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 static int subfacet_install(struct subfacet *,
420 const struct nlattr *actions, size_t actions_len,
421 struct dpif_flow_stats *, enum slow_path_reason);
422 static void subfacet_uninstall(struct subfacet *);
424 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
426 /* An exact-match instantiation of an OpenFlow flow.
428 * A facet associates a "struct flow", which represents the Open vSwitch
429 * userspace idea of an exact-match flow, with one or more subfacets. Each
430 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
431 * the facet. When the kernel module (or other dpif implementation) and Open
432 * vSwitch userspace agree on the definition of a flow key, there is exactly
433 * one subfacet per facet. If the dpif implementation supports more-specific
434 * flow matching than userspace, however, a facet can have more than one
435 * subfacet, each of which corresponds to some distinction in flow that
436 * userspace simply doesn't understand.
438 * Flow expiration works in terms of subfacets, so a facet must have at least
439 * one subfacet or it will never expire, leaking memory. */
442 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
443 struct list list_node; /* In owning rule's 'facets' list. */
444 struct rule_dpif *rule; /* Owning rule. */
447 struct list subfacets;
448 long long int used; /* Time last used; time created if not used. */
455 * - Do include packets and bytes sent "by hand", e.g. with
458 * - Do include packets and bytes that were obtained from the datapath
459 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
460 * DPIF_FP_ZERO_STATS).
462 * - Do not include packets or bytes that can be obtained from the
463 * datapath for any existing subfacet.
465 uint64_t packet_count; /* Number of packets received. */
466 uint64_t byte_count; /* Number of bytes received. */
468 /* Resubmit statistics. */
469 uint64_t prev_packet_count; /* Number of packets from last stats push. */
470 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
471 long long int prev_used; /* Used time from last stats push. */
474 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
475 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
476 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
478 /* Properties of datapath actions.
480 * Every subfacet has its own actions because actions can differ slightly
481 * between splintered and non-splintered subfacets due to the VLAN tag
482 * being initially different (present vs. absent). All of them have these
483 * properties in common so we just store one copy of them here. */
484 bool has_learn; /* Actions include NXAST_LEARN? */
485 bool has_normal; /* Actions output to OFPP_NORMAL? */
486 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
487 tag_type tags; /* Tags that would require revalidation. */
488 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
490 /* Storage for a single subfacet, to reduce malloc() time and space
491 * overhead. (A facet always has at least one subfacet and in the common
492 * case has exactly one subfacet. However, 'one_subfacet' may not
493 * always be valid, since it could have been removed after newer
494 * subfacets were pushed onto the 'subfacets' list.) */
495 struct subfacet one_subfacet;
497 long long int learn_rl; /* Rate limiter for facet_learn(). */
500 static struct facet *facet_create(struct rule_dpif *,
501 const struct flow *, uint32_t hash);
502 static void facet_remove(struct facet *);
503 static void facet_free(struct facet *);
505 static struct facet *facet_find(struct ofproto_dpif *,
506 const struct flow *, uint32_t hash);
507 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
508 const struct flow *, uint32_t hash);
509 static bool facet_revalidate(struct facet *);
510 static bool facet_check_consistency(struct facet *);
512 static void facet_flush_stats(struct facet *);
514 static void facet_update_time(struct facet *, long long int used);
515 static void facet_reset_counters(struct facet *);
516 static void facet_push_stats(struct facet *);
517 static void facet_learn(struct facet *);
518 static void facet_account(struct facet *);
519 static void push_all_stats(void);
521 static struct subfacet *facet_get_subfacet(struct facet *);
523 static bool facet_is_controller_flow(struct facet *);
526 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
530 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
531 struct list bundle_node; /* In struct ofbundle's "ports" list. */
532 struct cfm *cfm; /* Connectivity Fault Management, if any. */
533 tag_type tag; /* Tag associated with this port. */
534 bool may_enable; /* May be enabled in bonds. */
535 long long int carrier_seq; /* Carrier status changes. */
536 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
539 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
540 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
541 long long int stp_state_entered;
543 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
545 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
547 * This is deprecated. It is only for compatibility with broken device
548 * drivers in old versions of Linux that do not properly support VLANs when
549 * VLAN devices are not used. When broken device drivers are no longer in
550 * widespread use, we will delete these interfaces. */
551 uint16_t realdev_ofp_port;
555 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
556 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
557 * traffic egressing the 'ofport' with that priority should be marked with. */
558 struct priority_to_dscp {
559 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
560 uint32_t priority; /* Priority of this queue (see struct flow). */
562 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
565 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
567 * This is deprecated. It is only for compatibility with broken device drivers
568 * in old versions of Linux that do not properly support VLANs when VLAN
569 * devices are not used. When broken device drivers are no longer in
570 * widespread use, we will delete these interfaces. */
571 struct vlan_splinter {
572 struct hmap_node realdev_vid_node;
573 struct hmap_node vlandev_node;
574 uint16_t realdev_ofp_port;
575 uint16_t vlandev_ofp_port;
579 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
580 uint32_t realdev, ovs_be16 vlan_tci);
581 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
582 static void vsp_remove(struct ofport_dpif *);
583 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
585 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
587 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
590 static struct ofport_dpif *
591 ofport_dpif_cast(const struct ofport *ofport)
593 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
594 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
597 static void port_run(struct ofport_dpif *);
598 static void port_run_fast(struct ofport_dpif *);
599 static void port_wait(struct ofport_dpif *);
600 static int set_cfm(struct ofport *, const struct cfm_settings *);
601 static void ofport_clear_priorities(struct ofport_dpif *);
602 static void run_fast_rl(void);
604 struct dpif_completion {
605 struct list list_node;
606 struct ofoperation *op;
609 /* Extra information about a classifier table.
610 * Currently used just for optimized flow revalidation. */
612 /* If either of these is nonnull, then this table has a form that allows
613 * flows to be tagged to avoid revalidating most flows for the most common
614 * kinds of flow table changes. */
615 struct cls_table *catchall_table; /* Table that wildcards all fields. */
616 struct cls_table *other_table; /* Table with any other wildcard set. */
617 uint32_t basis; /* Keeps each table's tags separate. */
620 /* Reasons that we might need to revalidate every facet, and corresponding
623 * A value of 0 means that there is no need to revalidate.
625 * It would be nice to have some cleaner way to integrate with coverage
626 * counters, but with only a few reasons I guess this is good enough for
628 enum revalidate_reason {
629 REV_RECONFIGURE = 1, /* Switch configuration changed. */
630 REV_STP, /* Spanning tree protocol port status change. */
631 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
632 REV_FLOW_TABLE, /* Flow table changed. */
633 REV_INCONSISTENCY /* Facet self-check failed. */
635 COVERAGE_DEFINE(rev_reconfigure);
636 COVERAGE_DEFINE(rev_stp);
637 COVERAGE_DEFINE(rev_port_toggled);
638 COVERAGE_DEFINE(rev_flow_table);
639 COVERAGE_DEFINE(rev_inconsistency);
641 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
642 * These are datapath flows which have no associated ofproto, if they did we
643 * would use facets. */
645 struct hmap_node hmap_node;
650 /* All datapaths of a given type share a single dpif backer instance. */
655 struct timer next_expiration;
656 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
658 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
660 /* Facet revalidation flags applying to facets which use this backer. */
661 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
662 struct tag_set revalidate_set; /* Revalidate only matching facets. */
664 struct hmap drop_keys; /* Set of dropped odp keys. */
667 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
668 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
670 static void drop_key_clear(struct dpif_backer *);
671 static struct ofport_dpif *
672 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
674 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
676 struct avg_subfacet_rates {
677 double add_rate; /* Moving average of new flows created per minute. */
678 double del_rate; /* Moving average of flows deleted per minute. */
680 static void show_dp_rates(struct ds *ds, const char *heading,
681 const struct avg_subfacet_rates *rates);
682 static void exp_mavg(double *avg, int base, double new);
684 struct ofproto_dpif {
685 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
687 struct dpif_backer *backer;
689 /* Special OpenFlow rules. */
690 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
691 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
697 struct netflow *netflow;
698 struct dpif_sflow *sflow;
699 struct dpif_ipfix *ipfix;
700 struct hmap bundles; /* Contains "struct ofbundle"s. */
701 struct mac_learning *ml;
702 struct ofmirror *mirrors[MAX_MIRRORS];
704 bool has_bonded_bundles;
708 struct hmap subfacets;
709 struct governor *governor;
710 long long int consistency_rl;
713 struct table_dpif tables[N_TABLES];
715 /* Support for debugging async flow mods. */
716 struct list completions;
718 bool has_bundle_action; /* True when the first bundle action appears. */
719 struct netdev_stats stats; /* To account packets generated and consumed in
724 long long int stp_last_tick;
726 /* VLAN splinters. */
727 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
728 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
731 struct sset ports; /* Set of standard port names. */
732 struct sset ghost_ports; /* Ports with no datapath port. */
733 struct sset port_poll_set; /* Queued names for port_poll() reply. */
734 int port_poll_errno; /* Last errno for port_poll() reply. */
736 /* Per ofproto's dpif stats. */
740 /* Subfacet statistics.
742 * These keep track of the total number of subfacets added and deleted and
743 * flow life span. They are useful for computing the flow rates stats
744 * exposed via "ovs-appctl dpif/show". The goal is to learn about
745 * traffic patterns in ways that we can use later to improve Open vSwitch
746 * performance in new situations. */
747 long long int created; /* Time when it is created. */
748 unsigned int max_n_subfacet; /* Maximum number of flows */
750 /* The average number of subfacets... */
751 struct avg_subfacet_rates hourly; /* ...over the last hour. */
752 struct avg_subfacet_rates daily; /* ...over the last day. */
753 long long int last_minute; /* Last time 'hourly' was updated. */
755 /* Number of subfacets added or deleted since 'last_minute'. */
756 unsigned int subfacet_add_count;
757 unsigned int subfacet_del_count;
759 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
760 unsigned long long int total_subfacet_add_count;
761 unsigned long long int total_subfacet_del_count;
763 /* Sum of the number of milliseconds that each subfacet existed,
764 * over the subfacets that have been added and then later deleted. */
765 unsigned long long int total_subfacet_life_span;
767 /* Incremented by the number of currently existing subfacets, each
768 * time we pull statistics from the kernel. */
769 unsigned long long int total_subfacet_count;
771 /* Number of times we pull statistics from the kernel. */
772 unsigned long long int n_update_stats;
774 static unsigned long long int avg_subfacet_life_span(
775 const struct ofproto_dpif *);
776 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
777 static void update_moving_averages(struct ofproto_dpif *ofproto);
778 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
780 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
782 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
783 * for debugging the asynchronous flow_mod implementation.) */
786 /* All existing ofproto_dpif instances, indexed by ->up.name. */
787 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
789 static void ofproto_dpif_unixctl_init(void);
791 static struct ofproto_dpif *
792 ofproto_dpif_cast(const struct ofproto *ofproto)
794 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
795 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
798 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
800 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
802 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
803 const struct ofpbuf *,
804 const struct initial_vals *, struct ds *);
806 /* Packet processing. */
807 static void update_learning_table(struct ofproto_dpif *,
808 const struct flow *, int vlan,
811 #define FLOW_MISS_MAX_BATCH 50
812 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
814 /* Flow expiration. */
815 static int expire(struct dpif_backer *);
818 static void send_netflow_active_timeouts(struct ofproto_dpif *);
821 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
822 static size_t compose_sflow_action(const struct ofproto_dpif *,
823 struct ofpbuf *odp_actions,
824 const struct flow *, uint32_t odp_port);
825 static void compose_ipfix_action(const struct ofproto_dpif *,
826 struct ofpbuf *odp_actions,
827 const struct flow *);
828 static void add_mirror_actions(struct action_xlate_ctx *ctx,
829 const struct flow *flow);
830 /* Global variables. */
831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
833 /* Initial mappings of port to bridge mappings. */
834 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
836 /* Factory functions. */
839 init(const struct shash *iface_hints)
841 struct shash_node *node;
843 /* Make a local copy, since we don't own 'iface_hints' elements. */
844 SHASH_FOR_EACH(node, iface_hints) {
845 const struct iface_hint *orig_hint = node->data;
846 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
848 new_hint->br_name = xstrdup(orig_hint->br_name);
849 new_hint->br_type = xstrdup(orig_hint->br_type);
850 new_hint->ofp_port = orig_hint->ofp_port;
852 shash_add(&init_ofp_ports, node->name, new_hint);
857 enumerate_types(struct sset *types)
859 dp_enumerate_types(types);
863 enumerate_names(const char *type, struct sset *names)
865 struct ofproto_dpif *ofproto;
868 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
869 if (strcmp(type, ofproto->up.type)) {
872 sset_add(names, ofproto->up.name);
879 del(const char *type, const char *name)
884 error = dpif_open(name, type, &dpif);
886 error = dpif_delete(dpif);
893 port_open_type(const char *datapath_type, const char *port_type)
895 return dpif_port_open_type(datapath_type, port_type);
898 /* Type functions. */
900 static struct ofproto_dpif *
901 lookup_ofproto_dpif_by_port_name(const char *name)
903 struct ofproto_dpif *ofproto;
905 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
906 if (sset_contains(&ofproto->ports, name)) {
915 type_run(const char *type)
917 static long long int push_timer = LLONG_MIN;
918 struct dpif_backer *backer;
922 backer = shash_find_data(&all_dpif_backers, type);
924 /* This is not necessarily a problem, since backers are only
925 * created on demand. */
929 dpif_run(backer->dpif);
931 /* The most natural place to push facet statistics is when they're pulled
932 * from the datapath. However, when there are many flows in the datapath,
933 * this expensive operation can occur so frequently, that it reduces our
934 * ability to quickly set up flows. To reduce the cost, we push statistics
936 if (time_msec() > push_timer) {
937 push_timer = time_msec() + 2000;
941 if (backer->need_revalidate
942 || !tag_set_is_empty(&backer->revalidate_set)) {
943 struct tag_set revalidate_set = backer->revalidate_set;
944 bool need_revalidate = backer->need_revalidate;
945 struct ofproto_dpif *ofproto;
946 struct simap_node *node;
947 struct simap tmp_backers;
949 /* Handle tunnel garbage collection. */
950 simap_init(&tmp_backers);
951 simap_swap(&backer->tnl_backers, &tmp_backers);
953 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
954 struct ofport_dpif *iter;
956 if (backer != ofproto->backer) {
960 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
963 if (!iter->tnl_port) {
967 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
968 node = simap_find(&tmp_backers, dp_port);
970 simap_put(&backer->tnl_backers, dp_port, node->data);
971 simap_delete(&tmp_backers, node);
972 node = simap_find(&backer->tnl_backers, dp_port);
974 node = simap_find(&backer->tnl_backers, dp_port);
976 uint32_t odp_port = UINT32_MAX;
978 if (!dpif_port_add(backer->dpif, iter->up.netdev,
980 simap_put(&backer->tnl_backers, dp_port, odp_port);
981 node = simap_find(&backer->tnl_backers, dp_port);
986 iter->odp_port = node ? node->data : OVSP_NONE;
987 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
989 backer->need_revalidate = REV_RECONFIGURE;
994 SIMAP_FOR_EACH (node, &tmp_backers) {
995 dpif_port_del(backer->dpif, node->data);
997 simap_destroy(&tmp_backers);
999 switch (backer->need_revalidate) {
1000 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1001 case REV_STP: COVERAGE_INC(rev_stp); break;
1002 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1003 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1004 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1007 if (backer->need_revalidate) {
1008 /* Clear the drop_keys in case we should now be accepting some
1009 * formerly dropped flows. */
1010 drop_key_clear(backer);
1013 /* Clear the revalidation flags. */
1014 tag_set_init(&backer->revalidate_set);
1015 backer->need_revalidate = 0;
1017 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1018 struct facet *facet, *next;
1020 if (ofproto->backer != backer) {
1024 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1026 || tag_set_intersects(&revalidate_set, facet->tags)) {
1027 facet_revalidate(facet);
1034 if (timer_expired(&backer->next_expiration)) {
1035 int delay = expire(backer);
1036 timer_set_duration(&backer->next_expiration, delay);
1039 /* Check for port changes in the dpif. */
1040 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1041 struct ofproto_dpif *ofproto;
1042 struct dpif_port port;
1044 /* Don't report on the datapath's device. */
1045 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1049 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1050 &all_ofproto_dpifs) {
1051 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1056 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1057 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1058 /* The port was removed. If we know the datapath,
1059 * report it through poll_set(). If we don't, it may be
1060 * notifying us of a removal we initiated, so ignore it.
1061 * If there's a pending ENOBUFS, let it stand, since
1062 * everything will be reevaluated. */
1063 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1064 sset_add(&ofproto->port_poll_set, devname);
1065 ofproto->port_poll_errno = 0;
1067 } else if (!ofproto) {
1068 /* The port was added, but we don't know with which
1069 * ofproto we should associate it. Delete it. */
1070 dpif_port_del(backer->dpif, port.port_no);
1072 dpif_port_destroy(&port);
1078 if (error != EAGAIN) {
1079 struct ofproto_dpif *ofproto;
1081 /* There was some sort of error, so propagate it to all
1082 * ofprotos that use this backer. */
1083 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1084 &all_ofproto_dpifs) {
1085 if (ofproto->backer == backer) {
1086 sset_clear(&ofproto->port_poll_set);
1087 ofproto->port_poll_errno = error;
1096 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1100 /* Handle one or more batches of upcalls, until there's nothing left to do
1101 * or until we do a fixed total amount of work.
1103 * We do work in batches because it can be much cheaper to set up a number
1104 * of flows and fire off their patches all at once. We do multiple batches
1105 * because in some cases handling a packet can cause another packet to be
1106 * queued almost immediately as part of the return flow. Both
1107 * optimizations can make major improvements on some benchmarks and
1108 * presumably for real traffic as well. */
1110 while (work < max_batch) {
1111 int retval = handle_upcalls(backer, max_batch - work);
1122 type_run_fast(const char *type)
1124 struct dpif_backer *backer;
1126 backer = shash_find_data(&all_dpif_backers, type);
1128 /* This is not necessarily a problem, since backers are only
1129 * created on demand. */
1133 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1139 static long long int port_rl = LLONG_MIN;
1140 static unsigned int backer_rl = 0;
1142 if (time_msec() >= port_rl) {
1143 struct ofproto_dpif *ofproto;
1144 struct ofport_dpif *ofport;
1146 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1148 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1149 port_run_fast(ofport);
1152 port_rl = time_msec() + 200;
1155 /* XXX: We have to be careful not to do too much work in this function. If
1156 * we call dpif_backer_run_fast() too often, or with too large a batch,
1157 * performance improves signifcantly, but at a cost. It's possible for the
1158 * number of flows in the datapath to increase without bound, and for poll
1159 * loops to take 10s of seconds. The correct solution to this problem,
1160 * long term, is to separate flow miss handling into it's own thread so it
1161 * isn't affected by revalidations, and expirations. Until then, this is
1162 * the best we can do. */
1163 if (++backer_rl >= 10) {
1164 struct shash_node *node;
1167 SHASH_FOR_EACH (node, &all_dpif_backers) {
1168 dpif_backer_run_fast(node->data, 1);
1174 type_wait(const char *type)
1176 struct dpif_backer *backer;
1178 backer = shash_find_data(&all_dpif_backers, type);
1180 /* This is not necessarily a problem, since backers are only
1181 * created on demand. */
1185 timer_wait(&backer->next_expiration);
1188 /* Basic life-cycle. */
1190 static int add_internal_flows(struct ofproto_dpif *);
1192 static struct ofproto *
1195 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1196 return &ofproto->up;
1200 dealloc(struct ofproto *ofproto_)
1202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1207 close_dpif_backer(struct dpif_backer *backer)
1209 struct shash_node *node;
1211 ovs_assert(backer->refcount > 0);
1213 if (--backer->refcount) {
1217 drop_key_clear(backer);
1218 hmap_destroy(&backer->drop_keys);
1220 simap_destroy(&backer->tnl_backers);
1221 hmap_destroy(&backer->odp_to_ofport_map);
1222 node = shash_find(&all_dpif_backers, backer->type);
1224 shash_delete(&all_dpif_backers, node);
1225 dpif_close(backer->dpif);
1230 /* Datapath port slated for removal from datapath. */
1231 struct odp_garbage {
1232 struct list list_node;
1237 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1239 struct dpif_backer *backer;
1240 struct dpif_port_dump port_dump;
1241 struct dpif_port port;
1242 struct shash_node *node;
1243 struct list garbage_list;
1244 struct odp_garbage *garbage, *next;
1250 backer = shash_find_data(&all_dpif_backers, type);
1257 backer_name = xasprintf("ovs-%s", type);
1259 /* Remove any existing datapaths, since we assume we're the only
1260 * userspace controlling the datapath. */
1262 dp_enumerate_names(type, &names);
1263 SSET_FOR_EACH(name, &names) {
1264 struct dpif *old_dpif;
1266 /* Don't remove our backer if it exists. */
1267 if (!strcmp(name, backer_name)) {
1271 if (dpif_open(name, type, &old_dpif)) {
1272 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1274 dpif_delete(old_dpif);
1275 dpif_close(old_dpif);
1278 sset_destroy(&names);
1280 backer = xmalloc(sizeof *backer);
1282 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1285 VLOG_ERR("failed to open datapath of type %s: %s", type,
1291 backer->type = xstrdup(type);
1292 backer->refcount = 1;
1293 hmap_init(&backer->odp_to_ofport_map);
1294 hmap_init(&backer->drop_keys);
1295 timer_set_duration(&backer->next_expiration, 1000);
1296 backer->need_revalidate = 0;
1297 simap_init(&backer->tnl_backers);
1298 tag_set_init(&backer->revalidate_set);
1301 dpif_flow_flush(backer->dpif);
1303 /* Loop through the ports already on the datapath and remove any
1304 * that we don't need anymore. */
1305 list_init(&garbage_list);
1306 dpif_port_dump_start(&port_dump, backer->dpif);
1307 while (dpif_port_dump_next(&port_dump, &port)) {
1308 node = shash_find(&init_ofp_ports, port.name);
1309 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1310 garbage = xmalloc(sizeof *garbage);
1311 garbage->odp_port = port.port_no;
1312 list_push_front(&garbage_list, &garbage->list_node);
1315 dpif_port_dump_done(&port_dump);
1317 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1318 dpif_port_del(backer->dpif, garbage->odp_port);
1319 list_remove(&garbage->list_node);
1323 shash_add(&all_dpif_backers, type, backer);
1325 error = dpif_recv_set(backer->dpif, true);
1327 VLOG_ERR("failed to listen on datapath of type %s: %s",
1328 type, strerror(error));
1329 close_dpif_backer(backer);
1337 construct(struct ofproto *ofproto_)
1339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1340 struct shash_node *node, *next;
1345 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1350 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1351 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1353 ofproto->n_matches = 0;
1355 ofproto->netflow = NULL;
1356 ofproto->sflow = NULL;
1357 ofproto->ipfix = NULL;
1358 ofproto->stp = NULL;
1359 hmap_init(&ofproto->bundles);
1360 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1361 for (i = 0; i < MAX_MIRRORS; i++) {
1362 ofproto->mirrors[i] = NULL;
1364 ofproto->has_bonded_bundles = false;
1366 hmap_init(&ofproto->facets);
1367 hmap_init(&ofproto->subfacets);
1368 ofproto->governor = NULL;
1369 ofproto->consistency_rl = LLONG_MIN;
1371 for (i = 0; i < N_TABLES; i++) {
1372 struct table_dpif *table = &ofproto->tables[i];
1374 table->catchall_table = NULL;
1375 table->other_table = NULL;
1376 table->basis = random_uint32();
1379 list_init(&ofproto->completions);
1381 ofproto_dpif_unixctl_init();
1383 ofproto->has_mirrors = false;
1384 ofproto->has_bundle_action = false;
1386 hmap_init(&ofproto->vlandev_map);
1387 hmap_init(&ofproto->realdev_vid_map);
1389 sset_init(&ofproto->ports);
1390 sset_init(&ofproto->ghost_ports);
1391 sset_init(&ofproto->port_poll_set);
1392 ofproto->port_poll_errno = 0;
1394 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1395 struct iface_hint *iface_hint = node->data;
1397 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1398 /* Check if the datapath already has this port. */
1399 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1400 sset_add(&ofproto->ports, node->name);
1403 free(iface_hint->br_name);
1404 free(iface_hint->br_type);
1406 shash_delete(&init_ofp_ports, node);
1410 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1411 hash_string(ofproto->up.name, 0));
1412 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1414 ofproto_init_tables(ofproto_, N_TABLES);
1415 error = add_internal_flows(ofproto);
1416 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1419 ofproto->n_missed = 0;
1421 ofproto->max_n_subfacet = 0;
1422 ofproto->created = time_msec();
1423 ofproto->last_minute = ofproto->created;
1424 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1425 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1426 ofproto->subfacet_add_count = 0;
1427 ofproto->subfacet_del_count = 0;
1428 ofproto->total_subfacet_add_count = 0;
1429 ofproto->total_subfacet_del_count = 0;
1430 ofproto->total_subfacet_life_span = 0;
1431 ofproto->total_subfacet_count = 0;
1432 ofproto->n_update_stats = 0;
1438 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1439 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1441 struct ofputil_flow_mod fm;
1444 match_init_catchall(&fm.match);
1446 match_set_reg(&fm.match, 0, id);
1447 fm.new_cookie = htonll(0);
1448 fm.cookie = htonll(0);
1449 fm.cookie_mask = htonll(0);
1450 fm.table_id = TBL_INTERNAL;
1451 fm.command = OFPFC_ADD;
1452 fm.idle_timeout = 0;
1453 fm.hard_timeout = 0;
1457 fm.ofpacts = ofpacts->data;
1458 fm.ofpacts_len = ofpacts->size;
1460 error = ofproto_flow_mod(&ofproto->up, &fm);
1462 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1463 id, ofperr_to_string(error));
1467 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1468 ovs_assert(*rulep != NULL);
1474 add_internal_flows(struct ofproto_dpif *ofproto)
1476 struct ofpact_controller *controller;
1477 uint64_t ofpacts_stub[128 / 8];
1478 struct ofpbuf ofpacts;
1482 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1485 controller = ofpact_put_CONTROLLER(&ofpacts);
1486 controller->max_len = UINT16_MAX;
1487 controller->controller_id = 0;
1488 controller->reason = OFPR_NO_MATCH;
1489 ofpact_pad(&ofpacts);
1491 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1496 ofpbuf_clear(&ofpacts);
1497 error = add_internal_flow(ofproto, id++, &ofpacts,
1498 &ofproto->no_packet_in_rule);
1503 complete_operations(struct ofproto_dpif *ofproto)
1505 struct dpif_completion *c, *next;
1507 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1508 ofoperation_complete(c->op, 0);
1509 list_remove(&c->list_node);
1515 destruct(struct ofproto *ofproto_)
1517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1518 struct rule_dpif *rule, *next_rule;
1519 struct oftable *table;
1522 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1523 complete_operations(ofproto);
1525 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1526 struct cls_cursor cursor;
1528 cls_cursor_init(&cursor, &table->cls, NULL);
1529 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1530 ofproto_rule_destroy(&rule->up);
1534 for (i = 0; i < MAX_MIRRORS; i++) {
1535 mirror_destroy(ofproto->mirrors[i]);
1538 netflow_destroy(ofproto->netflow);
1539 dpif_sflow_destroy(ofproto->sflow);
1540 hmap_destroy(&ofproto->bundles);
1541 mac_learning_destroy(ofproto->ml);
1543 hmap_destroy(&ofproto->facets);
1544 hmap_destroy(&ofproto->subfacets);
1545 governor_destroy(ofproto->governor);
1547 hmap_destroy(&ofproto->vlandev_map);
1548 hmap_destroy(&ofproto->realdev_vid_map);
1550 sset_destroy(&ofproto->ports);
1551 sset_destroy(&ofproto->ghost_ports);
1552 sset_destroy(&ofproto->port_poll_set);
1554 close_dpif_backer(ofproto->backer);
1558 run_fast(struct ofproto *ofproto_)
1560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1561 struct ofport_dpif *ofport;
1563 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1564 port_run_fast(ofport);
1571 run(struct ofproto *ofproto_)
1573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1574 struct ofport_dpif *ofport;
1575 struct ofbundle *bundle;
1579 complete_operations(ofproto);
1582 error = run_fast(ofproto_);
1587 if (ofproto->netflow) {
1588 if (netflow_run(ofproto->netflow)) {
1589 send_netflow_active_timeouts(ofproto);
1592 if (ofproto->sflow) {
1593 dpif_sflow_run(ofproto->sflow);
1596 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1599 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1604 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1606 /* Check the consistency of a random facet, to aid debugging. */
1607 if (time_msec() >= ofproto->consistency_rl
1608 && !hmap_is_empty(&ofproto->facets)
1609 && !ofproto->backer->need_revalidate) {
1610 struct facet *facet;
1612 ofproto->consistency_rl = time_msec() + 250;
1614 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1615 struct facet, hmap_node);
1616 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1618 if (!facet_check_consistency(facet)) {
1619 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1624 if (ofproto->governor) {
1627 governor_run(ofproto->governor);
1629 /* If the governor has shrunk to its minimum size and the number of
1630 * subfacets has dwindled, then drop the governor entirely.
1632 * For hysteresis, the number of subfacets to drop the governor is
1633 * smaller than the number needed to trigger its creation. */
1634 n_subfacets = hmap_count(&ofproto->subfacets);
1635 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1636 && governor_is_idle(ofproto->governor)) {
1637 governor_destroy(ofproto->governor);
1638 ofproto->governor = NULL;
1646 wait(struct ofproto *ofproto_)
1648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1649 struct ofport_dpif *ofport;
1650 struct ofbundle *bundle;
1652 if (!clogged && !list_is_empty(&ofproto->completions)) {
1653 poll_immediate_wake();
1656 dpif_wait(ofproto->backer->dpif);
1657 dpif_recv_wait(ofproto->backer->dpif);
1658 if (ofproto->sflow) {
1659 dpif_sflow_wait(ofproto->sflow);
1661 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1662 poll_immediate_wake();
1664 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1667 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1668 bundle_wait(bundle);
1670 if (ofproto->netflow) {
1671 netflow_wait(ofproto->netflow);
1673 mac_learning_wait(ofproto->ml);
1675 if (ofproto->backer->need_revalidate) {
1676 /* Shouldn't happen, but if it does just go around again. */
1677 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1678 poll_immediate_wake();
1680 if (ofproto->governor) {
1681 governor_wait(ofproto->governor);
1686 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1688 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1691 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1695 flush(struct ofproto *ofproto_)
1697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1698 struct subfacet *subfacet, *next_subfacet;
1699 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1703 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1704 &ofproto->subfacets) {
1705 if (subfacet->path != SF_NOT_INSTALLED) {
1706 batch[n_batch++] = subfacet;
1707 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1708 subfacet_destroy_batch(ofproto, batch, n_batch);
1712 subfacet_destroy(subfacet);
1717 subfacet_destroy_batch(ofproto, batch, n_batch);
1722 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1723 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1725 *arp_match_ip = true;
1726 *actions = (OFPUTIL_A_OUTPUT |
1727 OFPUTIL_A_SET_VLAN_VID |
1728 OFPUTIL_A_SET_VLAN_PCP |
1729 OFPUTIL_A_STRIP_VLAN |
1730 OFPUTIL_A_SET_DL_SRC |
1731 OFPUTIL_A_SET_DL_DST |
1732 OFPUTIL_A_SET_NW_SRC |
1733 OFPUTIL_A_SET_NW_DST |
1734 OFPUTIL_A_SET_NW_TOS |
1735 OFPUTIL_A_SET_TP_SRC |
1736 OFPUTIL_A_SET_TP_DST |
1741 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1744 struct dpif_dp_stats s;
1746 strcpy(ots->name, "classifier");
1748 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1750 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1751 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1754 static struct ofport *
1757 struct ofport_dpif *port = xmalloc(sizeof *port);
1762 port_dealloc(struct ofport *port_)
1764 struct ofport_dpif *port = ofport_dpif_cast(port_);
1769 port_construct(struct ofport *port_)
1771 struct ofport_dpif *port = ofport_dpif_cast(port_);
1772 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1773 const struct netdev *netdev = port->up.netdev;
1774 struct dpif_port dpif_port;
1777 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1778 port->bundle = NULL;
1780 port->tag = tag_create_random();
1781 port->may_enable = true;
1782 port->stp_port = NULL;
1783 port->stp_state = STP_DISABLED;
1784 port->tnl_port = NULL;
1785 hmap_init(&port->priorities);
1786 port->realdev_ofp_port = 0;
1787 port->vlandev_vid = 0;
1788 port->carrier_seq = netdev_get_carrier_resets(netdev);
1790 if (netdev_vport_is_patch(netdev)) {
1791 /* By bailing out here, we don't submit the port to the sFlow module
1792 * to be considered for counter polling export. This is correct
1793 * because the patch port represents an interface that sFlow considers
1794 * to be "internal" to the switch as a whole, and therefore not an
1795 * candidate for counter polling. */
1796 port->odp_port = OVSP_NONE;
1800 error = dpif_port_query_by_name(ofproto->backer->dpif,
1801 netdev_vport_get_dpif_port(netdev),
1807 port->odp_port = dpif_port.port_no;
1809 if (netdev_get_tunnel_config(netdev)) {
1810 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1812 /* Sanity-check that a mapping doesn't already exist. This
1813 * shouldn't happen for non-tunnel ports. */
1814 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1815 VLOG_ERR("port %s already has an OpenFlow port number",
1817 dpif_port_destroy(&dpif_port);
1821 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1822 hash_int(port->odp_port, 0));
1824 dpif_port_destroy(&dpif_port);
1826 if (ofproto->sflow) {
1827 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1834 port_destruct(struct ofport *port_)
1836 struct ofport_dpif *port = ofport_dpif_cast(port_);
1837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1838 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1839 const char *devname = netdev_get_name(port->up.netdev);
1841 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1842 /* The underlying device is still there, so delete it. This
1843 * happens when the ofproto is being destroyed, since the caller
1844 * assumes that removal of attached ports will happen as part of
1846 if (!port->tnl_port) {
1847 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1849 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1852 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1853 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1856 tnl_port_del(port->tnl_port);
1857 sset_find_and_delete(&ofproto->ports, devname);
1858 sset_find_and_delete(&ofproto->ghost_ports, devname);
1859 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860 bundle_remove(port_);
1861 set_cfm(port_, NULL);
1862 if (ofproto->sflow) {
1863 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1866 ofport_clear_priorities(port);
1867 hmap_destroy(&port->priorities);
1871 port_modified(struct ofport *port_)
1873 struct ofport_dpif *port = ofport_dpif_cast(port_);
1875 if (port->bundle && port->bundle->bond) {
1876 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1881 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1883 struct ofport_dpif *port = ofport_dpif_cast(port_);
1884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1885 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1887 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1888 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1889 OFPUTIL_PC_NO_PACKET_IN)) {
1890 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1893 bundle_update(port->bundle);
1899 set_sflow(struct ofproto *ofproto_,
1900 const struct ofproto_sflow_options *sflow_options)
1902 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1903 struct dpif_sflow *ds = ofproto->sflow;
1905 if (sflow_options) {
1907 struct ofport_dpif *ofport;
1909 ds = ofproto->sflow = dpif_sflow_create();
1910 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1911 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1913 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1915 dpif_sflow_set_options(ds, sflow_options);
1918 dpif_sflow_destroy(ds);
1919 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1920 ofproto->sflow = NULL;
1928 struct ofproto *ofproto_,
1929 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1930 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1931 size_t n_flow_exporters_options)
1933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1934 struct dpif_ipfix *di = ofproto->ipfix;
1936 if (bridge_exporter_options || flow_exporters_options) {
1938 di = ofproto->ipfix = dpif_ipfix_create();
1940 dpif_ipfix_set_options(
1941 di, bridge_exporter_options, flow_exporters_options,
1942 n_flow_exporters_options);
1945 dpif_ipfix_destroy(di);
1946 ofproto->ipfix = NULL;
1953 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1955 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1962 struct ofproto_dpif *ofproto;
1964 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1965 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966 ofport->cfm = cfm_create(ofport->up.netdev);
1969 if (cfm_configure(ofport->cfm, s)) {
1975 cfm_destroy(ofport->cfm);
1981 get_cfm_status(const struct ofport *ofport_,
1982 struct ofproto_cfm_status *status)
1984 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987 status->faults = cfm_get_fault(ofport->cfm);
1988 status->remote_opstate = cfm_get_opup(ofport->cfm);
1989 status->health = cfm_get_health(ofport->cfm);
1990 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1997 /* Spanning Tree. */
2000 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2002 struct ofproto_dpif *ofproto = ofproto_;
2003 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2004 struct ofport_dpif *ofport;
2006 ofport = stp_port_get_aux(sp);
2008 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2009 ofproto->up.name, port_num);
2011 struct eth_header *eth = pkt->l2;
2013 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2014 if (eth_addr_is_zero(eth->eth_src)) {
2015 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2016 "with unknown MAC", ofproto->up.name, port_num);
2018 send_packet(ofport, pkt);
2024 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2026 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2028 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2030 /* Only revalidate flows if the configuration changed. */
2031 if (!s != !ofproto->stp) {
2032 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2036 if (!ofproto->stp) {
2037 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2038 send_bpdu_cb, ofproto);
2039 ofproto->stp_last_tick = time_msec();
2042 stp_set_bridge_id(ofproto->stp, s->system_id);
2043 stp_set_bridge_priority(ofproto->stp, s->priority);
2044 stp_set_hello_time(ofproto->stp, s->hello_time);
2045 stp_set_max_age(ofproto->stp, s->max_age);
2046 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2048 struct ofport *ofport;
2050 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2051 set_stp_port(ofport, NULL);
2054 stp_destroy(ofproto->stp);
2055 ofproto->stp = NULL;
2062 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2064 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2068 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2069 s->designated_root = stp_get_designated_root(ofproto->stp);
2070 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2079 update_stp_port_state(struct ofport_dpif *ofport)
2081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2082 enum stp_state state;
2084 /* Figure out new state. */
2085 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2089 if (ofport->stp_state != state) {
2090 enum ofputil_port_state of_state;
2093 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2094 netdev_get_name(ofport->up.netdev),
2095 stp_state_name(ofport->stp_state),
2096 stp_state_name(state));
2097 if (stp_learn_in_state(ofport->stp_state)
2098 != stp_learn_in_state(state)) {
2099 /* xxx Learning action flows should also be flushed. */
2100 mac_learning_flush(ofproto->ml,
2101 &ofproto->backer->revalidate_set);
2103 fwd_change = stp_forward_in_state(ofport->stp_state)
2104 != stp_forward_in_state(state);
2106 ofproto->backer->need_revalidate = REV_STP;
2107 ofport->stp_state = state;
2108 ofport->stp_state_entered = time_msec();
2110 if (fwd_change && ofport->bundle) {
2111 bundle_update(ofport->bundle);
2114 /* Update the STP state bits in the OpenFlow port description. */
2115 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2116 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2117 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2118 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2119 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2121 ofproto_port_set_state(&ofport->up, of_state);
2125 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2126 * caller is responsible for assigning STP port numbers and ensuring
2127 * there are no duplicates. */
2129 set_stp_port(struct ofport *ofport_,
2130 const struct ofproto_port_stp_settings *s)
2132 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2134 struct stp_port *sp = ofport->stp_port;
2136 if (!s || !s->enable) {
2138 ofport->stp_port = NULL;
2139 stp_port_disable(sp);
2140 update_stp_port_state(ofport);
2143 } else if (sp && stp_port_no(sp) != s->port_num
2144 && ofport == stp_port_get_aux(sp)) {
2145 /* The port-id changed, so disable the old one if it's not
2146 * already in use by another port. */
2147 stp_port_disable(sp);
2150 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2151 stp_port_enable(sp);
2153 stp_port_set_aux(sp, ofport);
2154 stp_port_set_priority(sp, s->priority);
2155 stp_port_set_path_cost(sp, s->path_cost);
2157 update_stp_port_state(ofport);
2163 get_stp_port_status(struct ofport *ofport_,
2164 struct ofproto_port_stp_status *s)
2166 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2167 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2168 struct stp_port *sp = ofport->stp_port;
2170 if (!ofproto->stp || !sp) {
2176 s->port_id = stp_port_get_id(sp);
2177 s->state = stp_port_get_state(sp);
2178 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2179 s->role = stp_port_get_role(sp);
2180 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2186 stp_run(struct ofproto_dpif *ofproto)
2189 long long int now = time_msec();
2190 long long int elapsed = now - ofproto->stp_last_tick;
2191 struct stp_port *sp;
2194 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2195 ofproto->stp_last_tick = now;
2197 while (stp_get_changed_port(ofproto->stp, &sp)) {
2198 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2201 update_stp_port_state(ofport);
2205 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2206 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2212 stp_wait(struct ofproto_dpif *ofproto)
2215 poll_timer_wait(1000);
2219 /* Returns true if STP should process 'flow'. */
2221 stp_should_process_flow(const struct flow *flow)
2223 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2227 stp_process_packet(const struct ofport_dpif *ofport,
2228 const struct ofpbuf *packet)
2230 struct ofpbuf payload = *packet;
2231 struct eth_header *eth = payload.data;
2232 struct stp_port *sp = ofport->stp_port;
2234 /* Sink packets on ports that have STP disabled when the bridge has
2236 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2240 /* Trim off padding on payload. */
2241 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2242 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2245 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2246 stp_received_bpdu(sp, payload.data, payload.size);
2250 static struct priority_to_dscp *
2251 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2253 struct priority_to_dscp *pdscp;
2256 hash = hash_int(priority, 0);
2257 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2258 if (pdscp->priority == priority) {
2266 ofport_clear_priorities(struct ofport_dpif *ofport)
2268 struct priority_to_dscp *pdscp, *next;
2270 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2271 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2277 set_queues(struct ofport *ofport_,
2278 const struct ofproto_port_queue *qdscp_list,
2281 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2283 struct hmap new = HMAP_INITIALIZER(&new);
2286 for (i = 0; i < n_qdscp; i++) {
2287 struct priority_to_dscp *pdscp;
2291 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2292 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2297 pdscp = get_priority(ofport, priority);
2299 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2301 pdscp = xmalloc(sizeof *pdscp);
2302 pdscp->priority = priority;
2304 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2307 if (pdscp->dscp != dscp) {
2309 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2312 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2315 if (!hmap_is_empty(&ofport->priorities)) {
2316 ofport_clear_priorities(ofport);
2317 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2320 hmap_swap(&new, &ofport->priorities);
2328 /* Expires all MAC learning entries associated with 'bundle' and forces its
2329 * ofproto to revalidate every flow.
2331 * Normally MAC learning entries are removed only from the ofproto associated
2332 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2333 * are removed from every ofproto. When patch ports and SLB bonds are in use
2334 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2335 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2336 * with the host from which it migrated. */
2338 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2340 struct ofproto_dpif *ofproto = bundle->ofproto;
2341 struct mac_learning *ml = ofproto->ml;
2342 struct mac_entry *mac, *next_mac;
2344 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2346 if (mac->port.p == bundle) {
2348 struct ofproto_dpif *o;
2350 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2352 struct mac_entry *e;
2354 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2357 mac_learning_expire(o->ml, e);
2363 mac_learning_expire(ml, mac);
2368 static struct ofbundle *
2369 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2371 struct ofbundle *bundle;
2373 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2374 &ofproto->bundles) {
2375 if (bundle->aux == aux) {
2382 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2383 * ones that are found to 'bundles'. */
2385 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2386 void **auxes, size_t n_auxes,
2387 struct hmapx *bundles)
2391 hmapx_init(bundles);
2392 for (i = 0; i < n_auxes; i++) {
2393 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2395 hmapx_add(bundles, bundle);
2401 bundle_update(struct ofbundle *bundle)
2403 struct ofport_dpif *port;
2405 bundle->floodable = true;
2406 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2407 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2408 || !stp_forward_in_state(port->stp_state)) {
2409 bundle->floodable = false;
2416 bundle_del_port(struct ofport_dpif *port)
2418 struct ofbundle *bundle = port->bundle;
2420 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2422 list_remove(&port->bundle_node);
2423 port->bundle = NULL;
2426 lacp_slave_unregister(bundle->lacp, port);
2429 bond_slave_unregister(bundle->bond, port);
2432 bundle_update(bundle);
2436 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2437 struct lacp_slave_settings *lacp)
2439 struct ofport_dpif *port;
2441 port = get_ofp_port(bundle->ofproto, ofp_port);
2446 if (port->bundle != bundle) {
2447 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2449 bundle_del_port(port);
2452 port->bundle = bundle;
2453 list_push_back(&bundle->ports, &port->bundle_node);
2454 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2455 || !stp_forward_in_state(port->stp_state)) {
2456 bundle->floodable = false;
2460 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2461 lacp_slave_register(bundle->lacp, port, lacp);
2468 bundle_destroy(struct ofbundle *bundle)
2470 struct ofproto_dpif *ofproto;
2471 struct ofport_dpif *port, *next_port;
2478 ofproto = bundle->ofproto;
2479 for (i = 0; i < MAX_MIRRORS; i++) {
2480 struct ofmirror *m = ofproto->mirrors[i];
2482 if (m->out == bundle) {
2484 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2485 || hmapx_find_and_delete(&m->dsts, bundle)) {
2486 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2491 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2492 bundle_del_port(port);
2495 bundle_flush_macs(bundle, true);
2496 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2498 free(bundle->trunks);
2499 lacp_destroy(bundle->lacp);
2500 bond_destroy(bundle->bond);
2505 bundle_set(struct ofproto *ofproto_, void *aux,
2506 const struct ofproto_bundle_settings *s)
2508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2509 bool need_flush = false;
2510 struct ofport_dpif *port;
2511 struct ofbundle *bundle;
2512 unsigned long *trunks;
2518 bundle_destroy(bundle_lookup(ofproto, aux));
2522 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2523 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2525 bundle = bundle_lookup(ofproto, aux);
2527 bundle = xmalloc(sizeof *bundle);
2529 bundle->ofproto = ofproto;
2530 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2531 hash_pointer(aux, 0));
2533 bundle->name = NULL;
2535 list_init(&bundle->ports);
2536 bundle->vlan_mode = PORT_VLAN_TRUNK;
2538 bundle->trunks = NULL;
2539 bundle->use_priority_tags = s->use_priority_tags;
2540 bundle->lacp = NULL;
2541 bundle->bond = NULL;
2543 bundle->floodable = true;
2545 bundle->src_mirrors = 0;
2546 bundle->dst_mirrors = 0;
2547 bundle->mirror_out = 0;
2550 if (!bundle->name || strcmp(s->name, bundle->name)) {
2552 bundle->name = xstrdup(s->name);
2557 if (!bundle->lacp) {
2558 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2559 bundle->lacp = lacp_create();
2561 lacp_configure(bundle->lacp, s->lacp);
2563 lacp_destroy(bundle->lacp);
2564 bundle->lacp = NULL;
2567 /* Update set of ports. */
2569 for (i = 0; i < s->n_slaves; i++) {
2570 if (!bundle_add_port(bundle, s->slaves[i],
2571 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2575 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2576 struct ofport_dpif *next_port;
2578 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2579 for (i = 0; i < s->n_slaves; i++) {
2580 if (s->slaves[i] == port->up.ofp_port) {
2585 bundle_del_port(port);
2589 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2591 if (list_is_empty(&bundle->ports)) {
2592 bundle_destroy(bundle);
2596 /* Set VLAN tagging mode */
2597 if (s->vlan_mode != bundle->vlan_mode
2598 || s->use_priority_tags != bundle->use_priority_tags) {
2599 bundle->vlan_mode = s->vlan_mode;
2600 bundle->use_priority_tags = s->use_priority_tags;
2605 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2606 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2608 if (vlan != bundle->vlan) {
2609 bundle->vlan = vlan;
2613 /* Get trunked VLANs. */
2614 switch (s->vlan_mode) {
2615 case PORT_VLAN_ACCESS:
2619 case PORT_VLAN_TRUNK:
2620 trunks = CONST_CAST(unsigned long *, s->trunks);
2623 case PORT_VLAN_NATIVE_UNTAGGED:
2624 case PORT_VLAN_NATIVE_TAGGED:
2625 if (vlan != 0 && (!s->trunks
2626 || !bitmap_is_set(s->trunks, vlan)
2627 || bitmap_is_set(s->trunks, 0))) {
2628 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2630 trunks = bitmap_clone(s->trunks, 4096);
2632 trunks = bitmap_allocate1(4096);
2634 bitmap_set1(trunks, vlan);
2635 bitmap_set0(trunks, 0);
2637 trunks = CONST_CAST(unsigned long *, s->trunks);
2644 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2645 free(bundle->trunks);
2646 if (trunks == s->trunks) {
2647 bundle->trunks = vlan_bitmap_clone(trunks);
2649 bundle->trunks = trunks;
2654 if (trunks != s->trunks) {
2659 if (!list_is_short(&bundle->ports)) {
2660 bundle->ofproto->has_bonded_bundles = true;
2662 if (bond_reconfigure(bundle->bond, s->bond)) {
2663 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2666 bundle->bond = bond_create(s->bond);
2667 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2670 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2671 bond_slave_register(bundle->bond, port, port->up.netdev);
2674 bond_destroy(bundle->bond);
2675 bundle->bond = NULL;
2678 /* If we changed something that would affect MAC learning, un-learn
2679 * everything on this port and force flow revalidation. */
2681 bundle_flush_macs(bundle, false);
2688 bundle_remove(struct ofport *port_)
2690 struct ofport_dpif *port = ofport_dpif_cast(port_);
2691 struct ofbundle *bundle = port->bundle;
2694 bundle_del_port(port);
2695 if (list_is_empty(&bundle->ports)) {
2696 bundle_destroy(bundle);
2697 } else if (list_is_short(&bundle->ports)) {
2698 bond_destroy(bundle->bond);
2699 bundle->bond = NULL;
2705 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2708 struct ofport_dpif *port = port_;
2709 uint8_t ea[ETH_ADDR_LEN];
2712 error = netdev_get_etheraddr(port->up.netdev, ea);
2714 struct ofpbuf packet;
2717 ofpbuf_init(&packet, 0);
2718 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2720 memcpy(packet_pdu, pdu, pdu_size);
2722 send_packet(port, &packet);
2723 ofpbuf_uninit(&packet);
2725 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2726 "%s (%s)", port->bundle->name,
2727 netdev_get_name(port->up.netdev), strerror(error));
2732 bundle_send_learning_packets(struct ofbundle *bundle)
2734 struct ofproto_dpif *ofproto = bundle->ofproto;
2735 int error, n_packets, n_errors;
2736 struct mac_entry *e;
2738 error = n_packets = n_errors = 0;
2739 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2740 if (e->port.p != bundle) {
2741 struct ofpbuf *learning_packet;
2742 struct ofport_dpif *port;
2746 /* The assignment to "port" is unnecessary but makes "grep"ing for
2747 * struct ofport_dpif more effective. */
2748 learning_packet = bond_compose_learning_packet(bundle->bond,
2752 ret = send_packet(port, learning_packet);
2753 ofpbuf_delete(learning_packet);
2763 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2764 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2765 "packets, last error was: %s",
2766 bundle->name, n_errors, n_packets, strerror(error));
2768 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2769 bundle->name, n_packets);
2774 bundle_run(struct ofbundle *bundle)
2777 lacp_run(bundle->lacp, send_pdu_cb);
2780 struct ofport_dpif *port;
2782 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2783 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2786 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2787 lacp_status(bundle->lacp));
2788 if (bond_should_send_learning_packets(bundle->bond)) {
2789 bundle_send_learning_packets(bundle);
2795 bundle_wait(struct ofbundle *bundle)
2798 lacp_wait(bundle->lacp);
2801 bond_wait(bundle->bond);
2808 mirror_scan(struct ofproto_dpif *ofproto)
2812 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2813 if (!ofproto->mirrors[idx]) {
2820 static struct ofmirror *
2821 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2825 for (i = 0; i < MAX_MIRRORS; i++) {
2826 struct ofmirror *mirror = ofproto->mirrors[i];
2827 if (mirror && mirror->aux == aux) {
2835 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2837 mirror_update_dups(struct ofproto_dpif *ofproto)
2841 for (i = 0; i < MAX_MIRRORS; i++) {
2842 struct ofmirror *m = ofproto->mirrors[i];
2845 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2849 for (i = 0; i < MAX_MIRRORS; i++) {
2850 struct ofmirror *m1 = ofproto->mirrors[i];
2857 for (j = i + 1; j < MAX_MIRRORS; j++) {
2858 struct ofmirror *m2 = ofproto->mirrors[j];
2860 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2861 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2862 m2->dup_mirrors |= m1->dup_mirrors;
2869 mirror_set(struct ofproto *ofproto_, void *aux,
2870 const struct ofproto_mirror_settings *s)
2872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873 mirror_mask_t mirror_bit;
2874 struct ofbundle *bundle;
2875 struct ofmirror *mirror;
2876 struct ofbundle *out;
2877 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2878 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2881 mirror = mirror_lookup(ofproto, aux);
2883 mirror_destroy(mirror);
2889 idx = mirror_scan(ofproto);
2891 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2893 ofproto->up.name, MAX_MIRRORS, s->name);
2897 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2898 mirror->ofproto = ofproto;
2901 mirror->out_vlan = -1;
2902 mirror->name = NULL;
2905 if (!mirror->name || strcmp(s->name, mirror->name)) {
2907 mirror->name = xstrdup(s->name);
2910 /* Get the new configuration. */
2911 if (s->out_bundle) {
2912 out = bundle_lookup(ofproto, s->out_bundle);
2914 mirror_destroy(mirror);
2920 out_vlan = s->out_vlan;
2922 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2923 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2925 /* If the configuration has not changed, do nothing. */
2926 if (hmapx_equals(&srcs, &mirror->srcs)
2927 && hmapx_equals(&dsts, &mirror->dsts)
2928 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2929 && mirror->out == out
2930 && mirror->out_vlan == out_vlan)
2932 hmapx_destroy(&srcs);
2933 hmapx_destroy(&dsts);
2937 hmapx_swap(&srcs, &mirror->srcs);
2938 hmapx_destroy(&srcs);
2940 hmapx_swap(&dsts, &mirror->dsts);
2941 hmapx_destroy(&dsts);
2943 free(mirror->vlans);
2944 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2947 mirror->out_vlan = out_vlan;
2949 /* Update bundles. */
2950 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2951 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2952 if (hmapx_contains(&mirror->srcs, bundle)) {
2953 bundle->src_mirrors |= mirror_bit;
2955 bundle->src_mirrors &= ~mirror_bit;
2958 if (hmapx_contains(&mirror->dsts, bundle)) {
2959 bundle->dst_mirrors |= mirror_bit;
2961 bundle->dst_mirrors &= ~mirror_bit;
2964 if (mirror->out == bundle) {
2965 bundle->mirror_out |= mirror_bit;
2967 bundle->mirror_out &= ~mirror_bit;
2971 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2972 ofproto->has_mirrors = true;
2973 mac_learning_flush(ofproto->ml,
2974 &ofproto->backer->revalidate_set);
2975 mirror_update_dups(ofproto);
2981 mirror_destroy(struct ofmirror *mirror)
2983 struct ofproto_dpif *ofproto;
2984 mirror_mask_t mirror_bit;
2985 struct ofbundle *bundle;
2992 ofproto = mirror->ofproto;
2993 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2994 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2996 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2997 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2998 bundle->src_mirrors &= ~mirror_bit;
2999 bundle->dst_mirrors &= ~mirror_bit;
3000 bundle->mirror_out &= ~mirror_bit;
3003 hmapx_destroy(&mirror->srcs);
3004 hmapx_destroy(&mirror->dsts);
3005 free(mirror->vlans);
3007 ofproto->mirrors[mirror->idx] = NULL;
3011 mirror_update_dups(ofproto);
3013 ofproto->has_mirrors = false;
3014 for (i = 0; i < MAX_MIRRORS; i++) {
3015 if (ofproto->mirrors[i]) {
3016 ofproto->has_mirrors = true;
3023 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3024 uint64_t *packets, uint64_t *bytes)
3026 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3027 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3030 *packets = *bytes = UINT64_MAX;
3036 *packets = mirror->packet_count;
3037 *bytes = mirror->byte_count;
3043 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3046 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3047 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3053 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3056 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3057 return bundle && bundle->mirror_out != 0;
3061 forward_bpdu_changed(struct ofproto *ofproto_)
3063 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3064 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3068 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3071 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3072 mac_learning_set_idle_time(ofproto->ml, idle_time);
3073 mac_learning_set_max_entries(ofproto->ml, max_entries);
3078 static struct ofport_dpif *
3079 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3081 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3082 return ofport ? ofport_dpif_cast(ofport) : NULL;
3085 static struct ofport_dpif *
3086 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3088 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3089 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3093 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3094 struct ofproto_port *ofproto_port,
3095 struct dpif_port *dpif_port)
3097 ofproto_port->name = dpif_port->name;
3098 ofproto_port->type = dpif_port->type;
3099 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3102 static struct ofport_dpif *
3103 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3105 const struct ofproto_dpif *ofproto;
3108 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3113 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3114 struct ofport *ofport;
3116 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3117 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3118 return ofport_dpif_cast(ofport);
3125 port_run_fast(struct ofport_dpif *ofport)
3127 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3128 struct ofpbuf packet;
3130 ofpbuf_init(&packet, 0);
3131 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3132 send_packet(ofport, &packet);
3133 ofpbuf_uninit(&packet);
3138 port_run(struct ofport_dpif *ofport)
3140 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3141 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3142 bool enable = netdev_get_carrier(ofport->up.netdev);
3144 ofport->carrier_seq = carrier_seq;
3146 port_run_fast(ofport);
3148 if (ofport->tnl_port
3149 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3150 &ofport->tnl_port)) {
3151 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3155 int cfm_opup = cfm_get_opup(ofport->cfm);
3157 cfm_run(ofport->cfm);
3158 enable = enable && !cfm_get_fault(ofport->cfm);
3160 if (cfm_opup >= 0) {
3161 enable = enable && cfm_opup;
3165 if (ofport->bundle) {
3166 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3167 if (carrier_changed) {
3168 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3172 if (ofport->may_enable != enable) {
3173 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3175 if (ofproto->has_bundle_action) {
3176 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3180 ofport->may_enable = enable;
3184 port_wait(struct ofport_dpif *ofport)
3187 cfm_wait(ofport->cfm);
3192 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3193 struct ofproto_port *ofproto_port)
3195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3196 struct dpif_port dpif_port;
3199 if (sset_contains(&ofproto->ghost_ports, devname)) {
3200 const char *type = netdev_get_type_from_name(devname);
3202 /* We may be called before ofproto->up.port_by_name is populated with
3203 * the appropriate ofport. For this reason, we must get the name and
3204 * type from the netdev layer directly. */
3206 const struct ofport *ofport;
3208 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3209 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3210 ofproto_port->name = xstrdup(devname);
3211 ofproto_port->type = xstrdup(type);
3217 if (!sset_contains(&ofproto->ports, devname)) {
3220 error = dpif_port_query_by_name(ofproto->backer->dpif,
3221 devname, &dpif_port);
3223 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3229 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3231 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3232 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3233 const char *devname = netdev_get_name(netdev);
3235 if (netdev_vport_is_patch(netdev)) {
3236 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3240 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3241 uint32_t port_no = UINT32_MAX;
3244 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3248 if (netdev_get_tunnel_config(netdev)) {
3249 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3253 if (netdev_get_tunnel_config(netdev)) {
3254 sset_add(&ofproto->ghost_ports, devname);
3256 sset_add(&ofproto->ports, devname);
3262 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3264 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3265 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3272 sset_find_and_delete(&ofproto->ghost_ports,
3273 netdev_get_name(ofport->up.netdev));
3274 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3275 if (!ofport->tnl_port) {
3276 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3278 /* The caller is going to close ofport->up.netdev. If this is a
3279 * bonded port, then the bond is using that netdev, so remove it
3280 * from the bond. The client will need to reconfigure everything
3281 * after deleting ports, so then the slave will get re-added. */
3282 bundle_remove(&ofport->up);
3289 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3291 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3296 error = netdev_get_stats(ofport->up.netdev, stats);
3298 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3299 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3301 /* ofproto->stats.tx_packets represents packets that we created
3302 * internally and sent to some port (e.g. packets sent with
3303 * send_packet()). Account for them as if they had come from
3304 * OFPP_LOCAL and got forwarded. */
3306 if (stats->rx_packets != UINT64_MAX) {
3307 stats->rx_packets += ofproto->stats.tx_packets;
3310 if (stats->rx_bytes != UINT64_MAX) {
3311 stats->rx_bytes += ofproto->stats.tx_bytes;
3314 /* ofproto->stats.rx_packets represents packets that were received on
3315 * some port and we processed internally and dropped (e.g. STP).
3316 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3318 if (stats->tx_packets != UINT64_MAX) {
3319 stats->tx_packets += ofproto->stats.rx_packets;
3322 if (stats->tx_bytes != UINT64_MAX) {
3323 stats->tx_bytes += ofproto->stats.rx_bytes;
3330 /* Account packets for LOCAL port. */
3332 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3333 size_t tx_size, size_t rx_size)
3335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3338 ofproto->stats.rx_packets++;
3339 ofproto->stats.rx_bytes += rx_size;
3342 ofproto->stats.tx_packets++;
3343 ofproto->stats.tx_bytes += tx_size;
3347 struct port_dump_state {
3352 struct ofproto_port port;
3357 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3359 *statep = xzalloc(sizeof(struct port_dump_state));
3364 port_dump_next(const struct ofproto *ofproto_, void *state_,
3365 struct ofproto_port *port)
3367 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3368 struct port_dump_state *state = state_;
3369 const struct sset *sset;
3370 struct sset_node *node;
3372 if (state->has_port) {
3373 ofproto_port_destroy(&state->port);
3374 state->has_port = false;
3376 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3377 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3380 error = port_query_by_name(ofproto_, node->name, &state->port);
3382 *port = state->port;
3383 state->has_port = true;
3385 } else if (error != ENODEV) {
3390 if (!state->ghost) {
3391 state->ghost = true;
3394 return port_dump_next(ofproto_, state_, port);
3401 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3403 struct port_dump_state *state = state_;
3405 if (state->has_port) {
3406 ofproto_port_destroy(&state->port);
3413 port_poll(const struct ofproto *ofproto_, char **devnamep)
3415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3417 if (ofproto->port_poll_errno) {
3418 int error = ofproto->port_poll_errno;
3419 ofproto->port_poll_errno = 0;
3423 if (sset_is_empty(&ofproto->port_poll_set)) {
3427 *devnamep = sset_pop(&ofproto->port_poll_set);
3432 port_poll_wait(const struct ofproto *ofproto_)
3434 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3435 dpif_port_poll_wait(ofproto->backer->dpif);
3439 port_is_lacp_current(const struct ofport *ofport_)
3441 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3442 return (ofport->bundle && ofport->bundle->lacp
3443 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3447 /* Upcall handling. */
3449 /* Flow miss batching.
3451 * Some dpifs implement operations faster when you hand them off in a batch.
3452 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3453 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3454 * more packets, plus possibly installing the flow in the dpif.
3456 * So far we only batch the operations that affect flow setup time the most.
3457 * It's possible to batch more than that, but the benefit might be minimal. */
3459 struct hmap_node hmap_node;
3460 struct ofproto_dpif *ofproto;
3462 enum odp_key_fitness key_fitness;
3463 const struct nlattr *key;
3465 struct initial_vals initial_vals;
3466 struct list packets;
3467 enum dpif_upcall_type upcall_type;
3468 uint32_t odp_in_port;
3471 struct flow_miss_op {
3472 struct dpif_op dpif_op;
3473 void *garbage; /* Pointer to pass to free(), NULL if none. */
3474 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3477 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3478 * OpenFlow controller as necessary according to their individual
3479 * configurations. */
3481 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3482 const struct flow *flow)
3484 struct ofputil_packet_in pin;
3486 pin.packet = packet->data;
3487 pin.packet_len = packet->size;
3488 pin.reason = OFPR_NO_MATCH;
3489 pin.controller_id = 0;
3494 pin.send_len = 0; /* not used for flow table misses */
3496 flow_get_metadata(flow, &pin.fmd);
3498 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3501 static enum slow_path_reason
3502 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3503 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3507 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3509 cfm_process_heartbeat(ofport->cfm, packet);
3512 } else if (ofport->bundle && ofport->bundle->lacp
3513 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3515 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3518 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3520 stp_process_packet(ofport, packet);
3528 static struct flow_miss *
3529 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3530 const struct flow *flow, uint32_t hash)
3532 struct flow_miss *miss;
3534 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3535 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3543 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3544 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3545 * 'miss' is associated with a subfacet the caller must also initialize the
3546 * returned op->subfacet, and if anything needs to be freed after processing
3547 * the op, the caller must initialize op->garbage also. */
3549 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3550 struct flow_miss_op *op)
3552 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3553 /* This packet was received on a VLAN splinter port. We
3554 * added a VLAN to the packet to make the packet resemble
3555 * the flow, but the actions were composed assuming that
3556 * the packet contained no VLAN. So, we must remove the
3557 * VLAN header from the packet before trying to execute the
3559 eth_pop_vlan(packet);
3563 op->dpif_op.type = DPIF_OP_EXECUTE;
3564 op->dpif_op.u.execute.key = miss->key;
3565 op->dpif_op.u.execute.key_len = miss->key_len;
3566 op->dpif_op.u.execute.packet = packet;
3569 /* Helper for handle_flow_miss_without_facet() and
3570 * handle_flow_miss_with_facet(). */
3572 handle_flow_miss_common(struct rule_dpif *rule,
3573 struct ofpbuf *packet, const struct flow *flow)
3575 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3577 ofproto->n_matches++;
3579 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3581 * Extra-special case for fail-open mode.
3583 * We are in fail-open mode and the packet matched the fail-open
3584 * rule, but we are connected to a controller too. We should send
3585 * the packet up to the controller in the hope that it will try to
3586 * set up a flow and thereby allow us to exit fail-open.
3588 * See the top-level comment in fail-open.c for more information.
3590 send_packet_in_miss(ofproto, packet, flow);
3594 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3595 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3596 * installing a datapath flow. The answer is usually "yes" (a return value of
3597 * true). However, for short flows the cost of bookkeeping is much higher than
3598 * the benefits, so when the datapath holds a large number of flows we impose
3599 * some heuristics to decide which flows are likely to be worth tracking. */
3601 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3602 struct flow_miss *miss, uint32_t hash)
3604 if (!ofproto->governor) {
3607 n_subfacets = hmap_count(&ofproto->subfacets);
3608 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3612 ofproto->governor = governor_create(ofproto->up.name);
3615 return governor_should_install_flow(ofproto->governor, hash,
3616 list_size(&miss->packets));
3619 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3620 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3621 * increment '*n_ops'. */
3623 handle_flow_miss_without_facet(struct flow_miss *miss,
3624 struct rule_dpif *rule,
3625 struct flow_miss_op *ops, size_t *n_ops)
3627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3628 long long int now = time_msec();
3629 struct action_xlate_ctx ctx;
3630 struct ofpbuf *packet;
3632 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3633 struct flow_miss_op *op = &ops[*n_ops];
3634 struct dpif_flow_stats stats;
3635 struct ofpbuf odp_actions;
3637 COVERAGE_INC(facet_suppress);
3639 handle_flow_miss_common(rule, packet, &miss->flow);
3641 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3643 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3644 rule_credit_stats(rule, &stats);
3646 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, &miss->initial_vals,
3647 rule, stats.tcp_flags, packet);
3648 ctx.resubmit_stats = &stats;
3649 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3652 if (odp_actions.size) {
3653 struct dpif_execute *execute = &op->dpif_op.u.execute;
3655 init_flow_miss_execute_op(miss, packet, op);
3656 execute->actions = odp_actions.data;
3657 execute->actions_len = odp_actions.size;
3658 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3662 ofpbuf_uninit(&odp_actions);
3667 /* Handles 'miss', which matches 'facet'. May add any required datapath
3668 * operations to 'ops', incrementing '*n_ops' for each new op.
3670 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3671 * This is really important only for new facets: if we just called time_msec()
3672 * here, then the new subfacet or its packets could look (occasionally) as
3673 * though it was used some time after the facet was used. That can make a
3674 * one-packet flow look like it has a nonzero duration, which looks odd in
3675 * e.g. NetFlow statistics. */
3677 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3679 struct flow_miss_op *ops, size_t *n_ops)
3681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3682 enum subfacet_path want_path;
3683 struct subfacet *subfacet;
3684 struct ofpbuf *packet;
3686 subfacet = subfacet_create(facet, miss, now);
3688 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3689 struct flow_miss_op *op = &ops[*n_ops];
3690 struct dpif_flow_stats stats;
3692 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3694 if (!subfacet->actions) {
3695 subfacet_make_actions(subfacet, packet);
3696 } else if (subfacet->slow) {
3697 struct action_xlate_ctx ctx;
3699 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
3700 &subfacet->initial_vals, facet->rule, 0,
3702 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
3703 facet->rule->up.ofpacts_len);
3706 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3707 subfacet_update_stats(subfacet, &stats);
3709 if (subfacet->actions_len) {
3710 struct dpif_execute *execute = &op->dpif_op.u.execute;
3712 init_flow_miss_execute_op(miss, packet, op);
3713 execute->actions = subfacet->actions;
3714 execute->actions_len = subfacet->actions_len;
3720 want_path = subfacet_want_path(subfacet->slow);
3721 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3722 struct flow_miss_op *op = &ops[(*n_ops)++];
3723 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3725 subfacet->path = want_path;
3728 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3729 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3730 put->key = miss->key;
3731 put->key_len = miss->key_len;
3732 if (want_path == SF_FAST_PATH) {
3733 put->actions = subfacet->actions;
3734 put->actions_len = subfacet->actions_len;
3736 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3737 op->stub, sizeof op->stub,
3738 &put->actions, &put->actions_len);
3744 /* Handles flow miss 'miss'. May add any required datapath operations
3745 * to 'ops', incrementing '*n_ops' for each new op. */
3747 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3750 struct ofproto_dpif *ofproto = miss->ofproto;
3751 struct facet *facet;
3755 /* The caller must ensure that miss->hmap_node.hash contains
3756 * flow_hash(miss->flow, 0). */
3757 hash = miss->hmap_node.hash;
3759 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3761 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3763 /* There does not exist a bijection between 'struct flow' and datapath
3764 * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental
3765 * assumption used throughout the facet and subfacet handling code.
3766 * Since we have to handle these misses in userspace anyway, we simply
3767 * skip facet creation, avoiding the problem alltogether. */
3768 if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3769 || !flow_miss_should_make_facet(ofproto, miss, hash)) {
3770 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3774 facet = facet_create(rule, &miss->flow, hash);
3779 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3782 static struct drop_key *
3783 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3786 struct drop_key *drop_key;
3788 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3789 &backer->drop_keys) {
3790 if (drop_key->key_len == key_len
3791 && !memcmp(drop_key->key, key, key_len)) {
3799 drop_key_clear(struct dpif_backer *backer)
3801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3802 struct drop_key *drop_key, *next;
3804 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3807 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3809 if (error && !VLOG_DROP_WARN(&rl)) {
3810 struct ds ds = DS_EMPTY_INITIALIZER;
3811 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3812 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3817 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3818 free(drop_key->key);
3823 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3824 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3825 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3826 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3827 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3828 * 'packet' ingressed.
3830 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3831 * 'flow''s in_port to OFPP_NONE.
3833 * This function does post-processing on data returned from
3834 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3835 * of the upcall processing logic. In particular, if the extracted in_port is
3836 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3837 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3838 * a VLAN header onto 'packet' (if it is nonnull).
3840 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3841 * to the VLAN TCI with which the packet was really received, that is, the
3842 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3843 * the value returned in flow->vlan_tci only for packets received on
3844 * VLAN splinters.) Also, if received on an IP tunnel, sets
3845 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3847 * Similarly, this function also includes some logic to help with tunnels. It
3848 * may modify 'flow' as necessary to make the tunneling implementation
3849 * transparent to the upcall processing logic.
3851 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3852 * or some other positive errno if there are other problems. */
3854 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3855 const struct nlattr *key, size_t key_len,
3856 struct flow *flow, enum odp_key_fitness *fitnessp,
3857 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3858 struct initial_vals *initial_vals)
3860 const struct ofport_dpif *port;
3861 enum odp_key_fitness fitness;
3864 fitness = odp_flow_key_to_flow(key, key_len, flow);
3865 if (fitness == ODP_FIT_ERROR) {
3871 initial_vals->vlan_tci = flow->vlan_tci;
3872 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3876 *odp_in_port = flow->in_port;
3879 if (tnl_port_should_receive(flow)) {
3880 const struct ofport *ofport = tnl_port_receive(flow);
3882 flow->in_port = OFPP_NONE;
3885 port = ofport_dpif_cast(ofport);
3887 /* We can't reproduce 'key' from 'flow'. */
3888 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3890 /* XXX: Since the tunnel module is not scoped per backer, it's
3891 * theoretically possible that we'll receive an ofport belonging to an
3892 * entirely different datapath. In practice, this can't happen because
3893 * no platforms has two separate datapaths which each support
3895 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3897 port = odp_port_to_ofport(backer, flow->in_port);
3899 flow->in_port = OFPP_NONE;
3903 flow->in_port = port->up.ofp_port;
3904 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3906 /* Make the packet resemble the flow, so that it gets sent to
3907 * an OpenFlow controller properly, so that it looks correct
3908 * for sFlow, and so that flow_extract() will get the correct
3909 * vlan_tci if it is called on 'packet'.
3911 * The allocated space inside 'packet' probably also contains
3912 * 'key', that is, both 'packet' and 'key' are probably part of
3913 * a struct dpif_upcall (see the large comment on that
3914 * structure definition), so pushing data on 'packet' is in
3915 * general not a good idea since it could overwrite 'key' or
3916 * free it as a side effect. However, it's OK in this special
3917 * case because we know that 'packet' is inside a Netlink
3918 * attribute: pushing 4 bytes will just overwrite the 4-byte
3919 * "struct nlattr", which is fine since we don't need that
3920 * header anymore. */
3921 eth_push_vlan(packet, flow->vlan_tci);
3923 /* We can't reproduce 'key' from 'flow'. */
3924 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3930 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3935 *fitnessp = fitness;
3941 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3944 struct dpif_upcall *upcall;
3945 struct flow_miss *miss;
3946 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3947 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3948 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3958 /* Construct the to-do list.
3960 * This just amounts to extracting the flow from each packet and sticking
3961 * the packets that have the same flow in the same "flow_miss" structure so
3962 * that we can process them together. */
3965 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3966 struct flow_miss *miss = &misses[n_misses];
3967 struct flow_miss *existing_miss;
3968 struct ofproto_dpif *ofproto;
3969 uint32_t odp_in_port;
3974 error = ofproto_receive(backer, upcall->packet, upcall->key,
3975 upcall->key_len, &flow, &miss->key_fitness,
3976 &ofproto, &odp_in_port, &miss->initial_vals);
3977 if (error == ENODEV) {
3978 struct drop_key *drop_key;
3980 /* Received packet on port for which we couldn't associate
3981 * an ofproto. This can happen if a port is removed while
3982 * traffic is being received. Print a rate-limited message
3983 * in case it happens frequently. Install a drop flow so
3984 * that future packets of the flow are inexpensively dropped
3986 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3989 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3991 drop_key = xmalloc(sizeof *drop_key);
3992 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3993 drop_key->key_len = upcall->key_len;
3995 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3996 hash_bytes(drop_key->key, drop_key->key_len, 0));
3997 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3998 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4006 ofproto->n_missed++;
4007 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4008 &flow.tunnel, flow.in_port, &miss->flow);
4010 /* Add other packets to a to-do list. */
4011 hash = flow_hash(&miss->flow, 0);
4012 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4013 if (!existing_miss) {
4014 hmap_insert(&todo, &miss->hmap_node, hash);
4015 miss->ofproto = ofproto;
4016 miss->key = upcall->key;
4017 miss->key_len = upcall->key_len;
4018 miss->upcall_type = upcall->type;
4019 miss->odp_in_port = odp_in_port;
4020 list_init(&miss->packets);
4024 miss = existing_miss;
4026 list_push_back(&miss->packets, &upcall->packet->list_node);
4029 /* Process each element in the to-do list, constructing the set of
4030 * operations to batch. */
4032 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4033 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4035 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4037 /* Execute batch. */
4038 for (i = 0; i < n_ops; i++) {
4039 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4041 dpif_operate(backer->dpif, dpif_ops, n_ops);
4044 for (i = 0; i < n_ops; i++) {
4045 free(flow_miss_ops[i].garbage);
4047 hmap_destroy(&todo);
4050 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4052 classify_upcall(const struct dpif_upcall *upcall)
4054 size_t userdata_len;
4055 union user_action_cookie cookie;
4057 /* First look at the upcall type. */
4058 switch (upcall->type) {
4059 case DPIF_UC_ACTION:
4065 case DPIF_N_UC_TYPES:
4067 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4071 /* "action" upcalls need a closer look. */
4072 if (!upcall->userdata) {
4073 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4076 userdata_len = nl_attr_get_size(upcall->userdata);
4077 if (userdata_len < sizeof cookie.type
4078 || userdata_len > sizeof cookie) {
4079 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4083 memset(&cookie, 0, sizeof cookie);
4084 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4085 if (userdata_len == sizeof cookie.sflow
4086 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4087 return SFLOW_UPCALL;
4088 } else if (userdata_len == sizeof cookie.slow_path
4089 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4091 } else if (userdata_len == sizeof cookie.flow_sample
4092 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4093 return FLOW_SAMPLE_UPCALL;
4094 } else if (userdata_len == sizeof cookie.ipfix
4095 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4096 return IPFIX_UPCALL;
4098 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4099 " and size %zu", cookie.type, userdata_len);
4105 handle_sflow_upcall(struct dpif_backer *backer,
4106 const struct dpif_upcall *upcall)
4108 struct ofproto_dpif *ofproto;
4109 union user_action_cookie cookie;
4111 uint32_t odp_in_port;
4113 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4114 &flow, NULL, &ofproto, &odp_in_port, NULL)
4115 || !ofproto->sflow) {
4119 memset(&cookie, 0, sizeof cookie);
4120 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4121 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4122 odp_in_port, &cookie);
4126 handle_flow_sample_upcall(struct dpif_backer *backer,
4127 const struct dpif_upcall *upcall)
4129 struct ofproto_dpif *ofproto;
4130 union user_action_cookie cookie;
4133 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4134 &flow, NULL, &ofproto, NULL, NULL)
4135 || !ofproto->ipfix) {
4139 memset(&cookie, 0, sizeof cookie);
4140 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4142 /* The flow reflects exactly the contents of the packet. Sample
4143 * the packet using it. */
4144 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4145 cookie.flow_sample.collector_set_id,
4146 cookie.flow_sample.probability,
4147 cookie.flow_sample.obs_domain_id,
4148 cookie.flow_sample.obs_point_id);
4152 handle_ipfix_upcall(struct dpif_backer *backer,
4153 const struct dpif_upcall *upcall)
4155 struct ofproto_dpif *ofproto;
4158 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4159 &flow, NULL, &ofproto, NULL, NULL)
4160 || !ofproto->ipfix) {
4164 /* The flow reflects exactly the contents of the packet. Sample
4165 * the packet using it. */
4166 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4170 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4172 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4173 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4174 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4179 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4182 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4183 struct dpif_upcall *upcall = &misses[n_misses];
4184 struct ofpbuf *buf = &miss_bufs[n_misses];
4187 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4188 sizeof miss_buf_stubs[n_misses]);
4189 error = dpif_recv(backer->dpif, upcall, buf);
4195 switch (classify_upcall(upcall)) {
4197 /* Handle it later. */
4202 handle_sflow_upcall(backer, upcall);
4206 case FLOW_SAMPLE_UPCALL:
4207 handle_flow_sample_upcall(backer, upcall);
4212 handle_ipfix_upcall(backer, upcall);
4222 /* Handle deferred MISS_UPCALL processing. */
4223 handle_miss_upcalls(backer, misses, n_misses);
4224 for (i = 0; i < n_misses; i++) {
4225 ofpbuf_uninit(&miss_bufs[i]);
4231 /* Flow expiration. */
4233 static int subfacet_max_idle(const struct ofproto_dpif *);
4234 static void update_stats(struct dpif_backer *);
4235 static void rule_expire(struct rule_dpif *);
4236 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4238 /* This function is called periodically by run(). Its job is to collect
4239 * updates for the flows that have been installed into the datapath, most
4240 * importantly when they last were used, and then use that information to
4241 * expire flows that have not been used recently.
4243 * Returns the number of milliseconds after which it should be called again. */
4245 expire(struct dpif_backer *backer)
4247 struct ofproto_dpif *ofproto;
4248 int max_idle = INT32_MAX;
4250 /* Periodically clear out the drop keys in an effort to keep them
4251 * relatively few. */
4252 drop_key_clear(backer);
4254 /* Update stats for each flow in the backer. */
4255 update_stats(backer);
4257 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4258 struct rule *rule, *next_rule;
4261 if (ofproto->backer != backer) {
4265 /* Keep track of the max number of flows per ofproto_dpif. */
4266 update_max_subfacet_count(ofproto);
4268 /* Expire subfacets that have been idle too long. */
4269 dp_max_idle = subfacet_max_idle(ofproto);
4270 expire_subfacets(ofproto, dp_max_idle);
4272 max_idle = MIN(max_idle, dp_max_idle);
4274 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4276 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4277 &ofproto->up.expirable) {
4278 rule_expire(rule_dpif_cast(rule));
4281 /* All outstanding data in existing flows has been accounted, so it's a
4282 * good time to do bond rebalancing. */
4283 if (ofproto->has_bonded_bundles) {
4284 struct ofbundle *bundle;
4286 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4288 bond_rebalance(bundle->bond, &backer->revalidate_set);
4294 return MIN(max_idle, 1000);
4297 /* Updates flow table statistics given that the datapath just reported 'stats'
4298 * as 'subfacet''s statistics. */
4300 update_subfacet_stats(struct subfacet *subfacet,
4301 const struct dpif_flow_stats *stats)
4303 struct facet *facet = subfacet->facet;
4305 if (stats->n_packets >= subfacet->dp_packet_count) {
4306 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4307 facet->packet_count += extra;
4309 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4312 if (stats->n_bytes >= subfacet->dp_byte_count) {
4313 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4315 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4318 subfacet->dp_packet_count = stats->n_packets;
4319 subfacet->dp_byte_count = stats->n_bytes;
4321 facet->tcp_flags |= stats->tcp_flags;
4323 subfacet_update_time(subfacet, stats->used);
4324 if (facet->accounted_bytes < facet->byte_count) {
4326 facet_account(facet);
4327 facet->accounted_bytes = facet->byte_count;
4331 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4332 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4334 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4335 const struct nlattr *key, size_t key_len)
4337 if (!VLOG_DROP_WARN(&rl)) {
4341 odp_flow_key_format(key, key_len, &s);
4342 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4346 COVERAGE_INC(facet_unexpected);
4347 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4350 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4352 * This function also pushes statistics updates to rules which each facet
4353 * resubmits into. Generally these statistics will be accurate. However, if a
4354 * facet changes the rule it resubmits into at some time in between
4355 * update_stats() runs, it is possible that statistics accrued to the
4356 * old rule will be incorrectly attributed to the new rule. This could be
4357 * avoided by calling update_stats() whenever rules are created or
4358 * deleted. However, the performance impact of making so many calls to the
4359 * datapath do not justify the benefit of having perfectly accurate statistics.
4361 * In addition, this function maintains per ofproto flow hit counts. The patch
4362 * port is not treated specially. e.g. A packet ingress from br0 patched into
4363 * br1 will increase the hit count of br0 by 1, however, does not affect
4364 * the hit or miss counts of br1.
4367 update_stats(struct dpif_backer *backer)
4369 const struct dpif_flow_stats *stats;
4370 struct dpif_flow_dump dump;
4371 const struct nlattr *key;
4372 struct ofproto_dpif *ofproto;
4375 dpif_flow_dump_start(&dump, backer->dpif);
4376 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4378 struct subfacet *subfacet;
4379 struct ofport_dpif *ofport;
4382 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4387 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4388 ofproto->n_update_stats++;
4390 ofport = get_ofp_port(ofproto, flow.in_port);
4391 if (ofport && ofport->tnl_port) {
4392 netdev_vport_inc_rx(ofport->up.netdev, stats);
4395 key_hash = odp_flow_key_hash(key, key_len);
4396 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4397 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4399 /* Update ofproto_dpif's hit count. */
4400 if (stats->n_packets > subfacet->dp_packet_count) {
4401 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4402 dpif_stats_update_hit_count(ofproto, delta);
4405 update_subfacet_stats(subfacet, stats);
4409 /* Stats are updated per-packet. */
4412 case SF_NOT_INSTALLED:
4414 delete_unexpected_flow(ofproto, key, key_len);
4419 dpif_flow_dump_done(&dump);
4421 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4422 update_moving_averages(ofproto);
4427 /* Calculates and returns the number of milliseconds of idle time after which
4428 * subfacets should expire from the datapath. When a subfacet expires, we fold
4429 * its statistics into its facet, and when a facet's last subfacet expires, we
4430 * fold its statistic into its rule. */
4432 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4435 * Idle time histogram.
4437 * Most of the time a switch has a relatively small number of subfacets.
4438 * When this is the case we might as well keep statistics for all of them
4439 * in userspace and to cache them in the kernel datapath for performance as
4442 * As the number of subfacets increases, the memory required to maintain
4443 * statistics about them in userspace and in the kernel becomes
4444 * significant. However, with a large number of subfacets it is likely
4445 * that only a few of them are "heavy hitters" that consume a large amount
4446 * of bandwidth. At this point, only heavy hitters are worth caching in
4447 * the kernel and maintaining in userspaces; other subfacets we can
4450 * The technique used to compute the idle time is to build a histogram with
4451 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4452 * that is installed in the kernel gets dropped in the appropriate bucket.
4453 * After the histogram has been built, we compute the cutoff so that only
4454 * the most-recently-used 1% of subfacets (but at least
4455 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4456 * the most-recently-used bucket of subfacets is kept, so actually an
4457 * arbitrary number of subfacets can be kept in any given expiration run
4458 * (though the next run will delete most of those unless they receive
4461 * This requires a second pass through the subfacets, in addition to the
4462 * pass made by update_stats(), because the former function never looks at
4463 * uninstallable subfacets.
4465 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4466 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4467 int buckets[N_BUCKETS] = { 0 };
4468 int total, subtotal, bucket;
4469 struct subfacet *subfacet;
4473 total = hmap_count(&ofproto->subfacets);
4474 if (total <= ofproto->up.flow_eviction_threshold) {
4475 return N_BUCKETS * BUCKET_WIDTH;
4478 /* Build histogram. */
4480 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4481 long long int idle = now - subfacet->used;
4482 int bucket = (idle <= 0 ? 0
4483 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4484 : (unsigned int) idle / BUCKET_WIDTH);
4488 /* Find the first bucket whose flows should be expired. */
4489 subtotal = bucket = 0;
4491 subtotal += buckets[bucket++];
4492 } while (bucket < N_BUCKETS &&
4493 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4495 if (VLOG_IS_DBG_ENABLED()) {
4499 ds_put_cstr(&s, "keep");
4500 for (i = 0; i < N_BUCKETS; i++) {
4502 ds_put_cstr(&s, ", drop");
4505 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4508 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4512 return bucket * BUCKET_WIDTH;
4516 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4518 /* Cutoff time for most flows. */
4519 long long int normal_cutoff = time_msec() - dp_max_idle;
4521 /* We really want to keep flows for special protocols around, so use a more
4522 * conservative cutoff. */
4523 long long int special_cutoff = time_msec() - 10000;
4525 struct subfacet *subfacet, *next_subfacet;
4526 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4530 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4531 &ofproto->subfacets) {
4532 long long int cutoff;
4534 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4537 if (subfacet->used < cutoff) {
4538 if (subfacet->path != SF_NOT_INSTALLED) {
4539 batch[n_batch++] = subfacet;
4540 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4541 subfacet_destroy_batch(ofproto, batch, n_batch);
4545 subfacet_destroy(subfacet);
4551 subfacet_destroy_batch(ofproto, batch, n_batch);
4555 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4556 * then delete it entirely. */
4558 rule_expire(struct rule_dpif *rule)
4560 struct facet *facet, *next_facet;
4564 if (rule->up.pending) {
4565 /* We'll have to expire it later. */
4569 /* Has 'rule' expired? */
4571 if (rule->up.hard_timeout
4572 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4573 reason = OFPRR_HARD_TIMEOUT;
4574 } else if (rule->up.idle_timeout
4575 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4576 reason = OFPRR_IDLE_TIMEOUT;
4581 COVERAGE_INC(ofproto_dpif_expired);
4583 /* Update stats. (This is a no-op if the rule expired due to an idle
4584 * timeout, because that only happens when the rule has no facets left.) */
4585 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4586 facet_remove(facet);
4589 /* Get rid of the rule. */
4590 ofproto_rule_expire(&rule->up, reason);
4595 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4597 * The caller must already have determined that no facet with an identical
4598 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4599 * the ofproto's classifier table.
4601 * 'hash' must be the return value of flow_hash(flow, 0).
4603 * The facet will initially have no subfacets. The caller should create (at
4604 * least) one subfacet with subfacet_create(). */
4605 static struct facet *
4606 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4608 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4609 struct facet *facet;
4611 facet = xzalloc(sizeof *facet);
4612 facet->used = time_msec();
4613 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4614 list_push_back(&rule->facets, &facet->list_node);
4616 facet->flow = *flow;
4617 list_init(&facet->subfacets);
4618 netflow_flow_init(&facet->nf_flow);
4619 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4621 facet->learn_rl = time_msec() + 500;
4627 facet_free(struct facet *facet)
4632 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4633 * 'packet', which arrived on 'in_port'. */
4635 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4636 const struct nlattr *odp_actions, size_t actions_len,
4637 struct ofpbuf *packet)
4639 struct odputil_keybuf keybuf;
4643 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4644 odp_flow_key_from_flow(&key, flow,
4645 ofp_port_to_odp_port(ofproto, flow->in_port));
4647 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4648 odp_actions, actions_len, packet);
4652 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4654 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4655 * rule's statistics, via subfacet_uninstall().
4657 * - Removes 'facet' from its rule and from ofproto->facets.
4660 facet_remove(struct facet *facet)
4662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4663 struct subfacet *subfacet, *next_subfacet;
4665 ovs_assert(!list_is_empty(&facet->subfacets));
4667 /* First uninstall all of the subfacets to get final statistics. */
4668 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4669 subfacet_uninstall(subfacet);
4672 /* Flush the final stats to the rule.
4674 * This might require us to have at least one subfacet around so that we
4675 * can use its actions for accounting in facet_account(), which is why we
4676 * have uninstalled but not yet destroyed the subfacets. */
4677 facet_flush_stats(facet);
4679 /* Now we're really all done so destroy everything. */
4680 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4681 &facet->subfacets) {
4682 subfacet_destroy__(subfacet);
4684 hmap_remove(&ofproto->facets, &facet->hmap_node);
4685 list_remove(&facet->list_node);
4689 /* Feed information from 'facet' back into the learning table to keep it in
4690 * sync with what is actually flowing through the datapath. */
4692 facet_learn(struct facet *facet)
4694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4695 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4696 struct subfacet, list_node);
4697 long long int now = time_msec();
4698 struct action_xlate_ctx ctx;
4700 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4704 facet->learn_rl = now + 500;
4706 if (!facet->has_learn
4707 && !facet->has_normal
4708 && (!facet->has_fin_timeout
4709 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4713 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4714 &subfacet->initial_vals,
4715 facet->rule, facet->tcp_flags, NULL);
4716 ctx.may_learn = true;
4717 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4718 facet->rule->up.ofpacts_len);
4722 facet_account(struct facet *facet)
4724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4725 struct subfacet *subfacet = facet_get_subfacet(facet);
4726 const struct nlattr *a;
4731 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4734 n_bytes = facet->byte_count - facet->accounted_bytes;
4736 /* This loop feeds byte counters to bond_account() for rebalancing to use
4737 * as a basis. We also need to track the actual VLAN on which the packet
4738 * is going to be sent to ensure that it matches the one passed to
4739 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4742 * We use the actions from an arbitrary subfacet because they should all
4743 * be equally valid for our purpose. */
4744 vlan_tci = facet->flow.vlan_tci;
4745 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4746 subfacet->actions, subfacet->actions_len) {
4747 const struct ovs_action_push_vlan *vlan;
4748 struct ofport_dpif *port;
4750 switch (nl_attr_type(a)) {
4751 case OVS_ACTION_ATTR_OUTPUT:
4752 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4753 if (port && port->bundle && port->bundle->bond) {
4754 bond_account(port->bundle->bond, &facet->flow,
4755 vlan_tci_to_vid(vlan_tci), n_bytes);
4759 case OVS_ACTION_ATTR_POP_VLAN:
4760 vlan_tci = htons(0);
4763 case OVS_ACTION_ATTR_PUSH_VLAN:
4764 vlan = nl_attr_get(a);
4765 vlan_tci = vlan->vlan_tci;
4771 /* Returns true if the only action for 'facet' is to send to the controller.
4772 * (We don't report NetFlow expiration messages for such facets because they
4773 * are just part of the control logic for the network, not real traffic). */
4775 facet_is_controller_flow(struct facet *facet)
4778 const struct rule *rule = &facet->rule->up;
4779 const struct ofpact *ofpacts = rule->ofpacts;
4780 size_t ofpacts_len = rule->ofpacts_len;
4782 if (ofpacts_len > 0 &&
4783 ofpacts->type == OFPACT_CONTROLLER &&
4784 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4791 /* Folds all of 'facet''s statistics into its rule. Also updates the
4792 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4793 * 'facet''s statistics in the datapath should have been zeroed and folded into
4794 * its packet and byte counts before this function is called. */
4796 facet_flush_stats(struct facet *facet)
4798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4799 struct subfacet *subfacet;
4801 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4802 ovs_assert(!subfacet->dp_byte_count);
4803 ovs_assert(!subfacet->dp_packet_count);
4806 facet_push_stats(facet);
4807 if (facet->accounted_bytes < facet->byte_count) {
4808 facet_account(facet);
4809 facet->accounted_bytes = facet->byte_count;
4812 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4813 struct ofexpired expired;
4814 expired.flow = facet->flow;
4815 expired.packet_count = facet->packet_count;
4816 expired.byte_count = facet->byte_count;
4817 expired.used = facet->used;
4818 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4821 facet->rule->packet_count += facet->packet_count;
4822 facet->rule->byte_count += facet->byte_count;
4824 /* Reset counters to prevent double counting if 'facet' ever gets
4826 facet_reset_counters(facet);
4828 netflow_flow_clear(&facet->nf_flow);
4829 facet->tcp_flags = 0;
4832 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4833 * Returns it if found, otherwise a null pointer.
4835 * 'hash' must be the return value of flow_hash(flow, 0).
4837 * The returned facet might need revalidation; use facet_lookup_valid()
4838 * instead if that is important. */
4839 static struct facet *
4840 facet_find(struct ofproto_dpif *ofproto,
4841 const struct flow *flow, uint32_t hash)
4843 struct facet *facet;
4845 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4846 if (flow_equal(flow, &facet->flow)) {
4854 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4855 * Returns it if found, otherwise a null pointer.
4857 * 'hash' must be the return value of flow_hash(flow, 0).
4859 * The returned facet is guaranteed to be valid. */
4860 static struct facet *
4861 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4864 struct facet *facet;
4866 facet = facet_find(ofproto, flow, hash);
4868 && (ofproto->backer->need_revalidate
4869 || tag_set_intersects(&ofproto->backer->revalidate_set,
4871 && !facet_revalidate(facet)) {
4878 /* Return a subfacet from 'facet'. A facet consists of one or more
4879 * subfacets, and this function returns one of them. */
4880 static struct subfacet *facet_get_subfacet(struct facet *facet)
4882 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4887 subfacet_path_to_string(enum subfacet_path path)
4890 case SF_NOT_INSTALLED:
4891 return "not installed";
4893 return "in fast path";
4895 return "in slow path";
4901 /* Returns the path in which a subfacet should be installed if its 'slow'
4902 * member has the specified value. */
4903 static enum subfacet_path
4904 subfacet_want_path(enum slow_path_reason slow)
4906 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4909 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4910 * supposing that its actions have been recalculated as 'want_actions' and that
4911 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4913 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4914 const struct ofpbuf *want_actions)
4916 enum subfacet_path want_path = subfacet_want_path(slow);
4917 return (want_path != subfacet->path
4918 || (want_path == SF_FAST_PATH
4919 && (subfacet->actions_len != want_actions->size
4920 || memcmp(subfacet->actions, want_actions->data,
4921 subfacet->actions_len))));
4925 facet_check_consistency(struct facet *facet)
4927 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4929 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4931 uint64_t odp_actions_stub[1024 / 8];
4932 struct ofpbuf odp_actions;
4934 struct rule_dpif *rule;
4935 struct subfacet *subfacet;
4936 bool may_log = false;
4939 /* Check the rule for consistency. */
4940 rule = rule_dpif_lookup(ofproto, &facet->flow);
4941 ok = rule == facet->rule;
4943 may_log = !VLOG_DROP_WARN(&rl);
4948 flow_format(&s, &facet->flow);
4949 ds_put_format(&s, ": facet associated with wrong rule (was "
4950 "table=%"PRIu8",", facet->rule->up.table_id);
4951 cls_rule_format(&facet->rule->up.cr, &s);
4952 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4954 cls_rule_format(&rule->up.cr, &s);
4955 ds_put_char(&s, ')');
4957 VLOG_WARN("%s", ds_cstr(&s));
4962 /* Check the datapath actions for consistency. */
4963 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4964 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4965 enum subfacet_path want_path;
4966 struct action_xlate_ctx ctx;
4969 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4970 &subfacet->initial_vals, rule, 0, NULL);
4971 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4974 if (subfacet->path == SF_NOT_INSTALLED) {
4975 /* This only happens if the datapath reported an error when we
4976 * tried to install the flow. Don't flag another error here. */
4980 want_path = subfacet_want_path(subfacet->slow);
4982 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4986 /* Inconsistency! */
4988 may_log = !VLOG_DROP_WARN(&rl);
4992 /* Rate-limited, skip reporting. */
4997 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4999 ds_put_cstr(&s, ": inconsistency in subfacet");
5000 if (want_path != subfacet->path) {
5001 enum odp_key_fitness fitness = subfacet->key_fitness;
5003 ds_put_format(&s, " (%s, fitness=%s)",
5004 subfacet_path_to_string(subfacet->path),
5005 odp_key_fitness_to_string(fitness));
5006 ds_put_format(&s, " (should have been %s)",
5007 subfacet_path_to_string(want_path));
5008 } else if (want_path == SF_FAST_PATH) {
5009 ds_put_cstr(&s, " (actions were: ");
5010 format_odp_actions(&s, subfacet->actions,
5011 subfacet->actions_len);
5012 ds_put_cstr(&s, ") (correct actions: ");
5013 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5014 ds_put_char(&s, ')');
5016 ds_put_cstr(&s, " (actions: ");
5017 format_odp_actions(&s, subfacet->actions,
5018 subfacet->actions_len);
5019 ds_put_char(&s, ')');
5021 VLOG_WARN("%s", ds_cstr(&s));
5024 ofpbuf_uninit(&odp_actions);
5029 /* Re-searches the classifier for 'facet':
5031 * - If the rule found is different from 'facet''s current rule, moves
5032 * 'facet' to the new rule and recompiles its actions.
5034 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5035 * where it is and recompiles its actions anyway.
5037 * - If any of 'facet''s subfacets correspond to a new flow according to
5038 * ofproto_receive(), 'facet' is removed.
5040 * Returns true if 'facet' is still valid. False if 'facet' was removed. */
5042 facet_revalidate(struct facet *facet)
5044 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5046 struct nlattr *odp_actions;
5049 struct actions *new_actions;
5051 struct action_xlate_ctx ctx;
5052 uint64_t odp_actions_stub[1024 / 8];
5053 struct ofpbuf odp_actions;
5055 struct rule_dpif *new_rule;
5056 struct subfacet *subfacet;
5059 COVERAGE_INC(facet_revalidate);
5061 /* Check that child subfacets still correspond to this facet. Tunnel
5062 * configuration changes could cause a subfacet's OpenFlow in_port to
5064 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5065 struct ofproto_dpif *recv_ofproto;
5066 struct flow recv_flow;
5069 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5070 subfacet->key_len, &recv_flow, NULL,
5071 &recv_ofproto, NULL, NULL);
5073 || recv_ofproto != ofproto
5074 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5075 facet_remove(facet);
5080 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5082 /* Calculate new datapath actions.
5084 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5085 * emit a NetFlow expiration and, if so, we need to have the old state
5086 * around to properly compose it. */
5088 /* If the datapath actions changed or the installability changed,
5089 * then we need to talk to the datapath. */
5092 memset(&ctx, 0, sizeof ctx);
5093 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5094 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5095 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5096 &subfacet->initial_vals, new_rule, 0, NULL);
5097 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5100 if (subfacet_should_install(subfacet, ctx.slow, &odp_actions)) {
5101 struct dpif_flow_stats stats;
5103 subfacet_install(subfacet, odp_actions.data, odp_actions.size,
5105 subfacet_update_stats(subfacet, &stats);
5108 new_actions = xcalloc(list_size(&facet->subfacets),
5109 sizeof *new_actions);
5111 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5113 new_actions[i].actions_len = odp_actions.size;
5118 ofpbuf_uninit(&odp_actions);
5121 facet_flush_stats(facet);
5124 /* Update 'facet' now that we've taken care of all the old state. */
5125 facet->tags = ctx.tags;
5126 facet->nf_flow.output_iface = ctx.nf_output_iface;
5127 facet->has_learn = ctx.has_learn;
5128 facet->has_normal = ctx.has_normal;
5129 facet->has_fin_timeout = ctx.has_fin_timeout;
5130 facet->mirrors = ctx.mirrors;
5133 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5134 subfacet->slow = ctx.slow;
5136 if (new_actions && new_actions[i].odp_actions) {
5137 free(subfacet->actions);
5138 subfacet->actions = new_actions[i].odp_actions;
5139 subfacet->actions_len = new_actions[i].actions_len;
5145 if (facet->rule != new_rule) {
5146 COVERAGE_INC(facet_changed_rule);
5147 list_remove(&facet->list_node);
5148 list_push_back(&new_rule->facets, &facet->list_node);
5149 facet->rule = new_rule;
5150 facet->used = new_rule->up.created;
5151 facet->prev_used = facet->used;
5157 /* Updates 'facet''s used time. Caller is responsible for calling
5158 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5160 facet_update_time(struct facet *facet, long long int used)
5162 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5163 if (used > facet->used) {
5165 ofproto_rule_update_used(&facet->rule->up, used);
5166 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5171 facet_reset_counters(struct facet *facet)
5173 facet->packet_count = 0;
5174 facet->byte_count = 0;
5175 facet->prev_packet_count = 0;
5176 facet->prev_byte_count = 0;
5177 facet->accounted_bytes = 0;
5181 facet_push_stats(struct facet *facet)
5183 struct dpif_flow_stats stats;
5185 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5186 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5187 ovs_assert(facet->used >= facet->prev_used);
5189 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5190 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5191 stats.used = facet->used;
5192 stats.tcp_flags = 0;
5194 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5195 facet->prev_packet_count = facet->packet_count;
5196 facet->prev_byte_count = facet->byte_count;
5197 facet->prev_used = facet->used;
5199 flow_push_stats(facet, &stats);
5201 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5202 facet->mirrors, stats.n_packets, stats.n_bytes);
5207 push_all_stats__(bool run_fast)
5209 static long long int rl = LLONG_MIN;
5210 struct ofproto_dpif *ofproto;
5212 if (time_msec() < rl) {
5216 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5217 struct facet *facet;
5219 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5220 facet_push_stats(facet);
5227 rl = time_msec() + 100;
5231 push_all_stats(void)
5233 push_all_stats__(true);
5237 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5239 rule->packet_count += stats->n_packets;
5240 rule->byte_count += stats->n_bytes;
5241 ofproto_rule_update_used(&rule->up, stats->used);
5244 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5245 * into given 'facet->rule''s actions and mirrors. */
5247 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5249 struct rule_dpif *rule = facet->rule;
5250 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5251 struct subfacet *subfacet = facet_get_subfacet(facet);
5252 struct action_xlate_ctx ctx;
5254 ofproto_rule_update_used(&rule->up, stats->used);
5256 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5257 &subfacet->initial_vals, rule, 0, NULL);
5258 ctx.resubmit_stats = stats;
5259 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5260 rule->up.ofpacts_len);
5265 static struct subfacet *
5266 subfacet_find(struct ofproto_dpif *ofproto,
5267 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5269 struct subfacet *subfacet;
5271 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5272 &ofproto->subfacets) {
5273 if (subfacet->key_len == key_len
5274 && !memcmp(key, subfacet->key, key_len)) {
5282 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5283 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5284 * existing subfacet if there is one, otherwise creates and returns a
5287 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5288 * which case the caller must populate the actions with
5289 * subfacet_make_actions(). */
5290 static struct subfacet *
5291 subfacet_create(struct facet *facet, struct flow_miss *miss,
5294 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5295 enum odp_key_fitness key_fitness = miss->key_fitness;
5296 const struct nlattr *key = miss->key;
5297 size_t key_len = miss->key_len;
5299 struct subfacet *subfacet;
5301 key_hash = odp_flow_key_hash(key, key_len);
5303 if (list_is_empty(&facet->subfacets)) {
5304 subfacet = &facet->one_subfacet;
5306 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5308 if (subfacet->facet == facet) {
5312 /* This shouldn't happen. */
5313 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5314 subfacet_destroy(subfacet);
5317 subfacet = xmalloc(sizeof *subfacet);
5320 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5321 list_push_back(&facet->subfacets, &subfacet->list_node);
5322 subfacet->facet = facet;
5323 subfacet->key_fitness = key_fitness;
5324 subfacet->key = xmemdup(key, key_len);
5325 subfacet->key_len = key_len;
5326 subfacet->used = now;
5327 subfacet->created = now;
5328 subfacet->dp_packet_count = 0;
5329 subfacet->dp_byte_count = 0;
5330 subfacet->actions_len = 0;
5331 subfacet->actions = NULL;
5333 subfacet->path = SF_NOT_INSTALLED;
5334 subfacet->initial_vals = miss->initial_vals;
5335 subfacet->odp_in_port = miss->odp_in_port;
5337 ofproto->subfacet_add_count++;
5341 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5342 * its facet within 'ofproto', and frees it. */
5344 subfacet_destroy__(struct subfacet *subfacet)
5346 struct facet *facet = subfacet->facet;
5347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5349 /* Update ofproto stats before uninstall the subfacet. */
5350 ofproto->subfacet_del_count++;
5351 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5353 subfacet_uninstall(subfacet);
5354 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5355 list_remove(&subfacet->list_node);
5356 free(subfacet->key);
5357 free(subfacet->actions);
5358 if (subfacet != &facet->one_subfacet) {
5363 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5364 * last remaining subfacet in its facet destroys the facet too. */
5366 subfacet_destroy(struct subfacet *subfacet)
5368 struct facet *facet = subfacet->facet;
5370 if (list_is_singleton(&facet->subfacets)) {
5371 /* facet_remove() needs at least one subfacet (it will remove it). */
5372 facet_remove(facet);
5374 subfacet_destroy__(subfacet);
5379 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5380 struct subfacet **subfacets, int n)
5382 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5383 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5384 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5387 for (i = 0; i < n; i++) {
5388 ops[i].type = DPIF_OP_FLOW_DEL;
5389 ops[i].u.flow_del.key = subfacets[i]->key;
5390 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5391 ops[i].u.flow_del.stats = &stats[i];
5395 dpif_operate(ofproto->backer->dpif, opsp, n);
5396 for (i = 0; i < n; i++) {
5397 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5398 subfacets[i]->path = SF_NOT_INSTALLED;
5399 subfacet_destroy(subfacets[i]);
5404 /* Composes the datapath actions for 'subfacet' based on its rule's actions. */
5406 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet)
5408 struct facet *facet = subfacet->facet;
5409 struct rule_dpif *rule = facet->rule;
5410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5412 struct action_xlate_ctx ctx;
5413 struct ofpbuf odp_actions;
5414 uint64_t stub[1024 / 8];
5416 ofpbuf_use_stub(&odp_actions, stub, sizeof stub);
5417 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5418 &subfacet->initial_vals, rule, 0, packet);
5419 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5420 facet->tags = ctx.tags;
5421 facet->has_learn = ctx.has_learn;
5422 facet->has_normal = ctx.has_normal;
5423 facet->has_fin_timeout = ctx.has_fin_timeout;
5424 facet->nf_flow.output_iface = ctx.nf_output_iface;
5425 facet->mirrors = ctx.mirrors;
5427 subfacet->slow = ctx.slow;
5429 ovs_assert(!subfacet->actions);
5430 subfacet->actions_len = odp_actions.size;
5431 subfacet->actions = ofpbuf_steal_data(&odp_actions);
5434 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5435 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5436 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5437 * since 'subfacet' was last updated.
5439 * Returns 0 if successful, otherwise a positive errno value. */
5441 subfacet_install(struct subfacet *subfacet,
5442 const struct nlattr *actions, size_t actions_len,
5443 struct dpif_flow_stats *stats,
5444 enum slow_path_reason slow)
5446 struct facet *facet = subfacet->facet;
5447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5448 enum subfacet_path path = subfacet_want_path(slow);
5449 uint64_t slow_path_stub[128 / 8];
5450 enum dpif_flow_put_flags flags;
5453 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5455 flags |= DPIF_FP_ZERO_STATS;
5458 if (path == SF_SLOW_PATH) {
5459 compose_slow_path(ofproto, &facet->flow, slow,
5460 slow_path_stub, sizeof slow_path_stub,
5461 &actions, &actions_len);
5464 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5465 subfacet->key_len, actions, actions_len, stats);
5468 subfacet_reset_dp_stats(subfacet, stats);
5472 subfacet->path = path;
5478 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5480 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5481 stats, subfacet->slow);
5484 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5486 subfacet_uninstall(struct subfacet *subfacet)
5488 if (subfacet->path != SF_NOT_INSTALLED) {
5489 struct rule_dpif *rule = subfacet->facet->rule;
5490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5491 struct dpif_flow_stats stats;
5494 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5495 subfacet->key_len, &stats);
5496 subfacet_reset_dp_stats(subfacet, &stats);
5498 subfacet_update_stats(subfacet, &stats);
5500 subfacet->path = SF_NOT_INSTALLED;
5502 ovs_assert(subfacet->dp_packet_count == 0);
5503 ovs_assert(subfacet->dp_byte_count == 0);
5507 /* Resets 'subfacet''s datapath statistics counters. This should be called
5508 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5509 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5510 * was reset in the datapath. 'stats' will be modified to include only
5511 * statistics new since 'subfacet' was last updated. */
5513 subfacet_reset_dp_stats(struct subfacet *subfacet,
5514 struct dpif_flow_stats *stats)
5517 && subfacet->dp_packet_count <= stats->n_packets
5518 && subfacet->dp_byte_count <= stats->n_bytes) {
5519 stats->n_packets -= subfacet->dp_packet_count;
5520 stats->n_bytes -= subfacet->dp_byte_count;
5523 subfacet->dp_packet_count = 0;
5524 subfacet->dp_byte_count = 0;
5527 /* Updates 'subfacet''s used time. The caller is responsible for calling
5528 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5530 subfacet_update_time(struct subfacet *subfacet, long long int used)
5532 if (used > subfacet->used) {
5533 subfacet->used = used;
5534 facet_update_time(subfacet->facet, used);
5538 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5540 * Because of the meaning of a subfacet's counters, it only makes sense to do
5541 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5542 * represents a packet that was sent by hand or if it represents statistics
5543 * that have been cleared out of the datapath. */
5545 subfacet_update_stats(struct subfacet *subfacet,
5546 const struct dpif_flow_stats *stats)
5548 if (stats->n_packets || stats->used > subfacet->used) {
5549 struct facet *facet = subfacet->facet;
5551 subfacet_update_time(subfacet, stats->used);
5552 facet->packet_count += stats->n_packets;
5553 facet->byte_count += stats->n_bytes;
5554 facet->tcp_flags |= stats->tcp_flags;
5555 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5561 static struct rule_dpif *
5562 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5564 struct rule_dpif *rule;
5566 rule = rule_dpif_lookup__(ofproto, flow, 0);
5571 return rule_dpif_miss_rule(ofproto, flow);
5574 static struct rule_dpif *
5575 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5578 struct cls_rule *cls_rule;
5579 struct classifier *cls;
5581 if (table_id >= N_TABLES) {
5585 cls = &ofproto->up.tables[table_id].cls;
5586 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5587 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5588 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5589 * are unavailable. */
5590 struct flow ofpc_normal_flow = *flow;
5591 ofpc_normal_flow.tp_src = htons(0);
5592 ofpc_normal_flow.tp_dst = htons(0);
5593 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5595 cls_rule = classifier_lookup(cls, flow);
5597 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5600 static struct rule_dpif *
5601 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5603 struct ofport_dpif *port;
5605 port = get_ofp_port(ofproto, flow->in_port);
5607 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5608 return ofproto->miss_rule;
5611 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5612 return ofproto->no_packet_in_rule;
5614 return ofproto->miss_rule;
5618 complete_operation(struct rule_dpif *rule)
5620 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5622 rule_invalidate(rule);
5624 struct dpif_completion *c = xmalloc(sizeof *c);
5625 c->op = rule->up.pending;
5626 list_push_back(&ofproto->completions, &c->list_node);
5628 ofoperation_complete(rule->up.pending, 0);
5632 static struct rule *
5635 struct rule_dpif *rule = xmalloc(sizeof *rule);
5640 rule_dealloc(struct rule *rule_)
5642 struct rule_dpif *rule = rule_dpif_cast(rule_);
5647 rule_construct(struct rule *rule_)
5649 struct rule_dpif *rule = rule_dpif_cast(rule_);
5650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5651 struct rule_dpif *victim;
5654 rule->packet_count = 0;
5655 rule->byte_count = 0;
5657 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5658 if (victim && !list_is_empty(&victim->facets)) {
5659 struct facet *facet;
5661 rule->facets = victim->facets;
5662 list_moved(&rule->facets);
5663 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5664 /* XXX: We're only clearing our local counters here. It's possible
5665 * that quite a few packets are unaccounted for in the datapath
5666 * statistics. These will be accounted to the new rule instead of
5667 * cleared as required. This could be fixed by clearing out the
5668 * datapath statistics for this facet, but currently it doesn't
5670 facet_reset_counters(facet);
5674 /* Must avoid list_moved() in this case. */
5675 list_init(&rule->facets);
5678 table_id = rule->up.table_id;
5680 rule->tag = victim->tag;
5681 } else if (table_id == 0) {
5686 miniflow_expand(&rule->up.cr.match.flow, &flow);
5687 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5688 ofproto->tables[table_id].basis);
5691 complete_operation(rule);
5696 rule_destruct(struct rule *rule_)
5698 struct rule_dpif *rule = rule_dpif_cast(rule_);
5699 struct facet *facet, *next_facet;
5701 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5702 facet_revalidate(facet);
5705 complete_operation(rule);
5709 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5711 struct rule_dpif *rule = rule_dpif_cast(rule_);
5712 struct facet *facet;
5714 /* push_all_stats() can handle flow misses which, when using the learn
5715 * action, can cause rules to be added and deleted. This can corrupt our
5716 * caller's datastructures which assume that rule_get_stats() doesn't have
5717 * an impact on the flow table. To be safe, we disable miss handling. */
5718 push_all_stats__(false);
5720 /* Start from historical data for 'rule' itself that are no longer tracked
5721 * in facets. This counts, for example, facets that have expired. */
5722 *packets = rule->packet_count;
5723 *bytes = rule->byte_count;
5725 /* Add any statistics that are tracked by facets. This includes
5726 * statistical data recently updated by ofproto_update_stats() as well as
5727 * stats for packets that were executed "by hand" via dpif_execute(). */
5728 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5729 *packets += facet->packet_count;
5730 *bytes += facet->byte_count;
5735 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5736 struct ofpbuf *packet)
5738 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5739 struct initial_vals initial_vals;
5740 struct dpif_flow_stats stats;
5741 struct action_xlate_ctx ctx;
5742 uint64_t odp_actions_stub[1024 / 8];
5743 struct ofpbuf odp_actions;
5745 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5746 rule_credit_stats(rule, &stats);
5748 initial_vals.vlan_tci = flow->vlan_tci;
5749 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5750 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5751 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5752 rule, stats.tcp_flags, packet);
5753 ctx.resubmit_stats = &stats;
5754 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5756 execute_odp_actions(ofproto, flow, odp_actions.data,
5757 odp_actions.size, packet);
5759 ofpbuf_uninit(&odp_actions);
5763 rule_execute(struct rule *rule, const struct flow *flow,
5764 struct ofpbuf *packet)
5766 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5767 ofpbuf_delete(packet);
5772 rule_modify_actions(struct rule *rule_)
5774 struct rule_dpif *rule = rule_dpif_cast(rule_);
5776 complete_operation(rule);
5779 /* Sends 'packet' out 'ofport'.
5780 * May modify 'packet'.
5781 * Returns 0 if successful, otherwise a positive errno value. */
5783 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5785 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5786 uint64_t odp_actions_stub[1024 / 8];
5787 struct ofpbuf key, odp_actions;
5788 struct odputil_keybuf keybuf;
5793 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5794 if (netdev_vport_is_patch(ofport->up.netdev)) {
5795 struct ofproto_dpif *peer_ofproto;
5796 struct dpif_flow_stats stats;
5797 struct ofport_dpif *peer;
5798 struct rule_dpif *rule;
5800 peer = ofport_get_peer(ofport);
5805 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5806 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5807 netdev_vport_inc_rx(peer->up.netdev, &stats);
5809 flow.in_port = peer->up.ofp_port;
5810 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5811 rule = rule_dpif_lookup(peer_ofproto, &flow);
5812 rule_dpif_execute(rule, &flow, packet);
5817 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5819 if (ofport->tnl_port) {
5820 struct dpif_flow_stats stats;
5822 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5823 if (odp_port == OVSP_NONE) {
5827 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5828 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5829 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5830 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5832 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5834 if (odp_port != ofport->odp_port) {
5835 eth_pop_vlan(packet);
5836 flow.vlan_tci = htons(0);
5840 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5841 odp_flow_key_from_flow(&key, &flow,
5842 ofp_port_to_odp_port(ofproto, flow.in_port));
5844 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5845 compose_ipfix_action(ofproto, &odp_actions, &flow);
5847 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5848 error = dpif_execute(ofproto->backer->dpif,
5850 odp_actions.data, odp_actions.size,
5852 ofpbuf_uninit(&odp_actions);
5855 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5856 ofproto->up.name, odp_port, strerror(error));
5858 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5862 /* OpenFlow to datapath action translation. */
5864 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5865 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5866 struct action_xlate_ctx *);
5867 static void xlate_normal(struct action_xlate_ctx *);
5869 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5870 * The action will state 'slow' as the reason that the action is in the slow
5871 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5872 * dump-flows" output to see why a flow is in the slow path.)
5874 * The 'stub_size' bytes in 'stub' will be used to store the action.
5875 * 'stub_size' must be large enough for the action.
5877 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5880 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5881 enum slow_path_reason slow,
5882 uint64_t *stub, size_t stub_size,
5883 const struct nlattr **actionsp, size_t *actions_lenp)
5885 union user_action_cookie cookie;
5888 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5889 cookie.slow_path.unused = 0;
5890 cookie.slow_path.reason = slow;
5892 ofpbuf_use_stack(&buf, stub, stub_size);
5893 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5894 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5895 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5897 put_userspace_action(ofproto, &buf, flow, &cookie,
5898 sizeof cookie.slow_path);
5900 *actionsp = buf.data;
5901 *actions_lenp = buf.size;
5905 put_userspace_action(const struct ofproto_dpif *ofproto,
5906 struct ofpbuf *odp_actions,
5907 const struct flow *flow,
5908 const union user_action_cookie *cookie,
5909 const size_t cookie_size)
5913 pid = dpif_port_get_pid(ofproto->backer->dpif,
5914 ofp_port_to_odp_port(ofproto, flow->in_port));
5916 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5919 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5920 * the number of packets out of UINT32_MAX to sample. The given
5921 * cookie is passed back in the callback for each sampled packet.
5924 compose_sample_action(const struct ofproto_dpif *ofproto,
5925 struct ofpbuf *odp_actions,
5926 const struct flow *flow,
5927 const uint32_t probability,
5928 const union user_action_cookie *cookie,
5929 const size_t cookie_size)
5931 size_t sample_offset, actions_offset;
5934 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5936 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5938 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5939 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
5942 nl_msg_end_nested(odp_actions, actions_offset);
5943 nl_msg_end_nested(odp_actions, sample_offset);
5944 return cookie_offset;
5948 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5949 ovs_be16 vlan_tci, uint32_t odp_port,
5950 unsigned int n_outputs, union user_action_cookie *cookie)
5954 cookie->type = USER_ACTION_COOKIE_SFLOW;
5955 cookie->sflow.vlan_tci = vlan_tci;
5957 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5958 * port information") for the interpretation of cookie->output. */
5959 switch (n_outputs) {
5961 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5962 cookie->sflow.output = 0x40000000 | 256;
5966 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5968 cookie->sflow.output = ifindex;
5973 /* 0x80000000 means "multiple output ports. */
5974 cookie->sflow.output = 0x80000000 | n_outputs;
5979 /* Compose SAMPLE action for sFlow bridge sampling. */
5981 compose_sflow_action(const struct ofproto_dpif *ofproto,
5982 struct ofpbuf *odp_actions,
5983 const struct flow *flow,
5986 uint32_t probability;
5987 union user_action_cookie cookie;
5989 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5993 probability = dpif_sflow_get_probability(ofproto->sflow);
5994 compose_sflow_cookie(ofproto, htons(0), odp_port,
5995 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5997 return compose_sample_action(ofproto, odp_actions, flow, probability,
5998 &cookie, sizeof cookie.sflow);
6002 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
6003 uint32_t obs_domain_id, uint32_t obs_point_id,
6004 union user_action_cookie *cookie)
6006 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6007 cookie->flow_sample.probability = probability;
6008 cookie->flow_sample.collector_set_id = collector_set_id;
6009 cookie->flow_sample.obs_domain_id = obs_domain_id;
6010 cookie->flow_sample.obs_point_id = obs_point_id;
6014 compose_ipfix_cookie(union user_action_cookie *cookie)
6016 cookie->type = USER_ACTION_COOKIE_IPFIX;
6019 /* Compose SAMPLE action for IPFIX bridge sampling. */
6021 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6022 struct ofpbuf *odp_actions,
6023 const struct flow *flow)
6025 uint32_t probability;
6026 union user_action_cookie cookie;
6028 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6032 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6033 compose_ipfix_cookie(&cookie);
6035 compose_sample_action(ofproto, odp_actions, flow, probability,
6036 &cookie, sizeof cookie.ipfix);
6039 /* SAMPLE action for sFlow must be first action in any given list of
6040 * actions. At this point we do not have all information required to
6041 * build it. So try to build sample action as complete as possible. */
6043 add_sflow_action(struct action_xlate_ctx *ctx)
6045 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6047 &ctx->flow, OVSP_NONE);
6048 ctx->sflow_odp_port = 0;
6049 ctx->sflow_n_outputs = 0;
6052 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6053 * of actions, eventually after the SAMPLE action for sFlow. */
6055 add_ipfix_action(struct action_xlate_ctx *ctx)
6057 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6060 /* Fix SAMPLE action according to data collected while composing ODP actions.
6061 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6062 * USERSPACE action's user-cookie which is required for sflow. */
6064 fix_sflow_action(struct action_xlate_ctx *ctx)
6066 const struct flow *base = &ctx->base_flow;
6067 union user_action_cookie *cookie;
6069 if (!ctx->user_cookie_offset) {
6073 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6074 sizeof cookie->sflow);
6075 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6077 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6078 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6082 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6085 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6086 ovs_be16 flow_vlan_tci;
6087 uint32_t flow_skb_mark;
6088 uint8_t flow_nw_tos;
6089 struct priority_to_dscp *pdscp;
6090 uint32_t out_port, odp_port;
6092 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6093 * before traversing a patch port. */
6094 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6097 xlate_report(ctx, "Nonexistent output port");
6099 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6100 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6102 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6103 xlate_report(ctx, "STP not in forwarding state, skipping output");
6107 if (netdev_vport_is_patch(ofport->up.netdev)) {
6108 struct ofport_dpif *peer = ofport_get_peer(ofport);
6109 struct flow old_flow = ctx->flow;
6110 const struct ofproto_dpif *peer_ofproto;
6111 enum slow_path_reason special;
6112 struct ofport_dpif *in_port;
6115 xlate_report(ctx, "Nonexistent patch port peer");
6119 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6120 if (peer_ofproto->backer != ctx->ofproto->backer) {
6121 xlate_report(ctx, "Patch port peer on a different datapath");
6125 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6126 ctx->flow.in_port = peer->up.ofp_port;
6127 ctx->flow.metadata = htonll(0);
6128 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6129 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6131 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6132 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6135 ctx->slow = special;
6136 } else if (!in_port || may_receive(in_port, ctx)) {
6137 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6138 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6140 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6141 * learning action look at the packet, then drop it. */
6142 struct flow old_base_flow = ctx->base_flow;
6143 size_t old_size = ctx->odp_actions->size;
6144 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6145 ctx->base_flow = old_base_flow;
6146 ctx->odp_actions->size = old_size;
6150 ctx->flow = old_flow;
6151 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6153 if (ctx->resubmit_stats) {
6154 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6155 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6161 flow_vlan_tci = ctx->flow.vlan_tci;
6162 flow_skb_mark = ctx->flow.skb_mark;
6163 flow_nw_tos = ctx->flow.nw_tos;
6165 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6167 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6168 ctx->flow.nw_tos |= pdscp->dscp;
6171 if (ofport->tnl_port) {
6172 /* Save tunnel metadata so that changes made due to
6173 * the Logical (tunnel) Port are not visible for any further
6174 * matches, while explicit set actions on tunnel metadata are.
6176 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6177 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6178 if (odp_port == OVSP_NONE) {
6179 xlate_report(ctx, "Tunneling decided against output");
6180 goto out; /* restore flow_nw_tos */
6183 if (ctx->resubmit_stats) {
6184 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6186 out_port = odp_port;
6187 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6189 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6191 odp_port = ofport->odp_port;
6192 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6193 ctx->flow.vlan_tci);
6194 if (out_port != odp_port) {
6195 ctx->flow.vlan_tci = htons(0);
6197 ctx->flow.skb_mark &= ~IPSEC_MARK;
6199 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6200 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6202 ctx->sflow_odp_port = odp_port;
6203 ctx->sflow_n_outputs++;
6204 ctx->nf_output_iface = ofp_port;
6207 ctx->flow.vlan_tci = flow_vlan_tci;
6208 ctx->flow.skb_mark = flow_skb_mark;
6210 ctx->flow.nw_tos = flow_nw_tos;
6214 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6216 compose_output_action__(ctx, ofp_port, true);
6220 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6222 struct ofproto_dpif *ofproto = ctx->ofproto;
6223 uint8_t table_id = ctx->table_id;
6225 if (table_id > 0 && table_id < N_TABLES) {
6226 struct table_dpif *table = &ofproto->tables[table_id];
6227 if (table->other_table) {
6228 ctx->tags |= (rule && rule->tag
6230 : rule_calculate_tag(&ctx->flow,
6231 &table->other_table->mask,
6237 /* Common rule processing in one place to avoid duplicating code. */
6238 static struct rule_dpif *
6239 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6242 if (ctx->resubmit_hook) {
6243 ctx->resubmit_hook(ctx, rule);
6245 if (rule == NULL && may_packet_in) {
6247 * check if table configuration flags
6248 * OFPTC_TABLE_MISS_CONTROLLER, default.
6249 * OFPTC_TABLE_MISS_CONTINUE,
6250 * OFPTC_TABLE_MISS_DROP
6251 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6253 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6255 if (rule && ctx->resubmit_stats) {
6256 rule_credit_stats(rule, ctx->resubmit_stats);
6262 xlate_table_action(struct action_xlate_ctx *ctx,
6263 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6265 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6266 struct rule_dpif *rule;
6267 uint16_t old_in_port = ctx->flow.in_port;
6268 uint8_t old_table_id = ctx->table_id;
6270 ctx->table_id = table_id;
6272 /* Look up a flow with 'in_port' as the input port. */
6273 ctx->flow.in_port = in_port;
6274 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6276 tag_the_flow(ctx, rule);
6278 /* Restore the original input port. Otherwise OFPP_NORMAL and
6279 * OFPP_IN_PORT will have surprising behavior. */
6280 ctx->flow.in_port = old_in_port;
6282 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6285 struct rule_dpif *old_rule = ctx->rule;
6289 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6290 ctx->rule = old_rule;
6294 ctx->table_id = old_table_id;
6296 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6298 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6299 MAX_RESUBMIT_RECURSION);
6300 ctx->max_resubmit_trigger = true;
6305 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6306 const struct ofpact_resubmit *resubmit)
6311 in_port = resubmit->in_port;
6312 if (in_port == OFPP_IN_PORT) {
6313 in_port = ctx->flow.in_port;
6316 table_id = resubmit->table_id;
6317 if (table_id == 255) {
6318 table_id = ctx->table_id;
6321 xlate_table_action(ctx, in_port, table_id, false);
6325 flood_packets(struct action_xlate_ctx *ctx, bool all)
6327 struct ofport_dpif *ofport;
6329 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6330 uint16_t ofp_port = ofport->up.ofp_port;
6332 if (ofp_port == ctx->flow.in_port) {
6337 compose_output_action__(ctx, ofp_port, false);
6338 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6339 compose_output_action(ctx, ofp_port);
6343 ctx->nf_output_iface = NF_OUT_FLOOD;
6347 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6348 enum ofp_packet_in_reason reason,
6349 uint16_t controller_id)
6351 struct ofputil_packet_in pin;
6352 struct ofpbuf *packet;
6354 ovs_assert(!ctx->slow || ctx->slow == SLOW_CONTROLLER);
6355 ctx->slow = SLOW_CONTROLLER;
6360 packet = ofpbuf_clone(ctx->packet);
6362 if (packet->l2 && packet->l3) {
6363 struct eth_header *eh;
6364 uint16_t mpls_depth;
6366 eth_pop_vlan(packet);
6369 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6370 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6372 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6373 eth_push_vlan(packet, ctx->flow.vlan_tci);
6376 mpls_depth = eth_mpls_depth(packet);
6378 if (mpls_depth < ctx->flow.mpls_depth) {
6379 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6380 } else if (mpls_depth > ctx->flow.mpls_depth) {
6381 pop_mpls(packet, ctx->flow.dl_type);
6382 } else if (mpls_depth) {
6383 set_mpls_lse(packet, ctx->flow.mpls_lse);
6387 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6388 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6389 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6393 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6394 packet_set_tcp_port(packet, ctx->flow.tp_src,
6396 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6397 packet_set_udp_port(packet, ctx->flow.tp_src,
6404 pin.packet = packet->data;
6405 pin.packet_len = packet->size;
6406 pin.reason = reason;
6407 pin.controller_id = controller_id;
6408 pin.table_id = ctx->table_id;
6409 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6412 flow_get_metadata(&ctx->flow, &pin.fmd);
6414 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6415 ofpbuf_delete(packet);
6419 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6421 ovs_assert(eth_type_mpls(eth_type));
6423 if (ctx->base_flow.mpls_depth) {
6424 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6425 ctx->flow.mpls_depth++;
6430 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6431 label = htonl(0x2); /* IPV6 Explicit Null. */
6433 label = htonl(0x0); /* IPV4 Explicit Null. */
6435 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6436 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6437 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6438 ctx->flow.mpls_depth = 1;
6440 ctx->flow.dl_type = eth_type;
6444 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6446 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6447 ovs_assert(!eth_type_mpls(eth_type));
6449 if (ctx->flow.mpls_depth) {
6450 ctx->flow.mpls_depth--;
6451 ctx->flow.mpls_lse = htonl(0);
6452 if (!ctx->flow.mpls_depth) {
6453 ctx->flow.dl_type = eth_type;
6459 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6461 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6462 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6466 if (ctx->flow.nw_ttl > 1) {
6472 for (i = 0; i < ids->n_controllers; i++) {
6473 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6477 /* Stop processing for current table. */
6483 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6485 if (!eth_type_mpls(ctx->flow.dl_type)) {
6489 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6494 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6496 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6498 if (!eth_type_mpls(ctx->flow.dl_type)) {
6504 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6507 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6509 /* Stop processing for current table. */
6515 xlate_output_action(struct action_xlate_ctx *ctx,
6516 uint16_t port, uint16_t max_len, bool may_packet_in)
6518 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6520 ctx->nf_output_iface = NF_OUT_DROP;
6524 compose_output_action(ctx, ctx->flow.in_port);
6527 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6533 flood_packets(ctx, false);
6536 flood_packets(ctx, true);
6538 case OFPP_CONTROLLER:
6539 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6545 if (port != ctx->flow.in_port) {
6546 compose_output_action(ctx, port);
6548 xlate_report(ctx, "skipping output to input port");
6553 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6554 ctx->nf_output_iface = NF_OUT_FLOOD;
6555 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6556 ctx->nf_output_iface = prev_nf_output_iface;
6557 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6558 ctx->nf_output_iface != NF_OUT_FLOOD) {
6559 ctx->nf_output_iface = NF_OUT_MULTI;
6564 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6565 const struct ofpact_output_reg *or)
6567 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6568 if (port <= UINT16_MAX) {
6569 xlate_output_action(ctx, port, or->max_len, false);
6574 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6575 const struct ofpact_enqueue *enqueue)
6577 uint16_t ofp_port = enqueue->port;
6578 uint32_t queue_id = enqueue->queue;
6579 uint32_t flow_priority, priority;
6582 /* Translate queue to priority. */
6583 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6584 queue_id, &priority);
6586 /* Fall back to ordinary output action. */
6587 xlate_output_action(ctx, enqueue->port, 0, false);
6591 /* Check output port. */
6592 if (ofp_port == OFPP_IN_PORT) {
6593 ofp_port = ctx->flow.in_port;
6594 } else if (ofp_port == ctx->flow.in_port) {
6598 /* Add datapath actions. */
6599 flow_priority = ctx->flow.skb_priority;
6600 ctx->flow.skb_priority = priority;
6601 compose_output_action(ctx, ofp_port);
6602 ctx->flow.skb_priority = flow_priority;
6604 /* Update NetFlow output port. */
6605 if (ctx->nf_output_iface == NF_OUT_DROP) {
6606 ctx->nf_output_iface = ofp_port;
6607 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6608 ctx->nf_output_iface = NF_OUT_MULTI;
6613 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6615 uint32_t skb_priority;
6617 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6618 queue_id, &skb_priority)) {
6619 ctx->flow.skb_priority = skb_priority;
6621 /* Couldn't translate queue to a priority. Nothing to do. A warning
6622 * has already been logged. */
6627 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6629 struct ofproto_dpif *ofproto = ofproto_;
6630 struct ofport_dpif *port;
6640 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6643 port = get_ofp_port(ofproto, ofp_port);
6644 return port ? port->may_enable : false;
6649 xlate_bundle_action(struct action_xlate_ctx *ctx,
6650 const struct ofpact_bundle *bundle)
6654 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6655 if (bundle->dst.field) {
6656 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6658 xlate_output_action(ctx, port, 0, false);
6663 xlate_learn_action(struct action_xlate_ctx *ctx,
6664 const struct ofpact_learn *learn)
6666 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6667 struct ofputil_flow_mod fm;
6668 uint64_t ofpacts_stub[1024 / 8];
6669 struct ofpbuf ofpacts;
6672 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6673 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6675 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6676 if (error && !VLOG_DROP_WARN(&rl)) {
6677 VLOG_WARN("learning action failed to modify flow table (%s)",
6678 ofperr_get_name(error));
6681 ofpbuf_uninit(&ofpacts);
6684 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6685 * means "infinite". */
6687 reduce_timeout(uint16_t max, uint16_t *timeout)
6689 if (max && (!*timeout || *timeout > max)) {
6695 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6696 const struct ofpact_fin_timeout *oft)
6698 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6699 struct rule_dpif *rule = ctx->rule;
6701 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6702 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6707 xlate_sample_action(struct action_xlate_ctx *ctx,
6708 const struct ofpact_sample *os)
6710 union user_action_cookie cookie;
6711 /* Scale the probability from 16-bit to 32-bit while representing
6712 * the same percentage. */
6713 uint32_t probability = (os->probability << 16) | os->probability;
6715 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6717 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6718 os->obs_domain_id, os->obs_point_id, &cookie);
6719 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6720 probability, &cookie, sizeof cookie.flow_sample);
6724 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6726 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6727 ? OFPUTIL_PC_NO_RECV_STP
6728 : OFPUTIL_PC_NO_RECV)) {
6732 /* Only drop packets here if both forwarding and learning are
6733 * disabled. If just learning is enabled, we need to have
6734 * OFPP_NORMAL and the learning action have a look at the packet
6735 * before we can drop it. */
6736 if (!stp_forward_in_state(port->stp_state)
6737 && !stp_learn_in_state(port->stp_state)) {
6745 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6747 if (is_ip_any(&ctx->base_flow)
6748 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6749 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6750 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6751 " but is not ECN capable");
6754 /* Set the ECN CE value in the tunneled packet. */
6755 ctx->flow.nw_tos |= IP_ECN_CE;
6763 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6764 struct action_xlate_ctx *ctx)
6766 bool was_evictable = true;
6767 const struct ofpact *a;
6770 /* Don't let the rule we're working on get evicted underneath us. */
6771 was_evictable = ctx->rule->up.evictable;
6772 ctx->rule->up.evictable = false;
6775 do_xlate_actions_again:
6776 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6777 struct ofpact_controller *controller;
6778 const struct ofpact_metadata *metadata;
6786 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6787 ofpact_get_OUTPUT(a)->max_len, true);
6790 case OFPACT_CONTROLLER:
6791 controller = ofpact_get_CONTROLLER(a);
6792 execute_controller_action(ctx, controller->max_len,
6794 controller->controller_id);
6797 case OFPACT_ENQUEUE:
6798 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6801 case OFPACT_SET_VLAN_VID:
6802 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6803 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6807 case OFPACT_SET_VLAN_PCP:
6808 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6809 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6814 case OFPACT_STRIP_VLAN:
6815 ctx->flow.vlan_tci = htons(0);
6818 case OFPACT_PUSH_VLAN:
6819 /* XXX 802.1AD(QinQ) */
6820 ctx->flow.vlan_tci = htons(VLAN_CFI);
6823 case OFPACT_SET_ETH_SRC:
6824 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6828 case OFPACT_SET_ETH_DST:
6829 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6833 case OFPACT_SET_IPV4_SRC:
6834 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6835 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6839 case OFPACT_SET_IPV4_DST:
6840 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6841 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6845 case OFPACT_SET_IPV4_DSCP:
6846 /* OpenFlow 1.0 only supports IPv4. */
6847 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6848 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6849 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6853 case OFPACT_SET_L4_SRC_PORT:
6854 if (is_ip_any(&ctx->flow)) {
6855 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6859 case OFPACT_SET_L4_DST_PORT:
6860 if (is_ip_any(&ctx->flow)) {
6861 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6865 case OFPACT_RESUBMIT:
6866 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6869 case OFPACT_SET_TUNNEL:
6870 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6873 case OFPACT_SET_QUEUE:
6874 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6877 case OFPACT_POP_QUEUE:
6878 ctx->flow.skb_priority = ctx->orig_skb_priority;
6881 case OFPACT_REG_MOVE:
6882 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6885 case OFPACT_REG_LOAD:
6886 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6889 case OFPACT_STACK_PUSH:
6890 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6894 case OFPACT_STACK_POP:
6895 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6899 case OFPACT_PUSH_MPLS:
6900 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6903 case OFPACT_POP_MPLS:
6904 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6907 case OFPACT_SET_MPLS_TTL:
6908 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6913 case OFPACT_DEC_MPLS_TTL:
6914 if (execute_dec_mpls_ttl_action(ctx)) {
6919 case OFPACT_DEC_TTL:
6920 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6926 /* Nothing to do. */
6929 case OFPACT_MULTIPATH:
6930 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6934 ctx->ofproto->has_bundle_action = true;
6935 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6938 case OFPACT_OUTPUT_REG:
6939 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6943 ctx->has_learn = true;
6944 if (ctx->may_learn) {
6945 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6953 case OFPACT_FIN_TIMEOUT:
6954 ctx->has_fin_timeout = true;
6955 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6958 case OFPACT_CLEAR_ACTIONS:
6960 * Nothing to do because writa-actions is not supported for now.
6961 * When writa-actions is supported, clear-actions also must
6962 * be supported at the same time.
6966 case OFPACT_WRITE_METADATA:
6967 metadata = ofpact_get_WRITE_METADATA(a);
6968 ctx->flow.metadata &= ~metadata->mask;
6969 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6972 case OFPACT_GOTO_TABLE: {
6973 /* It is assumed that goto-table is the last action. */
6974 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6975 struct rule_dpif *rule;
6977 ovs_assert(ctx->table_id < ogt->table_id);
6979 ctx->table_id = ogt->table_id;
6981 /* Look up a flow from the new table. */
6982 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6984 tag_the_flow(ctx, rule);
6986 rule = ctx_rule_hooks(ctx, rule, true);
6990 ctx->rule->up.evictable = was_evictable;
6993 was_evictable = rule->up.evictable;
6994 rule->up.evictable = false;
6996 /* Tail recursion removal. */
6997 ofpacts = rule->up.ofpacts;
6998 ofpacts_len = rule->up.ofpacts_len;
6999 goto do_xlate_actions_again;
7005 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7012 ctx->rule->up.evictable = was_evictable;
7017 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7018 struct ofproto_dpif *ofproto, const struct flow *flow,
7019 const struct initial_vals *initial_vals,
7020 struct rule_dpif *rule,
7021 uint8_t tcp_flags, const struct ofpbuf *packet)
7023 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
7025 /* Flow initialization rules:
7026 * - 'base_flow' must match the kernel's view of the packet at the
7027 * time that action processing starts. 'flow' represents any
7028 * transformations we wish to make through actions.
7029 * - By default 'base_flow' and 'flow' are the same since the input
7030 * packet matches the output before any actions are applied.
7031 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7032 * of the received packet as seen by the kernel. If we later output
7033 * to another device without any modifications this will cause us to
7034 * insert a new tag since the original one was stripped off by the
7036 * - Tunnel 'flow' is largely cleared when transitioning between
7037 * the input and output stages since it does not make sense to output
7038 * a packet with the exact headers that it was received with (i.e.
7039 * the destination IP is us). The one exception is the tun_id, which
7040 * is preserved to allow use in later resubmit lookups and loads into
7042 * - Tunnel 'base_flow' is completely cleared since that is what the
7043 * kernel does. If we wish to maintain the original values an action
7044 * needs to be generated. */
7046 ctx->ofproto = ofproto;
7048 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
7049 ctx->base_flow = ctx->flow;
7050 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7051 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
7052 ctx->flow.tunnel.tun_id = initial_tun_id;
7054 ctx->packet = packet;
7055 ctx->may_learn = packet != NULL;
7056 ctx->tcp_flags = tcp_flags;
7057 ctx->resubmit_hook = NULL;
7058 ctx->report_hook = NULL;
7059 ctx->resubmit_stats = NULL;
7062 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7063 * into datapath actions in 'odp_actions', using 'ctx'. */
7065 xlate_actions(struct action_xlate_ctx *ctx,
7066 const struct ofpact *ofpacts, size_t ofpacts_len,
7067 struct ofpbuf *odp_actions)
7069 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7070 * that in the future we always keep a copy of the original flow for
7071 * tracing purposes. */
7072 static bool hit_resubmit_limit;
7074 enum slow_path_reason special;
7075 struct ofport_dpif *in_port;
7076 struct flow orig_flow;
7078 COVERAGE_INC(ofproto_dpif_xlate);
7080 ofpbuf_clear(odp_actions);
7081 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7083 ctx->odp_actions = odp_actions;
7086 ctx->has_learn = false;
7087 ctx->has_normal = false;
7088 ctx->has_fin_timeout = false;
7089 ctx->nf_output_iface = NF_OUT_DROP;
7092 ctx->max_resubmit_trigger = false;
7093 ctx->orig_skb_priority = ctx->flow.skb_priority;
7097 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7099 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7100 /* Do this conditionally because the copy is expensive enough that it
7101 * shows up in profiles. */
7102 orig_flow = ctx->flow;
7105 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7106 switch (ctx->ofproto->up.frag_handling) {
7107 case OFPC_FRAG_NORMAL:
7108 /* We must pretend that transport ports are unavailable. */
7109 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7110 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7113 case OFPC_FRAG_DROP:
7116 case OFPC_FRAG_REASM:
7119 case OFPC_FRAG_NX_MATCH:
7120 /* Nothing to do. */
7123 case OFPC_INVALID_TTL_TO_CONTROLLER:
7128 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7129 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7131 ctx->slow = special;
7133 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7134 struct initial_vals initial_vals;
7135 size_t sample_actions_len;
7136 uint32_t local_odp_port;
7138 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7139 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
7141 add_sflow_action(ctx);
7142 add_ipfix_action(ctx);
7143 sample_actions_len = ctx->odp_actions->size;
7145 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7146 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7148 /* We've let OFPP_NORMAL and the learning action look at the
7149 * packet, so drop it now if forwarding is disabled. */
7150 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7151 ctx->odp_actions->size = sample_actions_len;
7155 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7156 if (!hit_resubmit_limit) {
7157 /* We didn't record the original flow. Make sure we do from
7159 hit_resubmit_limit = true;
7160 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7161 struct ds ds = DS_EMPTY_INITIALIZER;
7163 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7164 &initial_vals, &ds);
7165 VLOG_ERR("Trace triggered by excessive resubmit "
7166 "recursion:\n%s", ds_cstr(&ds));
7171 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7172 if (!connmgr_must_output_local(ctx->ofproto->up.connmgr, &ctx->flow,
7174 ctx->odp_actions->data,
7175 ctx->odp_actions->size)) {
7176 compose_output_action(ctx, OFPP_LOCAL);
7178 if (ctx->ofproto->has_mirrors) {
7179 add_mirror_actions(ctx, &orig_flow);
7181 fix_sflow_action(ctx);
7184 ofpbuf_uninit(&ctx->stack);
7187 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7188 * into datapath actions, using 'ctx', and discards the datapath actions. */
7190 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7191 const struct ofpact *ofpacts,
7194 uint64_t odp_actions_stub[1024 / 8];
7195 struct ofpbuf odp_actions;
7197 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7198 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7199 ofpbuf_uninit(&odp_actions);
7203 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7205 if (ctx->report_hook) {
7206 ctx->report_hook(ctx, s);
7210 /* OFPP_NORMAL implementation. */
7212 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7214 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7215 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7216 * the bundle on which the packet was received, returns the VLAN to which the
7219 * Both 'vid' and the return value are in the range 0...4095. */
7221 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7223 switch (in_bundle->vlan_mode) {
7224 case PORT_VLAN_ACCESS:
7225 return in_bundle->vlan;
7228 case PORT_VLAN_TRUNK:
7231 case PORT_VLAN_NATIVE_UNTAGGED:
7232 case PORT_VLAN_NATIVE_TAGGED:
7233 return vid ? vid : in_bundle->vlan;
7240 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7241 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7244 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7245 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7248 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7250 /* Allow any VID on the OFPP_NONE port. */
7251 if (in_bundle == &ofpp_none_bundle) {
7255 switch (in_bundle->vlan_mode) {
7256 case PORT_VLAN_ACCESS:
7259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7260 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7261 "packet received on port %s configured as VLAN "
7262 "%"PRIu16" access port",
7263 in_bundle->ofproto->up.name, vid,
7264 in_bundle->name, in_bundle->vlan);
7270 case PORT_VLAN_NATIVE_UNTAGGED:
7271 case PORT_VLAN_NATIVE_TAGGED:
7273 /* Port must always carry its native VLAN. */
7277 case PORT_VLAN_TRUNK:
7278 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7280 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7281 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7282 "received on port %s not configured for trunking "
7284 in_bundle->ofproto->up.name, vid,
7285 in_bundle->name, vid);
7297 /* Given 'vlan', the VLAN that a packet belongs to, and
7298 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7299 * that should be included in the 802.1Q header. (If the return value is 0,
7300 * then the 802.1Q header should only be included in the packet if there is a
7303 * Both 'vlan' and the return value are in the range 0...4095. */
7305 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7307 switch (out_bundle->vlan_mode) {
7308 case PORT_VLAN_ACCESS:
7311 case PORT_VLAN_TRUNK:
7312 case PORT_VLAN_NATIVE_TAGGED:
7315 case PORT_VLAN_NATIVE_UNTAGGED:
7316 return vlan == out_bundle->vlan ? 0 : vlan;
7324 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7327 struct ofport_dpif *port;
7329 ovs_be16 tci, old_tci;
7331 vid = output_vlan_to_vid(out_bundle, vlan);
7332 if (!out_bundle->bond) {
7333 port = ofbundle_get_a_port(out_bundle);
7335 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7338 /* No slaves enabled, so drop packet. */
7343 old_tci = ctx->flow.vlan_tci;
7345 if (tci || out_bundle->use_priority_tags) {
7346 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7348 tci |= htons(VLAN_CFI);
7351 ctx->flow.vlan_tci = tci;
7353 compose_output_action(ctx, port->up.ofp_port);
7354 ctx->flow.vlan_tci = old_tci;
7358 mirror_mask_ffs(mirror_mask_t mask)
7360 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7365 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7367 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7368 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7372 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7374 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7377 /* Returns an arbitrary interface within 'bundle'. */
7378 static struct ofport_dpif *
7379 ofbundle_get_a_port(const struct ofbundle *bundle)
7381 return CONTAINER_OF(list_front(&bundle->ports),
7382 struct ofport_dpif, bundle_node);
7386 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7388 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7392 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7394 struct ofproto_dpif *ofproto = ctx->ofproto;
7395 mirror_mask_t mirrors;
7396 struct ofbundle *in_bundle;
7399 const struct nlattr *a;
7402 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7403 ctx->packet != NULL, NULL);
7407 mirrors = in_bundle->src_mirrors;
7409 /* Drop frames on bundles reserved for mirroring. */
7410 if (in_bundle->mirror_out) {
7411 if (ctx->packet != NULL) {
7412 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7413 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7414 "%s, which is reserved exclusively for mirroring",
7415 ctx->ofproto->up.name, in_bundle->name);
7421 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7422 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7425 vlan = input_vid_to_vlan(in_bundle, vid);
7427 /* Look at the output ports to check for destination selections. */
7429 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7430 ctx->odp_actions->size) {
7431 enum ovs_action_attr type = nl_attr_type(a);
7432 struct ofport_dpif *ofport;
7434 if (type != OVS_ACTION_ATTR_OUTPUT) {
7438 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7439 if (ofport && ofport->bundle) {
7440 mirrors |= ofport->bundle->dst_mirrors;
7448 /* Restore the original packet before adding the mirror actions. */
7449 ctx->flow = *orig_flow;
7454 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7456 if (!vlan_is_mirrored(m, vlan)) {
7457 mirrors = zero_rightmost_1bit(mirrors);
7461 mirrors &= ~m->dup_mirrors;
7462 ctx->mirrors |= m->dup_mirrors;
7464 output_normal(ctx, m->out, vlan);
7465 } else if (vlan != m->out_vlan
7466 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7467 struct ofbundle *bundle;
7469 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7470 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7471 && !bundle->mirror_out) {
7472 output_normal(ctx, bundle, m->out_vlan);
7480 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7481 uint64_t packets, uint64_t bytes)
7487 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7490 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7493 /* In normal circumstances 'm' will not be NULL. However,
7494 * if mirrors are reconfigured, we can temporarily get out
7495 * of sync in facet_revalidate(). We could "correct" the
7496 * mirror list before reaching here, but doing that would
7497 * not properly account the traffic stats we've currently
7498 * accumulated for previous mirror configuration. */
7502 m->packet_count += packets;
7503 m->byte_count += bytes;
7507 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7508 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7509 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7511 is_gratuitous_arp(const struct flow *flow)
7513 return (flow->dl_type == htons(ETH_TYPE_ARP)
7514 && eth_addr_is_broadcast(flow->dl_dst)
7515 && (flow->nw_proto == ARP_OP_REPLY
7516 || (flow->nw_proto == ARP_OP_REQUEST
7517 && flow->nw_src == flow->nw_dst)));
7521 update_learning_table(struct ofproto_dpif *ofproto,
7522 const struct flow *flow, int vlan,
7523 struct ofbundle *in_bundle)
7525 struct mac_entry *mac;
7527 /* Don't learn the OFPP_NONE port. */
7528 if (in_bundle == &ofpp_none_bundle) {
7532 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7536 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7537 if (is_gratuitous_arp(flow)) {
7538 /* We don't want to learn from gratuitous ARP packets that are
7539 * reflected back over bond slaves so we lock the learning table. */
7540 if (!in_bundle->bond) {
7541 mac_entry_set_grat_arp_lock(mac);
7542 } else if (mac_entry_is_grat_arp_locked(mac)) {
7547 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7548 /* The log messages here could actually be useful in debugging,
7549 * so keep the rate limit relatively high. */
7550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7551 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7552 "on port %s in VLAN %d",
7553 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7554 in_bundle->name, vlan);
7556 mac->port.p = in_bundle;
7557 tag_set_add(&ofproto->backer->revalidate_set,
7558 mac_learning_changed(ofproto->ml, mac));
7562 static struct ofbundle *
7563 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7564 bool warn, struct ofport_dpif **in_ofportp)
7566 struct ofport_dpif *ofport;
7568 /* Find the port and bundle for the received packet. */
7569 ofport = get_ofp_port(ofproto, in_port);
7571 *in_ofportp = ofport;
7573 if (ofport && ofport->bundle) {
7574 return ofport->bundle;
7577 /* Special-case OFPP_NONE, which a controller may use as the ingress
7578 * port for traffic that it is sourcing. */
7579 if (in_port == OFPP_NONE) {
7580 return &ofpp_none_bundle;
7583 /* Odd. A few possible reasons here:
7585 * - We deleted a port but there are still a few packets queued up
7588 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7589 * we don't know about.
7591 * - The ofproto client didn't configure the port as part of a bundle.
7592 * This is particularly likely to happen if a packet was received on the
7593 * port after it was created, but before the client had a chance to
7594 * configure its bundle.
7597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7599 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7600 "port %"PRIu16, ofproto->up.name, in_port);
7605 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7606 * dropped. Returns true if they may be forwarded, false if they should be
7609 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7610 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7612 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7613 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7614 * checked by input_vid_is_valid().
7616 * May also add tags to '*tags', although the current implementation only does
7617 * so in one special case.
7620 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7623 struct ofproto_dpif *ofproto = ctx->ofproto;
7624 struct flow *flow = &ctx->flow;
7625 struct ofbundle *in_bundle = in_port->bundle;
7627 /* Drop frames for reserved multicast addresses
7628 * only if forward_bpdu option is absent. */
7629 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7630 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7634 if (in_bundle->bond) {
7635 struct mac_entry *mac;
7637 switch (bond_check_admissibility(in_bundle->bond, in_port,
7638 flow->dl_dst, &ctx->tags)) {
7643 xlate_report(ctx, "bonding refused admissibility, dropping");
7646 case BV_DROP_IF_MOVED:
7647 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7648 if (mac && mac->port.p != in_bundle &&
7649 (!is_gratuitous_arp(flow)
7650 || mac_entry_is_grat_arp_locked(mac))) {
7651 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7663 xlate_normal(struct action_xlate_ctx *ctx)
7665 struct ofport_dpif *in_port;
7666 struct ofbundle *in_bundle;
7667 struct mac_entry *mac;
7671 ctx->has_normal = true;
7673 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7674 ctx->packet != NULL, &in_port);
7676 xlate_report(ctx, "no input bundle, dropping");
7680 /* Drop malformed frames. */
7681 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7682 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7683 if (ctx->packet != NULL) {
7684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7685 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7686 "VLAN tag received on port %s",
7687 ctx->ofproto->up.name, in_bundle->name);
7689 xlate_report(ctx, "partial VLAN tag, dropping");
7693 /* Drop frames on bundles reserved for mirroring. */
7694 if (in_bundle->mirror_out) {
7695 if (ctx->packet != NULL) {
7696 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7697 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7698 "%s, which is reserved exclusively for mirroring",
7699 ctx->ofproto->up.name, in_bundle->name);
7701 xlate_report(ctx, "input port is mirror output port, dropping");
7706 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7707 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7708 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7711 vlan = input_vid_to_vlan(in_bundle, vid);
7713 /* Check other admissibility requirements. */
7714 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7718 /* Learn source MAC. */
7719 if (ctx->may_learn) {
7720 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7723 /* Determine output bundle. */
7724 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7727 if (mac->port.p != in_bundle) {
7728 xlate_report(ctx, "forwarding to learned port");
7729 output_normal(ctx, mac->port.p, vlan);
7731 xlate_report(ctx, "learned port is input port, dropping");
7734 struct ofbundle *bundle;
7736 xlate_report(ctx, "no learned MAC for destination, flooding");
7737 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7738 if (bundle != in_bundle
7739 && ofbundle_includes_vlan(bundle, vlan)
7740 && bundle->floodable
7741 && !bundle->mirror_out) {
7742 output_normal(ctx, bundle, vlan);
7745 ctx->nf_output_iface = NF_OUT_FLOOD;
7749 /* Optimized flow revalidation.
7751 * It's a difficult problem, in general, to tell which facets need to have
7752 * their actions recalculated whenever the OpenFlow flow table changes. We
7753 * don't try to solve that general problem: for most kinds of OpenFlow flow
7754 * table changes, we recalculate the actions for every facet. This is
7755 * relatively expensive, but it's good enough if the OpenFlow flow table
7756 * doesn't change very often.
7758 * However, we can expect one particular kind of OpenFlow flow table change to
7759 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7760 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7761 * table, we add a special case that applies to flow tables in which every rule
7762 * has the same form (that is, the same wildcards), except that the table is
7763 * also allowed to have a single "catch-all" flow that matches all packets. We
7764 * optimize this case by tagging all of the facets that resubmit into the table
7765 * and invalidating the same tag whenever a flow changes in that table. The
7766 * end result is that we revalidate just the facets that need it (and sometimes
7767 * a few more, but not all of the facets or even all of the facets that
7768 * resubmit to the table modified by MAC learning). */
7770 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7771 * into an OpenFlow table with the given 'basis'. */
7773 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7776 if (minimask_is_catchall(mask)) {
7779 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7780 return tag_create_deterministic(hash);
7784 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7785 * taggability of that table.
7787 * This function must be called after *each* change to a flow table. If you
7788 * skip calling it on some changes then the pointer comparisons at the end can
7789 * be invalid if you get unlucky. For example, if a flow removal causes a
7790 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7791 * different wildcards to be created with the same address, then this function
7792 * will incorrectly skip revalidation. */
7794 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7796 struct table_dpif *table = &ofproto->tables[table_id];
7797 const struct oftable *oftable = &ofproto->up.tables[table_id];
7798 struct cls_table *catchall, *other;
7799 struct cls_table *t;
7801 catchall = other = NULL;
7803 switch (hmap_count(&oftable->cls.tables)) {
7805 /* We could tag this OpenFlow table but it would make the logic a
7806 * little harder and it's a corner case that doesn't seem worth it
7812 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7813 if (cls_table_is_catchall(t)) {
7815 } else if (!other) {
7818 /* Indicate that we can't tag this by setting both tables to
7819 * NULL. (We know that 'catchall' is already NULL.) */
7826 /* Can't tag this table. */
7830 if (table->catchall_table != catchall || table->other_table != other) {
7831 table->catchall_table = catchall;
7832 table->other_table = other;
7833 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7837 /* Given 'rule' that has changed in some way (either it is a rule being
7838 * inserted, a rule being deleted, or a rule whose actions are being
7839 * modified), marks facets for revalidation to ensure that packets will be
7840 * forwarded correctly according to the new state of the flow table.
7842 * This function must be called after *each* change to a flow table. See
7843 * the comment on table_update_taggable() for more information. */
7845 rule_invalidate(const struct rule_dpif *rule)
7847 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7849 table_update_taggable(ofproto, rule->up.table_id);
7851 if (!ofproto->backer->need_revalidate) {
7852 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7854 if (table->other_table && rule->tag) {
7855 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7857 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7863 set_frag_handling(struct ofproto *ofproto_,
7864 enum ofp_config_flags frag_handling)
7866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7867 if (frag_handling != OFPC_FRAG_REASM) {
7868 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7876 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7877 const struct flow *flow,
7878 const struct ofpact *ofpacts, size_t ofpacts_len)
7880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7881 struct initial_vals initial_vals;
7882 struct odputil_keybuf keybuf;
7883 struct dpif_flow_stats stats;
7887 struct action_xlate_ctx ctx;
7888 uint64_t odp_actions_stub[1024 / 8];
7889 struct ofpbuf odp_actions;
7891 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7892 odp_flow_key_from_flow(&key, flow,
7893 ofp_port_to_odp_port(ofproto, flow->in_port));
7895 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7897 initial_vals.vlan_tci = flow->vlan_tci;
7898 initial_vals.tunnel_ip_tos = 0;
7899 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7900 packet_get_tcp_flags(packet, flow), packet);
7901 ctx.resubmit_stats = &stats;
7903 ofpbuf_use_stub(&odp_actions,
7904 odp_actions_stub, sizeof odp_actions_stub);
7905 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7906 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7907 odp_actions.data, odp_actions.size, packet);
7908 ofpbuf_uninit(&odp_actions);
7916 set_netflow(struct ofproto *ofproto_,
7917 const struct netflow_options *netflow_options)
7919 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7921 if (netflow_options) {
7922 if (!ofproto->netflow) {
7923 ofproto->netflow = netflow_create();
7925 return netflow_set_options(ofproto->netflow, netflow_options);
7927 netflow_destroy(ofproto->netflow);
7928 ofproto->netflow = NULL;
7934 get_netflow_ids(const struct ofproto *ofproto_,
7935 uint8_t *engine_type, uint8_t *engine_id)
7937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7939 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7943 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7945 if (!facet_is_controller_flow(facet) &&
7946 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7947 struct subfacet *subfacet;
7948 struct ofexpired expired;
7950 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7951 if (subfacet->path == SF_FAST_PATH) {
7952 struct dpif_flow_stats stats;
7954 subfacet_reinstall(subfacet, &stats);
7955 subfacet_update_stats(subfacet, &stats);
7959 expired.flow = facet->flow;
7960 expired.packet_count = facet->packet_count;
7961 expired.byte_count = facet->byte_count;
7962 expired.used = facet->used;
7963 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7968 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7970 struct facet *facet;
7972 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7973 send_active_timeout(ofproto, facet);
7977 static struct ofproto_dpif *
7978 ofproto_dpif_lookup(const char *name)
7980 struct ofproto_dpif *ofproto;
7982 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7983 hash_string(name, 0), &all_ofproto_dpifs) {
7984 if (!strcmp(ofproto->up.name, name)) {
7992 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7993 const char *argv[], void *aux OVS_UNUSED)
7995 struct ofproto_dpif *ofproto;
7998 ofproto = ofproto_dpif_lookup(argv[1]);
8000 unixctl_command_reply_error(conn, "no such bridge");
8003 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8005 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8006 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8010 unixctl_command_reply(conn, "table successfully flushed");
8014 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8015 const char *argv[], void *aux OVS_UNUSED)
8017 struct ds ds = DS_EMPTY_INITIALIZER;
8018 const struct ofproto_dpif *ofproto;
8019 const struct mac_entry *e;
8021 ofproto = ofproto_dpif_lookup(argv[1]);
8023 unixctl_command_reply_error(conn, "no such bridge");
8027 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8028 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8029 struct ofbundle *bundle = e->port.p;
8030 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8031 ofbundle_get_a_port(bundle)->odp_port,
8032 e->vlan, ETH_ADDR_ARGS(e->mac),
8033 mac_entry_age(ofproto->ml, e));
8035 unixctl_command_reply(conn, ds_cstr(&ds));
8040 struct action_xlate_ctx ctx;
8046 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8047 const struct rule_dpif *rule)
8049 ds_put_char_multiple(result, '\t', level);
8051 ds_put_cstr(result, "No match\n");
8055 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8056 table_id, ntohll(rule->up.flow_cookie));
8057 cls_rule_format(&rule->up.cr, result);
8058 ds_put_char(result, '\n');
8060 ds_put_char_multiple(result, '\t', level);
8061 ds_put_cstr(result, "OpenFlow ");
8062 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8063 ds_put_char(result, '\n');
8067 trace_format_flow(struct ds *result, int level, const char *title,
8068 struct trace_ctx *trace)
8070 ds_put_char_multiple(result, '\t', level);
8071 ds_put_format(result, "%s: ", title);
8072 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8073 ds_put_cstr(result, "unchanged");
8075 flow_format(result, &trace->ctx.flow);
8076 trace->flow = trace->ctx.flow;
8078 ds_put_char(result, '\n');
8082 trace_format_regs(struct ds *result, int level, const char *title,
8083 struct trace_ctx *trace)
8087 ds_put_char_multiple(result, '\t', level);
8088 ds_put_format(result, "%s:", title);
8089 for (i = 0; i < FLOW_N_REGS; i++) {
8090 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8092 ds_put_char(result, '\n');
8096 trace_format_odp(struct ds *result, int level, const char *title,
8097 struct trace_ctx *trace)
8099 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8101 ds_put_char_multiple(result, '\t', level);
8102 ds_put_format(result, "%s: ", title);
8103 format_odp_actions(result, odp_actions->data, odp_actions->size);
8104 ds_put_char(result, '\n');
8108 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8110 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8111 struct ds *result = trace->result;
8113 ds_put_char(result, '\n');
8114 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8115 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8116 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8117 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8121 trace_report(struct action_xlate_ctx *ctx, const char *s)
8123 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8124 struct ds *result = trace->result;
8126 ds_put_char_multiple(result, '\t', ctx->recurse);
8127 ds_put_cstr(result, s);
8128 ds_put_char(result, '\n');
8132 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8133 void *aux OVS_UNUSED)
8135 const char *dpname = argv[1];
8136 struct ofproto_dpif *ofproto;
8137 struct ofpbuf odp_key;
8138 struct ofpbuf *packet;
8139 struct initial_vals initial_vals;
8145 ofpbuf_init(&odp_key, 0);
8148 ofproto = ofproto_dpif_lookup(dpname);
8150 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8154 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8155 /* ofproto/trace dpname flow [-generate] */
8156 const char *flow_s = argv[2];
8157 const char *generate_s = argv[3];
8159 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8160 * flow. We guess which type it is based on whether 'flow_s' contains
8161 * an '(', since a datapath flow always contains '(') but an
8162 * OpenFlow-like flow should not (in fact it's allowed but I believe
8163 * that's not documented anywhere).
8165 * An alternative would be to try to parse 'flow_s' both ways, but then
8166 * it would be tricky giving a sensible error message. After all, do
8167 * you just say "syntax error" or do you present both error messages?
8168 * Both choices seem lousy. */
8169 if (strchr(flow_s, '(')) {
8172 /* Convert string to datapath key. */
8173 ofpbuf_init(&odp_key, 0);
8174 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8176 unixctl_command_reply_error(conn, "Bad flow syntax");
8180 /* The user might have specified the wrong ofproto but within the
8181 * same backer. That's OK, ofproto_receive() can find the right
8183 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8184 odp_key.size, &flow, NULL, &ofproto, NULL,
8186 unixctl_command_reply_error(conn, "Invalid flow");
8189 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8193 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8195 unixctl_command_reply_error(conn, error_s);
8200 initial_vals.vlan_tci = flow.vlan_tci;
8201 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8204 /* Generate a packet, if requested. */
8206 packet = ofpbuf_new(0);
8207 flow_compose(packet, &flow);
8209 } else if (argc == 7) {
8210 /* ofproto/trace dpname priority tun_id in_port mark packet */
8211 const char *priority_s = argv[2];
8212 const char *tun_id_s = argv[3];
8213 const char *in_port_s = argv[4];
8214 const char *mark_s = argv[5];
8215 const char *packet_s = argv[6];
8216 uint32_t in_port = atoi(in_port_s);
8217 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8218 uint32_t priority = atoi(priority_s);
8219 uint32_t mark = atoi(mark_s);
8222 msg = eth_from_hex(packet_s, &packet);
8224 unixctl_command_reply_error(conn, msg);
8228 ds_put_cstr(&result, "Packet: ");
8229 s = ofp_packet_to_string(packet->data, packet->size);
8230 ds_put_cstr(&result, s);
8233 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8234 flow.tunnel.tun_id = tun_id;
8235 initial_vals.vlan_tci = flow.vlan_tci;
8236 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8238 unixctl_command_reply_error(conn, "Bad command syntax");
8242 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8243 unixctl_command_reply(conn, ds_cstr(&result));
8246 ds_destroy(&result);
8247 ofpbuf_delete(packet);
8248 ofpbuf_uninit(&odp_key);
8252 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8253 const struct ofpbuf *packet,
8254 const struct initial_vals *initial_vals, struct ds *ds)
8256 struct rule_dpif *rule;
8258 ds_put_cstr(ds, "Flow: ");
8259 flow_format(ds, flow);
8260 ds_put_char(ds, '\n');
8262 rule = rule_dpif_lookup(ofproto, flow);
8264 trace_format_rule(ds, 0, 0, rule);
8265 if (rule == ofproto->miss_rule) {
8266 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8267 } else if (rule == ofproto->no_packet_in_rule) {
8268 ds_put_cstr(ds, "\nNo match, packets dropped because "
8269 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8273 uint64_t odp_actions_stub[1024 / 8];
8274 struct ofpbuf odp_actions;
8276 struct trace_ctx trace;
8279 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8282 ofpbuf_use_stub(&odp_actions,
8283 odp_actions_stub, sizeof odp_actions_stub);
8284 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8285 rule, tcp_flags, packet);
8286 trace.ctx.resubmit_hook = trace_resubmit;
8287 trace.ctx.report_hook = trace_report;
8288 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8291 ds_put_char(ds, '\n');
8292 trace_format_flow(ds, 0, "Final flow", &trace);
8293 ds_put_cstr(ds, "Datapath actions: ");
8294 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8295 ofpbuf_uninit(&odp_actions);
8297 if (trace.ctx.slow) {
8298 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8299 "slow path because it:");
8300 switch (trace.ctx.slow) {
8302 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8305 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8308 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8310 case SLOW_CONTROLLER:
8311 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8312 "to the OpenFlow controller.");
8322 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8323 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8326 unixctl_command_reply(conn, NULL);
8330 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8331 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8334 unixctl_command_reply(conn, NULL);
8337 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8338 * 'reply' describing the results. */
8340 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8342 struct facet *facet;
8346 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8347 if (!facet_check_consistency(facet)) {
8352 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8356 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8357 ofproto->up.name, errors);
8359 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8364 ofproto_dpif_self_check(struct unixctl_conn *conn,
8365 int argc, const char *argv[], void *aux OVS_UNUSED)
8367 struct ds reply = DS_EMPTY_INITIALIZER;
8368 struct ofproto_dpif *ofproto;
8371 ofproto = ofproto_dpif_lookup(argv[1]);
8373 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8374 "ofproto/list for help)");
8377 ofproto_dpif_self_check__(ofproto, &reply);
8379 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8380 ofproto_dpif_self_check__(ofproto, &reply);
8384 unixctl_command_reply(conn, ds_cstr(&reply));
8388 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8389 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8390 * to destroy 'ofproto_shash' and free the returned value. */
8391 static const struct shash_node **
8392 get_ofprotos(struct shash *ofproto_shash)
8394 const struct ofproto_dpif *ofproto;
8396 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8397 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8398 shash_add_nocopy(ofproto_shash, name, ofproto);
8401 return shash_sort(ofproto_shash);
8405 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8406 const char *argv[] OVS_UNUSED,
8407 void *aux OVS_UNUSED)
8409 struct ds ds = DS_EMPTY_INITIALIZER;
8410 struct shash ofproto_shash;
8411 const struct shash_node **sorted_ofprotos;
8414 shash_init(&ofproto_shash);
8415 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8416 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8417 const struct shash_node *node = sorted_ofprotos[i];
8418 ds_put_format(&ds, "%s\n", node->name);
8421 shash_destroy(&ofproto_shash);
8422 free(sorted_ofprotos);
8424 unixctl_command_reply(conn, ds_cstr(&ds));
8429 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8431 const struct shash_node **ports;
8433 struct avg_subfacet_rates lifetime;
8434 unsigned long long int minutes;
8435 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8437 minutes = (time_msec() - ofproto->created) / min_ms;
8440 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8442 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8445 lifetime.add_rate = 0.0;
8446 lifetime.del_rate = 0.0;
8449 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8450 dpif_name(ofproto->backer->dpif));
8452 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8453 ofproto->n_hit, ofproto->n_missed);
8454 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8455 " life span: %llu(ms)\n",
8456 hmap_count(&ofproto->subfacets),
8457 avg_subfacet_count(ofproto),
8458 ofproto->max_n_subfacet,
8459 avg_subfacet_life_span(ofproto));
8460 if (minutes >= 60) {
8461 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8463 if (minutes >= 60 * 24) {
8464 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8466 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8468 ports = shash_sort(&ofproto->up.port_by_name);
8469 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8470 const struct shash_node *node = ports[i];
8471 struct ofport *ofport = node->data;
8472 const char *name = netdev_get_name(ofport->netdev);
8473 const char *type = netdev_get_type(ofport->netdev);
8476 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8478 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8479 if (odp_port != OVSP_NONE) {
8480 ds_put_format(ds, "%"PRIu32":", odp_port);
8482 ds_put_cstr(ds, "none:");
8485 if (strcmp(type, "system")) {
8486 struct netdev *netdev;
8489 ds_put_format(ds, " (%s", type);
8491 error = netdev_open(name, type, &netdev);
8496 error = netdev_get_config(netdev, &config);
8498 const struct smap_node **nodes;
8501 nodes = smap_sort(&config);
8502 for (i = 0; i < smap_count(&config); i++) {
8503 const struct smap_node *node = nodes[i];
8504 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8505 node->key, node->value);
8509 smap_destroy(&config);
8511 netdev_close(netdev);
8513 ds_put_char(ds, ')');
8515 ds_put_char(ds, '\n');
8521 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8522 const char *argv[], void *aux OVS_UNUSED)
8524 struct ds ds = DS_EMPTY_INITIALIZER;
8525 const struct ofproto_dpif *ofproto;
8529 for (i = 1; i < argc; i++) {
8530 ofproto = ofproto_dpif_lookup(argv[i]);
8532 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8533 "for help)", argv[i]);
8534 unixctl_command_reply_error(conn, ds_cstr(&ds));
8537 show_dp_format(ofproto, &ds);
8540 struct shash ofproto_shash;
8541 const struct shash_node **sorted_ofprotos;
8544 shash_init(&ofproto_shash);
8545 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8546 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8547 const struct shash_node *node = sorted_ofprotos[i];
8548 show_dp_format(node->data, &ds);
8551 shash_destroy(&ofproto_shash);
8552 free(sorted_ofprotos);
8555 unixctl_command_reply(conn, ds_cstr(&ds));
8560 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8561 int argc OVS_UNUSED, const char *argv[],
8562 void *aux OVS_UNUSED)
8564 struct ds ds = DS_EMPTY_INITIALIZER;
8565 const struct ofproto_dpif *ofproto;
8566 struct subfacet *subfacet;
8568 ofproto = ofproto_dpif_lookup(argv[1]);
8570 unixctl_command_reply_error(conn, "no such bridge");
8574 update_stats(ofproto->backer);
8576 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8577 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8579 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8580 subfacet->dp_packet_count, subfacet->dp_byte_count);
8581 if (subfacet->used) {
8582 ds_put_format(&ds, "%.3fs",
8583 (time_msec() - subfacet->used) / 1000.0);
8585 ds_put_format(&ds, "never");
8587 if (subfacet->facet->tcp_flags) {
8588 ds_put_cstr(&ds, ", flags:");
8589 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8592 ds_put_cstr(&ds, ", actions:");
8593 if (subfacet->slow) {
8594 uint64_t slow_path_stub[128 / 8];
8595 const struct nlattr *actions;
8598 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8599 slow_path_stub, sizeof slow_path_stub,
8600 &actions, &actions_len);
8601 format_odp_actions(&ds, actions, actions_len);
8603 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8605 ds_put_char(&ds, '\n');
8608 unixctl_command_reply(conn, ds_cstr(&ds));
8613 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8614 int argc OVS_UNUSED, const char *argv[],
8615 void *aux OVS_UNUSED)
8617 struct ds ds = DS_EMPTY_INITIALIZER;
8618 struct ofproto_dpif *ofproto;
8620 ofproto = ofproto_dpif_lookup(argv[1]);
8622 unixctl_command_reply_error(conn, "no such bridge");
8626 flush(&ofproto->up);
8628 unixctl_command_reply(conn, ds_cstr(&ds));
8633 ofproto_dpif_unixctl_init(void)
8635 static bool registered;
8641 unixctl_command_register(
8643 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8644 2, 6, ofproto_unixctl_trace, NULL);
8645 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8646 ofproto_unixctl_fdb_flush, NULL);
8647 unixctl_command_register("fdb/show", "bridge", 1, 1,
8648 ofproto_unixctl_fdb_show, NULL);
8649 unixctl_command_register("ofproto/clog", "", 0, 0,
8650 ofproto_dpif_clog, NULL);
8651 unixctl_command_register("ofproto/unclog", "", 0, 0,
8652 ofproto_dpif_unclog, NULL);
8653 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8654 ofproto_dpif_self_check, NULL);
8655 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8656 ofproto_unixctl_dpif_dump_dps, NULL);
8657 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8658 ofproto_unixctl_dpif_show, NULL);
8659 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8660 ofproto_unixctl_dpif_dump_flows, NULL);
8661 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8662 ofproto_unixctl_dpif_del_flows, NULL);
8665 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8667 * This is deprecated. It is only for compatibility with broken device drivers
8668 * in old versions of Linux that do not properly support VLANs when VLAN
8669 * devices are not used. When broken device drivers are no longer in
8670 * widespread use, we will delete these interfaces. */
8673 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8676 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8678 if (realdev_ofp_port == ofport->realdev_ofp_port
8679 && vid == ofport->vlandev_vid) {
8683 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8685 if (ofport->realdev_ofp_port) {
8688 if (realdev_ofp_port && ofport->bundle) {
8689 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8690 * themselves be part of a bundle. */
8691 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8694 ofport->realdev_ofp_port = realdev_ofp_port;
8695 ofport->vlandev_vid = vid;
8697 if (realdev_ofp_port) {
8698 vsp_add(ofport, realdev_ofp_port, vid);
8705 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8707 return hash_2words(realdev_ofp_port, vid);
8710 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8711 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8712 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8713 * it would return the port number of eth0.9.
8715 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8716 * function just returns its 'realdev_odp_port' argument. */
8718 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8719 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8721 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8722 uint16_t realdev_ofp_port;
8723 int vid = vlan_tci_to_vid(vlan_tci);
8724 const struct vlan_splinter *vsp;
8726 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8727 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8728 hash_realdev_vid(realdev_ofp_port, vid),
8729 &ofproto->realdev_vid_map) {
8730 if (vsp->realdev_ofp_port == realdev_ofp_port
8731 && vsp->vid == vid) {
8732 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8736 return realdev_odp_port;
8739 static struct vlan_splinter *
8740 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8742 struct vlan_splinter *vsp;
8744 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8745 &ofproto->vlandev_map) {
8746 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8754 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8755 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8756 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8757 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8758 * eth0 and store 9 in '*vid'.
8760 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8761 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8764 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8765 uint16_t vlandev_ofp_port, int *vid)
8767 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8768 const struct vlan_splinter *vsp;
8770 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8775 return vsp->realdev_ofp_port;
8781 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8782 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8783 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8784 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8785 * always the case unless VLAN splinters are enabled), returns false without
8786 * making any changes. */
8788 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8793 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8798 /* Cause the flow to be processed as if it came in on the real device with
8799 * the VLAN device's VLAN ID. */
8800 flow->in_port = realdev;
8801 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8806 vsp_remove(struct ofport_dpif *port)
8808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8809 struct vlan_splinter *vsp;
8811 vsp = vlandev_find(ofproto, port->up.ofp_port);
8813 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8814 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8817 port->realdev_ofp_port = 0;
8819 VLOG_ERR("missing vlan device record");
8824 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8826 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8828 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8829 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8830 == realdev_ofp_port)) {
8831 struct vlan_splinter *vsp;
8833 vsp = xmalloc(sizeof *vsp);
8834 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8835 hash_int(port->up.ofp_port, 0));
8836 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8837 hash_realdev_vid(realdev_ofp_port, vid));
8838 vsp->realdev_ofp_port = realdev_ofp_port;
8839 vsp->vlandev_ofp_port = port->up.ofp_port;
8842 port->realdev_ofp_port = realdev_ofp_port;
8844 VLOG_ERR("duplicate vlan device record");
8849 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8851 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8852 return ofport ? ofport->odp_port : OVSP_NONE;
8855 static struct ofport_dpif *
8856 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8858 struct ofport_dpif *port;
8860 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8861 hash_int(odp_port, 0),
8862 &backer->odp_to_ofport_map) {
8863 if (port->odp_port == odp_port) {
8872 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8874 struct ofport_dpif *port;
8876 port = odp_port_to_ofport(ofproto->backer, odp_port);
8877 if (port && &ofproto->up == port->up.ofproto) {
8878 return port->up.ofp_port;
8883 static unsigned long long int
8884 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8886 unsigned long long int dc;
8887 unsigned long long int avg;
8889 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8890 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8896 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8900 if (ofproto->n_update_stats) {
8901 avg_c = (double)ofproto->total_subfacet_count
8902 / ofproto->n_update_stats;
8909 show_dp_rates(struct ds *ds, const char *heading,
8910 const struct avg_subfacet_rates *rates)
8912 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8913 heading, rates->add_rate, rates->del_rate);
8917 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8919 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8920 hmap_count(&ofproto->subfacets));
8923 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8924 * most heavily weighted element. 'base' designates the rate of decay: after
8925 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8928 exp_mavg(double *avg, int base, double new)
8930 *avg = (*avg * (base - 1) + new) / base;
8934 update_moving_averages(struct ofproto_dpif *ofproto)
8936 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8938 /* Update hourly averages on the minute boundaries. */
8939 if (time_msec() - ofproto->last_minute >= min_ms) {
8940 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8941 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8943 /* Update daily averages on the hour boundaries. */
8944 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8945 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8946 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8949 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8950 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8951 ofproto->subfacet_add_count = 0;
8952 ofproto->subfacet_del_count = 0;
8953 ofproto->last_minute += min_ms;
8958 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8960 ofproto->n_hit += delta;
8963 const struct ofproto_class ofproto_dpif_class = {
8998 port_is_lacp_current,
8999 NULL, /* rule_choose_table */
9006 rule_modify_actions,
9018 get_stp_port_status,
9025 is_mirror_output_bundle,
9026 forward_bpdu_changed,
9027 set_mac_table_config,