2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-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 /* The packet corresponding to 'flow', or a null pointer if we are
221 * revalidating without a packet to refer to. */
222 const struct ofpbuf *packet;
224 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
225 * actions update the flow table?
227 * We want to update these tables if we are actually processing a packet,
228 * or if we are accounting for packets that the datapath has processed, but
229 * not if we are just revalidating. */
232 /* The rule that we are currently translating, or NULL. */
233 struct rule_dpif *rule;
235 /* Union of the set of TCP flags seen so far in this flow. (Used only by
236 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
240 /* If nonnull, flow translation calls this function just before executing a
241 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
242 * when the recursion depth is exceeded.
244 * 'rule' is the rule being submitted into. It will be null if the
245 * resubmit or OFPP_TABLE action didn't find a matching rule.
247 * This is normally null so the client has to set it manually after
248 * calling action_xlate_ctx_init(). */
249 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
251 /* If nonnull, flow translation calls this function to report some
252 * significant decision, e.g. to explain why OFPP_NORMAL translation
253 * dropped a packet. */
254 void (*report_hook)(struct action_xlate_ctx *, const char *s);
256 /* If nonnull, flow translation credits the specified statistics to each
257 * rule reached through a resubmit or OFPP_TABLE action.
259 * This is normally null so the client has to set it manually after
260 * calling action_xlate_ctx_init(). */
261 const struct dpif_flow_stats *resubmit_stats;
263 /* xlate_actions() initializes and uses these members. The client might want
264 * to look at them after it returns. */
266 struct ofpbuf *odp_actions; /* Datapath actions. */
267 tag_type tags; /* Tags associated with actions. */
268 enum slow_path_reason slow; /* 0 if fast path may be used. */
269 bool has_learn; /* Actions include NXAST_LEARN? */
270 bool has_normal; /* Actions output to OFPP_NORMAL? */
271 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
272 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
273 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
275 /* xlate_actions() initializes and uses these members, but the client has no
276 * reason to look at them. */
278 int recurse; /* Recursion level, via xlate_table_action. */
279 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
280 struct flow base_flow; /* Flow at the last commit. */
281 uint32_t orig_skb_priority; /* Priority when packet arrived. */
282 uint8_t table_id; /* OpenFlow table ID where flow was found. */
283 uint32_t sflow_n_outputs; /* Number of output ports. */
284 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
285 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
286 bool exit; /* No further actions should be processed. */
287 struct flow orig_flow; /* Copy of original flow. */
290 /* Initial values of fields of the packet that may be changed during
291 * flow processing and needed later. */
292 struct initial_vals {
293 /* This is the value of vlan_tci in the packet as actually received from
294 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
295 * was received via a VLAN splinter. In that case, this value is 0
296 * (because the packet as actually received from the dpif had no 802.1Q
297 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
300 * This member should be removed when the VLAN splinters feature is no
304 /* If received on a tunnel, the IP TOS value of the tunnel. */
305 uint8_t tunnel_ip_tos;
308 static void action_xlate_ctx_init(struct action_xlate_ctx *,
309 struct ofproto_dpif *, const struct flow *,
310 const struct initial_vals *initial_vals,
312 uint8_t tcp_flags, const struct ofpbuf *);
313 static void xlate_actions(struct action_xlate_ctx *,
314 const struct ofpact *ofpacts, size_t ofpacts_len,
315 struct ofpbuf *odp_actions);
316 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts,
319 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
320 uint8_t table_id, bool may_packet_in);
322 static size_t put_userspace_action(const struct ofproto_dpif *,
323 struct ofpbuf *odp_actions,
325 const union user_action_cookie *);
327 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
328 enum slow_path_reason,
329 uint64_t *stub, size_t stub_size,
330 const struct nlattr **actionsp,
331 size_t *actions_lenp);
333 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
335 /* A subfacet (see "struct subfacet" below) has three possible installation
338 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
339 * case just after the subfacet is created, just before the subfacet is
340 * destroyed, or if the datapath returns an error when we try to install a
343 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
345 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
346 * ofproto_dpif is installed in the datapath.
349 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
350 SF_FAST_PATH, /* Full actions are installed. */
351 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
354 static const char *subfacet_path_to_string(enum subfacet_path);
356 /* A dpif flow and actions associated with a facet.
358 * See also the large comment on struct facet. */
361 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
362 struct list list_node; /* In struct facet's 'facets' list. */
363 struct facet *facet; /* Owning facet. */
365 enum odp_key_fitness key_fitness;
369 long long int used; /* Time last used; time created if not used. */
370 long long int created; /* Time created. */
372 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
373 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
377 * These should be essentially identical for every subfacet in a facet, but
378 * may differ in trivial ways due to VLAN splinters. */
379 size_t actions_len; /* Number of bytes in actions[]. */
380 struct nlattr *actions; /* Datapath actions. */
382 enum slow_path_reason slow; /* 0 if fast path may be used. */
383 enum subfacet_path path; /* Installed in datapath? */
385 /* Initial values of the packet that may be needed later. */
386 struct initial_vals initial_vals;
388 /* Datapath port the packet arrived on. This is needed to remove
389 * flows for ports that are no longer part of the bridge. Since the
390 * flow definition only has the OpenFlow port number and the port is
391 * no longer part of the bridge, we can't determine the datapath port
392 * number needed to delete the flow from the datapath. */
393 uint32_t odp_in_port;
396 #define SUBFACET_DESTROY_MAX_BATCH 50
398 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
400 static struct subfacet *subfacet_find(struct ofproto_dpif *,
401 const struct nlattr *key, size_t key_len,
403 static void subfacet_destroy(struct subfacet *);
404 static void subfacet_destroy__(struct subfacet *);
405 static void subfacet_destroy_batch(struct ofproto_dpif *,
406 struct subfacet **, int n);
407 static void subfacet_reset_dp_stats(struct subfacet *,
408 struct dpif_flow_stats *);
409 static void subfacet_update_time(struct subfacet *, long long int used);
410 static void subfacet_update_stats(struct subfacet *,
411 const struct dpif_flow_stats *);
412 static void subfacet_make_actions(struct subfacet *,
413 const struct ofpbuf *packet,
414 struct ofpbuf *odp_actions);
415 static int subfacet_install(struct subfacet *,
416 const struct nlattr *actions, size_t actions_len,
417 struct dpif_flow_stats *, enum slow_path_reason);
418 static void subfacet_uninstall(struct subfacet *);
420 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
422 /* An exact-match instantiation of an OpenFlow flow.
424 * A facet associates a "struct flow", which represents the Open vSwitch
425 * userspace idea of an exact-match flow, with one or more subfacets. Each
426 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
427 * the facet. When the kernel module (or other dpif implementation) and Open
428 * vSwitch userspace agree on the definition of a flow key, there is exactly
429 * one subfacet per facet. If the dpif implementation supports more-specific
430 * flow matching than userspace, however, a facet can have more than one
431 * subfacet, each of which corresponds to some distinction in flow that
432 * userspace simply doesn't understand.
434 * Flow expiration works in terms of subfacets, so a facet must have at least
435 * one subfacet or it will never expire, leaking memory. */
438 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
439 struct list list_node; /* In owning rule's 'facets' list. */
440 struct rule_dpif *rule; /* Owning rule. */
443 struct list subfacets;
444 long long int used; /* Time last used; time created if not used. */
451 * - Do include packets and bytes sent "by hand", e.g. with
454 * - Do include packets and bytes that were obtained from the datapath
455 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
456 * DPIF_FP_ZERO_STATS).
458 * - Do not include packets or bytes that can be obtained from the
459 * datapath for any existing subfacet.
461 uint64_t packet_count; /* Number of packets received. */
462 uint64_t byte_count; /* Number of bytes received. */
464 /* Resubmit statistics. */
465 uint64_t prev_packet_count; /* Number of packets from last stats push. */
466 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
467 long long int prev_used; /* Used time from last stats push. */
470 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
471 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
472 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
474 /* Properties of datapath actions.
476 * Every subfacet has its own actions because actions can differ slightly
477 * between splintered and non-splintered subfacets due to the VLAN tag
478 * being initially different (present vs. absent). All of them have these
479 * properties in common so we just store one copy of them here. */
480 bool has_learn; /* Actions include NXAST_LEARN? */
481 bool has_normal; /* Actions output to OFPP_NORMAL? */
482 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
483 tag_type tags; /* Tags that would require revalidation. */
484 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
486 /* Storage for a single subfacet, to reduce malloc() time and space
487 * overhead. (A facet always has at least one subfacet and in the common
488 * case has exactly one subfacet.) */
489 struct subfacet one_subfacet;
491 long long int learn_rl; /* Rate limiter for facet_learn(). */
494 static struct facet *facet_create(struct rule_dpif *,
495 const struct flow *, uint32_t hash);
496 static void facet_remove(struct facet *);
497 static void facet_free(struct facet *);
499 static struct facet *facet_find(struct ofproto_dpif *,
500 const struct flow *, uint32_t hash);
501 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
502 const struct flow *, uint32_t hash);
503 static void facet_revalidate(struct facet *);
504 static bool facet_check_consistency(struct facet *);
506 static void facet_flush_stats(struct facet *);
508 static void facet_update_time(struct facet *, long long int used);
509 static void facet_reset_counters(struct facet *);
510 static void facet_push_stats(struct facet *);
511 static void facet_learn(struct facet *);
512 static void facet_account(struct facet *);
513 static void push_all_stats(void);
515 static bool facet_is_controller_flow(struct facet *);
518 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
522 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
523 struct list bundle_node; /* In struct ofbundle's "ports" list. */
524 struct cfm *cfm; /* Connectivity Fault Management, if any. */
525 tag_type tag; /* Tag associated with this port. */
526 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
527 bool may_enable; /* May be enabled in bonds. */
528 long long int carrier_seq; /* Carrier status changes. */
529 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
532 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
533 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
534 long long int stp_state_entered;
536 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
538 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
540 * This is deprecated. It is only for compatibility with broken device
541 * drivers in old versions of Linux that do not properly support VLANs when
542 * VLAN devices are not used. When broken device drivers are no longer in
543 * widespread use, we will delete these interfaces. */
544 uint16_t realdev_ofp_port;
548 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
549 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
550 * traffic egressing the 'ofport' with that priority should be marked with. */
551 struct priority_to_dscp {
552 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
553 uint32_t priority; /* Priority of this queue (see struct flow). */
555 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
558 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
560 * This is deprecated. It is only for compatibility with broken device drivers
561 * in old versions of Linux that do not properly support VLANs when VLAN
562 * devices are not used. When broken device drivers are no longer in
563 * widespread use, we will delete these interfaces. */
564 struct vlan_splinter {
565 struct hmap_node realdev_vid_node;
566 struct hmap_node vlandev_node;
567 uint16_t realdev_ofp_port;
568 uint16_t vlandev_ofp_port;
572 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
573 uint32_t realdev, ovs_be16 vlan_tci);
574 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
575 static void vsp_remove(struct ofport_dpif *);
576 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
578 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
580 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
583 static struct ofport_dpif *
584 ofport_dpif_cast(const struct ofport *ofport)
586 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
587 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
590 static void port_run(struct ofport_dpif *);
591 static void port_run_fast(struct ofport_dpif *);
592 static void port_wait(struct ofport_dpif *);
593 static int set_cfm(struct ofport *, const struct cfm_settings *);
594 static void ofport_clear_priorities(struct ofport_dpif *);
595 static void run_fast_rl(void);
597 struct dpif_completion {
598 struct list list_node;
599 struct ofoperation *op;
602 /* Extra information about a classifier table.
603 * Currently used just for optimized flow revalidation. */
605 /* If either of these is nonnull, then this table has a form that allows
606 * flows to be tagged to avoid revalidating most flows for the most common
607 * kinds of flow table changes. */
608 struct cls_table *catchall_table; /* Table that wildcards all fields. */
609 struct cls_table *other_table; /* Table with any other wildcard set. */
610 uint32_t basis; /* Keeps each table's tags separate. */
613 /* Reasons that we might need to revalidate every facet, and corresponding
616 * A value of 0 means that there is no need to revalidate.
618 * It would be nice to have some cleaner way to integrate with coverage
619 * counters, but with only a few reasons I guess this is good enough for
621 enum revalidate_reason {
622 REV_RECONFIGURE = 1, /* Switch configuration changed. */
623 REV_STP, /* Spanning tree protocol port status change. */
624 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
625 REV_FLOW_TABLE, /* Flow table changed. */
626 REV_INCONSISTENCY /* Facet self-check failed. */
628 COVERAGE_DEFINE(rev_reconfigure);
629 COVERAGE_DEFINE(rev_stp);
630 COVERAGE_DEFINE(rev_port_toggled);
631 COVERAGE_DEFINE(rev_flow_table);
632 COVERAGE_DEFINE(rev_inconsistency);
634 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
635 * These are datapath flows which have no associated ofproto, if they did we
636 * would use facets. */
638 struct hmap_node hmap_node;
643 /* All datapaths of a given type share a single dpif backer instance. */
648 struct timer next_expiration;
649 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
651 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
653 /* Facet revalidation flags applying to facets which use this backer. */
654 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
655 struct tag_set revalidate_set; /* Revalidate only matching facets. */
657 struct hmap drop_keys; /* Set of dropped odp keys. */
660 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
661 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
663 static void drop_key_clear(struct dpif_backer *);
664 static struct ofport_dpif *
665 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
667 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
669 struct avg_subfacet_rates {
670 double add_rate; /* Moving average of new flows created per minute. */
671 double del_rate; /* Moving average of flows deleted per minute. */
673 static void show_dp_rates(struct ds *ds, const char *heading,
674 const struct avg_subfacet_rates *rates);
675 static void exp_mavg(double *avg, int base, double new);
677 struct ofproto_dpif {
678 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
680 struct dpif_backer *backer;
682 /* Special OpenFlow rules. */
683 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
684 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
690 struct netflow *netflow;
691 struct dpif_sflow *sflow;
692 struct hmap bundles; /* Contains "struct ofbundle"s. */
693 struct mac_learning *ml;
694 struct ofmirror *mirrors[MAX_MIRRORS];
696 bool has_bonded_bundles;
700 struct hmap subfacets;
701 struct governor *governor;
702 long long int consistency_rl;
705 struct table_dpif tables[N_TABLES];
707 /* Support for debugging async flow mods. */
708 struct list completions;
710 bool has_bundle_action; /* True when the first bundle action appears. */
711 struct netdev_stats stats; /* To account packets generated and consumed in
716 long long int stp_last_tick;
718 /* VLAN splinters. */
719 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
720 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
723 struct sset ports; /* Set of standard port names. */
724 struct sset ghost_ports; /* Ports with no datapath port. */
725 struct sset port_poll_set; /* Queued names for port_poll() reply. */
726 int port_poll_errno; /* Last errno for port_poll() reply. */
728 /* Per ofproto's dpif stats. */
732 /* Subfacet statistics.
734 * These keep track of the total number of subfacets added and deleted and
735 * flow life span. They are useful for computing the flow rates stats
736 * exposed via "ovs-appctl dpif/show". The goal is to learn about
737 * traffic patterns in ways that we can use later to improve Open vSwitch
738 * performance in new situations. */
739 long long int created; /* Time when it is created. */
740 unsigned int max_n_subfacet; /* Maximum number of flows */
742 /* The average number of subfacets... */
743 struct avg_subfacet_rates hourly; /* ...over the last hour. */
744 struct avg_subfacet_rates daily; /* ...over the last day. */
745 long long int last_minute; /* Last time 'hourly' was updated. */
747 /* Number of subfacets added or deleted since 'last_minute'. */
748 unsigned int subfacet_add_count;
749 unsigned int subfacet_del_count;
751 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
752 unsigned long long int total_subfacet_add_count;
753 unsigned long long int total_subfacet_del_count;
755 /* Sum of the number of milliseconds that each subfacet existed,
756 * over the subfacets that have been added and then later deleted. */
757 unsigned long long int total_subfacet_life_span;
759 /* Incremented by the number of currently existing subfacets, each
760 * time we pull statistics from the kernel. */
761 unsigned long long int total_subfacet_count;
763 /* Number of times we pull statistics from the kernel. */
764 unsigned long long int n_update_stats;
766 static unsigned long long int avg_subfacet_life_span(
767 const struct ofproto_dpif *);
768 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
769 static void update_moving_averages(struct ofproto_dpif *ofproto);
770 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
772 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
774 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
775 * for debugging the asynchronous flow_mod implementation.) */
778 /* All existing ofproto_dpif instances, indexed by ->up.name. */
779 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
781 static void ofproto_dpif_unixctl_init(void);
783 static struct ofproto_dpif *
784 ofproto_dpif_cast(const struct ofproto *ofproto)
786 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
787 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
790 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
792 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
794 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
795 const struct ofpbuf *,
796 const struct initial_vals *, struct ds *);
798 /* Packet processing. */
799 static void update_learning_table(struct ofproto_dpif *,
800 const struct flow *, int vlan,
803 #define FLOW_MISS_MAX_BATCH 50
804 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
806 /* Flow expiration. */
807 static int expire(struct dpif_backer *);
810 static void send_netflow_active_timeouts(struct ofproto_dpif *);
813 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
814 static size_t compose_sflow_action(const struct ofproto_dpif *,
815 struct ofpbuf *odp_actions,
816 const struct flow *, uint32_t odp_port);
817 static void add_mirror_actions(struct action_xlate_ctx *ctx,
818 const struct flow *flow);
819 /* Global variables. */
820 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
822 /* Initial mappings of port to bridge mappings. */
823 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
825 /* Factory functions. */
828 init(const struct shash *iface_hints)
830 struct shash_node *node;
832 /* Make a local copy, since we don't own 'iface_hints' elements. */
833 SHASH_FOR_EACH(node, iface_hints) {
834 const struct iface_hint *orig_hint = node->data;
835 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
837 new_hint->br_name = xstrdup(orig_hint->br_name);
838 new_hint->br_type = xstrdup(orig_hint->br_type);
839 new_hint->ofp_port = orig_hint->ofp_port;
841 shash_add(&init_ofp_ports, node->name, new_hint);
846 enumerate_types(struct sset *types)
848 dp_enumerate_types(types);
852 enumerate_names(const char *type, struct sset *names)
854 struct ofproto_dpif *ofproto;
857 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
858 if (strcmp(type, ofproto->up.type)) {
861 sset_add(names, ofproto->up.name);
868 del(const char *type, const char *name)
873 error = dpif_open(name, type, &dpif);
875 error = dpif_delete(dpif);
882 port_open_type(const char *datapath_type, const char *port_type)
884 return dpif_port_open_type(datapath_type, port_type);
887 /* Type functions. */
889 static struct ofproto_dpif *
890 lookup_ofproto_dpif_by_port_name(const char *name)
892 struct ofproto_dpif *ofproto;
894 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
895 if (sset_contains(&ofproto->ports, name)) {
904 type_run(const char *type)
906 static long long int push_timer = LLONG_MIN;
907 struct dpif_backer *backer;
911 backer = shash_find_data(&all_dpif_backers, type);
913 /* This is not necessarily a problem, since backers are only
914 * created on demand. */
918 dpif_run(backer->dpif);
920 /* The most natural place to push facet statistics is when they're pulled
921 * from the datapath. However, when there are many flows in the datapath,
922 * this expensive operation can occur so frequently, that it reduces our
923 * ability to quickly set up flows. To reduce the cost, we push statistics
925 if (time_msec() > push_timer) {
926 push_timer = time_msec() + 2000;
930 if (backer->need_revalidate
931 || !tag_set_is_empty(&backer->revalidate_set)) {
932 struct tag_set revalidate_set = backer->revalidate_set;
933 bool need_revalidate = backer->need_revalidate;
934 struct ofproto_dpif *ofproto;
935 struct simap_node *node;
936 struct simap tmp_backers;
938 /* Handle tunnel garbage collection. */
939 simap_init(&tmp_backers);
940 simap_swap(&backer->tnl_backers, &tmp_backers);
942 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
943 struct ofport_dpif *iter;
945 if (backer != ofproto->backer) {
949 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
952 if (!iter->tnl_port) {
956 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
957 node = simap_find(&tmp_backers, dp_port);
959 simap_put(&backer->tnl_backers, dp_port, node->data);
960 simap_delete(&tmp_backers, node);
961 node = simap_find(&backer->tnl_backers, dp_port);
963 node = simap_find(&backer->tnl_backers, dp_port);
965 uint32_t odp_port = UINT32_MAX;
967 if (!dpif_port_add(backer->dpif, iter->up.netdev,
969 simap_put(&backer->tnl_backers, dp_port, odp_port);
970 node = simap_find(&backer->tnl_backers, dp_port);
975 iter->odp_port = node ? node->data : OVSP_NONE;
976 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
978 backer->need_revalidate = REV_RECONFIGURE;
983 SIMAP_FOR_EACH (node, &tmp_backers) {
984 dpif_port_del(backer->dpif, node->data);
986 simap_destroy(&tmp_backers);
988 switch (backer->need_revalidate) {
989 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
990 case REV_STP: COVERAGE_INC(rev_stp); break;
991 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
992 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
993 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
996 if (backer->need_revalidate) {
997 /* Clear the drop_keys in case we should now be accepting some
998 * formerly dropped flows. */
999 drop_key_clear(backer);
1002 /* Clear the revalidation flags. */
1003 tag_set_init(&backer->revalidate_set);
1004 backer->need_revalidate = 0;
1006 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1007 struct facet *facet, *next;
1009 if (ofproto->backer != backer) {
1013 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1015 || tag_set_intersects(&revalidate_set, facet->tags)) {
1016 facet_revalidate(facet);
1023 if (timer_expired(&backer->next_expiration)) {
1024 int delay = expire(backer);
1025 timer_set_duration(&backer->next_expiration, delay);
1028 /* Check for port changes in the dpif. */
1029 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1030 struct ofproto_dpif *ofproto;
1031 struct dpif_port port;
1033 /* Don't report on the datapath's device. */
1034 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1038 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1039 &all_ofproto_dpifs) {
1040 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1045 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1046 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1047 /* The port was removed. If we know the datapath,
1048 * report it through poll_set(). If we don't, it may be
1049 * notifying us of a removal we initiated, so ignore it.
1050 * If there's a pending ENOBUFS, let it stand, since
1051 * everything will be reevaluated. */
1052 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1053 sset_add(&ofproto->port_poll_set, devname);
1054 ofproto->port_poll_errno = 0;
1056 } else if (!ofproto) {
1057 /* The port was added, but we don't know with which
1058 * ofproto we should associate it. Delete it. */
1059 dpif_port_del(backer->dpif, port.port_no);
1061 dpif_port_destroy(&port);
1067 if (error != EAGAIN) {
1068 struct ofproto_dpif *ofproto;
1070 /* There was some sort of error, so propagate it to all
1071 * ofprotos that use this backer. */
1072 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1073 &all_ofproto_dpifs) {
1074 if (ofproto->backer == backer) {
1075 sset_clear(&ofproto->port_poll_set);
1076 ofproto->port_poll_errno = error;
1085 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1089 /* Handle one or more batches of upcalls, until there's nothing left to do
1090 * or until we do a fixed total amount of work.
1092 * We do work in batches because it can be much cheaper to set up a number
1093 * of flows and fire off their patches all at once. We do multiple batches
1094 * because in some cases handling a packet can cause another packet to be
1095 * queued almost immediately as part of the return flow. Both
1096 * optimizations can make major improvements on some benchmarks and
1097 * presumably for real traffic as well. */
1099 while (work < max_batch) {
1100 int retval = handle_upcalls(backer, max_batch - work);
1111 type_run_fast(const char *type)
1113 struct dpif_backer *backer;
1115 backer = shash_find_data(&all_dpif_backers, type);
1117 /* This is not necessarily a problem, since backers are only
1118 * created on demand. */
1122 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1128 static long long int port_rl = LLONG_MIN;
1129 static unsigned int backer_rl = 0;
1131 if (time_msec() >= port_rl) {
1132 struct ofproto_dpif *ofproto;
1133 struct ofport_dpif *ofport;
1135 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1137 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1138 port_run_fast(ofport);
1141 port_rl = time_msec() + 200;
1144 /* XXX: We have to be careful not to do too much work in this function. If
1145 * we call dpif_backer_run_fast() too often, or with too large a batch,
1146 * performance improves signifcantly, but at a cost. It's possible for the
1147 * number of flows in the datapath to increase without bound, and for poll
1148 * loops to take 10s of seconds. The correct solution to this problem,
1149 * long term, is to separate flow miss handling into it's own thread so it
1150 * isn't affected by revalidations, and expirations. Until then, this is
1151 * the best we can do. */
1152 if (++backer_rl >= 10) {
1153 struct shash_node *node;
1156 SHASH_FOR_EACH (node, &all_dpif_backers) {
1157 dpif_backer_run_fast(node->data, 1);
1163 type_wait(const char *type)
1165 struct dpif_backer *backer;
1167 backer = shash_find_data(&all_dpif_backers, type);
1169 /* This is not necessarily a problem, since backers are only
1170 * created on demand. */
1174 timer_wait(&backer->next_expiration);
1177 /* Basic life-cycle. */
1179 static int add_internal_flows(struct ofproto_dpif *);
1181 static struct ofproto *
1184 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1185 return &ofproto->up;
1189 dealloc(struct ofproto *ofproto_)
1191 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1196 close_dpif_backer(struct dpif_backer *backer)
1198 struct shash_node *node;
1200 ovs_assert(backer->refcount > 0);
1202 if (--backer->refcount) {
1206 drop_key_clear(backer);
1207 hmap_destroy(&backer->drop_keys);
1209 simap_destroy(&backer->tnl_backers);
1210 hmap_destroy(&backer->odp_to_ofport_map);
1211 node = shash_find(&all_dpif_backers, backer->type);
1213 shash_delete(&all_dpif_backers, node);
1214 dpif_close(backer->dpif);
1219 /* Datapath port slated for removal from datapath. */
1220 struct odp_garbage {
1221 struct list list_node;
1226 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1228 struct dpif_backer *backer;
1229 struct dpif_port_dump port_dump;
1230 struct dpif_port port;
1231 struct shash_node *node;
1232 struct list garbage_list;
1233 struct odp_garbage *garbage, *next;
1239 backer = shash_find_data(&all_dpif_backers, type);
1246 backer_name = xasprintf("ovs-%s", type);
1248 /* Remove any existing datapaths, since we assume we're the only
1249 * userspace controlling the datapath. */
1251 dp_enumerate_names(type, &names);
1252 SSET_FOR_EACH(name, &names) {
1253 struct dpif *old_dpif;
1255 /* Don't remove our backer if it exists. */
1256 if (!strcmp(name, backer_name)) {
1260 if (dpif_open(name, type, &old_dpif)) {
1261 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1263 dpif_delete(old_dpif);
1264 dpif_close(old_dpif);
1267 sset_destroy(&names);
1269 backer = xmalloc(sizeof *backer);
1271 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1274 VLOG_ERR("failed to open datapath of type %s: %s", type,
1280 backer->type = xstrdup(type);
1281 backer->refcount = 1;
1282 hmap_init(&backer->odp_to_ofport_map);
1283 hmap_init(&backer->drop_keys);
1284 timer_set_duration(&backer->next_expiration, 1000);
1285 backer->need_revalidate = 0;
1286 simap_init(&backer->tnl_backers);
1287 tag_set_init(&backer->revalidate_set);
1290 dpif_flow_flush(backer->dpif);
1292 /* Loop through the ports already on the datapath and remove any
1293 * that we don't need anymore. */
1294 list_init(&garbage_list);
1295 dpif_port_dump_start(&port_dump, backer->dpif);
1296 while (dpif_port_dump_next(&port_dump, &port)) {
1297 node = shash_find(&init_ofp_ports, port.name);
1298 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1299 garbage = xmalloc(sizeof *garbage);
1300 garbage->odp_port = port.port_no;
1301 list_push_front(&garbage_list, &garbage->list_node);
1304 dpif_port_dump_done(&port_dump);
1306 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1307 dpif_port_del(backer->dpif, garbage->odp_port);
1308 list_remove(&garbage->list_node);
1312 shash_add(&all_dpif_backers, type, backer);
1314 error = dpif_recv_set(backer->dpif, true);
1316 VLOG_ERR("failed to listen on datapath of type %s: %s",
1317 type, strerror(error));
1318 close_dpif_backer(backer);
1326 construct(struct ofproto *ofproto_)
1328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1329 struct shash_node *node, *next;
1334 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1339 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1340 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1342 ofproto->n_matches = 0;
1344 ofproto->netflow = NULL;
1345 ofproto->sflow = NULL;
1346 ofproto->stp = NULL;
1347 hmap_init(&ofproto->bundles);
1348 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1349 for (i = 0; i < MAX_MIRRORS; i++) {
1350 ofproto->mirrors[i] = NULL;
1352 ofproto->has_bonded_bundles = false;
1354 hmap_init(&ofproto->facets);
1355 hmap_init(&ofproto->subfacets);
1356 ofproto->governor = NULL;
1357 ofproto->consistency_rl = LLONG_MIN;
1359 for (i = 0; i < N_TABLES; i++) {
1360 struct table_dpif *table = &ofproto->tables[i];
1362 table->catchall_table = NULL;
1363 table->other_table = NULL;
1364 table->basis = random_uint32();
1367 list_init(&ofproto->completions);
1369 ofproto_dpif_unixctl_init();
1371 ofproto->has_mirrors = false;
1372 ofproto->has_bundle_action = false;
1374 hmap_init(&ofproto->vlandev_map);
1375 hmap_init(&ofproto->realdev_vid_map);
1377 sset_init(&ofproto->ports);
1378 sset_init(&ofproto->ghost_ports);
1379 sset_init(&ofproto->port_poll_set);
1380 ofproto->port_poll_errno = 0;
1382 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1383 struct iface_hint *iface_hint = node->data;
1385 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1386 /* Check if the datapath already has this port. */
1387 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1388 sset_add(&ofproto->ports, node->name);
1391 free(iface_hint->br_name);
1392 free(iface_hint->br_type);
1394 shash_delete(&init_ofp_ports, node);
1398 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1399 hash_string(ofproto->up.name, 0));
1400 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1402 ofproto_init_tables(ofproto_, N_TABLES);
1403 error = add_internal_flows(ofproto);
1404 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1407 ofproto->n_missed = 0;
1409 ofproto->max_n_subfacet = 0;
1410 ofproto->created = time_msec();
1411 ofproto->last_minute = ofproto->created;
1412 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1413 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1414 ofproto->subfacet_add_count = 0;
1415 ofproto->subfacet_del_count = 0;
1416 ofproto->total_subfacet_add_count = 0;
1417 ofproto->total_subfacet_del_count = 0;
1418 ofproto->total_subfacet_life_span = 0;
1419 ofproto->total_subfacet_count = 0;
1420 ofproto->n_update_stats = 0;
1426 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1427 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1429 struct ofputil_flow_mod fm;
1432 match_init_catchall(&fm.match);
1434 match_set_reg(&fm.match, 0, id);
1435 fm.new_cookie = htonll(0);
1436 fm.cookie = htonll(0);
1437 fm.cookie_mask = htonll(0);
1438 fm.table_id = TBL_INTERNAL;
1439 fm.command = OFPFC_ADD;
1440 fm.idle_timeout = 0;
1441 fm.hard_timeout = 0;
1445 fm.ofpacts = ofpacts->data;
1446 fm.ofpacts_len = ofpacts->size;
1448 error = ofproto_flow_mod(&ofproto->up, &fm);
1450 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1451 id, ofperr_to_string(error));
1455 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1456 ovs_assert(*rulep != NULL);
1462 add_internal_flows(struct ofproto_dpif *ofproto)
1464 struct ofpact_controller *controller;
1465 uint64_t ofpacts_stub[128 / 8];
1466 struct ofpbuf ofpacts;
1470 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1473 controller = ofpact_put_CONTROLLER(&ofpacts);
1474 controller->max_len = UINT16_MAX;
1475 controller->controller_id = 0;
1476 controller->reason = OFPR_NO_MATCH;
1477 ofpact_pad(&ofpacts);
1479 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1484 ofpbuf_clear(&ofpacts);
1485 error = add_internal_flow(ofproto, id++, &ofpacts,
1486 &ofproto->no_packet_in_rule);
1491 complete_operations(struct ofproto_dpif *ofproto)
1493 struct dpif_completion *c, *next;
1495 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1496 ofoperation_complete(c->op, 0);
1497 list_remove(&c->list_node);
1503 destruct(struct ofproto *ofproto_)
1505 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1506 struct rule_dpif *rule, *next_rule;
1507 struct oftable *table;
1510 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1511 complete_operations(ofproto);
1513 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1514 struct cls_cursor cursor;
1516 cls_cursor_init(&cursor, &table->cls, NULL);
1517 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1518 ofproto_rule_destroy(&rule->up);
1522 for (i = 0; i < MAX_MIRRORS; i++) {
1523 mirror_destroy(ofproto->mirrors[i]);
1526 netflow_destroy(ofproto->netflow);
1527 dpif_sflow_destroy(ofproto->sflow);
1528 hmap_destroy(&ofproto->bundles);
1529 mac_learning_destroy(ofproto->ml);
1531 hmap_destroy(&ofproto->facets);
1532 hmap_destroy(&ofproto->subfacets);
1533 governor_destroy(ofproto->governor);
1535 hmap_destroy(&ofproto->vlandev_map);
1536 hmap_destroy(&ofproto->realdev_vid_map);
1538 sset_destroy(&ofproto->ports);
1539 sset_destroy(&ofproto->ghost_ports);
1540 sset_destroy(&ofproto->port_poll_set);
1542 close_dpif_backer(ofproto->backer);
1546 run_fast(struct ofproto *ofproto_)
1548 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1549 struct ofport_dpif *ofport;
1551 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1552 port_run_fast(ofport);
1559 run(struct ofproto *ofproto_)
1561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562 struct ofport_dpif *ofport;
1563 struct ofbundle *bundle;
1567 complete_operations(ofproto);
1570 error = run_fast(ofproto_);
1575 if (ofproto->netflow) {
1576 if (netflow_run(ofproto->netflow)) {
1577 send_netflow_active_timeouts(ofproto);
1580 if (ofproto->sflow) {
1581 dpif_sflow_run(ofproto->sflow);
1584 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1587 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1592 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1594 /* Check the consistency of a random facet, to aid debugging. */
1595 if (time_msec() >= ofproto->consistency_rl
1596 && !hmap_is_empty(&ofproto->facets)
1597 && !ofproto->backer->need_revalidate) {
1598 struct facet *facet;
1600 ofproto->consistency_rl = time_msec() + 250;
1602 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1603 struct facet, hmap_node);
1604 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1606 if (!facet_check_consistency(facet)) {
1607 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1612 if (ofproto->governor) {
1615 governor_run(ofproto->governor);
1617 /* If the governor has shrunk to its minimum size and the number of
1618 * subfacets has dwindled, then drop the governor entirely.
1620 * For hysteresis, the number of subfacets to drop the governor is
1621 * smaller than the number needed to trigger its creation. */
1622 n_subfacets = hmap_count(&ofproto->subfacets);
1623 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1624 && governor_is_idle(ofproto->governor)) {
1625 governor_destroy(ofproto->governor);
1626 ofproto->governor = NULL;
1634 wait(struct ofproto *ofproto_)
1636 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1637 struct ofport_dpif *ofport;
1638 struct ofbundle *bundle;
1640 if (!clogged && !list_is_empty(&ofproto->completions)) {
1641 poll_immediate_wake();
1644 dpif_wait(ofproto->backer->dpif);
1645 dpif_recv_wait(ofproto->backer->dpif);
1646 if (ofproto->sflow) {
1647 dpif_sflow_wait(ofproto->sflow);
1649 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1650 poll_immediate_wake();
1652 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1655 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1656 bundle_wait(bundle);
1658 if (ofproto->netflow) {
1659 netflow_wait(ofproto->netflow);
1661 mac_learning_wait(ofproto->ml);
1663 if (ofproto->backer->need_revalidate) {
1664 /* Shouldn't happen, but if it does just go around again. */
1665 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1666 poll_immediate_wake();
1668 if (ofproto->governor) {
1669 governor_wait(ofproto->governor);
1674 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1676 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1678 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1679 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1683 flush(struct ofproto *ofproto_)
1685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1686 struct subfacet *subfacet, *next_subfacet;
1687 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1691 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1692 &ofproto->subfacets) {
1693 if (subfacet->path != SF_NOT_INSTALLED) {
1694 batch[n_batch++] = subfacet;
1695 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1696 subfacet_destroy_batch(ofproto, batch, n_batch);
1700 subfacet_destroy(subfacet);
1705 subfacet_destroy_batch(ofproto, batch, n_batch);
1710 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1711 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1713 *arp_match_ip = true;
1714 *actions = (OFPUTIL_A_OUTPUT |
1715 OFPUTIL_A_SET_VLAN_VID |
1716 OFPUTIL_A_SET_VLAN_PCP |
1717 OFPUTIL_A_STRIP_VLAN |
1718 OFPUTIL_A_SET_DL_SRC |
1719 OFPUTIL_A_SET_DL_DST |
1720 OFPUTIL_A_SET_NW_SRC |
1721 OFPUTIL_A_SET_NW_DST |
1722 OFPUTIL_A_SET_NW_TOS |
1723 OFPUTIL_A_SET_TP_SRC |
1724 OFPUTIL_A_SET_TP_DST |
1729 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1732 struct dpif_dp_stats s;
1734 strcpy(ots->name, "classifier");
1736 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1738 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1739 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1742 static struct ofport *
1745 struct ofport_dpif *port = xmalloc(sizeof *port);
1750 port_dealloc(struct ofport *port_)
1752 struct ofport_dpif *port = ofport_dpif_cast(port_);
1757 port_construct(struct ofport *port_)
1759 struct ofport_dpif *port = ofport_dpif_cast(port_);
1760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1761 const struct netdev *netdev = port->up.netdev;
1762 struct dpif_port dpif_port;
1765 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1766 port->bundle = NULL;
1768 port->tag = tag_create_random();
1769 port->may_enable = true;
1770 port->stp_port = NULL;
1771 port->stp_state = STP_DISABLED;
1772 port->tnl_port = NULL;
1773 hmap_init(&port->priorities);
1774 port->realdev_ofp_port = 0;
1775 port->vlandev_vid = 0;
1776 port->carrier_seq = netdev_get_carrier_resets(netdev);
1778 if (netdev_vport_is_patch(netdev)) {
1779 /* XXX By bailing out here, we don't do required sFlow work. */
1780 port->odp_port = OVSP_NONE;
1784 error = dpif_port_query_by_name(ofproto->backer->dpif,
1785 netdev_vport_get_dpif_port(netdev),
1791 port->odp_port = dpif_port.port_no;
1793 if (netdev_get_tunnel_config(netdev)) {
1794 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1796 /* Sanity-check that a mapping doesn't already exist. This
1797 * shouldn't happen for non-tunnel ports. */
1798 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1799 VLOG_ERR("port %s already has an OpenFlow port number",
1801 dpif_port_destroy(&dpif_port);
1805 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1806 hash_int(port->odp_port, 0));
1808 dpif_port_destroy(&dpif_port);
1810 if (ofproto->sflow) {
1811 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1818 port_destruct(struct ofport *port_)
1820 struct ofport_dpif *port = ofport_dpif_cast(port_);
1821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1822 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1823 const char *devname = netdev_get_name(port->up.netdev);
1825 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1826 /* The underlying device is still there, so delete it. This
1827 * happens when the ofproto is being destroyed, since the caller
1828 * assumes that removal of attached ports will happen as part of
1830 if (!port->tnl_port) {
1831 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1833 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1837 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1840 tnl_port_del(port->tnl_port);
1841 sset_find_and_delete(&ofproto->ports, devname);
1842 sset_find_and_delete(&ofproto->ghost_ports, devname);
1843 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1844 bundle_remove(port_);
1845 set_cfm(port_, NULL);
1846 if (ofproto->sflow) {
1847 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1850 ofport_clear_priorities(port);
1851 hmap_destroy(&port->priorities);
1855 port_modified(struct ofport *port_)
1857 struct ofport_dpif *port = ofport_dpif_cast(port_);
1859 if (port->bundle && port->bundle->bond) {
1860 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1865 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1867 struct ofport_dpif *port = ofport_dpif_cast(port_);
1868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1869 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1871 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1872 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1873 OFPUTIL_PC_NO_PACKET_IN)) {
1874 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1876 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1877 bundle_update(port->bundle);
1883 set_sflow(struct ofproto *ofproto_,
1884 const struct ofproto_sflow_options *sflow_options)
1886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1887 struct dpif_sflow *ds = ofproto->sflow;
1889 if (sflow_options) {
1891 struct ofport_dpif *ofport;
1893 ds = ofproto->sflow = dpif_sflow_create();
1894 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1895 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1897 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1899 dpif_sflow_set_options(ds, sflow_options);
1902 dpif_sflow_destroy(ds);
1903 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1904 ofproto->sflow = NULL;
1911 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1913 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1920 struct ofproto_dpif *ofproto;
1922 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1923 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1924 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1927 if (cfm_configure(ofport->cfm, s)) {
1933 cfm_destroy(ofport->cfm);
1939 get_cfm_fault(const struct ofport *ofport_)
1941 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1943 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1947 get_cfm_opup(const struct ofport *ofport_)
1949 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1951 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1955 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1958 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1969 get_cfm_health(const struct ofport *ofport_)
1971 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1973 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1976 /* Spanning Tree. */
1979 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1981 struct ofproto_dpif *ofproto = ofproto_;
1982 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1983 struct ofport_dpif *ofport;
1985 ofport = stp_port_get_aux(sp);
1987 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1988 ofproto->up.name, port_num);
1990 struct eth_header *eth = pkt->l2;
1992 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1993 if (eth_addr_is_zero(eth->eth_src)) {
1994 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1995 "with unknown MAC", ofproto->up.name, port_num);
1997 send_packet(ofport, pkt);
2003 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2005 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2007 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2009 /* Only revalidate flows if the configuration changed. */
2010 if (!s != !ofproto->stp) {
2011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2015 if (!ofproto->stp) {
2016 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2017 send_bpdu_cb, ofproto);
2018 ofproto->stp_last_tick = time_msec();
2021 stp_set_bridge_id(ofproto->stp, s->system_id);
2022 stp_set_bridge_priority(ofproto->stp, s->priority);
2023 stp_set_hello_time(ofproto->stp, s->hello_time);
2024 stp_set_max_age(ofproto->stp, s->max_age);
2025 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2027 struct ofport *ofport;
2029 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2030 set_stp_port(ofport, NULL);
2033 stp_destroy(ofproto->stp);
2034 ofproto->stp = NULL;
2041 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2047 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2048 s->designated_root = stp_get_designated_root(ofproto->stp);
2049 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2058 update_stp_port_state(struct ofport_dpif *ofport)
2060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2061 enum stp_state state;
2063 /* Figure out new state. */
2064 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2068 if (ofport->stp_state != state) {
2069 enum ofputil_port_state of_state;
2072 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2073 netdev_get_name(ofport->up.netdev),
2074 stp_state_name(ofport->stp_state),
2075 stp_state_name(state));
2076 if (stp_learn_in_state(ofport->stp_state)
2077 != stp_learn_in_state(state)) {
2078 /* xxx Learning action flows should also be flushed. */
2079 mac_learning_flush(ofproto->ml,
2080 &ofproto->backer->revalidate_set);
2082 fwd_change = stp_forward_in_state(ofport->stp_state)
2083 != stp_forward_in_state(state);
2085 ofproto->backer->need_revalidate = REV_STP;
2086 ofport->stp_state = state;
2087 ofport->stp_state_entered = time_msec();
2089 if (fwd_change && ofport->bundle) {
2090 bundle_update(ofport->bundle);
2093 /* Update the STP state bits in the OpenFlow port description. */
2094 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2095 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2096 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2097 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2098 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2100 ofproto_port_set_state(&ofport->up, of_state);
2104 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2105 * caller is responsible for assigning STP port numbers and ensuring
2106 * there are no duplicates. */
2108 set_stp_port(struct ofport *ofport_,
2109 const struct ofproto_port_stp_settings *s)
2111 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2112 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2113 struct stp_port *sp = ofport->stp_port;
2115 if (!s || !s->enable) {
2117 ofport->stp_port = NULL;
2118 stp_port_disable(sp);
2119 update_stp_port_state(ofport);
2122 } else if (sp && stp_port_no(sp) != s->port_num
2123 && ofport == stp_port_get_aux(sp)) {
2124 /* The port-id changed, so disable the old one if it's not
2125 * already in use by another port. */
2126 stp_port_disable(sp);
2129 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2130 stp_port_enable(sp);
2132 stp_port_set_aux(sp, ofport);
2133 stp_port_set_priority(sp, s->priority);
2134 stp_port_set_path_cost(sp, s->path_cost);
2136 update_stp_port_state(ofport);
2142 get_stp_port_status(struct ofport *ofport_,
2143 struct ofproto_port_stp_status *s)
2145 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2146 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2147 struct stp_port *sp = ofport->stp_port;
2149 if (!ofproto->stp || !sp) {
2155 s->port_id = stp_port_get_id(sp);
2156 s->state = stp_port_get_state(sp);
2157 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2158 s->role = stp_port_get_role(sp);
2159 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2165 stp_run(struct ofproto_dpif *ofproto)
2168 long long int now = time_msec();
2169 long long int elapsed = now - ofproto->stp_last_tick;
2170 struct stp_port *sp;
2173 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2174 ofproto->stp_last_tick = now;
2176 while (stp_get_changed_port(ofproto->stp, &sp)) {
2177 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2180 update_stp_port_state(ofport);
2184 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2185 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2191 stp_wait(struct ofproto_dpif *ofproto)
2194 poll_timer_wait(1000);
2198 /* Returns true if STP should process 'flow'. */
2200 stp_should_process_flow(const struct flow *flow)
2202 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2206 stp_process_packet(const struct ofport_dpif *ofport,
2207 const struct ofpbuf *packet)
2209 struct ofpbuf payload = *packet;
2210 struct eth_header *eth = payload.data;
2211 struct stp_port *sp = ofport->stp_port;
2213 /* Sink packets on ports that have STP disabled when the bridge has
2215 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2219 /* Trim off padding on payload. */
2220 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2221 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2224 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2225 stp_received_bpdu(sp, payload.data, payload.size);
2229 static struct priority_to_dscp *
2230 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2232 struct priority_to_dscp *pdscp;
2235 hash = hash_int(priority, 0);
2236 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2237 if (pdscp->priority == priority) {
2245 ofport_clear_priorities(struct ofport_dpif *ofport)
2247 struct priority_to_dscp *pdscp, *next;
2249 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2250 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2256 set_queues(struct ofport *ofport_,
2257 const struct ofproto_port_queue *qdscp_list,
2260 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2262 struct hmap new = HMAP_INITIALIZER(&new);
2265 for (i = 0; i < n_qdscp; i++) {
2266 struct priority_to_dscp *pdscp;
2270 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2271 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2276 pdscp = get_priority(ofport, priority);
2278 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2280 pdscp = xmalloc(sizeof *pdscp);
2281 pdscp->priority = priority;
2283 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2286 if (pdscp->dscp != dscp) {
2288 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2291 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2294 if (!hmap_is_empty(&ofport->priorities)) {
2295 ofport_clear_priorities(ofport);
2296 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2299 hmap_swap(&new, &ofport->priorities);
2307 /* Expires all MAC learning entries associated with 'bundle' and forces its
2308 * ofproto to revalidate every flow.
2310 * Normally MAC learning entries are removed only from the ofproto associated
2311 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2312 * are removed from every ofproto. When patch ports and SLB bonds are in use
2313 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2314 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2315 * with the host from which it migrated. */
2317 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2319 struct ofproto_dpif *ofproto = bundle->ofproto;
2320 struct mac_learning *ml = ofproto->ml;
2321 struct mac_entry *mac, *next_mac;
2323 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2324 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2325 if (mac->port.p == bundle) {
2327 struct ofproto_dpif *o;
2329 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2331 struct mac_entry *e;
2333 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2336 mac_learning_expire(o->ml, e);
2342 mac_learning_expire(ml, mac);
2347 static struct ofbundle *
2348 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2350 struct ofbundle *bundle;
2352 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2353 &ofproto->bundles) {
2354 if (bundle->aux == aux) {
2361 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2362 * ones that are found to 'bundles'. */
2364 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2365 void **auxes, size_t n_auxes,
2366 struct hmapx *bundles)
2370 hmapx_init(bundles);
2371 for (i = 0; i < n_auxes; i++) {
2372 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2374 hmapx_add(bundles, bundle);
2380 bundle_update(struct ofbundle *bundle)
2382 struct ofport_dpif *port;
2384 bundle->floodable = true;
2385 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2386 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2387 || !stp_forward_in_state(port->stp_state)) {
2388 bundle->floodable = false;
2395 bundle_del_port(struct ofport_dpif *port)
2397 struct ofbundle *bundle = port->bundle;
2399 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2401 list_remove(&port->bundle_node);
2402 port->bundle = NULL;
2405 lacp_slave_unregister(bundle->lacp, port);
2408 bond_slave_unregister(bundle->bond, port);
2411 bundle_update(bundle);
2415 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2416 struct lacp_slave_settings *lacp,
2417 uint32_t bond_stable_id)
2419 struct ofport_dpif *port;
2421 port = get_ofp_port(bundle->ofproto, ofp_port);
2426 if (port->bundle != bundle) {
2427 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2429 bundle_del_port(port);
2432 port->bundle = bundle;
2433 list_push_back(&bundle->ports, &port->bundle_node);
2434 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2435 || !stp_forward_in_state(port->stp_state)) {
2436 bundle->floodable = false;
2440 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2441 lacp_slave_register(bundle->lacp, port, lacp);
2444 port->bond_stable_id = bond_stable_id;
2450 bundle_destroy(struct ofbundle *bundle)
2452 struct ofproto_dpif *ofproto;
2453 struct ofport_dpif *port, *next_port;
2460 ofproto = bundle->ofproto;
2461 for (i = 0; i < MAX_MIRRORS; i++) {
2462 struct ofmirror *m = ofproto->mirrors[i];
2464 if (m->out == bundle) {
2466 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2467 || hmapx_find_and_delete(&m->dsts, bundle)) {
2468 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2473 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2474 bundle_del_port(port);
2477 bundle_flush_macs(bundle, true);
2478 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2480 free(bundle->trunks);
2481 lacp_destroy(bundle->lacp);
2482 bond_destroy(bundle->bond);
2487 bundle_set(struct ofproto *ofproto_, void *aux,
2488 const struct ofproto_bundle_settings *s)
2490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2491 bool need_flush = false;
2492 struct ofport_dpif *port;
2493 struct ofbundle *bundle;
2494 unsigned long *trunks;
2500 bundle_destroy(bundle_lookup(ofproto, aux));
2504 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2505 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2507 bundle = bundle_lookup(ofproto, aux);
2509 bundle = xmalloc(sizeof *bundle);
2511 bundle->ofproto = ofproto;
2512 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2513 hash_pointer(aux, 0));
2515 bundle->name = NULL;
2517 list_init(&bundle->ports);
2518 bundle->vlan_mode = PORT_VLAN_TRUNK;
2520 bundle->trunks = NULL;
2521 bundle->use_priority_tags = s->use_priority_tags;
2522 bundle->lacp = NULL;
2523 bundle->bond = NULL;
2525 bundle->floodable = true;
2527 bundle->src_mirrors = 0;
2528 bundle->dst_mirrors = 0;
2529 bundle->mirror_out = 0;
2532 if (!bundle->name || strcmp(s->name, bundle->name)) {
2534 bundle->name = xstrdup(s->name);
2539 if (!bundle->lacp) {
2540 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2541 bundle->lacp = lacp_create();
2543 lacp_configure(bundle->lacp, s->lacp);
2545 lacp_destroy(bundle->lacp);
2546 bundle->lacp = NULL;
2549 /* Update set of ports. */
2551 for (i = 0; i < s->n_slaves; i++) {
2552 if (!bundle_add_port(bundle, s->slaves[i],
2553 s->lacp ? &s->lacp_slaves[i] : NULL,
2554 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2558 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2559 struct ofport_dpif *next_port;
2561 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2562 for (i = 0; i < s->n_slaves; i++) {
2563 if (s->slaves[i] == port->up.ofp_port) {
2568 bundle_del_port(port);
2572 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2574 if (list_is_empty(&bundle->ports)) {
2575 bundle_destroy(bundle);
2579 /* Set VLAN tagging mode */
2580 if (s->vlan_mode != bundle->vlan_mode
2581 || s->use_priority_tags != bundle->use_priority_tags) {
2582 bundle->vlan_mode = s->vlan_mode;
2583 bundle->use_priority_tags = s->use_priority_tags;
2588 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2589 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2591 if (vlan != bundle->vlan) {
2592 bundle->vlan = vlan;
2596 /* Get trunked VLANs. */
2597 switch (s->vlan_mode) {
2598 case PORT_VLAN_ACCESS:
2602 case PORT_VLAN_TRUNK:
2603 trunks = CONST_CAST(unsigned long *, s->trunks);
2606 case PORT_VLAN_NATIVE_UNTAGGED:
2607 case PORT_VLAN_NATIVE_TAGGED:
2608 if (vlan != 0 && (!s->trunks
2609 || !bitmap_is_set(s->trunks, vlan)
2610 || bitmap_is_set(s->trunks, 0))) {
2611 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2613 trunks = bitmap_clone(s->trunks, 4096);
2615 trunks = bitmap_allocate1(4096);
2617 bitmap_set1(trunks, vlan);
2618 bitmap_set0(trunks, 0);
2620 trunks = CONST_CAST(unsigned long *, s->trunks);
2627 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2628 free(bundle->trunks);
2629 if (trunks == s->trunks) {
2630 bundle->trunks = vlan_bitmap_clone(trunks);
2632 bundle->trunks = trunks;
2637 if (trunks != s->trunks) {
2642 if (!list_is_short(&bundle->ports)) {
2643 bundle->ofproto->has_bonded_bundles = true;
2645 if (bond_reconfigure(bundle->bond, s->bond)) {
2646 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2649 bundle->bond = bond_create(s->bond);
2650 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2653 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2654 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2658 bond_destroy(bundle->bond);
2659 bundle->bond = NULL;
2662 /* If we changed something that would affect MAC learning, un-learn
2663 * everything on this port and force flow revalidation. */
2665 bundle_flush_macs(bundle, false);
2672 bundle_remove(struct ofport *port_)
2674 struct ofport_dpif *port = ofport_dpif_cast(port_);
2675 struct ofbundle *bundle = port->bundle;
2678 bundle_del_port(port);
2679 if (list_is_empty(&bundle->ports)) {
2680 bundle_destroy(bundle);
2681 } else if (list_is_short(&bundle->ports)) {
2682 bond_destroy(bundle->bond);
2683 bundle->bond = NULL;
2689 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2691 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2692 struct ofport_dpif *port = port_;
2693 uint8_t ea[ETH_ADDR_LEN];
2696 error = netdev_get_etheraddr(port->up.netdev, ea);
2698 struct ofpbuf packet;
2701 ofpbuf_init(&packet, 0);
2702 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2704 memcpy(packet_pdu, pdu, pdu_size);
2706 send_packet(port, &packet);
2707 ofpbuf_uninit(&packet);
2709 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2710 "%s (%s)", port->bundle->name,
2711 netdev_get_name(port->up.netdev), strerror(error));
2716 bundle_send_learning_packets(struct ofbundle *bundle)
2718 struct ofproto_dpif *ofproto = bundle->ofproto;
2719 int error, n_packets, n_errors;
2720 struct mac_entry *e;
2722 error = n_packets = n_errors = 0;
2723 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2724 if (e->port.p != bundle) {
2725 struct ofpbuf *learning_packet;
2726 struct ofport_dpif *port;
2730 /* The assignment to "port" is unnecessary but makes "grep"ing for
2731 * struct ofport_dpif more effective. */
2732 learning_packet = bond_compose_learning_packet(bundle->bond,
2736 ret = send_packet(port, learning_packet);
2737 ofpbuf_delete(learning_packet);
2747 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2748 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2749 "packets, last error was: %s",
2750 bundle->name, n_errors, n_packets, strerror(error));
2752 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2753 bundle->name, n_packets);
2758 bundle_run(struct ofbundle *bundle)
2761 lacp_run(bundle->lacp, send_pdu_cb);
2764 struct ofport_dpif *port;
2766 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2767 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2770 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2771 lacp_status(bundle->lacp));
2772 if (bond_should_send_learning_packets(bundle->bond)) {
2773 bundle_send_learning_packets(bundle);
2779 bundle_wait(struct ofbundle *bundle)
2782 lacp_wait(bundle->lacp);
2785 bond_wait(bundle->bond);
2792 mirror_scan(struct ofproto_dpif *ofproto)
2796 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2797 if (!ofproto->mirrors[idx]) {
2804 static struct ofmirror *
2805 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2809 for (i = 0; i < MAX_MIRRORS; i++) {
2810 struct ofmirror *mirror = ofproto->mirrors[i];
2811 if (mirror && mirror->aux == aux) {
2819 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2821 mirror_update_dups(struct ofproto_dpif *ofproto)
2825 for (i = 0; i < MAX_MIRRORS; i++) {
2826 struct ofmirror *m = ofproto->mirrors[i];
2829 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2833 for (i = 0; i < MAX_MIRRORS; i++) {
2834 struct ofmirror *m1 = ofproto->mirrors[i];
2841 for (j = i + 1; j < MAX_MIRRORS; j++) {
2842 struct ofmirror *m2 = ofproto->mirrors[j];
2844 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2845 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2846 m2->dup_mirrors |= m1->dup_mirrors;
2853 mirror_set(struct ofproto *ofproto_, void *aux,
2854 const struct ofproto_mirror_settings *s)
2856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2857 mirror_mask_t mirror_bit;
2858 struct ofbundle *bundle;
2859 struct ofmirror *mirror;
2860 struct ofbundle *out;
2861 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2862 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2865 mirror = mirror_lookup(ofproto, aux);
2867 mirror_destroy(mirror);
2873 idx = mirror_scan(ofproto);
2875 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2877 ofproto->up.name, MAX_MIRRORS, s->name);
2881 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2882 mirror->ofproto = ofproto;
2885 mirror->out_vlan = -1;
2886 mirror->name = NULL;
2889 if (!mirror->name || strcmp(s->name, mirror->name)) {
2891 mirror->name = xstrdup(s->name);
2894 /* Get the new configuration. */
2895 if (s->out_bundle) {
2896 out = bundle_lookup(ofproto, s->out_bundle);
2898 mirror_destroy(mirror);
2904 out_vlan = s->out_vlan;
2906 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2907 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2909 /* If the configuration has not changed, do nothing. */
2910 if (hmapx_equals(&srcs, &mirror->srcs)
2911 && hmapx_equals(&dsts, &mirror->dsts)
2912 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2913 && mirror->out == out
2914 && mirror->out_vlan == out_vlan)
2916 hmapx_destroy(&srcs);
2917 hmapx_destroy(&dsts);
2921 hmapx_swap(&srcs, &mirror->srcs);
2922 hmapx_destroy(&srcs);
2924 hmapx_swap(&dsts, &mirror->dsts);
2925 hmapx_destroy(&dsts);
2927 free(mirror->vlans);
2928 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2931 mirror->out_vlan = out_vlan;
2933 /* Update bundles. */
2934 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2935 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2936 if (hmapx_contains(&mirror->srcs, bundle)) {
2937 bundle->src_mirrors |= mirror_bit;
2939 bundle->src_mirrors &= ~mirror_bit;
2942 if (hmapx_contains(&mirror->dsts, bundle)) {
2943 bundle->dst_mirrors |= mirror_bit;
2945 bundle->dst_mirrors &= ~mirror_bit;
2948 if (mirror->out == bundle) {
2949 bundle->mirror_out |= mirror_bit;
2951 bundle->mirror_out &= ~mirror_bit;
2955 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2956 ofproto->has_mirrors = true;
2957 mac_learning_flush(ofproto->ml,
2958 &ofproto->backer->revalidate_set);
2959 mirror_update_dups(ofproto);
2965 mirror_destroy(struct ofmirror *mirror)
2967 struct ofproto_dpif *ofproto;
2968 mirror_mask_t mirror_bit;
2969 struct ofbundle *bundle;
2976 ofproto = mirror->ofproto;
2977 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2978 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2980 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2981 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2982 bundle->src_mirrors &= ~mirror_bit;
2983 bundle->dst_mirrors &= ~mirror_bit;
2984 bundle->mirror_out &= ~mirror_bit;
2987 hmapx_destroy(&mirror->srcs);
2988 hmapx_destroy(&mirror->dsts);
2989 free(mirror->vlans);
2991 ofproto->mirrors[mirror->idx] = NULL;
2995 mirror_update_dups(ofproto);
2997 ofproto->has_mirrors = false;
2998 for (i = 0; i < MAX_MIRRORS; i++) {
2999 if (ofproto->mirrors[i]) {
3000 ofproto->has_mirrors = true;
3007 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3008 uint64_t *packets, uint64_t *bytes)
3010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3011 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3014 *packets = *bytes = UINT64_MAX;
3020 *packets = mirror->packet_count;
3021 *bytes = mirror->byte_count;
3027 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3029 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3031 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3037 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3040 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3041 return bundle && bundle->mirror_out != 0;
3045 forward_bpdu_changed(struct ofproto *ofproto_)
3047 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3048 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3052 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3056 mac_learning_set_idle_time(ofproto->ml, idle_time);
3057 mac_learning_set_max_entries(ofproto->ml, max_entries);
3062 static struct ofport_dpif *
3063 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3065 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3066 return ofport ? ofport_dpif_cast(ofport) : NULL;
3069 static struct ofport_dpif *
3070 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3072 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3073 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3077 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3078 struct ofproto_port *ofproto_port,
3079 struct dpif_port *dpif_port)
3081 ofproto_port->name = dpif_port->name;
3082 ofproto_port->type = dpif_port->type;
3083 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3086 static struct ofport_dpif *
3087 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3089 const struct ofproto_dpif *ofproto;
3092 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3097 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3098 struct ofport *ofport;
3100 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3101 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3102 return ofport_dpif_cast(ofport);
3109 port_run_fast(struct ofport_dpif *ofport)
3111 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3112 struct ofpbuf packet;
3114 ofpbuf_init(&packet, 0);
3115 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3116 send_packet(ofport, &packet);
3117 ofpbuf_uninit(&packet);
3122 port_run(struct ofport_dpif *ofport)
3124 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3125 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3126 bool enable = netdev_get_carrier(ofport->up.netdev);
3128 ofport->carrier_seq = carrier_seq;
3130 port_run_fast(ofport);
3132 if (ofport->tnl_port
3133 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3134 &ofport->tnl_port)) {
3135 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3139 int cfm_opup = cfm_get_opup(ofport->cfm);
3141 cfm_run(ofport->cfm);
3142 enable = enable && !cfm_get_fault(ofport->cfm);
3144 if (cfm_opup >= 0) {
3145 enable = enable && cfm_opup;
3149 if (ofport->bundle) {
3150 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3151 if (carrier_changed) {
3152 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3156 if (ofport->may_enable != enable) {
3157 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3159 if (ofproto->has_bundle_action) {
3160 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3164 ofport->may_enable = enable;
3168 port_wait(struct ofport_dpif *ofport)
3171 cfm_wait(ofport->cfm);
3176 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3177 struct ofproto_port *ofproto_port)
3179 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3180 struct dpif_port dpif_port;
3183 if (sset_contains(&ofproto->ghost_ports, devname)) {
3184 const char *type = netdev_get_type_from_name(devname);
3186 /* We may be called before ofproto->up.port_by_name is populated with
3187 * the appropriate ofport. For this reason, we must get the name and
3188 * type from the netdev layer directly. */
3190 const struct ofport *ofport;
3192 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3193 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3194 ofproto_port->name = xstrdup(devname);
3195 ofproto_port->type = xstrdup(type);
3201 if (!sset_contains(&ofproto->ports, devname)) {
3204 error = dpif_port_query_by_name(ofproto->backer->dpif,
3205 devname, &dpif_port);
3207 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3213 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3216 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3217 const char *devname = netdev_get_name(netdev);
3219 if (netdev_vport_is_patch(netdev)) {
3220 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3224 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3225 uint32_t port_no = UINT32_MAX;
3228 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3232 if (netdev_get_tunnel_config(netdev)) {
3233 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3237 if (netdev_get_tunnel_config(netdev)) {
3238 sset_add(&ofproto->ghost_ports, devname);
3240 sset_add(&ofproto->ports, devname);
3246 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3249 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3256 sset_find_and_delete(&ofproto->ghost_ports,
3257 netdev_get_name(ofport->up.netdev));
3258 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3259 if (!ofport->tnl_port) {
3260 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3262 /* The caller is going to close ofport->up.netdev. If this is a
3263 * bonded port, then the bond is using that netdev, so remove it
3264 * from the bond. The client will need to reconfigure everything
3265 * after deleting ports, so then the slave will get re-added. */
3266 bundle_remove(&ofport->up);
3273 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3275 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3280 error = netdev_get_stats(ofport->up.netdev, stats);
3282 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3283 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3285 /* ofproto->stats.tx_packets represents packets that we created
3286 * internally and sent to some port (e.g. packets sent with
3287 * send_packet()). Account for them as if they had come from
3288 * OFPP_LOCAL and got forwarded. */
3290 if (stats->rx_packets != UINT64_MAX) {
3291 stats->rx_packets += ofproto->stats.tx_packets;
3294 if (stats->rx_bytes != UINT64_MAX) {
3295 stats->rx_bytes += ofproto->stats.tx_bytes;
3298 /* ofproto->stats.rx_packets represents packets that were received on
3299 * some port and we processed internally and dropped (e.g. STP).
3300 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3302 if (stats->tx_packets != UINT64_MAX) {
3303 stats->tx_packets += ofproto->stats.rx_packets;
3306 if (stats->tx_bytes != UINT64_MAX) {
3307 stats->tx_bytes += ofproto->stats.rx_bytes;
3314 /* Account packets for LOCAL port. */
3316 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3317 size_t tx_size, size_t rx_size)
3319 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3322 ofproto->stats.rx_packets++;
3323 ofproto->stats.rx_bytes += rx_size;
3326 ofproto->stats.tx_packets++;
3327 ofproto->stats.tx_bytes += tx_size;
3331 struct port_dump_state {
3336 struct ofproto_port port;
3341 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3343 *statep = xzalloc(sizeof(struct port_dump_state));
3348 port_dump_next(const struct ofproto *ofproto_, void *state_,
3349 struct ofproto_port *port)
3351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3352 struct port_dump_state *state = state_;
3353 const struct sset *sset;
3354 struct sset_node *node;
3356 if (state->has_port) {
3357 ofproto_port_destroy(&state->port);
3358 state->has_port = false;
3360 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3361 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3364 error = port_query_by_name(ofproto_, node->name, &state->port);
3366 *port = state->port;
3367 state->has_port = true;
3369 } else if (error != ENODEV) {
3374 if (!state->ghost) {
3375 state->ghost = true;
3378 return port_dump_next(ofproto_, state_, port);
3385 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3387 struct port_dump_state *state = state_;
3389 if (state->has_port) {
3390 ofproto_port_destroy(&state->port);
3397 port_poll(const struct ofproto *ofproto_, char **devnamep)
3399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3401 if (ofproto->port_poll_errno) {
3402 int error = ofproto->port_poll_errno;
3403 ofproto->port_poll_errno = 0;
3407 if (sset_is_empty(&ofproto->port_poll_set)) {
3411 *devnamep = sset_pop(&ofproto->port_poll_set);
3416 port_poll_wait(const struct ofproto *ofproto_)
3418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3419 dpif_port_poll_wait(ofproto->backer->dpif);
3423 port_is_lacp_current(const struct ofport *ofport_)
3425 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3426 return (ofport->bundle && ofport->bundle->lacp
3427 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3431 /* Upcall handling. */
3433 /* Flow miss batching.
3435 * Some dpifs implement operations faster when you hand them off in a batch.
3436 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3437 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3438 * more packets, plus possibly installing the flow in the dpif.
3440 * So far we only batch the operations that affect flow setup time the most.
3441 * It's possible to batch more than that, but the benefit might be minimal. */
3443 struct hmap_node hmap_node;
3444 struct ofproto_dpif *ofproto;
3446 enum odp_key_fitness key_fitness;
3447 const struct nlattr *key;
3449 struct initial_vals initial_vals;
3450 struct list packets;
3451 enum dpif_upcall_type upcall_type;
3452 uint32_t odp_in_port;
3455 struct flow_miss_op {
3456 struct dpif_op dpif_op;
3457 void *garbage; /* Pointer to pass to free(), NULL if none. */
3458 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3461 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3462 * OpenFlow controller as necessary according to their individual
3463 * configurations. */
3465 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3466 const struct flow *flow)
3468 struct ofputil_packet_in pin;
3470 pin.packet = packet->data;
3471 pin.packet_len = packet->size;
3472 pin.reason = OFPR_NO_MATCH;
3473 pin.controller_id = 0;
3478 pin.send_len = 0; /* not used for flow table misses */
3480 flow_get_metadata(flow, &pin.fmd);
3482 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3485 static enum slow_path_reason
3486 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3487 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3491 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3493 cfm_process_heartbeat(ofport->cfm, packet);
3496 } else if (ofport->bundle && ofport->bundle->lacp
3497 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3499 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3502 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3504 stp_process_packet(ofport, packet);
3512 static struct flow_miss *
3513 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3514 const struct flow *flow, uint32_t hash)
3516 struct flow_miss *miss;
3518 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3519 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3527 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3528 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3529 * 'miss' is associated with a subfacet the caller must also initialize the
3530 * returned op->subfacet, and if anything needs to be freed after processing
3531 * the op, the caller must initialize op->garbage also. */
3533 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3534 struct flow_miss_op *op)
3536 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3537 /* This packet was received on a VLAN splinter port. We
3538 * added a VLAN to the packet to make the packet resemble
3539 * the flow, but the actions were composed assuming that
3540 * the packet contained no VLAN. So, we must remove the
3541 * VLAN header from the packet before trying to execute the
3543 eth_pop_vlan(packet);
3547 op->dpif_op.type = DPIF_OP_EXECUTE;
3548 op->dpif_op.u.execute.key = miss->key;
3549 op->dpif_op.u.execute.key_len = miss->key_len;
3550 op->dpif_op.u.execute.packet = packet;
3553 /* Helper for handle_flow_miss_without_facet() and
3554 * handle_flow_miss_with_facet(). */
3556 handle_flow_miss_common(struct rule_dpif *rule,
3557 struct ofpbuf *packet, const struct flow *flow)
3559 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3561 ofproto->n_matches++;
3563 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3565 * Extra-special case for fail-open mode.
3567 * We are in fail-open mode and the packet matched the fail-open
3568 * rule, but we are connected to a controller too. We should send
3569 * the packet up to the controller in the hope that it will try to
3570 * set up a flow and thereby allow us to exit fail-open.
3572 * See the top-level comment in fail-open.c for more information.
3574 send_packet_in_miss(ofproto, packet, flow);
3578 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3579 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3580 * installing a datapath flow. The answer is usually "yes" (a return value of
3581 * true). However, for short flows the cost of bookkeeping is much higher than
3582 * the benefits, so when the datapath holds a large number of flows we impose
3583 * some heuristics to decide which flows are likely to be worth tracking. */
3585 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3586 struct flow_miss *miss, uint32_t hash)
3588 if (!ofproto->governor) {
3591 n_subfacets = hmap_count(&ofproto->subfacets);
3592 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3596 ofproto->governor = governor_create(ofproto->up.name);
3599 return governor_should_install_flow(ofproto->governor, hash,
3600 list_size(&miss->packets));
3603 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3604 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3605 * increment '*n_ops'. */
3607 handle_flow_miss_without_facet(struct flow_miss *miss,
3608 struct rule_dpif *rule,
3609 struct flow_miss_op *ops, size_t *n_ops)
3611 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3612 long long int now = time_msec();
3613 struct action_xlate_ctx ctx;
3614 struct ofpbuf *packet;
3616 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3617 struct flow_miss_op *op = &ops[*n_ops];
3618 struct dpif_flow_stats stats;
3619 struct ofpbuf odp_actions;
3621 COVERAGE_INC(facet_suppress);
3623 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3625 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3626 rule_credit_stats(rule, &stats);
3628 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3629 &miss->initial_vals, rule, 0, packet);
3630 ctx.resubmit_stats = &stats;
3631 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3634 if (odp_actions.size) {
3635 struct dpif_execute *execute = &op->dpif_op.u.execute;
3637 init_flow_miss_execute_op(miss, packet, op);
3638 execute->actions = odp_actions.data;
3639 execute->actions_len = odp_actions.size;
3640 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3644 ofpbuf_uninit(&odp_actions);
3649 /* Handles 'miss', which matches 'facet'. May add any required datapath
3650 * operations to 'ops', incrementing '*n_ops' for each new op.
3652 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3653 * This is really important only for new facets: if we just called time_msec()
3654 * here, then the new subfacet or its packets could look (occasionally) as
3655 * though it was used some time after the facet was used. That can make a
3656 * one-packet flow look like it has a nonzero duration, which looks odd in
3657 * e.g. NetFlow statistics. */
3659 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3661 struct flow_miss_op *ops, size_t *n_ops)
3663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3664 enum subfacet_path want_path;
3665 struct subfacet *subfacet;
3666 struct ofpbuf *packet;
3668 subfacet = subfacet_create(facet, miss, now);
3670 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3671 struct flow_miss_op *op = &ops[*n_ops];
3672 struct dpif_flow_stats stats;
3673 struct ofpbuf odp_actions;
3675 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3677 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3678 if (!subfacet->actions || subfacet->slow) {
3679 subfacet_make_actions(subfacet, packet, &odp_actions);
3682 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3683 subfacet_update_stats(subfacet, &stats);
3685 if (subfacet->actions_len) {
3686 struct dpif_execute *execute = &op->dpif_op.u.execute;
3688 init_flow_miss_execute_op(miss, packet, op);
3689 if (!subfacet->slow) {
3690 execute->actions = subfacet->actions;
3691 execute->actions_len = subfacet->actions_len;
3692 ofpbuf_uninit(&odp_actions);
3694 execute->actions = odp_actions.data;
3695 execute->actions_len = odp_actions.size;
3696 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3701 ofpbuf_uninit(&odp_actions);
3705 want_path = subfacet_want_path(subfacet->slow);
3706 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3707 struct flow_miss_op *op = &ops[(*n_ops)++];
3708 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3710 subfacet->path = want_path;
3713 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3714 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3715 put->key = miss->key;
3716 put->key_len = miss->key_len;
3717 if (want_path == SF_FAST_PATH) {
3718 put->actions = subfacet->actions;
3719 put->actions_len = subfacet->actions_len;
3721 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3722 op->stub, sizeof op->stub,
3723 &put->actions, &put->actions_len);
3729 /* Handles flow miss 'miss'. May add any required datapath operations
3730 * to 'ops', incrementing '*n_ops' for each new op. */
3732 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3735 struct ofproto_dpif *ofproto = miss->ofproto;
3736 struct facet *facet;
3740 /* The caller must ensure that miss->hmap_node.hash contains
3741 * flow_hash(miss->flow, 0). */
3742 hash = miss->hmap_node.hash;
3744 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3746 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3748 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3749 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3753 facet = facet_create(rule, &miss->flow, hash);
3758 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3761 static struct drop_key *
3762 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3765 struct drop_key *drop_key;
3767 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3768 &backer->drop_keys) {
3769 if (drop_key->key_len == key_len
3770 && !memcmp(drop_key->key, key, key_len)) {
3778 drop_key_clear(struct dpif_backer *backer)
3780 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3781 struct drop_key *drop_key, *next;
3783 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3786 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3788 if (error && !VLOG_DROP_WARN(&rl)) {
3789 struct ds ds = DS_EMPTY_INITIALIZER;
3790 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3791 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3796 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3797 free(drop_key->key);
3802 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3803 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3804 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3805 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3806 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3807 * 'packet' ingressed.
3809 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3810 * 'flow''s in_port to OFPP_NONE.
3812 * This function does post-processing on data returned from
3813 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3814 * of the upcall processing logic. In particular, if the extracted in_port is
3815 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3816 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3817 * a VLAN header onto 'packet' (if it is nonnull).
3819 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3820 * to the VLAN TCI with which the packet was really received, that is, the
3821 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3822 * the value returned in flow->vlan_tci only for packets received on
3823 * VLAN splinters.) Also, if received on an IP tunnel, sets
3824 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3826 * Similarly, this function also includes some logic to help with tunnels. It
3827 * may modify 'flow' as necessary to make the tunneling implementation
3828 * transparent to the upcall processing logic.
3830 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3831 * or some other positive errno if there are other problems. */
3833 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3834 const struct nlattr *key, size_t key_len,
3835 struct flow *flow, enum odp_key_fitness *fitnessp,
3836 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3837 struct initial_vals *initial_vals)
3839 const struct ofport_dpif *port;
3840 enum odp_key_fitness fitness;
3843 fitness = odp_flow_key_to_flow(key, key_len, flow);
3844 if (fitness == ODP_FIT_ERROR) {
3850 initial_vals->vlan_tci = flow->vlan_tci;
3851 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3855 *odp_in_port = flow->in_port;
3858 if (tnl_port_should_receive(flow)) {
3859 const struct ofport *ofport = tnl_port_receive(flow);
3861 flow->in_port = OFPP_NONE;
3864 port = ofport_dpif_cast(ofport);
3866 /* We can't reproduce 'key' from 'flow'. */
3867 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3869 /* XXX: Since the tunnel module is not scoped per backer, it's
3870 * theoretically possible that we'll receive an ofport belonging to an
3871 * entirely different datapath. In practice, this can't happen because
3872 * no platforms has two separate datapaths which each support
3874 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3876 port = odp_port_to_ofport(backer, flow->in_port);
3878 flow->in_port = OFPP_NONE;
3882 flow->in_port = port->up.ofp_port;
3883 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3885 /* Make the packet resemble the flow, so that it gets sent to
3886 * an OpenFlow controller properly, so that it looks correct
3887 * for sFlow, and so that flow_extract() will get the correct
3888 * vlan_tci if it is called on 'packet'.
3890 * The allocated space inside 'packet' probably also contains
3891 * 'key', that is, both 'packet' and 'key' are probably part of
3892 * a struct dpif_upcall (see the large comment on that
3893 * structure definition), so pushing data on 'packet' is in
3894 * general not a good idea since it could overwrite 'key' or
3895 * free it as a side effect. However, it's OK in this special
3896 * case because we know that 'packet' is inside a Netlink
3897 * attribute: pushing 4 bytes will just overwrite the 4-byte
3898 * "struct nlattr", which is fine since we don't need that
3899 * header anymore. */
3900 eth_push_vlan(packet, flow->vlan_tci);
3902 /* We can't reproduce 'key' from 'flow'. */
3903 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3909 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3914 *fitnessp = fitness;
3920 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3923 struct dpif_upcall *upcall;
3924 struct flow_miss *miss;
3925 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3926 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3927 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3937 /* Construct the to-do list.
3939 * This just amounts to extracting the flow from each packet and sticking
3940 * the packets that have the same flow in the same "flow_miss" structure so
3941 * that we can process them together. */
3944 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3945 struct flow_miss *miss = &misses[n_misses];
3946 struct flow_miss *existing_miss;
3947 struct ofproto_dpif *ofproto;
3948 uint32_t odp_in_port;
3953 error = ofproto_receive(backer, upcall->packet, upcall->key,
3954 upcall->key_len, &flow, &miss->key_fitness,
3955 &ofproto, &odp_in_port, &miss->initial_vals);
3956 if (error == ENODEV) {
3957 struct drop_key *drop_key;
3959 /* Received packet on port for which we couldn't associate
3960 * an ofproto. This can happen if a port is removed while
3961 * traffic is being received. Print a rate-limited message
3962 * in case it happens frequently. Install a drop flow so
3963 * that future packets of the flow are inexpensively dropped
3965 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3968 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3970 drop_key = xmalloc(sizeof *drop_key);
3971 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3972 drop_key->key_len = upcall->key_len;
3974 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3975 hash_bytes(drop_key->key, drop_key->key_len, 0));
3976 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3977 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3985 ofproto->n_missed++;
3986 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3987 &flow.tunnel, flow.in_port, &miss->flow);
3989 /* Add other packets to a to-do list. */
3990 hash = flow_hash(&miss->flow, 0);
3991 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3992 if (!existing_miss) {
3993 hmap_insert(&todo, &miss->hmap_node, hash);
3994 miss->ofproto = ofproto;
3995 miss->key = upcall->key;
3996 miss->key_len = upcall->key_len;
3997 miss->upcall_type = upcall->type;
3998 miss->odp_in_port = odp_in_port;
3999 list_init(&miss->packets);
4003 miss = existing_miss;
4005 list_push_back(&miss->packets, &upcall->packet->list_node);
4008 /* Process each element in the to-do list, constructing the set of
4009 * operations to batch. */
4011 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4012 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4014 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4016 /* Execute batch. */
4017 for (i = 0; i < n_ops; i++) {
4018 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4020 dpif_operate(backer->dpif, dpif_ops, n_ops);
4023 for (i = 0; i < n_ops; i++) {
4024 free(flow_miss_ops[i].garbage);
4026 hmap_destroy(&todo);
4029 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
4030 classify_upcall(const struct dpif_upcall *upcall)
4032 union user_action_cookie cookie;
4034 /* First look at the upcall type. */
4035 switch (upcall->type) {
4036 case DPIF_UC_ACTION:
4042 case DPIF_N_UC_TYPES:
4044 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4048 /* "action" upcalls need a closer look. */
4049 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
4050 switch (cookie.type) {
4051 case USER_ACTION_COOKIE_SFLOW:
4052 return SFLOW_UPCALL;
4054 case USER_ACTION_COOKIE_SLOW_PATH:
4057 case USER_ACTION_COOKIE_UNSPEC:
4059 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
4065 handle_sflow_upcall(struct dpif_backer *backer,
4066 const struct dpif_upcall *upcall)
4068 struct ofproto_dpif *ofproto;
4069 union user_action_cookie cookie;
4071 uint32_t odp_in_port;
4073 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4074 &flow, NULL, &ofproto, &odp_in_port, NULL)
4075 || !ofproto->sflow) {
4079 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
4080 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4081 odp_in_port, &cookie);
4085 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4087 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4088 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4089 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4094 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4097 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4098 struct dpif_upcall *upcall = &misses[n_misses];
4099 struct ofpbuf *buf = &miss_bufs[n_misses];
4102 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4103 sizeof miss_buf_stubs[n_misses]);
4104 error = dpif_recv(backer->dpif, upcall, buf);
4110 switch (classify_upcall(upcall)) {
4112 /* Handle it later. */
4117 handle_sflow_upcall(backer, upcall);
4127 /* Handle deferred MISS_UPCALL processing. */
4128 handle_miss_upcalls(backer, misses, n_misses);
4129 for (i = 0; i < n_misses; i++) {
4130 ofpbuf_uninit(&miss_bufs[i]);
4136 /* Flow expiration. */
4138 static int subfacet_max_idle(const struct ofproto_dpif *);
4139 static void update_stats(struct dpif_backer *);
4140 static void rule_expire(struct rule_dpif *);
4141 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4143 /* This function is called periodically by run(). Its job is to collect
4144 * updates for the flows that have been installed into the datapath, most
4145 * importantly when they last were used, and then use that information to
4146 * expire flows that have not been used recently.
4148 * Returns the number of milliseconds after which it should be called again. */
4150 expire(struct dpif_backer *backer)
4152 struct ofproto_dpif *ofproto;
4153 int max_idle = INT32_MAX;
4155 /* Periodically clear out the drop keys in an effort to keep them
4156 * relatively few. */
4157 drop_key_clear(backer);
4159 /* Update stats for each flow in the backer. */
4160 update_stats(backer);
4162 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4163 struct rule *rule, *next_rule;
4166 if (ofproto->backer != backer) {
4170 /* Keep track of the max number of flows per ofproto_dpif. */
4171 update_max_subfacet_count(ofproto);
4173 /* Expire subfacets that have been idle too long. */
4174 dp_max_idle = subfacet_max_idle(ofproto);
4175 expire_subfacets(ofproto, dp_max_idle);
4177 max_idle = MIN(max_idle, dp_max_idle);
4179 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4181 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4182 &ofproto->up.expirable) {
4183 rule_expire(rule_dpif_cast(rule));
4186 /* All outstanding data in existing flows has been accounted, so it's a
4187 * good time to do bond rebalancing. */
4188 if (ofproto->has_bonded_bundles) {
4189 struct ofbundle *bundle;
4191 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4193 bond_rebalance(bundle->bond, &backer->revalidate_set);
4199 return MIN(max_idle, 1000);
4202 /* Updates flow table statistics given that the datapath just reported 'stats'
4203 * as 'subfacet''s statistics. */
4205 update_subfacet_stats(struct subfacet *subfacet,
4206 const struct dpif_flow_stats *stats)
4208 struct facet *facet = subfacet->facet;
4210 if (stats->n_packets >= subfacet->dp_packet_count) {
4211 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4212 facet->packet_count += extra;
4214 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4217 if (stats->n_bytes >= subfacet->dp_byte_count) {
4218 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4220 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4223 subfacet->dp_packet_count = stats->n_packets;
4224 subfacet->dp_byte_count = stats->n_bytes;
4226 facet->tcp_flags |= stats->tcp_flags;
4228 subfacet_update_time(subfacet, stats->used);
4229 if (facet->accounted_bytes < facet->byte_count) {
4231 facet_account(facet);
4232 facet->accounted_bytes = facet->byte_count;
4236 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4237 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4239 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4240 const struct nlattr *key, size_t key_len)
4242 if (!VLOG_DROP_WARN(&rl)) {
4246 odp_flow_key_format(key, key_len, &s);
4247 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4251 COVERAGE_INC(facet_unexpected);
4252 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4255 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4257 * This function also pushes statistics updates to rules which each facet
4258 * resubmits into. Generally these statistics will be accurate. However, if a
4259 * facet changes the rule it resubmits into at some time in between
4260 * update_stats() runs, it is possible that statistics accrued to the
4261 * old rule will be incorrectly attributed to the new rule. This could be
4262 * avoided by calling update_stats() whenever rules are created or
4263 * deleted. However, the performance impact of making so many calls to the
4264 * datapath do not justify the benefit of having perfectly accurate statistics.
4266 * In addition, this function maintains per ofproto flow hit counts. The patch
4267 * port is not treated specially. e.g. A packet ingress from br0 patched into
4268 * br1 will increase the hit count of br0 by 1, however, does not affect
4269 * the hit or miss counts of br1.
4272 update_stats(struct dpif_backer *backer)
4274 const struct dpif_flow_stats *stats;
4275 struct dpif_flow_dump dump;
4276 const struct nlattr *key;
4277 struct ofproto_dpif *ofproto;
4280 dpif_flow_dump_start(&dump, backer->dpif);
4281 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4283 struct subfacet *subfacet;
4284 struct ofport_dpif *ofport;
4287 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4292 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4293 ofproto->n_update_stats++;
4295 ofport = get_ofp_port(ofproto, flow.in_port);
4296 if (ofport && ofport->tnl_port) {
4297 netdev_vport_inc_rx(ofport->up.netdev, stats);
4300 key_hash = odp_flow_key_hash(key, key_len);
4301 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4302 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4304 /* Update ofproto_dpif's hit count. */
4305 if (stats->n_packets > subfacet->dp_packet_count) {
4306 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4307 dpif_stats_update_hit_count(ofproto, delta);
4310 update_subfacet_stats(subfacet, stats);
4314 /* Stats are updated per-packet. */
4317 case SF_NOT_INSTALLED:
4319 delete_unexpected_flow(ofproto, key, key_len);
4324 dpif_flow_dump_done(&dump);
4326 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4327 update_moving_averages(ofproto);
4332 /* Calculates and returns the number of milliseconds of idle time after which
4333 * subfacets should expire from the datapath. When a subfacet expires, we fold
4334 * its statistics into its facet, and when a facet's last subfacet expires, we
4335 * fold its statistic into its rule. */
4337 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4340 * Idle time histogram.
4342 * Most of the time a switch has a relatively small number of subfacets.
4343 * When this is the case we might as well keep statistics for all of them
4344 * in userspace and to cache them in the kernel datapath for performance as
4347 * As the number of subfacets increases, the memory required to maintain
4348 * statistics about them in userspace and in the kernel becomes
4349 * significant. However, with a large number of subfacets it is likely
4350 * that only a few of them are "heavy hitters" that consume a large amount
4351 * of bandwidth. At this point, only heavy hitters are worth caching in
4352 * the kernel and maintaining in userspaces; other subfacets we can
4355 * The technique used to compute the idle time is to build a histogram with
4356 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4357 * that is installed in the kernel gets dropped in the appropriate bucket.
4358 * After the histogram has been built, we compute the cutoff so that only
4359 * the most-recently-used 1% of subfacets (but at least
4360 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4361 * the most-recently-used bucket of subfacets is kept, so actually an
4362 * arbitrary number of subfacets can be kept in any given expiration run
4363 * (though the next run will delete most of those unless they receive
4366 * This requires a second pass through the subfacets, in addition to the
4367 * pass made by update_stats(), because the former function never looks at
4368 * uninstallable subfacets.
4370 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4371 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4372 int buckets[N_BUCKETS] = { 0 };
4373 int total, subtotal, bucket;
4374 struct subfacet *subfacet;
4378 total = hmap_count(&ofproto->subfacets);
4379 if (total <= ofproto->up.flow_eviction_threshold) {
4380 return N_BUCKETS * BUCKET_WIDTH;
4383 /* Build histogram. */
4385 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4386 long long int idle = now - subfacet->used;
4387 int bucket = (idle <= 0 ? 0
4388 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4389 : (unsigned int) idle / BUCKET_WIDTH);
4393 /* Find the first bucket whose flows should be expired. */
4394 subtotal = bucket = 0;
4396 subtotal += buckets[bucket++];
4397 } while (bucket < N_BUCKETS &&
4398 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4400 if (VLOG_IS_DBG_ENABLED()) {
4404 ds_put_cstr(&s, "keep");
4405 for (i = 0; i < N_BUCKETS; i++) {
4407 ds_put_cstr(&s, ", drop");
4410 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4413 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4417 return bucket * BUCKET_WIDTH;
4421 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4423 /* Cutoff time for most flows. */
4424 long long int normal_cutoff = time_msec() - dp_max_idle;
4426 /* We really want to keep flows for special protocols around, so use a more
4427 * conservative cutoff. */
4428 long long int special_cutoff = time_msec() - 10000;
4430 struct subfacet *subfacet, *next_subfacet;
4431 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4435 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4436 &ofproto->subfacets) {
4437 long long int cutoff;
4439 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4442 if (subfacet->used < cutoff) {
4443 if (subfacet->path != SF_NOT_INSTALLED) {
4444 batch[n_batch++] = subfacet;
4445 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4446 subfacet_destroy_batch(ofproto, batch, n_batch);
4450 subfacet_destroy(subfacet);
4456 subfacet_destroy_batch(ofproto, batch, n_batch);
4460 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4461 * then delete it entirely. */
4463 rule_expire(struct rule_dpif *rule)
4465 struct facet *facet, *next_facet;
4469 if (rule->up.pending) {
4470 /* We'll have to expire it later. */
4474 /* Has 'rule' expired? */
4476 if (rule->up.hard_timeout
4477 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4478 reason = OFPRR_HARD_TIMEOUT;
4479 } else if (rule->up.idle_timeout
4480 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4481 reason = OFPRR_IDLE_TIMEOUT;
4486 COVERAGE_INC(ofproto_dpif_expired);
4488 /* Update stats. (This is a no-op if the rule expired due to an idle
4489 * timeout, because that only happens when the rule has no facets left.) */
4490 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4491 facet_remove(facet);
4494 /* Get rid of the rule. */
4495 ofproto_rule_expire(&rule->up, reason);
4500 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4502 * The caller must already have determined that no facet with an identical
4503 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4504 * the ofproto's classifier table.
4506 * 'hash' must be the return value of flow_hash(flow, 0).
4508 * The facet will initially have no subfacets. The caller should create (at
4509 * least) one subfacet with subfacet_create(). */
4510 static struct facet *
4511 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4513 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4514 struct facet *facet;
4516 facet = xzalloc(sizeof *facet);
4517 facet->used = time_msec();
4518 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4519 list_push_back(&rule->facets, &facet->list_node);
4521 facet->flow = *flow;
4522 list_init(&facet->subfacets);
4523 netflow_flow_init(&facet->nf_flow);
4524 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4526 facet->learn_rl = time_msec() + 500;
4532 facet_free(struct facet *facet)
4537 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4538 * 'packet', which arrived on 'in_port'. */
4540 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4541 const struct nlattr *odp_actions, size_t actions_len,
4542 struct ofpbuf *packet)
4544 struct odputil_keybuf keybuf;
4548 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4549 odp_flow_key_from_flow(&key, flow,
4550 ofp_port_to_odp_port(ofproto, flow->in_port));
4552 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4553 odp_actions, actions_len, packet);
4557 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4559 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4560 * rule's statistics, via subfacet_uninstall().
4562 * - Removes 'facet' from its rule and from ofproto->facets.
4565 facet_remove(struct facet *facet)
4567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4568 struct subfacet *subfacet, *next_subfacet;
4570 ovs_assert(!list_is_empty(&facet->subfacets));
4572 /* First uninstall all of the subfacets to get final statistics. */
4573 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4574 subfacet_uninstall(subfacet);
4577 /* Flush the final stats to the rule.
4579 * This might require us to have at least one subfacet around so that we
4580 * can use its actions for accounting in facet_account(), which is why we
4581 * have uninstalled but not yet destroyed the subfacets. */
4582 facet_flush_stats(facet);
4584 /* Now we're really all done so destroy everything. */
4585 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4586 &facet->subfacets) {
4587 subfacet_destroy__(subfacet);
4589 hmap_remove(&ofproto->facets, &facet->hmap_node);
4590 list_remove(&facet->list_node);
4594 /* Feed information from 'facet' back into the learning table to keep it in
4595 * sync with what is actually flowing through the datapath. */
4597 facet_learn(struct facet *facet)
4599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4600 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4601 struct subfacet, list_node);
4602 long long int now = time_msec();
4603 struct action_xlate_ctx ctx;
4605 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4609 facet->learn_rl = now + 500;
4611 if (!facet->has_learn
4612 && !facet->has_normal
4613 && (!facet->has_fin_timeout
4614 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4618 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4619 &subfacet->initial_vals,
4620 facet->rule, facet->tcp_flags, NULL);
4621 ctx.may_learn = true;
4622 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4623 facet->rule->up.ofpacts_len);
4627 facet_account(struct facet *facet)
4629 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4630 struct subfacet *subfacet;
4631 const struct nlattr *a;
4636 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4639 n_bytes = facet->byte_count - facet->accounted_bytes;
4641 /* This loop feeds byte counters to bond_account() for rebalancing to use
4642 * as a basis. We also need to track the actual VLAN on which the packet
4643 * is going to be sent to ensure that it matches the one passed to
4644 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4647 * We use the actions from an arbitrary subfacet because they should all
4648 * be equally valid for our purpose. */
4649 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4650 struct subfacet, list_node);
4651 vlan_tci = facet->flow.vlan_tci;
4652 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4653 subfacet->actions, subfacet->actions_len) {
4654 const struct ovs_action_push_vlan *vlan;
4655 struct ofport_dpif *port;
4657 switch (nl_attr_type(a)) {
4658 case OVS_ACTION_ATTR_OUTPUT:
4659 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4660 if (port && port->bundle && port->bundle->bond) {
4661 bond_account(port->bundle->bond, &facet->flow,
4662 vlan_tci_to_vid(vlan_tci), n_bytes);
4666 case OVS_ACTION_ATTR_POP_VLAN:
4667 vlan_tci = htons(0);
4670 case OVS_ACTION_ATTR_PUSH_VLAN:
4671 vlan = nl_attr_get(a);
4672 vlan_tci = vlan->vlan_tci;
4678 /* Returns true if the only action for 'facet' is to send to the controller.
4679 * (We don't report NetFlow expiration messages for such facets because they
4680 * are just part of the control logic for the network, not real traffic). */
4682 facet_is_controller_flow(struct facet *facet)
4685 const struct rule *rule = &facet->rule->up;
4686 const struct ofpact *ofpacts = rule->ofpacts;
4687 size_t ofpacts_len = rule->ofpacts_len;
4689 if (ofpacts_len > 0 &&
4690 ofpacts->type == OFPACT_CONTROLLER &&
4691 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4698 /* Folds all of 'facet''s statistics into its rule. Also updates the
4699 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4700 * 'facet''s statistics in the datapath should have been zeroed and folded into
4701 * its packet and byte counts before this function is called. */
4703 facet_flush_stats(struct facet *facet)
4705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4706 struct subfacet *subfacet;
4708 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4709 ovs_assert(!subfacet->dp_byte_count);
4710 ovs_assert(!subfacet->dp_packet_count);
4713 facet_push_stats(facet);
4714 if (facet->accounted_bytes < facet->byte_count) {
4715 facet_account(facet);
4716 facet->accounted_bytes = facet->byte_count;
4719 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4720 struct ofexpired expired;
4721 expired.flow = facet->flow;
4722 expired.packet_count = facet->packet_count;
4723 expired.byte_count = facet->byte_count;
4724 expired.used = facet->used;
4725 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4728 facet->rule->packet_count += facet->packet_count;
4729 facet->rule->byte_count += facet->byte_count;
4731 /* Reset counters to prevent double counting if 'facet' ever gets
4733 facet_reset_counters(facet);
4735 netflow_flow_clear(&facet->nf_flow);
4736 facet->tcp_flags = 0;
4739 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4740 * Returns it if found, otherwise a null pointer.
4742 * 'hash' must be the return value of flow_hash(flow, 0).
4744 * The returned facet might need revalidation; use facet_lookup_valid()
4745 * instead if that is important. */
4746 static struct facet *
4747 facet_find(struct ofproto_dpif *ofproto,
4748 const struct flow *flow, uint32_t hash)
4750 struct facet *facet;
4752 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4753 if (flow_equal(flow, &facet->flow)) {
4761 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4762 * Returns it if found, otherwise a null pointer.
4764 * 'hash' must be the return value of flow_hash(flow, 0).
4766 * The returned facet is guaranteed to be valid. */
4767 static struct facet *
4768 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4771 struct facet *facet;
4773 facet = facet_find(ofproto, flow, hash);
4775 && (ofproto->backer->need_revalidate
4776 || tag_set_intersects(&ofproto->backer->revalidate_set,
4778 facet_revalidate(facet);
4780 /* facet_revalidate() may have destroyed 'facet'. */
4781 facet = facet_find(ofproto, flow, hash);
4788 subfacet_path_to_string(enum subfacet_path path)
4791 case SF_NOT_INSTALLED:
4792 return "not installed";
4794 return "in fast path";
4796 return "in slow path";
4802 /* Returns the path in which a subfacet should be installed if its 'slow'
4803 * member has the specified value. */
4804 static enum subfacet_path
4805 subfacet_want_path(enum slow_path_reason slow)
4807 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4810 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4811 * supposing that its actions have been recalculated as 'want_actions' and that
4812 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4814 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4815 const struct ofpbuf *want_actions)
4817 enum subfacet_path want_path = subfacet_want_path(slow);
4818 return (want_path != subfacet->path
4819 || (want_path == SF_FAST_PATH
4820 && (subfacet->actions_len != want_actions->size
4821 || memcmp(subfacet->actions, want_actions->data,
4822 subfacet->actions_len))));
4826 facet_check_consistency(struct facet *facet)
4828 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4832 uint64_t odp_actions_stub[1024 / 8];
4833 struct ofpbuf odp_actions;
4835 struct rule_dpif *rule;
4836 struct subfacet *subfacet;
4837 bool may_log = false;
4840 /* Check the rule for consistency. */
4841 rule = rule_dpif_lookup(ofproto, &facet->flow);
4842 ok = rule == facet->rule;
4844 may_log = !VLOG_DROP_WARN(&rl);
4849 flow_format(&s, &facet->flow);
4850 ds_put_format(&s, ": facet associated with wrong rule (was "
4851 "table=%"PRIu8",", facet->rule->up.table_id);
4852 cls_rule_format(&facet->rule->up.cr, &s);
4853 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4855 cls_rule_format(&rule->up.cr, &s);
4856 ds_put_char(&s, ')');
4858 VLOG_WARN("%s", ds_cstr(&s));
4863 /* Check the datapath actions for consistency. */
4864 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4865 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4866 enum subfacet_path want_path;
4867 struct action_xlate_ctx ctx;
4870 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4871 &subfacet->initial_vals, rule, 0, NULL);
4872 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4875 if (subfacet->path == SF_NOT_INSTALLED) {
4876 /* This only happens if the datapath reported an error when we
4877 * tried to install the flow. Don't flag another error here. */
4881 want_path = subfacet_want_path(subfacet->slow);
4882 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4883 /* The actions for slow-path flows may legitimately vary from one
4884 * packet to the next. We're done. */
4888 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4892 /* Inconsistency! */
4894 may_log = !VLOG_DROP_WARN(&rl);
4898 /* Rate-limited, skip reporting. */
4903 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4905 ds_put_cstr(&s, ": inconsistency in subfacet");
4906 if (want_path != subfacet->path) {
4907 enum odp_key_fitness fitness = subfacet->key_fitness;
4909 ds_put_format(&s, " (%s, fitness=%s)",
4910 subfacet_path_to_string(subfacet->path),
4911 odp_key_fitness_to_string(fitness));
4912 ds_put_format(&s, " (should have been %s)",
4913 subfacet_path_to_string(want_path));
4914 } else if (want_path == SF_FAST_PATH) {
4915 ds_put_cstr(&s, " (actions were: ");
4916 format_odp_actions(&s, subfacet->actions,
4917 subfacet->actions_len);
4918 ds_put_cstr(&s, ") (correct actions: ");
4919 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4920 ds_put_char(&s, ')');
4922 ds_put_cstr(&s, " (actions: ");
4923 format_odp_actions(&s, subfacet->actions,
4924 subfacet->actions_len);
4925 ds_put_char(&s, ')');
4927 VLOG_WARN("%s", ds_cstr(&s));
4930 ofpbuf_uninit(&odp_actions);
4935 /* Re-searches the classifier for 'facet':
4937 * - If the rule found is different from 'facet''s current rule, moves
4938 * 'facet' to the new rule and recompiles its actions.
4940 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4941 * where it is and recompiles its actions anyway.
4943 * - If any of 'facet''s subfacets correspond to a new flow according to
4944 * ofproto_receive(), 'facet' is removed. */
4946 facet_revalidate(struct facet *facet)
4948 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4950 struct nlattr *odp_actions;
4953 struct actions *new_actions;
4955 struct action_xlate_ctx ctx;
4956 uint64_t odp_actions_stub[1024 / 8];
4957 struct ofpbuf odp_actions;
4959 struct rule_dpif *new_rule;
4960 struct subfacet *subfacet;
4963 COVERAGE_INC(facet_revalidate);
4965 /* Check that child subfacets still correspond to this facet. Tunnel
4966 * configuration changes could cause a subfacet's OpenFlow in_port to
4968 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4969 struct ofproto_dpif *recv_ofproto;
4970 struct flow recv_flow;
4973 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4974 subfacet->key_len, &recv_flow, NULL,
4975 &recv_ofproto, NULL, NULL);
4977 || recv_ofproto != ofproto
4978 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4979 facet_remove(facet);
4984 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4986 /* Calculate new datapath actions.
4988 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4989 * emit a NetFlow expiration and, if so, we need to have the old state
4990 * around to properly compose it. */
4992 /* If the datapath actions changed or the installability changed,
4993 * then we need to talk to the datapath. */
4996 memset(&ctx, 0, sizeof ctx);
4997 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4998 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4999 enum slow_path_reason slow;
5001 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5002 &subfacet->initial_vals, new_rule, 0, NULL);
5003 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5006 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5007 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5008 struct dpif_flow_stats stats;
5010 subfacet_install(subfacet,
5011 odp_actions.data, odp_actions.size, &stats, slow);
5012 subfacet_update_stats(subfacet, &stats);
5015 new_actions = xcalloc(list_size(&facet->subfacets),
5016 sizeof *new_actions);
5018 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5020 new_actions[i].actions_len = odp_actions.size;
5025 ofpbuf_uninit(&odp_actions);
5028 facet_flush_stats(facet);
5031 /* Update 'facet' now that we've taken care of all the old state. */
5032 facet->tags = ctx.tags;
5033 facet->nf_flow.output_iface = ctx.nf_output_iface;
5034 facet->has_learn = ctx.has_learn;
5035 facet->has_normal = ctx.has_normal;
5036 facet->has_fin_timeout = ctx.has_fin_timeout;
5037 facet->mirrors = ctx.mirrors;
5040 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5041 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5043 if (new_actions && new_actions[i].odp_actions) {
5044 free(subfacet->actions);
5045 subfacet->actions = new_actions[i].odp_actions;
5046 subfacet->actions_len = new_actions[i].actions_len;
5052 if (facet->rule != new_rule) {
5053 COVERAGE_INC(facet_changed_rule);
5054 list_remove(&facet->list_node);
5055 list_push_back(&new_rule->facets, &facet->list_node);
5056 facet->rule = new_rule;
5057 facet->used = new_rule->up.created;
5058 facet->prev_used = facet->used;
5062 /* Updates 'facet''s used time. Caller is responsible for calling
5063 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5065 facet_update_time(struct facet *facet, long long int used)
5067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5068 if (used > facet->used) {
5070 ofproto_rule_update_used(&facet->rule->up, used);
5071 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5076 facet_reset_counters(struct facet *facet)
5078 facet->packet_count = 0;
5079 facet->byte_count = 0;
5080 facet->prev_packet_count = 0;
5081 facet->prev_byte_count = 0;
5082 facet->accounted_bytes = 0;
5086 facet_push_stats(struct facet *facet)
5088 struct dpif_flow_stats stats;
5090 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5091 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5092 ovs_assert(facet->used >= facet->prev_used);
5094 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5095 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5096 stats.used = facet->used;
5097 stats.tcp_flags = 0;
5099 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5100 facet->prev_packet_count = facet->packet_count;
5101 facet->prev_byte_count = facet->byte_count;
5102 facet->prev_used = facet->used;
5104 flow_push_stats(facet, &stats);
5106 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5107 facet->mirrors, stats.n_packets, stats.n_bytes);
5112 push_all_stats__(bool run_fast)
5114 static long long int rl = LLONG_MIN;
5115 struct ofproto_dpif *ofproto;
5117 if (time_msec() < rl) {
5121 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5122 struct facet *facet;
5124 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5125 facet_push_stats(facet);
5132 rl = time_msec() + 100;
5136 push_all_stats(void)
5138 push_all_stats__(true);
5142 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5144 rule->packet_count += stats->n_packets;
5145 rule->byte_count += stats->n_bytes;
5146 ofproto_rule_update_used(&rule->up, stats->used);
5149 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5150 * into given 'facet->rule''s actions and mirrors. */
5152 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5154 struct rule_dpif *rule = facet->rule;
5155 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5156 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
5157 struct subfacet, list_node);
5158 struct action_xlate_ctx ctx;
5160 ofproto_rule_update_used(&rule->up, stats->used);
5162 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5163 &subfacet->initial_vals, rule, 0, NULL);
5164 ctx.resubmit_stats = stats;
5165 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5166 rule->up.ofpacts_len);
5171 static struct subfacet *
5172 subfacet_find(struct ofproto_dpif *ofproto,
5173 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5175 struct subfacet *subfacet;
5177 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5178 &ofproto->subfacets) {
5179 if (subfacet->key_len == key_len
5180 && !memcmp(key, subfacet->key, key_len)) {
5188 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5189 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5190 * existing subfacet if there is one, otherwise creates and returns a
5193 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5194 * which case the caller must populate the actions with
5195 * subfacet_make_actions(). */
5196 static struct subfacet *
5197 subfacet_create(struct facet *facet, struct flow_miss *miss,
5200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5201 enum odp_key_fitness key_fitness = miss->key_fitness;
5202 const struct nlattr *key = miss->key;
5203 size_t key_len = miss->key_len;
5205 struct subfacet *subfacet;
5207 key_hash = odp_flow_key_hash(key, key_len);
5209 if (list_is_empty(&facet->subfacets)) {
5210 subfacet = &facet->one_subfacet;
5212 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5214 if (subfacet->facet == facet) {
5218 /* This shouldn't happen. */
5219 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5220 subfacet_destroy(subfacet);
5223 subfacet = xmalloc(sizeof *subfacet);
5226 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5227 list_push_back(&facet->subfacets, &subfacet->list_node);
5228 subfacet->facet = facet;
5229 subfacet->key_fitness = key_fitness;
5230 subfacet->key = xmemdup(key, key_len);
5231 subfacet->key_len = key_len;
5232 subfacet->used = now;
5233 subfacet->created = now;
5234 subfacet->dp_packet_count = 0;
5235 subfacet->dp_byte_count = 0;
5236 subfacet->actions_len = 0;
5237 subfacet->actions = NULL;
5238 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5241 subfacet->path = SF_NOT_INSTALLED;
5242 subfacet->initial_vals = miss->initial_vals;
5243 subfacet->odp_in_port = miss->odp_in_port;
5245 ofproto->subfacet_add_count++;
5249 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5250 * its facet within 'ofproto', and frees it. */
5252 subfacet_destroy__(struct subfacet *subfacet)
5254 struct facet *facet = subfacet->facet;
5255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5257 /* Update ofproto stats before uninstall the subfacet. */
5258 ofproto->subfacet_del_count++;
5259 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5261 subfacet_uninstall(subfacet);
5262 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5263 list_remove(&subfacet->list_node);
5264 free(subfacet->key);
5265 free(subfacet->actions);
5266 if (subfacet != &facet->one_subfacet) {
5271 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5272 * last remaining subfacet in its facet destroys the facet too. */
5274 subfacet_destroy(struct subfacet *subfacet)
5276 struct facet *facet = subfacet->facet;
5278 if (list_is_singleton(&facet->subfacets)) {
5279 /* facet_remove() needs at least one subfacet (it will remove it). */
5280 facet_remove(facet);
5282 subfacet_destroy__(subfacet);
5287 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5288 struct subfacet **subfacets, int n)
5290 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5291 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5292 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5295 for (i = 0; i < n; i++) {
5296 ops[i].type = DPIF_OP_FLOW_DEL;
5297 ops[i].u.flow_del.key = subfacets[i]->key;
5298 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5299 ops[i].u.flow_del.stats = &stats[i];
5303 dpif_operate(ofproto->backer->dpif, opsp, n);
5304 for (i = 0; i < n; i++) {
5305 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5306 subfacets[i]->path = SF_NOT_INSTALLED;
5307 subfacet_destroy(subfacets[i]);
5312 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5313 * Translates the actions into 'odp_actions', which the caller must have
5314 * initialized and is responsible for uninitializing. */
5316 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5317 struct ofpbuf *odp_actions)
5319 struct facet *facet = subfacet->facet;
5320 struct rule_dpif *rule = facet->rule;
5321 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5323 struct action_xlate_ctx ctx;
5325 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5326 &subfacet->initial_vals, rule, 0, packet);
5327 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5328 facet->tags = ctx.tags;
5329 facet->has_learn = ctx.has_learn;
5330 facet->has_normal = ctx.has_normal;
5331 facet->has_fin_timeout = ctx.has_fin_timeout;
5332 facet->nf_flow.output_iface = ctx.nf_output_iface;
5333 facet->mirrors = ctx.mirrors;
5335 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5336 if (subfacet->actions_len != odp_actions->size
5337 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5338 free(subfacet->actions);
5339 subfacet->actions_len = odp_actions->size;
5340 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5344 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5345 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5346 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5347 * since 'subfacet' was last updated.
5349 * Returns 0 if successful, otherwise a positive errno value. */
5351 subfacet_install(struct subfacet *subfacet,
5352 const struct nlattr *actions, size_t actions_len,
5353 struct dpif_flow_stats *stats,
5354 enum slow_path_reason slow)
5356 struct facet *facet = subfacet->facet;
5357 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5358 enum subfacet_path path = subfacet_want_path(slow);
5359 uint64_t slow_path_stub[128 / 8];
5360 enum dpif_flow_put_flags flags;
5363 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5365 flags |= DPIF_FP_ZERO_STATS;
5368 if (path == SF_SLOW_PATH) {
5369 compose_slow_path(ofproto, &facet->flow, slow,
5370 slow_path_stub, sizeof slow_path_stub,
5371 &actions, &actions_len);
5374 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5375 subfacet->key_len, actions, actions_len, stats);
5378 subfacet_reset_dp_stats(subfacet, stats);
5382 subfacet->path = path;
5388 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5390 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5391 stats, subfacet->slow);
5394 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5396 subfacet_uninstall(struct subfacet *subfacet)
5398 if (subfacet->path != SF_NOT_INSTALLED) {
5399 struct rule_dpif *rule = subfacet->facet->rule;
5400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5401 struct dpif_flow_stats stats;
5404 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5405 subfacet->key_len, &stats);
5406 subfacet_reset_dp_stats(subfacet, &stats);
5408 subfacet_update_stats(subfacet, &stats);
5410 subfacet->path = SF_NOT_INSTALLED;
5412 ovs_assert(subfacet->dp_packet_count == 0);
5413 ovs_assert(subfacet->dp_byte_count == 0);
5417 /* Resets 'subfacet''s datapath statistics counters. This should be called
5418 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5419 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5420 * was reset in the datapath. 'stats' will be modified to include only
5421 * statistics new since 'subfacet' was last updated. */
5423 subfacet_reset_dp_stats(struct subfacet *subfacet,
5424 struct dpif_flow_stats *stats)
5427 && subfacet->dp_packet_count <= stats->n_packets
5428 && subfacet->dp_byte_count <= stats->n_bytes) {
5429 stats->n_packets -= subfacet->dp_packet_count;
5430 stats->n_bytes -= subfacet->dp_byte_count;
5433 subfacet->dp_packet_count = 0;
5434 subfacet->dp_byte_count = 0;
5437 /* Updates 'subfacet''s used time. The caller is responsible for calling
5438 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5440 subfacet_update_time(struct subfacet *subfacet, long long int used)
5442 if (used > subfacet->used) {
5443 subfacet->used = used;
5444 facet_update_time(subfacet->facet, used);
5448 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5450 * Because of the meaning of a subfacet's counters, it only makes sense to do
5451 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5452 * represents a packet that was sent by hand or if it represents statistics
5453 * that have been cleared out of the datapath. */
5455 subfacet_update_stats(struct subfacet *subfacet,
5456 const struct dpif_flow_stats *stats)
5458 if (stats->n_packets || stats->used > subfacet->used) {
5459 struct facet *facet = subfacet->facet;
5461 subfacet_update_time(subfacet, stats->used);
5462 facet->packet_count += stats->n_packets;
5463 facet->byte_count += stats->n_bytes;
5464 facet->tcp_flags |= stats->tcp_flags;
5465 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5471 static struct rule_dpif *
5472 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5474 struct rule_dpif *rule;
5476 rule = rule_dpif_lookup__(ofproto, flow, 0);
5481 return rule_dpif_miss_rule(ofproto, flow);
5484 static struct rule_dpif *
5485 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5488 struct cls_rule *cls_rule;
5489 struct classifier *cls;
5491 if (table_id >= N_TABLES) {
5495 cls = &ofproto->up.tables[table_id].cls;
5496 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5497 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5498 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5499 * are unavailable. */
5500 struct flow ofpc_normal_flow = *flow;
5501 ofpc_normal_flow.tp_src = htons(0);
5502 ofpc_normal_flow.tp_dst = htons(0);
5503 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5505 cls_rule = classifier_lookup(cls, flow);
5507 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5510 static struct rule_dpif *
5511 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5513 struct ofport_dpif *port;
5515 port = get_ofp_port(ofproto, flow->in_port);
5517 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5518 return ofproto->miss_rule;
5521 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5522 return ofproto->no_packet_in_rule;
5524 return ofproto->miss_rule;
5528 complete_operation(struct rule_dpif *rule)
5530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5532 rule_invalidate(rule);
5534 struct dpif_completion *c = xmalloc(sizeof *c);
5535 c->op = rule->up.pending;
5536 list_push_back(&ofproto->completions, &c->list_node);
5538 ofoperation_complete(rule->up.pending, 0);
5542 static struct rule *
5545 struct rule_dpif *rule = xmalloc(sizeof *rule);
5550 rule_dealloc(struct rule *rule_)
5552 struct rule_dpif *rule = rule_dpif_cast(rule_);
5557 rule_construct(struct rule *rule_)
5559 struct rule_dpif *rule = rule_dpif_cast(rule_);
5560 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5561 struct rule_dpif *victim;
5564 rule->packet_count = 0;
5565 rule->byte_count = 0;
5567 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5568 if (victim && !list_is_empty(&victim->facets)) {
5569 struct facet *facet;
5571 rule->facets = victim->facets;
5572 list_moved(&rule->facets);
5573 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5574 /* XXX: We're only clearing our local counters here. It's possible
5575 * that quite a few packets are unaccounted for in the datapath
5576 * statistics. These will be accounted to the new rule instead of
5577 * cleared as required. This could be fixed by clearing out the
5578 * datapath statistics for this facet, but currently it doesn't
5580 facet_reset_counters(facet);
5584 /* Must avoid list_moved() in this case. */
5585 list_init(&rule->facets);
5588 table_id = rule->up.table_id;
5590 rule->tag = victim->tag;
5591 } else if (table_id == 0) {
5596 miniflow_expand(&rule->up.cr.match.flow, &flow);
5597 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5598 ofproto->tables[table_id].basis);
5601 complete_operation(rule);
5606 rule_destruct(struct rule *rule_)
5608 struct rule_dpif *rule = rule_dpif_cast(rule_);
5609 struct facet *facet, *next_facet;
5611 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5612 facet_revalidate(facet);
5615 complete_operation(rule);
5619 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5621 struct rule_dpif *rule = rule_dpif_cast(rule_);
5622 struct facet *facet;
5624 /* push_all_stats() can handle flow misses which, when using the learn
5625 * action, can cause rules to be added and deleted. This can corrupt our
5626 * caller's datastructures which assume that rule_get_stats() doesn't have
5627 * an impact on the flow table. To be safe, we disable miss handling. */
5628 push_all_stats__(false);
5630 /* Start from historical data for 'rule' itself that are no longer tracked
5631 * in facets. This counts, for example, facets that have expired. */
5632 *packets = rule->packet_count;
5633 *bytes = rule->byte_count;
5635 /* Add any statistics that are tracked by facets. This includes
5636 * statistical data recently updated by ofproto_update_stats() as well as
5637 * stats for packets that were executed "by hand" via dpif_execute(). */
5638 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5639 *packets += facet->packet_count;
5640 *bytes += facet->byte_count;
5645 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5646 struct ofpbuf *packet)
5648 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5649 struct initial_vals initial_vals;
5650 struct dpif_flow_stats stats;
5651 struct action_xlate_ctx ctx;
5652 uint64_t odp_actions_stub[1024 / 8];
5653 struct ofpbuf odp_actions;
5655 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5656 rule_credit_stats(rule, &stats);
5658 initial_vals.vlan_tci = flow->vlan_tci;
5659 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5660 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5661 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5662 rule, stats.tcp_flags, packet);
5663 ctx.resubmit_stats = &stats;
5664 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5666 execute_odp_actions(ofproto, flow, odp_actions.data,
5667 odp_actions.size, packet);
5669 ofpbuf_uninit(&odp_actions);
5673 rule_execute(struct rule *rule, const struct flow *flow,
5674 struct ofpbuf *packet)
5676 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5677 ofpbuf_delete(packet);
5682 rule_modify_actions(struct rule *rule_)
5684 struct rule_dpif *rule = rule_dpif_cast(rule_);
5686 complete_operation(rule);
5689 /* Sends 'packet' out 'ofport'.
5690 * May modify 'packet'.
5691 * Returns 0 if successful, otherwise a positive errno value. */
5693 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5695 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5696 uint64_t odp_actions_stub[1024 / 8];
5697 struct ofpbuf key, odp_actions;
5698 struct odputil_keybuf keybuf;
5703 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5704 if (netdev_vport_is_patch(ofport->up.netdev)) {
5705 struct ofproto_dpif *peer_ofproto;
5706 struct dpif_flow_stats stats;
5707 struct ofport_dpif *peer;
5708 struct rule_dpif *rule;
5710 peer = ofport_get_peer(ofport);
5715 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5716 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5717 netdev_vport_inc_rx(peer->up.netdev, &stats);
5719 flow.in_port = peer->up.ofp_port;
5720 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5721 rule = rule_dpif_lookup(peer_ofproto, &flow);
5722 rule_dpif_execute(rule, &flow, packet);
5727 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5729 if (ofport->tnl_port) {
5730 struct dpif_flow_stats stats;
5732 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5733 if (odp_port == OVSP_NONE) {
5737 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5738 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5739 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5740 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5742 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5744 if (odp_port != ofport->odp_port) {
5745 eth_pop_vlan(packet);
5746 flow.vlan_tci = htons(0);
5750 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5751 odp_flow_key_from_flow(&key, &flow,
5752 ofp_port_to_odp_port(ofproto, flow.in_port));
5754 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5756 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5757 error = dpif_execute(ofproto->backer->dpif,
5759 odp_actions.data, odp_actions.size,
5761 ofpbuf_uninit(&odp_actions);
5764 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5765 ofproto->up.name, odp_port, strerror(error));
5767 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5771 /* OpenFlow to datapath action translation. */
5773 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5774 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5775 struct action_xlate_ctx *);
5776 static void xlate_normal(struct action_xlate_ctx *);
5778 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5779 * The action will state 'slow' as the reason that the action is in the slow
5780 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5781 * dump-flows" output to see why a flow is in the slow path.)
5783 * The 'stub_size' bytes in 'stub' will be used to store the action.
5784 * 'stub_size' must be large enough for the action.
5786 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5789 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5790 enum slow_path_reason slow,
5791 uint64_t *stub, size_t stub_size,
5792 const struct nlattr **actionsp, size_t *actions_lenp)
5794 union user_action_cookie cookie;
5797 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5798 cookie.slow_path.unused = 0;
5799 cookie.slow_path.reason = slow;
5801 ofpbuf_use_stack(&buf, stub, stub_size);
5802 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5803 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5804 odp_put_userspace_action(pid, &cookie, &buf);
5806 put_userspace_action(ofproto, &buf, flow, &cookie);
5808 *actionsp = buf.data;
5809 *actions_lenp = buf.size;
5813 put_userspace_action(const struct ofproto_dpif *ofproto,
5814 struct ofpbuf *odp_actions,
5815 const struct flow *flow,
5816 const union user_action_cookie *cookie)
5820 pid = dpif_port_get_pid(ofproto->backer->dpif,
5821 ofp_port_to_odp_port(ofproto, flow->in_port));
5823 return odp_put_userspace_action(pid, cookie, odp_actions);
5827 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5828 ovs_be16 vlan_tci, uint32_t odp_port,
5829 unsigned int n_outputs, union user_action_cookie *cookie)
5833 cookie->type = USER_ACTION_COOKIE_SFLOW;
5834 cookie->sflow.vlan_tci = vlan_tci;
5836 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5837 * port information") for the interpretation of cookie->output. */
5838 switch (n_outputs) {
5840 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5841 cookie->sflow.output = 0x40000000 | 256;
5845 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5847 cookie->sflow.output = ifindex;
5852 /* 0x80000000 means "multiple output ports. */
5853 cookie->sflow.output = 0x80000000 | n_outputs;
5858 /* Compose SAMPLE action for sFlow. */
5860 compose_sflow_action(const struct ofproto_dpif *ofproto,
5861 struct ofpbuf *odp_actions,
5862 const struct flow *flow,
5865 uint32_t probability;
5866 union user_action_cookie cookie;
5867 size_t sample_offset, actions_offset;
5870 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5874 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5876 /* Number of packets out of UINT_MAX to sample. */
5877 probability = dpif_sflow_get_probability(ofproto->sflow);
5878 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5880 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5881 compose_sflow_cookie(ofproto, htons(0), odp_port,
5882 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5883 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5885 nl_msg_end_nested(odp_actions, actions_offset);
5886 nl_msg_end_nested(odp_actions, sample_offset);
5887 return cookie_offset;
5890 /* SAMPLE action must be first action in any given list of actions.
5891 * At this point we do not have all information required to build it. So try to
5892 * build sample action as complete as possible. */
5894 add_sflow_action(struct action_xlate_ctx *ctx)
5896 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5898 &ctx->flow, OVSP_NONE);
5899 ctx->sflow_odp_port = 0;
5900 ctx->sflow_n_outputs = 0;
5903 /* Fix SAMPLE action according to data collected while composing ODP actions.
5904 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5905 * USERSPACE action's user-cookie which is required for sflow. */
5907 fix_sflow_action(struct action_xlate_ctx *ctx)
5909 const struct flow *base = &ctx->base_flow;
5910 union user_action_cookie *cookie;
5912 if (!ctx->user_cookie_offset) {
5916 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5918 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5920 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5921 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5925 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5928 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5929 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5930 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5931 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5932 struct priority_to_dscp *pdscp;
5933 uint32_t out_port, odp_port;
5935 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5936 * before traversing a patch port. */
5937 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5940 xlate_report(ctx, "Nonexistent output port");
5942 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5943 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5945 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5946 xlate_report(ctx, "STP not in forwarding state, skipping output");
5950 if (netdev_vport_is_patch(ofport->up.netdev)) {
5951 struct ofport_dpif *peer = ofport_get_peer(ofport);
5952 struct flow old_flow = ctx->flow;
5953 const struct ofproto_dpif *peer_ofproto;
5954 enum slow_path_reason special;
5955 struct ofport_dpif *in_port;
5958 xlate_report(ctx, "Nonexistent patch port peer");
5962 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5963 if (peer_ofproto->backer != ctx->ofproto->backer) {
5964 xlate_report(ctx, "Patch port peer on a different datapath");
5968 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5969 ctx->flow.in_port = peer->up.ofp_port;
5970 ctx->flow.metadata = htonll(0);
5971 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5972 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5974 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5975 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5978 ctx->slow |= special;
5979 } else if (!in_port || may_receive(in_port, ctx)) {
5980 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5981 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5983 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5984 * learning action look at the packet, then drop it. */
5985 struct flow old_base_flow = ctx->base_flow;
5986 size_t old_size = ctx->odp_actions->size;
5987 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5988 ctx->base_flow = old_base_flow;
5989 ctx->odp_actions->size = old_size;
5993 ctx->flow = old_flow;
5994 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5996 if (ctx->resubmit_stats) {
5997 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5998 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6004 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6006 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6007 ctx->flow.nw_tos |= pdscp->dscp;
6010 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
6011 if (ofport->tnl_port) {
6012 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6013 if (odp_port == OVSP_NONE) {
6014 xlate_report(ctx, "Tunneling decided against output");
6018 if (ctx->resubmit_stats) {
6019 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6021 out_port = odp_port;
6022 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6025 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6026 ctx->flow.vlan_tci);
6027 if (out_port != odp_port) {
6028 ctx->flow.vlan_tci = htons(0);
6030 ctx->flow.skb_mark &= ~IPSEC_MARK;
6032 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6033 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6035 ctx->sflow_odp_port = odp_port;
6036 ctx->sflow_n_outputs++;
6037 ctx->nf_output_iface = ofp_port;
6038 ctx->flow.tunnel.tun_id = flow_tun_id;
6039 ctx->flow.vlan_tci = flow_vlan_tci;
6040 ctx->flow.nw_tos = flow_nw_tos;
6044 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6046 compose_output_action__(ctx, ofp_port, true);
6050 xlate_table_action(struct action_xlate_ctx *ctx,
6051 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6053 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6054 struct ofproto_dpif *ofproto = ctx->ofproto;
6055 struct rule_dpif *rule;
6056 uint16_t old_in_port;
6057 uint8_t old_table_id;
6059 old_table_id = ctx->table_id;
6060 ctx->table_id = table_id;
6062 /* Look up a flow with 'in_port' as the input port. */
6063 old_in_port = ctx->flow.in_port;
6064 ctx->flow.in_port = in_port;
6065 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
6068 if (table_id > 0 && table_id < N_TABLES) {
6069 struct table_dpif *table = &ofproto->tables[table_id];
6070 if (table->other_table) {
6071 ctx->tags |= (rule && rule->tag
6073 : rule_calculate_tag(&ctx->flow,
6074 &table->other_table->mask,
6079 /* Restore the original input port. Otherwise OFPP_NORMAL and
6080 * OFPP_IN_PORT will have surprising behavior. */
6081 ctx->flow.in_port = old_in_port;
6083 if (ctx->resubmit_hook) {
6084 ctx->resubmit_hook(ctx, rule);
6087 if (rule == NULL && may_packet_in) {
6089 * check if table configuration flags
6090 * OFPTC_TABLE_MISS_CONTROLLER, default.
6091 * OFPTC_TABLE_MISS_CONTINUE,
6092 * OFPTC_TABLE_MISS_DROP
6093 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6095 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
6099 struct rule_dpif *old_rule = ctx->rule;
6101 if (ctx->resubmit_stats) {
6102 rule_credit_stats(rule, ctx->resubmit_stats);
6107 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6108 ctx->rule = old_rule;
6112 ctx->table_id = old_table_id;
6114 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6116 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6117 MAX_RESUBMIT_RECURSION);
6118 ctx->max_resubmit_trigger = true;
6123 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6124 const struct ofpact_resubmit *resubmit)
6129 in_port = resubmit->in_port;
6130 if (in_port == OFPP_IN_PORT) {
6131 in_port = ctx->flow.in_port;
6134 table_id = resubmit->table_id;
6135 if (table_id == 255) {
6136 table_id = ctx->table_id;
6139 xlate_table_action(ctx, in_port, table_id, false);
6143 flood_packets(struct action_xlate_ctx *ctx, bool all)
6145 struct ofport_dpif *ofport;
6147 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6148 uint16_t ofp_port = ofport->up.ofp_port;
6150 if (ofp_port == ctx->flow.in_port) {
6155 compose_output_action__(ctx, ofp_port, false);
6156 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6157 compose_output_action(ctx, ofp_port);
6161 ctx->nf_output_iface = NF_OUT_FLOOD;
6165 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6166 enum ofp_packet_in_reason reason,
6167 uint16_t controller_id)
6169 struct ofputil_packet_in pin;
6170 struct ofpbuf *packet;
6172 ctx->slow |= SLOW_CONTROLLER;
6177 packet = ofpbuf_clone(ctx->packet);
6179 if (packet->l2 && packet->l3) {
6180 struct eth_header *eh;
6182 eth_pop_vlan(packet);
6185 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
6186 * LLC frame. Calculating the Ethernet type of these frames is more
6187 * trouble than seems appropriate for a simple assertion. */
6188 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
6189 || eh->eth_type == ctx->flow.dl_type);
6191 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6192 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6194 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6195 eth_push_vlan(packet, ctx->flow.vlan_tci);
6199 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6200 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6201 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6205 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6206 packet_set_tcp_port(packet, ctx->flow.tp_src,
6208 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6209 packet_set_udp_port(packet, ctx->flow.tp_src,
6216 pin.packet = packet->data;
6217 pin.packet_len = packet->size;
6218 pin.reason = reason;
6219 pin.controller_id = controller_id;
6220 pin.table_id = ctx->table_id;
6221 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6224 flow_get_metadata(&ctx->flow, &pin.fmd);
6226 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6227 ofpbuf_delete(packet);
6231 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6233 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6234 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6238 if (ctx->flow.nw_ttl > 1) {
6244 for (i = 0; i < ids->n_controllers; i++) {
6245 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6249 /* Stop processing for current table. */
6255 xlate_output_action(struct action_xlate_ctx *ctx,
6256 uint16_t port, uint16_t max_len, bool may_packet_in)
6258 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6260 ctx->nf_output_iface = NF_OUT_DROP;
6264 compose_output_action(ctx, ctx->flow.in_port);
6267 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6273 flood_packets(ctx, false);
6276 flood_packets(ctx, true);
6278 case OFPP_CONTROLLER:
6279 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6285 if (port != ctx->flow.in_port) {
6286 compose_output_action(ctx, port);
6288 xlate_report(ctx, "skipping output to input port");
6293 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6294 ctx->nf_output_iface = NF_OUT_FLOOD;
6295 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6296 ctx->nf_output_iface = prev_nf_output_iface;
6297 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6298 ctx->nf_output_iface != NF_OUT_FLOOD) {
6299 ctx->nf_output_iface = NF_OUT_MULTI;
6304 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6305 const struct ofpact_output_reg *or)
6307 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6308 if (port <= UINT16_MAX) {
6309 xlate_output_action(ctx, port, or->max_len, false);
6314 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6315 const struct ofpact_enqueue *enqueue)
6317 uint16_t ofp_port = enqueue->port;
6318 uint32_t queue_id = enqueue->queue;
6319 uint32_t flow_priority, priority;
6322 /* Translate queue to priority. */
6323 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6324 queue_id, &priority);
6326 /* Fall back to ordinary output action. */
6327 xlate_output_action(ctx, enqueue->port, 0, false);
6331 /* Check output port. */
6332 if (ofp_port == OFPP_IN_PORT) {
6333 ofp_port = ctx->flow.in_port;
6334 } else if (ofp_port == ctx->flow.in_port) {
6338 /* Add datapath actions. */
6339 flow_priority = ctx->flow.skb_priority;
6340 ctx->flow.skb_priority = priority;
6341 compose_output_action(ctx, ofp_port);
6342 ctx->flow.skb_priority = flow_priority;
6344 /* Update NetFlow output port. */
6345 if (ctx->nf_output_iface == NF_OUT_DROP) {
6346 ctx->nf_output_iface = ofp_port;
6347 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6348 ctx->nf_output_iface = NF_OUT_MULTI;
6353 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6355 uint32_t skb_priority;
6357 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6358 queue_id, &skb_priority)) {
6359 ctx->flow.skb_priority = skb_priority;
6361 /* Couldn't translate queue to a priority. Nothing to do. A warning
6362 * has already been logged. */
6366 struct xlate_reg_state {
6372 xlate_autopath(struct action_xlate_ctx *ctx,
6373 const struct ofpact_autopath *ap)
6375 uint16_t ofp_port = ap->port;
6376 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6378 if (!port || !port->bundle) {
6379 ofp_port = OFPP_NONE;
6380 } else if (port->bundle->bond) {
6381 /* Autopath does not support VLAN hashing. */
6382 struct ofport_dpif *slave = bond_choose_output_slave(
6383 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6385 ofp_port = slave->up.ofp_port;
6388 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6392 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6394 struct ofproto_dpif *ofproto = ofproto_;
6395 struct ofport_dpif *port;
6405 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6408 port = get_ofp_port(ofproto, ofp_port);
6409 return port ? port->may_enable : false;
6414 xlate_bundle_action(struct action_xlate_ctx *ctx,
6415 const struct ofpact_bundle *bundle)
6419 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6420 if (bundle->dst.field) {
6421 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6423 xlate_output_action(ctx, port, 0, false);
6428 xlate_learn_action(struct action_xlate_ctx *ctx,
6429 const struct ofpact_learn *learn)
6431 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6432 struct ofputil_flow_mod fm;
6433 uint64_t ofpacts_stub[1024 / 8];
6434 struct ofpbuf ofpacts;
6437 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6438 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6440 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6441 if (error && !VLOG_DROP_WARN(&rl)) {
6442 VLOG_WARN("learning action failed to modify flow table (%s)",
6443 ofperr_get_name(error));
6446 ofpbuf_uninit(&ofpacts);
6449 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6450 * means "infinite". */
6452 reduce_timeout(uint16_t max, uint16_t *timeout)
6454 if (max && (!*timeout || *timeout > max)) {
6460 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6461 const struct ofpact_fin_timeout *oft)
6463 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6464 struct rule_dpif *rule = ctx->rule;
6466 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6467 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6472 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6474 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6475 ? OFPUTIL_PC_NO_RECV_STP
6476 : OFPUTIL_PC_NO_RECV)) {
6480 /* Only drop packets here if both forwarding and learning are
6481 * disabled. If just learning is enabled, we need to have
6482 * OFPP_NORMAL and the learning action have a look at the packet
6483 * before we can drop it. */
6484 if (!stp_forward_in_state(port->stp_state)
6485 && !stp_learn_in_state(port->stp_state)) {
6493 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6495 if (is_ip_any(&ctx->base_flow)
6496 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6497 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6498 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6499 " but is not ECN capable");
6502 /* Set the ECN CE value in the tunneled packet. */
6503 ctx->flow.nw_tos |= IP_ECN_CE;
6511 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6512 struct action_xlate_ctx *ctx)
6514 bool was_evictable = true;
6515 const struct ofpact *a;
6518 /* Don't let the rule we're working on get evicted underneath us. */
6519 was_evictable = ctx->rule->up.evictable;
6520 ctx->rule->up.evictable = false;
6522 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6523 struct ofpact_controller *controller;
6524 const struct ofpact_metadata *metadata;
6532 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6533 ofpact_get_OUTPUT(a)->max_len, true);
6536 case OFPACT_CONTROLLER:
6537 controller = ofpact_get_CONTROLLER(a);
6538 execute_controller_action(ctx, controller->max_len,
6540 controller->controller_id);
6543 case OFPACT_ENQUEUE:
6544 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6547 case OFPACT_SET_VLAN_VID:
6548 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6549 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6553 case OFPACT_SET_VLAN_PCP:
6554 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6555 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6560 case OFPACT_STRIP_VLAN:
6561 ctx->flow.vlan_tci = htons(0);
6564 case OFPACT_PUSH_VLAN:
6565 /* XXX 802.1AD(QinQ) */
6566 ctx->flow.vlan_tci = htons(VLAN_CFI);
6569 case OFPACT_SET_ETH_SRC:
6570 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6574 case OFPACT_SET_ETH_DST:
6575 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6579 case OFPACT_SET_IPV4_SRC:
6580 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6583 case OFPACT_SET_IPV4_DST:
6584 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6587 case OFPACT_SET_IPV4_DSCP:
6588 /* OpenFlow 1.0 only supports IPv4. */
6589 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6590 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6591 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6595 case OFPACT_SET_L4_SRC_PORT:
6596 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6599 case OFPACT_SET_L4_DST_PORT:
6600 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6603 case OFPACT_RESUBMIT:
6604 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6607 case OFPACT_SET_TUNNEL:
6608 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6611 case OFPACT_SET_QUEUE:
6612 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6615 case OFPACT_POP_QUEUE:
6616 ctx->flow.skb_priority = ctx->orig_skb_priority;
6619 case OFPACT_REG_MOVE:
6620 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6623 case OFPACT_REG_LOAD:
6624 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6627 case OFPACT_DEC_TTL:
6628 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6634 /* Nothing to do. */
6637 case OFPACT_MULTIPATH:
6638 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6641 case OFPACT_AUTOPATH:
6642 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6646 ctx->ofproto->has_bundle_action = true;
6647 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6650 case OFPACT_OUTPUT_REG:
6651 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6655 ctx->has_learn = true;
6656 if (ctx->may_learn) {
6657 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6665 case OFPACT_FIN_TIMEOUT:
6666 ctx->has_fin_timeout = true;
6667 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6670 case OFPACT_CLEAR_ACTIONS:
6672 * Nothing to do because writa-actions is not supported for now.
6673 * When writa-actions is supported, clear-actions also must
6674 * be supported at the same time.
6678 case OFPACT_WRITE_METADATA:
6679 metadata = ofpact_get_WRITE_METADATA(a);
6680 ctx->flow.metadata &= ~metadata->mask;
6681 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6684 case OFPACT_GOTO_TABLE: {
6685 /* XXX remove recursion */
6686 /* It is assumed that goto-table is last action */
6687 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6688 ovs_assert(ctx->table_id < ogt->table_id);
6689 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6697 ctx->rule->up.evictable = was_evictable;
6702 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6703 struct ofproto_dpif *ofproto, const struct flow *flow,
6704 const struct initial_vals *initial_vals,
6705 struct rule_dpif *rule,
6706 uint8_t tcp_flags, const struct ofpbuf *packet)
6708 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6710 /* Flow initialization rules:
6711 * - 'base_flow' must match the kernel's view of the packet at the
6712 * time that action processing starts. 'flow' represents any
6713 * transformations we wish to make through actions.
6714 * - By default 'base_flow' and 'flow' are the same since the input
6715 * packet matches the output before any actions are applied.
6716 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6717 * of the received packet as seen by the kernel. If we later output
6718 * to another device without any modifications this will cause us to
6719 * insert a new tag since the original one was stripped off by the
6721 * - Tunnel 'flow' is largely cleared when transitioning between
6722 * the input and output stages since it does not make sense to output
6723 * a packet with the exact headers that it was received with (i.e.
6724 * the destination IP is us). The one exception is the tun_id, which
6725 * is preserved to allow use in later resubmit lookups and loads into
6727 * - Tunnel 'base_flow' is completely cleared since that is what the
6728 * kernel does. If we wish to maintain the original values an action
6729 * needs to be generated. */
6731 ctx->ofproto = ofproto;
6733 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6734 ctx->base_flow = ctx->flow;
6735 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6736 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6737 ctx->flow.tunnel.tun_id = initial_tun_id;
6739 ctx->packet = packet;
6740 ctx->may_learn = packet != NULL;
6741 ctx->tcp_flags = tcp_flags;
6742 ctx->resubmit_hook = NULL;
6743 ctx->report_hook = NULL;
6744 ctx->resubmit_stats = NULL;
6747 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6748 * into datapath actions in 'odp_actions', using 'ctx'. */
6750 xlate_actions(struct action_xlate_ctx *ctx,
6751 const struct ofpact *ofpacts, size_t ofpacts_len,
6752 struct ofpbuf *odp_actions)
6754 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6755 * that in the future we always keep a copy of the original flow for
6756 * tracing purposes. */
6757 static bool hit_resubmit_limit;
6759 enum slow_path_reason special;
6760 struct ofport_dpif *in_port;
6762 COVERAGE_INC(ofproto_dpif_xlate);
6764 ofpbuf_clear(odp_actions);
6765 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6767 ctx->odp_actions = odp_actions;
6770 ctx->has_learn = false;
6771 ctx->has_normal = false;
6772 ctx->has_fin_timeout = false;
6773 ctx->nf_output_iface = NF_OUT_DROP;
6776 ctx->max_resubmit_trigger = false;
6777 ctx->orig_skb_priority = ctx->flow.skb_priority;
6781 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6782 /* Do this conditionally because the copy is expensive enough that it
6783 * shows up in profiles.
6785 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6786 * believe that I wasn't using it without initializing it if I kept it
6787 * in a local variable. */
6788 ctx->orig_flow = ctx->flow;
6791 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6792 switch (ctx->ofproto->up.frag_handling) {
6793 case OFPC_FRAG_NORMAL:
6794 /* We must pretend that transport ports are unavailable. */
6795 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6796 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6799 case OFPC_FRAG_DROP:
6802 case OFPC_FRAG_REASM:
6805 case OFPC_FRAG_NX_MATCH:
6806 /* Nothing to do. */
6809 case OFPC_INVALID_TTL_TO_CONTROLLER:
6814 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6815 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6817 ctx->slow |= special;
6819 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6820 struct initial_vals initial_vals;
6821 uint32_t local_odp_port;
6823 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6824 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6826 add_sflow_action(ctx);
6828 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6829 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6831 /* We've let OFPP_NORMAL and the learning action look at the
6832 * packet, so drop it now if forwarding is disabled. */
6833 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6834 ofpbuf_clear(ctx->odp_actions);
6835 add_sflow_action(ctx);
6839 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6840 if (!hit_resubmit_limit) {
6841 /* We didn't record the original flow. Make sure we do from
6843 hit_resubmit_limit = true;
6844 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6845 struct ds ds = DS_EMPTY_INITIALIZER;
6847 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6848 &initial_vals, &ds);
6849 VLOG_ERR("Trace triggered by excessive resubmit "
6850 "recursion:\n%s", ds_cstr(&ds));
6855 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6856 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6858 ctx->odp_actions->data,
6859 ctx->odp_actions->size)) {
6860 ctx->slow |= SLOW_IN_BAND;
6862 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6864 compose_output_action(ctx, OFPP_LOCAL);
6867 if (ctx->ofproto->has_mirrors) {
6868 add_mirror_actions(ctx, &ctx->orig_flow);
6870 fix_sflow_action(ctx);
6874 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6875 * into datapath actions, using 'ctx', and discards the datapath actions. */
6877 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6878 const struct ofpact *ofpacts,
6881 uint64_t odp_actions_stub[1024 / 8];
6882 struct ofpbuf odp_actions;
6884 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6885 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6886 ofpbuf_uninit(&odp_actions);
6890 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6892 if (ctx->report_hook) {
6893 ctx->report_hook(ctx, s);
6897 /* OFPP_NORMAL implementation. */
6899 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6901 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6902 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6903 * the bundle on which the packet was received, returns the VLAN to which the
6906 * Both 'vid' and the return value are in the range 0...4095. */
6908 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6910 switch (in_bundle->vlan_mode) {
6911 case PORT_VLAN_ACCESS:
6912 return in_bundle->vlan;
6915 case PORT_VLAN_TRUNK:
6918 case PORT_VLAN_NATIVE_UNTAGGED:
6919 case PORT_VLAN_NATIVE_TAGGED:
6920 return vid ? vid : in_bundle->vlan;
6927 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6928 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6931 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6932 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6935 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6937 /* Allow any VID on the OFPP_NONE port. */
6938 if (in_bundle == &ofpp_none_bundle) {
6942 switch (in_bundle->vlan_mode) {
6943 case PORT_VLAN_ACCESS:
6946 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6947 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6948 "packet received on port %s configured as VLAN "
6949 "%"PRIu16" access port",
6950 in_bundle->ofproto->up.name, vid,
6951 in_bundle->name, in_bundle->vlan);
6957 case PORT_VLAN_NATIVE_UNTAGGED:
6958 case PORT_VLAN_NATIVE_TAGGED:
6960 /* Port must always carry its native VLAN. */
6964 case PORT_VLAN_TRUNK:
6965 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6967 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6968 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6969 "received on port %s not configured for trunking "
6971 in_bundle->ofproto->up.name, vid,
6972 in_bundle->name, vid);
6984 /* Given 'vlan', the VLAN that a packet belongs to, and
6985 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6986 * that should be included in the 802.1Q header. (If the return value is 0,
6987 * then the 802.1Q header should only be included in the packet if there is a
6990 * Both 'vlan' and the return value are in the range 0...4095. */
6992 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6994 switch (out_bundle->vlan_mode) {
6995 case PORT_VLAN_ACCESS:
6998 case PORT_VLAN_TRUNK:
6999 case PORT_VLAN_NATIVE_TAGGED:
7002 case PORT_VLAN_NATIVE_UNTAGGED:
7003 return vlan == out_bundle->vlan ? 0 : vlan;
7011 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7014 struct ofport_dpif *port;
7016 ovs_be16 tci, old_tci;
7018 vid = output_vlan_to_vid(out_bundle, vlan);
7019 if (!out_bundle->bond) {
7020 port = ofbundle_get_a_port(out_bundle);
7022 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7025 /* No slaves enabled, so drop packet. */
7030 old_tci = ctx->flow.vlan_tci;
7032 if (tci || out_bundle->use_priority_tags) {
7033 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7035 tci |= htons(VLAN_CFI);
7038 ctx->flow.vlan_tci = tci;
7040 compose_output_action(ctx, port->up.ofp_port);
7041 ctx->flow.vlan_tci = old_tci;
7045 mirror_mask_ffs(mirror_mask_t mask)
7047 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7052 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7054 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7055 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7059 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7061 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7064 /* Returns an arbitrary interface within 'bundle'. */
7065 static struct ofport_dpif *
7066 ofbundle_get_a_port(const struct ofbundle *bundle)
7068 return CONTAINER_OF(list_front(&bundle->ports),
7069 struct ofport_dpif, bundle_node);
7073 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7075 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7079 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7081 struct ofproto_dpif *ofproto = ctx->ofproto;
7082 mirror_mask_t mirrors;
7083 struct ofbundle *in_bundle;
7086 const struct nlattr *a;
7089 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7090 ctx->packet != NULL, NULL);
7094 mirrors = in_bundle->src_mirrors;
7096 /* Drop frames on bundles reserved for mirroring. */
7097 if (in_bundle->mirror_out) {
7098 if (ctx->packet != NULL) {
7099 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7100 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7101 "%s, which is reserved exclusively for mirroring",
7102 ctx->ofproto->up.name, in_bundle->name);
7108 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7109 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7112 vlan = input_vid_to_vlan(in_bundle, vid);
7114 /* Look at the output ports to check for destination selections. */
7116 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7117 ctx->odp_actions->size) {
7118 enum ovs_action_attr type = nl_attr_type(a);
7119 struct ofport_dpif *ofport;
7121 if (type != OVS_ACTION_ATTR_OUTPUT) {
7125 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7126 if (ofport && ofport->bundle) {
7127 mirrors |= ofport->bundle->dst_mirrors;
7135 /* Restore the original packet before adding the mirror actions. */
7136 ctx->flow = *orig_flow;
7141 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7143 if (!vlan_is_mirrored(m, vlan)) {
7144 mirrors = zero_rightmost_1bit(mirrors);
7148 mirrors &= ~m->dup_mirrors;
7149 ctx->mirrors |= m->dup_mirrors;
7151 output_normal(ctx, m->out, vlan);
7152 } else if (vlan != m->out_vlan
7153 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7154 struct ofbundle *bundle;
7156 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7157 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7158 && !bundle->mirror_out) {
7159 output_normal(ctx, bundle, m->out_vlan);
7167 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7168 uint64_t packets, uint64_t bytes)
7174 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7177 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7180 /* In normal circumstances 'm' will not be NULL. However,
7181 * if mirrors are reconfigured, we can temporarily get out
7182 * of sync in facet_revalidate(). We could "correct" the
7183 * mirror list before reaching here, but doing that would
7184 * not properly account the traffic stats we've currently
7185 * accumulated for previous mirror configuration. */
7189 m->packet_count += packets;
7190 m->byte_count += bytes;
7194 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7195 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7196 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7198 is_gratuitous_arp(const struct flow *flow)
7200 return (flow->dl_type == htons(ETH_TYPE_ARP)
7201 && eth_addr_is_broadcast(flow->dl_dst)
7202 && (flow->nw_proto == ARP_OP_REPLY
7203 || (flow->nw_proto == ARP_OP_REQUEST
7204 && flow->nw_src == flow->nw_dst)));
7208 update_learning_table(struct ofproto_dpif *ofproto,
7209 const struct flow *flow, int vlan,
7210 struct ofbundle *in_bundle)
7212 struct mac_entry *mac;
7214 /* Don't learn the OFPP_NONE port. */
7215 if (in_bundle == &ofpp_none_bundle) {
7219 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7223 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7224 if (is_gratuitous_arp(flow)) {
7225 /* We don't want to learn from gratuitous ARP packets that are
7226 * reflected back over bond slaves so we lock the learning table. */
7227 if (!in_bundle->bond) {
7228 mac_entry_set_grat_arp_lock(mac);
7229 } else if (mac_entry_is_grat_arp_locked(mac)) {
7234 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7235 /* The log messages here could actually be useful in debugging,
7236 * so keep the rate limit relatively high. */
7237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7238 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7239 "on port %s in VLAN %d",
7240 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7241 in_bundle->name, vlan);
7243 mac->port.p = in_bundle;
7244 tag_set_add(&ofproto->backer->revalidate_set,
7245 mac_learning_changed(ofproto->ml, mac));
7249 static struct ofbundle *
7250 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7251 bool warn, struct ofport_dpif **in_ofportp)
7253 struct ofport_dpif *ofport;
7255 /* Find the port and bundle for the received packet. */
7256 ofport = get_ofp_port(ofproto, in_port);
7258 *in_ofportp = ofport;
7260 if (ofport && ofport->bundle) {
7261 return ofport->bundle;
7264 /* Special-case OFPP_NONE, which a controller may use as the ingress
7265 * port for traffic that it is sourcing. */
7266 if (in_port == OFPP_NONE) {
7267 return &ofpp_none_bundle;
7270 /* Odd. A few possible reasons here:
7272 * - We deleted a port but there are still a few packets queued up
7275 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7276 * we don't know about.
7278 * - The ofproto client didn't configure the port as part of a bundle.
7279 * This is particularly likely to happen if a packet was received on the
7280 * port after it was created, but before the client had a chance to
7281 * configure its bundle.
7284 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7286 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7287 "port %"PRIu16, ofproto->up.name, in_port);
7292 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7293 * dropped. Returns true if they may be forwarded, false if they should be
7296 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7297 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7299 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7300 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7301 * checked by input_vid_is_valid().
7303 * May also add tags to '*tags', although the current implementation only does
7304 * so in one special case.
7307 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7310 struct ofproto_dpif *ofproto = ctx->ofproto;
7311 struct flow *flow = &ctx->flow;
7312 struct ofbundle *in_bundle = in_port->bundle;
7314 /* Drop frames for reserved multicast addresses
7315 * only if forward_bpdu option is absent. */
7316 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7317 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7321 if (in_bundle->bond) {
7322 struct mac_entry *mac;
7324 switch (bond_check_admissibility(in_bundle->bond, in_port,
7325 flow->dl_dst, &ctx->tags)) {
7330 xlate_report(ctx, "bonding refused admissibility, dropping");
7333 case BV_DROP_IF_MOVED:
7334 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7335 if (mac && mac->port.p != in_bundle &&
7336 (!is_gratuitous_arp(flow)
7337 || mac_entry_is_grat_arp_locked(mac))) {
7338 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7350 xlate_normal(struct action_xlate_ctx *ctx)
7352 struct ofport_dpif *in_port;
7353 struct ofbundle *in_bundle;
7354 struct mac_entry *mac;
7358 ctx->has_normal = true;
7360 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7361 ctx->packet != NULL, &in_port);
7363 xlate_report(ctx, "no input bundle, dropping");
7367 /* Drop malformed frames. */
7368 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7369 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7370 if (ctx->packet != NULL) {
7371 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7372 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7373 "VLAN tag received on port %s",
7374 ctx->ofproto->up.name, in_bundle->name);
7376 xlate_report(ctx, "partial VLAN tag, dropping");
7380 /* Drop frames on bundles reserved for mirroring. */
7381 if (in_bundle->mirror_out) {
7382 if (ctx->packet != NULL) {
7383 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7384 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7385 "%s, which is reserved exclusively for mirroring",
7386 ctx->ofproto->up.name, in_bundle->name);
7388 xlate_report(ctx, "input port is mirror output port, dropping");
7393 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7394 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7395 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7398 vlan = input_vid_to_vlan(in_bundle, vid);
7400 /* Check other admissibility requirements. */
7401 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7405 /* Learn source MAC. */
7406 if (ctx->may_learn) {
7407 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7410 /* Determine output bundle. */
7411 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7414 if (mac->port.p != in_bundle) {
7415 xlate_report(ctx, "forwarding to learned port");
7416 output_normal(ctx, mac->port.p, vlan);
7418 xlate_report(ctx, "learned port is input port, dropping");
7421 struct ofbundle *bundle;
7423 xlate_report(ctx, "no learned MAC for destination, flooding");
7424 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7425 if (bundle != in_bundle
7426 && ofbundle_includes_vlan(bundle, vlan)
7427 && bundle->floodable
7428 && !bundle->mirror_out) {
7429 output_normal(ctx, bundle, vlan);
7432 ctx->nf_output_iface = NF_OUT_FLOOD;
7436 /* Optimized flow revalidation.
7438 * It's a difficult problem, in general, to tell which facets need to have
7439 * their actions recalculated whenever the OpenFlow flow table changes. We
7440 * don't try to solve that general problem: for most kinds of OpenFlow flow
7441 * table changes, we recalculate the actions for every facet. This is
7442 * relatively expensive, but it's good enough if the OpenFlow flow table
7443 * doesn't change very often.
7445 * However, we can expect one particular kind of OpenFlow flow table change to
7446 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7447 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7448 * table, we add a special case that applies to flow tables in which every rule
7449 * has the same form (that is, the same wildcards), except that the table is
7450 * also allowed to have a single "catch-all" flow that matches all packets. We
7451 * optimize this case by tagging all of the facets that resubmit into the table
7452 * and invalidating the same tag whenever a flow changes in that table. The
7453 * end result is that we revalidate just the facets that need it (and sometimes
7454 * a few more, but not all of the facets or even all of the facets that
7455 * resubmit to the table modified by MAC learning). */
7457 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7458 * into an OpenFlow table with the given 'basis'. */
7460 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7463 if (minimask_is_catchall(mask)) {
7466 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7467 return tag_create_deterministic(hash);
7471 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7472 * taggability of that table.
7474 * This function must be called after *each* change to a flow table. If you
7475 * skip calling it on some changes then the pointer comparisons at the end can
7476 * be invalid if you get unlucky. For example, if a flow removal causes a
7477 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7478 * different wildcards to be created with the same address, then this function
7479 * will incorrectly skip revalidation. */
7481 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7483 struct table_dpif *table = &ofproto->tables[table_id];
7484 const struct oftable *oftable = &ofproto->up.tables[table_id];
7485 struct cls_table *catchall, *other;
7486 struct cls_table *t;
7488 catchall = other = NULL;
7490 switch (hmap_count(&oftable->cls.tables)) {
7492 /* We could tag this OpenFlow table but it would make the logic a
7493 * little harder and it's a corner case that doesn't seem worth it
7499 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7500 if (cls_table_is_catchall(t)) {
7502 } else if (!other) {
7505 /* Indicate that we can't tag this by setting both tables to
7506 * NULL. (We know that 'catchall' is already NULL.) */
7513 /* Can't tag this table. */
7517 if (table->catchall_table != catchall || table->other_table != other) {
7518 table->catchall_table = catchall;
7519 table->other_table = other;
7520 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7524 /* Given 'rule' that has changed in some way (either it is a rule being
7525 * inserted, a rule being deleted, or a rule whose actions are being
7526 * modified), marks facets for revalidation to ensure that packets will be
7527 * forwarded correctly according to the new state of the flow table.
7529 * This function must be called after *each* change to a flow table. See
7530 * the comment on table_update_taggable() for more information. */
7532 rule_invalidate(const struct rule_dpif *rule)
7534 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7536 table_update_taggable(ofproto, rule->up.table_id);
7538 if (!ofproto->backer->need_revalidate) {
7539 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7541 if (table->other_table && rule->tag) {
7542 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7544 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7550 set_frag_handling(struct ofproto *ofproto_,
7551 enum ofp_config_flags frag_handling)
7553 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7554 if (frag_handling != OFPC_FRAG_REASM) {
7555 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7563 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7564 const struct flow *flow,
7565 const struct ofpact *ofpacts, size_t ofpacts_len)
7567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7568 struct initial_vals initial_vals;
7569 struct odputil_keybuf keybuf;
7570 struct dpif_flow_stats stats;
7574 struct action_xlate_ctx ctx;
7575 uint64_t odp_actions_stub[1024 / 8];
7576 struct ofpbuf odp_actions;
7578 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7579 odp_flow_key_from_flow(&key, flow,
7580 ofp_port_to_odp_port(ofproto, flow->in_port));
7582 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7584 initial_vals.vlan_tci = flow->vlan_tci;
7585 initial_vals.tunnel_ip_tos = 0;
7586 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7587 packet_get_tcp_flags(packet, flow), packet);
7588 ctx.resubmit_stats = &stats;
7590 ofpbuf_use_stub(&odp_actions,
7591 odp_actions_stub, sizeof odp_actions_stub);
7592 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7593 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7594 odp_actions.data, odp_actions.size, packet);
7595 ofpbuf_uninit(&odp_actions);
7603 set_netflow(struct ofproto *ofproto_,
7604 const struct netflow_options *netflow_options)
7606 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7608 if (netflow_options) {
7609 if (!ofproto->netflow) {
7610 ofproto->netflow = netflow_create();
7612 return netflow_set_options(ofproto->netflow, netflow_options);
7614 netflow_destroy(ofproto->netflow);
7615 ofproto->netflow = NULL;
7621 get_netflow_ids(const struct ofproto *ofproto_,
7622 uint8_t *engine_type, uint8_t *engine_id)
7624 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7626 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7630 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7632 if (!facet_is_controller_flow(facet) &&
7633 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7634 struct subfacet *subfacet;
7635 struct ofexpired expired;
7637 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7638 if (subfacet->path == SF_FAST_PATH) {
7639 struct dpif_flow_stats stats;
7641 subfacet_reinstall(subfacet, &stats);
7642 subfacet_update_stats(subfacet, &stats);
7646 expired.flow = facet->flow;
7647 expired.packet_count = facet->packet_count;
7648 expired.byte_count = facet->byte_count;
7649 expired.used = facet->used;
7650 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7655 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7657 struct facet *facet;
7659 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7660 send_active_timeout(ofproto, facet);
7664 static struct ofproto_dpif *
7665 ofproto_dpif_lookup(const char *name)
7667 struct ofproto_dpif *ofproto;
7669 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7670 hash_string(name, 0), &all_ofproto_dpifs) {
7671 if (!strcmp(ofproto->up.name, name)) {
7679 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7680 const char *argv[], void *aux OVS_UNUSED)
7682 struct ofproto_dpif *ofproto;
7685 ofproto = ofproto_dpif_lookup(argv[1]);
7687 unixctl_command_reply_error(conn, "no such bridge");
7690 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7692 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7693 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7697 unixctl_command_reply(conn, "table successfully flushed");
7701 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7702 const char *argv[], void *aux OVS_UNUSED)
7704 struct ds ds = DS_EMPTY_INITIALIZER;
7705 const struct ofproto_dpif *ofproto;
7706 const struct mac_entry *e;
7708 ofproto = ofproto_dpif_lookup(argv[1]);
7710 unixctl_command_reply_error(conn, "no such bridge");
7714 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7715 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7716 struct ofbundle *bundle = e->port.p;
7717 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7718 ofbundle_get_a_port(bundle)->odp_port,
7719 e->vlan, ETH_ADDR_ARGS(e->mac),
7720 mac_entry_age(ofproto->ml, e));
7722 unixctl_command_reply(conn, ds_cstr(&ds));
7727 struct action_xlate_ctx ctx;
7733 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7734 const struct rule_dpif *rule)
7736 ds_put_char_multiple(result, '\t', level);
7738 ds_put_cstr(result, "No match\n");
7742 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7743 table_id, ntohll(rule->up.flow_cookie));
7744 cls_rule_format(&rule->up.cr, result);
7745 ds_put_char(result, '\n');
7747 ds_put_char_multiple(result, '\t', level);
7748 ds_put_cstr(result, "OpenFlow ");
7749 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7750 ds_put_char(result, '\n');
7754 trace_format_flow(struct ds *result, int level, const char *title,
7755 struct trace_ctx *trace)
7757 ds_put_char_multiple(result, '\t', level);
7758 ds_put_format(result, "%s: ", title);
7759 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7760 ds_put_cstr(result, "unchanged");
7762 flow_format(result, &trace->ctx.flow);
7763 trace->flow = trace->ctx.flow;
7765 ds_put_char(result, '\n');
7769 trace_format_regs(struct ds *result, int level, const char *title,
7770 struct trace_ctx *trace)
7774 ds_put_char_multiple(result, '\t', level);
7775 ds_put_format(result, "%s:", title);
7776 for (i = 0; i < FLOW_N_REGS; i++) {
7777 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7779 ds_put_char(result, '\n');
7783 trace_format_odp(struct ds *result, int level, const char *title,
7784 struct trace_ctx *trace)
7786 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7788 ds_put_char_multiple(result, '\t', level);
7789 ds_put_format(result, "%s: ", title);
7790 format_odp_actions(result, odp_actions->data, odp_actions->size);
7791 ds_put_char(result, '\n');
7795 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7797 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7798 struct ds *result = trace->result;
7800 ds_put_char(result, '\n');
7801 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7802 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7803 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7804 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7808 trace_report(struct action_xlate_ctx *ctx, const char *s)
7810 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7811 struct ds *result = trace->result;
7813 ds_put_char_multiple(result, '\t', ctx->recurse);
7814 ds_put_cstr(result, s);
7815 ds_put_char(result, '\n');
7819 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7820 void *aux OVS_UNUSED)
7822 const char *dpname = argv[1];
7823 struct ofproto_dpif *ofproto;
7824 struct ofpbuf odp_key;
7825 struct ofpbuf *packet;
7826 struct initial_vals initial_vals;
7832 ofpbuf_init(&odp_key, 0);
7835 ofproto = ofproto_dpif_lookup(dpname);
7837 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7841 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7842 /* ofproto/trace dpname flow [-generate] */
7843 const char *flow_s = argv[2];
7844 const char *generate_s = argv[3];
7846 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7847 * flow. We guess which type it is based on whether 'flow_s' contains
7848 * an '(', since a datapath flow always contains '(') but an
7849 * OpenFlow-like flow should not (in fact it's allowed but I believe
7850 * that's not documented anywhere).
7852 * An alternative would be to try to parse 'flow_s' both ways, but then
7853 * it would be tricky giving a sensible error message. After all, do
7854 * you just say "syntax error" or do you present both error messages?
7855 * Both choices seem lousy. */
7856 if (strchr(flow_s, '(')) {
7859 /* Convert string to datapath key. */
7860 ofpbuf_init(&odp_key, 0);
7861 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7863 unixctl_command_reply_error(conn, "Bad flow syntax");
7867 /* The user might have specified the wrong ofproto but within the
7868 * same backer. That's OK, ofproto_receive() can find the right
7870 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7871 odp_key.size, &flow, NULL, &ofproto, NULL,
7873 unixctl_command_reply_error(conn, "Invalid flow");
7876 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7880 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7882 unixctl_command_reply_error(conn, error_s);
7887 initial_vals.vlan_tci = flow.vlan_tci;
7888 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7891 /* Generate a packet, if requested. */
7893 packet = ofpbuf_new(0);
7894 flow_compose(packet, &flow);
7896 } else if (argc == 7) {
7897 /* ofproto/trace dpname priority tun_id in_port mark packet */
7898 const char *priority_s = argv[2];
7899 const char *tun_id_s = argv[3];
7900 const char *in_port_s = argv[4];
7901 const char *mark_s = argv[5];
7902 const char *packet_s = argv[6];
7903 uint32_t in_port = atoi(in_port_s);
7904 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7905 uint32_t priority = atoi(priority_s);
7906 uint32_t mark = atoi(mark_s);
7909 msg = eth_from_hex(packet_s, &packet);
7911 unixctl_command_reply_error(conn, msg);
7915 ds_put_cstr(&result, "Packet: ");
7916 s = ofp_packet_to_string(packet->data, packet->size);
7917 ds_put_cstr(&result, s);
7920 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7921 flow.tunnel.tun_id = tun_id;
7922 initial_vals.vlan_tci = flow.vlan_tci;
7923 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7925 unixctl_command_reply_error(conn, "Bad command syntax");
7929 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7930 unixctl_command_reply(conn, ds_cstr(&result));
7933 ds_destroy(&result);
7934 ofpbuf_delete(packet);
7935 ofpbuf_uninit(&odp_key);
7939 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7940 const struct ofpbuf *packet,
7941 const struct initial_vals *initial_vals, struct ds *ds)
7943 struct rule_dpif *rule;
7945 ds_put_cstr(ds, "Flow: ");
7946 flow_format(ds, flow);
7947 ds_put_char(ds, '\n');
7949 rule = rule_dpif_lookup(ofproto, flow);
7951 trace_format_rule(ds, 0, 0, rule);
7952 if (rule == ofproto->miss_rule) {
7953 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7954 } else if (rule == ofproto->no_packet_in_rule) {
7955 ds_put_cstr(ds, "\nNo match, packets dropped because "
7956 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7960 uint64_t odp_actions_stub[1024 / 8];
7961 struct ofpbuf odp_actions;
7963 struct trace_ctx trace;
7966 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7969 ofpbuf_use_stub(&odp_actions,
7970 odp_actions_stub, sizeof odp_actions_stub);
7971 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7972 rule, tcp_flags, packet);
7973 trace.ctx.resubmit_hook = trace_resubmit;
7974 trace.ctx.report_hook = trace_report;
7975 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7978 ds_put_char(ds, '\n');
7979 trace_format_flow(ds, 0, "Final flow", &trace);
7980 ds_put_cstr(ds, "Datapath actions: ");
7981 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7982 ofpbuf_uninit(&odp_actions);
7984 if (trace.ctx.slow) {
7985 enum slow_path_reason slow;
7987 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7988 "slow path because it:");
7989 for (slow = trace.ctx.slow; slow; ) {
7990 enum slow_path_reason bit = rightmost_1bit(slow);
7994 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7997 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8000 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8003 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8006 ds_put_cstr(ds, "\n\t (The datapath actions are "
8007 "incomplete--for complete actions, "
8008 "please supply a packet.)");
8011 case SLOW_CONTROLLER:
8012 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8013 "to the OpenFlow controller.");
8016 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8017 "than the datapath supports.");
8024 if (slow & ~SLOW_MATCH) {
8025 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8026 "the special slow-path processing.");
8033 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8034 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8037 unixctl_command_reply(conn, NULL);
8041 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8042 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8045 unixctl_command_reply(conn, NULL);
8048 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8049 * 'reply' describing the results. */
8051 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8053 struct facet *facet;
8057 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8058 if (!facet_check_consistency(facet)) {
8063 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8067 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8068 ofproto->up.name, errors);
8070 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8075 ofproto_dpif_self_check(struct unixctl_conn *conn,
8076 int argc, const char *argv[], void *aux OVS_UNUSED)
8078 struct ds reply = DS_EMPTY_INITIALIZER;
8079 struct ofproto_dpif *ofproto;
8082 ofproto = ofproto_dpif_lookup(argv[1]);
8084 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8085 "ofproto/list for help)");
8088 ofproto_dpif_self_check__(ofproto, &reply);
8090 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8091 ofproto_dpif_self_check__(ofproto, &reply);
8095 unixctl_command_reply(conn, ds_cstr(&reply));
8099 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8100 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8101 * to destroy 'ofproto_shash' and free the returned value. */
8102 static const struct shash_node **
8103 get_ofprotos(struct shash *ofproto_shash)
8105 const struct ofproto_dpif *ofproto;
8107 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8108 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8109 shash_add_nocopy(ofproto_shash, name, ofproto);
8112 return shash_sort(ofproto_shash);
8116 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8117 const char *argv[] OVS_UNUSED,
8118 void *aux OVS_UNUSED)
8120 struct ds ds = DS_EMPTY_INITIALIZER;
8121 struct shash ofproto_shash;
8122 const struct shash_node **sorted_ofprotos;
8125 shash_init(&ofproto_shash);
8126 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8127 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8128 const struct shash_node *node = sorted_ofprotos[i];
8129 ds_put_format(&ds, "%s\n", node->name);
8132 shash_destroy(&ofproto_shash);
8133 free(sorted_ofprotos);
8135 unixctl_command_reply(conn, ds_cstr(&ds));
8140 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8142 const struct shash_node **ports;
8144 struct avg_subfacet_rates lifetime;
8145 unsigned long long int minutes;
8146 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8148 minutes = (time_msec() - ofproto->created) / min_ms;
8151 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8153 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8156 lifetime.add_rate = 0.0;
8157 lifetime.del_rate = 0.0;
8160 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8161 dpif_name(ofproto->backer->dpif));
8163 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8164 ofproto->n_hit, ofproto->n_missed);
8165 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8166 " life span: %llu(ms)\n",
8167 hmap_count(&ofproto->subfacets),
8168 avg_subfacet_count(ofproto),
8169 ofproto->max_n_subfacet,
8170 avg_subfacet_life_span(ofproto));
8171 if (minutes >= 60) {
8172 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8174 if (minutes >= 60 * 24) {
8175 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8177 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8179 ports = shash_sort(&ofproto->up.port_by_name);
8180 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8181 const struct shash_node *node = ports[i];
8182 struct ofport *ofport = node->data;
8183 const char *name = netdev_get_name(ofport->netdev);
8184 const char *type = netdev_get_type(ofport->netdev);
8187 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8189 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8190 if (odp_port != OVSP_NONE) {
8191 ds_put_format(ds, "%"PRIu32":", odp_port);
8193 ds_put_cstr(ds, "none:");
8196 if (strcmp(type, "system")) {
8197 struct netdev *netdev;
8200 ds_put_format(ds, " (%s", type);
8202 error = netdev_open(name, type, &netdev);
8207 error = netdev_get_config(netdev, &config);
8209 const struct smap_node **nodes;
8212 nodes = smap_sort(&config);
8213 for (i = 0; i < smap_count(&config); i++) {
8214 const struct smap_node *node = nodes[i];
8215 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8216 node->key, node->value);
8220 smap_destroy(&config);
8222 netdev_close(netdev);
8224 ds_put_char(ds, ')');
8226 ds_put_char(ds, '\n');
8232 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8233 const char *argv[], void *aux OVS_UNUSED)
8235 struct ds ds = DS_EMPTY_INITIALIZER;
8236 const struct ofproto_dpif *ofproto;
8240 for (i = 1; i < argc; i++) {
8241 ofproto = ofproto_dpif_lookup(argv[i]);
8243 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8244 "for help)", argv[i]);
8245 unixctl_command_reply_error(conn, ds_cstr(&ds));
8248 show_dp_format(ofproto, &ds);
8251 struct shash ofproto_shash;
8252 const struct shash_node **sorted_ofprotos;
8255 shash_init(&ofproto_shash);
8256 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8257 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8258 const struct shash_node *node = sorted_ofprotos[i];
8259 show_dp_format(node->data, &ds);
8262 shash_destroy(&ofproto_shash);
8263 free(sorted_ofprotos);
8266 unixctl_command_reply(conn, ds_cstr(&ds));
8271 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8272 int argc OVS_UNUSED, const char *argv[],
8273 void *aux OVS_UNUSED)
8275 struct ds ds = DS_EMPTY_INITIALIZER;
8276 const struct ofproto_dpif *ofproto;
8277 struct subfacet *subfacet;
8279 ofproto = ofproto_dpif_lookup(argv[1]);
8281 unixctl_command_reply_error(conn, "no such bridge");
8285 update_stats(ofproto->backer);
8287 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8288 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8290 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8291 subfacet->dp_packet_count, subfacet->dp_byte_count);
8292 if (subfacet->used) {
8293 ds_put_format(&ds, "%.3fs",
8294 (time_msec() - subfacet->used) / 1000.0);
8296 ds_put_format(&ds, "never");
8298 if (subfacet->facet->tcp_flags) {
8299 ds_put_cstr(&ds, ", flags:");
8300 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8303 ds_put_cstr(&ds, ", actions:");
8304 if (subfacet->slow) {
8305 uint64_t slow_path_stub[128 / 8];
8306 const struct nlattr *actions;
8309 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8310 slow_path_stub, sizeof slow_path_stub,
8311 &actions, &actions_len);
8312 format_odp_actions(&ds, actions, actions_len);
8314 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8316 ds_put_char(&ds, '\n');
8319 unixctl_command_reply(conn, ds_cstr(&ds));
8324 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8325 int argc OVS_UNUSED, const char *argv[],
8326 void *aux OVS_UNUSED)
8328 struct ds ds = DS_EMPTY_INITIALIZER;
8329 struct ofproto_dpif *ofproto;
8331 ofproto = ofproto_dpif_lookup(argv[1]);
8333 unixctl_command_reply_error(conn, "no such bridge");
8337 flush(&ofproto->up);
8339 unixctl_command_reply(conn, ds_cstr(&ds));
8344 ofproto_dpif_unixctl_init(void)
8346 static bool registered;
8352 unixctl_command_register(
8354 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8355 2, 6, ofproto_unixctl_trace, NULL);
8356 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8357 ofproto_unixctl_fdb_flush, NULL);
8358 unixctl_command_register("fdb/show", "bridge", 1, 1,
8359 ofproto_unixctl_fdb_show, NULL);
8360 unixctl_command_register("ofproto/clog", "", 0, 0,
8361 ofproto_dpif_clog, NULL);
8362 unixctl_command_register("ofproto/unclog", "", 0, 0,
8363 ofproto_dpif_unclog, NULL);
8364 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8365 ofproto_dpif_self_check, NULL);
8366 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8367 ofproto_unixctl_dpif_dump_dps, NULL);
8368 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8369 ofproto_unixctl_dpif_show, NULL);
8370 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8371 ofproto_unixctl_dpif_dump_flows, NULL);
8372 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8373 ofproto_unixctl_dpif_del_flows, NULL);
8376 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8378 * This is deprecated. It is only for compatibility with broken device drivers
8379 * in old versions of Linux that do not properly support VLANs when VLAN
8380 * devices are not used. When broken device drivers are no longer in
8381 * widespread use, we will delete these interfaces. */
8384 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8387 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8389 if (realdev_ofp_port == ofport->realdev_ofp_port
8390 && vid == ofport->vlandev_vid) {
8394 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8396 if (ofport->realdev_ofp_port) {
8399 if (realdev_ofp_port && ofport->bundle) {
8400 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8401 * themselves be part of a bundle. */
8402 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8405 ofport->realdev_ofp_port = realdev_ofp_port;
8406 ofport->vlandev_vid = vid;
8408 if (realdev_ofp_port) {
8409 vsp_add(ofport, realdev_ofp_port, vid);
8416 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8418 return hash_2words(realdev_ofp_port, vid);
8421 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8422 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8423 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8424 * it would return the port number of eth0.9.
8426 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8427 * function just returns its 'realdev_odp_port' argument. */
8429 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8430 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8432 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8433 uint16_t realdev_ofp_port;
8434 int vid = vlan_tci_to_vid(vlan_tci);
8435 const struct vlan_splinter *vsp;
8437 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8438 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8439 hash_realdev_vid(realdev_ofp_port, vid),
8440 &ofproto->realdev_vid_map) {
8441 if (vsp->realdev_ofp_port == realdev_ofp_port
8442 && vsp->vid == vid) {
8443 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8447 return realdev_odp_port;
8450 static struct vlan_splinter *
8451 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8453 struct vlan_splinter *vsp;
8455 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8456 &ofproto->vlandev_map) {
8457 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8465 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8466 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8467 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8468 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8469 * eth0 and store 9 in '*vid'.
8471 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8472 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8475 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8476 uint16_t vlandev_ofp_port, int *vid)
8478 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8479 const struct vlan_splinter *vsp;
8481 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8486 return vsp->realdev_ofp_port;
8492 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8493 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8494 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8495 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8496 * always the case unless VLAN splinters are enabled), returns false without
8497 * making any changes. */
8499 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8504 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8509 /* Cause the flow to be processed as if it came in on the real device with
8510 * the VLAN device's VLAN ID. */
8511 flow->in_port = realdev;
8512 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8517 vsp_remove(struct ofport_dpif *port)
8519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8520 struct vlan_splinter *vsp;
8522 vsp = vlandev_find(ofproto, port->up.ofp_port);
8524 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8525 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8528 port->realdev_ofp_port = 0;
8530 VLOG_ERR("missing vlan device record");
8535 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8537 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8539 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8540 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8541 == realdev_ofp_port)) {
8542 struct vlan_splinter *vsp;
8544 vsp = xmalloc(sizeof *vsp);
8545 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8546 hash_int(port->up.ofp_port, 0));
8547 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8548 hash_realdev_vid(realdev_ofp_port, vid));
8549 vsp->realdev_ofp_port = realdev_ofp_port;
8550 vsp->vlandev_ofp_port = port->up.ofp_port;
8553 port->realdev_ofp_port = realdev_ofp_port;
8555 VLOG_ERR("duplicate vlan device record");
8560 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8562 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8563 return ofport ? ofport->odp_port : OVSP_NONE;
8566 static struct ofport_dpif *
8567 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8569 struct ofport_dpif *port;
8571 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8572 hash_int(odp_port, 0),
8573 &backer->odp_to_ofport_map) {
8574 if (port->odp_port == odp_port) {
8583 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8585 struct ofport_dpif *port;
8587 port = odp_port_to_ofport(ofproto->backer, odp_port);
8588 if (port && &ofproto->up == port->up.ofproto) {
8589 return port->up.ofp_port;
8594 static unsigned long long int
8595 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8597 unsigned long long int dc;
8598 unsigned long long int avg;
8600 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8601 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8607 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8611 if (ofproto->n_update_stats) {
8612 avg_c = (double)ofproto->total_subfacet_count
8613 / ofproto->n_update_stats;
8620 show_dp_rates(struct ds *ds, const char *heading,
8621 const struct avg_subfacet_rates *rates)
8623 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
8624 heading, rates->add_rate, rates->del_rate);
8628 update_max_subfacet_count(struct ofproto_dpif *ofproto)
8630 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
8631 hmap_count(&ofproto->subfacets));
8634 /* Compute exponentially weighted moving average, adding 'new' as the newest,
8635 * most heavily weighted element. 'base' designates the rate of decay: after
8636 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
8639 exp_mavg(double *avg, int base, double new)
8641 *avg = (*avg * (base - 1) + new) / base;
8645 update_moving_averages(struct ofproto_dpif *ofproto)
8647 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8649 /* Update hourly averages on the minute boundaries. */
8650 if (time_msec() - ofproto->last_minute >= min_ms) {
8651 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
8652 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
8654 /* Update daily averages on the hour boundaries. */
8655 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
8656 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
8657 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
8660 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
8661 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
8662 ofproto->subfacet_add_count = 0;
8663 ofproto->subfacet_del_count = 0;
8664 ofproto->last_minute += min_ms;
8669 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8671 ofproto->n_hit += delta;
8674 const struct ofproto_class ofproto_dpif_class = {
8709 port_is_lacp_current,
8710 NULL, /* rule_choose_table */
8717 rule_modify_actions,
8726 get_cfm_remote_mpids,
8731 get_stp_port_status,
8738 is_mirror_output_bundle,
8739 forward_bpdu_changed,
8740 set_mac_table_config,