables if we are actually processing a packet,
233 * or if we are accounting for packets that the datapath has processed, but
234 * not if we are just revalidating. */
237 /* The rule that we are currently translating, or NULL. */
238 struct rule_dpif *rule;
240 /* Union of the set of TCP flags seen so far in this flow. (Used only by
241 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
245 /* If nonnull, flow translation calls this function just before executing a
246 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
247 * when the recursion depth is exceeded.
249 * 'rule' is the rule being submitted into. It will be null if the
250 * resubmit or OFPP_TABLE action didn't find a matching rule.
252 * This is normally null so the client has to set it manually after
253 * calling action_xlate_ctx_init(). */
254 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
256 /* If nonnull, flow translation calls this function to report some
257 * significant decision, e.g. to explain why OFPP_NORMAL translation
258 * dropped a packet. */
259 void (*report_hook)(struct action_xlate_ctx *, const char *s);
261 /* If nonnull, flow translation credits the specified statistics to each
262 * rule reached through a resubmit or OFPP_TABLE action.
264 * This is normally null so the client has to set it manually after
265 * calling action_xlate_ctx_init(). */
266 const struct dpif_flow_stats *resubmit_stats;
268 /* xlate_actions() initializes and uses these members. The client might want
269 * to look at them after it returns. */
271 struct ofpbuf *odp_actions; /* Datapath actions. */
272 tag_type tags; /* Tags associated with actions. */
273 enum slow_path_reason slow; /* 0 if fast path may be used. */
274 bool has_learn; /* Actions include NXAST_LEARN? */
275 bool has_normal; /* Actions output to OFPP_NORMAL? */
276 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
277 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
278 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
280 /* xlate_actions() initializes and uses these members, but the client has no
281 * reason to look at them. */
283 int recurse; /* Recursion level, via xlate_table_action. */
284 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
285 struct flow base_flow; /* Flow at the last commit. */
286 uint32_t orig_skb_priority; /* Priority when packet arrived. */
287 uint8_t table_id; /* OpenFlow table ID where flow was found. */
288 uint32_t sflow_n_outputs; /* Number of output ports. */
289 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
290 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
291 bool exit; /* No further actions should be processed. */
294 /* Initial values of fields of the packet that may be changed during
295 * flow processing and needed later. */
296 struct initial_vals {
297 /* This is the value of vlan_tci in the packet as actually received from
298 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
299 * was received via a VLAN splinter. In that case, this value is 0
300 * (because the packet as actually received from the dpif had no 802.1Q
301 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
304 * This member should be removed when the VLAN splinters feature is no
308 /* If received on a tunnel, the IP TOS value of the tunnel. */
309 uint8_t tunnel_ip_tos;
312 static void action_xlate_ctx_init(struct action_xlate_ctx *,
313 struct ofproto_dpif *, const struct flow *,
314 const struct initial_vals *initial_vals,
316 uint8_t tcp_flags, const struct ofpbuf *);
317 static void xlate_actions(struct action_xlate_ctx *,
318 const struct ofpact *ofpacts, size_t ofpacts_len,
319 struct ofpbuf *odp_actions);
320 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
321 const struct ofpact *ofpacts,
323 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
324 uint8_t table_id, bool may_packet_in);
326 static size_t put_userspace_action(const struct ofproto_dpif *,
327 struct ofpbuf *odp_actions,
329 const union user_action_cookie *,
332 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
333 enum slow_path_reason,
334 uint64_t *stub, size_t stub_size,
335 const struct nlattr **actionsp,
336 size_t *actions_lenp);
338 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
340 /* A subfacet (see "struct subfacet" below) has three possible installation
343 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
344 * case just after the subfacet is created, just before the subfacet is
345 * destroyed, or if the datapath returns an error when we try to install a
348 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
350 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
351 * ofproto_dpif is installed in the datapath.
354 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
355 SF_FAST_PATH, /* Full actions are installed. */
356 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
359 static const char *subfacet_path_to_string(enum subfacet_path);
361 /* A dpif flow and actions associated with a facet.
363 * See also the large comment on struct facet. */
366 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
367 struct list list_node; /* In struct facet's 'facets' list. */
368 struct facet *facet; /* Owning facet. */
370 enum odp_key_fitness key_fitness;
374 long long int used; /* Time last used; time created if not used. */
375 long long int created; /* Time created. */
377 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
378 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
382 * These should be essentially identical for every subfacet in a facet, but
383 * may differ in trivial ways due to VLAN splinters. */
384 size_t actions_len; /* Number of bytes in actions[]. */
385 struct nlattr *actions; /* Datapath actions. */
387 enum slow_path_reason slow; /* 0 if fast path may be used. */
388 enum subfacet_path path; /* Installed in datapath? */
390 /* Initial values of the packet that may be needed later. */
391 struct initial_vals initial_vals;
393 /* Datapath port the packet arrived on. This is needed to remove
394 * flows for ports that are no longer part of the bridge. Since the
395 * flow definition only has the OpenFlow port number and the port is
396 * no longer part of the bridge, we can't determine the datapath port
397 * number needed to delete the flow from the datapath. */
398 uint32_t odp_in_port;
401 #define SUBFACET_DESTROY_MAX_BATCH 50
403 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
405 static struct subfacet *subfacet_find(struct ofproto_dpif *,
406 const struct nlattr *key, size_t key_len,
408 static void subfacet_destroy(struct subfacet *);
409 static void subfacet_destroy__(struct subfacet *);
410 static void subfacet_destroy_batch(struct ofproto_dpif *,
411 struct subfacet **, int n);
412 static void subfacet_reset_dp_stats(struct subfacet *,
413 struct dpif_flow_stats *);
414 static void subfacet_update_time(struct subfacet *, long long int used);
415 static void subfacet_update_stats(struct subfacet *,
416 const struct dpif_flow_stats *);
417 static void subfacet_make_actions(struct subfacet *,
418 const struct ofpbuf *packet,
419 struct ofpbuf *odp_actions);
420 static int subfacet_install(struct subfacet *,
421 const struct nlattr *actions, size_t actions_len,
422 struct dpif_flow_stats *, enum slow_path_reason);
423 static void subfacet_uninstall(struct subfacet *);
425 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
427 /* An exact-match instantiation of an OpenFlow flow.
429 * A facet associates a "struct flow", which represents the Open vSwitch
430 * userspace idea of an exact-match flow, with one or more subfacets. Each
431 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
432 * the facet. When the kernel module (or other dpif implementation) and Open
433 * vSwitch userspace agree on the definition of a flow key, there is exactly
434 * one subfacet per facet. If the dpif implementation supports more-specific
435 * flow matching than userspace, however, a facet can have more than one
436 * subfacet, each of which corresponds to some distinction in flow that
437 * userspace simply doesn't understand.
439 * Flow expiration works in terms of subfacets, so a facet must have at least
440 * one subfacet or it will never expire, leaking memory. */
443 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
444 struct list list_node; /* In owning rule's 'facets' list. */
445 struct rule_dpif *rule; /* Owning rule. */
448 struct list subfacets;
449 long long int used; /* Time last used; time created if not used. */
456 * - Do include packets and bytes sent "by hand", e.g. with
459 * - Do include packets and bytes that were obtained from the datapath
460 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
461 * DPIF_FP_ZERO_STATS).
463 * - Do not include packets or bytes that can be obtained from the
464 * datapath for any existing subfacet.
466 uint64_t packet_count; /* Number of packets received. */
467 uint64_t byte_count; /* Number of bytes received. */
469 /* Resubmit statistics. */
470 uint64_t prev_packet_count; /* Number of packets from last stats push. */
471 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
472 long long int prev_used; /* Used time from last stats push. */
475 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
476 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
477 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
479 /* Properties of datapath actions.
481 * Every subfacet has its own actions because actions can differ slightly
482 * between splintered and non-splintered subfacets due to the VLAN tag
483 * being initially different (present vs. absent). All of them have these
484 * properties in common so we just store one copy of them here. */
485 bool has_learn; /* Actions include NXAST_LEARN? */
486 bool has_normal; /* Actions output to OFPP_NORMAL? */
487 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
488 tag_type tags; /* Tags that would require revalidation. */
489 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
491 /* Storage for a single subfacet, to reduce malloc() time and space
492 * overhead. (A facet always has at least one subfacet and in the common
493 * case has exactly one subfacet. However, 'one_subfacet' may not
494 * always be valid, since it could have been removed after newer
495 * subfacets were pushed onto the 'subfacets' list.) */
496 struct subfacet one_subfacet;
498 long long int learn_rl; /* Rate limiter for facet_learn(). */
501 static struct facet *facet_create(struct rule_dpif *,
502 const struct flow *, uint32_t hash);
503 static void facet_remove(struct facet *);
504 static void facet_free(struct facet *);
506 static struct facet *facet_find(struct ofproto_dpif *,
507 const struct flow *, uint32_t hash);
508 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
509 const struct flow *, uint32_t hash);
510 static void facet_revalidate(struct facet *);
511 static bool facet_check_consistency(struct facet *);
513 static void facet_flush_stats(struct facet *);
515 static void facet_update_time(struct facet *, long long int used);
516 static void facet_reset_counters(struct facet *);
517 static void facet_push_stats(struct facet *);
518 static void facet_learn(struct facet *);
519 static void facet_account(struct facet *);
520 static void push_all_stats(void);
522 static struct subfacet *facet_get_subfacet(struct facet *);
524 static bool facet_is_controller_flow(struct facet *);
527 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
531 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
532 struct list bundle_node; /* In struct ofbundle's "ports" list. */
533 struct cfm *cfm; /* Connectivity Fault Management, if any. */
534 tag_type tag; /* Tag associated with this port. */
535 bool may_enable; /* May be enabled in bonds. */
536 long long int carrier_seq; /* Carrier status changes. */
537 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
540 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
541 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
542 long long int stp_state_entered;
544 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
546 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
548 * This is deprecated. It is only for compatibility with broken device
549 * drivers in old versions of Linux that do not properly support VLANs when
550 * VLAN devices are not used. When broken device drivers are no longer in
551 * widespread use, we will delete these interfaces. */
552 uint16_t realdev_ofp_port;
556 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
557 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
558 * traffic egressing the 'ofport' with that priority should be marked with. */
559 struct priority_to_dscp {
560 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
561 uint32_t priority; /* Priority of this queue (see struct flow). */
563 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
566 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
568 * This is deprecated. It is only for compatibility with broken device drivers
569 * in old versions of Linux that do not properly support VLANs when VLAN
570 * devices are not used. When broken device drivers are no longer in
571 * widespread use, we will delete these interfaces. */
572 struct vlan_splinter {
573 struct hmap_node realdev_vid_node;
574 struct hmap_node vlandev_node;
575 uint16_t realdev_ofp_port;
576 uint16_t vlandev_ofp_port;
580 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
581 uint32_t realdev, ovs_be16 vlan_tci);
582 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
583 static void vsp_remove(struct ofport_dpif *);
584 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
586 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
588 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
591 static struct ofport_dpif *
592 ofport_dpif_cast(const struct ofport *ofport)
594 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
595 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
598 static void port_run(struct ofport_dpif *);
599 static void port_run_fast(struct ofport_dpif *);
600 static void port_wait(struct ofport_dpif *);
601 static int set_cfm(struct ofport *, const struct cfm_settings *);
602 static void ofport_clear_priorities(struct ofport_dpif *);
603 static void run_fast_rl(void);
605 struct dpif_completion {
606 struct list list_node;
607 struct ofoperation *op;
610 /* Extra information about a classifier table.
611 * Currently used just for optimized flow revalidation. */
613 /* If either of these is nonnull, then this table has a form that allows
614 * flows to be tagged to avoid revalidating most flows for the most common
615 * kinds of flow table changes. */
616 struct cls_table *catchall_table; /* Table that wildcards all fields. */
617 struct cls_table *other_table; /* Table with any other wildcard set. */
618 uint32_t basis; /* Keeps each table's tags separate. */
621 /* Reasons that we might need to revalidate every facet, and corresponding
624 * A value of 0 means that there is no need to revalidate.
626 * It would be nice to have some cleaner way to integrate with coverage
627 * counters, but with only a few reasons I guess this is good enough for
629 enum revalidate_reason {
630 REV_RECONFIGURE = 1, /* Switch configuration changed. */
631 REV_STP, /* Spanning tree protocol port status change. */
632 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
633 REV_FLOW_TABLE, /* Flow table changed. */
634 REV_INCONSISTENCY /* Facet self-check failed. */
636 COVERAGE_DEFINE(rev_reconfigure);
637 COVERAGE_DEFINE(rev_stp);
638 COVERAGE_DEFINE(rev_port_toggled);
639 COVERAGE_DEFINE(rev_flow_table);
640 COVERAGE_DEFINE(rev_inconsistency);
642 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
643 * These are datapath flows which have no associated ofproto, if they did we
644 * would use facets. */
646 struct hmap_node hmap_node;
651 /* All datapaths of a given type share a single dpif backer instance. */
656 struct timer next_expiration;
657 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
659 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
661 /* Facet revalidation flags applying to facets which use this backer. */
662 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
663 struct tag_set revalidate_set; /* Revalidate only matching facets. */
665 struct hmap drop_keys; /* Set of dropped odp keys. */
668 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
669 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
671 static void drop_key_clear(struct dpif_backer *);
672 static struct ofport_dpif *
673 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
675 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
677 struct avg_subfacet_rates {
678 double add_rate; /* Moving average of new flows created per minute. */
679 double del_rate; /* Moving average of flows deleted per minute. */
681 static void show_dp_rates(struct ds *ds, const char *heading,
682 const struct avg_subfacet_rates *rates);
683 static void exp_mavg(double *avg, int base, double new);
685 struct ofproto_dpif {
686 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
688 struct dpif_backer *backer;
690 /* Special OpenFlow rules. */
691 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
692 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
698 struct netflow *netflow;
699 struct dpif_sflow *sflow;
700 struct dpif_ipfix *ipfix;
701 struct hmap bundles; /* Contains "struct ofbundle"s. */
702 struct mac_learning *ml;
703 struct ofmirror *mirrors[MAX_MIRRORS];
705 bool has_bonded_bundles;
709 struct hmap subfacets;
710 struct governor *governor;
711 long long int consistency_rl;
714 struct table_dpif tables[N_TABLES];
716 /* Support for debugging async flow mods. */
717 struct list completions;
719 bool has_bundle_action; /* True when the first bundle action appears. */
720 struct netdev_stats stats; /* To account packets generated and consumed in
725 long long int stp_last_tick;
727 /* VLAN splinters. */
728 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
729 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
732 struct sset ports; /* Set of standard port names. */
733 struct sset ghost_ports; /* Ports with no datapath port. */
734 struct sset port_poll_set; /* Queued names for port_poll() reply. */
735 int port_poll_errno; /* Last errno for port_poll() reply. */
737 /* Per ofproto's dpif stats. */
741 /* Subfacet statistics.
743 * These keep track of the total number of subfacets added and deleted and
744 * flow life span. They are useful for computing the flow rates stats
745 * exposed via "ovs-appctl dpif/show". The goal is to learn about
746 * traffic patterns in ways that we can use later to improve Open vSwitch
747 * performance in new situations. */
748 long long int created; /* Time when it is created. */
749 unsigned int max_n_subfacet; /* Maximum number of flows */
751 /* The average number of subfacets... */
752 struct avg_subfacet_rates hourly; /* ...over the last hour. */
753 struct avg_subfacet_rates daily; /* ...over the last day. */
754 long long int last_minute; /* Last time 'hourly' was updated. */
756 /* Number of subfacets added or deleted since 'last_minute'. */
757 unsigned int subfacet_add_count;
758 unsigned int subfacet_del_count;
760 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
761 unsigned long long int total_subfacet_add_count;
762 unsigned long long int total_subfacet_del_count;
764 /* Sum of the number of milliseconds that each subfacet existed,
765 * over the subfacets that have been added and then later deleted. */
766 unsigned long long int total_subfacet_life_span;
768 /* Incremented by the number of currently existing subfacets, each
769 * time we pull statistics from the kernel. */
770 unsigned long long int total_subfacet_count;
772 /* Number of times we pull statistics from the kernel. */
773 unsigned long long int n_update_stats;
775 static unsigned long long int avg_subfacet_life_span(
776 const struct ofproto_dpif *);
777 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
778 static void update_moving_averages(struct ofproto_dpif *ofproto);
779 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
781 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
783 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
784 * for debugging the asynchronous flow_mod implementation.) */
787 /* All existing ofproto_dpif instances, indexed by ->up.name. */
788 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
790 static void ofproto_dpif_unixctl_init(void);
792 static struct ofproto_dpif *
793 ofproto_dpif_cast(const struct ofproto *ofproto)
795 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
796 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
799 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
801 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
803 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
804 const struct ofpbuf *,
805 const struct initial_vals *, struct ds *);
807 /* Packet processing. */
808 static void update_learning_table(struct ofproto_dpif *,
809 const struct flow *, int vlan,
812 #define FLOW_MISS_MAX_BATCH 50
813 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
815 /* Flow expiration. */
816 static int expire(struct dpif_backer *);
819 static void send_netflow_active_timeouts(struct ofproto_dpif *);
822 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
823 static size_t compose_sflow_action(const struct ofproto_dpif *,
824 struct ofpbuf *odp_actions,
825 const struct flow *, uint32_t odp_port);
826 static void compose_ipfix_action(const struct ofproto_dpif *,
827 struct ofpbuf *odp_actions,
828 const struct flow *);
829 static void add_mirror_actions(struct action_xlate_ctx *ctx,
830 const struct flow *flow);
831 /* Global variables. */
832 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
834 /* Initial mappings of port to bridge mappings. */
835 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
837 /* Factory functions. */
840 init(const struct shash *iface_hints)
842 struct shash_node *node;
844 /* Make a local copy, since we don't own 'iface_hints' elements. */
845 SHASH_FOR_EACH(node, iface_hints) {
846 const struct iface_hint *orig_hint = node->data;
847 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
849 new_hint->br_name = xstrdup(orig_hint->br_name);
850 new_hint->br_type = xstrdup(orig_hint->br_type);
851 new_hint->ofp_port = orig_hint->ofp_port;
853 shash_add(&init_ofp_ports, node->name, new_hint);
858 enumerate_types(struct sset *types)
860 dp_enumerate_types(types);
864 enumerate_names(const char *type, struct sset *names)
866 struct ofproto_dpif *ofproto;
869 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
870 if (strcmp(type, ofproto->up.type)) {
873 sset_add(names, ofproto->up.name);
880 del(const char *type, const char *name)
885 error = dpif_open(name, type, &dpif);
887 error = dpif_delete(dpif);
894 port_open_type(const char *datapath_type, const char *port_type)
896 return dpif_port_open_type(datapath_type, port_type);
899 /* Type functions. */
901 static struct ofproto_dpif *
902 lookup_ofproto_dpif_by_port_name(const char *name)
904 struct ofproto_dpif *ofproto;
906 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
907 if (sset_contains(&ofproto->ports, name)) {
916 type_run(const char *type)
918 static long long int push_timer = LLONG_MIN;
919 struct dpif_backer *backer;
923 backer = shash_find_data(&all_dpif_backers, type);
925 /* This is not necessarily a problem, since backers are only
926 * created on demand. */
930 dpif_run(backer->dpif);
932 /* The most natural place to push facet statistics is when they're pulled
933 * from the datapath. However, when there are many flows in the datapath,
934 * this expensive operation can occur so frequently, that it reduces our
935 * ability to quickly set up flows. To reduce the cost, we push statistics
937 if (time_msec() > push_timer) {
938 push_timer = time_msec() + 2000;
942 if (backer->need_revalidate
943 || !tag_set_is_empty(&backer->revalidate_set)) {
944 struct tag_set revalidate_set = backer->revalidate_set;
945 bool need_revalidate = backer->need_revalidate;
946 struct ofproto_dpif *ofproto;
947 struct simap_node *node;
948 struct simap tmp_backers;
950 /* Handle tunnel garbage collection. */
951 simap_init(&tmp_backers);
952 simap_swap(&backer->tnl_backers, &tmp_backers);
954 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
955 struct ofport_dpif *iter;
957 if (backer != ofproto->backer) {
961 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
964 if (!iter->tnl_port) {
968 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
969 node = simap_find(&tmp_backers, dp_port);
971 simap_put(&backer->tnl_backers, dp_port, node->data);
972 simap_delete(&tmp_backers, node);
973 node = simap_find(&backer->tnl_backers, dp_port);
975 node = simap_find(&backer->tnl_backers, dp_port);
977 uint32_t odp_port = UINT32_MAX;
979 if (!dpif_port_add(backer->dpif, iter->up.netdev,
981 simap_put(&backer->tnl_backers, dp_port, odp_port);
982 node = simap_find(&backer->tnl_backers, dp_port);
987 iter->odp_port = node ? node->data : OVSP_NONE;
988 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
990 backer->need_revalidate = REV_RECONFIGURE;
995 SIMAP_FOR_EACH (node, &tmp_backers) {
996 dpif_port_del(backer->dpif, node->data);
998 simap_destroy(&tmp_backers);
1000 switch (backer->need_revalidate) {
1001 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1002 case REV_STP: COVERAGE_INC(rev_stp); break;
1003 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1004 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1005 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1008 if (backer->need_revalidate) {
1009 /* Clear the drop_keys in case we should now be accepting some
1010 * formerly dropped flows. */
1011 drop_key_clear(backer);
1014 /* Clear the revalidation flags. */
1015 tag_set_init(&backer->revalidate_set);
1016 backer->need_revalidate = 0;
1018 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1019 struct facet *facet, *next;
1021 if (ofproto->backer != backer) {
1025 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1027 || tag_set_intersects(&revalidate_set, facet->tags)) {
1028 facet_revalidate(facet);
1035 if (timer_expired(&backer->next_expiration)) {
1036 int delay = expire(backer);
1037 timer_set_duration(&backer->next_expiration, delay);
1040 /* Check for port changes in the dpif. */
1041 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1042 struct ofproto_dpif *ofproto;
1043 struct dpif_port port;
1045 /* Don't report on the datapath's device. */
1046 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1050 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1051 &all_ofproto_dpifs) {
1052 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1057 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1058 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1059 /* The port was removed. If we know the datapath,
1060 * report it through poll_set(). If we don't, it may be
1061 * notifying us of a removal we initiated, so ignore it.
1062 * If there's a pending ENOBUFS, let it stand, since
1063 * everything will be reevaluated. */
1064 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1065 sset_add(&ofproto->port_poll_set, devname);
1066 ofproto->port_poll_errno = 0;
1068 } else if (!ofproto) {
1069 /* The port was added, but we don't know with which
1070 * ofproto we should associate it. Delete it. */
1071 dpif_port_del(backer->dpif, port.port_no);
1073 dpif_port_destroy(&port);
1079 if (error != EAGAIN) {
1080 struct ofproto_dpif *ofproto;
1082 /* There was some sort of error, so propagate it to all
1083 * ofprotos that use this backer. */
1084 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1085 &all_ofproto_dpifs) {
1086 if (ofproto->backer == backer) {
1087 sset_clear(&ofproto->port_poll_set);
1088 ofproto->port_poll_errno = error;
1097 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1101 /* Handle one or more batches of upcalls, until there's nothing left to do
1102 * or until we do a fixed total amount of work.
1104 * We do work in batches because it can be much cheaper to set up a number
1105 * of flows and fire off their patches all at once. We do multiple batches
1106 * because in some cases handling a packet can cause another packet to be
1107 * queued almost immediately as part of the return flow. Both
1108 * optimizations can make major improvements on some benchmarks and
1109 * presumably for real traffic as well. */
1111 while (work < max_batch) {
1112 int retval = handle_upcalls(backer, max_batch - work);
1123 type_run_fast(const char *type)
1125 struct dpif_backer *backer;
1127 backer = shash_find_data(&all_dpif_backers, type);
1129 /* This is not necessarily a problem, since backers are only
1130 * created on demand. */
1134 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1140 static long long int port_rl = LLONG_MIN;
1141 static unsigned int backer_rl = 0;
1143 if (time_msec() >= port_rl) {
1144 struct ofproto_dpif *ofproto;
1145 struct ofport_dpif *ofport;
1147 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1149 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1150 port_run_fast(ofport);
1153 port_rl = time_msec() + 200;
1156 /* XXX: We have to be careful not to do too much work in this function. If
1157 * we call dpif_backer_run_fast() too often, or with too large a batch,
1158 * performance improves signifcantly, but at a cost. It's possible for the
1159 * number of flows in the datapath to increase without bound, and for poll
1160 * loops to take 10s of seconds. The correct solution to this problem,
1161 * long term, is to separate flow miss handling into it's own thread so it
1162 * isn't affected by revalidations, and expirations. Until then, this is
1163 * the best we can do. */
1164 if (++backer_rl >= 10) {
1165 struct shash_node *node;
1168 SHASH_FOR_EACH (node, &all_dpif_backers) {
1169 dpif_backer_run_fast(node->data, 1);
1175 type_wait(const char *type)
1177 struct dpif_backer *backer;
1179 backer = shash_find_data(&all_dpif_backers, type);
1181 /* This is not necessarily a problem, since backers are only
1182 * created on demand. */
1186 timer_wait(&backer->next_expiration);
1189 /* Basic life-cycle. */
1191 static int add_internal_flows(struct ofproto_dpif *);
1193 static struct ofproto *
1196 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1197 return &ofproto->up;
1201 dealloc(struct ofproto *ofproto_)
1203 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1208 close_dpif_backer(struct dpif_backer *backer)
1210 struct shash_node *node;
1212 ovs_assert(backer->refcount > 0);
1214 if (--backer->refcount) {
1218 drop_key_clear(backer);
1219 hmap_destroy(&backer->drop_keys);
1221 simap_destroy(&backer->tnl_backers);
1222 hmap_destroy(&backer->odp_to_ofport_map);
1223 node = shash_find(&all_dpif_backers, backer->type);
1225 shash_delete(&all_dpif_backers, node);
1226 dpif_close(backer->dpif);
1231 /* Datapath port slated for removal from datapath. */
1232 struct odp_garbage {
1233 struct list list_node;
1238 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1240 struct dpif_backer *backer;
1241 struct dpif_port_dump port_dump;
1242 struct dpif_port port;
1243 struct shash_node *node;
1244 struct list garbage_list;
1245 struct odp_garbage *garbage, *next;
1251 backer = shash_find_data(&all_dpif_backers, type);
1258 backer_name = xasprintf("ovs-%s", type);
1260 /* Remove any existing datapaths, since we assume we're the only
1261 * userspace controlling the datapath. */
1263 dp_enumerate_names(type, &names);
1264 SSET_FOR_EACH(name, &names) {
1265 struct dpif *old_dpif;
1267 /* Don't remove our backer if it exists. */
1268 if (!strcmp(name, backer_name)) {
1272 if (dpif_open(name, type, &old_dpif)) {
1273 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1275 dpif_delete(old_dpif);
1276 dpif_close(old_dpif);
1279 sset_destroy(&names);
1281 backer = xmalloc(sizeof *backer);
1283 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1286 VLOG_ERR("failed to open datapath of type %s: %s", type,
1292 backer->type = xstrdup(type);
1293 backer->refcount = 1;
1294 hmap_init(&backer->odp_to_ofport_map);
1295 hmap_init(&backer->drop_keys);
1296 timer_set_duration(&backer->next_expiration, 1000);
1297 backer->need_revalidate = 0;
1298 simap_init(&backer->tnl_backers);
1299 tag_set_init(&backer->revalidate_set);
1302 dpif_flow_flush(backer->dpif);
1304 /* Loop through the ports already on the datapath and remove any
1305 * that we don't need anymore. */
1306 list_init(&garbage_list);
1307 dpif_port_dump_start(&port_dump, backer->dpif);
1308 while (dpif_port_dump_next(&port_dump, &port)) {
1309 node = shash_find(&init_ofp_ports, port.name);
1310 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1311 garbage = xmalloc(sizeof *garbage);
1312 garbage->odp_port = port.port_no;
1313 list_push_front(&garbage_list, &garbage->list_node);
1316 dpif_port_dump_done(&port_dump);
1318 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1319 dpif_port_del(backer->dpif, garbage->odp_port);
1320 list_remove(&garbage->list_node);
1324 shash_add(&all_dpif_backers, type, backer);
1326 error = dpif_recv_set(backer->dpif, true);
1328 VLOG_ERR("failed to listen on datapath of type %s: %s",
1329 type, strerror(error));
1330 close_dpif_backer(backer);
1338 construct(struct ofproto *ofproto_)
1340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1341 struct shash_node *node, *next;
1346 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1351 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1352 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1354 ofproto->n_matches = 0;
1356 ofproto->netflow = NULL;
1357 ofproto->sflow = NULL;
1358 ofproto->ipfix = NULL;
1359 ofproto->stp = NULL;
1360 hmap_init(&ofproto->bundles);
1361 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1362 for (i = 0; i < MAX_MIRRORS; i++) {
1363 ofproto->mirrors[i] = NULL;
1365 ofproto->has_bonded_bundles = false;
1367 hmap_init(&ofproto->facets);
1368 hmap_init(&ofproto->subfacets);
1369 ofproto->governor = NULL;
1370 ofproto->consistency_rl = LLONG_MIN;
1372 for (i = 0; i < N_TABLES; i++) {
1373 struct table_dpif *table = &ofproto->tables[i];
1375 table->catchall_table = NULL;
1376 table->other_table = NULL;
1377 table->basis = random_uint32();
1380 list_init(&ofproto->completions);
1382 ofproto_dpif_unixctl_init();
1384 ofproto->has_mirrors = false;
1385 ofproto->has_bundle_action = false;
1387 hmap_init(&ofproto->vlandev_map);
1388 hmap_init(&ofproto->realdev_vid_map);
1390 sset_init(&ofproto->ports);
1391 sset_init(&ofproto->ghost_ports);
1392 sset_init(&ofproto->port_poll_set);
1393 ofproto->port_poll_errno = 0;
1395 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1396 struct iface_hint *iface_hint = node->data;
1398 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1399 /* Check if the datapath already has this port. */
1400 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1401 sset_add(&ofproto->ports, node->name);
1404 free(iface_hint->br_name);
1405 free(iface_hint->br_type);
1407 shash_delete(&init_ofp_ports, node);
1411 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1412 hash_string(ofproto->up.name, 0));
1413 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1415 ofproto_init_tables(ofproto_, N_TABLES);
1416 error = add_internal_flows(ofproto);
1417 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1420 ofproto->n_missed = 0;
1422 ofproto->max_n_subfacet = 0;
1423 ofproto->created = time_msec();
1424 ofproto->last_minute = ofproto->created;
1425 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1426 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1427 ofproto->subfacet_add_count = 0;
1428 ofproto->subfacet_del_count = 0;
1429 ofproto->total_subfacet_add_count = 0;
1430 ofproto->total_subfacet_del_count = 0;
1431 ofproto->total_subfacet_life_span = 0;
1432 ofproto->total_subfacet_count = 0;
1433 ofproto->n_update_stats = 0;
1439 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1440 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1442 struct ofputil_flow_mod fm;
1445 match_init_catchall(&fm.match);
1447 match_set_reg(&fm.match, 0, id);
1448 fm.new_cookie = htonll(0);
1449 fm.cookie = htonll(0);
1450 fm.cookie_mask = htonll(0);
1451 fm.table_id = TBL_INTERNAL;
1452 fm.command = OFPFC_ADD;
1453 fm.idle_timeout = 0;
1454 fm.hard_timeout = 0;
1458 fm.ofpacts = ofpacts->data;
1459 fm.ofpacts_len = ofpacts->size;
1461 error = ofproto_flow_mod(&ofproto->up, &fm);
1463 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1464 id, ofperr_to_string(error));
1468 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1469 ovs_assert(*rulep != NULL);
1475 add_internal_flows(struct ofproto_dpif *ofproto)
1477 struct ofpact_controller *controller;
1478 uint64_t ofpacts_stub[128 / 8];
1479 struct ofpbuf ofpacts;
1483 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1486 controller = ofpact_put_CONTROLLER(&ofpacts);
1487 controller->max_len = UINT16_MAX;
1488 controller->controller_id = 0;
1489 controller->reason = OFPR_NO_MATCH;
1490 ofpact_pad(&ofpacts);
1492 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1497 ofpbuf_clear(&ofpacts);
1498 error = add_internal_flow(ofproto, id++, &ofpacts,
1499 &ofproto->no_packet_in_rule);
1504 complete_operations(struct ofproto_dpif *ofproto)
1506 struct dpif_completion *c, *next;
1508 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1509 ofoperation_complete(c->op, 0);
1510 list_remove(&c->list_node);
1516 destruct(struct ofproto *ofproto_)
1518 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1519 struct rule_dpif *rule, *next_rule;
1520 struct oftable *table;
1523 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1524 complete_operations(ofproto);
1526 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1527 struct cls_cursor cursor;
1529 cls_cursor_init(&cursor, &table->cls, NULL);
1530 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1531 ofproto_rule_destroy(&rule->up);
1535 for (i = 0; i < MAX_MIRRORS; i++) {
1536 mirror_destroy(ofproto->mirrors[i]);
1539 netflow_destroy(ofproto->netflow);
1540 dpif_sflow_destroy(ofproto->sflow);
1541 hmap_destroy(&ofproto->bundles);
1542 mac_learning_destroy(ofproto->ml);
1544 hmap_destroy(&ofproto->facets);
1545 hmap_destroy(&ofproto->subfacets);
1546 governor_destroy(ofproto->governor);
1548 hmap_destroy(&ofproto->vlandev_map);
1549 hmap_destroy(&ofproto->realdev_vid_map);
1551 sset_destroy(&ofproto->ports);
1552 sset_destroy(&ofproto->ghost_ports);
1553 sset_destroy(&ofproto->port_poll_set);
1555 close_dpif_backer(ofproto->backer);
1559 run_fast(struct ofproto *ofproto_)
1561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562 struct ofport_dpif *ofport;
1564 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1565 port_run_fast(ofport);
1572 run(struct ofproto *ofproto_)
1574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1575 struct ofport_dpif *ofport;
1576 struct ofbundle *bundle;
1580 complete_operations(ofproto);
1583 error = run_fast(ofproto_);
1588 if (ofproto->netflow) {
1589 if (netflow_run(ofproto->netflow)) {
1590 send_netflow_active_timeouts(ofproto);
1593 if (ofproto->sflow) {
1594 dpif_sflow_run(ofproto->sflow);
1597 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1600 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1605 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1607 /* Check the consistency of a random facet, to aid debugging. */
1608 if (time_msec() >= ofproto->consistency_rl
1609 && !hmap_is_empty(&ofproto->facets)
1610 && !ofproto->backer->need_revalidate) {
1611 struct facet *facet;
1613 ofproto->consistency_rl = time_msec() + 250;
1615 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1616 struct facet, hmap_node);
1617 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1619 if (!facet_check_consistency(facet)) {
1620 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1625 if (ofproto->governor) {
1628 governor_run(ofproto->governor);
1630 /* If the governor has shrunk to its minimum size and the number of
1631 * subfacets has dwindled, then drop the governor entirely.
1633 * For hysteresis, the number of subfacets to drop the governor is
1634 * smaller than the number needed to trigger its creation. */
1635 n_subfacets = hmap_count(&ofproto->subfacets);
1636 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1637 && governor_is_idle(ofproto->governor)) {
1638 governor_destroy(ofproto->governor);
1639 ofproto->governor = NULL;
1647 wait(struct ofproto *ofproto_)
1649 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1650 struct ofport_dpif *ofport;
1651 struct ofbundle *bundle;
1653 if (!clogged && !list_is_empty(&ofproto->completions)) {
1654 poll_immediate_wake();
1657 dpif_wait(ofproto->backer->dpif);
1658 dpif_recv_wait(ofproto->backer->dpif);
1659 if (ofproto->sflow) {
1660 dpif_sflow_wait(ofproto->sflow);
1662 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1663 poll_immediate_wake();
1665 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1668 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1669 bundle_wait(bundle);
1671 if (ofproto->netflow) {
1672 netflow_wait(ofproto->netflow);
1674 mac_learning_wait(ofproto->ml);
1676 if (ofproto->backer->need_revalidate) {
1677 /* Shouldn't happen, but if it does just go around again. */
1678 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1679 poll_immediate_wake();
1681 if (ofproto->governor) {
1682 governor_wait(ofproto->governor);
1687 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1689 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1691 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1692 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1696 flush(struct ofproto *ofproto_)
1698 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699 struct subfacet *subfacet, *next_subfacet;
1700 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1704 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1705 &ofproto->subfacets) {
1706 if (subfacet->path != SF_NOT_INSTALLED) {
1707 batch[n_batch++] = subfacet;
1708 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1709 subfacet_destroy_batch(ofproto, batch, n_batch);
1713 subfacet_destroy(subfacet);
1718 subfacet_destroy_batch(ofproto, batch, n_batch);
1723 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1724 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1726 *arp_match_ip = true;
1727 *actions = (OFPUTIL_A_OUTPUT |
1728 OFPUTIL_A_SET_VLAN_VID |
1729 OFPUTIL_A_SET_VLAN_PCP |
1730 OFPUTIL_A_STRIP_VLAN |
1731 OFPUTIL_A_SET_DL_SRC |
1732 OFPUTIL_A_SET_DL_DST |
1733 OFPUTIL_A_SET_NW_SRC |
1734 OFPUTIL_A_SET_NW_DST |
1735 OFPUTIL_A_SET_NW_TOS |
1736 OFPUTIL_A_SET_TP_SRC |
1737 OFPUTIL_A_SET_TP_DST |
1742 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1745 struct dpif_dp_stats s;
1747 strcpy(ots->name, "classifier");
1749 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1751 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1752 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1755 static struct ofport *
1758 struct ofport_dpif *port = xmalloc(sizeof *port);
1763 port_dealloc(struct ofport *port_)
1765 struct ofport_dpif *port = ofport_dpif_cast(port_);
1770 port_construct(struct ofport *port_)
1772 struct ofport_dpif *port = ofport_dpif_cast(port_);
1773 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1774 const struct netdev *netdev = port->up.netdev;
1775 struct dpif_port dpif_port;
1778 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1779 port->bundle = NULL;
1781 port->tag = tag_create_random();
1782 port->may_enable = true;
1783 port->stp_port = NULL;
1784 port->stp_state = STP_DISABLED;
1785 port->tnl_port = NULL;
1786 hmap_init(&port->priorities);
1787 port->realdev_ofp_port = 0;
1788 port->vlandev_vid = 0;
1789 port->carrier_seq = netdev_get_carrier_resets(netdev);
1791 if (netdev_vport_is_patch(netdev)) {
1792 /* By bailing out here, we don't submit the port to the sFlow module
1793 * to be considered for counter polling export. This is correct
1794 * because the patch port represents an interface that sFlow considers
1795 * to be "internal" to the switch as a whole, and therefore not an
1796 * candidate for counter polling. */
1797 port->odp_port = OVSP_NONE;
1801 error = dpif_port_query_by_name(ofproto->backer->dpif,
1802 netdev_vport_get_dpif_port(netdev),
1808 port->odp_port = dpif_port.port_no;
1810 if (netdev_get_tunnel_config(netdev)) {
1811 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1813 /* Sanity-check that a mapping doesn't already exist. This
1814 * shouldn't happen for non-tunnel ports. */
1815 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1816 VLOG_ERR("port %s already has an OpenFlow port number",
1818 dpif_port_destroy(&dpif_port);
1822 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1823 hash_int(port->odp_port, 0));
1825 dpif_port_destroy(&dpif_port);
1827 if (ofproto->sflow) {
1828 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1835 port_destruct(struct ofport *port_)
1837 struct ofport_dpif *port = ofport_dpif_cast(port_);
1838 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1839 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1840 const char *devname = netdev_get_name(port->up.netdev);
1842 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1843 /* The underlying device is still there, so delete it. This
1844 * happens when the ofproto is being destroyed, since the caller
1845 * assumes that removal of attached ports will happen as part of
1847 if (!port->tnl_port) {
1848 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1850 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1854 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1857 tnl_port_del(port->tnl_port);
1858 sset_find_and_delete(&ofproto->ports, devname);
1859 sset_find_and_delete(&ofproto->ghost_ports, devname);
1860 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861 bundle_remove(port_);
1862 set_cfm(port_, NULL);
1863 if (ofproto->sflow) {
1864 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1867 ofport_clear_priorities(port);
1868 hmap_destroy(&port->priorities);
1872 port_modified(struct ofport *port_)
1874 struct ofport_dpif *port = ofport_dpif_cast(port_);
1876 if (port->bundle && port->bundle->bond) {
1877 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1882 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1884 struct ofport_dpif *port = ofport_dpif_cast(port_);
1885 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1886 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1888 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1889 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1890 OFPUTIL_PC_NO_PACKET_IN)) {
1891 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1893 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1894 bundle_update(port->bundle);
1900 set_sflow(struct ofproto *ofproto_,
1901 const struct ofproto_sflow_options *sflow_options)
1903 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1904 struct dpif_sflow *ds = ofproto->sflow;
1906 if (sflow_options) {
1908 struct ofport_dpif *ofport;
1910 ds = ofproto->sflow = dpif_sflow_create();
1911 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1912 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1914 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1916 dpif_sflow_set_options(ds, sflow_options);
1919 dpif_sflow_destroy(ds);
1920 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1921 ofproto->sflow = NULL;
1929 struct ofproto *ofproto_,
1930 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1931 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1932 size_t n_flow_exporters_options)
1934 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1935 struct dpif_ipfix *di = ofproto->ipfix;
1937 if (bridge_exporter_options || flow_exporters_options) {
1939 di = ofproto->ipfix = dpif_ipfix_create();
1941 dpif_ipfix_set_options(
1942 di, bridge_exporter_options, flow_exporters_options,
1943 n_flow_exporters_options);
1946 dpif_ipfix_destroy(di);
1947 ofproto->ipfix = NULL;
1954 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1956 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1963 struct ofproto_dpif *ofproto;
1965 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1966 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1967 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1970 if (cfm_configure(ofport->cfm, s)) {
1976 cfm_destroy(ofport->cfm);
1982 get_cfm_status(const struct ofport *ofport_,
1983 struct ofproto_cfm_status *status)
1985 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1988 status->faults = cfm_get_fault(ofport->cfm);
1989 status->remote_opstate = cfm_get_opup(ofport->cfm);
1990 status->health = cfm_get_health(ofport->cfm);
1991 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1998 /* Spanning Tree. */
2001 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2003 struct ofproto_dpif *ofproto = ofproto_;
2004 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2005 struct ofport_dpif *ofport;
2007 ofport = stp_port_get_aux(sp);
2009 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2010 ofproto->up.name, port_num);
2012 struct eth_header *eth = pkt->l2;
2014 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2015 if (eth_addr_is_zero(eth->eth_src)) {
2016 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2017 "with unknown MAC", ofproto->up.name, port_num);
2019 send_packet(ofport, pkt);
2025 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2027 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2031 /* Only revalidate flows if the configuration changed. */
2032 if (!s != !ofproto->stp) {
2033 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2037 if (!ofproto->stp) {
2038 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2039 send_bpdu_cb, ofproto);
2040 ofproto->stp_last_tick = time_msec();
2043 stp_set_bridge_id(ofproto->stp, s->system_id);
2044 stp_set_bridge_priority(ofproto->stp, s->priority);
2045 stp_set_hello_time(ofproto->stp, s->hello_time);
2046 stp_set_max_age(ofproto->stp, s->max_age);
2047 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2049 struct ofport *ofport;
2051 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2052 set_stp_port(ofport, NULL);
2055 stp_destroy(ofproto->stp);
2056 ofproto->stp = NULL;
2063 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2069 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2070 s->designated_root = stp_get_designated_root(ofproto->stp);
2071 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2080 update_stp_port_state(struct ofport_dpif *ofport)
2082 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2083 enum stp_state state;
2085 /* Figure out new state. */
2086 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2090 if (ofport->stp_state != state) {
2091 enum ofputil_port_state of_state;
2094 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2095 netdev_get_name(ofport->up.netdev),
2096 stp_state_name(ofport->stp_state),
2097 stp_state_name(state));
2098 if (stp_learn_in_state(ofport->stp_state)
2099 != stp_learn_in_state(state)) {
2100 /* xxx Learning action flows should also be flushed. */
2101 mac_learning_flush(ofproto->ml,
2102 &ofproto->backer->revalidate_set);
2104 fwd_change = stp_forward_in_state(ofport->stp_state)
2105 != stp_forward_in_state(state);
2107 ofproto->backer->need_revalidate = REV_STP;
2108 ofport->stp_state = state;
2109 ofport->stp_state_entered = time_msec();
2111 if (fwd_change && ofport->bundle) {
2112 bundle_update(ofport->bundle);
2115 /* Update the STP state bits in the OpenFlow port description. */
2116 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2117 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2118 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2119 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2120 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2122 ofproto_port_set_state(&ofport->up, of_state);
2126 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2127 * caller is responsible for assigning STP port numbers and ensuring
2128 * there are no duplicates. */
2130 set_stp_port(struct ofport *ofport_,
2131 const struct ofproto_port_stp_settings *s)
2133 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2135 struct stp_port *sp = ofport->stp_port;
2137 if (!s || !s->enable) {
2139 ofport->stp_port = NULL;
2140 stp_port_disable(sp);
2141 update_stp_port_state(ofport);
2144 } else if (sp && stp_port_no(sp) != s->port_num
2145 && ofport == stp_port_get_aux(sp)) {
2146 /* The port-id changed, so disable the old one if it's not
2147 * already in use by another port. */
2148 stp_port_disable(sp);
2151 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2152 stp_port_enable(sp);
2154 stp_port_set_aux(sp, ofport);
2155 stp_port_set_priority(sp, s->priority);
2156 stp_port_set_path_cost(sp, s->path_cost);
2158 update_stp_port_state(ofport);
2164 get_stp_port_status(struct ofport *ofport_,
2165 struct ofproto_port_stp_status *s)
2167 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2169 struct stp_port *sp = ofport->stp_port;
2171 if (!ofproto->stp || !sp) {
2177 s->port_id = stp_port_get_id(sp);
2178 s->state = stp_port_get_state(sp);
2179 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2180 s->role = stp_port_get_role(sp);
2181 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2187 stp_run(struct ofproto_dpif *ofproto)
2190 long long int now = time_msec();
2191 long long int elapsed = now - ofproto->stp_last_tick;
2192 struct stp_port *sp;
2195 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2196 ofproto->stp_last_tick = now;
2198 while (stp_get_changed_port(ofproto->stp, &sp)) {
2199 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2202 update_stp_port_state(ofport);
2206 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2207 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2213 stp_wait(struct ofproto_dpif *ofproto)
2216 poll_timer_wait(1000);
2220 /* Returns true if STP should process 'flow'. */
2222 stp_should_process_flow(const struct flow *flow)
2224 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2228 stp_process_packet(const struct ofport_dpif *ofport,
2229 const struct ofpbuf *packet)
2231 struct ofpbuf payload = *packet;
2232 struct eth_header *eth = payload.data;
2233 struct stp_port *sp = ofport->stp_port;
2235 /* Sink packets on ports that have STP disabled when the bridge has
2237 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2241 /* Trim off padding on payload. */
2242 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2243 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2246 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2247 stp_received_bpdu(sp, payload.data, payload.size);
2251 static struct priority_to_dscp *
2252 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2254 struct priority_to_dscp *pdscp;
2257 hash = hash_int(priority, 0);
2258 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2259 if (pdscp->priority == priority) {
2267 ofport_clear_priorities(struct ofport_dpif *ofport)
2269 struct priority_to_dscp *pdscp, *next;
2271 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2272 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2278 set_queues(struct ofport *ofport_,
2279 const struct ofproto_port_queue *qdscp_list,
2282 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2283 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2284 struct hmap new = HMAP_INITIALIZER(&new);
2287 for (i = 0; i < n_qdscp; i++) {
2288 struct priority_to_dscp *pdscp;
2292 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2293 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2298 pdscp = get_priority(ofport, priority);
2300 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2302 pdscp = xmalloc(sizeof *pdscp);
2303 pdscp->priority = priority;
2305 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2308 if (pdscp->dscp != dscp) {
2310 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2313 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2316 if (!hmap_is_empty(&ofport->priorities)) {
2317 ofport_clear_priorities(ofport);
2318 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2321 hmap_swap(&new, &ofport->priorities);
2329 /* Expires all MAC learning entries associated with 'bundle' and forces its
2330 * ofproto to revalidate every flow.
2332 * Normally MAC learning entries are removed only from the ofproto associated
2333 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2334 * are removed from every ofproto. When patch ports and SLB bonds are in use
2335 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2336 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2337 * with the host from which it migrated. */
2339 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2341 struct ofproto_dpif *ofproto = bundle->ofproto;
2342 struct mac_learning *ml = ofproto->ml;
2343 struct mac_entry *mac, *next_mac;
2345 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2346 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2347 if (mac->port.p == bundle) {
2349 struct ofproto_dpif *o;
2351 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2353 struct mac_entry *e;
2355 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2358 mac_learning_expire(o->ml, e);
2364 mac_learning_expire(ml, mac);
2369 static struct ofbundle *
2370 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2372 struct ofbundle *bundle;
2374 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2375 &ofproto->bundles) {
2376 if (bundle->aux == aux) {
2383 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2384 * ones that are found to 'bundles'. */
2386 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2387 void **auxes, size_t n_auxes,
2388 struct hmapx *bundles)
2392 hmapx_init(bundles);
2393 for (i = 0; i < n_auxes; i++) {
2394 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2396 hmapx_add(bundles, bundle);
2402 bundle_update(struct ofbundle *bundle)
2404 struct ofport_dpif *port;
2406 bundle->floodable = true;
2407 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2408 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2409 || !stp_forward_in_state(port->stp_state)) {
2410 bundle->floodable = false;
2417 bundle_del_port(struct ofport_dpif *port)
2419 struct ofbundle *bundle = port->bundle;
2421 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2423 list_remove(&port->bundle_node);
2424 port->bundle = NULL;
2427 lacp_slave_unregister(bundle->lacp, port);
2430 bond_slave_unregister(bundle->bond, port);
2433 bundle_update(bundle);
2437 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2438 struct lacp_slave_settings *lacp)
2440 struct ofport_dpif *port;
2442 port = get_ofp_port(bundle->ofproto, ofp_port);
2447 if (port->bundle != bundle) {
2448 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2450 bundle_del_port(port);
2453 port->bundle = bundle;
2454 list_push_back(&bundle->ports, &port->bundle_node);
2455 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2456 || !stp_forward_in_state(port->stp_state)) {
2457 bundle->floodable = false;
2461 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2462 lacp_slave_register(bundle->lacp, port, lacp);
2469 bundle_destroy(struct ofbundle *bundle)
2471 struct ofproto_dpif *ofproto;
2472 struct ofport_dpif *port, *next_port;
2479 ofproto = bundle->ofproto;
2480 for (i = 0; i < MAX_MIRRORS; i++) {
2481 struct ofmirror *m = ofproto->mirrors[i];
2483 if (m->out == bundle) {
2485 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2486 || hmapx_find_and_delete(&m->dsts, bundle)) {
2487 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2492 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2493 bundle_del_port(port);
2496 bundle_flush_macs(bundle, true);
2497 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2499 free(bundle->trunks);
2500 lacp_destroy(bundle->lacp);
2501 bond_destroy(bundle->bond);
2506 bundle_set(struct ofproto *ofproto_, void *aux,
2507 const struct ofproto_bundle_settings *s)
2509 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2510 bool need_flush = false;
2511 struct ofport_dpif *port;
2512 struct ofbundle *bundle;
2513 unsigned long *trunks;
2519 bundle_destroy(bundle_lookup(ofproto, aux));
2523 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2524 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2526 bundle = bundle_lookup(ofproto, aux);
2528 bundle = xmalloc(sizeof *bundle);
2530 bundle->ofproto = ofproto;
2531 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2532 hash_pointer(aux, 0));
2534 bundle->name = NULL;
2536 list_init(&bundle->ports);
2537 bundle->vlan_mode = PORT_VLAN_TRUNK;
2539 bundle->trunks = NULL;
2540 bundle->use_priority_tags = s->use_priority_tags;
2541 bundle->lacp = NULL;
2542 bundle->bond = NULL;
2544 bundle->floodable = true;
2546 bundle->src_mirrors = 0;
2547 bundle->dst_mirrors = 0;
2548 bundle->mirror_out = 0;
2551 if (!bundle->name || strcmp(s->name, bundle->name)) {
2553 bundle->name = xstrdup(s->name);
2558 if (!bundle->lacp) {
2559 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2560 bundle->lacp = lacp_create();
2562 lacp_configure(bundle->lacp, s->lacp);
2564 lacp_destroy(bundle->lacp);
2565 bundle->lacp = NULL;
2568 /* Update set of ports. */
2570 for (i = 0; i < s->n_slaves; i++) {
2571 if (!bundle_add_port(bundle, s->slaves[i],
2572 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2576 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2577 struct ofport_dpif *next_port;
2579 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2580 for (i = 0; i < s->n_slaves; i++) {
2581 if (s->slaves[i] == port->up.ofp_port) {
2586 bundle_del_port(port);
2590 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2592 if (list_is_empty(&bundle->ports)) {
2593 bundle_destroy(bundle);
2597 /* Set VLAN tagging mode */
2598 if (s->vlan_mode != bundle->vlan_mode
2599 || s->use_priority_tags != bundle->use_priority_tags) {
2600 bundle->vlan_mode = s->vlan_mode;
2601 bundle->use_priority_tags = s->use_priority_tags;
2606 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2607 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2609 if (vlan != bundle->vlan) {
2610 bundle->vlan = vlan;
2614 /* Get trunked VLANs. */
2615 switch (s->vlan_mode) {
2616 case PORT_VLAN_ACCESS:
2620 case PORT_VLAN_TRUNK:
2621 trunks = CONST_CAST(unsigned long *, s->trunks);
2624 case PORT_VLAN_NATIVE_UNTAGGED:
2625 case PORT_VLAN_NATIVE_TAGGED:
2626 if (vlan != 0 && (!s->trunks
2627 || !bitmap_is_set(s->trunks, vlan)
2628 || bitmap_is_set(s->trunks, 0))) {
2629 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2631 trunks = bitmap_clone(s->trunks, 4096);
2633 trunks = bitmap_allocate1(4096);
2635 bitmap_set1(trunks, vlan);
2636 bitmap_set0(trunks, 0);
2638 trunks = CONST_CAST(unsigned long *, s->trunks);
2645 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2646 free(bundle->trunks);
2647 if (trunks == s->trunks) {
2648 bundle->trunks = vlan_bitmap_clone(trunks);
2650 bundle->trunks = trunks;
2655 if (trunks != s->trunks) {
2660 if (!list_is_short(&bundle->ports)) {
2661 bundle->ofproto->has_bonded_bundles = true;
2663 if (bond_reconfigure(bundle->bond, s->bond)) {
2664 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2667 bundle->bond = bond_create(s->bond);
2668 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2671 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2672 bond_slave_register(bundle->bond, port, port->up.netdev);
2675 bond_destroy(bundle->bond);
2676 bundle->bond = NULL;
2679 /* If we changed something that would affect MAC learning, un-learn
2680 * everything on this port and force flow revalidation. */
2682 bundle_flush_macs(bundle, false);
2689 bundle_remove(struct ofport *port_)
2691 struct ofport_dpif *port = ofport_dpif_cast(port_);
2692 struct ofbundle *bundle = port->bundle;
2695 bundle_del_port(port);
2696 if (list_is_empty(&bundle->ports)) {
2697 bundle_destroy(bundle);
2698 } else if (list_is_short(&bundle->ports)) {
2699 bond_destroy(bundle->bond);
2700 bundle->bond = NULL;
2706 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2708 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2709 struct ofport_dpif *port = port_;
2710 uint8_t ea[ETH_ADDR_LEN];
2713 error = netdev_get_etheraddr(port->up.netdev, ea);
2715 struct ofpbuf packet;
2718 ofpbuf_init(&packet, 0);
2719 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2721 memcpy(packet_pdu, pdu, pdu_size);
2723 send_packet(port, &packet);
2724 ofpbuf_uninit(&packet);
2726 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2727 "%s (%s)", port->bundle->name,
2728 netdev_get_name(port->up.netdev), strerror(error));
2733 bundle_send_learning_packets(struct ofbundle *bundle)
2735 struct ofproto_dpif *ofproto = bundle->ofproto;
2736 int error, n_packets, n_errors;
2737 struct mac_entry *e;
2739 error = n_packets = n_errors = 0;
2740 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2741 if (e->port.p != bundle) {
2742 struct ofpbuf *learning_packet;
2743 struct ofport_dpif *port;
2747 /* The assignment to "port" is unnecessary but makes "grep"ing for
2748 * struct ofport_dpif more effective. */
2749 learning_packet = bond_compose_learning_packet(bundle->bond,
2753 ret = send_packet(port, learning_packet);
2754 ofpbuf_delete(learning_packet);
2764 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2765 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2766 "packets, last error was: %s",
2767 bundle->name, n_errors, n_packets, strerror(error));
2769 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2770 bundle->name, n_packets);
2775 bundle_run(struct ofbundle *bundle)
2778 lacp_run(bundle->lacp, send_pdu_cb);
2781 struct ofport_dpif *port;
2783 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2784 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2787 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2788 lacp_status(bundle->lacp));
2789 if (bond_should_send_learning_packets(bundle->bond)) {
2790 bundle_send_learning_packets(bundle);
2796 bundle_wait(struct ofbundle *bundle)
2799 lacp_wait(bundle->lacp);
2802 bond_wait(bundle->bond);
2809 mirror_scan(struct ofproto_dpif *ofproto)
2813 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2814 if (!ofproto->mirrors[idx]) {
2821 static struct ofmirror *
2822 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2826 for (i = 0; i < MAX_MIRRORS; i++) {
2827 struct ofmirror *mirror = ofproto->mirrors[i];
2828 if (mirror && mirror->aux == aux) {
2836 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2838 mirror_update_dups(struct ofproto_dpif *ofproto)
2842 for (i = 0; i < MAX_MIRRORS; i++) {
2843 struct ofmirror *m = ofproto->mirrors[i];
2846 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2850 for (i = 0; i < MAX_MIRRORS; i++) {
2851 struct ofmirror *m1 = ofproto->mirrors[i];
2858 for (j = i + 1; j < MAX_MIRRORS; j++) {
2859 struct ofmirror *m2 = ofproto->mirrors[j];
2861 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2862 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2863 m2->dup_mirrors |= m1->dup_mirrors;
2870 mirror_set(struct ofproto *ofproto_, void *aux,
2871 const struct ofproto_mirror_settings *s)
2873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2874 mirror_mask_t mirror_bit;
2875 struct ofbundle *bundle;
2876 struct ofmirror *mirror;
2877 struct ofbundle *out;
2878 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2879 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2882 mirror = mirror_lookup(ofproto, aux);
2884 mirror_destroy(mirror);
2890 idx = mirror_scan(ofproto);
2892 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2894 ofproto->up.name, MAX_MIRRORS, s->name);
2898 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2899 mirror->ofproto = ofproto;
2902 mirror->out_vlan = -1;
2903 mirror->name = NULL;
2906 if (!mirror->name || strcmp(s->name, mirror->name)) {
2908 mirror->name = xstrdup(s->name);
2911 /* Get the new configuration. */
2912 if (s->out_bundle) {
2913 out = bundle_lookup(ofproto, s->out_bundle);
2915 mirror_destroy(mirror);
2921 out_vlan = s->out_vlan;
2923 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2924 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2926 /* If the configuration has not changed, do nothing. */
2927 if (hmapx_equals(&srcs, &mirror->srcs)
2928 && hmapx_equals(&dsts, &mirror->dsts)
2929 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2930 && mirror->out == out
2931 && mirror->out_vlan == out_vlan)
2933 hmapx_destroy(&srcs);
2934 hmapx_destroy(&dsts);
2938 hmapx_swap(&srcs, &mirror->srcs);
2939 hmapx_destroy(&srcs);
2941 hmapx_swap(&dsts, &mirror->dsts);
2942 hmapx_destroy(&dsts);
2944 free(mirror->vlans);
2945 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2948 mirror->out_vlan = out_vlan;
2950 /* Update bundles. */
2951 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2952 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2953 if (hmapx_contains(&mirror->srcs, bundle)) {
2954 bundle->src_mirrors |= mirror_bit;
2956 bundle->src_mirrors &= ~mirror_bit;
2959 if (hmapx_contains(&mirror->dsts, bundle)) {
2960 bundle->dst_mirrors |= mirror_bit;
2962 bundle->dst_mirrors &= ~mirror_bit;
2965 if (mirror->out == bundle) {
2966 bundle->mirror_out |= mirror_bit;
2968 bundle->mirror_out &= ~mirror_bit;
2972 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2973 ofproto->has_mirrors = true;
2974 mac_learning_flush(ofproto->ml,
2975 &ofproto->backer->revalidate_set);
2976 mirror_update_dups(ofproto);
2982 mirror_destroy(struct ofmirror *mirror)
2984 struct ofproto_dpif *ofproto;
2985 mirror_mask_t mirror_bit;
2986 struct ofbundle *bundle;
2993 ofproto = mirror->ofproto;
2994 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2995 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2997 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2998 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2999 bundle->src_mirrors &= ~mirror_bit;
3000 bundle->dst_mirrors &= ~mirror_bit;
3001 bundle->mirror_out &= ~mirror_bit;
3004 hmapx_destroy(&mirror->srcs);
3005 hmapx_destroy(&mirror->dsts);
3006 free(mirror->vlans);
3008 ofproto->mirrors[mirror->idx] = NULL;
3012 mirror_update_dups(ofproto);
3014 ofproto->has_mirrors = false;
3015 for (i = 0; i < MAX_MIRRORS; i++) {
3016 if (ofproto->mirrors[i]) {
3017 ofproto->has_mirrors = true;
3024 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3025 uint64_t *packets, uint64_t *bytes)
3027 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3028 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3031 *packets = *bytes = UINT64_MAX;
3037 *packets = mirror->packet_count;
3038 *bytes = mirror->byte_count;
3044 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3047 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3048 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3054 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3056 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3057 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3058 return bundle && bundle->mirror_out != 0;
3062 forward_bpdu_changed(struct ofproto *ofproto_)
3064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3065 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3069 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3072 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3073 mac_learning_set_idle_time(ofproto->ml, idle_time);
3074 mac_learning_set_max_entries(ofproto->ml, max_entries);
3079 static struct ofport_dpif *
3080 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3082 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3083 return ofport ? ofport_dpif_cast(ofport) : NULL;
3086 static struct ofport_dpif *
3087 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3089 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3090 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3094 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3095 struct ofproto_port *ofproto_port,
3096 struct dpif_port *dpif_port)
3098 ofproto_port->name = dpif_port->name;
3099 ofproto_port->type = dpif_port->type;
3100 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3103 static struct ofport_dpif *
3104 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3106 const struct ofproto_dpif *ofproto;
3109 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3114 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3115 struct ofport *ofport;
3117 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3118 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3119 return ofport_dpif_cast(ofport);
3126 port_run_fast(struct ofport_dpif *ofport)
3128 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3129 struct ofpbuf packet;
3131 ofpbuf_init(&packet, 0);
3132 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3133 send_packet(ofport, &packet);
3134 ofpbuf_uninit(&packet);
3139 port_run(struct ofport_dpif *ofport)
3141 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3142 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3143 bool enable = netdev_get_carrier(ofport->up.netdev);
3145 ofport->carrier_seq = carrier_seq;
3147 port_run_fast(ofport);
3149 if (ofport->tnl_port
3150 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3151 &ofport->tnl_port)) {
3152 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3156 int cfm_opup = cfm_get_opup(ofport->cfm);
3158 cfm_run(ofport->cfm);
3159 enable = enable && !cfm_get_fault(ofport->cfm);
3161 if (cfm_opup >= 0) {
3162 enable = enable && cfm_opup;
3166 if (ofport->bundle) {
3167 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3168 if (carrier_changed) {
3169 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3173 if (ofport->may_enable != enable) {
3174 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3176 if (ofproto->has_bundle_action) {
3177 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3181 ofport->may_enable = enable;
3185 port_wait(struct ofport_dpif *ofport)
3188 cfm_wait(ofport->cfm);
3193 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3194 struct ofproto_port *ofproto_port)
3196 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3197 struct dpif_port dpif_port;
3200 if (sset_contains(&ofproto->ghost_ports, devname)) {
3201 const char *type = netdev_get_type_from_name(devname);
3203 /* We may be called before ofproto->up.port_by_name is populated with
3204 * the appropriate ofport. For this reason, we must get the name and
3205 * type from the netdev layer directly. */
3207 const struct ofport *ofport;
3209 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3210 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3211 ofproto_port->name = xstrdup(devname);
3212 ofproto_port->type = xstrdup(type);
3218 if (!sset_contains(&ofproto->ports, devname)) {
3221 error = dpif_port_query_by_name(ofproto->backer->dpif,
3222 devname, &dpif_port);
3224 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3230 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3232 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3233 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3234 const char *devname = netdev_get_name(netdev);
3236 if (netdev_vport_is_patch(netdev)) {
3237 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3241 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3242 uint32_t port_no = UINT32_MAX;
3245 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3249 if (netdev_get_tunnel_config(netdev)) {
3250 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3254 if (netdev_get_tunnel_config(netdev)) {
3255 sset_add(&ofproto->ghost_ports, devname);
3257 sset_add(&ofproto->ports, devname);
3263 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3265 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3266 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3273 sset_find_and_delete(&ofproto->ghost_ports,
3274 netdev_get_name(ofport->up.netdev));
3275 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3276 if (!ofport->tnl_port) {
3277 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3279 /* The caller is going to close ofport->up.netdev. If this is a
3280 * bonded port, then the bond is using that netdev, so remove it
3281 * from the bond. The client will need to reconfigure everything
3282 * after deleting ports, so then the slave will get re-added. */
3283 bundle_remove(&ofport->up);
3290 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3292 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3297 error = netdev_get_stats(ofport->up.netdev, stats);
3299 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3302 /* ofproto->stats.tx_packets represents packets that we created
3303 * internally and sent to some port (e.g. packets sent with
3304 * send_packet()). Account for them as if they had come from
3305 * OFPP_LOCAL and got forwarded. */
3307 if (stats->rx_packets != UINT64_MAX) {
3308 stats->rx_packets += ofproto->stats.tx_packets;
3311 if (stats->rx_bytes != UINT64_MAX) {
3312 stats->rx_bytes += ofproto->stats.tx_bytes;
3315 /* ofproto->stats.rx_packets represents packets that were received on
3316 * some port and we processed internally and dropped (e.g. STP).
3317 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3319 if (stats->tx_packets != UINT64_MAX) {
3320 stats->tx_packets += ofproto->stats.rx_packets;
3323 if (stats->tx_bytes != UINT64_MAX) {
3324 stats->tx_bytes += ofproto->stats.rx_bytes;
3331 /* Account packets for LOCAL port. */
3333 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3334 size_t tx_size, size_t rx_size)
3336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3339 ofproto->stats.rx_packets++;
3340 ofproto->stats.rx_bytes += rx_size;
3343 ofproto->stats.tx_packets++;
3344 ofproto->stats.tx_bytes += tx_size;
3348 struct port_dump_state {
3353 struct ofproto_port port;
3358 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3360 *statep = xzalloc(sizeof(struct port_dump_state));
3365 port_dump_next(const struct ofproto *ofproto_, void *state_,
3366 struct ofproto_port *port)
3368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3369 struct port_dump_state *state = state_;
3370 const struct sset *sset;
3371 struct sset_node *node;
3373 if (state->has_port) {
3374 ofproto_port_destroy(&state->port);
3375 state->has_port = false;
3377 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3378 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3381 error = port_query_by_name(ofproto_, node->name, &state->port);
3383 *port = state->port;
3384 state->has_port = true;
3386 } else if (error != ENODEV) {
3391 if (!state->ghost) {
3392 state->ghost = true;
3395 return port_dump_next(ofproto_, state_, port);
3402 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3404 struct port_dump_state *state = state_;
3406 if (state->has_port) {
3407 ofproto_port_destroy(&state->port);
3414 port_poll(const struct ofproto *ofproto_, char **devnamep)
3416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3418 if (ofproto->port_poll_errno) {
3419 int error = ofproto->port_poll_errno;
3420 ofproto->port_poll_errno = 0;
3424 if (sset_is_empty(&ofproto->port_poll_set)) {
3428 *devnamep = sset_pop(&ofproto->port_poll_set);
3433 port_poll_wait(const struct ofproto *ofproto_)
3435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3436 dpif_port_poll_wait(ofproto->backer->dpif);
3440 port_is_lacp_current(const struct ofport *ofport_)
3442 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3443 return (ofport->bundle && ofport->bundle->lacp
3444 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3448 /* Upcall handling. */
3450 /* Flow miss batching.
3452 * Some dpifs implement operations faster when you hand them off in a batch.
3453 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3454 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3455 * more packets, plus possibly installing the flow in the dpif.
3457 * So far we only batch the operations that affect flow setup time the most.
3458 * It's possible to batch more than that, but the benefit might be minimal. */
3460 struct hmap_node hmap_node;
3461 struct ofproto_dpif *ofproto;
3463 enum odp_key_fitness key_fitness;
3464 const struct nlattr *key;
3466 struct initial_vals initial_vals;
3467 struct list packets;
3468 enum dpif_upcall_type upcall_type;
3469 uint32_t odp_in_port;
3472 struct flow_miss_op {
3473 struct dpif_op dpif_op;
3474 void *garbage; /* Pointer to pass to free(), NULL if none. */
3475 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3478 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3479 * OpenFlow controller as necessary according to their individual
3480 * configurations. */
3482 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3483 const struct flow *flow)
3485 struct ofputil_packet_in pin;
3487 pin.packet = packet->data;
3488 pin.packet_len = packet->size;
3489 pin.reason = OFPR_NO_MATCH;
3490 pin.controller_id = 0;
3495 pin.send_len = 0; /* not used for flow table misses */
3497 flow_get_metadata(flow, &pin.fmd);
3499 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3502 static enum slow_path_reason
3503 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3504 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3508 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3510 cfm_process_heartbeat(ofport->cfm, packet);
3513 } else if (ofport->bundle && ofport->bundle->lacp
3514 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3516 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3519 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3521 stp_process_packet(ofport, packet);
3529 static struct flow_miss *
3530 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3531 const struct flow *flow, uint32_t hash)
3533 struct flow_miss *miss;
3535 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3536 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3544 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3545 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3546 * 'miss' is associated with a subfacet the caller must also initialize the
3547 * returned op->subfacet, and if anything needs to be freed after processing
3548 * the op, the caller must initialize op->garbage also. */
3550 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3551 struct flow_miss_op *op)
3553 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3554 /* This packet was received on a VLAN splinter port. We
3555 * added a VLAN to the packet to make the packet resemble
3556 * the flow, but the actions were composed assuming that
3557 * the packet contained no VLAN. So, we must remove the
3558 * VLAN header from the packet before trying to execute the
3560 eth_pop_vlan(packet);
3564 op->dpif_op.type = DPIF_OP_EXECUTE;
3565 op->dpif_op.u.execute.key = miss->key;
3566 op->dpif_op.u.execute.key_len = miss->key_len;
3567 op->dpif_op.u.execute.packet = packet;
3570 /* Helper for handle_flow_miss_without_facet() and
3571 * handle_flow_miss_with_facet(). */
3573 handle_flow_miss_common(struct rule_dpif *rule,
3574 struct ofpbuf *packet, const struct flow *flow)
3576 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3578 ofproto->n_matches++;
3580 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3582 * Extra-special case for fail-open mode.
3584 * We are in fail-open mode and the packet matched the fail-open
3585 * rule, but we are connected to a controller too. We should send
3586 * the packet up to the controller in the hope that it will try to
3587 * set up a flow and thereby allow us to exit fail-open.
3589 * See the top-level comment in fail-open.c for more information.
3591 send_packet_in_miss(ofproto, packet, flow);
3595 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3596 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3597 * installing a datapath flow. The answer is usually "yes" (a return value of
3598 * true). However, for short flows the cost of bookkeeping is much higher than
3599 * the benefits, so when the datapath holds a large number of flows we impose
3600 * some heuristics to decide which flows are likely to be worth tracking. */
3602 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3603 struct flow_miss *miss, uint32_t hash)
3605 if (!ofproto->governor) {
3608 n_subfacets = hmap_count(&ofproto->subfacets);
3609 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3613 ofproto->governor = governor_create(ofproto->up.name);
3616 return governor_should_install_flow(ofproto->governor, hash,
3617 list_size(&miss->packets));
3620 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3621 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3622 * increment '*n_ops'. */
3624 handle_flow_miss_without_facet(struct flow_miss *miss,
3625 struct rule_dpif *rule,
3626 struct flow_miss_op *ops, size_t *n_ops)
3628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3629 long long int now = time_msec();
3630 struct action_xlate_ctx ctx;
3631 struct ofpbuf *packet;
3633 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3634 struct flow_miss_op *op = &ops[*n_ops];
3635 struct dpif_flow_stats stats;
3636 struct ofpbuf odp_actions;
3638 COVERAGE_INC(facet_suppress);
3640 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3642 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3643 rule_credit_stats(rule, &stats);
3645 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3646 &miss->initial_vals, rule, 0, packet);
3647 ctx.resubmit_stats = &stats;
3648 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3651 if (odp_actions.size) {
3652 struct dpif_execute *execute = &op->dpif_op.u.execute;
3654 init_flow_miss_execute_op(miss, packet, op);
3655 execute->actions = odp_actions.data;
3656 execute->actions_len = odp_actions.size;
3657 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3661 ofpbuf_uninit(&odp_actions);
3666 /* Handles 'miss', which matches 'facet'. May add any required datapath
3667 * operations to 'ops', incrementing '*n_ops' for each new op.
3669 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3670 * This is really important only for new facets: if we just called time_msec()
3671 * here, then the new subfacet or its packets could look (occasionally) as
3672 * though it was used some time after the facet was used. That can make a
3673 * one-packet flow look like it has a nonzero duration, which looks odd in
3674 * e.g. NetFlow statistics. */
3676 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3678 struct flow_miss_op *ops, size_t *n_ops)
3680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3681 enum subfacet_path want_path;
3682 struct subfacet *subfacet;
3683 struct ofpbuf *packet;
3685 subfacet = subfacet_create(facet, miss, now);
3687 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3688 struct flow_miss_op *op = &ops[*n_ops];
3689 struct dpif_flow_stats stats;
3690 struct ofpbuf odp_actions;
3692 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3694 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3695 if (!subfacet->actions || subfacet->slow) {
3696 subfacet_make_actions(subfacet, packet, &odp_actions);
3699 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3700 subfacet_update_stats(subfacet, &stats);
3702 if (subfacet->actions_len) {
3703 struct dpif_execute *execute = &op->dpif_op.u.execute;
3705 init_flow_miss_execute_op(miss, packet, op);
3706 if (!subfacet->slow) {
3707 execute->actions = subfacet->actions;
3708 execute->actions_len = subfacet->actions_len;
3709 ofpbuf_uninit(&odp_actions);
3711 execute->actions = odp_actions.data;
3712 execute->actions_len = odp_actions.size;
3713 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3718 ofpbuf_uninit(&odp_actions);
3722 want_path = subfacet_want_path(subfacet->slow);
3723 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3724 struct flow_miss_op *op = &ops[(*n_ops)++];
3725 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3727 subfacet->path = want_path;
3730 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3731 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3732 put->key = miss->key;
3733 put->key_len = miss->key_len;
3734 if (want_path == SF_FAST_PATH) {
3735 put->actions = subfacet->actions;
3736 put->actions_len = subfacet->actions_len;
3738 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3739 op->stub, sizeof op->stub,
3740 &put->actions, &put->actions_len);
3746 /* Handles flow miss 'miss'. May add any required datapath operations
3747 * to 'ops', incrementing '*n_ops' for each new op. */
3749 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3752 struct ofproto_dpif *ofproto = miss->ofproto;
3753 struct facet *facet;
3757 /* The caller must ensure that miss->hmap_node.hash contains
3758 * flow_hash(miss->flow, 0). */
3759 hash = miss->hmap_node.hash;
3761 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3763 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3765 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3766 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3770 facet = facet_create(rule, &miss->flow, hash);
3775 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3778 static struct drop_key *
3779 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3782 struct drop_key *drop_key;
3784 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3785 &backer->drop_keys) {
3786 if (drop_key->key_len == key_len
3787 && !memcmp(drop_key->key, key, key_len)) {
3795 drop_key_clear(struct dpif_backer *backer)
3797 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3798 struct drop_key *drop_key, *next;
3800 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3803 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3805 if (error && !VLOG_DROP_WARN(&rl)) {
3806 struct ds ds = DS_EMPTY_INITIALIZER;
3807 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3808 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3813 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3814 free(drop_key->key);
3819 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3820 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3821 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3822 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3823 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3824 * 'packet' ingressed.
3826 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3827 * 'flow''s in_port to OFPP_NONE.
3829 * This function does post-processing on data returned from
3830 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3831 * of the upcall processing logic. In particular, if the extracted in_port is
3832 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3833 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3834 * a VLAN header onto 'packet' (if it is nonnull).
3836 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3837 * to the VLAN TCI with which the packet was really received, that is, the
3838 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3839 * the value returned in flow->vlan_tci only for packets received on
3840 * VLAN splinters.) Also, if received on an IP tunnel, sets
3841 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3843 * Similarly, this function also includes some logic to help with tunnels. It
3844 * may modify 'flow' as necessary to make the tunneling implementation
3845 * transparent to the upcall processing logic.
3847 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3848 * or some other positive errno if there are other problems. */
3850 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3851 const struct nlattr *key, size_t key_len,
3852 struct flow *flow, enum odp_key_fitness *fitnessp,
3853 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3854 struct initial_vals *initial_vals)
3856 const struct ofport_dpif *port;
3857 enum odp_key_fitness fitness;
3860 fitness = odp_flow_key_to_flow(key, key_len, flow);
3861 if (fitness == ODP_FIT_ERROR) {
3867 initial_vals->vlan_tci = flow->vlan_tci;
3868 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3872 *odp_in_port = flow->in_port;
3875 if (tnl_port_should_receive(flow)) {
3876 const struct ofport *ofport = tnl_port_receive(flow);
3878 flow->in_port = OFPP_NONE;
3881 port = ofport_dpif_cast(ofport);
3883 /* We can't reproduce 'key' from 'flow'. */
3884 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3886 /* XXX: Since the tunnel module is not scoped per backer, it's
3887 * theoretically possible that we'll receive an ofport belonging to an
3888 * entirely different datapath. In practice, this can't happen because
3889 * no platforms has two separate datapaths which each support
3891 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3893 port = odp_port_to_ofport(backer, flow->in_port);
3895 flow->in_port = OFPP_NONE;
3899 flow->in_port = port->up.ofp_port;
3900 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3902 /* Make the packet resemble the flow, so that it gets sent to
3903 * an OpenFlow controller properly, so that it looks correct
3904 * for sFlow, and so that flow_extract() will get the correct
3905 * vlan_tci if it is called on 'packet'.
3907 * The allocated space inside 'packet' probably also contains
3908 * 'key', that is, both 'packet' and 'key' are probably part of
3909 * a struct dpif_upcall (see the large comment on that
3910 * structure definition), so pushing data on 'packet' is in
3911 * general not a good idea since it could overwrite 'key' or
3912 * free it as a side effect. However, it's OK in this special
3913 * case because we know that 'packet' is inside a Netlink
3914 * attribute: pushing 4 bytes will just overwrite the 4-byte
3915 * "struct nlattr", which is fine since we don't need that
3916 * header anymore. */
3917 eth_push_vlan(packet, flow->vlan_tci);
3919 /* We can't reproduce 'key' from 'flow'. */
3920 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3926 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3931 *fitnessp = fitness;
3937 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3940 struct dpif_upcall *upcall;
3941 struct flow_miss *miss;
3942 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3943 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3944 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3954 /* Construct the to-do list.
3956 * This just amounts to extracting the flow from each packet and sticking
3957 * the packets that have the same flow in the same "flow_miss" structure so
3958 * that we can process them together. */
3961 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3962 struct flow_miss *miss = &misses[n_misses];
3963 struct flow_miss *existing_miss;
3964 struct ofproto_dpif *ofproto;
3965 uint32_t odp_in_port;
3970 error = ofproto_receive(backer, upcall->packet, upcall->key,
3971 upcall->key_len, &flow, &miss->key_fitness,
3972 &ofproto, &odp_in_port, &miss->initial_vals);
3973 if (error == ENODEV) {
3974 struct drop_key *drop_key;
3976 /* Received packet on port for which we couldn't associate
3977 * an ofproto. This can happen if a port is removed while
3978 * traffic is being received. Print a rate-limited message
3979 * in case it happens frequently. Install a drop flow so
3980 * that future packets of the flow are inexpensively dropped
3982 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3985 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3987 drop_key = xmalloc(sizeof *drop_key);
3988 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3989 drop_key->key_len = upcall->key_len;
3991 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3992 hash_bytes(drop_key->key, drop_key->key_len, 0));
3993 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3994 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4002 ofproto->n_missed++;
4003 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4004 &flow.tunnel, flow.in_port, &miss->flow);
4006 /* Add other packets to a to-do list. */
4007 hash = flow_hash(&miss->flow, 0);
4008 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4009 if (!existing_miss) {
4010 hmap_insert(&todo, &miss->hmap_node, hash);
4011 miss->ofproto = ofproto;
4012 miss->key = upcall->key;
4013 miss->key_len = upcall->key_len;
4014 miss->upcall_type = upcall->type;
4015 miss->odp_in_port = odp_in_port;
4016 list_init(&miss->packets);
4020 miss = existing_miss;
4022 list_push_back(&miss->packets, &upcall->packet->list_node);
4025 /* Process each element in the to-do list, constructing the set of
4026 * operations to batch. */
4028 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4029 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4031 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4033 /* Execute batch. */
4034 for (i = 0; i < n_ops; i++) {
4035 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4037 dpif_operate(backer->dpif, dpif_ops, n_ops);
4040 for (i = 0; i < n_ops; i++) {
4041 free(flow_miss_ops[i].garbage);
4043 hmap_destroy(&todo);
4046 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4048 classify_upcall(const struct dpif_upcall *upcall)
4050 size_t userdata_len;
4051 union user_action_cookie cookie;
4053 /* First look at the upcall type. */
4054 switch (upcall->type) {
4055 case DPIF_UC_ACTION:
4061 case DPIF_N_UC_TYPES:
4063 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4067 /* "action" upcalls need a closer look. */
4068 if (!upcall->userdata) {
4069 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4072 userdata_len = nl_attr_get_size(upcall->userdata);
4073 if (userdata_len < sizeof cookie.type
4074 || userdata_len > sizeof cookie) {
4075 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4079 memset(&cookie, 0, sizeof cookie);
4080 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4081 if (userdata_len == sizeof cookie.sflow
4082 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4083 return SFLOW_UPCALL;
4084 } else if (userdata_len == sizeof cookie.slow_path
4085 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4087 } else if (userdata_len == sizeof cookie.flow_sample
4088 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4089 return FLOW_SAMPLE_UPCALL;
4090 } else if (userdata_len == sizeof cookie.ipfix
4091 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4092 return IPFIX_UPCALL;
4094 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4095 " and size %zu", cookie.type, userdata_len);
4101 handle_sflow_upcall(struct dpif_backer *backer,
4102 const struct dpif_upcall *upcall)
4104 struct ofproto_dpif *ofproto;
4105 union user_action_cookie cookie;
4107 uint32_t odp_in_port;
4109 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4110 &flow, NULL, &ofproto, &odp_in_port, NULL)
4111 || !ofproto->sflow) {
4115 memset(&cookie, 0, sizeof cookie);
4116 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4117 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4118 odp_in_port, &cookie);
4122 handle_flow_sample_upcall(struct dpif_backer *backer,
4123 const struct dpif_upcall *upcall)
4125 struct ofproto_dpif *ofproto;
4126 union user_action_cookie cookie;
4129 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4130 &flow, NULL, &ofproto, NULL, NULL)
4131 || !ofproto->ipfix) {
4135 memset(&cookie, 0, sizeof cookie);
4136 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4138 /* The flow reflects exactly the contents of the packet. Sample
4139 * the packet using it. */
4140 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4141 cookie.flow_sample.collector_set_id,
4142 cookie.flow_sample.probability,
4143 cookie.flow_sample.obs_domain_id,
4144 cookie.flow_sample.obs_point_id);
4148 handle_ipfix_upcall(struct dpif_backer *backer,
4149 const struct dpif_upcall *upcall)
4151 struct ofproto_dpif *ofproto;
4154 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4155 &flow, NULL, &ofproto, NULL, NULL)
4156 || !ofproto->ipfix) {
4160 /* The flow reflects exactly the contents of the packet. Sample
4161 * the packet using it. */
4162 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4166 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4168 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4169 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4170 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4175 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4178 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4179 struct dpif_upcall *upcall = &misses[n_misses];
4180 struct ofpbuf *buf = &miss_bufs[n_misses];
4183 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4184 sizeof miss_buf_stubs[n_misses]);
4185 error = dpif_recv(backer->dpif, upcall, buf);
4191 switch (classify_upcall(upcall)) {
4193 /* Handle it later. */
4198 handle_sflow_upcall(backer, upcall);
4202 case FLOW_SAMPLE_UPCALL:
4203 handle_flow_sample_upcall(backer, upcall);
4208 handle_ipfix_upcall(backer, upcall);
4218 /* Handle deferred MISS_UPCALL processing. */
4219 handle_miss_upcalls(backer, misses, n_misses);
4220 for (i = 0; i < n_misses; i++) {
4221 ofpbuf_uninit(&miss_bufs[i]);
4227 /* Flow expiration. */
4229 static int subfacet_max_idle(const struct ofproto_dpif *);
4230 static void update_stats(struct dpif_backer *);
4231 static void rule_expire(struct rule_dpif *);
4232 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4234 /* This function is called periodically by run(). Its job is to collect
4235 * updates for the flows that have been installed into the datapath, most
4236 * importantly when they last were used, and then use that information to
4237 * expire flows that have not been used recently.
4239 * Returns the number of milliseconds after which it should be called again. */
4241 expire(struct dpif_backer *backer)
4243 struct ofproto_dpif *ofproto;
4244 int max_idle = INT32_MAX;
4246 /* Periodically clear out the drop keys in an effort to keep them
4247 * relatively few. */
4248 drop_key_clear(backer);
4250 /* Update stats for each flow in the backer. */
4251 update_stats(backer);
4253 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4254 struct rule *rule, *next_rule;
4257 if (ofproto->backer != backer) {
4261 /* Keep track of the max number of flows per ofproto_dpif. */
4262 update_max_subfacet_count(ofproto);
4264 /* Expire subfacets that have been idle too long. */
4265 dp_max_idle = subfacet_max_idle(ofproto);
4266 expire_subfacets(ofproto, dp_max_idle);
4268 max_idle = MIN(max_idle, dp_max_idle);
4270 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4272 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4273 &ofproto->up.expirable) {
4274 rule_expire(rule_dpif_cast(rule));
4277 /* All outstanding data in existing flows has been accounted, so it's a
4278 * good time to do bond rebalancing. */
4279 if (ofproto->has_bonded_bundles) {
4280 struct ofbundle *bundle;
4282 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4284 bond_rebalance(bundle->bond, &backer->revalidate_set);
4290 return MIN(max_idle, 1000);
4293 /* Updates flow table statistics given that the datapath just reported 'stats'
4294 * as 'subfacet''s statistics. */
4296 update_subfacet_stats(struct subfacet *subfacet,
4297 const struct dpif_flow_stats *stats)
4299 struct facet *facet = subfacet->facet;
4301 if (stats->n_packets >= subfacet->dp_packet_count) {
4302 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4303 facet->packet_count += extra;
4305 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4308 if (stats->n_bytes >= subfacet->dp_byte_count) {
4309 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4311 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4314 subfacet->dp_packet_count = stats->n_packets;
4315 subfacet->dp_byte_count = stats->n_bytes;
4317 facet->tcp_flags |= stats->tcp_flags;
4319 subfacet_update_time(subfacet, stats->used);
4320 if (facet->accounted_bytes < facet->byte_count) {
4322 facet_account(facet);
4323 facet->accounted_bytes = facet->byte_count;
4327 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4328 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4330 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4331 const struct nlattr *key, size_t key_len)
4333 if (!VLOG_DROP_WARN(&rl)) {
4337 odp_flow_key_format(key, key_len, &s);
4338 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4342 COVERAGE_INC(facet_unexpected);
4343 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4346 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4348 * This function also pushes statistics updates to rules which each facet
4349 * resubmits into. Generally these statistics will be accurate. However, if a
4350 * facet changes the rule it resubmits into at some time in between
4351 * update_stats() runs, it is possible that statistics accrued to the
4352 * old rule will be incorrectly attributed to the new rule. This could be
4353 * avoided by calling update_stats() whenever rules are created or
4354 * deleted. However, the performance impact of making so many calls to the
4355 * datapath do not justify the benefit of having perfectly accurate statistics.
4357 * In addition, this function maintains per ofproto flow hit counts. The patch
4358 * port is not treated specially. e.g. A packet ingress from br0 patched into
4359 * br1 will increase the hit count of br0 by 1, however, does not affect
4360 * the hit or miss counts of br1.
4363 update_stats(struct dpif_backer *backer)
4365 const struct dpif_flow_stats *stats;
4366 struct dpif_flow_dump dump;
4367 const struct nlattr *key;
4368 struct ofproto_dpif *ofproto;
4371 dpif_flow_dump_start(&dump, backer->dpif);
4372 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4374 struct subfacet *subfacet;
4375 struct ofport_dpif *ofport;
4378 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4383 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4384 ofproto->n_update_stats++;
4386 ofport = get_ofp_port(ofproto, flow.in_port);
4387 if (ofport && ofport->tnl_port) {
4388 netdev_vport_inc_rx(ofport->up.netdev, stats);
4391 key_hash = odp_flow_key_hash(key, key_len);
4392 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4393 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4395 /* Update ofproto_dpif's hit count. */
4396 if (stats->n_packets > subfacet->dp_packet_count) {
4397 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4398 dpif_stats_update_hit_count(ofproto, delta);
4401 update_subfacet_stats(subfacet, stats);
4405 /* Stats are updated per-packet. */
4408 case SF_NOT_INSTALLED:
4410 delete_unexpected_flow(ofproto, key, key_len);
4415 dpif_flow_dump_done(&dump);
4417 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4418 update_moving_averages(ofproto);
4423 /* Calculates and returns the number of milliseconds of idle time after which
4424 * subfacets should expire from the datapath. When a subfacet expires, we fold
4425 * its statistics into its facet, and when a facet's last subfacet expires, we
4426 * fold its statistic into its rule. */
4428 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4431 * Idle time histogram.
4433 * Most of the time a switch has a relatively small number of subfacets.
4434 * When this is the case we might as well keep statistics for all of them
4435 * in userspace and to cache them in the kernel datapath for performance as
4438 * As the number of subfacets increases, the memory required to maintain
4439 * statistics about them in userspace and in the kernel becomes
4440 * significant. However, with a large number of subfacets it is likely
4441 * that only a few of them are "heavy hitters" that consume a large amount
4442 * of bandwidth. At this point, only heavy hitters are worth caching in
4443 * the kernel and maintaining in userspaces; other subfacets we can
4446 * The technique used to compute the idle time is to build a histogram with
4447 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4448 * that is installed in the kernel gets dropped in the appropriate bucket.
4449 * After the histogram has been built, we compute the cutoff so that only
4450 * the most-recently-used 1% of subfacets (but at least
4451 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4452 * the most-recently-used bucket of subfacets is kept, so actually an
4453 * arbitrary number of subfacets can be kept in any given expiration run
4454 * (though the next run will delete most of those unless they receive
4457 * This requires a second pass through the subfacets, in addition to the
4458 * pass made by update_stats(), because the former function never looks at
4459 * uninstallable subfacets.
4461 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4462 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4463 int buckets[N_BUCKETS] = { 0 };
4464 int total, subtotal, bucket;
4465 struct subfacet *subfacet;
4469 total = hmap_count(&ofproto->subfacets);
4470 if (total <= ofproto->up.flow_eviction_threshold) {
4471 return N_BUCKETS * BUCKET_WIDTH;
4474 /* Build histogram. */
4476 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4477 long long int idle = now - subfacet->used;
4478 int bucket = (idle <= 0 ? 0
4479 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4480 : (unsigned int) idle / BUCKET_WIDTH);
4484 /* Find the first bucket whose flows should be expired. */
4485 subtotal = bucket = 0;
4487 subtotal += buckets[bucket++];
4488 } while (bucket < N_BUCKETS &&
4489 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4491 if (VLOG_IS_DBG_ENABLED()) {
4495 ds_put_cstr(&s, "keep");
4496 for (i = 0; i < N_BUCKETS; i++) {
4498 ds_put_cstr(&s, ", drop");
4501 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4504 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4508 return bucket * BUCKET_WIDTH;
4512 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4514 /* Cutoff time for most flows. */
4515 long long int normal_cutoff = time_msec() - dp_max_idle;
4517 /* We really want to keep flows for special protocols around, so use a more
4518 * conservative cutoff. */
4519 long long int special_cutoff = time_msec() - 10000;
4521 struct subfacet *subfacet, *next_subfacet;
4522 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4526 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4527 &ofproto->subfacets) {
4528 long long int cutoff;
4530 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4533 if (subfacet->used < cutoff) {
4534 if (subfacet->path != SF_NOT_INSTALLED) {
4535 batch[n_batch++] = subfacet;
4536 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4537 subfacet_destroy_batch(ofproto, batch, n_batch);
4541 subfacet_destroy(subfacet);
4547 subfacet_destroy_batch(ofproto, batch, n_batch);
4551 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4552 * then delete it entirely. */
4554 rule_expire(struct rule_dpif *rule)
4556 struct facet *facet, *next_facet;
4560 if (rule->up.pending) {
4561 /* We'll have to expire it later. */
4565 /* Has 'rule' expired? */
4567 if (rule->up.hard_timeout
4568 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4569 reason = OFPRR_HARD_TIMEOUT;
4570 } else if (rule->up.idle_timeout
4571 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4572 reason = OFPRR_IDLE_TIMEOUT;
4577 COVERAGE_INC(ofproto_dpif_expired);
4579 /* Update stats. (This is a no-op if the rule expired due to an idle
4580 * timeout, because that only happens when the rule has no facets left.) */
4581 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4582 facet_remove(facet);
4585 /* Get rid of the rule. */
4586 ofproto_rule_expire(&rule->up, reason);
4591 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4593 * The caller must already have determined that no facet with an identical
4594 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4595 * the ofproto's classifier table.
4597 * 'hash' must be the return value of flow_hash(flow, 0).
4599 * The facet will initially have no subfacets. The caller should create (at
4600 * least) one subfacet with subfacet_create(). */
4601 static struct facet *
4602 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4604 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4605 struct facet *facet;
4607 facet = xzalloc(sizeof *facet);
4608 facet->used = time_msec();
4609 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4610 list_push_back(&rule->facets, &facet->list_node);
4612 facet->flow = *flow;
4613 list_init(&facet->subfacets);
4614 netflow_flow_init(&facet->nf_flow);
4615 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4617 facet->learn_rl = time_msec() + 500;
4623 facet_free(struct facet *facet)
4628 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4629 * 'packet', which arrived on 'in_port'. */
4631 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4632 const struct nlattr *odp_actions, size_t actions_len,
4633 struct ofpbuf *packet)
4635 struct odputil_keybuf keybuf;
4639 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4640 odp_flow_key_from_flow(&key, flow,
4641 ofp_port_to_odp_port(ofproto, flow->in_port));
4643 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4644 odp_actions, actions_len, packet);
4648 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4650 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4651 * rule's statistics, via subfacet_uninstall().
4653 * - Removes 'facet' from its rule and from ofproto->facets.
4656 facet_remove(struct facet *facet)
4658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4659 struct subfacet *subfacet, *next_subfacet;
4661 ovs_assert(!list_is_empty(&facet->subfacets));
4663 /* First uninstall all of the subfacets to get final statistics. */
4664 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4665 subfacet_uninstall(subfacet);
4668 /* Flush the final stats to the rule.
4670 * This might require us to have at least one subfacet around so that we
4671 * can use its actions for accounting in facet_account(), which is why we
4672 * have uninstalled but not yet destroyed the subfacets. */
4673 facet_flush_stats(facet);
4675 /* Now we're really all done so destroy everything. */
4676 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4677 &facet->subfacets) {
4678 subfacet_destroy__(subfacet);
4680 hmap_remove(&ofproto->facets, &facet->hmap_node);
4681 list_remove(&facet->list_node);
4685 /* Feed information from 'facet' back into the learning table to keep it in
4686 * sync with what is actually flowing through the datapath. */
4688 facet_learn(struct facet *facet)
4690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4691 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4692 struct subfacet, list_node);
4693 long long int now = time_msec();
4694 struct action_xlate_ctx ctx;
4696 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4700 facet->learn_rl = now + 500;
4702 if (!facet->has_learn
4703 && !facet->has_normal
4704 && (!facet->has_fin_timeout
4705 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4709 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4710 &subfacet->initial_vals,
4711 facet->rule, facet->tcp_flags, NULL);
4712 ctx.may_learn = true;
4713 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4714 facet->rule->up.ofpacts_len);
4718 facet_account(struct facet *facet)
4720 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4721 struct subfacet *subfacet = facet_get_subfacet(facet);
4722 const struct nlattr *a;
4727 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4730 n_bytes = facet->byte_count - facet->accounted_bytes;
4732 /* This loop feeds byte counters to bond_account() for rebalancing to use
4733 * as a basis. We also need to track the actual VLAN on which the packet
4734 * is going to be sent to ensure that it matches the one passed to
4735 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4738 * We use the actions from an arbitrary subfacet because they should all
4739 * be equally valid for our purpose. */
4740 vlan_tci = facet->flow.vlan_tci;
4741 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4742 subfacet->actions, subfacet->actions_len) {
4743 const struct ovs_action_push_vlan *vlan;
4744 struct ofport_dpif *port;
4746 switch (nl_attr_type(a)) {
4747 case OVS_ACTION_ATTR_OUTPUT:
4748 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4749 if (port && port->bundle && port->bundle->bond) {
4750 bond_account(port->bundle->bond, &facet->flow,
4751 vlan_tci_to_vid(vlan_tci), n_bytes);
4755 case OVS_ACTION_ATTR_POP_VLAN:
4756 vlan_tci = htons(0);
4759 case OVS_ACTION_ATTR_PUSH_VLAN:
4760 vlan = nl_attr_get(a);
4761 vlan_tci = vlan->vlan_tci;
4767 /* Returns true if the only action for 'facet' is to send to the controller.
4768 * (We don't report NetFlow expiration messages for such facets because they
4769 * are just part of the control logic for the network, not real traffic). */
4771 facet_is_controller_flow(struct facet *facet)
4774 const struct rule *rule = &facet->rule->up;
4775 const struct ofpact *ofpacts = rule->ofpacts;
4776 size_t ofpacts_len = rule->ofpacts_len;
4778 if (ofpacts_len > 0 &&
4779 ofpacts->type == OFPACT_CONTROLLER &&
4780 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4787 /* Folds all of 'facet''s statistics into its rule. Also updates the
4788 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4789 * 'facet''s statistics in the datapath should have been zeroed and folded into
4790 * its packet and byte counts before this function is called. */
4792 facet_flush_stats(struct facet *facet)
4794 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4795 struct subfacet *subfacet;
4797 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4798 ovs_assert(!subfacet->dp_byte_count);
4799 ovs_assert(!subfacet->dp_packet_count);
4802 facet_push_stats(facet);
4803 if (facet->accounted_bytes < facet->byte_count) {
4804 facet_account(facet);
4805 facet->accounted_bytes = facet->byte_count;
4808 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4809 struct ofexpired expired;
4810 expired.flow = facet->flow;
4811 expired.packet_count = facet->packet_count;
4812 expired.byte_count = facet->byte_count;
4813 expired.used = facet->used;
4814 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4817 facet->rule->packet_count += facet->packet_count;
4818 facet->rule->byte_count += facet->byte_count;
4820 /* Reset counters to prevent double counting if 'facet' ever gets
4822 facet_reset_counters(facet);
4824 netflow_flow_clear(&facet->nf_flow);
4825 facet->tcp_flags = 0;
4828 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4829 * Returns it if found, otherwise a null pointer.
4831 * 'hash' must be the return value of flow_hash(flow, 0).
4833 * The returned facet might need revalidation; use facet_lookup_valid()
4834 * instead if that is important. */
4835 static struct facet *
4836 facet_find(struct ofproto_dpif *ofproto,
4837 const struct flow *flow, uint32_t hash)
4839 struct facet *facet;
4841 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4842 if (flow_equal(flow, &facet->flow)) {
4850 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4851 * Returns it if found, otherwise a null pointer.
4853 * 'hash' must be the return value of flow_hash(flow, 0).
4855 * The returned facet is guaranteed to be valid. */
4856 static struct facet *
4857 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4860 struct facet *facet;
4862 facet = facet_find(ofproto, flow, hash);
4864 && (ofproto->backer->need_revalidate
4865 || tag_set_intersects(&ofproto->backer->revalidate_set,
4867 facet_revalidate(facet);
4869 /* facet_revalidate() may have destroyed 'facet'. */
4870 facet = facet_find(ofproto, flow, hash);
4876 /* Return a subfacet from 'facet'. A facet consists of one or more
4877 * subfacets, and this function returns one of them. */
4878 static struct subfacet *facet_get_subfacet(struct facet *facet)
4880 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4885 subfacet_path_to_string(enum subfacet_path path)
4888 case SF_NOT_INSTALLED:
4889 return "not installed";
4891 return "in fast path";
4893 return "in slow path";
4899 /* Returns the path in which a subfacet should be installed if its 'slow'
4900 * member has the specified value. */
4901 static enum subfacet_path
4902 subfacet_want_path(enum slow_path_reason slow)
4904 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4907 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4908 * supposing that its actions have been recalculated as 'want_actions' and that
4909 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4911 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4912 const struct ofpbuf *want_actions)
4914 enum subfacet_path want_path = subfacet_want_path(slow);
4915 return (want_path != subfacet->path
4916 || (want_path == SF_FAST_PATH
4917 && (subfacet->actions_len != want_actions->size
4918 || memcmp(subfacet->actions, want_actions->data,
4919 subfacet->actions_len))));
4923 facet_check_consistency(struct facet *facet)
4925 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4929 uint64_t odp_actions_stub[1024 / 8];
4930 struct ofpbuf odp_actions;
4932 struct rule_dpif *rule;
4933 struct subfacet *subfacet;
4934 bool may_log = false;
4937 /* Check the rule for consistency. */
4938 rule = rule_dpif_lookup(ofproto, &facet->flow);
4939 ok = rule == facet->rule;
4941 may_log = !VLOG_DROP_WARN(&rl);
4946 flow_format(&s, &facet->flow);
4947 ds_put_format(&s, ": facet associated with wrong rule (was "
4948 "table=%"PRIu8",", facet->rule->up.table_id);
4949 cls_rule_format(&facet->rule->up.cr, &s);
4950 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4952 cls_rule_format(&rule->up.cr, &s);
4953 ds_put_char(&s, ')');
4955 VLOG_WARN("%s", ds_cstr(&s));
4960 /* Check the datapath actions for consistency. */
4961 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4962 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4963 enum subfacet_path want_path;
4964 struct action_xlate_ctx ctx;
4967 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4968 &subfacet->initial_vals, rule, 0, NULL);
4969 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4972 if (subfacet->path == SF_NOT_INSTALLED) {
4973 /* This only happens if the datapath reported an error when we
4974 * tried to install the flow. Don't flag another error here. */
4978 want_path = subfacet_want_path(subfacet->slow);
4979 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4980 /* The actions for slow-path flows may legitimately vary from one
4981 * packet to the next. We're done. */
4985 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4989 /* Inconsistency! */
4991 may_log = !VLOG_DROP_WARN(&rl);
4995 /* Rate-limited, skip reporting. */
5000 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5002 ds_put_cstr(&s, ": inconsistency in subfacet");
5003 if (want_path != subfacet->path) {
5004 enum odp_key_fitness fitness = subfacet->key_fitness;
5006 ds_put_format(&s, " (%s, fitness=%s)",
5007 subfacet_path_to_string(subfacet->path),
5008 odp_key_fitness_to_string(fitness));
5009 ds_put_format(&s, " (should have been %s)",
5010 subfacet_path_to_string(want_path));
5011 } else if (want_path == SF_FAST_PATH) {
5012 ds_put_cstr(&s, " (actions were: ");
5013 format_odp_actions(&s, subfacet->actions,
5014 subfacet->actions_len);
5015 ds_put_cstr(&s, ") (correct actions: ");
5016 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5017 ds_put_char(&s, ')');
5019 ds_put_cstr(&s, " (actions: ");
5020 format_odp_actions(&s, subfacet->actions,
5021 subfacet->actions_len);
5022 ds_put_char(&s, ')');
5024 VLOG_WARN("%s", ds_cstr(&s));
5027 ofpbuf_uninit(&odp_actions);
5032 /* Re-searches the classifier for 'facet':
5034 * - If the rule found is different from 'facet''s current rule, moves
5035 * 'facet' to the new rule and recompiles its actions.
5037 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5038 * where it is and recompiles its actions anyway.
5040 * - If any of 'facet''s subfacets correspond to a new flow according to
5041 * ofproto_receive(), 'facet' is removed. */
5043 facet_revalidate(struct facet *facet)
5045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5047 struct nlattr *odp_actions;
5050 struct actions *new_actions;
5052 struct action_xlate_ctx ctx;
5053 uint64_t odp_actions_stub[1024 / 8];
5054 struct ofpbuf odp_actions;
5056 struct rule_dpif *new_rule;
5057 struct subfacet *subfacet;
5060 COVERAGE_INC(facet_revalidate);
5062 /* Check that child subfacets still correspond to this facet. Tunnel
5063 * configuration changes could cause a subfacet's OpenFlow in_port to
5065 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5066 struct ofproto_dpif *recv_ofproto;
5067 struct flow recv_flow;
5070 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5071 subfacet->key_len, &recv_flow, NULL,
5072 &recv_ofproto, NULL, NULL);
5074 || recv_ofproto != ofproto
5075 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5076 facet_remove(facet);
5081 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5083 /* Calculate new datapath actions.
5085 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5086 * emit a NetFlow expiration and, if so, we need to have the old state
5087 * around to properly compose it. */
5089 /* If the datapath actions changed or the installability changed,
5090 * then we need to talk to the datapath. */
5093 memset(&ctx, 0, sizeof ctx);
5094 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5095 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5096 enum slow_path_reason slow;
5098 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5099 &subfacet->initial_vals, new_rule, 0, NULL);
5100 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5103 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5104 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5105 struct dpif_flow_stats stats;
5107 subfacet_install(subfacet,
5108 odp_actions.data, odp_actions.size, &stats, slow);
5109 subfacet_update_stats(subfacet, &stats);
5112 new_actions = xcalloc(list_size(&facet->subfacets),
5113 sizeof *new_actions);
5115 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5117 new_actions[i].actions_len = odp_actions.size;
5122 ofpbuf_uninit(&odp_actions);
5125 facet_flush_stats(facet);
5128 /* Update 'facet' now that we've taken care of all the old state. */
5129 facet->tags = ctx.tags;
5130 facet->nf_flow.output_iface = ctx.nf_output_iface;
5131 facet->has_learn = ctx.has_learn;
5132 facet->has_normal = ctx.has_normal;
5133 facet->has_fin_timeout = ctx.has_fin_timeout;
5134 facet->mirrors = ctx.mirrors;
5137 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5138 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5140 if (new_actions && new_actions[i].odp_actions) {
5141 free(subfacet->actions);
5142 subfacet->actions = new_actions[i].odp_actions;
5143 subfacet->actions_len = new_actions[i].actions_len;
5149 if (facet->rule != new_rule) {
5150 COVERAGE_INC(facet_changed_rule);
5151 list_remove(&facet->list_node);
5152 list_push_back(&new_rule->facets, &facet->list_node);
5153 facet->rule = new_rule;
5154 facet->used = new_rule->up.created;
5155 facet->prev_used = facet->used;
5159 /* Updates 'facet''s used time. Caller is responsible for calling
5160 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5162 facet_update_time(struct facet *facet, long long int used)
5164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5165 if (used > facet->used) {
5167 ofproto_rule_update_used(&facet->rule->up, used);
5168 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5173 facet_reset_counters(struct facet *facet)
5175 facet->packet_count = 0;
5176 facet->byte_count = 0;
5177 facet->prev_packet_count = 0;
5178 facet->prev_byte_count = 0;
5179 facet->accounted_bytes = 0;
5183 facet_push_stats(struct facet *facet)
5185 struct dpif_flow_stats stats;
5187 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5188 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5189 ovs_assert(facet->used >= facet->prev_used);
5191 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5192 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5193 stats.used = facet->used;
5194 stats.tcp_flags = 0;
5196 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5197 facet->prev_packet_count = facet->packet_count;
5198 facet->prev_byte_count = facet->byte_count;
5199 facet->prev_used = facet->used;
5201 flow_push_stats(facet, &stats);
5203 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5204 facet->mirrors, stats.n_packets, stats.n_bytes);
5209 push_all_stats__(bool run_fast)
5211 static long long int rl = LLONG_MIN;
5212 struct ofproto_dpif *ofproto;
5214 if (time_msec() < rl) {
5218 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5219 struct facet *facet;
5221 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5222 facet_push_stats(facet);
5229 rl = time_msec() + 100;
5233 push_all_stats(void)
5235 push_all_stats__(true);
5239 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5241 rule->packet_count += stats->n_packets;
5242 rule->byte_count += stats->n_bytes;
5243 ofproto_rule_update_used(&rule->up, stats->used);
5246 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5247 * into given 'facet->rule''s actions and mirrors. */
5249 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5251 struct rule_dpif *rule = facet->rule;
5252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5253 struct subfacet *subfacet = facet_get_subfacet(facet);
5254 struct action_xlate_ctx ctx;
5256 ofproto_rule_update_used(&rule->up, stats->used);
5258 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5259 &subfacet->initial_vals, rule, 0, NULL);
5260 ctx.resubmit_stats = stats;
5261 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5262 rule->up.ofpacts_len);
5267 static struct subfacet *
5268 subfacet_find(struct ofproto_dpif *ofproto,
5269 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5271 struct subfacet *subfacet;
5273 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5274 &ofproto->subfacets) {
5275 if (subfacet->key_len == key_len
5276 && !memcmp(key, subfacet->key, key_len)) {
5284 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5285 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5286 * existing subfacet if there is one, otherwise creates and returns a
5289 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5290 * which case the caller must populate the actions with
5291 * subfacet_make_actions(). */
5292 static struct subfacet *
5293 subfacet_create(struct facet *facet, struct flow_miss *miss,
5296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5297 enum odp_key_fitness key_fitness = miss->key_fitness;
5298 const struct nlattr *key = miss->key;
5299 size_t key_len = miss->key_len;
5301 struct subfacet *subfacet;
5303 key_hash = odp_flow_key_hash(key, key_len);
5305 if (list_is_empty(&facet->subfacets)) {
5306 subfacet = &facet->one_subfacet;
5308 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5310 if (subfacet->facet == facet) {
5314 /* This shouldn't happen. */
5315 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5316 subfacet_destroy(subfacet);
5319 subfacet = xmalloc(sizeof *subfacet);
5322 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5323 list_push_back(&facet->subfacets, &subfacet->list_node);
5324 subfacet->facet = facet;
5325 subfacet->key_fitness = key_fitness;
5326 subfacet->key = xmemdup(key, key_len);
5327 subfacet->key_len = key_len;
5328 subfacet->used = now;
5329 subfacet->created = now;
5330 subfacet->dp_packet_count = 0;
5331 subfacet->dp_byte_count = 0;
5332 subfacet->actions_len = 0;
5333 subfacet->actions = NULL;
5334 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5337 subfacet->path = SF_NOT_INSTALLED;
5338 subfacet->initial_vals = miss->initial_vals;
5339 subfacet->odp_in_port = miss->odp_in_port;
5341 ofproto->subfacet_add_count++;
5345 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5346 * its facet within 'ofproto', and frees it. */
5348 subfacet_destroy__(struct subfacet *subfacet)
5350 struct facet *facet = subfacet->facet;
5351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5353 /* Update ofproto stats before uninstall the subfacet. */
5354 ofproto->subfacet_del_count++;
5355 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5357 subfacet_uninstall(subfacet);
5358 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5359 list_remove(&subfacet->list_node);
5360 free(subfacet->key);
5361 free(subfacet->actions);
5362 if (subfacet != &facet->one_subfacet) {
5367 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5368 * last remaining subfacet in its facet destroys the facet too. */
5370 subfacet_destroy(struct subfacet *subfacet)
5372 struct facet *facet = subfacet->facet;
5374 if (list_is_singleton(&facet->subfacets)) {
5375 /* facet_remove() needs at least one subfacet (it will remove it). */
5376 facet_remove(facet);
5378 subfacet_destroy__(subfacet);
5383 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5384 struct subfacet **subfacets, int n)
5386 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5387 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5388 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5391 for (i = 0; i < n; i++) {
5392 ops[i].type = DPIF_OP_FLOW_DEL;
5393 ops[i].u.flow_del.key = subfacets[i]->key;
5394 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5395 ops[i].u.flow_del.stats = &stats[i];
5399 dpif_operate(ofproto->backer->dpif, opsp, n);
5400 for (i = 0; i < n; i++) {
5401 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5402 subfacets[i]->path = SF_NOT_INSTALLED;
5403 subfacet_destroy(subfacets[i]);
5408 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5409 * Translates the actions into 'odp_actions', which the caller must have
5410 * initialized and is responsible for uninitializing. */
5412 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5413 struct ofpbuf *odp_actions)
5415 struct facet *facet = subfacet->facet;
5416 struct rule_dpif *rule = facet->rule;
5417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5419 struct action_xlate_ctx ctx;
5421 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5422 &subfacet->initial_vals, rule, 0, packet);
5423 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5424 facet->tags = ctx.tags;
5425 facet->has_learn = ctx.has_learn;
5426 facet->has_normal = ctx.has_normal;
5427 facet->has_fin_timeout = ctx.has_fin_timeout;
5428 facet->nf_flow.output_iface = ctx.nf_output_iface;
5429 facet->mirrors = ctx.mirrors;
5431 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5432 if (subfacet->actions_len != odp_actions->size
5433 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5434 free(subfacet->actions);
5435 subfacet->actions_len = odp_actions->size;
5436 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5440 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5441 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5442 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5443 * since 'subfacet' was last updated.
5445 * Returns 0 if successful, otherwise a positive errno value. */
5447 subfacet_install(struct subfacet *subfacet,
5448 const struct nlattr *actions, size_t actions_len,
5449 struct dpif_flow_stats *stats,
5450 enum slow_path_reason slow)
5452 struct facet *facet = subfacet->facet;
5453 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5454 enum subfacet_path path = subfacet_want_path(slow);
5455 uint64_t slow_path_stub[128 / 8];
5456 enum dpif_flow_put_flags flags;
5459 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5461 flags |= DPIF_FP_ZERO_STATS;
5464 if (path == SF_SLOW_PATH) {
5465 compose_slow_path(ofproto, &facet->flow, slow,
5466 slow_path_stub, sizeof slow_path_stub,
5467 &actions, &actions_len);
5470 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5471 subfacet->key_len, actions, actions_len, stats);
5474 subfacet_reset_dp_stats(subfacet, stats);
5478 subfacet->path = path;
5484 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5486 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5487 stats, subfacet->slow);
5490 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5492 subfacet_uninstall(struct subfacet *subfacet)
5494 if (subfacet->path != SF_NOT_INSTALLED) {
5495 struct rule_dpif *rule = subfacet->facet->rule;
5496 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5497 struct dpif_flow_stats stats;
5500 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5501 subfacet->key_len, &stats);
5502 subfacet_reset_dp_stats(subfacet, &stats);
5504 subfacet_update_stats(subfacet, &stats);
5506 subfacet->path = SF_NOT_INSTALLED;
5508 ovs_assert(subfacet->dp_packet_count == 0);
5509 ovs_assert(subfacet->dp_byte_count == 0);
5513 /* Resets 'subfacet''s datapath statistics counters. This should be called
5514 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5515 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5516 * was reset in the datapath. 'stats' will be modified to include only
5517 * statistics new since 'subfacet' was last updated. */
5519 subfacet_reset_dp_stats(struct subfacet *subfacet,
5520 struct dpif_flow_stats *stats)
5523 && subfacet->dp_packet_count <= stats->n_packets
5524 && subfacet->dp_byte_count <= stats->n_bytes) {
5525 stats->n_packets -= subfacet->dp_packet_count;
5526 stats->n_bytes -= subfacet->dp_byte_count;
5529 subfacet->dp_packet_count = 0;
5530 subfacet->dp_byte_count = 0;
5533 /* Updates 'subfacet''s used time. The caller is responsible for calling
5534 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5536 subfacet_update_time(struct subfacet *subfacet, long long int used)
5538 if (used > subfacet->used) {
5539 subfacet->used = used;
5540 facet_update_time(subfacet->facet, used);
5544 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5546 * Because of the meaning of a subfacet's counters, it only makes sense to do
5547 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5548 * represents a packet that was sent by hand or if it represents statistics
5549 * that have been cleared out of the datapath. */
5551 subfacet_update_stats(struct subfacet *subfacet,
5552 const struct dpif_flow_stats *stats)
5554 if (stats->n_packets || stats->used > subfacet->used) {
5555 struct facet *facet = subfacet->facet;
5557 subfacet_update_time(subfacet, stats->used);
5558 facet->packet_count += stats->n_packets;
5559 facet->byte_count += stats->n_bytes;
5560 facet->tcp_flags |= stats->tcp_flags;
5561 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5567 static struct rule_dpif *
5568 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5570 struct rule_dpif *rule;
5572 rule = rule_dpif_lookup__(ofproto, flow, 0);
5577 return rule_dpif_miss_rule(ofproto, flow);
5580 static struct rule_dpif *
5581 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5584 struct cls_rule *cls_rule;
5585 struct classifier *cls;
5587 if (table_id >= N_TABLES) {
5591 cls = &ofproto->up.tables[table_id].cls;
5592 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5593 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5594 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5595 * are unavailable. */
5596 struct flow ofpc_normal_flow = *flow;
5597 ofpc_normal_flow.tp_src = htons(0);
5598 ofpc_normal_flow.tp_dst = htons(0);
5599 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5601 cls_rule = classifier_lookup(cls, flow);
5603 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5606 static struct rule_dpif *
5607 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5609 struct ofport_dpif *port;
5611 port = get_ofp_port(ofproto, flow->in_port);
5613 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5614 return ofproto->miss_rule;
5617 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5618 return ofproto->no_packet_in_rule;
5620 return ofproto->miss_rule;
5624 complete_operation(struct rule_dpif *rule)
5626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5628 rule_invalidate(rule);
5630 struct dpif_completion *c = xmalloc(sizeof *c);
5631 c->op = rule->up.pending;
5632 list_push_back(&ofproto->completions, &c->list_node);
5634 ofoperation_complete(rule->up.pending, 0);
5638 static struct rule *
5641 struct rule_dpif *rule = xmalloc(sizeof *rule);
5646 rule_dealloc(struct rule *rule_)
5648 struct rule_dpif *rule = rule_dpif_cast(rule_);
5653 rule_construct(struct rule *rule_)
5655 struct rule_dpif *rule = rule_dpif_cast(rule_);
5656 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5657 struct rule_dpif *victim;
5660 rule->packet_count = 0;
5661 rule->byte_count = 0;
5663 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5664 if (victim && !list_is_empty(&victim->facets)) {
5665 struct facet *facet;
5667 rule->facets = victim->facets;
5668 list_moved(&rule->facets);
5669 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5670 /* XXX: We're only clearing our local counters here. It's possible
5671 * that quite a few packets are unaccounted for in the datapath
5672 * statistics. These will be accounted to the new rule instead of
5673 * cleared as required. This could be fixed by clearing out the
5674 * datapath statistics for this facet, but currently it doesn't
5676 facet_reset_counters(facet);
5680 /* Must avoid list_moved() in this case. */
5681 list_init(&rule->facets);
5684 table_id = rule->up.table_id;
5686 rule->tag = victim->tag;
5687 } else if (table_id == 0) {
5692 miniflow_expand(&rule->up.cr.match.flow, &flow);
5693 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5694 ofproto->tables[table_id].basis);
5697 complete_operation(rule);
5702 rule_destruct(struct rule *rule_)
5704 struct rule_dpif *rule = rule_dpif_cast(rule_);
5705 struct facet *facet, *next_facet;
5707 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5708 facet_revalidate(facet);
5711 complete_operation(rule);
5715 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5717 struct rule_dpif *rule = rule_dpif_cast(rule_);
5718 struct facet *facet;
5720 /* push_all_stats() can handle flow misses which, when using the learn
5721 * action, can cause rules to be added and deleted. This can corrupt our
5722 * caller's datastructures which assume that rule_get_stats() doesn't have
5723 * an impact on the flow table. To be safe, we disable miss handling. */
5724 push_all_stats__(false);
5726 /* Start from historical data for 'rule' itself that are no longer tracked
5727 * in facets. This counts, for example, facets that have expired. */
5728 *packets = rule->packet_count;
5729 *bytes = rule->byte_count;
5731 /* Add any statistics that are tracked by facets. This includes
5732 * statistical data recently updated by ofproto_update_stats() as well as
5733 * stats for packets that were executed "by hand" via dpif_execute(). */
5734 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5735 *packets += facet->packet_count;
5736 *bytes += facet->byte_count;
5741 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5742 struct ofpbuf *packet)
5744 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5745 struct initial_vals initial_vals;
5746 struct dpif_flow_stats stats;
5747 struct action_xlate_ctx ctx;
5748 uint64_t odp_actions_stub[1024 / 8];
5749 struct ofpbuf odp_actions;
5751 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5752 rule_credit_stats(rule, &stats);
5754 initial_vals.vlan_tci = flow->vlan_tci;
5755 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5756 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5757 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5758 rule, stats.tcp_flags, packet);
5759 ctx.resubmit_stats = &stats;
5760 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5762 execute_odp_actions(ofproto, flow, odp_actions.data,
5763 odp_actions.size, packet);
5765 ofpbuf_uninit(&odp_actions);
5769 rule_execute(struct rule *rule, const struct flow *flow,
5770 struct ofpbuf *packet)
5772 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5773 ofpbuf_delete(packet);
5778 rule_modify_actions(struct rule *rule_)
5780 struct rule_dpif *rule = rule_dpif_cast(rule_);
5782 complete_operation(rule);
5785 /* Sends 'packet' out 'ofport'.
5786 * May modify 'packet'.
5787 * Returns 0 if successful, otherwise a positive errno value. */
5789 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5791 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5792 uint64_t odp_actions_stub[1024 / 8];
5793 struct ofpbuf key, odp_actions;
5794 struct odputil_keybuf keybuf;
5799 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5800 if (netdev_vport_is_patch(ofport->up.netdev)) {
5801 struct ofproto_dpif *peer_ofproto;
5802 struct dpif_flow_stats stats;
5803 struct ofport_dpif *peer;
5804 struct rule_dpif *rule;
5806 peer = ofport_get_peer(ofport);
5811 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5812 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5813 netdev_vport_inc_rx(peer->up.netdev, &stats);
5815 flow.in_port = peer->up.ofp_port;
5816 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5817 rule = rule_dpif_lookup(peer_ofproto, &flow);
5818 rule_dpif_execute(rule, &flow, packet);
5823 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5825 if (ofport->tnl_port) {
5826 struct dpif_flow_stats stats;
5828 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5829 if (odp_port == OVSP_NONE) {
5833 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5834 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5835 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5836 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5838 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5840 if (odp_port != ofport->odp_port) {
5841 eth_pop_vlan(packet);
5842 flow.vlan_tci = htons(0);
5846 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5847 odp_flow_key_from_flow(&key, &flow,
5848 ofp_port_to_odp_port(ofproto, flow.in_port));
5850 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5851 compose_ipfix_action(ofproto, &odp_actions, &flow);
5853 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5854 error = dpif_execute(ofproto->backer->dpif,
5856 odp_actions.data, odp_actions.size,
5858 ofpbuf_uninit(&odp_actions);
5861 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5862 ofproto->up.name, odp_port, strerror(error));
5864 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5868 /* OpenFlow to datapath action translation. */
5870 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5871 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5872 struct action_xlate_ctx *);
5873 static void xlate_normal(struct action_xlate_ctx *);
5875 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5876 * The action will state 'slow' as the reason that the action is in the slow
5877 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5878 * dump-flows" output to see why a flow is in the slow path.)
5880 * The 'stub_size' bytes in 'stub' will be used to store the action.
5881 * 'stub_size' must be large enough for the action.
5883 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5886 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5887 enum slow_path_reason slow,
5888 uint64_t *stub, size_t stub_size,
5889 const struct nlattr **actionsp, size_t *actions_lenp)
5891 union user_action_cookie cookie;
5894 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5895 cookie.slow_path.unused = 0;
5896 cookie.slow_path.reason = slow;
5898 ofpbuf_use_stack(&buf, stub, stub_size);
5899 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5900 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5901 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5903 put_userspace_action(ofproto, &buf, flow, &cookie,
5904 sizeof cookie.slow_path);
5906 *actionsp = buf.data;
5907 *actions_lenp = buf.size;
5911 put_userspace_action(const struct ofproto_dpif *ofproto,
5912 struct ofpbuf *odp_actions,
5913 const struct flow *flow,
5914 const union user_action_cookie *cookie,
5915 const size_t cookie_size)
5919 pid = dpif_port_get_pid(ofproto->backer->dpif,
5920 ofp_port_to_odp_port(ofproto, flow->in_port));
5922 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5925 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5926 * the number of packets out of UINT32_MAX to sample. The given
5927 * cookie is passed back in the callback for each sampled packet.
5930 compose_sample_action(const struct ofproto_dpif *ofproto,
5931 struct ofpbuf *odp_actions,
5932 const struct flow *flow,
5933 const uint32_t probability,
5934 const union user_action_cookie *cookie,
5935 const size_t cookie_size)
5937 size_t sample_offset, actions_offset;
5940 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5942 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5944 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5945 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5948 nl_msg_end_nested(odp_actions, actions_offset);
5949 nl_msg_end_nested(odp_actions, sample_offset);
5950 return cookie_offset;
5954 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5955 ovs_be16 vlan_tci, uint32_t odp_port,
5956 unsigned int n_outputs, union user_action_cookie *cookie)
5960 cookie->type = USER_ACTION_COOKIE_SFLOW;
5961 cookie->sflow.vlan_tci = vlan_tci;
5963 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5964 * port information") for the interpretation of cookie->output. */
5965 switch (n_outputs) {
5967 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5968 cookie->sflow.output = 0x40000000 | 256;
5972 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5974 cookie->sflow.output = ifindex;
5979 /* 0x80000000 means "multiple output ports. */
5980 cookie->sflow.output = 0x80000000 | n_outputs;
5985 /* Compose SAMPLE action for sFlow bridge sampling. */
5987 compose_sflow_action(const struct ofproto_dpif *ofproto,
5988 struct ofpbuf *odp_actions,
5989 const struct flow *flow,
5992 uint32_t probability;
5993 union user_action_cookie cookie;
5995 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5999 probability = dpif_sflow_get_probability(ofproto->sflow);
6000 compose_sflow_cookie(ofproto, htons(0), odp_port,
6001 odp_port == OVSP_NONE ? 0 : 1, &cookie);
6003 return compose_sample_action(ofproto, odp_actions, flow, probability,
6004 &cookie, sizeof cookie.sflow);
6008 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
6009 uint32_t obs_domain_id, uint32_t obs_point_id,
6010 union user_action_cookie *cookie)
6012 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6013 cookie->flow_sample.probability = probability;
6014 cookie->flow_sample.collector_set_id = collector_set_id;
6015 cookie->flow_sample.obs_domain_id = obs_domain_id;
6016 cookie->flow_sample.obs_point_id = obs_point_id;
6020 compose_ipfix_cookie(union user_action_cookie *cookie)
6022 cookie->type = USER_ACTION_COOKIE_IPFIX;
6025 /* Compose SAMPLE action for IPFIX bridge sampling. */
6027 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6028 struct ofpbuf *odp_actions,
6029 const struct flow *flow)
6031 uint32_t probability;
6032 union user_action_cookie cookie;
6034 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6038 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6039 compose_ipfix_cookie(&cookie);
6041 compose_sample_action(ofproto, odp_actions, flow, probability,
6042 &cookie, sizeof cookie.ipfix);
6045 /* SAMPLE action for sFlow must be first action in any given list of
6046 * actions. At this point we do not have all information required to
6047 * build it. So try to build sample action as complete as possible. */
6049 add_sflow_action(struct action_xlate_ctx *ctx)
6051 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6053 &ctx->flow, OVSP_NONE);
6054 ctx->sflow_odp_port = 0;
6055 ctx->sflow_n_outputs = 0;
6058 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6059 * of actions, eventually after the SAMPLE action for sFlow. */
6061 add_ipfix_action(struct action_xlate_ctx *ctx)
6063 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6066 /* Fix SAMPLE action according to data collected while composing ODP actions.
6067 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6068 * USERSPACE action's user-cookie which is required for sflow. */
6070 fix_sflow_action(struct action_xlate_ctx *ctx)
6072 const struct flow *base = &ctx->base_flow;
6073 union user_action_cookie *cookie;
6075 if (!ctx->user_cookie_offset) {
6079 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6080 sizeof cookie->sflow);
6081 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6083 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6084 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6088 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6091 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6092 ovs_be16 flow_vlan_tci;
6093 uint32_t flow_skb_mark;
6094 uint8_t flow_nw_tos;
6095 struct priority_to_dscp *pdscp;
6096 uint32_t out_port, odp_port;
6098 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6099 * before traversing a patch port. */
6100 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6103 xlate_report(ctx, "Nonexistent output port");
6105 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6106 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6108 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6109 xlate_report(ctx, "STP not in forwarding state, skipping output");
6113 if (netdev_vport_is_patch(ofport->up.netdev)) {
6114 struct ofport_dpif *peer = ofport_get_peer(ofport);
6115 struct flow old_flow = ctx->flow;
6116 const struct ofproto_dpif *peer_ofproto;
6117 enum slow_path_reason special;
6118 struct ofport_dpif *in_port;
6121 xlate_report(ctx, "Nonexistent patch port peer");
6125 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6126 if (peer_ofproto->backer != ctx->ofproto->backer) {
6127 xlate_report(ctx, "Patch port peer on a different datapath");
6131 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6132 ctx->flow.in_port = peer->up.ofp_port;
6133 ctx->flow.metadata = htonll(0);
6134 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6135 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6137 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6138 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6141 ctx->slow |= special;
6142 } else if (!in_port || may_receive(in_port, ctx)) {
6143 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6144 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6146 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6147 * learning action look at the packet, then drop it. */
6148 struct flow old_base_flow = ctx->base_flow;
6149 size_t old_size = ctx->odp_actions->size;
6150 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6151 ctx->base_flow = old_base_flow;
6152 ctx->odp_actions->size = old_size;
6156 ctx->flow = old_flow;
6157 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6159 if (ctx->resubmit_stats) {
6160 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6161 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6167 flow_vlan_tci = ctx->flow.vlan_tci;
6168 flow_skb_mark = ctx->flow.skb_mark;
6169 flow_nw_tos = ctx->flow.nw_tos;
6171 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6173 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6174 ctx->flow.nw_tos |= pdscp->dscp;
6177 if (ofport->tnl_port) {
6178 /* Save tunnel metadata so that changes made due to
6179 * the Logical (tunnel) Port are not visible for any further
6180 * matches, while explicit set actions on tunnel metadata are.
6182 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6183 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6184 if (odp_port == OVSP_NONE) {
6185 xlate_report(ctx, "Tunneling decided against output");
6186 goto out; /* restore flow_nw_tos */
6189 if (ctx->resubmit_stats) {
6190 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6192 out_port = odp_port;
6193 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6195 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6197 odp_port = ofport->odp_port;
6198 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6199 ctx->flow.vlan_tci);
6200 if (out_port != odp_port) {
6201 ctx->flow.vlan_tci = htons(0);
6203 ctx->flow.skb_mark &= ~IPSEC_MARK;
6205 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6206 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6208 ctx->sflow_odp_port = odp_port;
6209 ctx->sflow_n_outputs++;
6210 ctx->nf_output_iface = ofp_port;
6213 ctx->flow.vlan_tci = flow_vlan_tci;
6214 ctx->flow.skb_mark = flow_skb_mark;
6216 ctx->flow.nw_tos = flow_nw_tos;
6220 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6222 compose_output_action__(ctx, ofp_port, true);
6226 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6228 struct ofproto_dpif *ofproto = ctx->ofproto;
6229 uint8_t table_id = ctx->table_id;
6231 if (table_id > 0 && table_id < N_TABLES) {
6232 struct table_dpif *table = &ofproto->tables[table_id];
6233 if (table->other_table) {
6234 ctx->tags |= (rule && rule->tag
6236 : rule_calculate_tag(&ctx->flow,
6237 &table->other_table->mask,
6243 /* Common rule processing in one place to avoid duplicating code. */
6244 static struct rule_dpif *
6245 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6248 if (ctx->resubmit_hook) {
6249 ctx->resubmit_hook(ctx, rule);
6251 if (rule == NULL && may_packet_in) {
6253 * check if table configuration flags
6254 * OFPTC_TABLE_MISS_CONTROLLER, default.
6255 * OFPTC_TABLE_MISS_CONTINUE,
6256 * OFPTC_TABLE_MISS_DROP
6257 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6259 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6261 if (rule && ctx->resubmit_stats) {
6262 rule_credit_stats(rule, ctx->resubmit_stats);
6268 xlate_table_action(struct action_xlate_ctx *ctx,
6269 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6271 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6272 struct rule_dpif *rule;
6273 uint16_t old_in_port = ctx->flow.in_port;
6274 uint8_t old_table_id = ctx->table_id;
6276 ctx->table_id = table_id;
6278 /* Look up a flow with 'in_port' as the input port. */
6279 ctx->flow.in_port = in_port;
6280 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6282 tag_the_flow(ctx, rule);
6284 /* Restore the original input port. Otherwise OFPP_NORMAL and
6285 * OFPP_IN_PORT will have surprising behavior. */
6286 ctx->flow.in_port = old_in_port;
6288 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6291 struct rule_dpif *old_rule = ctx->rule;
6295 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6296 ctx->rule = old_rule;
6300 ctx->table_id = old_table_id;
6302 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6304 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6305 MAX_RESUBMIT_RECURSION);
6306 ctx->max_resubmit_trigger = true;
6311 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6312 const struct ofpact_resubmit *resubmit)
6317 in_port = resubmit->in_port;
6318 if (in_port == OFPP_IN_PORT) {
6319 in_port = ctx->flow.in_port;
6322 table_id = resubmit->table_id;
6323 if (table_id == 255) {
6324 table_id = ctx->table_id;
6327 xlate_table_action(ctx, in_port, table_id, false);
6331 flood_packets(struct action_xlate_ctx *ctx, bool all)
6333 struct ofport_dpif *ofport;
6335 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6336 uint16_t ofp_port = ofport->up.ofp_port;
6338 if (ofp_port == ctx->flow.in_port) {
6343 compose_output_action__(ctx, ofp_port, false);
6344 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6345 compose_output_action(ctx, ofp_port);
6349 ctx->nf_output_iface = NF_OUT_FLOOD;
6353 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6354 enum ofp_packet_in_reason reason,
6355 uint16_t controller_id)
6357 struct ofputil_packet_in pin;
6358 struct ofpbuf *packet;
6360 ctx->slow |= SLOW_CONTROLLER;
6365 packet = ofpbuf_clone(ctx->packet);
6367 if (packet->l2 && packet->l3) {
6368 struct eth_header *eh;
6369 uint16_t mpls_depth;
6371 eth_pop_vlan(packet);
6374 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6375 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6377 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6378 eth_push_vlan(packet, ctx->flow.vlan_tci);
6381 mpls_depth = eth_mpls_depth(packet);
6383 if (mpls_depth < ctx->flow.mpls_depth) {
6384 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6385 } else if (mpls_depth > ctx->flow.mpls_depth) {
6386 pop_mpls(packet, ctx->flow.dl_type);
6387 } else if (mpls_depth) {
6388 set_mpls_lse(packet, ctx->flow.mpls_lse);
6392 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6393 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6394 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6398 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6399 packet_set_tcp_port(packet, ctx->flow.tp_src,
6401 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6402 packet_set_udp_port(packet, ctx->flow.tp_src,
6409 pin.packet = packet->data;
6410 pin.packet_len = packet->size;
6411 pin.reason = reason;
6412 pin.controller_id = controller_id;
6413 pin.table_id = ctx->table_id;
6414 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6417 flow_get_metadata(&ctx->flow, &pin.fmd);
6419 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6420 ofpbuf_delete(packet);
6424 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6426 ovs_assert(eth_type_mpls(eth_type));
6428 if (ctx->base_flow.mpls_depth) {
6429 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6430 ctx->flow.mpls_depth++;
6435 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6436 label = htonl(0x2); /* IPV6 Explicit Null. */
6438 label = htonl(0x0); /* IPV4 Explicit Null. */
6440 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6441 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6442 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6443 ctx->flow.mpls_depth = 1;
6445 ctx->flow.dl_type = eth_type;
6449 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6451 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6452 ovs_assert(!eth_type_mpls(eth_type));
6454 if (ctx->flow.mpls_depth) {
6455 ctx->flow.mpls_depth--;
6456 ctx->flow.mpls_lse = htonl(0);
6457 if (!ctx->flow.mpls_depth) {
6458 ctx->flow.dl_type = eth_type;
6464 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6466 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6467 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6471 if (ctx->flow.nw_ttl > 1) {
6477 for (i = 0; i < ids->n_controllers; i++) {
6478 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6482 /* Stop processing for current table. */
6488 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6490 if (!eth_type_mpls(ctx->flow.dl_type)) {
6494 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6499 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6501 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6503 if (!eth_type_mpls(ctx->flow.dl_type)) {
6509 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6512 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6514 /* Stop processing for current table. */
6520 xlate_output_action(struct action_xlate_ctx *ctx,
6521 uint16_t port, uint16_t max_len, bool may_packet_in)
6523 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6525 ctx->nf_output_iface = NF_OUT_DROP;
6529 compose_output_action(ctx, ctx->flow.in_port);
6532 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6538 flood_packets(ctx, false);
6541 flood_packets(ctx, true);
6543 case OFPP_CONTROLLER:
6544 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6550 if (port != ctx->flow.in_port) {
6551 compose_output_action(ctx, port);
6553 xlate_report(ctx, "skipping output to input port");
6558 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6559 ctx->nf_output_iface = NF_OUT_FLOOD;
6560 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6561 ctx->nf_output_iface = prev_nf_output_iface;
6562 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6563 ctx->nf_output_iface != NF_OUT_FLOOD) {
6564 ctx->nf_output_iface = NF_OUT_MULTI;
6569 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6570 const struct ofpact_output_reg *or)
6572 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6573 if (port <= UINT16_MAX) {
6574 xlate_output_action(ctx, port, or->max_len, false);
6579 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6580 const struct ofpact_enqueue *enqueue)
6582 uint16_t ofp_port = enqueue->port;
6583 uint32_t queue_id = enqueue->queue;
6584 uint32_t flow_priority, priority;
6587 /* Translate queue to priority. */
6588 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6589 queue_id, &priority);
6591 /* Fall back to ordinary output action. */
6592 xlate_output_action(ctx, enqueue->port, 0, false);
6596 /* Check output port. */
6597 if (ofp_port == OFPP_IN_PORT) {
6598 ofp_port = ctx->flow.in_port;
6599 } else if (ofp_port == ctx->flow.in_port) {
6603 /* Add datapath actions. */
6604 flow_priority = ctx->flow.skb_priority;
6605 ctx->flow.skb_priority = priority;
6606 compose_output_action(ctx, ofp_port);
6607 ctx->flow.skb_priority = flow_priority;
6609 /* Update NetFlow output port. */
6610 if (ctx->nf_output_iface == NF_OUT_DROP) {
6611 ctx->nf_output_iface = ofp_port;
6612 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6613 ctx->nf_output_iface = NF_OUT_MULTI;
6618 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6620 uint32_t skb_priority;
6622 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6623 queue_id, &skb_priority)) {
6624 ctx->flow.skb_priority = skb_priority;
6626 /* Couldn't translate queue to a priority. Nothing to do. A warning
6627 * has already been logged. */
6632 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6634 struct ofproto_dpif *ofproto = ofproto_;
6635 struct ofport_dpif *port;
6645 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6648 port = get_ofp_port(ofproto, ofp_port);
6649 return port ? port->may_enable : false;
6654 xlate_bundle_action(struct action_xlate_ctx *ctx,
6655 const struct ofpact_bundle *bundle)
6659 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6660 if (bundle->dst.field) {
6661 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6663 xlate_output_action(ctx, port, 0, false);
6668 xlate_learn_action(struct action_xlate_ctx *ctx,
6669 const struct ofpact_learn *learn)
6671 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6672 struct ofputil_flow_mod fm;
6673 uint64_t ofpacts_stub[1024 / 8];
6674 struct ofpbuf ofpacts;
6677 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6678 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6680 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6681 if (error && !VLOG_DROP_WARN(&rl)) {
6682 VLOG_WARN("learning action failed to modify flow table (%s)",
6683 ofperr_get_name(error));
6686 ofpbuf_uninit(&ofpacts);
6689 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6690 * means "infinite". */
6692 reduce_timeout(uint16_t max, uint16_t *timeout)
6694 if (max && (!*timeout || *timeout > max)) {
6700 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6701 const struct ofpact_fin_timeout *oft)
6703 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6704 struct rule_dpif *rule = ctx->rule;
6706 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6707 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6712 xlate_sample_action(struct action_xlate_ctx *ctx,
6713 const struct ofpact_sample *os)
6715 union user_action_cookie cookie;
6716 /* Scale the probability from 16-bit to 32-bit while representing
6717 * the same percentage. */
6718 uint32_t probability = (os->probability << 16) | os->probability;
6720 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6722 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6723 os->obs_domain_id, os->obs_point_id, &cookie);
6724 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6725 probability, &cookie, sizeof cookie.flow_sample);
6729 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6731 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6732 ? OFPUTIL_PC_NO_RECV_STP
6733 : OFPUTIL_PC_NO_RECV)) {
6737 /* Only drop packets here if both forwarding and learning are
6738 * disabled. If just learning is enabled, we need to have
6739 * OFPP_NORMAL and the learning action have a look at the packet
6740 * before we can drop it. */
6741 if (!stp_forward_in_state(port->stp_state)
6742 && !stp_learn_in_state(port->stp_state)) {
6750 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6752 if (is_ip_any(&ctx->base_flow)
6753 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6754 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6755 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6756 " but is not ECN capable");
6759 /* Set the ECN CE value in the tunneled packet. */
6760 ctx->flow.nw_tos |= IP_ECN_CE;
6768 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6769 struct action_xlate_ctx *ctx)
6771 bool was_evictable = true;
6772 const struct ofpact *a;
6775 /* Don't let the rule we're working on get evicted underneath us. */
6776 was_evictable = ctx->rule->up.evictable;
6777 ctx->rule->up.evictable = false;
6780 do_xlate_actions_again:
6781 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6782 struct ofpact_controller *controller;
6783 const struct ofpact_metadata *metadata;
6791 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6792 ofpact_get_OUTPUT(a)->max_len, true);
6795 case OFPACT_CONTROLLER:
6796 controller = ofpact_get_CONTROLLER(a);
6797 execute_controller_action(ctx, controller->max_len,
6799 controller->controller_id);
6802 case OFPACT_ENQUEUE:
6803 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6806 case OFPACT_SET_VLAN_VID:
6807 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6808 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6812 case OFPACT_SET_VLAN_PCP:
6813 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6814 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6819 case OFPACT_STRIP_VLAN:
6820 ctx->flow.vlan_tci = htons(0);
6823 case OFPACT_PUSH_VLAN:
6824 /* XXX 802.1AD(QinQ) */
6825 ctx->flow.vlan_tci = htons(VLAN_CFI);
6828 case OFPACT_SET_ETH_SRC:
6829 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6833 case OFPACT_SET_ETH_DST:
6834 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6838 case OFPACT_SET_IPV4_SRC:
6839 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6840 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6844 case OFPACT_SET_IPV4_DST:
6845 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6846 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6850 case OFPACT_SET_IPV4_DSCP:
6851 /* OpenFlow 1.0 only supports IPv4. */
6852 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6853 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6854 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6858 case OFPACT_SET_L4_SRC_PORT:
6859 if (is_ip_any(&ctx->flow)) {
6860 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6864 case OFPACT_SET_L4_DST_PORT:
6865 if (is_ip_any(&ctx->flow)) {
6866 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6870 case OFPACT_RESUBMIT:
6871 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6874 case OFPACT_SET_TUNNEL:
6875 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6878 case OFPACT_SET_QUEUE:
6879 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6882 case OFPACT_POP_QUEUE:
6883 ctx->flow.skb_priority = ctx->orig_skb_priority;
6886 case OFPACT_REG_MOVE:
6887 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6890 case OFPACT_REG_LOAD:
6891 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6894 case OFPACT_STACK_PUSH:
6895 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6899 case OFPACT_STACK_POP:
6900 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6904 case OFPACT_PUSH_MPLS:
6905 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6908 case OFPACT_POP_MPLS:
6909 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6912 case OFPACT_SET_MPLS_TTL:
6913 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6918 case OFPACT_DEC_MPLS_TTL:
6919 if (execute_dec_mpls_ttl_action(ctx)) {
6924 case OFPACT_DEC_TTL:
6925 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6931 /* Nothing to do. */
6934 case OFPACT_MULTIPATH:
6935 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6939 ctx->ofproto->has_bundle_action = true;
6940 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6943 case OFPACT_OUTPUT_REG:
6944 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6948 ctx->has_learn = true;
6949 if (ctx->may_learn) {
6950 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6958 case OFPACT_FIN_TIMEOUT:
6959 ctx->has_fin_timeout = true;
6960 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6963 case OFPACT_CLEAR_ACTIONS:
6965 * Nothing to do because writa-actions is not supported for now.
6966 * When writa-actions is supported, clear-actions also must
6967 * be supported at the same time.
6971 case OFPACT_WRITE_METADATA:
6972 metadata = ofpact_get_WRITE_METADATA(a);
6973 ctx->flow.metadata &= ~metadata->mask;
6974 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6977 case OFPACT_GOTO_TABLE: {
6978 /* It is assumed that goto-table is the last action. */
6979 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6980 struct rule_dpif *rule;
6982 ovs_assert(ctx->table_id < ogt->table_id);
6984 ctx->table_id = ogt->table_id;
6986 /* Look up a flow from the new table. */
6987 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6989 tag_the_flow(ctx, rule);
6991 rule = ctx_rule_hooks(ctx, rule, true);
6995 ctx->rule->up.evictable = was_evictable;
6998 was_evictable = rule->up.evictable;
6999 rule->up.evictable = false;
7001 /* Tail recursion removal. */
7002 ofpacts = rule->up.ofpacts;
7003 ofpacts_len = rule->up.ofpacts_len;
7004 goto do_xlate_actions_again;
7010 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7017 ctx->rule->up.evictable = was_evictable;
7022 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7023 struct ofproto_dpif *ofproto, const struct flow *flow,
7024 const struct initial_vals *initial_vals,
7025 struct rule_dpif *rule,
7026 uint8_t tcp_flags, const struct ofpbuf *packet)
7028 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
7030 /* Flow initialization rules:
7031 * - 'base_flow' must match the kernel's view of the packet at the
7032 * time that action processing starts. 'flow' represents any
7033 * transformations we wish to make through actions.
7034 * - By default 'base_flow' and 'flow' are the same since the input
7035 * packet matches the output before any actions are applied.
7036 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7037 * of the received packet as seen by the kernel. If we later output
7038 * to another device without any modifications this will cause us to
7039 * insert a new tag since the original one was stripped off by the
7041 * - Tunnel 'flow' is largely cleared when transitioning between
7042 * the input and output stages since it does not make sense to output
7043 * a packet with the exact headers that it was received with (i.e.
7044 * the destination IP is us). The one exception is the tun_id, which
7045 * is preserved to allow use in later resubmit lookups and loads into
7047 * - Tunnel 'base_flow' is completely cleared since that is what the
7048 * kernel does. If we wish to maintain the original values an action
7049 * needs to be generated. */
7051 ctx->ofproto = ofproto;
7053 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
7054 ctx->base_flow = ctx->flow;
7055 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7056 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
7057 ctx->flow.tunnel.tun_id = initial_tun_id;
7059 ctx->packet = packet;
7060 ctx->may_learn = packet != NULL;
7061 ctx->tcp_flags = tcp_flags;
7062 ctx->resubmit_hook = NULL;
7063 ctx->report_hook = NULL;
7064 ctx->resubmit_stats = NULL;
7067 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7068 * into datapath actions in 'odp_actions', using 'ctx'. */
7070 xlate_actions(struct action_xlate_ctx *ctx,
7071 const struct ofpact *ofpacts, size_t ofpacts_len,
7072 struct ofpbuf *odp_actions)
7074 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7075 * that in the future we always keep a copy of the original flow for
7076 * tracing purposes. */
7077 static bool hit_resubmit_limit;
7079 enum slow_path_reason special;
7080 struct ofport_dpif *in_port;
7081 struct flow orig_flow;
7083 COVERAGE_INC(ofproto_dpif_xlate);
7085 ofpbuf_clear(odp_actions);
7086 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7088 ctx->odp_actions = odp_actions;
7091 ctx->has_learn = false;
7092 ctx->has_normal = false;
7093 ctx->has_fin_timeout = false;
7094 ctx->nf_output_iface = NF_OUT_DROP;
7097 ctx->max_resubmit_trigger = false;
7098 ctx->orig_skb_priority = ctx->flow.skb_priority;
7102 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7104 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7105 /* Do this conditionally because the copy is expensive enough that it
7106 * shows up in profiles. */
7107 orig_flow = ctx->flow;
7110 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7111 switch (ctx->ofproto->up.frag_handling) {
7112 case OFPC_FRAG_NORMAL:
7113 /* We must pretend that transport ports are unavailable. */
7114 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7115 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7118 case OFPC_FRAG_DROP:
7121 case OFPC_FRAG_REASM:
7124 case OFPC_FRAG_NX_MATCH:
7125 /* Nothing to do. */
7128 case OFPC_INVALID_TTL_TO_CONTROLLER:
7133 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7134 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7136 ctx->slow |= special;
7138 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7139 struct initial_vals initial_vals;
7140 uint32_t local_odp_port;
7142 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7143 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
7145 add_sflow_action(ctx);
7146 add_ipfix_action(ctx);
7148 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7149 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7151 /* We've let OFPP_NORMAL and the learning action look at the
7152 * packet, so drop it now if forwarding is disabled. */
7153 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7154 ofpbuf_clear(ctx->odp_actions);
7155 add_sflow_action(ctx);
7156 add_ipfix_action(ctx);
7160 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7161 if (!hit_resubmit_limit) {
7162 /* We didn't record the original flow. Make sure we do from
7164 hit_resubmit_limit = true;
7165 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7166 struct ds ds = DS_EMPTY_INITIALIZER;
7168 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7169 &initial_vals, &ds);
7170 VLOG_ERR("Trace triggered by excessive resubmit "
7171 "recursion:\n%s", ds_cstr(&ds));
7176 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7177 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
7179 ctx->odp_actions->data,
7180 ctx->odp_actions->size)) {
7181 ctx->slow |= SLOW_IN_BAND;
7183 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
7185 compose_output_action(ctx, OFPP_LOCAL);
7188 if (ctx->ofproto->has_mirrors) {
7189 add_mirror_actions(ctx, &orig_flow);
7191 fix_sflow_action(ctx);
7194 ofpbuf_uninit(&ctx->stack);
7197 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7198 * into datapath actions, using 'ctx', and discards the datapath actions. */
7200 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7201 const struct ofpact *ofpacts,
7204 uint64_t odp_actions_stub[1024 / 8];
7205 struct ofpbuf odp_actions;
7207 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7208 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7209 ofpbuf_uninit(&odp_actions);
7213 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7215 if (ctx->report_hook) {
7216 ctx->report_hook(ctx, s);
7220 /* OFPP_NORMAL implementation. */
7222 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7224 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7225 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7226 * the bundle on which the packet was received, returns the VLAN to which the
7229 * Both 'vid' and the return value are in the range 0...4095. */
7231 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7233 switch (in_bundle->vlan_mode) {
7234 case PORT_VLAN_ACCESS:
7235 return in_bundle->vlan;
7238 case PORT_VLAN_TRUNK:
7241 case PORT_VLAN_NATIVE_UNTAGGED:
7242 case PORT_VLAN_NATIVE_TAGGED:
7243 return vid ? vid : in_bundle->vlan;
7250 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7251 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7254 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7255 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7258 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7260 /* Allow any VID on the OFPP_NONE port. */
7261 if (in_bundle == &ofpp_none_bundle) {
7265 switch (in_bundle->vlan_mode) {
7266 case PORT_VLAN_ACCESS:
7269 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7270 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7271 "packet received on port %s configured as VLAN "
7272 "%"PRIu16" access port",
7273 in_bundle->ofproto->up.name, vid,
7274 in_bundle->name, in_bundle->vlan);
7280 case PORT_VLAN_NATIVE_UNTAGGED:
7281 case PORT_VLAN_NATIVE_TAGGED:
7283 /* Port must always carry its native VLAN. */
7287 case PORT_VLAN_TRUNK:
7288 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7290 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7291 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7292 "received on port %s not configured for trunking "
7294 in_bundle->ofproto->up.name, vid,
7295 in_bundle->name, vid);
7307 /* Given 'vlan', the VLAN that a packet belongs to, and
7308 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7309 * that should be included in the 802.1Q header. (If the return value is 0,
7310 * then the 802.1Q header should only be included in the packet if there is a
7313 * Both 'vlan' and the return value are in the range 0...4095. */
7315 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7317 switch (out_bundle->vlan_mode) {
7318 case PORT_VLAN_ACCESS:
7321 case PORT_VLAN_TRUNK:
7322 case PORT_VLAN_NATIVE_TAGGED:
7325 case PORT_VLAN_NATIVE_UNTAGGED:
7326 return vlan == out_bundle->vlan ? 0 : vlan;
7334 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7337 struct ofport_dpif *port;
7339 ovs_be16 tci, old_tci;
7341 vid = output_vlan_to_vid(out_bundle, vlan);
7342 if (!out_bundle->bond) {
7343 port = ofbundle_get_a_port(out_bundle);
7345 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7348 /* No slaves enabled, so drop packet. */
7353 old_tci = ctx->flow.vlan_tci;
7355 if (tci || out_bundle->use_priority_tags) {
7356 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7358 tci |= htons(VLAN_CFI);
7361 ctx->flow.vlan_tci = tci;
7363 compose_output_action(ctx, port->up.ofp_port);
7364 ctx->flow.vlan_tci = old_tci;
7368 mirror_mask_ffs(mirror_mask_t mask)
7370 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7375 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7377 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7378 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7382 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7384 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7387 /* Returns an arbitrary interface within 'bundle'. */
7388 static struct ofport_dpif *
7389 ofbundle_get_a_port(const struct ofbundle *bundle)
7391 return CONTAINER_OF(list_front(&bundle->ports),
7392 struct ofport_dpif, bundle_node);
7396 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7398 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7402 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7404 struct ofproto_dpif *ofproto = ctx->ofproto;
7405 mirror_mask_t mirrors;
7406 struct ofbundle *in_bundle;
7409 const struct nlattr *a;
7412 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7413 ctx->packet != NULL, NULL);
7417 mirrors = in_bundle->src_mirrors;
7419 /* Drop frames on bundles reserved for mirroring. */
7420 if (in_bundle->mirror_out) {
7421 if (ctx->packet != NULL) {
7422 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7423 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7424 "%s, which is reserved exclusively for mirroring",
7425 ctx->ofproto->up.name, in_bundle->name);
7431 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7432 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7435 vlan = input_vid_to_vlan(in_bundle, vid);
7437 /* Look at the output ports to check for destination selections. */
7439 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7440 ctx->odp_actions->size) {
7441 enum ovs_action_attr type = nl_attr_type(a);
7442 struct ofport_dpif *ofport;
7444 if (type != OVS_ACTION_ATTR_OUTPUT) {
7448 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7449 if (ofport && ofport->bundle) {
7450 mirrors |= ofport->bundle->dst_mirrors;
7458 /* Restore the original packet before adding the mirror actions. */
7459 ctx->flow = *orig_flow;
7464 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7466 if (!vlan_is_mirrored(m, vlan)) {
7467 mirrors = zero_rightmost_1bit(mirrors);
7471 mirrors &= ~m->dup_mirrors;
7472 ctx->mirrors |= m->dup_mirrors;
7474 output_normal(ctx, m->out, vlan);
7475 } else if (vlan != m->out_vlan
7476 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7477 struct ofbundle *bundle;
7479 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7480 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7481 && !bundle->mirror_out) {
7482 output_normal(ctx, bundle, m->out_vlan);
7490 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7491 uint64_t packets, uint64_t bytes)
7497 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7500 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7503 /* In normal circumstances 'm' will not be NULL. However,
7504 * if mirrors are reconfigured, we can temporarily get out
7505 * of sync in facet_revalidate(). We could "correct" the
7506 * mirror list before reaching here, but doing that would
7507 * not properly account the traffic stats we've currently
7508 * accumulated for previous mirror configuration. */
7512 m->packet_count += packets;
7513 m->byte_count += bytes;
7517 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7518 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7519 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7521 is_gratuitous_arp(const struct flow *flow)
7523 return (flow->dl_type == htons(ETH_TYPE_ARP)
7524 && eth_addr_is_broadcast(flow->dl_dst)
7525 && (flow->nw_proto == ARP_OP_REPLY
7526 || (flow->nw_proto == ARP_OP_REQUEST
7527 && flow->nw_src == flow->nw_dst)));
7531 update_learning_table(struct ofproto_dpif *ofproto,
7532 const struct flow *flow, int vlan,
7533 struct ofbundle *in_bundle)
7535 struct mac_entry *mac;
7537 /* Don't learn the OFPP_NONE port. */
7538 if (in_bundle == &ofpp_none_bundle) {
7542 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7546 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7547 if (is_gratuitous_arp(flow)) {
7548 /* We don't want to learn from gratuitous ARP packets that are
7549 * reflected back over bond slaves so we lock the learning table. */
7550 if (!in_bundle->bond) {
7551 mac_entry_set_grat_arp_lock(mac);
7552 } else if (mac_entry_is_grat_arp_locked(mac)) {
7557 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7558 /* The log messages here could actually be useful in debugging,
7559 * so keep the rate limit relatively high. */
7560 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7561 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7562 "on port %s in VLAN %d",
7563 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7564 in_bundle->name, vlan);
7566 mac->port.p = in_bundle;
7567 tag_set_add(&ofproto->backer->revalidate_set,
7568 mac_learning_changed(ofproto->ml, mac));
7572 static struct ofbundle *
7573 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7574 bool warn, struct ofport_dpif **in_ofportp)
7576 struct ofport_dpif *ofport;
7578 /* Find the port and bundle for the received packet. */
7579 ofport = get_ofp_port(ofproto, in_port);
7581 *in_ofportp = ofport;
7583 if (ofport && ofport->bundle) {
7584 return ofport->bundle;
7587 /* Special-case OFPP_NONE, which a controller may use as the ingress
7588 * port for traffic that it is sourcing. */
7589 if (in_port == OFPP_NONE) {
7590 return &ofpp_none_bundle;
7593 /* Odd. A few possible reasons here:
7595 * - We deleted a port but there are still a few packets queued up
7598 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7599 * we don't know about.
7601 * - The ofproto client didn't configure the port as part of a bundle.
7602 * This is particularly likely to happen if a packet was received on the
7603 * port after it was created, but before the client had a chance to
7604 * configure its bundle.
7607 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7609 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7610 "port %"PRIu16, ofproto->up.name, in_port);
7615 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7616 * dropped. Returns true if they may be forwarded, false if they should be
7619 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7620 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7622 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7623 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7624 * checked by input_vid_is_valid().
7626 * May also add tags to '*tags', although the current implementation only does
7627 * so in one special case.
7630 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7633 struct ofproto_dpif *ofproto = ctx->ofproto;
7634 struct flow *flow = &ctx->flow;
7635 struct ofbundle *in_bundle = in_port->bundle;
7637 /* Drop frames for reserved multicast addresses
7638 * only if forward_bpdu option is absent. */
7639 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7640 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7644 if (in_bundle->bond) {
7645 struct mac_entry *mac;
7647 switch (bond_check_admissibility(in_bundle->bond, in_port,
7648 flow->dl_dst, &ctx->tags)) {
7653 xlate_report(ctx, "bonding refused admissibility, dropping");
7656 case BV_DROP_IF_MOVED:
7657 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7658 if (mac && mac->port.p != in_bundle &&
7659 (!is_gratuitous_arp(flow)
7660 || mac_entry_is_grat_arp_locked(mac))) {
7661 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7673 xlate_normal(struct action_xlate_ctx *ctx)
7675 struct ofport_dpif *in_port;
7676 struct ofbundle *in_bundle;
7677 struct mac_entry *mac;
7681 ctx->has_normal = true;
7683 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7684 ctx->packet != NULL, &in_port);
7686 xlate_report(ctx, "no input bundle, dropping");
7690 /* Drop malformed frames. */
7691 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7692 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7693 if (ctx->packet != NULL) {
7694 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7695 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7696 "VLAN tag received on port %s",
7697 ctx->ofproto->up.name, in_bundle->name);
7699 xlate_report(ctx, "partial VLAN tag, dropping");
7703 /* Drop frames on bundles reserved for mirroring. */
7704 if (in_bundle->mirror_out) {
7705 if (ctx->packet != NULL) {
7706 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7707 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7708 "%s, which is reserved exclusively for mirroring",
7709 ctx->ofproto->up.name, in_bundle->name);
7711 xlate_report(ctx, "input port is mirror output port, dropping");
7716 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7717 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7718 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7721 vlan = input_vid_to_vlan(in_bundle, vid);
7723 /* Check other admissibility requirements. */
7724 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7728 /* Learn source MAC. */
7729 if (ctx->may_learn) {
7730 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7733 /* Determine output bundle. */
7734 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7737 if (mac->port.p != in_bundle) {
7738 xlate_report(ctx, "forwarding to learned port");
7739 output_normal(ctx, mac->port.p, vlan);
7741 xlate_report(ctx, "learned port is input port, dropping");
7744 struct ofbundle *bundle;
7746 xlate_report(ctx, "no learned MAC for destination, flooding");
7747 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7748 if (bundle != in_bundle
7749 && ofbundle_includes_vlan(bundle, vlan)
7750 && bundle->floodable
7751 && !bundle->mirror_out) {
7752 output_normal(ctx, bundle, vlan);
7755 ctx->nf_output_iface = NF_OUT_FLOOD;
7759 /* Optimized flow revalidation.
7761 * It's a difficult problem, in general, to tell which facets need to have
7762 * their actions recalculated whenever the OpenFlow flow table changes. We
7763 * don't try to solve that general problem: for most kinds of OpenFlow flow
7764 * table changes, we recalculate the actions for every facet. This is
7765 * relatively expensive, but it's good enough if the OpenFlow flow table
7766 * doesn't change very often.
7768 * However, we can expect one particular kind of OpenFlow flow table change to
7769 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7770 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7771 * table, we add a special case that applies to flow tables in which every rule
7772 * has the same form (that is, the same wildcards), except that the table is
7773 * also allowed to have a single "catch-all" flow that matches all packets. We
7774 * optimize this case by tagging all of the facets that resubmit into the table
7775 * and invalidating the same tag whenever a flow changes in that table. The
7776 * end result is that we revalidate just the facets that need it (and sometimes
7777 * a few more, but not all of the facets or even all of the facets that
7778 * resubmit to the table modified by MAC learning). */
7780 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7781 * into an OpenFlow table with the given 'basis'. */
7783 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7786 if (minimask_is_catchall(mask)) {
7789 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7790 return tag_create_deterministic(hash);
7794 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7795 * taggability of that table.
7797 * This function must be called after *each* change to a flow table. If you
7798 * skip calling it on some changes then the pointer comparisons at the end can
7799 * be invalid if you get unlucky. For example, if a flow removal causes a
7800 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7801 * different wildcards to be created with the same address, then this function
7802 * will incorrectly skip revalidation. */
7804 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7806 struct table_dpif *table = &ofproto->tables[table_id];
7807 const struct oftable *oftable = &ofproto->up.tables[table_id];
7808 struct cls_table *catchall, *other;
7809 struct cls_table *t;
7811 catchall = other = NULL;
7813 switch (hmap_count(&oftable->cls.tables)) {
7815 /* We could tag this OpenFlow table but it would make the logic a
7816 * little harder and it's a corner case that doesn't seem worth it
7822 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7823 if (cls_table_is_catchall(t)) {
7825 } else if (!other) {
7828 /* Indicate that we can't tag this by setting both tables to
7829 * NULL. (We know that 'catchall' is already NULL.) */
7836 /* Can't tag this table. */
7840 if (table->catchall_table != catchall || table->other_table != other) {
7841 table->catchall_table = catchall;
7842 table->other_table = other;
7843 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7847 /* Given 'rule' that has changed in some way (either it is a rule being
7848 * inserted, a rule being deleted, or a rule whose actions are being
7849 * modified), marks facets for revalidation to ensure that packets will be
7850 * forwarded correctly according to the new state of the flow table.
7852 * This function must be called after *each* change to a flow table. See
7853 * the comment on table_update_taggable() for more information. */
7855 rule_invalidate(const struct rule_dpif *rule)
7857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7859 table_update_taggable(ofproto, rule->up.table_id);
7861 if (!ofproto->backer->need_revalidate) {
7862 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7864 if (table->other_table && rule->tag) {
7865 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7867 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7873 set_frag_handling(struct ofproto *ofproto_,
7874 enum ofp_config_flags frag_handling)
7876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7877 if (frag_handling != OFPC_FRAG_REASM) {
7878 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7886 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7887 const struct flow *flow,
7888 const struct ofpact *ofpacts, size_t ofpacts_len)
7890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7891 struct initial_vals initial_vals;
7892 struct odputil_keybuf keybuf;
7893 struct dpif_flow_stats stats;
7897 struct action_xlate_ctx ctx;
7898 uint64_t odp_actions_stub[1024 / 8];
7899 struct ofpbuf odp_actions;
7901 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7902 odp_flow_key_from_flow(&key, flow,
7903 ofp_port_to_odp_port(ofproto, flow->in_port));
7905 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7907 initial_vals.vlan_tci = flow->vlan_tci;
7908 initial_vals.tunnel_ip_tos = 0;
7909 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7910 packet_get_tcp_flags(packet, flow), packet);
7911 ctx.resubmit_stats = &stats;
7913 ofpbuf_use_stub(&odp_actions,
7914 odp_actions_stub, sizeof odp_actions_stub);
7915 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7916 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7917 odp_actions.data, odp_actions.size, packet);
7918 ofpbuf_uninit(&odp_actions);
7926 set_netflow(struct ofproto *ofproto_,
7927 const struct netflow_options *netflow_options)
7929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7931 if (netflow_options) {
7932 if (!ofproto->netflow) {
7933 ofproto->netflow = netflow_create();
7935 return netflow_set_options(ofproto->netflow, netflow_options);
7937 netflow_destroy(ofproto->netflow);
7938 ofproto->netflow = NULL;
7944 get_netflow_ids(const struct ofproto *ofproto_,
7945 uint8_t *engine_type, uint8_t *engine_id)
7947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7949 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7953 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7955 if (!facet_is_controller_flow(facet) &&
7956 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7957 struct subfacet *subfacet;
7958 struct ofexpired expired;
7960 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7961 if (subfacet->path == SF_FAST_PATH) {
7962 struct dpif_flow_stats stats;
7964 subfacet_reinstall(subfacet, &stats);
7965 subfacet_update_stats(subfacet, &stats);
7969 expired.flow = facet->flow;
7970 expired.packet_count = facet->packet_count;
7971 expired.byte_count = facet->byte_count;
7972 expired.used = facet->used;
7973 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7978 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7980 struct facet *facet;
7982 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7983 send_active_timeout(ofproto, facet);
7987 static struct ofproto_dpif *
7988 ofproto_dpif_lookup(const char *name)
7990 struct ofproto_dpif *ofproto;
7992 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7993 hash_string(name, 0), &all_ofproto_dpifs) {
7994 if (!strcmp(ofproto->up.name, name)) {
8002 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8003 const char *argv[], void *aux OVS_UNUSED)
8005 struct ofproto_dpif *ofproto;
8008 ofproto = ofproto_dpif_lookup(argv[1]);
8010 unixctl_command_reply_error(conn, "no such bridge");
8013 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8015 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8016 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8020 unixctl_command_reply(conn, "table successfully flushed");
8024 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8025 const char *argv[], void *aux OVS_UNUSED)
8027 struct ds ds = DS_EMPTY_INITIALIZER;
8028 const struct ofproto_dpif *ofproto;
8029 const struct mac_entry *e;
8031 ofproto = ofproto_dpif_lookup(argv[1]);
8033 unixctl_command_reply_error(conn, "no such bridge");
8037 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8038 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8039 struct ofbundle *bundle = e->port.p;
8040 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8041 ofbundle_get_a_port(bundle)->odp_port,
8042 e->vlan, ETH_ADDR_ARGS(e->mac),
8043 mac_entry_age(ofproto->ml, e));
8045 unixctl_command_reply(conn, ds_cstr(&ds));
8050 struct action_xlate_ctx ctx;
8056 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8057 const struct rule_dpif *rule)
8059 ds_put_char_multiple(result, '\t', level);
8061 ds_put_cstr(result, "No match\n");
8065 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8066 table_id, ntohll(rule->up.flow_cookie));
8067 cls_rule_format(&rule->up.cr, result);
8068 ds_put_char(result, '\n');
8070 ds_put_char_multiple(result, '\t', level);
8071 ds_put_cstr(result, "OpenFlow ");
8072 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8073 ds_put_char(result, '\n');
8077 trace_format_flow(struct ds *result, int level, const char *title,
8078 struct trace_ctx *trace)
8080 ds_put_char_multiple(result, '\t', level);
8081 ds_put_format(result, "%s: ", title);
8082 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8083 ds_put_cstr(result, "unchanged");
8085 flow_format(result, &trace->ctx.flow);
8086 trace->flow = trace->ctx.flow;
8088 ds_put_char(result, '\n');
8092 trace_format_regs(struct ds *result, int level, const char *title,
8093 struct trace_ctx *trace)
8097 ds_put_char_multiple(result, '\t', level);
8098 ds_put_format(result, "%s:", title);
8099 for (i = 0; i < FLOW_N_REGS; i++) {
8100 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8102 ds_put_char(result, '\n');
8106 trace_format_odp(struct ds *result, int level, const char *title,
8107 struct trace_ctx *trace)
8109 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8111 ds_put_char_multiple(result, '\t', level);
8112 ds_put_format(result, "%s: ", title);
8113 format_odp_actions(result, odp_actions->data, odp_actions->size);
8114 ds_put_char(result, '\n');
8118 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8120 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8121 struct ds *result = trace->result;
8123 ds_put_char(result, '\n');
8124 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8125 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8126 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8127 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8131 trace_report(struct action_xlate_ctx *ctx, const char *s)
8133 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8134 struct ds *result = trace->result;
8136 ds_put_char_multiple(result, '\t', ctx->recurse);
8137 ds_put_cstr(result, s);
8138 ds_put_char(result, '\n');
8142 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8143 void *aux OVS_UNUSED)
8145 const char *dpname = argv[1];
8146 struct ofproto_dpif *ofproto;
8147 struct ofpbuf odp_key;
8148 struct ofpbuf *packet;
8149 struct initial_vals initial_vals;
8155 ofpbuf_init(&odp_key, 0);
8158 ofproto = ofproto_dpif_lookup(dpname);
8160 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8164 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8165 /* ofproto/trace dpname flow [-generate] */
8166 const char *flow_s = argv[2];
8167 const char *generate_s = argv[3];
8169 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8170 * flow. We guess which type it is based on whether 'flow_s' contains
8171 * an '(', since a datapath flow always contains '(') but an
8172 * OpenFlow-like flow should not (in fact it's allowed but I believe
8173 * that's not documented anywhere).
8175 * An alternative would be to try to parse 'flow_s' both ways, but then
8176 * it would be tricky giving a sensible error message. After all, do
8177 * you just say "syntax error" or do you present both error messages?
8178 * Both choices seem lousy. */
8179 if (strchr(flow_s, '(')) {
8182 /* Convert string to datapath key. */
8183 ofpbuf_init(&odp_key, 0);
8184 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8186 unixctl_command_reply_error(conn, "Bad flow syntax");
8190 /* The user might have specified the wrong ofproto but within the
8191 * same backer. That's OK, ofproto_receive() can find the right
8193 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8194 odp_key.size, &flow, NULL, &ofproto, NULL,
8196 unixctl_command_reply_error(conn, "Invalid flow");
8199 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8203 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8205 unixctl_command_reply_error(conn, error_s);
8210 initial_vals.vlan_tci = flow.vlan_tci;
8211 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8214 /* Generate a packet, if requested. */
8216 packet = ofpbuf_new(0);
8217 flow_compose(packet, &flow);
8219 } else if (argc == 7) {
8220 /* ofproto/trace dpname priority tun_id in_port mark packet */
8221 const char *priority_s = argv[2];
8222 const char *tun_id_s = argv[3];
8223 const char *in_port_s = argv[4];
8224 const char *mark_s = argv[5];
8225 const char *packet_s = argv[6];
8226 uint32_t in_port = atoi(in_port_s);
8227 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8228 uint32_t priority = atoi(priority_s);
8229 uint32_t mark = atoi(mark_s);
8232 msg = eth_from_hex(packet_s, &packet);
8234 unixctl_command_reply_error(conn, msg);
8238 ds_put_cstr(&result, "Packet: ");
8239 s = ofp_packet_to_string(packet->data, packet->size);
8240 ds_put_cstr(&result, s);
8243 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8244 flow.tunnel.tun_id = tun_id;
8245 initial_vals.vlan_tci = flow.vlan_tci;
8246 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8248 unixctl_command_reply_error(conn, "Bad command syntax");
8252 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8253 unixctl_command_reply(conn, ds_cstr(&result));
8256 ds_destroy(&result);
8257 ofpbuf_delete(packet);
8258 ofpbuf_uninit(&odp_key);
8262 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8263 const struct ofpbuf *packet,
8264 const struct initial_vals *initial_vals, struct ds *ds)
8266 struct rule_dpif *rule;
8268 ds_put_cstr(ds, "Flow: ");
8269 flow_format(ds, flow);
8270 ds_put_char(ds, '\n');
8272 rule = rule_dpif_lookup(ofproto, flow);
8274 trace_format_rule(ds, 0, 0, rule);
8275 if (rule == ofproto->miss_rule) {
8276 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8277 } else if (rule == ofproto->no_packet_in_rule) {
8278 ds_put_cstr(ds, "\nNo match, packets dropped because "
8279 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8283 uint64_t odp_actions_stub[1024 / 8];
8284 struct ofpbuf odp_actions;
8286 struct trace_ctx trace;
8289 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8292 ofpbuf_use_stub(&odp_actions,
8293 odp_actions_stub, sizeof odp_actions_stub);
8294 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8295 rule, tcp_flags, packet);
8296 trace.ctx.resubmit_hook = trace_resubmit;
8297 trace.ctx.report_hook = trace_report;
8298 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8301 ds_put_char(ds, '\n');
8302 trace_format_flow(ds, 0, "Final flow", &trace);
8303 ds_put_cstr(ds, "Datapath actions: ");
8304 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8305 ofpbuf_uninit(&odp_actions);
8307 if (trace.ctx.slow) {
8308 enum slow_path_reason slow;
8310 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8311 "slow path because it:");
8312 for (slow = trace.ctx.slow; slow; ) {
8313 enum slow_path_reason bit = rightmost_1bit(slow);
8317 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8320 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8323 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8326 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8329 ds_put_cstr(ds, "\n\t (The datapath actions are "
8330 "incomplete--for complete actions, "
8331 "please supply a packet.)");
8334 case SLOW_CONTROLLER:
8335 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8336 "to the OpenFlow controller.");
8339 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8340 "than the datapath supports.");
8347 if (slow & ~SLOW_MATCH) {
8348 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8349 "the special slow-path processing.");
8356 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8357 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8360 unixctl_command_reply(conn, NULL);
8364 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8365 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8368 unixctl_command_reply(conn, NULL);
8371 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8372 * 'reply' describing the results. */
8374 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8376 struct facet *facet;
8380 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8381 if (!facet_check_consistency(facet)) {
8386 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8390 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8391 ofproto->up.name, errors);
8393 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8398 ofproto_dpif_self_check(struct unixctl_conn *conn,
8399 int argc, const char *argv[], void *aux OVS_UNUSED)
8401 struct ds reply = DS_EMPTY_INITIALIZER;
8402 struct ofproto_dpif *ofproto;
8405 ofproto = ofproto_dpif_lookup(argv[1]);
8407 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8408 "ofproto/list for help)");
8411 ofproto_dpif_self_check__(ofproto, &reply);
8413 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8414 ofproto_dpif_self_check__(ofproto, &reply);
8418 unixctl_command_reply(conn, ds_cstr(&reply));
8422 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8423 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8424 * to destroy 'ofproto_shash' and free the returned value. */
8425 static const struct shash_node **
8426 get_ofprotos(struct shash *ofproto_shash)
8428 const struct ofproto_dpif *ofproto;
8430 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8431 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8432 shash_add_nocopy(ofproto_shash, name, ofproto);
8435 return shash_sort(ofproto_shash);
8439 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8440 const char *argv[] OVS_UNUSED,
8441 void *aux OVS_UNUSED)
8443 struct ds ds = DS_EMPTY_INITIALIZER;
8444 struct shash ofproto_shash;
8445 const struct shash_node **sorted_ofprotos;
8448 shash_init(&ofproto_shash);
8449 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8450 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8451 const struct shash_node *node = sorted_ofprotos[i];
8452 ds_put_format(&ds, "%s\n", node->name);
8455 shash_destroy(&ofproto_shash);
8456 free(sorted_ofprotos);
8458 unixctl_command_reply(conn, ds_cstr(&ds));
8463 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8465 const struct shash_node **ports;
8467 struct avg_subfacet_rates lifetime;
8468 unsigned long long int minutes;
8469 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8471 minutes = (time_msec() - ofproto->created) / min_ms;
8474 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8476 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8479 lifetime.add_rate = 0.0;
8480 lifetime.del_rate = 0.0;
8483 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8484 dpif_name(ofproto->backer->dpif));
8486 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8487 ofproto->n_hit, ofproto->n_missed);
8488 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8489 " life span: %llu(ms)\n",
8490 hmap_count(&ofproto->subfacets),
8491 avg_subfacet_count(ofproto),
8492 ofproto->max_n_subfacet,
8493 avg_subfacet_life_span(ofproto));
8494 if (minutes >= 60) {
8495 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8497 if (minutes >= 60 * 24) {
8498 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8500 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8502 ports = shash_sort(&ofproto->up.port_by_name);
8503 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8504 const struct shash_node *node = ports[i];
8505 struct ofport *ofport = node->data;
8506 const char *name = netdev_get_name(ofport->netdev);
8507 const char *type = netdev_get_type(ofport->netdev);
8510 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8512 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8513 if (odp_port != OVSP_NONE) {
8514 ds_put_format(ds, "%"PRIu32":", odp_port);
8516 ds_put_cstr(ds, "none:");
8519 if (strcmp(type, "system")) {
8520 struct netdev *netdev;
8523 ds_put_format(ds, " (%s", type);
8525 error = netdev_open(name, type, &netdev);
8530 error = netdev_get_config(netdev, &config);
8532 const struct smap_node **nodes;
8535 nodes = smap_sort(&config);
8536 for (i = 0; i < smap_count(&config); i++) {
8537 const struct smap_node *node = nodes[i];
8538 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8539 node->key, node->value);
8543 smap_destroy(&config);
8545 netdev_close(netdev);
8547 ds_put_char(ds, ')');
8549 ds_put_char(ds, '\n');
8555 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8556 const char *argv[], void *aux OVS_UNUSED)
8558 struct ds ds = DS_EMPTY_INITIALIZER;
8559 const struct ofproto_dpif *ofproto;
8563 for (i = 1; i < argc; i++) {
8564 ofproto = ofproto_dpif_lookup(argv[i]);
8566 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8567 "for help)", argv[i]);
8568 unixctl_command_reply_error(conn, ds_cstr(&ds));
8571 show_dp_format(ofproto, &ds);
8574 struct shash ofproto_shash;
8575 const struct shash_node **sorted_ofprotos;
8578 shash_init(&ofproto_shash);
8579 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8580 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8581 const struct shash_node *node = sorted_ofprotos[i];
8582 show_dp_format(node->data, &ds);
8585 shash_destroy(&ofproto_shash);
8586 free(sorted_ofprotos);
8589 unixctl_command_reply(conn, ds_cstr(&ds));
8594 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8595 int argc OVS_UNUSED, const char *argv[],
8596 void *aux OVS_UNUSED)
8598 struct ds ds = DS_EMPTY_INITIALIZER;
8599 const struct ofproto_dpif *ofproto;
8600 struct subfacet *subfacet;
8602 ofproto = ofproto_dpif_lookup(argv[1]);
8604 unixctl_command_reply_error(conn, "no such bridge");
8608 update_stats(ofproto->backer);
8610 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8611 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8613 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8614 subfacet->dp_packet_count, subfacet->dp_byte_count);
8615 if (subfacet->used) {
8616 ds_put_format(&ds, "%.3fs",
8617 (time_msec() - subfacet->used) / 1000.0);
8619 ds_put_format(&ds, "never");
8621 if (subfacet->facet->tcp_flags) {
8622 ds_put_cstr(&ds, ", flags:");
8623 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8626 ds_put_cstr(&ds, ", actions:");
8627 if (subfacet->slow) {
8628 uint64_t slow_path_stub[128 / 8];
8629 const struct nlattr *actions;
8632 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8633 slow_path_stub, sizeof slow_path_stub,
8634 &actions, &actions_len);
8635 format_odp_actions(&ds, actions, actions_len);
8637 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8639 ds_put_char(&ds, '\n');
8642 unixctl_command_reply(conn, ds_cstr(&ds));
8647 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8648 int argc OVS_UNUSED, const char *argv[],
8649 void *aux OVS_UNUSED)
8651 struct ds ds = DS_EMPTY_INITIALIZER;
8652 struct ofproto_dpif *ofproto;
8654 ofproto = ofproto_dpif_lookup(argv[1]);
8656 unixctl_command_reply_error(conn, "no such bridge");
8660 flush(&ofproto->up);
8662 unixctl_command_reply(conn, ds_cstr(&ds));
8667 ofproto_dpif_unixctl_init(void)
8669 static bool registered;
8675 unixctl_command_register(
8677 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8678 2, 6, ofproto_unixctl_trace, NULL);
8679 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8680 ofproto_unixctl_fdb_flush, NULL);
8681 unixctl_command_register("fdb/show", "bridge", 1, 1,
8682 ofproto_unixctl_fdb_show, NULL);
8683 unixctl_command_register("ofproto/clog", "", 0, 0,
8684 ofproto_dpif_clog, NULL);
8685 unixctl_command_register("ofproto/unclog", "", 0, 0,
8686 ofproto_dpif_unclog, NULL);
8687 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8688 ofproto_dpif_self_check, NULL);
8689 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8690 ofproto_unixctl_dpif_dump_dps, NULL);
8691 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8692 ofproto_unixctl_dpif_show, NULL);
8693 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8694 ofproto_unixctl_dpif_dump_flows, NULL);
8695 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8696 ofproto_unixctl_dpif_del_flows, NULL);
8699 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8701 * This is deprecated. It is only for compatibility with broken device drivers
8702 * in old versions of Linux that do not properly support VLANs when VLAN
8703 * devices are not used. When broken device drivers are no longer in
8704 * widespread use, we will delete these interfaces. */
8707 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8710 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8712 if (realdev_ofp_port == ofport->realdev_ofp_port
8713 && vid == ofport->vlandev_vid) {
8717 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8719 if (ofport->realdev_ofp_port) {
8722 if (realdev_ofp_port && ofport->bundle) {
8723 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8724 * themselves be part of a bundle. */
8725 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8728 ofport->realdev_ofp_port = realdev_ofp_port;
8729 ofport->vlandev_vid = vid;
8731 if (realdev_ofp_port) {
8732 vsp_add(ofport, realdev_ofp_port, vid);
8739 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8741 return hash_2words(realdev_ofp_port, vid);
8744 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8745 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8746 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8747 * it would return the port number of eth0.9.
8749 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8750 * function just returns its 'realdev_odp_port' argument. */
8752 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8753 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8755 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8756 uint16_t realdev_ofp_port;
8757 int vid = vlan_tci_to_vid(vlan_tci);
8758 const struct vlan_splinter *vsp;
8760 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8761 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8762 hash_realdev_vid(realdev_ofp_port, vid),
8763 &ofproto->realdev_vid_map) {
8764 if (vsp->realdev_ofp_port == realdev_ofp_port
8765 && vsp->vid == vid) {
8766 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8770 return realdev_odp_port;
8773 static struct vlan_splinter *
8774 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8776 struct vlan_splinter *vsp;
8778 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8779 &ofproto->vlandev_map) {
8780 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8788 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8789 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8790 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8791 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8792 * eth0 and store 9 in '*vid'.
8794 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8795 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8798 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8799 uint16_t vlandev_ofp_port, int *vid)
8801 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8802 const struct vlan_splinter *vsp;
8804 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8809 return vsp->realdev_ofp_port;
8815 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8816 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8817 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8818 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8819 * always the case unless VLAN splinters are enabled), returns false without
8820 * making any changes. */
8822 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8827 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8832 /* Cause the flow to be processed as if it came in on the real device with
8833 * the VLAN device's VLAN ID. */
8834 flow->in_port = realdev;
8835 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8840 vsp_remove(struct ofport_dpif *port)
8842 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8843 struct vlan_splinter *vsp;
8845 vsp = vlandev_find(ofproto, port->up.ofp_port);
8847 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8848 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8851 port->realdev_ofp_port = 0;
8853 VLOG_ERR("missing vlan device record");
8858 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8862 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8863 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8864 == realdev_ofp_port)) {
8865 struct vlan_splinter *vsp;
8867 vsp = xmalloc(sizeof *vsp);
8868 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8869 hash_int(port->up.ofp_port, 0));
8870 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8871 hash_realdev_vid(realdev_ofp_port, vid));
8872 vsp->realdev_ofp_port = realdev_ofp_port;
8873 vsp->vlandev_ofp_port = port->up.ofp_port;
8876 port->realdev_ofp_port = realdev_ofp_port;
8878 VLOG_ERR("duplicate vlan device record");
8883 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8885 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8886 return ofport ? ofport->odp_port : OVSP_NONE;
8889 static struct ofport_dpif *
8890 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8892 struct ofport_dpif *port;
8894 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8895 hash_int(odp_port, 0),
8896 &backer->odp_to_ofport_map) {
8897 if (port->odp_port == odp_port) {
8906 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8908 struct ofport_dpif *port;
8910 port = odp_port_to_ofport(ofproto->backer, odp_port);
8911 if (port && &ofproto->up == port->up.ofproto) {
8912 return port->up.ofp_port;
8917 static unsigned long long int
8918 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8920 unsigned long long int dc;
8921 unsigned long long int avg;
8923 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8924 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8930 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8934 if (ofproto->n_update_stats) {
8935 avg_c = (double)ofproto->total_subfacet_count
8936 / ofproto->n_update_stats;
8943 show_dp_rates(struct ds *ds, const char *heading,
8944 const struct avg_subfacet_rates *rates)
8946 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8947 heading, rates->add_rate, rates->del_rate);
8951 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8953 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8954 hmap_count(&ofproto->subfacets));
8957 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8958 * most heavily weighted element. 'base' designates the rate of decay: after
8959 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8962 exp_mavg(double *avg, int base, double new)
8964 *avg = (*avg * (base - 1) + new) / base;
8968 update_moving_averages(struct ofproto_dpif *ofproto)
8970 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8972 /* Update hourly averages on the minute boundaries. */
8973 if (time_msec() - ofproto->last_minute >= min_ms) {
8974 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8975 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8977 /* Update daily averages on the hour boundaries. */
8978 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8979 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8980 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8983 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8984 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8985 ofproto->subfacet_add_count = 0;
8986 ofproto->subfacet_del_count = 0;
8987 ofproto->last_minute += min_ms;
8992 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8994 ofproto->n_hit += delta;
8997 const struct ofproto_class ofproto_dpif_class = {
9032 port_is_lacp_current,
9033 NULL, /* rule_choose_table */
9040 rule_modify_actions,
9052 get_stp_port_status,
9059 is_mirror_output_bundle,
9060 forward_bpdu_changed,
9061 set_mac_table_config,